Terahertz (THz) technology has experienced rapid development in the past two decades. Growing numbers of interdisciplinary applications are emerging, including materials science, physics, communications, and security as well as biomedicine. THz biophotonics involves studies applying THz photonic technology in biomedicine, which has attracted attention due to the unique features of THz waves, such as the high sensitivity to water, resonance with biomolecules, favorable spatial resolution, capacity to probe the water–biomolecule interactions, and nonionizing photon energy. Despite the great potential, THz biophotonics is still at an early stage of development. There is a lack of standards for instrumentation, measurement protocols, and data analysis, which makes it difficult to make comparisons among all the work published. In this article, we give a comprehensive review of the key findings that have underpinned research into biomedical applications of THz technology. In particular, we will focus on the advances made in general THz instrumentation and specific THz-based instruments for biomedical applications. We will also discuss the theories describing the interaction between THz light and biomedical samples. We aim to provide an overview of both basic biomedical research as well as pre-clinical and clinical applications under investigation. The paper aims to provide a clear picture of the achievements, challenges, and future perspectives of THz biophotonics.

Over the last two decades, terahertz (THz) biophotonics has been rapidly growing as an important interdisciplinary branch of THz science and biomedical engineering. Being loosely defined as the electromagnetic waves in the frequency range of 0.1–10 THz, THz waves have some intermediate properties between microwaves and infrared, which are driving growth in this research field. For example, the hydrogen-bond network of water has broadband relaxation modes ranging from GHz to THz frequencies. This is the origin of the high permittivity and absorption of water in the THz band that make THz waves sensitive to water.1 One of the most famous and impressive THz images that have promoted the development of THz biophotonics to some extent is the leaf image scanned by Hu and Nuss in 1995,2 showing a dramatic change in the THz response for the leaf after 48 h due to the loss of water. Apart from water, intermolecular vibrations and rotations of some small biomolecules, such as nucleobase and amino acids,3,4 also lie within the THz range, providing unique fingerprints which can be specifically detected. Additionally, the shorter wavelengths offer THz waves more optical properties than microwaves. Indeed, THz has an alternative name of “far-infrared,” which was commonly used before the 21st century. THz waves are usually manipulated by optics in a form of beams in free space. Compared to the waveguiding approach generally used for microwaves, a THz optical system can be easily arranged into a spectroscopy or an imaging setup. The shorter wavelength also brings another important merit of better spatial resolution compared to microwaves. This, combined with its nonionizing nature due to the low photon energy and the wavelength-comparable depth of penetration into tissues, means that THz waves have the potential to be a good modality for in vivo imaging and diagnosis of superficial tissues. Moreover, the high sensitivity to tissue water content variation and limited bulk scattering render THz imaging an attractive diagnostic tool for diseases/pathologies/injuries that are defined by or correlated with abnormal surface tissue water content. The above characteristics have endowed THz waves with great potential for applications in the field of biomedicine, hence pushing forward a popular research field called THz biophotonics.

THz biomedical applications span a wide range of scales from molecules to cells and tissues. However, as a newly developed technique, there are no technical standards set for the measurements, and great effort is being put into the THz instrumentation, interaction theory, data analysis, and investigation of THz biomedical effects. THz instrumentation builds the foundation for many types of applications. The theory, modeling, and data processing have also experienced promising development over the years, and they undoubtedly also play important roles. Recently, concerns about the biological safety of intense THz radiation have also been frequently discussed. These research areas not only improve the technology and broaden the applications, but also result in a large divergence, making them difficult to compare between different works and evaluate the state of art. As an interdisciplinary branch between THz science and biomedical engineering, it is important that an overview of these different aspects be presented to allow readers from different fields to gain a comprehensive view. Several reviews related to the field of THz biophotonics have been published over the last ten years. There are topical reviews on THz cancer diagnosis by Yu et al.,5 Peng et al.,6 and Zaytsev et al.7 Reviews on skin include those by Nikitkina et al.8 and Wang et al.,9 and there are more general reviews on biomedical applications by Fan et al.,10 Yang et al.,11 and Sun et al.12 However, most of these articles have been focused on the application aspect or a specific topic. In this manuscript, we aim to provide a comprehensive overview of THz biophotonics, the scope of which is illustrated in Fig. 1. Covering these different aspects of THz biophotonics is helpful for understanding the development, achievements, challenges, and future perspectives, which are essential for the development of this promising interdisciplinary area.

FIG. 1.

The scope of THz biophotonics covered in this review.

FIG. 1.

The scope of THz biophotonics covered in this review.

Close modal

Emission, detection, and beam-guiding components are the three essential elements for a THz system. The historical name of “THz gap” originates from the lack of efficient THz emission and detection devices. Nowadays, this gap has not been completely filled as can be seen if the power, efficiency, and sensitivity of sources and detectors are compared to those available for visible wavelengths. However, considerable progress has been made toward narrowing the gap and to enable growing amounts of applications. THz systems applied to THz biophotonics can be classified into two categories: pulsed time-domain THz systems and continuous-wave (CW) THz systems. The two categories do not simply differ in emission mode; they are very different in several aspects, namely, sources and detectors, experimental protocols, and information acquired. A comprehensive overview of THz radiation and detection techniques is not within the scope of this manuscript, but this can be found from other reviews or handbooks.13,14 Here, we focus on the systems that have been applied or have a high potential to be applied in THz biomedical research, highlighting their advantages, shortfalls, and scenarios of applications.

1. Pulsed time-domain THz systems

The name “pulsed time-domain THz systems” describes the two common features of this category: pulsed THz emission and time-domain detection. They are also commonly named as THz time-domain spectroscopy (THz-TDS) systems. Figure 2(a) shows the simplified fundamental optical setup for a pulsed time-domain THz system, which includes the essential components of a femtosecond (fs) laser, a beam splitter and an optical delay line (ODL). These optics control the beam delay between the pumping and probing beams sent to the THz emitter and detector, respectively, such that the THz waveform can be sampled in the time-domain.14 A practical system may contain a lot more optics according to the specific user requirements, such as the laser amplifier, attenuators, waveplates, gratings, optical choppers, etc. In the THz optics region, a couple of laser-driven techniques based on different mechanisms can be used for the generation and detection of THz-waves, for example, ultra-fast carriers in photoconductive antennas (PCAs), optical rectification and electro-optic (EO) sampling in nonlinear crystals, plasma in air or liquids, and spintronic currents in a ferromagnetic/heavy-metal structure. These techniques all provide some outstanding features that make pulsed time-domain THz systems highly useful for numerous applications. For example, they all provide a picosecond level time-resolution, a broad bandwidth, and a field-detection ability comprising both magnitude and phase information for the spectrum. However, they can have different characteristics including the bandwidth, power, and flexibility, etc. In the following, we will introduce the two conventional techniques based on PCAs and EO crystals, and a relatively new emerging technology based on spintronic emission.

FIG. 2.

Pulsed time-domain THz systems: (a) fundamental optical setting for a pulsed time-domain THz system. The THz optics region can be applied with various THz emission and detection devices, such as (b) photoconductive antennas, (c) nonlinear crystals, or (d) THz spintronic emitter. M: mirror, BS: beam splitter, ODL: optical delay line, EO: electro-optic, QWP: quarter-wave plate, WP: Wollaston prism, and PD: photodetector.

FIG. 2.

Pulsed time-domain THz systems: (a) fundamental optical setting for a pulsed time-domain THz system. The THz optics region can be applied with various THz emission and detection devices, such as (b) photoconductive antennas, (c) nonlinear crystals, or (d) THz spintronic emitter. M: mirror, BS: beam splitter, ODL: optical delay line, EO: electro-optic, QWP: quarter-wave plate, WP: Wollaston prism, and PD: photodetector.

Close modal
a. Photoconductive antenna

The invention of the photoconductive switch in 1980s has had a great impact on the whole THz society including THz biophotonics.15 Tremendous amounts of work have been reported based on this technique. Figure 2(b) shows how a PCA works as a THz emitter and a detector. On the emitter side, the biased voltage on the electrodes accelerates the ultrafast carriers photoexcited by the femtosecond (fs) laser. This forms a transient current that radiates a picosecond pulse containing broadband THz frequencies. The electrodes act as the antenna to radiate the THz waves in the direction of laser propagation, and a hyperhemispherical silicon (Si) lens is used to collimate the radiated beam. On the detector side, the THz waves are focused by the Si lens. The laser-excited carriers in the semiconductor are accelerated by the THz electric field. This forms a current to be detected and lock-in amplified. As the carrier lifetime is much shorter than the THz pulse duration, the generated current represents the THz electric field at the moment it interacts with the fs pulse, and the complete THz waveform is achieved by shifting the delay line to sample the THz electric field at different positions in the time-domain. PCAs have two unique features that make them especially suitable for biomedical applications. First, the metallic electrodes on the antenna allow direct control of the output polarization. Second, the compatibility with fibers enables a much higher flexibility in the optical arrangement. Polarization states usually need to be specified in biomedical research as the high absorption of many biosamples requires a reflection geometry to be used, making the measurement polarization-dependent. PCAs are usually linear in polarization for emission and detection depending on the electrode design. The polarization direction can be simply set by mechanically controlling the orientation of the antennas. Electrical control of the polarization direction has also been achieved by designing cross-electrodes.16 The fiber-coupled modality makes it possible to freely move or rotate the antenna without affecting the laser-antenna alignment. This is usually achieved by using a fs fiber-laser with the wavelength centered at 1550 nm. In this case, all the THz optics can be assembled into a compact and portable probe head for flexible sensing,17,18 or even in a prototype of an endoscope.19 They can also be assembled on translation stages to perform a raster scan by moving the optics rather than the sample;20 this is especially useful for in vivo studies. The fiber-connection also makes it possible to freely rotate the incident angle of a reflection system, particularly required in ellipsometry, which sets the incident angle close to the Brewster angle of a specific sample.21 

b. Electro-optic crystal

Optical rectification is another mechanism that can be used to generate broadband THz radiation using a EO crystal, as shown in Fig. 2(c). This technique utilizes the large second-order nonlinear susceptibility of the crystal to induce a polarization proportional to the envelope of the fs pulse, giving rise to THz radiation.22 The emitted THz wave is regarded as being rectified from the optical beam. EO crystals can also be used to coherently detect the THz electric field in time-domain, which is called EO sampling. This approach does not directly measure the THz electric field or the induced current; instead, it measures the change of the polarization of the optical beam due to the presence of the THz wave in the EO crystal. The EO technique has three major advantages over PCAs. First, without being limited by the carrier lifetime, EO generation and detection can support a much shorter pulse in the time-domain, providing a better temporal hence depth resolution. Fourier transforming a shorter pulse also provides a broader bandwidth, of which the upper frequency limit can be up to the mid-infrared depending on the crystal used.23 This is akin to depth resolution in optical coherence tomography (OCT) where increasing optical bandwidth leads to decreased voxel depth.24 Second, it can generate THz pulses with very high peak-intensity up to sub-mJ and peak electric field up to MV/cm.25 The high peak power has been utilized to investigate the biological effects of intense THz radiation, which will be detailed in Sec. IV. Third, without the need for electrodes, direct spatial patterning on the pumping or probing beam can be used to perform ghost imaging (i.e., compressed sensing).26–28 The major limitation of the EO technique is the need of laser amplification required by the nonlinear process, which restricts the repetition rate to a kHz level that fast-sampling of the THz waveforms becomes unavailable.

c. THz spintronic emitter

THz spintronic emitters are a new kind of THz source which have emerged in the last five years. The emitter is a multiple layered structure including a ferromagnetic and a nonmagnetic thin film. The generation mechanism is based on the inverse spin Hall effect to produce spin current, which comes from the diffusion of the spin-polarized electrons photoexcited in the ferromagnetic layer by the fs pulse. This spin current is then converted into a charge current in the nonmagnetic metal film, giving rise to THz radiation.29 The detection of the emitted THz waves can be done by either EO sampling or a PCA. The former can detect frequencies over 10 THz, which is sometimes needed as spintronic emitters can cover a bandwidth up to 30 THz.30,31 The latter approach provides simpler optics and a better flexibility, as shown in Fig. 2(d). THz spintronic emitters have many exciting features. Apart from the potential ultrabroad bandwidth, their emission efficiency remains nearly constant with the input pumping power from few mW to few tens of mW, with the pumping wavelengths from 800 to 1550 nm.30 They also have a low requirement on pumping pulse energy; thus, lasers with a high repetition rate can be used without further amplification. The low-cost and compact fs-fiber lasers can be directly integrated, providing the flexibility of the THz optics discussed for the PCAs. The polarization can also be simply controlled by rotating the orientation of the magnetic field applied to the emitter. Note that usually only a few mT is required; hence, bulky magnets are unnecessary. Similar to the EO crystals, direct patterning on the pumping beam can also be achieved in THz spintronic emitters for compressed sensing, as reported by Chen et al.32 The simple structure enables the fabrication of the emitter on a bendable substrate,33 providing more flexibility. It can also be scaled to a large area up to 7.5 cm in diameter, as presented by Seifert et al.34 The large device can generate intense THz pulses up to 300 kV cm−1, offering options for investigating THz nonthermal biomedical effects (see Sec. IV). As a relatively new technique, there are still very few biomedical applications based on spintronic emitters reported. However, given the previously mentioned advantages and the rapid development in progress, spintronics has the potential to have a greater role in the future.

2. Continuous-wave and other THz systems

a. Generation

There are many ways to generate CW THz waves, ranging from vacuum devices, such as the backward-wave oscillator (BWO), solid-state devices, such as Gunn diodes, to four types of photonic devices, including gas lasers, quantum cascade lasers (QCLs), THz parametric sources, and THz photomixers. These sources have very different operating mechanisms and characteristics. BWOs and solid-state diodes are electronic devices originally developed from the microwave side. They rely on frequency multipliers to operate at higher frequencies, with a significant sacrifice on the power. For example, the multiplied output power of solid-state diodes decreases from hundreds of mW to sub-mW when the frequency is increased from 0.1 to 1 THz.35 Diode multiplier chains can also be integrated as the extender heads with a vector network analyzer (VNA). The extender heads work both as the emitter and the detector, by harmonic generation or mixing the fundamental microwave source up to THz frequencies. VNAs measure both amplitude and phase, reflection, and transmission simultaneously with a high signal-to-noise ratio (SNR). These characteristics make them very useful in testing the performance of a device by coupling to the waveguide ports, while they can also be employed in a quasi-optical arrangement for material characterization.36 Free-electron lasers (FELs) are also vacuum sources but operate through a very different mechanism. FELs generate coherent THz radiation by converting the kinetic energy of a relativistic electron beam to electromagnetic waves in a periodic static magnetic field. THz FELs have a few irreplaceable features, including a high tunability of frequency in a ultrabroad band, the extremely high peak power up to gigawatts, and the flexible emission mode from CW to ultra-short pulses less than a picosecond.37,38 The drawback is that conventional FELs are very large facilities with a high cost and complexity; hence, they are more suitable for fundamental scientific research, such as investigating THz biological effects due to the intense radiation.39,40 However, table-top FELs are also available due to the lower requirement of the electron-beam energy for THz emission, which offers more opportunities for other biomedical applications, such as imaging.41 On the other hand, gas lasers and QCLs are developed from the infrared side. Gas lasers emit extremely narrow-linewidth THz radiation from particular gases, such as methanol, typically pumped by a CO2 laser. A QCL-pumped compact THz gas laser has also been recently reported.42 The operating frequency can be tuned by using different gases and pump lines; however, the resulting spectrum is discrete. QCLs are quantum solid-state devices utilizing the intersubband transitions. They have received great attention over the last two decades as research pushes to overcome the limitations in the working frequency, operating temperature, and output power of QCLs. For example, recent work by Hu's group demonstrates a compact THz QCL operating at 4 THz at temperatures up to 250 K, which can be cooled by portable thermoelectric coolers instead of bulky cryogenics.43 Different from the above source, THz parametric sources and photomixers can both be continuously tuned over a broad band, but they are very different in terms of power level and operation mode. THz parametric sources including THz parametric generators (TPGs) and THz parametric oscillators (TPOs), provide high peak-power pulses with a low repetition rate at tens to hundreds of Hz. They have a high gain efficiency over 1–3 THz. Photomixing is similar to pulsed PCAs, but they modulate the photoexcited carriers at a frequency equal to the frequency difference between two CW laser beams, thus emitting CW THz waves. They operate with a relatively low power at the μW level.

b. Detection

There are also several methods for CW THz detection, usually selected according to the characteristics of the source used and experimental requirements. For fundamental intensity detection, pyroelectric detectors, bolometers, and Golay cells can be used. They are sensitive single-pixel detectors, responsive across a broad band and adaptive to most kinds of sources. THz matrix cameras, usually based on microbolometer arrays or pyroelectric arrays, can be used for fast THz imaging if bright THz illumination is available. Schottky diodes are frequency mixers that down-convert the detected signal to a much lower frequency. They require a local oscillator as another THz source to provide a frequency close to that of the detected signal, with the difference frequency being measured. Some devices work as both a generator and a detector, such as THz parametric detection in nonlinear crystals,44 using laser feedback interferometry in QCLs,45 and photoconductive detection in THz photomixers.46 The latter two approaches are interferometric-based detection; thus, they are phase-sensitive. Most of the other detection mechanisms only measure the intensity; additional interferometric optical settings and/or algorithms are needed to extract the phase.47 

c. Characteristics and applications

Despite the divergent mechanisms, CW THz systems have a common advantage of narrow linewidth, compared to the typical 2 GHz resolution limit of time-domain systems. CW sources, such as parametric generators (100 MHz linewidth),48 QCLs (tens of MHz linewidth),49 and gas lasers (kHz linewidth)42 have a much better resolution. Combined with the frequency-tuning ability for most of CW sources, they provide an excellent spectral resolution for spectroscopy, which has been utilized in gas sensing. However, it should also be pointed out that the absorption features of most biomedical samples with spectral fingerprints are relatively wide, typically over a few tens of GHz, which can be well resolved by pulsed time-domain systems.3,4,50 CW systems have some limitations for spectroscopy. Vacuum and solid-state devices have a limited tuning bandwidth due to the power reduction with frequency upconversion. THz parametric generators and THz photomixers have a broader tuning range but they are typically slow. Another advantage of many CW THz sources is the high output power at their optimal frequencies, which enables a deeper penetration depth and a better SNR. The high brightness is essential for real-time imaging with a THz camera, since the beam energy has to be distributed over a large number of pixels. The high power is also required for scattering scanning nearfield optical microscopy (s-SNOM). In s-SNOM, only a tiny amount of the input THz light is scattered from the scanning tip and detected due to the large mismatch between the THz spot size and the tip, and a high harmonic component is extracted to suppress the background reflection;51 these introduce a significant loss. For example, using the second-order harmonic component, less than one millionth of the input energy can be detected.52 A strong input power is necessary to provide a good SNR.

CW THz systems are often used in biomedical imaging, partially due to the more robust and simpler optical settings, and faster acquisition speed compared to pulsed time-domain systems with a comparable SNR. This can be seen from Table I, where we summarize a few different CW THz systems using different combinations of sources and detectors for THz biomedical imaging and spectroscopy. The diverse selections of different devices show the high flexibility of CW THz systems. It is difficult to directly compare the performance of these systems given the diverse performance on various aspects. A potential direction for CW THz systems is to develop high-brightness, compact, sensitive, and low-cost real-time imaging systems, similar to those in the visible regime. This is needed to progress in vivo biomedical applications and will also promote the THz applications in many other fields, such as security and industry.

TABLE I.

Different CW THz systems for biomedical applications.

TypeSourceDetectorυ (THz)SampleReference
Imaging BWO Golay cell 0.279 Chicken bone Li et al.53  
  Pyroelectric detector 0.52–0.71 Leaf, pig tongue, chicken heart, pig cervical canal, liver-cancer, and breast cancer samples Dobroiu et al.54  
 Gunn diode Golay cell 0.32 Burned porcine skin Lu et al.55  
  Schottky diode 0.3 Human colonic tissues Chen et al.56  
  Pyroelectric detector 0.1 In vivo rabbit cornea Taylor et al.57  
 Diode multiplier chain Schottky diode 0.65 In vivo human cornea Sung et al.58  
 Gas laser Pyroelectric array 2.52 Human hepatocellular carcinoma Rong et al.59  
  Bolometer 0.584 Fresh normal and cancerous human colonic tissue Doradla et al.60  
  Golay cell 2.52 Rat traumatic brain injury Shi et al.61  
  Schottky diode 2.5 Mouse brain slices Siegel et al.62  
 QCL Bolometer 3.7 Sample model of liver, fat, muscle and tendon, rat brain, rat liver Kim et al.63  
  Microbolometer array 2.8 Human skin slice Locatelli et al.64  
  QCL 2.59 Porcine tissues Lim et al.65  
 Photomixer Bolometer 0.23 Human liver Knobloch et al.66  
  Photomixer Canine's basal cell tumor Siebert et al.67  
 TPO Pyroelectric detector 1–2.5 Drugs Kawase et al.68  
  Bolometer 0.9–2.5 Pork and chicken tissues Wang et al.69  
Spectroscopy/point scan Photomixer Golay cell 0.2–1 Human skin in vivo Bennett et al.70  
  Photomixer 0.4–1.2 Bacillus thuringiensis Zhang et al.71  
  Photomixer 0.1 Cornea of human eye in vivo Safonova et al.72  
 VNA extender source VNA extender detector 0.22–0.33 Ex vivo lamb eyes Tamminen et al.73  
TypeSourceDetectorυ (THz)SampleReference
Imaging BWO Golay cell 0.279 Chicken bone Li et al.53  
  Pyroelectric detector 0.52–0.71 Leaf, pig tongue, chicken heart, pig cervical canal, liver-cancer, and breast cancer samples Dobroiu et al.54  
 Gunn diode Golay cell 0.32 Burned porcine skin Lu et al.55  
  Schottky diode 0.3 Human colonic tissues Chen et al.56  
  Pyroelectric detector 0.1 In vivo rabbit cornea Taylor et al.57  
 Diode multiplier chain Schottky diode 0.65 In vivo human cornea Sung et al.58  
 Gas laser Pyroelectric array 2.52 Human hepatocellular carcinoma Rong et al.59  
  Bolometer 0.584 Fresh normal and cancerous human colonic tissue Doradla et al.60  
  Golay cell 2.52 Rat traumatic brain injury Shi et al.61  
  Schottky diode 2.5 Mouse brain slices Siegel et al.62  
 QCL Bolometer 3.7 Sample model of liver, fat, muscle and tendon, rat brain, rat liver Kim et al.63  
  Microbolometer array 2.8 Human skin slice Locatelli et al.64  
  QCL 2.59 Porcine tissues Lim et al.65  
 Photomixer Bolometer 0.23 Human liver Knobloch et al.66  
  Photomixer Canine's basal cell tumor Siebert et al.67  
 TPO Pyroelectric detector 1–2.5 Drugs Kawase et al.68  
  Bolometer 0.9–2.5 Pork and chicken tissues Wang et al.69  
Spectroscopy/point scan Photomixer Golay cell 0.2–1 Human skin in vivo Bennett et al.70  
  Photomixer 0.4–1.2 Bacillus thuringiensis Zhang et al.71  
  Photomixer 0.1 Cornea of human eye in vivo Safonova et al.72  
 VNA extender source VNA extender detector 0.22–0.33 Ex vivo lamb eyes Tamminen et al.73  

The experimental configuration determines how the THz beam is guided in free space, interacts with the sample, and is detected. Numerous configurations have been proposed for different applications, especially for biophotonics due to the complexity of biological systems. For a specific experiment, the configuration is usually designed according to the sample properties (e.g., transparent or absorptive, solid or liquid) and the measurement types (e.g., imaging or spectroscopy, in vivo or in vitro). In the following, both the conventional and several novel configurations for THz biophotonics will be introduced, with their features and adaptability being highlighted.

1. Conventional configurations

Transmission and reflection are the two most conventional configurations. Transmission is a very straightforward modality for sample measurement. As shown in Fig. 3(a), a sample is directly placed in the THz beam region with the beam being either focused or collimated. The beam is partially reflected and absorbed, with the rest being received by the detector following transmission through the sample; this is described as the sample signal. By comparing this sample signal to the reference signal, generally acquired by measuring the transmitted signal in the absence of the sample, one is able to analyze the sample properties. This is particularly useful for pulsed time-domain THz measurements, in which both broadband amplitude and phase can be obtained, and the complex dielectric properties of the sample can be determined by Beer–Lambert's law. Another key advantage of the transmission configuration is the high sensitivity to most THz-transparent samples, originating from the long interaction distance between the THz light and the sample. Mathematically, it can be seen from the time-domain that the pulse can be delayed for a long distance by having a large sample thickness, as long as the attenuation is acceptable. Equivalently, this means the phase variation in the frequency-domain has no upper limit. In contrast, the phase of a signal reflected off a sample can only vary within 180°, that is, only the pulse shape is changed without a delay. Therefore, transmission is usually the best choice for transparent bulk samples. However, fresh biological samples are absorptive as they have a large water concentration, and they are usually difficult to cut into thin films. Some dehydrated biological slides are too thin to provide a long THz–sample interaction distance. In vivo measurements can only be performed with a reflection geometry. Therefore, in many cases a reflection configuration is required to overcome these issues.

FIG. 3.

Some commonly used THz configurations: (a) transmission, (b) window-based reflection, (c) prism-based reflection, and (d) ellipsometry.

FIG. 3.

Some commonly used THz configurations: (a) transmission, (b) window-based reflection, (c) prism-based reflection, and (d) ellipsometry.

Close modal

A reflection measurement can be performed by directly illuminating THz light onto a free-standing sample, or through a certain supporting medium, such as a window or a prism. Free-standing reflection is mostly used when sample characterization is not required. This is because the reflection plane of a free-standing sample will have certain height positioning error compared to that of a reference medium, resulting in the sample-reference relative phase error, known as the “phase uncertainty” problem as the accuracy bottleneck.74,75 This configuration has been mainly applied in CW THz imaging when only the intensity information is used to form reflectivity images, or when the absolute phase is used to form holographic images.59,60,64 Using a window or a prism to support the sample is more frequently used for biomedical studies. The supporting medium not only provides a stable reflection plane to eliminate the positioning error between the sample and reference (air) reflections, but also introduces a flat interface to optimize the focusing. This configuration requires good contact with the sample; thus, it can be used to measure liquids, fresh tissues, or skin in vivo. Dehydrated samples can be directly prepared on the supporting medium, in order to prevent the presence of air gaps caused by the rigid surface. Characterizing solid-state samples in reflection can be done using a reference-free configuration of ellipsometry, which will be introduced in Sec. II B 2.

A reflection configuration is specified by various systematic parameters, including the angle of incidence, the polarization, the properties of the supporting medium, and the light–sample interaction structure. These properties should be designed to optimize the measurement sensitivity. For example, Møller compares the sensitivities of transmission (for different sample thicknesses), ordinary (silicon-supported) reflection and attenuated total internal reflection (ATR) for transparent heptane and absorptive water, which shows a huge difference in the measured optical responses.76,77 The analysis shows that the absorptive water is best characterized by the ATR geometry. A similar analysis of free-standing reflection and quartz-supported reflection can be found in Refs. 74 and 75, showing that a reflection configuration has the best sensitivity for samples with properties close to that of the supporting window. For thin films, a sandwich structure is shown to be capable of efficiently enhancing the contrast.78 The selection of the structure and incident angle may also affect the imaging capability. For example, a prism setup generally only allows single-point spectroscopy due to the alignment variation in a two-dimensional scan. For other imaging setups at a large incident angle, the focused beam spot becomes elliptical and a worse resolution in the scanning direction parallel to the incident plane should be considered. Compared to transmission, reflection measurements are more critical in the optical alignment and more sensitive to various errors. Placing a sample slightly out of focus has little influence in transmission, but positioning and tilting errors will sensitively affect the reflection. During the measurement, the largest error source usually comes from the phase. As discussed above, the limited 180° phase variation range indicates that a small phase error could significantly change the characterization results. The phase can be influenced by many factors, such as delay-stage or laser jittering error, temperature-induced refractive-index change in fibers or other optical components, and sample positioning error. These errors should be carefully controlled in a reflection measurement.

2. Ellipsometry

Ellipsometry is a reference-free characterization technique, sometimes known as a self-reference method. It has been widely applied in the infrared and ultra-violet regimes as an accurate technology for investigating the dielectric function of different materials.79 The fundamental mechanism is the polarization-dependent reflections of samples according to the Fresnel equations. It measures the p- and s-reflections off a sample and expresses their ratio as a function of the sample properties. By solving this inverse problem of light–matter interactions, the dielectric properties of the investigated material can be extracted. In this way, no reference medium is required. Therefore, ellipsometry is very suitable for solid-state absorptive materials since they cannot be well characterized by either transmission, free-standing reflection, or medium-support reflection.21,80,81 Ellipsometry also shows an excellent sensitivity for thin-film characterization, providing a higher phase contrast than transmission.21,82

Since the polarization-dependency is the foundation of the sensitivity, ellipsometry is usually carried out close to the Brewster angle of a sample where the largest difference between the p- and s-reflection coefficients can be found. This can be easily understood from the opposite example, such as at normal incidence. In that case, the p- and s-reflections are completely identical and a ratio of 1 will always be obtained regardless of the properties of the measured sample. Measuring close to the Brewster angle brings two difficulties. First, at this region the p-reflection varies significantly with the incident angle, especially for materials with a large refractive index. This means the incident angle has to be set and read precisely. Using fully fiber-coupled PCAs allows flexible setting of the incident angle, and using large f-number optics can efficiently reduce the beam divergence to reduce the incident angle error. The only price is the larger beam spot due to the diffraction limit. Second, as the p-reflection is much weaker than the s-reflection close to the Brewster angle, the polarization has to be controlled precisely to make sure no s-component is leaked and detected when measuring the p-signal. Some calibration methods have been proposed to improve the polarization accuracy.21,83 Due to the critical requirements on the optical setup and polarization control, utilizing ellipsometry for biomedical applications is still rare, but recently Chen et al. proposed an example of in vivo characterization of skin,84 showing the ability to probe the hydration and cellular structure simultaneously. The ellipsometry configuration in this work consists of two right-angle prisms, one silicon prism on top and one gold-coated prism (i.e., perfect mirrors) on the bottom mounted symmetrically, as shown in Fig. 4. Putting the double-prism system in a transmission configuration offers two alternative beam paths, as shown in Fig. 4(a). The two positions provide two incident angles of 33° and 57° to the Si-subject/air interface. The large refractive index of the Si prism compared to biosamples gives rise to very distinct reflection coefficients at the two angles. Switching between the two angles can be conveniently achieved by vertically moving the double-prism system with a translation stage without changing the THz optics. Additionally, using the polarization control with three polarizers introduced in the previous work,21 four uncorrelated spectra from four configurations given by different combinations of incident angles and polarizations can be obtained. This significantly enriches the spectral information obtained, this will be necessary for characterizing samples with a complicated structure, such as skin (e.g., multiple layers and anisotropic). The application to skin characterization will be explained in detail in Sec. V D 2. Although in this work the skin was in contact with a prism to enable a flat reflection interface, theoretically ellipsometry can be applied to noncontact in vivo measurements without a phase uncertainty, which provides an ideal modality for the clinical applications. Some hard tissues, such as bones and teeth, or dehydrated tissues, can also be investigated with THz ellipsometry. Therefore, it has great potential to be further explored for biomedical applications in the future.

FIG. 4.

Multi-configuration ellipsometer for skin characterization: (a) the double-prism system used to introduce two alternative incident angles of θi1 and θi2. Ep and Es represent the electric fields with p- and s-polarizations, respectively. (b) The integration of the double-prism system with a transmission system to provide four uncorrelated reflected spectra. P1, P2, and P3 are three polarizers, with P1 and P3 fixed at 45° and P2 being rotated to select the p- or s-signals. Reproduced with permission from Chen et al., Adv. Photonics Res. 2, 2000024 (2021).84 Copyright 2021 Authors, licensed under a Creative Commons Attribution (CC BY) license.

FIG. 4.

Multi-configuration ellipsometer for skin characterization: (a) the double-prism system used to introduce two alternative incident angles of θi1 and θi2. Ep and Es represent the electric fields with p- and s-polarizations, respectively. (b) The integration of the double-prism system with a transmission system to provide four uncorrelated reflected spectra. P1, P2, and P3 are three polarizers, with P1 and P3 fixed at 45° and P2 being rotated to select the p- or s-signals. Reproduced with permission from Chen et al., Adv. Photonics Res. 2, 2000024 (2021).84 Copyright 2021 Authors, licensed under a Creative Commons Attribution (CC BY) license.

Close modal

3. Microscopy and the near-field

The THz wavelengths of 0.03–3 mm prevent far-field imaging techniques from resolving many important spatial details at the micrometer level; thus, near-field techniques are needed to overcome this limit. The most impressive near-field images have spatial resolution of 10 nm and temporal resolution of 50 fs;85 this is achieved by placing an AFM tip in the near-field of the sample. Unfortunately, methods based on placing metallic tips next to the sample are not compatible with biological samples for three reasons: first, raster scanning is too slow although it might be worth the wait; second, biosamples cannot be flat on a nanometer scale like solid-state samples; and third, sample hydration further complicates an already complex setup given the high THz absorption of water. This is why most THz imaging of biological materials has been performed in the far-field. Nevertheless, there have been studies showing near-field THz imaging of biosamples.56,86–95 These references use different systems and will be discussed below.

a. THz waveguide/fiber coupling

The work done by the group of Chi-Kuang Sun56,90–92 all utilizes the same basic concept: couple free-space THz radiation onto a THz fiber or waveguide; then, this fiber/waveguide is raster scanned in the near-field of a sample with a far-field detector collecting all the radiation. They usually use a 320 GHz Gunn oscillator as their source and a Schottky diode or a Golay cell as the detector. Such a technique can also measure over a broadband frequency, typically driven by VNAs to measure both the phase and amplitude. Regardless of their single-frequency source, they have been able to distinguish between cancerous and healthy tissues56,90,91 and also observe differences in the transmitted THz signal in the blood vessels of mouse ears due to an insulin injection.92 The biggest problem with this work is the use of a waveguide or a THz fiber as these generally work well only for a few frequencies. Further, the physical raster scanning mechanism will most likely limit this technology to applications that need good resolution but do not need the results in real-time.

b. Optical resolution THz imaging

Optical resolution THz imaging is based on placing the sample on an EO crystal emitter (or detector) and using a visible light pump (or probe) to spatially map-out the THz field variations due to the sample. This idea was demonstrated in 2011 in a previous study87 although not for biosamples. It was in 2020 that Okada et al.86 used this technique to image breast cancer in a tissue sample. Since the THz emission or detection is localized to where the visible light pump or probe is, placing the sample in the near-field results in a THz image with optical wavelength resolution. The authors used an optical CCD array to map out the THz near-field interaction of a sample placed onto top of a EO crystal detector,87 whereas Okada placed their sample onto an EO crystal emitter and raster scanned the optical pump beam that generates the THz radiation.86 Both techniques achieve around 15–20 μm resolution with image acquisition taking about 1 min; however, the major difference comes from laser systems powering these setups. A previous study87 used a 1 kHz regenerative amplified laser because the optical probe beam needs to be spread over the sample area with enough fluence, whereas in another study86 the optical beam is focused to a spot so the 80 MHz rate fiber-laser achieves the necessary fluences. A fiber-based laser is much smaller and cheaper, making it much more commercially appealing; however, both have been used to achieve similar physical resolution while measuring the amplitude and phase over similar frequency ranges. For this reason, this technique is likely to use fiber-based systems in the future for fundamental research of biological systems.

c. Single-pixel THz imaging with optical resolution

The study of Barr et al.88 and Stantchev et al.89 pioneered the use of a single-pixel THz camera with the spatial THz-light modulator placed in the near-field of the sample. A single-pixel camera works by spatially patterning a beam of radiation with many patterns and then recording the resulting transmissions or reflections from an object with a single-pixel detector. The image of the object is reconstructed via computer post-processing.96 Their first work on this approach in 201889 used an optical pump to photoexcite charge-carriers in undoped silicon and move the plasma frequency from below to above the THz frequency range, thus switching the semiconductor response from dielectric to conducting. Further, they synchronized the THz pulses such that they passed through the silicon a few picoseconds after optical excitation, long before the charge carriers had diffused thus the spatial pattern was not blurred due to carrier diffusion. This results in the THz resolution being determined by the distance between the object and modulator.97 The problem is that the spatial modulator requires optical pulses with large energy per pulse per area that is only achievable with 1 kHz repetition-rate regenerative amplified lasers, which makes the whole system slow, big, and expensive. In their next work,88 they used a CW THz source and high-powered LED as the optical pump for a silicon photo-modulator, with the modification that the measurement was performed in a total-internal-reflection geometry as opposed to ordinary transmission. This of course results in carrier diffusion limiting the THz resolution; however, it greatly reduces the system size and cost as well as improving the acquisition speed (due to higher signal to noise ratio) and can still give subwavelength resolution for low THz frequencies. The interesting part of this technology is that the carrier injection/depletion in the modulator can also be achieved by electrical biasing instead of optical pumping,98 whereby the resolution would be limited by the photo-lithographic techniques used to manufacture the modulator. However, such electrical modulator arrays are currently in heavy development with recent work showing a 16 × 16 graphene array with a 1 kHz switch-rate.99 

d. On-chip near-field THz imaging

The final technique we mention here is the development of an integrated system-on-a-chip near-field imaging device working at 0.55 THz by Pfeiffer.93,94 The device works by having a local oscillator connected to a split-ring resonator and finally connected to a detector. When the split-ring resonator is exposed to the sample, its resonance frequency will shift and thus the signal arriving at the detector will change. The split-ring resonator is only affected by the sample via inductive coupling to enhance the resolution to about 10 μm. As a proof-of-principle in an earlier study,93 an image of a human fingerprint is taken and in another study,94 they image a breast tissue sample. The downsides of this device is the single frequency operation and that the inductive coupling mechanism results in the measurement of only the real permittivity, which is unlikely to be enough for research purposes during clinical trials. However, all of the components are compatible with CMOS technology; thus, this device including all the readout electronics is very small, a few centimeters, and can be powered by a 5 V USB cable with 0.6 W of power consumption.94 Therefore, these devices have the potential to be used by households and not just hospitals or laboratories.

4. Handheld, endoscopic, and robotic geometries

Flexibility is crucial for in vivo applications due to the need to measure different locations of the body. Unlike ex vivo samples that can be prepared into ideal flat slides with a desired thickness, living tissues usually have irregular shapes, rough surfaces, and are in inconvenient positions. Developing THz instrumentation that is compact, portable, and flexible to adapt with these applications is a hot area of research. There are mainly three strategies, handheld, endoscopic, and robotic geometries. The flexibility of these strategies is usually provided by fibers or flexible waveguides, opposed to free-space beam-guiding where the alignment is usually so critical that there is little freedom of movement.

With their compact size and the adaptability to be coupled with fiber lasers, PCAs have been designed into various handheld probing structures. One of the earliest designs is the handheld probe-head product by TeraView. The earlier product from 2004 included one emitter surrounded by six detectors.100,101 A later design was reduced to one emitter and one detector, but with two rotating Risley prisms to enable a 1D scan along a 1.5 cm line,17,102 as shown in Figs. 5(a) and 5(b). The THz beam is focused on top of the quartz window at the end. The probe head can be freely moved to scan any sample contacting the quartz window without affecting the alignment, similar to a clinical ultrasonic probe. Mini-Z and micro-Z are another two commercial products by Zomega that were developed around 2010103,104 in which all the optics are assembled in a single box. They are more portable but less flexible. The handheld THz probe can also be used as an otoscope, as proposed by Ji et al.105 The THz probe head consists of an emitter based on photo-Dember currents and a PCA detector, assembled with a commercial optical otoscope, as shown in Fig. 5(c). The THz beam was well collimated over a long propagation distance. Different sizes of ear specula can be integrated, with the smallest diameter of the ear speculum head down to 2.3 mm with a length of 4.1 cm. Recent work by Harris et al. developed a portable head employing a two-dimensional gimballed mirror and a f–θ lens to perform a fast 2D scan without using linear stages.18 It has a field-of-view of 12 × 19 mm2 and provides consistent resolutions over the imaging area. Using these compact and handheld devices, it is possible to probe regions which are inaccessible using traditional reflection configurations.

FIG. 5.

Handheld THz probe heads. (a) TeraView 1D scan portable probe head and (b) its inner optical beam path. Reproduced with permission from Parrott et al., J. Biomed. Opt. 16, 106010 (2011).17 Copyright 2011 SPIE. (c) THz otoscope. Reprinted with permission from Ji et al., Biomed. Opt. Express 7, 1201 (2016).105 Copyright 2016 Optical Society of America.

FIG. 5.

Handheld THz probe heads. (a) TeraView 1D scan portable probe head and (b) its inner optical beam path. Reproduced with permission from Parrott et al., J. Biomed. Opt. 16, 106010 (2011).17 Copyright 2011 SPIE. (c) THz otoscope. Reprinted with permission from Ji et al., Biomed. Opt. Express 7, 1201 (2016).105 Copyright 2016 Optical Society of America.

Close modal

Endoscopy is another potential THz biomedical application. Probing inside a human body is obviously more challenging. A probing structure that is highly bendable, compact, easy to focus, and anti-jamming is required. Fiber-coupled PCAs are a potential approach for this purpose. A structure for a miniaturized fiber-coupled emitter and detector was proposed by Ji et al.19 Both the emitter and detector are highly compact in a size of only 2 × 4 × 6 mm3, small enough to be inserted into a human body. Using fibers and waveguides to directly guide the THz waves in a bendable medium is another strategy. This technique typically uses a CW THz source and the fiber or waveguide is optimized to the operating frequency. For example, Lu et al. used plastic fibers to guide both the incident and reflected narrow-bandwidth THz waves in a single channel.55 A long propagation distance and a small bending loss can be achieved. However, due to the weak confinement of the field, it is sensitive to the environment around the fiber, which limits its practical usage. A more robust prototype would be a metal-coated THz waveguide, as proposed by Doradla et al. It completely confines the field inside the waveguide.106 The sample has to be placed closed to the waveguide tip to achieve the optimized resolution. Overall, the current state of the THz endoscopic techniques still have a long way to go before they are ready for practical applications. One important limitation is the imaging ability, which is essential for endoscopy to obtain spatial information. However, all the reported techniques can only detect a single pixel and the demonstrated imaging experiments were done by moving the sample. Other technical issues should also be further addressed, such as the robustness of the alignment against different measured distances and complicated sample surfaces, and the consideration of the diffraction limit.

Although a handheld geometry provides opportunities to measure areas that are difficult to make contact with a bulk reflection system, a small window is still required on the probe head to align the THz beam to the measured region. The contact with the window brings an issue of occlusion for in vivo measurements, which refers to water being accumulated on the skin surface due to the prevention of water evaporation by the window.108 The occlusion rate is highest at the start of the contact but also last for a long time and its influence on the THz reflectivity can be observed up to 30 min.84 This can be clearly seen by imaging a region of healthy skin and observing the varied reflectivity across the image.108 The reason is that the signals at different pixels are obtained at different times with different occlusion states, and this time-induced spatial variation affects the contrast of the tissue itself. In addition, the pressure exerted on the window can also affect the THz response.109,110 To overcome these issues, a noncontact reflection setup with a high degree of flexibility to adapt with the nonflat surface is needed. A robotic geometry is a possible solution.

Sung et al. demonstrated an example of quasi-robotic THz imaging of the cornea ex vivo and in vivo.58,111 They used two motorized mirrors to change the illuminating position on the parabolic mirror, thus controlling the THz beam to focus it onto different positions of a spherical surface at a normal incidence. Based on this setup, they scanned ex vivo porcine eyes and in vivo human eyes,58 with each row scan completed within 1 s and each image obtained within 20 s. Recently, the intraocular pressure (IOP) for ex vivo samples was also measured by utilizing multiple stages to control the focus on a spherical surface.112 These techniques, although successful, are limited to spherical samples of a particular size. Using a multi-dimensional robotic arm can provide a more general solution.

Nowadays, the use of robotic-arm assisted systems is becoming more common in many different fields and the THz field is not an exception. THz spectroscopy has taken the advantage of robotic systems in the quality control of paint layers for the automotive industry,113,114 which has now been introduced to various THz applications. Stübling et al. presented a robotic system able to position the THz probes perpendicular to the curved surface of a sample.115 The technique is based on the fringe patterns projected over the sample surface which is recorded by a camera to analyze the shape of an arbitrary object. With this novel technology, the same research group has been able to examine wood infested by insects and a collection of ancient pottery.116,117 They have also demonstrated an exciting application to analyze human remains,107 as shown in Figs. 6(a)–6(c).

FIG. 6.

Robotic configuration: (a) robotic arm integrated with the THz sensor and optical 3D scanner; (b) photo of the THz optical sensor and the beam path; and (c) optical 3D scan of the investigated mummy hand. The THz measurement area is highlighted in red. Reproduced with permission from Stübling et al., Sci. Rep. 9, 3390 (2019).107 Copyright 2019 Authors, licensed under a Creative Commons Attribution (CC BY) license.

FIG. 6.

Robotic configuration: (a) robotic arm integrated with the THz sensor and optical 3D scanner; (b) photo of the THz optical sensor and the beam path; and (c) optical 3D scan of the investigated mummy hand. The THz measurement area is highlighted in red. Reproduced with permission from Stübling et al., Sci. Rep. 9, 3390 (2019).107 Copyright 2019 Authors, licensed under a Creative Commons Attribution (CC BY) license.

Close modal

In the study, they showed that the resolution of THz images is comparable with that of more mature techniques, such as CT and micro-CT scans. However, since the surface reconstruction is computational-based thus time-consuming, it is only effective for still objects. For living samples, the need for speed in data acquisition and robot positioning is mandatory due involuntary movements of the living being. Novel methods in surface profile acquisition have to be developed. Fortunately, with the invention of 3D vision systems, the topology of a sample can be reconstructed even at rates of 20 frames per seconds; this has proved to be helpful for capsule endoscopy,118,119 for instance. We envisage that the inclusion of 3D stereo cameras on robotic-based THz systems for real-time noncontact examination will open new opportunities for THz in vivo biological imaging, which may serve as a complementary tool for clinical diagnosis.

Typical THz systems utilize components that are sufficiently large to limit diffraction but are still small with respect to the wavelength. Beam propagation in quasi-optical THz systems can be modeled via Gaussian beam theory or physical optics theory.73 One major difference between THz and optical waves is that the scattering, especially Rayleigh scattering can be ignored in most cases for THz waves due to the long wavelength.120 The inner microscopic structure of a sample is also usually much smaller than the THz wavelength, thus very little diffuse reflection will occur. Therefore, most materials are “mirror-like” in the THz regime and mainly give specular reflections. This simplifies the sample preparation, theoretical analysis, and characterization. The THz–matter interactions can be conveniently described by Fresnel's equations to express the reflection or transmission as a function of the sample properties.79 As a result, the sample properties can be extracted by solving the inverse problem of the transfer function.77 Biosamples also follow these characteristics, while in some tissue studies there are wavelength-comparable features that may cause Mie scattering and require additional analysis.121,122 In general, the THz response of most homogeneous samples is solely determined by its dielectric function. Solid-state small biomolecules in the crystalline arrangement provide unique absorption features in the THz spectrum. Most of the other dehydrated samples are featureless due to the large number of vibration modes. Hydrated or liquid samples have their THz response mainly dominated by the water concentration, which can be interpreted by the dielectric model of water or the effective medium theories (EMTs). In this section, we will first overview biomolecules that exhibit spectral features in the THz range. Next, the water dielectric model and the EMTs will be introduced to understand the THz response of hydrated samples. Exploring the dielectric behaviors of different samples provides a way to understand the origin of the THz contrast and what can be explored with THz light, which will be discussed at the end.

Using spectral features is an ideal way for both spectroscopic and imaging to sensitively detect a target. The absorption fingerprints provide unique contrast at specific frequencies, serving as features for label-free sensing. For biomedical research, the observation of THz absorption features caused by the vibrational and rotational transition modes has been found mainly in small biomolecules. The spectral features in nucleobases and nucleosides,3,123 amino acids,4,124 uric acid,125 purine and adenine,126 peptides,127,128 and short-chain polypeptides129 have been reported. Large biomolecules, such as proteins and DNA, have a larger number of vibrational modes that collapse into a smooth spectrum, hence usually do not exhibit clear fingerprints.130 To best observe the spectral features, the sample should usually be dry, solid-state, and in a single or poly-crystalline arrangement. The fact that spectral features are not observed in aqueous samples is not because they are covered by the high absorption of the solvent, instead, they disappear due to the changes in the intermolecular structure. THz waves mainly probe the intermolecular vibrations in a long-range ordering structure. An obvious example is the dramatic changes of the spectra of crystalline and amorphous glucose and lactose samples.131,132Figure 7 clearly shows the variation of the lactose absorbance at different crystallinity.132 More examples of spectral-feature dependency on the molecular structure can be seen in Sec. V A 1. The restriction to small and crystalline biomolecules significantly limits the practical detection based on the spectral features. Most biosamples contain macromolecules of proteins and DNA, and could contain a large water content; thus, they do not show any characteristic absorption.

FIG. 7.

Lactose absorbance in 0.1–1.0 THz at different crystallinity. Reproduced with permission from Warnecke et al., Vib. Spectrosc. 102, 39–46 (2019).132 Copyright 2019 Elsevier.

FIG. 7.

Lactose absorbance in 0.1–1.0 THz at different crystallinity. Reproduced with permission from Warnecke et al., Vib. Spectrosc. 102, 39–46 (2019).132 Copyright 2019 Elsevier.

Close modal

1. Water dielectric model

Water is the core substance for all living things on Earth. It undoubtedly plays a key role in biomedical research, especially for THz biophotonics due to the uniquely high absorption by water in the THz regime. Therefore, it is essential to model the dielectric function of water to study many water-rich biosamples. From static field to microwave (below 100 GHz), water can be precisely described by a broad Debye relaxation centered at about 20 GHz, originating from the collective rearrangement dynamics of the water hydrogen-bond network. The end of the Debye relaxation reaches the THz regime, where it fails to successfully fit data above 0.1 THz. Historically, there is disagreement on how to fit the high-frequency response of water, but adding a secondary Debye relaxation term has been widely adapted.133 This gives the so-called double-Debye model which can describe the water dielectric behavior up to 2 THz. Above that, an additional damped harmonic oscillation mode arising from the intermolecular stretching vibration at around 5 THz should be added to describe the dielectric function up to 10 THz.1 Therefore, the selection of the dielectric model depends on the frequency range of a measurement.

In practical biomedical applications, the double-Debye model is more frequently used. This is partially because most THz biomedical experiments, especially tissue studies, are carried below 2 THz to obtain a better SNR. Another reason is that to fit the oscillation mode centered at around 5 THz accurately, a much broader bandwidth (e.g., >4 THz) is needed. According to the double-Debye model, the permittivity can be expressed as134,135

ε̃(ω)=ε+εsε11+iωτ1+ε1ε1+iωτ2,
(1)

where εs is the static permittivity and ε is the permittivity in the high frequency limit. Their combinations with ε1 decide the strength of the two relaxations. τ1 and τ2 are the slow and fast relaxation time, respectively. The double-Debye model has been shown to be applicable to the dielectric behavior of water-rich biosamples. Pickwell et al. pioneered the application of the double-Debye model to describe the THz properties of skin136,137 and basal cell carcinoma.138 A good match was found, and the use of the double-Debye model enables the simulation of the THz response using the model parameters. Later research by other groups has further employed the model for burns injuries,139,140 breast tissues,141 brain gliomas,142 and blood.143 High correlations between some double-Debye parameters and the healthy status of the tissue were reported, indicating the potential usage of these parameters for cancer identification and classification.144,145

Since the two Debye relaxations originated from the hydrogen-bond rearrangement of water molecules, the application range of the double-Debye model is theoretically limited to samples with a high water-concentration. The permittivity of nonwater substances, such as fat, does not follow double-Debye behavior; thus, the presence of significant concentrations of these in samples would reduce the fitting credibility. In that case, EMTs, which consider the material as a homogeneous mixture of multiple compositions and express the effective permittivity as a function of the volume fraction of each composition, can be better applied.

2. Effective medium theories

EMTs were proposed over a century ago to determine the effective dielectric property of a medium consisting of more than one inclusion.146 The fundamental requirement for the validation of an EMT is that the included composition should have a size far smaller than the wavelength and be homogeneously distributed, hence inducing no scattering. Based on different assumptions and derivations, there are various EMTs with slightly different areas of applications. A comprehensive overview and comparison of these EMTs in the THz regime can be found in Ref. 147. Among them, the Maxwell-Garnett (MG), Bruggeman (BM), and Landau–Lifshitz–Looyenga (LLL) models are the three most commonly used EMTs for THz applications. However, the MG model requires the guest medium embedded in the host medium to have a low volume fraction. Mathematically, this is represented by its asymmetric characteristics which mean that exchanging the permittivities and volume fractions of the two inclusions result in a different effective permittivity. For biomedical samples of which the volume fraction of water can vary in a large range, the MG model is not suitable. In contrast, BM and LLL allow arbitrary volume fractions to be included. The BM and LLL models for a two-component composite can be represented by the following equations, respectively:

f1ε1εeffε1+2εeff+f2ε2εeffε2+2εeff=0,
(2)
εeff3=f1ε13+f2ε23,
(3)

where, f1 (ε1) and f2 (ε2) are the volume fractions (permittivities) of components 1 and 2, respectively. f1 and f2 satisfy f1+f2=1. εeff is the effective permittivity of the composite. Both the BM and LLL models are symmetric, as exchanging the permittivities and volume fractions of the two components gives the same result. The major difference between them is the shape of the guest medium assumed in the theoretical derivations. The BM model assumes spherical particles while the LLL model allows arbitrary shapes. In principle, LLL has a broader range of applications. However, it is usually difficult to define the shape of the inclusions for biosamples. For example, tissues contain structures from mesoscopic vasculature to microscopic cells, to nanoscale DNA and proteins. More importantly, the BM and LLL models show very little difference in the calculated effective permittivities, as shown in Fig. 8, with only a very small deviation from the MG model.148 The experimental results from water-basil mixtures fit well with all the three models. In practical applications, the uncertainty on the permittivities of the biological background (i.e., dehydrated sample) has a much larger impact on the calculation; hence, the difference among these models can be completely ignored. Therefore, instead of considering the adaptability of the model, more attention should be given to the data used for the model.

FIG. 8.

Comparison of different EMTs with the experimental data: (a) real part and (b) imaginary part of the dielectric function calculated from the Maxwell-Garnet, Bruggeman, and Landau–Lifshitz–Looyenga models, respectively. The experimental results are for a water–basil mixture data at 0.4 THz. Reproduced with permission from Hernandez-Cardoso et al., Appl. Opt. 59, D6 (2020).148 Copyright 2020 Optical Society of America.

FIG. 8.

Comparison of different EMTs with the experimental data: (a) real part and (b) imaginary part of the dielectric function calculated from the Maxwell-Garnet, Bruggeman, and Landau–Lifshitz–Looyenga models, respectively. The experimental results are for a water–basil mixture data at 0.4 THz. Reproduced with permission from Hernandez-Cardoso et al., Appl. Opt. 59, D6 (2020).148 Copyright 2020 Optical Society of America.

Close modal

A common usage of EMTs in THz biomedical applications is to represent a sample as a mixture of water and a biological background from the nonwater substances. In this way, the permittivity of the sample can be effectively expressed by the permittivities and volume fractions of the two components.149 Most dehydrated samples have a low refractive index and absorption, being transparent to the THz light. Their large contrast to the high refractive index and absorption of water makes the effective permittivity of a sample highly sensitive and mainly dominated by the water concentration, giving rise to the contrast between different samples or different regions of a tissue. In practical applications, the properties of the biological background and water are usually assumed to be known a priori. Fitting the EMT to the permittivity of a measured sample returns the water concentration.150 From another aspect, with the knowledge on the properties of water and the biological background, the frequency-dependent permittivity of a sample is reduced to a single unknown variable of the water concentration. In characterizations, this greatly reduces the number of unknown parameters, making it possible to characterize some complicated structures. For instance, such an approach has been employed for characterizing skin in vivo which contains multiple layers or depth-dependent water concentrations.151,152 The limitation of the EMTs is the difficulty in obtaining an accurate biological background. Ideally it should be obtained by measuring the completely dehydrated sample being investigated. However, in many cases this is unavailable especially for in vivo applications. A typical solution is to adopt the values from the literature, but the sources are very limited and could have large errors compared to those of the investigated sample. For example, the biological background of human skin is usually estimated using that of dehydrated porcine skin.70,151 In our recent review, we show a large variation in the dehydrated properties from different studies, which could lead to a large difference in the extracted water concentration.9 It should also be noted that applying EMTs is not the only way to extract the water concentration. Nuclear magnetic resonance (NMR) is another approach to evaluate the water content,153 which has been used to verify the origin of contrast for THz images.154 For ex vivo samples where dehydration is possible, the water concentration can be directly calculated using gravimetric measurements.

We have given an overview above of small biomolecules which exhibit spectral features and hydrated samples of which the dielectric properties are mainly dominated by the water concentration. These provide the two major origins of contrast: the unique “colors” provided by the spectral features, and the “darkness” provided by the water content. There is also a third origin of contrast from the biological background itself. Although most dehydrated samples are transparent to THz light, they still exhibit slightly different dielectric properties. He et al. measured dehydrated fat, muscle and skin samples.149 Their refractive indices vary from 1.5 to 1.8, and the absorption of dehydrated fat could be four-times smaller than that of dehydrated muscle and skin. The experiment clearly shows the contrast from the biological background, which is very important in analyzing its contribution in the contrast between diseased and healthy tissues. The investigations on liver cancer and liver cirrhosis,66,155 human breast tumor,156 colon cancer,157 and brain tissues158,159 have all found differences in the THz dielectric properties between healthy and diseased areas in the dehydrated state. Although most studies have found that water dominates the contrast, in some tissues the biological background can offer a larger difference. In the study of liver cirrhosis, 50%–66% of the absorption difference is attributed to the inner structural differences between diseased and healthy tissues, which means the contribution of water is less than a half.155 The origin of the biological background contrast could be complicated. Researchers have considered the contrast coming from the differences in the lymphatic systems, vasculature and cell density.60,156,159 Other subwavelength structural changes or composition variations, such as protein concentration,122 could also lead to a difference. Further, Stantchev et al.89 found that the permittivity of bovine articular cartilage is polarization dependent with variation on a subwavelength scale in line with the alignment of the collagen fibrils. The contrast in the biological background can be investigated by in vitro samples, while more studies are needed to explore how different factors affect the THz dielectric properties. Tissue morphology may also affect the THz response, for example in the cornea where the tissue layer results in lossy longitudinal modes that serve as the origin of contrast. Similarly, the layered structure of skin will also affect the frequency-dependent THz reflectivity.

In most practical applications, THz spectra are featureless with the contrast mainly contributed by the latter two factors, that is, the water concentration and biological background. The differences between two compared targets, such as the cancerous and normal areas, is potentially small. We will introduce several techniques and methods to enhance the contrast, such as by using biosensors (Sec. V B), by designing a sensitive multiple-layer structure (Sec. V C 2), or by using contrast enhancing agents, such as nanoparticles (Sec. V C 3).

Currently, the safety power levels set by the International Commission on Nonionizing Radiation Protection for frequencies from 2 to 300 GHz is 1 mW/cm2 for 6 min of exposure.160 For frequencies above this range, there are currently no safety limits until one reaches IR-laser safety specifications set by the international commission. However, The American National Standards Institute sets the safety exposure for IR lasers with wavelengths from 1.4 μm to 1 mm (214 to 0.3 THz) to be at 0.1 W/cm2 for 10 s or more.161 They, however, only consider thermally induced damage whereas THz effects on biomatter can be split into two broad categories of thermal and nonthermal effects. The thermal effects are usually of concern to CW sources due to the strong absorption of THz waves by water162,163 and because pulsed sources usually have low powers per unit time but high peak field strength values. The nonthermal effects of THz radiation are very unique from other frequency regimes because, while the low photon energies are not strong enough break chemical bonds, linear or nonlinear resonances in DNA can occur.164,165 These nonthermal effects were first discussed by Fröhlich in the 1970s and 1980s166 with the idea that THz radiation forms coherent states in biomatter, often called Fröhlich condensates. He considered a system of quantum oscillators surrounded by a thermal reservoir and he showed that this system can oscillate as a single coherent mode, somewhat like a Bose–Einstein condensate, instead of as a set of free oscillators each with a particular resonance frequency. He claimed that such effects will perturb the biochemical kinetics of biological objects, hence the need to consider the entire system as a whole rather than focusing only on single parts.

Some of the earliest investigations into whether THz radiation can damage biological function was the project named “THz-Bridge” funded by the Europe with the project ending in 2004.167 It showed that under low power conditions, THz radiation did not induce any adverse effects in most experimental conditions; however, there was a study where both genotoxic and epigenetic effects were induced in lymphocytes following exposure to CW 100 GHz radiation of 0.05 mW/cm2. This was at intensities much lower than the current safety levels, and the cells were without the protection of the human body. The project concluded that more measurements under different exposure conditions, such as CW and pulsed sources with different powers, are needed before THz radiation can be safely adopted for commercial purposes under standardized protocols. More recent studies have found further biological effects of THz radiation with previous studies162,168–170 being reviews dedicated to the biological effects of THz radiation.

The bioeffects of THz radiation are unique. A series of studies by Fedorov171–173 found that exposure to THz radiation (1 ps pulses, 0.1–2.2 THz, 76 MHz repetition rate, 8.5 mW of power) would shorten the lifespans of drosophila flies. This suggests a response that affects the entire system that cannot be explained by thermal effects only. There is also a 2013 study by Titova et al. showing the activation of DNA damage response in artificial skin cells due to exposure to pulsed THz radiation (1 kHz pulse rate, 1 μJ per pulse, 0.1–2 THz),174 and the cell response was different when exposed to pulsed UV light. The mechanism that activated this DNA damage response was not known at the time of writing. In 2020, Tachizaki et al.175 showed that THz pulses (1 kHz rep-rate, 0.5 MV/cm, 0.2–1.1 THz) caused the regulation of genes related to zinc–finger proteins in stem cells indicating that the concentration of Zn2+ ions was affected by the intense THz pulses. Most recently in 2021, the Hegmann group showed that THz pulses (1 kHz rep-rate, 0.03–2.4 μJ energy per pulse, peak electric fields of 27–240 kVcm1) incident on artificially cultivated human skin cells inhibited DNA signaling pathways associated with some types of cancers,176 offering potential for cancer therapy. An earlier study by Gun-Sik Park's group at Seoul National University also found that THz pulses (1 kHz rep-rate, 0.26 nJ/pulse, 0.3–2.5 THz) induce a genetic response in mouse skin similar to that of a puncture wound and that the genome analysis was very different to that caused by UV radiation, thermal damage and neuron radiation exposure.177 Additionally, they observed that exposure to this type of THz radiation resulted in the wound taking longer to heal. There were also the observations of Ramundo-Orlando et al. that permeability of cell membranes was affected by pulsed THz radiation.178 Note that not all studies observed effects due to THz pulses. Angeluts et al.179 found no evidence to suggest that THz radiation (1 kHz rep-rate, 550 pJ/pulse, 0.6–6 THz) damages the DNA of human blood leukocytes.

The references in the above paragraph all used pulsed THz radiation and in such sources thermal effects can be ignored because the temperature increases are below 0.1 K, which is not enough to trigger a thermally induced stress response in cells. However, CW sources are capable of inducing heating effects although it should be noted that low enough CW powers can still induce a response without triggering any thermal response in the biological system. In one study, Franchini et al.180 looked at how human fibroblasts respond to pulsed and CW THz radiation. They used a compact free-electron laser to generate pulses between 100 and 150 GHz with time length 50 ps and their CW source generated a single frequency at 25 GHz. They found that THz exposure for both pulsed and CW sources do not induce any thermal effects (due to low CW powers); however, they did observe aneuploidy in their samples for both types of radiation with the speculation that different mechanisms are involved. Increasing the CW power levels was done by. in the study by Echchgadda et al.181 where they exposed jurkat cells to a 2.52 THz CW source with 636 mWcm2 power and observed a 6 °C temperature increase; therefore, they had to compare their results to jurkat cells that were heated to the same temperature for the same duration. The heated cells do show significant changes in their genome expressions; however, the THz exposed cells showed the activation of very different signaling pathways. This indicates that while CW THz radiation can heat the sample, its effects are not simply thermal. The same group has also shown that the majority of jurkat cells die after 20 minutes of exposure to their CW 2.52 THz source with 227 mWcm2 power.182 The final study we review from this group is Ref. 183, where they exposed human keratinocytes (skin cells) for 20 min to a CW source with different frequencies (1.4, 2.52, and 3.11 THz) keeping the power density at 44.2 mWcm2 for each frequency. This low power density was not enough to induce a thermal response. Their results showed that each frequency triggered specific metabolic or signaling pathways that were not triggered by the other frequencies. This indicates that a careful study needs to performed to find the safe exposure levels at each frequency.

Kiseliov et al. studied the potential of THz radiation to suppress tumor growth (Guerin carcinoma) in rats and compared it to x-ray exposure.184,185 Their source was a CW hydrogen cyanide gas laser operating at 0.89 THz with power densities of 1600 μWcm2, and they exposed the tumors for 10 min once every three days. They found that THz radiation and x rays both slow down tumor growth compared to a control group. However, the mechanisms are very different as the x rays kill all cells whereas the THz wave induced an inflammatory response where the immune system targeted the cancer cells. In the same study, they also exposed tumors for 10 min once every day with a power density of 400 μWcm2. In this scenario, they found that tumor growth actually accelerated compared to their control. This study shows the great effort required to determine and develop any potential therapeutic uses of THz radiation. A study that makes a further relevant point is that performed by Bock et al.186 where they exposed mouse stem cells to pulsed THz radiation (1 kHz rep-rate, 1–20 THz frequency range, 1 μJ per pulse) from air-plasma. They found that THz irradiation accelerates cell differentiation by activating a specific transcription factor (peroxisome proliferator-activated receptor gamma) with molecular dynamics computer simulations coinciding with the gene specific response they observed. Their work indicates that THz radiation could be used for cellular reprogramming.186 

Mizuno et al. performed a series of in vivo measurements investigating the effects of THz radiation on corneal tissues.187,188 They irradiated the corneal tissue of each rabbit with a 162 GHz CW source with powers ranging from 60 to 600 mW cm−2 for 6 min. The corneal thickness, opacity, and temperature were measured and they also checked for corneal epithelium damage. They observed some thinning of the cornea and the development of corneal opacity following the measurement. However, they also monitored the corneas for nine days following the measurements and found that some of these effects recovered with increasing time following the exposure. The study classified the severity of the effects according to the power level of the exposure. They observed no damage to the corneal tissues at 60 and 120 mW cm−2 in any of their corneal assessment techniques. They also measured changes in the THz reflectivity following the exposure and propose that some of the damage observed could be linked to dehydration effects caused by the increase in temperature attributed to the high power exposures.

Currently, there is no consensus in regard to what are safe THz power levels partly because a lot of studies have not observed any effects due to THz exposure (see Refs. 179,189, and 190) while simultaneously others have observed effects (see references in the above paragraphs). It should be noted that most studies use different methods of evaluating biological damage and their exposure conditions are also different. Further, the exact mechanism via which THz radiation affects living organisms is unclear. The so-called Fröhlich condensate is the most accepted theoretical mechanism; however, it was first experimentally observed in 2015 by Lundholm et al.191 in a lysozyme protein. They shone a CW 0.4 THz light on a lysozyme protein and used the x-ray crystallographic method to obtain the vibrational modes of the protein structure. They observed changes to the atomic displacement of a long helical part of the protein, and that the overall atomic positions were more ordered (less displacement) with the THz light on. The most likely explanation of their results is a Fröhlich condensate. This, unfortunately, is only a single-step toward understanding the physical mechanisms via which THz radiation affects entire biological systems which consist of many proteins and molecules, and highlights the research left to be undertaken.

THz biomedical applications can be found for molecules in a nanometer level up to tissues at a centimeter level. Different approaches for sample preparation, measurements, and data analysis have been used depending on the different characteristics of the samples. Earlier studies gave more attention to investigating the fundamental THz dielectric properties of biosamples. This was followed by numerous studies targeting sensitive detection. Recently, increasing numbers of measurements and images of macroscopic tissues have been performed, especially relating to in vivo applications, showing the trend of THz biophotonics toward more practical applications. In this section, we present an overview of biomedical applications using THz technologies. In particular, we first introduce the fundamental earlier studies of biomolecules including their solid phase and aqueous solutions. Then, different sensing techniques proposed for sensitive detection of various biosamples will be presented. Another large branch of applications, covering tissues investigated either in vitro or in vivo, will be comprehensively reviewed. We will show the progress that has been made in different areas to date and discuss their trends of development.

THz biomolecular spectroscopy is one of the earliest THz biomedical applications, which can be classified into a field of THz-biochemistry. It starts from the far-infrared studies using Fourier transform infrared spectroscopy (FTIR) from the 1960s.192 The invention of THz-TDS filled the lower spectral range of 0.1–4 THz, providing a better SNR and accuracy in this region. The motivation of THz spectroscopy of biomolecules comes from the fact that THz waves can sensitively probe the phonon vibrations and intermolecular interactions, such as the hydrogen-bonding activities and collective vibrations, which play critical roles in many biological processes, such as protein folding and conformational changes or DNA transcription. This is in contrast to the infrared region which mainly probes the intramolecular vibrations and rotations. The sensitivity to the long-range arrangement makes the molecular THz response highly dependent on the ordering structure and the solvation environment. This mechanism determines the characteristics mentioned earlier that spectral features of sharp absorption peaks are only observed in small biomolecules in their solid crystalline form. For macromolecules, THz waves probe a large number of collective modes that give rise to a smooth spectral absorption.130,193 For amorphous samples, the high level of spatial disorder results in complicated intermolecular interactions from the random molecular orientations, giving rise to inhomogeneous broadening of the features.131,132 Long-range order is also absent in aqueous solutions. In this case, the solvated molecules interact with the surrounding water molecules, forming a hydration shell with properties different from that of bulk water,194 thus offering a chance to be probed by THz spectroscopy. Therefore, in terms of applications, THz molecular spectroscopy can be classified into three categories: investigating the molecular dynamics in solid-phase samples, investigating the molecular interaction with water, and improving the sensitivity of molecular detection.

1. Solid-phase samples

Most investigated solid-state biomolecular samples for THz spectroscopy are usually poly- or single-crystalline small molecules which provide highly ordered lattice structures. The connected intermolecular vibrations controlled by noncovalent forces, such as H-bonds, give rise to absorption at some specific frequencies in the THz range, resulting in sharp peaks as the spectral features. As a result, THz light is highly sensitive to the molecular arrangement. For example, isomers, which may present similar infrared spectral features due to the similar intramolecular vibrations, can be distinguished from the low-frequency intermolecular modes in THz spectroscopy.125 For the measurement, polycrystalline samples are usually prepared by mixing the investigated biomolecules with THz-transparent polymer powders, such as polyethylene and compressing the mixture into pellets. This method provides flexibility in the fraction ratios between the two compounds and arbitrary thicknesses, such that a long sample–THz interaction can be achieved. THz-TDS is commonly used, which can probe the low-frequency modes down to subterahertz and enables the simultaneous extraction of both the refractive index and the absorption coefficient. FTIR was mainly used in some earlier works, providing a broader bandwidth up to mid-infrared. In both systems, a focused THz beam directly passes through the pellet sample in a transmission geometry. The sample thickness is typically maximized to allow a sufficiently long THz–sample interaction distance, while attention should be given to ensuring the sample attenuation does not exceed the dynamic range of the system, especially at the peak regions. Fabry–Pérot oscillations in the spectrum due to the beam bouncing within the sample may occur. These features should be distinguished from the sample spectral features. Another way to prepare polycrystalline samples is by drop-casting or spin-coating solutions on a substrate to make planar ordering polycrystalline thin films. Compared to polycrystalline pellets which are randomly orientated in the 3D directions, the crystal orientations of the film are only random in the 2D planar direction, significantly reducing the inhomogeneous broadening of the spectral features. The narrower feature absorption makes it possible to separate some close vibrational modes. However, it is difficult to measure such thin films in the traditional transmission geometry due to the low sensitivity. Instead, waveguides have shown a very promising sensitivity for broadband characterization of such thin films,195,196 which will be detailed in Sec. V B 2.

The above measurement protocol for pellets has been widely adapted to investigate many important biomolecules. The observation of spectral features is usually limited to small biomolecules when the number of oscillation modes is small. Large biomolecules are usually featureless in the THz band. This is the result of the increased number of vibrational modes due to the large number of degrees of freedom, which causes the spectrum to collapse into a featureless spectrum. For example, spectral features are found in nucleoside and nucleobases,3 but not in DNA or RNA,197,198 in amino acids and short-chain peptides,4,129,199 smoothed in tetrameric peptides,127 and not seen in proteins.200,201 Other small therapeutic biomolecules also exhibit THz spectral features due to collective vibrations, such as various types of saccharides,131 uric acid,125 purine and adenine,126 and many pharmaceutical drugs.50 Due to the high sensitivity to the intermolecular network, a promising application is using THz spectroscopy to monitor polymorphic conversion in situ. Figure 9 from Ref. 202 shows the solid–solid conversion of carbamazepine form III to I at isothermal conditions (433 K), recorded at 5-min intervals. In such an application, THz spectroscopy demonstrates its advantages of low photon-energy (no heating effect), fast acquisition, and noninvasive measurement.

FIG. 9.

Solid–solid conversion of carbamazepine form III (black) to I (light gray) at isothermal conditions (433 K), recorded at a 5-min interval. Reproduced with permission from Zeitler et al., Thermochim. Acta 436, 71–77 (2005).202 Copyright 2005 Elsevier.

FIG. 9.

Solid–solid conversion of carbamazepine form III (black) to I (light gray) at isothermal conditions (433 K), recorded at a 5-min interval. Reproduced with permission from Zeitler et al., Thermochim. Acta 436, 71–77 (2005).202 Copyright 2005 Elsevier.

Close modal

Another common feature observed for the small biomolecules is the sharpened and enhanced absorption peaks at low temperatures. This sometimes enables two peaks to be resolved at low temperatures, but they are collapsed into one at room temperature.3,126 The positions of the feature peaks are also temperature sensitive, while the relationships with temperatures vary with different peaks and at different temperatures, due to the different mechanisms dominating the shift.131 To understand the origins of the observed features, researchers usually apply various molecular modeling calculations or simulation methods to analyze the experimental results, such as normal mode calculations,200,203 density functional theory calculations,3,123 and molecular dynamics simulations.128,204 For macromolecules which present no spectral features, THz spectroscopy is still sensitively responsive to their molecular structures. In this case, the detection is based on a featureless dielectric difference. For example, applications have shown that THz dielectric spectroscopy can be used to probe protein conformation and mutation,200 and detect the binding state of DNA.198 However, the absence of a specific fingerprint reduces the sensitivity and the contrast to other dielectric substances.

2. Interaction with water

Although solid-phase samples may exhibit spectral features in the THz range, the most ideal approach for studying biomolecules is to investigate them in a condition mimicking their in vivo biological environment. This requires an aqueous phase, with a proper pH value, temperature, and other requirements on the inclusions. Among these conditions, it is critically important that the mechanism and dynamics of biomolecules interacting with water is explored, which essentially determines many biological structures and functions, especially for proteins. An important branch of THz biomolecular spectroscopy is the study of solvated biomolecules, usually by means of THz absorption or dielectric spectroscopy. The most obvious obstacle, however, is the high THz absorption by water that strongly limits the light–matter interaction distance, hence the accuracy. Havenith's group has made a great contribution to this area. They probed the fast relaxation dynamics of water solutions using a Germanium laser operating at 2.3–2.9 THz with a high output power up to 1 W. This enables the measurement of a large sample thickness to provide a high sensitivity. Their early work in 2006 measured the absorption of lactose solution as a function of the lactose concentrations.205 The first observation is the concentration-dependent absorption from the measurement does not follow the weighted sum of the two components (i.e., lactose and water). Instead, a three-component model including the solute, bulk water (i.e., free water), and hydration water (i.e., solvation water or bound water) shows a better agreement. Further analysis reveals a hydration shell of 5.13 × 0.24 Å around the lactose, corresponding to 123 hydration water molecules around each lactose molecule, which have reached beyond the first hydration layer. These observations are based on the sensitivity of THz to the rearrangement of hydrogen-bonds in water, which is affected by the solute. Their following work in 2007 studied water–protein interactions,206 which is important for understanding how water affects protein structures and functions. A nonmonotonic trend of the THz absorption against the protein concentration is found. This relationship is explained by a hydration shell of 20 Å around each protein molecule and the overlapping of the shells at high concentrations. The observed hydration shell thickness is much larger than that observed by scattering methods, potentially due to the different definitions of hydrated water. The analysis of the hydrogen-bond correlation function shows increasing hydrogen-bond lifetime with decreasing distance to the protein, showing the distance-dependent water properties in the THz regime. A further step forward is the investigations of the group into the influence of protein mutations and the pH value,207 protein structural flexibility,194 and the thermal denaturation effect on the solvation dynamics.208 They have also investigated the protein–solvent interaction upon the folding process (which happens on a millisecond time scale), by using a kinetic THz absorption (KITA) setup to initiate the refolding reaction relative to the probing time of the THz pulse.209 Details about these investigations can also be found in Refs. 210 and 211.

The concept of a long-range hydration shell with distinct THz properties brings a new perspective in which to view the protein hydration dynamics. This theory has been supported by other groups. Tamminen et al. studied the free and bound water in gelatin hydrogels with different total-water contents in the frequency range of 220–330 GHz. They verified the constant nonfreezing bound-water mass with a distinct property compared to pure water.212 Arikawa et al. used THz-TDS to measure disaccharide-water solutions.213 The use of the ATR geometry circumvents the high attenuation problem of water in a transmission measurement; thus, a broad bandwidth of 0.5–2.6 THz can be investigated. They used the three-component model with local field correlation to extract the properties of the hydration water. Bye et al. used an intense THz synchrotron source to study different concentrations of bovine serum albumin (BSA) solutions in 0.3–3.3 THz.214 An observed nonlinear concentration dependency was explained by the presence of a 15 Å hydration shell around the protein. Sushko et al. compared proteins of different molecular weights with different concentrations using a quasi-optical VNA in the WR-3 waveguide band (0.22–0.325 THz). They also found a similar thickness for the hydration shell of 16–25 Å.215 A larger range of molecular weights from 0.075 to 250 kDa, covering amino acids to proteins solvated in water is reported by Grognot and Gollot.216 They applied a broadband THz-TDS in a Si-based ATR configuration to improve the sensitivity for absorptive liquids. Very different concentration dependencies for different sizes of molecules were found, as can be seen from the relative refractive index and extinction coefficient in Fig. 10. For small biomolecules, the hydration shell is relatively large compared to the molecule. The property of the hydration water in the shell makes a great contribution to the solution, hence the refractive index can increase and the extinction coefficient drops slowly with the increasing concentration. In contrast, the hydration shell is relatively small compared to large molecules. Therefore, the effective properties of the solution agree well with the theory that the biomolecules act as empty and transparent cavities to replace the water molecules.210 Reduced refractive index and extinction coefficient were observed with the increasing concentration. The solute is not limited to proteins and their components. Phospholipid bilayers, which is a main component for biomembranes, also showed a long-range hydration shell up to four to five layers, as reported by Hishida and Tanaka.217 All these studies demonstrate the unique role of THz spectroscopy in exploring the biomolecule–water interactions. The thickness of the hydration shell revealed by THz spectroscopy is larger than that observed by inelastic neutron scattering or nuclear magnetic resonance. This is because of the different definitions of the hydration shell caused by the different relaxation dynamics probed by these techniques. Inelastic neutron scattering or nuclear magnetic resonance probe the slower dynamics of 1091011 s, corresponding to the water molecules immediately adjacent to the solute, while THz spectroscopy measures dynamics at 1013 s such that the slightly perturbed water molecules at long distances are included.215,217

FIG. 10.

Concentration-dependent relative (a) refractive index and (b) extinction coefficient of biomolecules with different molecular weights, including glycine (0.075 kDa), serine (0.105 kDa), gluthatione (0.307 kDa), peptide (0.458 kDa), albumin (66 kDa), and catalase (250 kDa). Reproduced with permission from Grognot et al., J. Phys. Chem. B 121, 9508–9512 (2017).216 Copyright 2017 American Chemical Society.

FIG. 10.

Concentration-dependent relative (a) refractive index and (b) extinction coefficient of biomolecules with different molecular weights, including glycine (0.075 kDa), serine (0.105 kDa), gluthatione (0.307 kDa), peptide (0.458 kDa), albumin (66 kDa), and catalase (250 kDa). Reproduced with permission from Grognot et al., J. Phys. Chem. B 121, 9508–9512 (2017).216 Copyright 2017 American Chemical Society.

Close modal

Given that spectral features are rarely available in most biosamples, differentiating the relatively small dielectric differences requires high accuracy. In addition, the volume of most biosamples is small, the THz light can only interact with a thin film, and the measurement sensitivity is usually low. An important branch of THz biomedical research is to improve the sensitivity of biosensors with the aim to induce larger changes in the THz response with only trace amounts of samples. For this purpose, it is actually competing with many other sensing techniques, ranging from surface plasmon resonance (SPR), fluorescence to colorimetric methods. One of the biggest challenges of using THz waves for sensing is the long wavelength compared to the sample thickness, which is usually at a molecular level or subwavelength. To overcome this issue, various techniques have been proposed to efficiently enhance the light–matter interaction.218,219 Here, we overview the three most widely applied techniques, including metamaterials (MMs), waveguides, and THz surface plasmon polaritons (SPPs).

1. Metamaterials

MMs are artificial media consisting of subwavelength periodic structures (usually metallic) that have effective electromagnetic properties tailored by their structures. The high flexibility of the structural design makes it possible to realize almost arbitrary responses of the THz waves; thus, MMs are not just limited to THz sensing, and they have also been widely used in THz functional devices for beam control, modulation and polarization manipulation.220,221 This is also the reason that MMs have become one of the hottest THz topics in the past ten years. The fundamental principle of MMs is based on exciting localized plasmonic modes in the metallic structures resonating at a single frequency or at some particular frequencies. The localized resonance gives rise to strongly enhanced and subwavelength-confined fields at the resonant metal-dielectric boundaries, typically at the gap regions. Any tiny changes in the dielectric environment caused by the analyte will significantly change the resonance condition, thus introducing a detectable variation in the resonant frequencies and/or intensities. The tightly focused field strongly enhances the light–matter interaction and requires only trace amounts of samples to produce a measurable change.

For sensing applications, planar MMs, which are also called metasurfaces, are more commonly used for convenient sample preparations. As the measurement is mainly based on the shift or the amplitude change of a resonance, a MM sensor with a sharp resonant peak or dip, mathematically represented by a high quality-factor (Q-factor), is preferred. The performance of a MM sensor can be evaluated by many aspects, sometimes depending on the sample types. For example, the sensitivity (S) is usually defined as the frequency shift per unit refractive index change, which is convenient for films with well-defined bulk THz properties. For dehydrated molecules on a sensor with an inhomogeneous and ultrathin thickness, it is difficult to define the refractive index, and one may need a measure of frequency shift per unit weight per unit area. The full width at half maximum (FWHM), or sometimes the Q-factor, is used to describe how robust the detection is against measurement errors. The figure of merit (FOM) combines the above two quantities by FOM = S/FWHM, providing a more comprehensive evaluation. To compare the ability to detect trace amount of the sample, the sample thickness or the volume will also be considered, such as measuring the volumetric sensitivity as the sensitivity per unit volume. The high flexibility on the MM patterns allows many different structures to optimize these aspects, thus there are a great numbers of studies in this area. In this manuscript, we will overview how the sensitivity, the sample volume, and the Q-factor were improved by different groups for biomedical sensing. Readers may also refer to the topical reviews in Refs. 218 and 222.

A lot of effort has been dedicated to increasing the sensitivity, which can be improved by various approaches. Conventional MMs include split-ring resonators, dipole resonators, or mesh resonator metallic structures fabricated on a bulk substrate. One well-known limitation in the traditional design is the field-confinement in the high-permittivity bulk substrate. Typically, a Si or other semiconductor substrate with a permittivity over 10 and a thickness of hundreds of micrometers is used. The confined field in the substrate weakens the interaction with the analyte, which can be solved by using an ultra-thin or low permittivity substrate.223 Tao et al. compared two sensors based on the same pattern on a thick (500 μm) Si and a thin (1 μm) SiN substrate, respectively. The sensitivity was improved more than an order of magnitude by the use of the thin substrate.224 The same group proposed a MM sensor with the same pattern but on a paper substrate in the next year. The low permittivity and low absorption of paper also enables a strong field-analyte interaction. The use of a paper substrate, enabled by their novel shadow mask patterning fabrication method, has the advantage of the porosity feature, which is favorable for sample absorption. 100 μl glucose solution was deposited on the paper sensor and dried. They showed that glucose concentration down to 3 mmol l−1 can be detected, comparable with the clinical techniques used for blood glucose detection.225 Different detection sensitivities for glucose and urea were also found, which implies specific-target detection. Another approach to reduce the substrate influence is partially removing the substrate at the nonpattern regions. Using a focused photon beam, Chiam et al. fabricated a high aspect ratio MM sensor with 10 μm high split-ring resonators. Doing this leaves more field-enhancing space for the analyte previously occupied by the substrate, this significantly improves the sensitivity.226 One potential drawback for this approach is the relatively large sample volume needed to fill the removed substrate, as the strongest field enhancing plane is located at the metallic layer. Another fabrication method of etching only the gap region of the split-ring resonator may ease this issue.227Figures 11(a) and 11(b) show the top-view and cross-section SEM images of the resonator with an etched trench. The increased sample volume is negligible, and an improvement factor of up to 2.7 times can be achieved. The substrate influence can also be eliminated by adopting a perfect absorber design. A MM absorber usually consists of three layers, namely, the ground metallic plane, the middle dielectric layer, and the MM patterns on top. The metallic ground perfectly isolates the field interaction with the lower substrate. Further, it also provides a high Q factor over 10 (traditional MM is below 10).231 The metallic ground plane can also be an individual flexible substrate, as proposed by Yahiaoui et al.232 Apart from the substrate, the sensitivity can also be improved specifically for a targeted sample. Lee et al. fabricated a nano-slit antenna MM sensor, with the structural dimensions designed to match the MM resonant frequencies to the vibrational frequencies of D-glucose and fructose, respectively.228 By drop-casting the glucose or fructose solution on the sensor, a polycrystalline thin-film was formed which provides sharp vibrational peaks with very high absorption at the resonant frequencies of the sensor. In this way, the sensor provides selectivity to only be highly sensitive to the targeted samples. Figures 11(c) and 11(d) show the measured results of the normalized spectra using the glucose sensor to measure sucrose and D-glucose respectively. The resonant peak of the sensor at 1.4 THz matches the glucose vibrational mode. A much higher sensitivity can be found in the glucose measurement. Recently, novel nanomaterials have also been introduced to improve the sensitivity. By adding a monolayer of graphene on top of a MM absorber,233 or replacing the metallic dipoles with carbon nanotube (CNT) dipoles,234 a higher sensitivity was observed, as reported by Ying's group. They demonstrated the sensitive detection of 100 nM thrombin binding to the aptamers on the graphene sensor and the clear reflectance variation of the CNT sensor by glucose solutions at concentrations from 0 to 100 ng/ml. They attributed the improved sensitivity to the efficient absorption of the analyte by the graphene or CNTs through p–p stacking and van der Waals forces. However, the added analyte introduces negligible modulation to the graphene or the CNT; thus, there is very little contribution from the conductivity variation. In addition, cleaning the surface of graphene or CNTs could be difficult, which may limit their practical usages. Nanoparticles have the surface plasmon resonance usually at visible wavelengths due to the nanometer-scale structure, which cannot be utilized for THz sensing. However, when combined with MM, their larger permittivity will cause changes in the electric field especially at the enhancing regions. The use of nanoparticles may also provide a homogeneous monolayer distribution. Adding gold nanoparticles has been shown to cause a highly efficient improvement on the detection sensitivity for both avidin235 and epidermal growth factor receptor (EGFR).236 

FIG. 11.

THz metamaterial sensors: (a) top-view and (b) cross-section SEM images of the metamaterial pattern with a trench etched at the field-enhancing region. (a) and (b) are reproduced with permission from Meng et al., Opt. Express 27, 23164 (2019).227 Copyright 2019 Authors, licensed under a Creative Commons Attribution (CC BY) license. Normalized THz spectra measured with the glucose antenna metamaterial sensor for (c) sucrose and (d) D-glucose molecules. (c) and (d) are reproduced with permission from Lee et al., Sci. Rep. 5, 15459 (2015).228 Copyright 2015 Authors, licensed under a Creative Commons Attribution (CC BY) license. (e) Schematic of the microfluidic bow-tie metamaterial sensor. The corresponding simulation in (f) shows that the resonant THz field is enhanced at the microfluid regions. (e) and (f) are reproduced with permission from Zhang et al., IEEE Trans. Terahertz Sci. Technol. 9, 209–214 (2019).229 Copyright 2019 IEEE. (g) Symmetric-broken double split ring resonator design. The scale bar is 15 μm. The corresponding (h) measured transmission spectra of the metamaterial sensor. (g) and (h) are reproduced with permission from Yang et al., Opt. Express 25, 15938 (2017).230 Copyright 2017 Optical Society of America.

FIG. 11.

THz metamaterial sensors: (a) top-view and (b) cross-section SEM images of the metamaterial pattern with a trench etched at the field-enhancing region. (a) and (b) are reproduced with permission from Meng et al., Opt. Express 27, 23164 (2019).227 Copyright 2019 Authors, licensed under a Creative Commons Attribution (CC BY) license. Normalized THz spectra measured with the glucose antenna metamaterial sensor for (c) sucrose and (d) D-glucose molecules. (c) and (d) are reproduced with permission from Lee et al., Sci. Rep. 5, 15459 (2015).228 Copyright 2015 Authors, licensed under a Creative Commons Attribution (CC BY) license. (e) Schematic of the microfluidic bow-tie metamaterial sensor. The corresponding simulation in (f) shows that the resonant THz field is enhanced at the microfluid regions. (e) and (f) are reproduced with permission from Zhang et al., IEEE Trans. Terahertz Sci. Technol. 9, 209–214 (2019).229 Copyright 2019 IEEE. (g) Symmetric-broken double split ring resonator design. The scale bar is 15 μm. The corresponding (h) measured transmission spectra of the metamaterial sensor. (g) and (h) are reproduced with permission from Yang et al., Opt. Express 25, 15938 (2017).230 Copyright 2017 Optical Society of America.

Close modal

Improving the sensitivity can usually reduce the amount of sample required for successful sensing. There is also research focused on directly reducing the volume needed. The enhanced field at the resonant locations of traditional MM sensors is confined within few tens of micrometers in height. Given that the frequency resolution limit of THz-TDS is about 2 GHz, the minimum thickness of a sample that can cause a measurable changes is about 100 nm.231,237 Thinner layers require an extremely confined field compared to the wavelength. A solution provided by Park et al. is a single nanoslit antenna.238 In this structure, the analyte in the nanoslit interacts with the strongly confined field; thus, a very small amount of sample is required, while the transmission is very weak. The approach of narrowing the gap is also adopted in another design consisting of annular nanogaps fabricated by atomic layer lithography.239 The device showed the ability to detect a 1-nm-thick layer of Al2O3, six orders smaller than the wavelength. Alternatively, the sample volume can also be minimized by reducing the lateral area. For example, Withayachumnankul et al. attached the MM sensor close to the THz emission crystal to interact with a THz near-field.240 The sample can be as small as 0.2λ. More recently, a spintronic THz emitter has been integrated with MM sensors monolithically by Bai et al.241 Hela cells and pseudomonas prepared on the asymmetric double-split ring resonator MM sensors directly interact with the THz waves emitted from the backside of the substrate, with a lateral spatial resolution down to 500 μm. This resolution can be further improved by reducing the substrate thickness (2 mm in this study). The area can also be reduced by concentrating the analyte to only the field-enhancing regions by guiding the sample using a microfluidic device, as reported by Zhang et al.229 and shown in Figs. 11(e) and 11(f), and they presented the ability to detect BSA solutions down to 0.13 μl.

A larger amplitude change can be induced by the analyte near the resonant frequency if a higher Q-factor can be provided; thus, a better robustness against experimental errors can be obtained. Some research has also been dedicated to improving the Q-factor by sharpening the resonant peak or dip. This is usually realized by breaking the symmetry of the MM pattern. A theoretical study of asymmetric THz MMs has been given by Signh et al.242 This design enables a Q-factor up to 65 in the quadrupole mode of a split-ring resonator with asymmetric double slits223 and up to 100 in an octupolar mode of a double split-ring resonator structure.230Figures 11(g) and 11(h) show the MM unit design and the corresponding experimental transmission spectra. The octupolar mode is at about 1.3 THz.

The great effort given to THz MM sensors has improved their performance from many aspects. Sample loading and measurement are simple with excellent sensitivity. However, the mechanism of localized plasmonic resonance is sensitive to the dielectric loss of the sample. Many aqueous samples, especially water solutions, have extremely high absorption in the THz range. Measuring these absorptive materials introduces significant loss that obviously broadens the resonant peak, resulting in a poor Q-factor, as can be seen in Refs. 229, 243, and 244. Therefore, most of the MM sensing applications have been shown on low-loss samples, such as dehydrated materials. Developments leading to structures specifically designed for absorptive samples are in high demand.

2. Waveguides

Waveguides direct the propagation of a THz beam by confining the field near the guiding structure; thus, they can also be used to sense a small amount of analyte. There are many forms of THz waveguides. Among them, the parallel plate waveguide (PPWG) is one of the most commonly used structures due to a few advantages. PPWGs support the propagation of a TEM mode with almost zero dispersion, similar to free space THz waves, while confining one of the spatial dimensions of the electric field to a subwavelength scale, providing efficient interaction with thin-film samples. The absence of resonance enables a direct characterization of the sample optical properties. A PPWG is simply made of two metallic plates parallel to each other, with a subwavelength air-space (typically of the order of 100 μm) between them. Usually, two Si plano-cylindrical lenses are attached to the edges of the PPWG to couple the free-space THz light into and out of the waveguide, as shown in Fig. 12. In this way, the THz wave is compressed in the z direction to a subwavelength level in order to efficiently interact with any thin-film sample deposited on the waveguide inner surfaces. The near perfect conductivity of metals in the THz regime offers nearly zero dispersion, no cutoff frequency, and low absorption of 0.1 cm−1 in the range of 0.1–4.5 THz.245 The metallic plates can also be replaced by metallic films to provide a high flexibility.246 A PPWG can also be designed in a tapered structure such that the thickness of the gap gradually reduces from the input/output to the middle. In this way, the middle thickness can be reduced down to 8 μm to strongly confine the field without significantly affecting the coupling efficiency between the THz light and the PPWG.

FIG. 12.

Schematic of a typical parallel plate waveguide and its coupling to the free-space THz waves.

FIG. 12.

Schematic of a typical parallel plate waveguide and its coupling to the free-space THz waves.

Close modal

The Grischkowsky group has made great contributions in PPWG sensing. A successful application of thin-film water sensing has been demonstrated by Zhang and Grischkowsky.247 They controlled the temperature and the pressure of a chamber, such that a 20-nm thin film of water was condensed on the inner surfaces of the PPWG inside the chamber. The PPWG enables a long interaction length up to 63.5 mm (i.e., the length of the PPWG) with the water film. Both the refractive index and absorption coefficient can be sensitively characterized. The two major limitations of this method are first, the water film thickness cannot be physically measured and it was determined by equating the refractive index at 0.86 THz to that of bulk water. Such a method cannot be applied to other samples with unknown optical properties. Second, strong water-vapor absorption lines appeared in the characterization results, which are caused by the different vapor pressures in the sample and reference measurements. Another significant advantage of PPWGs is the ability to characterize the samples with spectral features. This is not only because it provides a high sensitivity to thin films. As mentioned earlier, polycrystalline thin films have planar ordering and have one less dimensional disorder compared to pellet samples; thus, they provide sharper feature peaks by reducing the inhomogeneous broadening. Melinger et al. investigated 1,2-dicyanobenzene (12DCB), tetracyanoquinodimethane (TCNQ) and 1,3-dicyanobenzene (13DCB) drop-casting on the inner surface of the PPWG to form polycrystalline thin films.195,196Figure 13 shows an example comparing the spectra of 12DCB measured in a pellet and in a PPWG.196 Compared to the three broad features for the pellet sample, the PPWG can resolve six narrow absorption features. The FWHM of the peaks can be significantly reduced by up to five times compared to the pellet samples. They have also studied different crystalline forms induced by different concentrations and preparation methods,195 which correspond to different absorption characteristics. In another study by Laman et al., 2,4-DNT sample was measured by the same PPWG method and compared to its pellet form at 11 K. The PPWG sensing identified 19 features with many of them having linewidths below 15 GHz, compared to only seven spectral features and about an eight times broader width observed in the pellet sample. They have also studied seven small biomolecules, including deoxycytidine, nucleosides and amino acids, to demonstrate the applications to biomedical samples.248 The drawback of using PPWG for spectral sensing is the difficulty in extracting the absolute absorption coefficients, due to the slight changes in the frequency-dependent transmission when the PPWG is reassembled to load/remove the sample.

FIG. 13.

The spectra of 12DCB transmitted through (a) a pellet and (b) a parallel plate waveguide covered with 12DCB film, at 295 and 77 K, respectively. The insets show the corresponding absorbance spectra (only the absorbance at 295 K for the pellet sample is shown). Reproduced with permission from Melinger et al., Appl. Phys. Lett. 89, 251110 (2006).196 Copyright 2006 AIP Publishing.

FIG. 13.

The spectra of 12DCB transmitted through (a) a pellet and (b) a parallel plate waveguide covered with 12DCB film, at 295 and 77 K, respectively. The insets show the corresponding absorbance spectra (only the absorbance at 295 K for the pellet sample is shown). Reproduced with permission from Melinger et al., Appl. Phys. Lett. 89, 251110 (2006).196 Copyright 2006 AIP Publishing.

Close modal

Resonance can be introduced to PPWGs for sensing via frequency shifts, in a way similar to that of MM sensors. There are several methods reported. For example, Harsha inserted a dielectric Bragg grating into the PPWG.249 The height of the dielectric grating is only 900 nm and thus introduces very little perturbation, and the waves still propagate in the broadband TEM mode. The structure provides very sharp resonance dips as narrow as 6 GHz and a Q-factor up to 430, significantly higher than traditional MM sensors. Mendis et al. adopted another approach by machining a rectangular groove on one of the waveguide plates.250 The TE1 mode is excited, and an extremely narrow resonance with a linewidth of 3 GHz was formed at 0.293 THz. Liquids can be conveniently filled into the groove to shift the resonant frequencies. A thin stainless steel film with a single slit inserted into the middle of a tapered PPWG can also introduce a sharp resonance with a linewidth of 11 GHz.251 The same group has also investigated the application of this structure in the sensing of a thin-film dielectric layer, showing the sensitivity to the sample length, thickness and refractive index.252 

Other waveguide structures have also been proposed for sensing applications. Microstrip-line waveguides are usually directly integrated with a PCA emitter and detector in an on-chip configuration. Coplanar waveguide lines combined with VNA extenders have been proposed for liquid and cell characterization.253,254 Byrne et al. showed the application of such a sensor for measuring lactose pellets;255 it was able to resolve the consistent absorption dip observed in transmission measurements. Compared to free-space TDS, this technique has the advantages of compactness and no multiple reflections from the system. The lactose spectrum appeared as a straight line with an obvious absorption dip. However, it requires the sample to be placed close to the device to sufficiently interact with the evanescent wave, while direct contact may damage the device since it is integrated with the emitter and detector. Improvements to the robustness can be made using the design proposed by Ohkubo et al.256 They illuminated the pumping and probing beam from the bottom of the device and protected the antenna and the strip line by a thin cover layer. By changing the thickness of the cover layer, they were able to control the interaction efficiency between the THz light and the aqueous sample. Band-pass or band-stop filters can also be integrated with microstrip-line waveguides such that the sensing is measured in a quantity of frequency shift. Wood et al. showed a significant frequency shift of the stop band induced by photoresist layers of a few micrometers.257 Nagel et al. demonstrated that a bandpass microstrip filter can sensitively differentiate the denaturized and hybridized DNA films.258 Dielectric waveguide sensing has also been proposed. In the work by Li et al., a Bragg waveguide with a defect in the first layer showed the ability to sense thin-film lactose. However, it is inconvenient to coat the sample onto the inner surface of the waveguide. The work by Fan et al.259 favors tiny sample sensing. They guided the THz wave along a PMMA cylindrical tube with one side of the tube machined with a grating structure. The output spectrum from the waveguide contained resonance from the interference effect. The grating sharpened the resonance dips and served as a platform to load a small amount of analyte, with the dip position shift determined by the properties of the analyte.

Most of the waveguide structures provide a tight field confinement and a long propagation length. This characteristic is highly favorable for thin-film sensing because a long interaction length can be achieved in the planar direction of the sample. However, similar to the MM sensors, sensitive detection of absorptive aqueous samples has not yet been demonstrated. Although we have seen the characterization of an ultrathin 20-nm water film,247 the special condensation method to form the thin film cannot be applied for many other aqueous samples. Resonance methods, such as Mendis's work,250 are expected to face the same issue as that of MM sensors, namely, that the resonance dip will be significantly broadened by the absorptive sample.

3. THz surface plasmon polaritons

SPR is a technique which was originally used for sensitive probing at visible wavelengths. It is based on the excitation of SPPs, referring to the coupling oscillations of electromagnetic waves and the electron gas density at the metal-dielectric boundary. The resonant wave is nonradiative, traveling only along and tightly confined to the boundary, hence giving the term “surface” and providing the ability to sense thin-film samples. Optical SPPs are generally excited near the plasma frequency of a metal in a prism-coupling configuration. Since the plasma frequencies of all metals are far above the THz range, SPPs at the metal–dielectric interface are very weakly confined at THz frequencies. THz SPPs have been mainly achieved by three approaches: (a) prism-coupled SPR based on doped semiconductors; (b) aperture-coupled SPPs along a conductor–dielectric interface; and (c) spoof SPPs using artificial periodic structures.

The dispersion relation of surface plasmon waves at a conductor–dielectric interface is described by the following equation:260 

kSPP(ω)=ωcεcεdεc+εd=kfree(nSPP+iκSPP),
(4)

where kSPP and kfree are the wavevectors of the SPPs and free space wave, respectively; εc and εd are the relative permittivities of the conductor and dielectric, respectively; and nSPPκSPP are the SPP effective refractive index and extinction coefficient. Exciting SPPs requires nSPP larger than 1. Therefore, free-space radiation cannot directly couple with SPPs due to the smaller wavevector. A prism (i.e., n >1) in an Otto or Kretschmann configuration is usually used to match the wavevector.261,262 For the conductor, a negative real part of εc is also required, which is found below the plasma frequency at which the real part of the permittivity changes its sign. Semiconductors with a proper doping level can tune the plasma frequency to the THz regime, such as doping silicon at a level of 10−16 to 10−17 cm−3.263,264 InAs and InSb have been experimentally demonstrated to be materials capable of supporting well confined THz SPRs, and their magneto-optical response further enables the control of the resonance via external magnetic fields.265,266 However, the quality of the SPR based on these semiconducting materials is relatively low compared to the gold SPR in the visible range. This is fundamentally determined by the dispersion behavior of the semiconductor near the plasma frequency. Generally, an ideal SPR has a wavevector that dramatically increases when approaching the plasma frequency, providing a tightly confined field and a sensitive angular/frequency dependence of the reflectivity. When these are satisfied, a sharp SPR dip in the frequency-domain or in the incident-angle domain can be obtained, providing a high Q-factor and a high sensitivity of the dip shift to tiny changes in the analyte properties.267,268 Compared to the visible devices, the spectral reflectivity of THz SPR is flat and broad, providing a low sensitivity.265 Although the decay length in the dielectric is subwavelength, it is still relatively large compared to MM sensors or waveguides, limiting its thin-film sensing ability. This is potentially why this approach has not been widely applied in practical sensing applications. Apart from bulk semiconductors, graphene has also been theoretically proved to be able to support THz SPR when it is heavily doped or stacked as multiple layers.269,270 Theoretical analysis shows better dispersion characteristics, reflected by the narrow FWHM of the resonance. However, so far there has been no fabricated graphene SPR sensor demonstrated in the THz band.

Another approach to excite THz SPPs is by scattering the THz light at an aperture formed by a gap between a metal blade and a conductor, as first proposed by Saxler et al.271 The SPPs propagating along the conductor over a certain distance will be re-coupled into free space by another blade aperture and will be received by the detector. Due to the continuum of the wave vectors of the scattered evanescent field, the broadband spectrum all can be coupled to the SPPs. The measured signal is a low-pass continuous spectrum up to the plasma frequency. This is in contrast to the above prism-SPR in which only one wavelength will be in resonance and an absorption dip is expected. In Saxler's work, gold was used for the generation of SPPs. Dielectric thin films of 3.8 and 9 μm coated on the gold showed different transmitted amplitudes. However, as discussed above, gold is obviously not an ideal selection for THz SPPs which yields in a very poor field confinement. The later work by Isaac et al. improved this method by using InSb,272 which reduced the decay length of the SPPs in air by over a factor of 10. They also showed a onefold phase change when measuring a thin-film polymer on InSb compared to that on gold. The same configuration has also been used for defect detection by Yang et al.273 They presented the ability to measure a particle on the InSb with a size down to 11 μm. However, the properties of the SPPs are determined by the materials. The InSb used has the same limitation as that in the prism-coupled SPR, that is, the field confinement is still relatively weak. For example, in Issac's work,272 they estimated the decay length into the air at 1.2 THz to be 178 μm using InSb (70% of the wavelength in free space), which is not significantly subwavelength.

A more practical method for THz SPP sensing is launching spoof SPPs on a structured periodic conductor.274 The patterned periodic conductor, which can be regarded as a metamaterial, exhibits a similar dispersion relation to that of conductors near the plasma frequency. Therefore, the term “spoof” was allocated to differentiate it from the natural SPPs. Structured conductors have two important advantages over the natural THz SPP conductors. First, better dispersion characteristics can usually be achieved for a more tightly confined field. Second, the plasma frequency can be freely adjusted by scaling the structure. To excite SPPs, the coupling methods used for the natural conductors can be applied. A very similar configuration to that described above for aperture-coupled SPPs is applied for a groove-structured heavily doped Si design.275 The period, width, and depth of the rectangular grooves are 442, 221, and 100 μm, respectively, giving rise to a low-pass profile with a sharp stop band close to 0.3 THz. The rapid transmission drop is attributed to the rapid decrease in the group velocity approaching the stop band. The highly doped Si can also be directly replaced by metals, as shown in the work by Williams et al.276 A copper metamaterial structure with periodic pits was investigated with the same aperture-coupling method, as shown in Figs. 14(a) and 14(b). The apertures h1 and h2 in Fig. 14(b) are used to couple in and out the SPPs, while the aperture h3 can be adjusted to verify the field confinement. Figure 14(c) shows the measured spectra when h3 = 1000 and 200 μm, respectively, compared to the calculation. The artificial structure provides a better dispersion profile such that a sharp cutoff frequency is observed in the spectrum at the Brillouin zone boundary. In addition, the use of metal reduces the Ohmic losses compared to doped semiconductors. Other coupling methods have also been reported. Li et al. fabricated a V-groove structure made of heavily doped Si.277 They directly illuminated a normal-incident light to an extra coupling groove 5 mm away from the V-groove structure, which is more convenient than the aperture coupling method. A sharp stop band edge was also observed, owing to the bent dispersion curve. From the propagation point of view, these devices can also be regarded as plasmonic waveguides. The Otto configuration can also be applied. In that case, the detected reflection is expected to have a narrow dip at the resonant frequency, in contrast to a stop band. Ng et al. used a wax prism in an Otto configuration to couple the THz light to a groove structure deposited with gold.278 The schematic and the experimental configuration of the groove metamaterials are shown in Figs. 15(a)–15(c). By optimizing the gap between the prism and the grooves, a deep and narrow resonant dip, accompanied by a dramatic phase change was formed. A high FOM up to 49 was achieved for nitrogen sensing and a sensitivity of 0.52 THz RIU−1 was found from the fitting. They have shown sensitive detection of different analytes that obviously shift the resonant dip, as shown in Fig. 15(d). However, samples, such as water and glycerin with a large absorption, have a significant damping effect, which obviously broaden the resonance. Using a prism with a higher refractive index can further improve the field confinement in such a configuration, as demonstrated by Huang et al.279 They have also proposed a more accurate method for measuring the frequency shift by looking at the phase change (i.e., the first derivative of the phase spectra), which offers a higher Q-factor than the absorption spectra.280 However, similar broadband resonances were found for absorptive liquids, indicating the common damping effect caused by the large sample absorption.

FIG. 14.

Aperture-coupled THz spoof surface plasmon polaritons: (a) illustration of the experimental arrangement and the coupling between the free-space THz light and the surface plasmon polaritons; (b) schematic of the plasmonic metamaterial with periodic pits; and (c) measured spectra for one of the samples (sample II) when h3 = 1000 and 200 μm, respectively, compared to the calculation. The dashed vertical line indicates the Brillouin zone boundary. Reproduced with permission from Williams et al., Nat. Photonics 2, 175–179 (2008).276 Copyright 2008 Nature Publishing Group.

FIG. 14.

Aperture-coupled THz spoof surface plasmon polaritons: (a) illustration of the experimental arrangement and the coupling between the free-space THz light and the surface plasmon polaritons; (b) schematic of the plasmonic metamaterial with periodic pits; and (c) measured spectra for one of the samples (sample II) when h3 = 1000 and 200 μm, respectively, compared to the calculation. The dashed vertical line indicates the Brillouin zone boundary. Reproduced with permission from Williams et al., Nat. Photonics 2, 175–179 (2008).276 Copyright 2008 Nature Publishing Group.

Close modal
FIG. 15.

Prism-coupled THz spoof surface plasmon polaritons: (a) schematic and (b) optical microscope of the spoof surface plasmon polaritons groove structure; (c) illustration of the Otto-configuration to couple the THz light to SPPs; and (d) measured spectra of different sample fluids. Reproduced with permission from Ng et al., Adv. Opt. Mater. 1, 543–548 (2013).278 Copyright 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim.

FIG. 15.

Prism-coupled THz spoof surface plasmon polaritons: (a) schematic and (b) optical microscope of the spoof surface plasmon polaritons groove structure; (c) illustration of the Otto-configuration to couple the THz light to SPPs; and (d) measured spectra of different sample fluids. Reproduced with permission from Ng et al., Adv. Opt. Mater. 1, 543–548 (2013).278 Copyright 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim.

Close modal

According to the coupling and detecting method, THz SPP sensing may have similar characteristics to those of MM sensors and waveguides. Aperture-coupled SPPs can be regarded as plasmonic waveguides such that a long interaction distance with the thin-film sample can be achieved. Prism-coupled SPR has the merit of a sharp resonance similar to MMs such that a sensitive detection can be achieved by monitoring the shift of the resonance. However, overall THz SPP is also a refractive-index sensing technique that has a worse performance with a higher sample absorption.

We have introduced various techniques for THz sensing, which have been mainly applied for thin-film samples, typically molecules, cells, and biomolecular solutions. Tissues have a much larger scale with spatial and frequency-dependent properties; thus, they are usually directly measured by label-free THz spectroscopy and imaging. Applications relating to disease diagnosis have been widely reported over the last two decades. They can be classified into the two distinct categories of in vitro and in vivo studies, which are very different in terms of the sample properties, preparation methods and measurement protocols. In the following, we will first overview the in vitro THz applications.

The current state of THz technology is still primarily confined to university laboratories due to critical experimental requirements. For example, THz imaging is mostly performed by a slow raster scanning. Reflection measurements require a highly flat sample surface for precise alignment, and typically good contact with a window is needed for accurate phase acquisition. In vitro tissues can better meet these critical requirements as they have a more ideal sample shape, and measurement variables, such as the applied pressure on the window or the subject-variance, are more easily controlled than for in vivo studies. They can also be accessed and handled more conveniently. In addition, in vitro studies are necessary to investigate and understand the origin of contrast, which is essential for further applications in clinical cancer detection and delineation. Some conceptual experiments can only be performed on in vitro samples. Excised tissues may also aid the histopathological examination. Therefore, in vitro studies are of great importance in THz biological applications. Next, we will talk about the sample preparation methods for in vitro measurements, followed by the most common in vitro application of pathological detection, especially neoplasm detection.

1. Sample preparation

In vitro samples for THz measurements can be either fresh or dehydrated, and there are several methods which can be used to prepare the samples. Freshly excised tissues can be directly measured, which have the advantage of preserving the original sample properties and contrast, while the large water content results in a high absorption close to that of pure water. This means that a thin slide, typically less than 500 μm, is needed to measure them in transmission. This can be achieved but for some tissues it could be challenging. The high absorption also limits the bandwidth due to the stronger attenuation at higher frequencies.149 Therefore, fresh samples can be more conveniently measured in reflection since they are usually soft enough to contact well with a window.60,156 Another problem for fresh samples is that they can be quickly dehydrated after being excised, especially when exposed to dry air or nitrogen in the experimental chamber. Structural and chemical changes could also happen with time. Fan et al. proposed a novel method to embed the fresh sample in gelatin,122,281 as shown in the inset of Fig. 16(a). Figures 16(a) and 16(b) show that the THz properties of the gelatin preserved skin sample have negligible changes for up to 35 h. In comparison, the sample without gelatin preservation (i.e., control sample) showed dramatic property alterations after 9 h. Although fresh tissues to the best extent mimic the in vivo conditions, freezing the fresh sample to effectively reduce the water absorption could be useful in many aspects. For example, it allows a precise transmission measurement by allowing a larger sample thickness and may sometimes provide a better contrast between diseased and healthy tissues by reducing the high background absorption from water. Since water has a monotonically increasing absorption at THz frequencies, freezing the sample also improves the transparency at high frequencies; thus, a broader bandwidth can be measured. Hoshina et al. showed the greatly extended measurable bandwidth of a 1.7 mm porcine tissue in transmission at −33 °C.282 Sim et al. showed an enhanced contrast between oral cancer and normal mucosa after freezing the sample at −20 °C compared to the results obtained at room temperature.283 A hidden tumor inside the tissue was also observed in the frozen sample. He et al. further investigated the influence on the sample properties in the freeze–thaw process and found that by combining snap-freezing and fast-thawing, it was possible to remove the hysteresis in the slow-thawing process.284 

FIG. 16.

The changes in the refractive index with increasing time of (a) the gelatin preserved sample and (b) the control sample. Inset of (a) shows the diagram of the skin sample embedded in gelatin during the measurement. Reproduced with permission from Fan et al., Phys. Med. Biol. 60, 2703 (2015).281 Copyright 2015 Institute of Physics and Engineering in Medicine.

FIG. 16.

The changes in the refractive index with increasing time of (a) the gelatin preserved sample and (b) the control sample. Inset of (a) shows the diagram of the skin sample embedded in gelatin during the measurement. Reproduced with permission from Fan et al., Phys. Med. Biol. 60, 2703 (2015).281 Copyright 2015 Institute of Physics and Engineering in Medicine.

Close modal

Dehydrated samples have also been broadly studied. There are a few reasons; first, it provides a high transparency in transmission. Second, it enables the background contrast from the nonwater component to be investigated. Third, the sample can be easily handled with the properties remaining unchanged over a long time. The most widely adopted method is to fix the sample in formalin,155,285 which could be followed by ethanol dehydration and paraffin-embedding.102,149 Sun et al. studied the effect of formalin-fixing on the THz properties of samples.285 The observation of the reduced refractive index and absorption coefficient with the increasing fixing time indicates the reduction in the water concentration caused by the formalin. In addition, new intermolecular interactions between the sample and formalin could occur. The fixed sample becomes rigid and could be distorted. A good contact with a window could be difficult, and scattering may occur affecting the results of the measurement. To prepare samples with a controllable surface flatness and thickness, the fixed sample is usually further soaked in increasing concentrations of ethanol to remove the remaining water and embedded in paraffin wax to enable precise sectioning by a microtome. Another potential influence of formalin-fixing is that it could change the protein and DNA structure. Png et al. proposed the lyophilization (freeze drying) method to overcome this issue.286 The lyophilization procedure includes snap freezing the sample and placing it in a low-pressure (10–100 Pa), low-temperature (−74 °C) environment to sublimate the water. The method was claimed to better retain sample freshness.

2. Pathological detection

Neoplasms are the most common type of pathological changes being studied by THz in vitro. A neoplasm refers to a type of abnormal and excessive growth of tissue, which will normally but not necessarily develop into a tumor. Early diagnosis of neoplasms has a great impact on human health. THz is one of the technologies with the potential to provide good contrast between neoplasms and normal tissues. Given that noninvasive measurement is the most ideal modality for cancer detection, skin is potentially the most suitable target as it is relatively easy to access and THz waves can penetrate into certain depths of skin. Therefore, for in vitro studies, skin tissues are one of the earliest investigated samples. Skin cancers can be categorized into two main types, melanoma and nonmelanoma cancers. Most of the investigations have been on nonmelanoma cancers, including basal cell carcinomas (BCCs) and squamous cell carcinomas (SCCs). Early work by Woodward et al. demonstrated the image contrast in the reflected THz amplitude at a fixed time position between BCCs and healthy tissues.287 The higher reflectivity from the BCC area was confirmed by Wallace et al. in the study of 19 fresh BCC lesions.288 Further analysis of the spectral properties of BCCs and normal areas attributes the difference to the increased water content in the diseased tissues.289Figure 17 shows that both the refractive index and absorption coefficient of BCC are higher than that of the normal skin.138 The difference can be clearly reflected by the fitted double-Debye parameters. Another technology using a CW THz imaging system showed that both BCCs and SCCs can be distinguished from healthy tissues by their cross-polarization reflections.290 Oral cancer is another kind of neoplasm that could potentially be measured noninvasively, although with the current state of THz systems such an application has not yet been tested in vivo. Sim et al. measured frozen fresh oral cancer specimens and compared the results to measurements taken at room temperature; there was a larger contrast between oral cancer and normal mucosa at frozen temperatures.283 Further, a lower refractive index and absorption coefficient was observed for the cancerous tissues, which is different from the previous skin studies. Sun et al. utilized a multiple-layer structure to enhance the imaging sensitivity. Dehydrated oral cancer tissues were deposited on a quartz window and then covered by water. Such a quartz–sample–water sandwich structure gave a significant contrast between healthy and cancerous areas, while the THz transmission images could not clearly differentiate the two areas due to the small thickness of dehydrated sample slice.78 

FIG. 17.

(a) The refractive index and (b) absorption coefficient of normal skin tissue and basal cell carcinomas. The error bars represent 95% confidence limits. Reproduced with permission from Pickwell et al., J. Biomed. Opt. 10, 064021 (2005).138 Copyright 2005 SPIE.

FIG. 17.

(a) The refractive index and (b) absorption coefficient of normal skin tissue and basal cell carcinomas. The error bars represent 95% confidence limits. Reproduced with permission from Pickwell et al., J. Biomed. Opt. 10, 064021 (2005).138 Copyright 2005 SPIE.

Close modal

Neoplasms inside the human body are very difficult to measure in vivo even with state-of-the-art THz technology, mainly due to the difficulty in aligning beams to an irregular surface, the slow imaging speed and the strong attenuation of THz light by water. Studies of excised specimens avoid many of the technical issues associated with in vivo measurements. As can be seen from the following examples, the studies of either fresh or dehydrated samples show the essential THz contrast between normal and diseased areas. Breast cancer was one of the first types of cancer studied to investigate the ability of THz measurements to observe contrast between healthy and cancerous tissues. The study by Fitzgerald et al. shows a high correlation between the THz images and the photomicrograph for the tumor area of 22 freshly excised breast tissue specimens.156 Using transmission measurements of the sample, a larger refractive index and absorption compared to healthy fibrous tissue was found,291 in agreement with the characteristics observed for skin cancers. The contrast can also be provided by other measured parameters. Figures 18(a) and 18(b) show the THz reflected images plotting Emax and Emax/Emin, respectively, compared to the histology image in Fig. 18(c). The pink region in the histology image indicates the cancerous tissue, and this can be seen to correspond to the yellow region in the Emax/Emin image, indicating that this parameter is capable of differentiating the breast tumor from the surrounding adipose tissue. In fact, the broadband detection enables the extraction of a large number of parameters, which can be investigated in order to differentiate between different tissue types. Automatic classifications of cancerous and healthy breast tissues can be done by adopting machine learning methods, such as support vector machine (SVM), to account for multiple features.102,292 A similar data analysis has been applied for many other diseases. We will introduce numerous advanced data analysis methods independently in Sec. VI. Brain tumors have recently attracted more attention. Oh et al. measured both freshly excised and paraffin-embedded rat brain tissue.158 A higher hydration level was found in tumor areas leading to a higher refractive index and absorption, providing a good contrast to the normal areas in the THz images. Figures 18(d)–18(g) compare the visual, THz and magnetic resonance (MR) images of fresh brain tissues, where the samples in (d)–(f) contain brain tumor and (g) is a normal brain tissue. This contrast between healthy and cancerous regions is shown in the THz images in Fig. 18 by the red regions and is consistent with the bright regions in the MR images caused by a higher water concentration. The THz images are given by the peak-to-peak reflected amplitudes normalized to the metal reflection. This further confirms that water could be the primary source of contrast for cancer detection. When the water was removed, the cell density still affected the THz response, but it was found that it contributes less to the contrast. The observations are in line with the spectral analysis of fresh normal and tumor rat brain tissues by Yamaguchi et al.293 They have also considered the higher cell density and water content in the tumor as the reasons for their larger refractive index and absorption. The linear discriminant analysis was used to classify the data with an accuracy up to 98%. Ji et al. studied both fresh mouse brain samples and human specimens of different grades.154 They showed clear contrast at 0.5 THz among glioma, normal white, and gray matter. The higher reflection in the gray matter compared to the white matter could be the result of the intrinsic differences in water and lipid concentrations. Glioma of different grades were also investigated by Gavdush et al.122 Fresh specimens (26 in total) from patients were embedded in gelatin and measured, with a higher refractive index in the tumor region for grade I to IV. The increased water content caused by the abnormal vascularity and edema is considered as the origin of contrast. A recent topical review on THz diagnosis of human brain tumor is given in Ref. 294.

FIG. 18.

THz reflected images of breast tissue plotted by (a) Emax, (b) Emax/Emin, respectively, compared to (c) the hisotology image. (a)–(c) are reproduced with permission from Ashworth et al., Opt. Express 17, 12444–12454 (2009).291 Copyright 2009 Optical Society of America. (d)–(g) Visual, THz, and magnetic resonance (MR) images of brain tissues (d)–(f) with and (g) without brain tumor. The THz image size is 4 × 3 cm2. (d)–(g) are reproduced with permission from Oh et al., Biomed. Opt. Express 5, 2837 (2014).158 Copyright 2014 Optical Society of America.

FIG. 18.

THz reflected images of breast tissue plotted by (a) Emax, (b) Emax/Emin, respectively, compared to (c) the hisotology image. (a)–(c) are reproduced with permission from Ashworth et al., Opt. Express 17, 12444–12454 (2009).291 Copyright 2009 Optical Society of America. (d)–(g) Visual, THz, and magnetic resonance (MR) images of brain tissues (d)–(f) with and (g) without brain tumor. The THz image size is 4 × 3 cm2. (d)–(g) are reproduced with permission from Oh et al., Biomed. Opt. Express 5, 2837 (2014).158 Copyright 2014 Optical Society of America.

Close modal

Neoplasms in many other locations, such as the stomach,295,296 liver,106,297 colon,60,298,299 and cervix,300 have also been broadly explored. These studies provide the essential knowledge of the THz response between different cancerous and healthy tissues. Apart from neoplasms, other pathological tissue changes have also been studied. Liver cirrhosis was studied by Sy et al. by measuring both fresh and fixed samples.155 The results show a considerably large contrast contribution of 50%–66% from the biological background, even higher than that by water. Alzheimer's disease was studied, and it was found that the disease will increase the contrast between the gray matter and white matter to 4.2%, compared to 2% in the normal brain tissues.301 The demyelination of white matter in Alzheimer's tissues could be the potential cause of the difference. Cornea hydration is related to diseases including keratoconus and Fuchs' dystrophy.302In vitro hydration sensing of porcine corneas has been done by Bennett et al.303 They used EMT on different hydration levels, and it shows a good fit, with a linear relationship between the THz reflectivity and the water concentration being observed. Most of these studies conclude that water is the major source of contrast, and typically a higher water content is observed in the cancerous/diseased area. However, tissues of different types, locations and grades, can have different biological compositions and structural changes that may contribute to the specific THz dielectric properties, and in some cases, these factors could dominate the contrast. Investigating samples in both fresh and dehydrated states in the same experiment makes it possible to better explore the origin of contrast. Although to some extent consistency in the observed contrast is determined by the THz systems used, sample preparation techniques, measurement protocols, and data processing, numerous variables among these processes could alter the experimental results, introducing difficulties in comparisons. Also, experimental errors could be difficult to comprehensively estimate, such as the systematic error, nonideal contact, or scattering from the rough surface. Currently, it is difficult to establish a gold standard for THz in vitro studies of tissues, as this field is still under rapid development. However, attention should be given to controlling these variables and developing a robust protocol in future studies to improve data credibility and comparability.

3. Other applications

In vitro measurements are also important in many other studies. For example, the fundamental investigation of the EMT adaptability for biological samples is based on the measurement of both fresh and dehydrated samples,149 as we have also mentioned in Sec. III B 2. In vitro studies can also be applied to explore agents to enhance the penetration depth or the imaging contrast. Oh et al. demonstrated that glycerol can effectively enhance the penetration depth of THz waves into fresh tissues.304Figure 19(a) shows their experimental setup. A metal plate was placed above the mice skin. Figure 19(b) shows the reflected THz pulse of the skin with/without glycerol. Applying the glycerol significantly enlarged the second pulse reflected from the skin–metal interface and reduced the separation between the two reflections. The results correspond to a decrease in both the refractive index and absorption coefficient. They have also demonstrated the potential application of detecting tumors underneath a thick skin. In Fig. 19(c), Matrigel drops (normally a gel form except at a temperature of 4 °C) on the backside of the sample are pictured and can be compared to the location of the contrast in the THz images. Figure 19(d) shows the peak-to-peak value of the entire waveform measured from the sample, while Fig. 19(e) shows the peak to peak after the initial reflection has been removed so only the reflection from the metal plate remains. In Fig. 19(e), the shape of artificial tumor made by the Matrigel drops can be observed as the dark region in the image when glycerol has been applied to the sample, but only very weak contrast was found in the sample without applying the glycerol. Another work by the authors showed that gold nanoparticles could also be a potential contrast-enhancing agent.305 Cancer cells with/without adding gold nanoparticles illuminated by IR radiation showed significant/no changes on the reflection. However, it should be noted that THz does not interact with the nanoparticles directly due to the much larger wavelength compared to the particle dimensions. The changes are purely temperature-based by coupling the IR radiation with the surface plasmon resonance of the nanoparticles to heat up the cells. THz imaging was utilized as a thermometer to monitor the changes in the properties of water induced by the temperature. Gold nanoparticles have also been used to assist laser tissue soldering, which is an emerging concept in surgical medicine. Utilizing the time-resolution of THz-TDS and the high sensitivity to water, Dong et al. demonstrated 3D reconstruction of photothermal damaged porcine skin induced by tissue soldering.306 Another application of THz in vitro studies is drug delivery, which can also be monitored by THz imaging based on the high sensitivity to water and drug concentrations. Kim et al. applied dimethyl sulfoxide (DMSO) containing ketoprofen onto a mouse skin sample and demonstrated that THz imaging can reveal the spatial distribution and penetration depth of the drug solution.307 As the drug was being absorbed, the refractive index of the sample reduced, giving rise to an index matching with the quartz window; thus, the reflected signal decreased with the time. Their further study demonstrated that THz imaging can also distinguish different concentrations of the drug solutions.308 Wang et al. used in vitro porcine skin to evaluate the transdermal drug delivery efficiency by different application methods, including topical application and via microneedles to enhance the permeability of skin.20 The results show that THz imaging can be used to evaluate different methods for transdermal drug delivery.

FIG. 19.

(a) Schematic of the experimental setup. (b) Reflected THz pulses without and with applying glycerol. (c) Visible image and THz images plotted using (d) the peak-to-peak values of the whole waveform and (e) the peak-to-peak values of the waveform after cut off the main pulse. Reproduce with permission from Oh et al., Opt. Express 21, 21299 (2013).304 Copyright 2013 Optical Society of America.

FIG. 19.

(a) Schematic of the experimental setup. (b) Reflected THz pulses without and with applying glycerol. (c) Visible image and THz images plotted using (d) the peak-to-peak values of the whole waveform and (e) the peak-to-peak values of the waveform after cut off the main pulse. Reproduce with permission from Oh et al., Opt. Express 21, 21299 (2013).304 Copyright 2013 Optical Society of America.

Close modal

The nonionizing nature of THz radiation means that it has great potential for safely imaging biological tissues in vivo. Furthermore, it enables a noninvasive detection without additionally extracting the region of interest prior to measuring it, e.g., for a biopsy examination. This also removes the need to use techniques to preserve the sample which could alter the properties of the region. However, most THz in vivo measurements have been limited to areas that can be accessed noninvasively, typically skin, and in some studies the cornea has been measured as well. This is because of the high absorption of THz light by tissue such that THz light cannot detectably penetrate deeper than a few hundreds of micrometers into biological tissues. More importantly, as in vivo measurements can only be performed in a reflection configuration, a flat surface in most cases is required to align and receive the THz light. This is typically achieved by contacting a well-aligned imaging window on the measured area. Even in a surgery in which the area inside the body can be exposed to THz light, the need of a window is not an ideal modality for clinical applications, as we will later introduce in the examples.154,309 Although some concepts, such as THz endoscope, have been proposed, they are still only single-pixel detection and require precise alignment, which is far from practical. For the available areas such as skin, there are challenges which arise when a living subject is measured such as the movement of the subject during the measurement, and an increased exposure of the sample to other environmental factors which could alter the properties of the region.

As outlined above, many studies use an imaging window to aid the alignment of the sample and to flatten the soft tissue making it easier to image without the need to account for the curvature of the biosample surface. In some studies, a processing technique is used to remove the influence of the lower reflection from the imaging window.310 For skin which is composed of multiple layers, a deconvolution is usually performed on the measured response of the sample with an air reference. The result is described as an impulse function and can be used to identify the presence of hidden layers in the sample and to quantify changes in the water content of the sample. Other more advanced processing techniques which are being researched to enhance the information that can be extracted from THz measurements are explained in detail in Sec. VI.

This section will summarize some of the key developments in the field of in vivo assessment with THz spectroscopy or imaging, highlighting different biological changes which have been shown to induce a contrast at THz frequencies. We will also introduce some of the work which has been done to address other variables which can influence the THz response of biological tissues and what experimental approaches have been employed to reduce the effects of these.

1. Skin assessment

The first investigation in which human skin was measured in vivo in laboratory conditions was performed by Cole et al. in 2001.311 They demonstrated the ability of THz pulsed imaging to identify the thickness of the stratum corneum (SC) in regions, such as the palm where the SC is thick enough such that the reflection at the SC–epidermis interface can be distinguished from that from the quartz–SC interface. By performing a raster scan to acquire an image, it was possible to present results similar to an ultrasound B-scan in which the different layers in the skin can be identified as a function of depth. However, due to the high water content, the accessible depth is wavelength-comparable; thus, it mainly provides superficial skin information. They also found that they could observe changes in the hydration of the skin with THz measurements, for example, following the application of a wet towel to the skin surface. It was also observed that the THz response of skin changed throughout the measurement while the skin was in contact with the imaging window, when using a setup, such as the one shown in Fig. 20(a). This was attributed to the occlusive effect of the imaging window, as it prevents the natural process of water loss through the skin surface causing water to accumulate in the skin with increasing contact time. Figure 20(b) shows microscope images of skin increasing times into occlusion, and the smoothing of the skin throughout this process can be observed. Figures 20(c) and 20(d) show the effect of occlusion on the measured THz response of skin during 10 min of occlusion, (c) shows the processed waveforms increasing times into occlusion where they have been shifted horizontally for clarity, and (d) shows the amplitude of these waveforms as a function of time. Sun et al. demonstrated that this occlusion effect can be modeled using a biexponential function; this is demonstrated by the red line in Fig. 20(d).108 A more complete summary of the work performed by Sun et al. will be presented in Sec. V D 4 along with approaches which can be used to remove the effects of occlusion from results obtained with contact THz measurements of the skin.

FIG. 20.

(a) An example of an experimental setup used to measure the skin on the volar forearm in a reflection geometry. Reproduce with permission from Lindley-Hatcher et al., J. Phys. 3, 014001 (2020).312 Copyright 2020 Authors, licensed under a Creative Commons Attribution (CC BY) license. (b) Microscope images of the skin after increasing occlusion times. Reproduced with permission from Chen et al., Adv. Photonics Res. 2, 2000024 (2021).84 Copyright 2021 Authors, licensed under a Creative Commons Attribution (CC BY) license. (c) Processed THz signals obtained at 1-min intervals when measuring skin for 10 min under occlusion, the waveforms have been horizontally shifted for clarity. (d) The peak to peak amplitude of the processed waveform plotted against occlusion time and fitted with a biexponential function shown in red.

FIG. 20.

(a) An example of an experimental setup used to measure the skin on the volar forearm in a reflection geometry. Reproduce with permission from Lindley-Hatcher et al., J. Phys. 3, 014001 (2020).312 Copyright 2020 Authors, licensed under a Creative Commons Attribution (CC BY) license. (b) Microscope images of the skin after increasing occlusion times. Reproduced with permission from Chen et al., Adv. Photonics Res. 2, 2000024 (2021).84 Copyright 2021 Authors, licensed under a Creative Commons Attribution (CC BY) license. (c) Processed THz signals obtained at 1-min intervals when measuring skin for 10 min under occlusion, the waveforms have been horizontally shifted for clarity. (d) The peak to peak amplitude of the processed waveform plotted against occlusion time and fitted with a biexponential function shown in red.

Close modal

Pickwell et al. performed a series of in vivo measurements on 20 human volunteers to explore the repeatability of THz measurements of the skin and the efficacy of theoretical models at describing the interaction of THz radiation with living skin.136 The palm and the dorsal and volar forearm were measured weekly for four weeks using a contact reflection geometry. The volar forearm was found to give the most repeatable results; this is thought to be because it is the region with the least exposure to environmental conditions, such as sunlight, and does not have as much hair as the dorsal forearm. The double Debye model introduced in Sec. III B 1 was used in combination with the finite difference time domain (FDTD) model proposed in Ref. 137 to simulate the THz response of skin. Two types of models were used for the skin structure; the volar and dorsal forearm were assumed to be single-layer structures as the SC is too thin in these regions to be resolved with THz wavelengths, while the palm was modeled as a two layer structure to account for the thicker SC. In all regions, there was good correlation between the measured and simulated impulse functions of the skin, confirming the efficacy of this model in describing in vivo THz–skin interactions. Any discrepancies between the measured and simulated results were attributed to the oversimplification of the structure of the skin used in the model. Zaystev et al. also performed in vivo THz measurements in a range of the locations across the body extracting the refractive index and absorption coefficient.313 They were able to observe variation in the THz optical properties of the skin across the body, which could be attributed to changes in the SC thickness or hydration.

Echchgadda et al. measured the regions on the dorsal forearm, volar forearm, and the palm of 35 human subjects and extracted the refractive index and absorption coefficients of each of the regions.314 They also measured the transepidermal water loss (TEWL) and the extent of melanin pigmentation—two common techniques for skin assessment. TEWL measurements give an indication of the skin barrier function and are, therefore, also linked to the hydration state of the skin while the melanin pigmentation describes the skin tone. The study found no significant differences in the absorption coefficients of the three regions measured. However, it was found that the palm had significantly lower refractive index than the two forearm regions and that these changes are most significant at lower frequencies. The study also found moderate correlation between the slope of the absorption coefficient and the TEWL of the skin suggesting that THz measurements of the skin are effective for hydration assessment.

Up until this point, the summary of the use of THz radiation for in vivo skin assessment has been focused on techniques with resolution limited by the wavelength. However, near-field imaging techniques are also being developed, as introduced in Sec. II B 3. Tseng et al. applied such an approach to in vivo imaging.92 This study took the unusual approach of utilizing a transmission geometry to perform the measurements; this was possible as the region being imaged was the ear of a mouse, which was found to have a thickness of around 0.2 mm, making it thin enough for the THz radiation to pass through the tissue and be detected with sufficient signal to noise. Using this approach, they were able to perform measurements with sub wavelength resolution in which it was possible to observe blood vessels in the ear. A dip could be seen in the transmission through the tissue in the location of the blood vessel due to the higher water content in the blood compared to the surrounding tissue, and examples of these results can be seen in Fig. 21. They were also able to apply FDTD simulations of blood vessels to investigate the impact of changing the refractive index and absorption coefficient of the blood and surrounding tissue on the quality of the fit. They also assess the suitability of the technique for performing longer term monitoring of blood vessels by measuring the effect of injecting the mouse with insulin. Changes in the properties of the blood vessel in the mouse that has been injected with insulin can be seen compared to those of the control mouse which received no such treatment. This technique shows promise for imaging sub-wavelength features of the skin; however, for most human measurements, a transmission geometry will not be feasible so the technique should be developed into an approach, which can be used to image thicker regions through which THz radiation cannot penetrate.

FIG. 21.

In vivo measurements of the capillaries in mouse ears: (a) optical images of the regions, (b) THz transmittance images, (c) THz images normalized to the peak transmittance between the vessels, and (d) normalized transmittance cross section along the arrow. Reproduced with permission from Tseng et al., Opt. Express 23, 25058–25071 (2015).92 Copyright 2015 Optical Society of America.

FIG. 21.

In vivo measurements of the capillaries in mouse ears: (a) optical images of the regions, (b) THz transmittance images, (c) THz images normalized to the peak transmittance between the vessels, and (d) normalized transmittance cross section along the arrow. Reproduced with permission from Tseng et al., Opt. Express 23, 25058–25071 (2015).92 Copyright 2015 Optical Society of America.

Close modal

2. Quantifying hydration changes in the skin

In addition to measuring the refractive index and absorption coefficient of the skin at THz frequencies or changes in the THz pulse reflected from the skin surface, it is possible to use some of the models introduced in Sec. III B 2 to extract further information about the skin, such as the percentage water content or SC thickness. Other studies deliberately change the hydration properties of the skin to observe the sensitivity of THz measurements to such changes. In this section, we will summarize some of the key developments in this area; for the full details of these models, we refer the reader to a review of the models describing the interaction of THz radiation with the skin.9,148

Sun et al. were able to extract the diffusivity of occluded skin.152 They modeled the water distribution in the skin using a swelling model for the SC and proposed three different functions to describe the diffusivity. These models were used to produce a hydration profile for the skin which was entered into the stratified media model first proposed by Bennett et al.70 By fitting to the measured results with these models, the hydration profile and diffusivity in the skin could be extracted. The measurements were performed on five healthy subjects, and the skin was measured throughout 20 min of occlusion by the imaging window. The study revealed that the function in which diffusivity was expressed as proportional to the water gradient across the SC gave the best fitting results, it was then possible to infer that it is this hydration gradient which is driving the movement of water through the skin during the occlusion. The convection velocity of water as a function of occlusion time was also extracted, and this confirmed that the rate of water movement decreased with occlusion time as the gradient across the SC decreases as the upper layers of the SC become saturated with water. A summary of these key results can be seen in Fig. 22(a), which plots the hydration profile in the skin as a function of occlusion time on the horizontal axis, and the SC thickness is plotted on the vertical axis so that the swelling of the SC as the water content increases can be seen. In this example, the high absorption of water shows a unique feature of the THz range. Although the detectable depth of the hydration profile is limited to mainly the stratum corneum, it provides high sensitivity measurements and can determine the sub-wavelength water-content variation.

FIG. 22.

(a) A colormap of the dynamic changes in water concentration in the surface of the skin as a function of skin depth throughout occlusion. Reproduced with permission from Sun et al., J. Biophotonics 12, e201800145 (2019).152 Copyright 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (b) THz line scan images increasing time after 10 min silicone gel sheets, 10 min wet bandage (WB), 4 h silicone gel sheets, and 12 h silicone gel sheets application (left to right). Reproduced with permission from Wang et al., Biomed. Opt. Express 10, 3584–3590 (2019).151 Copyright 2019 Optical Society of America.

FIG. 22.

(a) A colormap of the dynamic changes in water concentration in the surface of the skin as a function of skin depth throughout occlusion. Reproduced with permission from Sun et al., J. Biophotonics 12, e201800145 (2019).152 Copyright 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (b) THz line scan images increasing time after 10 min silicone gel sheets, 10 min wet bandage (WB), 4 h silicone gel sheets, and 12 h silicone gel sheets application (left to right). Reproduced with permission from Wang et al., Biomed. Opt. Express 10, 3584–3590 (2019).151 Copyright 2019 Optical Society of America.

Close modal

Wang et al. performed a study taking THz point and line scans to quantitatively assess the efficacy of different burn treatments at increasing the hydration of the skin.151 They measured the volar forearms of ten human subjects and used a SC–epidermis bilayer model of the skin alongside the LLL EMT to extract the changes in the SC hydration and thickness in response to the treatments. The point scans compare the effects of silicone gel sheets (SGSs) (an occlusive layer which is stuck to the skin for the duration of the treatment) to a silicone gel sample which is topically applied. They found that the effects of the silicone gel were minimal, in contrast the amplitude of the time-domain signal was significantly reduced following the application of the SGS, and the magnitude of this reduction increased the longer the sheet was applied. Using the EMT, they were able to find that the water content of the epidermis did not significantly increase following the application of the sheet, however the hydration of the SC was found to increase from 24% to 58% after 4 h SGS application, the thickness of the SC was also found to have been significantly increased. They also observed that skin that had been treated with the SGS had a less significant response to the occlusive effect of the imaging window than untreated skin, as the treated skin was already saturated with water. Further, a line scan was also acquired of a region of the forearm where a thin strip (5 mm) of SGS had been applied in the direction perpendicular to the line scan so that treated and untreated skin could be measured. They measured the skin before applying the SGS and up to 2 h after removal where the SGS had been applied for 10 min, 4 h, and 12 h. An example of these results is shown in Fig. 22(b), where the blue regions indicate skin with increased hydration and each row in the figure shows the line scan of the region with increasing time after the removal of the SGS. For comparison they also measured the response of the skin to the application of a wet bandage (WB). They found that the longer the SGS is applied the longer it takes the hydration levels of the skin to recover back to their initial state. They also found that the wet bandage increased the hydration the most significantly however these changes did not last and could only be seen up to 5 min after the removal of the bandage. It was observed that there were no lateral effects of the SGS sheeting on the hydration levels of the surrounding tissues, while the wet bandage increased the hydration of the tissues surrounding it up to double its own width. This study demonstrates the potential of THz imaging for evaluating the efficacy of different hydration techniques for burn treatments and providing advice as to the correct application durations for occlusive bandages.

The study performed by Lindley-Hatcher et al. involved obtaining point scan THz measurements from 20 subjects before and after the application of three different types of skin moisturizers.312 They measured four regions of the volar forearm, one which remained untreated as a control and the other three were treated with an oil, water and water in oil-based moisturizer, respectively. The regions were all measured before the treatment and again 20 min after the application of the sample. The regions were all also measured using the commercial gold standard for skin hydration assessment: the corneometer. The study found that the THz measurements were able to distinguish between the different skin responses to the different sample types and moderate correlation was found between the THz amplitude of untreated skin and the measured corneometer response. The corneometer clearly indicated changes in the hydration state of skin following the application of the samples; however, it appeared to be less able to distinguish between the different effects of the samples, for example, the water-based and water-in-oil samples appeared to have the same effect on the response of skin measured with the corneometer. This study suggests that THz sensing could be a useful tool for assessing changes induced in the skin by different moisturizer samples and even help to identify which moisturizers would be most effective for certain skin types.

In addition to using the standard reflection geometry in which a single polarization state is used to take THz measurements of the skin, new geometries and approaches to in vivo THz skin assessment are also being developed. For example, Chen et al. developed an ellipsometry system capable of measuring the skin using s and p polarizations at two incident angles.84 Using the experimental setup introduced in Sec. II B 2 and shown in Fig. 4, they were able to obtain four complementary spectral ratios describing the skin; this enhanced the ability to resolve the complex model of skin. They used an anisotropic model for the SC to account for the alternating layers of lipids and corneocytes that makeup the SC, as shown in Fig. 23(a). Using the data obtained with this experimental setup, they were able to verify the birefringent nature of the SC. By measuring the skin throughout 31 min of occlusion, it could be observed that both the ordinary and extraordinary components of the refractive index and absorption coefficient increased with occlusion time as water built up in the surface of the skin while the properties of the epidermis remained relatively stable throughout occlusion, as shown in Fig. 23(b). They also extracted the dispersion and birefringence of the SC as a function of occlusion time and linked these variables to the hydration and structure of the skin respectively. They conclude that occlusion increases the amount of water in SC and decreases the inhomogeneity. This experimental configuration and model for the skin make it possible to learn about changes in the skin on a cellular level through noninvasive measurements; this opens the door to more opportunities to use THz measurements to learn about changes in the skin.

FIG. 23.

(a) The model of the skin used, showing the alternating lipid and corneocyte layers; and (b) the changes in the refractive index and extinction coefficient of the ordinary and extraordinary components of the stratum corneum and of the epidermis throughout 31 min of occlusion. Reproduced with permission from Chen et al., Adv. Photonics Res. 2, 2000024 (2021).84 Copyright 2021 Authors, licensed under a Creative Commons Attribution (CC BY) license.

FIG. 23.

(a) The model of the skin used, showing the alternating lipid and corneocyte layers; and (b) the changes in the refractive index and extinction coefficient of the ordinary and extraordinary components of the stratum corneum and of the epidermis throughout 31 min of occlusion. Reproduced with permission from Chen et al., Adv. Photonics Res. 2, 2000024 (2021).84 Copyright 2021 Authors, licensed under a Creative Commons Attribution (CC BY) license.

Close modal

3. Monitoring burns and scars

There has also been significant interest in the ability of THz imaging to objectively assess and classify the severity of burns. The links between burn severity and the dehydration of the tissue and the ability of pulsed THz techniques to identify different layers in the skin suggest that THz imaging could be very useful in this field. Additionally, THz techniques have high sensitivity to water content of tissues while materials, such as bandages and gauze, are nearly transparent, meaning that burns could be viewed through a bandage. This section summarizes some of the spectroscopic and imaging techniques which have been tested as potential approaches for burn assessment.

Tewari et al. measured the THz reflectivity of the skin of a rat up to 7 h after the tissue was burned.315 A 35 × 35 mm2 image was obtained, which took less than 8 min to acquire, with the region maintaining contact with an imaging window throughout the measurement. They calculated the water content of the skin using the stratified media model of the skin with an EMT. These measurements were performed every 15 min for the first hour and then every 30 min subsequently for up to 7 h following the injury. Figures 24(a)–24(d) show THz images of the change in water content of the skin before the burn and increasing time points after a thermal contact injury with a cross shape, where the lighter regions indicate regions of higher water content. The study revealed an initial increase in the water content in the region local to the burn which is associated with the inflammatory response of the skin. Following this, a region of increased water content was observed directly where the skin was injured, surrounded by a region of decreased water content. This suggests that THz imaging has the potential to provide insight into the behavior of skin following injuries, such as burns, and provide guidance on the correct methods and timescales to use when providing treatment.

FIG. 24.

(a)–(d) THz images of the water content of rat skin increasing time following a burn, reproduced with permission from Tewari et al., J. Biomed. Opt. 17, 040503 (2012).315 Copyright 2012 SPIE. (e) THz images of the refractive index (left) and absorption coefficient (middle) at 1 THz of a scar six months following the injury, compared to the visible image (right) of the same region. Reproduced with permission from Fan et al., J. Biphotonics 10, 1143–1151 (2017).316 Copyright 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (f) Images of partial and full thickness burns of the THz reflectivity and the THz reflectivity combined with a visible image of the region increasing time after a burn. Reproduced with permission from Tewari et al., Biomed. Opt. Express 10, 322–337 (2019).317 Copyright 2019 Optical Society of America.

FIG. 24.

(a)–(d) THz images of the water content of rat skin increasing time following a burn, reproduced with permission from Tewari et al., J. Biomed. Opt. 17, 040503 (2012).315 Copyright 2012 SPIE. (e) THz images of the refractive index (left) and absorption coefficient (middle) at 1 THz of a scar six months following the injury, compared to the visible image (right) of the same region. Reproduced with permission from Fan et al., J. Biphotonics 10, 1143–1151 (2017).316 Copyright 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (f) Images of partial and full thickness burns of the THz reflectivity and the THz reflectivity combined with a visible image of the region increasing time after a burn. Reproduced with permission from Tewari et al., Biomed. Opt. Express 10, 322–337 (2019).317 Copyright 2019 Optical Society of America.

Close modal

Following this study, the group identified water as a source of contrast in these images and confirmed that THz reflection measurements are sensitive to the movement of water in the skin following a thermal contact injury. This was done by comparing THz reflectivity images of full and partial thickness burns on rats to T2 weighted MRI measurements of the same regions. In this study, the spatial-temporal contrast displayed in the image series is more important than the reflectivity of individual pixels. This contrast provides sufficient information for burn depth/severity characterization. These studies find that there is good correlation between the two imaging techniques for both types of burn.318,319

There has also been work by Tewari et al. to explore the combining of THz images with optical images of the burn regions.317 This is done using two markers which are visible at THz and optical frequencies, making it possible to calibrate the two images. Such an approach makes it possible to more accurately link the features in the THz image to the location on the skin relative to the burn. In the study, nine rats received either a full or partial thickness burn and the burnt area was imaged continuously for 435 min following the injury, with the region remaining in contact with an imaging window for the duration of this time. Figure 24(f) shows the images of the full and partial thickness burns increasing times after the injury at THz frequencies alone and an optical image combined with the THz results. In the THz images, the reflectivity is plotted with the colormap where the light regions show regions of increased reflectivity indicating increased tissue water content compared to the surrounding darker regions. Moreover, the concentric-like spatial distribution of water relative to the injured region correlates with the known concentrically arranged damage zones in burn injuries. The results revealed the dynamic changes in the THz response of the tissue around the burnt region, which could be an indication of the movement of water in response to the injury. It is possible that the prolonged contact between the skin and imaging window also had an impact on the water distribution throughout the study; this effect is explained in more detail in Sec. V D 4.

Arbab et al. explored an approach to the application of THz spectroscopy for the classification of burns.140 Unlike the approach explained above which used an incoherent imaging technique, this study used THz-TDS to measure the spectral response of a single region of a burn. They used nine rats in their study, five of which were inflicted with a second degree burn and four with a third degree burn; regions left uninjured are also marked on each rat to act as a control region. Both regions are measured on each rat immediately following the burn and 72 h following the burn. The results were compared to histology results obtained following the excision of the burnt area at the end of the study. They used an image processing technique to define a quantitative variable, the density of skin structures (DOS), to characterize the histology results. They also define a spectral variable Z, which is used to classify the burn from the measured THz reflectivity spectrum,

Z=a·R+b·S,
(5)

where R is the average reflectivity across the measured frequency range and S is the spectral slope of the reflectivity. a and b are arbitrary constants optimized to maximize the specificity of Z. Using these parameters, they demonstrate the ability of THz spectroscopy to identify healthy and burnt skin. They also find good correlation between DOS and Z with a correlation coefficient value of −0.797 for the results obtained 72 h post-injury.

This group then expanded upon the study by performing measurements on porcine skin, as this is thought to be a better model for human skin.320 Three pigs were measured and on each pig three regions were burnt for increasing durations to increase the severity and depth of the burn. The results of the THz measurements were compared to the current gold standard for burn depth assessment: vimentin immunohistochemistry (IHC). As before, they defined a spectral parameter Z to characterize the burns from the THz measurements,

Z=a·SA+b·SS,
(6)

where SA is the spectral area and SS is the spectral slope. They found that there was good correlation between the burn depth and the spectral parameter Z, suggesting that THz measurements could be an effective way to quickly assess the severity and depth of a burn. However, it should be recognized that the dermis layer, which is partially or fully damaged in the second- or third-degree of burn injuries respectively, has a thickness of 1–4 mm, where the THz waves becomes considerably weak to interact with. Therefore, the attenuation of water may fundamentally limit the sensitivity in evaluating burn severity. It is hoped that this work could be used to develop THz spectroscopy into a technique that can predict healing outcomes and identify which treatment option would be most effective.

Fan et al. found in their study that the ability of THz imaging to identify disruption in the skin is not limited to thermal injuries, such as burns but can also include scars caused by abrasion.316 They measured four human scars, three of which were old scars which had finished the healing process while the last one was a recent scar which was measured at multiple time points so that the healing process could be observed. By producing THz images of the refractive index and absorption coefficient, they found contrast in both variables. They found that as the recent scar healed, the contrast in the absorption coefficient diminished while some contrast in the refractive index remained as shown in Fig. 24(e); this was attributed to the deposition of collagen leading to structural changes in the tissue causing the scar. They were also able to observe the scar at two weeks post injury through a plaster: they found they could still identify the contrast in the measured impulse function of the skin.

4. Parameters affecting the THz response of skin

The strong sensitivity of THz radiation to the water content of tissues can be a challenge as well as a strength of the technique as there are many variables which can influence skin hydration. It is important that these variables are carefully controlled in order to obtain repeatable results, which are meaningful and can be compared between different subjects.

As mentioned in Sec. V D 1, the occlusion of the skin by an imaging window prevents the natural loss of water from the surface of the skin resulting in an increase in the water content in the SC throughout a measurement. Sun et al. conducted a systematic review of other variables with the potential to contribute to the observed decrease in the amplitude throughout the measurement including the movement of the imaging window, thermal expansion of the quartz window, and the buildup of a perspiration layer. The study confirmed that the effect of these variables was negligible and identified the buildup of water in the SC as the dominant factor. A key challenge of acquiring an image of the skin is that the optics must be raster scanned across the whole region, meaning that there is a long time delay between the measurement of the first point and the last point. Figure 25(a) shows the variation in the THz image of the skin with increasing time into occlusion. Sun et al. proposed that the biexponential model could be used to compensate for the effect of occlusion from THz images of the skin. In addition to removing the effect of occlusion from the results, there is also potential that the occlusion curves of skin could be used to identify abnormalities, such as the presence of cancer beneath the skin surface, which may affect the water dynamics in the skin.

FIG. 25.

(a) Images of the same region of skin with increasing time into occlusion, showing the decrease in peak to peak amplitude with increasing occlusion time. Reproduced with permission from Sun et al., J. Biophotonics 11, e201700111 (2018).108 Copyright 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (b) THz images of the skin acquired under different contact pressures between the skin and the imaging window. Reproduced with permission from Wang et al., Biomed. Opt. Express 9, 6467 (2018).109 Copyright 2018 Optical Society of America.

FIG. 25.

(a) Images of the same region of skin with increasing time into occlusion, showing the decrease in peak to peak amplitude with increasing occlusion time. Reproduced with permission from Sun et al., J. Biophotonics 11, e201700111 (2018).108 Copyright 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (b) THz images of the skin acquired under different contact pressures between the skin and the imaging window. Reproduced with permission from Wang et al., Biomed. Opt. Express 9, 6467 (2018).109 Copyright 2018 Optical Society of America.

Close modal

In addition to occlusion, another variable which will affect the THz response of skin is the contact pressure between the skin and the imaging window. Wang et al. investigated the impact of pressure by incorporating pressure sensors into the THz system and measured the THz response of skin at five different pressures.109 The study found that the reflectivity of skin decreased by up to 30% as the pressure increased, while the refractive index increased by up to 10%. Images of the peak-to-peak amplitude at different contact pressures are shown in Fig. 25(b), and the clear effect of pressure on the measured THz response of skin can be seen. This was attributed to an increase in hydration and an increase in the density of the biological tissues. The results from the skin modeling confirmed that the thickness of the SC decreases as pressure increases and the surface hydration of the SC increases.

To account for the observations that occlusion time and contact pressure influence the THz response of skin, Lindley-Hatcher et al. compiled a robust protocol for in vivo skin measurements using a reflection THz system.110 They also introduced a variable called the normalized relative change, which can be used to remove the effects of natural variation of the skin between measurements by subtracting the variation in the control region from the variation in the treated region. They found that by using this protocol and processing technique they were able to obtain repeatable measurements of the changes in human skin following the application of a commercial moisturizer.

These initial studies have identified some of the parameters which can influence the measured THz response of skin and have quantified the magnitude of these changes. Other parameters which can also have an impact include the environmental conditions, such as temperature and humidity. It is important that these parameters are controlled as effectively as possible by the experimental protocol if in vivo THz measurements are to give robust and repeatable results.

5. Corneal assessment

In addition to imaging the skin, THz radiation has been tested as a potential technique for assessing corneal hydration, as the eye can be accessed without passing through any other tissues. The first in vivo images of the cornea at THz frequencies were obtained by Taylor et al. who measured the corneas of five rabbits using an imaging window to flatten the curve of the tissue.57,302 The corneas were measured every 10–15 min over a period of 2 h while the eye was exposed to a stream of air to dehydrate the eye. In addition to the THz measurements, the central corneal thickness was measured using ultrasound; this can be used to approximate the corneal water content; however, the accuracy of such a conversion is questionable. Very little correlation was found between the THz reflectivity of the cornea and the measured corneal thickness. Following electromagnetic modeling, the study raises the question that the technique employed to dehydrate the eye using air may be altering the corneal thickness in isolation from the hydration leading to observable changes in the ultrasound results but not in the THz images.

The THz spectral signatures of the cornea observed arise from lossy longitudinal modes, it is assumed that these exist due to the uniform structure of the cornea comprised of a thin collagen layer backed by water. The lossy nature of these modes means that the upper limit of the usable illumination band for studying the cornea may not extend too much beyond the millimeter wave region. A subsequent study by Chen et al. applied the more appropriate diffusion model to represent water content gradients.321 Corneal tissue water content gradients have been confirmed in cornea. Castoro et al. observed a substantial bound water component in ex vivo bovine corneal samples along the optical axis.322 This suggests that the cornea cannot be modeled using only free water components.

Following on from this initial study, the group advanced their approach further by developing a noncontact imaging technique for the cornea, as mentioned in Sec. II B 4, which takes advantage of the limited variation of the shape of the cornea between subjects and the relatively uniform geometry.111 The move to noncontact imaging not only removes the risk of the imaging window interfering with the corneal hydration during the measurement, it also increases the applicability of such an approach for in vivo human measurements as it is more comfortable than maintaining contact between the eye and an imaging window for the duration of the measurement. Using this system, Sung et al. were able to acquire the first noncontact THz images of the human cornea of a single subject and subsequently plot the THz reflectivity map of the cornea. This study highlights the potential of THz imaging for checking the health of the eye; however, they also encountered challenges with controlling the movement of the subject. There are practical difficulties with in vivo measurements which must be dealt with before quick, repeatable noncontact measurements of the eye can be obtained with such a system as part of routine eye checks.

Tamminen et al. produced hydrogel-based corneal phantoms to test the efficacy of THz spectroscopy for corneal assessments and validated the accuracy of the results obtained.73 Using this approach, they were able to extract the absolute thickness of the cornea and verify the results by comparing the results with measurements of the central corneal thickness obtained with OCT, which is the present gold standard for corneal assessment. The technique was also able to provide absolute estimates of the anterior and posterior water content of the corneal phantom and characterize phantom sphericity. It has also been proposed that THz spectroscopy could be used for sensing increases in intraocular pressure (IOP), in the recently obtained patent.323 The first experimental demonstration of correlation between THz reflectivity and IOP was reported in a previous study.121 Statistically significant increases in reflectivity between 0.4 and 0.6 THz corresponded to an increase in IOP. Endothelial layer damage, resulting from IOP increases, was confirmed using SEM of the corneal biopsy samples.

Additionally, work was performed by Ozheredov et al. to demonstrate the ability of THz imaging to observe the tear film in the human eye.72 By asking subjects to keep their eye open for as long as possible before blinking and measuring the THz reflectivity of the eye using a noncontact geometry, a decrease in the reflectivity of the eye as a function of time could be observed. This decrease is associated with the drying of the eye as the tear film declines; they also observe the recovery of the reflectivity following blinking; this suggests that the tear film is restored by blinking. This study was then expanded upon by performing a similar investigation on 29 subjects measuring both eyes, the gradient of the decay of the reflectivity between blinks was obtained using a linear fit.324 These results were used to categorize whether subjects had dry eye syndrome and compared to other approaches used to assess the health of the eye. It appears that subjects who reported suffering from dry eyes were more likely to have a greater decrease in the THz reflectivity of the eye between blinks. They also performed a study exploring the optimal model for describing the dehydration of the eye between blinks.325 By applying an EMT, they were able to show that an exponential decay is the function which best described the decay of the reflection coefficient between blinks, and they were able to learn more about the structural changes in the tear film layer and how this influences the penetration depth of the THz radiation.

6. Cancer diagnosis

There remains a high demand for a nonionizing imaging technique which can quickly identify the presence of cancer and accurately outline the tumor margins. THz imaging has the potential to fill this gap and significant progress has been made in investigating this possibility. The first in vivo study which demonstrated the ability of THz imaging to identify the presence of cancer was performed by Wallace et al. in 2004.288 As with many of the following studies, the focus was on skin cancer as it can be accessed and imaged noninvasively using THz reflection imaging. In the Wallace study, five subjects with basal cell carcinoma (BCC) on their arm were imaged, and THz images of the regions were produced. The study found a broadening of the THz pulse in the cancerous regions and an increase in the peak amplitude. There was good correlation between the THz images and the subsequent histological results. At the time, it was acknowledged that the origin of this observed contrast was not well understood, and an up to date summary of the current understanding of the origin of contrast is given in Sec. III C.

More recently Zaystev et al. conducted a series of studies in which they investigated the ability of THz measurements to identify the presence of skin cancer in vivo.326–328 In addition to the BCC examined by Wallace et al., they also studied the common precursor to melanoma: dysplastic skin nevi. They compared the dielectric permittivity of these dysplastic nevi to that of nondysplastic nevi which are more likely to remain benign. By plotting the Cole–Cole diagrams of these results, they were able to demonstrate that THz measurements were able to distinguish between healthy skin and dysplastic and nondysplastic skin nevi. It should be noted that these studies were performed with a relatively small sample size of four subjects.

While the primary focus of research into potential applications of THz imaging to cancer diagnosis has been on skin cancers, there has also been some work demonstrating the contrast between healthy and cancerous tissues in other regions. These studies lose the advantage of being noninvasive imaging techniques, but may serve as an assisting operating examination tool. As mentioned, many current THz in vivo measurements are still based on a window-tissue raster scan modality. This can be seen from the early example of in vivo BCC imaging by Wallace et al.,288 as well as the more recent in vivo studies of rat-brain tumors154,309 described in the following paragraph. Such a modality is not ideal for invasive measurements, as a noncontact and fast imaging system would be preferable for aiding cancer removal and ensuring that the correct margins are used to define the tissue extraction.

One example of invasive in vivo THz measurements of cancer is the application to the identification of brain gliomas. Ji et al. performed an ex vivo study of the sensitivity of THz measurements to the presence of gliomas in brain tissue of mice, as part of this they also measured a single mouse brain in vivo by creating a cranial window to access the brain.154 The study revealed that the THz image displayed a contrast between the healthy and cancerous regions; this was verified using MRI and fluorescence images of the same region and further ex vivo images following the extraction of the brain. Wu et al. extended this in their study in which 15 mice with brain gliomas were imaged in vivo and compared to the results from five healthy mice which underwent the same surgical procedures.309 The study compared the THz reflection images of the brain taken following the creation of a cranial window with MRI and visual images and showed good correlation in the cancerous regions identified. After the THz images had been acquired, the brain was extracted and a histological examination was also performed. Some of the results from this study are shown in Fig. 26, where the cancerous tissue can be seen in the THz image as the red region which corresponds well to the location of the black rings marked on the ex vivo optical image showing the location of the cancer. These studies identify the ability of THz imaging to reveal the presence of cancer in other tissue types; however, a noncontact measurement scheme with a fast-imaging speed is needed for clinical applications.

FIG. 26.

Brains of mice with glioma (1–3) and a healthy mouse (4) imaged in vivo with MRI (a), visible light (b) and THz (c) and ex vivo images (d), and histological results (e) of the same regions. Reproduced with permission from Wu et al., Biomed. Opt. Express 10, 3953 (2019).309 Copyright 2019 Optical Society of America.

FIG. 26.

Brains of mice with glioma (1–3) and a healthy mouse (4) imaged in vivo with MRI (a), visible light (b) and THz (c) and ex vivo images (d), and histological results (e) of the same regions. Reproduced with permission from Wu et al., Biomed. Opt. Express 10, 3953 (2019).309 Copyright 2019 Optical Society of America.

Close modal

Finally, it should also be noted that in a few specific cases in vivo THz measurements have been performed using a transmission geometry rather than using reflection. Chen et al. used a transmission geometry to image human breast cancer which had been placed inside the subcutaneous tissue of a mouse.329 Mouse skin is very thin and elastic so it could be stretched thin enough that the transmission measurement could be performed and it was found that the cancerous region could be identified from the THz image. However, it is questionable how such an approach could be transferred to in vivo human measurements. It is useful to observe that breast cancer can be observed in vivo even under such conditions, and that the sensitivity shown in THz images in previous studies is not unique to the ex vivo conditions used.

7. Clinical solutions

There are many more potential applications for THz imaging and spectroscopy that have been explored. A brief summary of some of the recent advances will be addressed here. Once such application for THz imaging which has been investigated is the assessment of reconstructive skin flaps to determine if they will be successful. To do this, Bajwa et al. performed THz measurements on six rats, three of which had a flap of skin fully excised then returned to the region and held in place with sutures.330 The other three rats had the region partially excised leaving two sides of the square of tissue intact; the tissue was raised from the skin surface then returned to its original position. These regions were measured with THz radiation 24 h, 48 h, and 7 days following this procedure, and THz reflectivity images of the regions were produced. At these time points, the regions were also clinically assessed to evaluate the health of the tissue. At the end of the study, the regions were removed, and a histological examination was performed. It was found that the fully excised region showed a significant contrast in the THz image from the surrounding healthy tissue 24 h following the procedure; such differences were observable in the clinical examination 48 h following, as shown in Fig. 27. Upon histological examination, it was confirmed that this tissue was necrotic. However, no such contrast was visible in the THz images of the partially excised regions. This suggests that THz images are able identify the drop in tissue water content, indicating that a skin graft is not successful prior to a clinical examination of the region. This could be a useful tool for rapid quantitative assessments of skin health and indicate the potential success of reconstructive surgeries at earlier stages.

FIG. 27.

The visible and THz images of fully and partially excised skin flaps, respectively, increasing time following the procedure. Reproduced with permission from Bajwa et al., Biomed. Opt. Express 8, 460 (2017).330 Copyright 2016 Optical Society of America.

FIG. 27.

The visible and THz images of fully and partially excised skin flaps, respectively, increasing time following the procedure. Reproduced with permission from Bajwa et al., Biomed. Opt. Express 8, 460 (2017).330 Copyright 2016 Optical Society of America.

Close modal

THz imaging has also been investigated as a potential method for the objective diagnosis of diabetic foot syndrome. Diabetic foot syndrome is common in patients with diabetes and can cause ulceration on the foot which in many cases results in partial or total amputation of the limb. Presently, there is no gold standard technique for diagnosis and currently used methods are very subjective. Hernandez-Cardoso et al. constructed a setup to perform THz reflection measurements on the foot including a chair for the subject to sit on while their foot was in contact with the imaging window, shown in Fig. 28(a).331 In an initial study, they measured the right foot of 33 healthy volunteers and 38 subjects diagnosed with diabetes; from these images, they could use 21 from the control group and 12 from the diabetic group, as some subjects struggled to maintain stable contact with the imaging window for the duration of the measurement. They extracted the water content of the skin using the LLL EMT and produced an image of the foot with each pixel representing the water content in that location. It is clear from these results that subjects in the diabetic group had significantly drier skin, as shown in Figs. 28(b)–28(f). As part of the further analysis of these results, they found that by using a three color coding system for the pixels based on the water content, it was possible to identify a potential threshold for diabetic foot diagnosis.332,333 They coded pixels with water content above 55.8% as green, those between 47% and 55.8% as yellow, and those below 47% as red. They defined a threshold parameter as

GR=%GreenPixels%RedPixels.
(7)
FIG. 28.

Assessing diabetic foot syndrome with THz imaging: (a) the experimental setup used to image the foot with the skin in contact with an imaging window; (b) THz images of the water content of the feet of healthy and diabetic subjects; and (c) the distribution of the water content in the different foot regions for healthy and diabetic subjects. Reproduced with permission from Hernandez-Cardoso et al., Sci. Rep. 7, 42124 (2017).331 Copyright 2017 Authors, licensed under a Creative Commons Attribution (CC BY) license.

FIG. 28.

Assessing diabetic foot syndrome with THz imaging: (a) the experimental setup used to image the foot with the skin in contact with an imaging window; (b) THz images of the water content of the feet of healthy and diabetic subjects; and (c) the distribution of the water content in the different foot regions for healthy and diabetic subjects. Reproduced with permission from Hernandez-Cardoso et al., Sci. Rep. 7, 42124 (2017).331 Copyright 2017 Authors, licensed under a Creative Commons Attribution (CC BY) license.

Close modal

Plotting this variable for diabetic and nondiabetic subjects, they found that all healthy subjects had GR>18% whereas 82% of diabetic subjects had GR<18%. The most recent results on this front334 dealt with a much larger cohort that includes 98 nondiabetics and 80 diabetics. This is probably the largest human population ever imaged with THz for medical research purposes. The diabetics were classified according to their clinical history as well as neurological and vascular assessments of their feet. First, the new study shows that patients with a history of ulceration or amputation or with peripheral neuropathy can be distinguished from those without these problems using the THz technique. Second, it provides clear indication that the skin dehydration of the feet of diabetic patients is related to the neurological “feedback loop” that regulates the hydration of the feet, rather than the irrigation. It should be noted that this is the only known application presently under research in which the pressure applied to the region by the imaging window mimics the normal environmental conditions for the tissue being tested, making this application unique among THz in vivo investigations.

Finally, there has also been some speculation on the ability of THz measurements to identify changes in blood glucose levels. For example, Cherkasova et al. investigated the sensitivity of THz measurements to glycerol when injected, ingested, or topically applied to the skin.335,336 They measured some changes in the skin following the increase in blood glucose. Additionally, Chen et al. performed a series of near-field measurements on capillaries in the ears of diabetic mice.95 They were able to identify the capillaries in the ear of a mouse from the image by the higher absorption coefficient value in the capillary compared to the surrounding tissue. They monitored the changes in the THz images of the ears of diabetic mice and observed changes in the measured absorption coefficient. However, in these studies, it is not possible to isolate the observed changes to be solely linked to changes in the blood glucose. These approaches to blood glucose measurement still lack the accuracy and specificity to replace current techniques and new methods need to be developed to enable THz devices to measure blood which is often too deep below the surface of the skin to be accessed by THz radiation.

It should also be noted that work has recently been performed exploring the potential of THz spectroscopy to differentiate between diabetic and nondiabetic blood samples by processing the blood to form blood plasma pellets.337,338 A change in the refractive index of around 10% was observed between pellets from diabetic and nondiabetic subjects. This approach enables the removal of the effect of water from the THz spectra while also producing samples which can be stored at room temperature for a long time, making it possible to transport and/or repeat the measurements in the future if needed.

8. Passive in vivo imaging

It should also be mentioned that the potential of THz imaging has been investigated using the THz radiation produced by the body itself, rather than using a THz emitter to illuminate the sample and measure the reflected response, an image of the region of interest is taken at THz frequencies. For example, Berlovskaya et al. measured the background THz radiation emitted by subjects to obtain IR-THz images and explored the possibility of using these images to identify the psychoemotional state of a subject using noncontact imaging.339 They measured 38 subjects in cases of electrical stimulation, cognitive load, and physical exercise. Through the classification of the pixels in the image into different clusters, they could observe differences in the stress responses of different subjects; some appeared to have an increased blood supply under stress while others had a decrease. They were also able to observe the periodic patterns in the region around the nose caused by breathing.

Another study was performed using a THz detector to obtain measurements of the radiation emitted by the human body in the frequency range of 0.1–1.5 THz.340 The system was raster scanned to acquire an image of the human hand. In the image, the fingers can be clearly observed as there was sufficient sensitivity provided by the detector despite the low temperature contrast between the human body and the room temperature background.

Owda et al. performed a series of simulations of the emissivity of human skin in the frequency range from 30 to 300 GHz and found that the emissivity of skin is expected to decrease with increasing skin hydration.341 They then measured the skin of 30 subjects in four regions; the inner and outer wrist and the volar and dorsal forearm. The study revealed differences in the measurements of male and female skin and also in the different regions measured. This study was extended further as 60 subjects were then measured in six regions on the arm and hand using the same approach to passive sensing.342 They extended the parameters being investigated to also include ethnicity and body mass index. This technique has potential for being used in security as it is able to measure the emissivity of human skin and would be able to identify regions which have abnormalities indicating the presence of concealed objects.

As discussed in Sec. III, most biosamples exhibit no absorption fingerprints in the THz region; hence, there is usually no distinct feature for convenient sample classification or identification. Separation between two different sample classes usually relies on small dielectric differences at some particular frequencies. Fortunately, mainly promoted by the commercialized THz-TDS systems, broadband THz measurements are now widely available, which provide multi-dimensional data in the time, frequency, and even spatial domains. These data not only offer comprehensive information, but also introduce difficulties in data analysis. For example, the refractive indices or absorption coefficients at particular frequencies, the amplitude in the time domain, and the pulse position can all potentially provide a contrast between two sample classes. It is difficult to decide which parameters should be used without an accurate prior knowledge. Advanced data analysis comprehensively taking this high-dimensional information into account is in high demand to achieve a better accuracy.

Usually, the data analysis for THz biomedical research is a binary classification problem, such as deciding whether a signal is reflected from the cancerous area or the normal area.295 Multi-class classification may be required in some cases, such as determining the degree of cancer or injuries.61 For continuous variations, such as the concentration of protein solutions, it becomes a regression problem.343 In either case, the essential data processing typically contains three steps: (a) data acquisition; (b) feature extraction; and (c) classification, as depicted in the flowchart in Fig. 29. We overview the recent investigations which apply these techniques for THz biomedical research. Table II summarizes the system used for data acquisition, feature extraction methods, classifiers, and their applications. For the data acquisition, most measurements are based on THz-TDS systems owing to the advantageous broadband information.344–346 A VNA has also been used, which also offers a broadband signal in the frequency-domain.347 Using these systems, refractive indices or absorption coefficients which represent the sample properties independent of the measurement setup can be used for further processing. Time-domain amplitudes, transmission spectra, and other measured values can also be utilized, but they are specifically related to the measurement configuration. If imaging was performed, useful information may also be found from the spatial features. However, usually due to the limited number of samples and the small field of view, spatial information is rarely used. For THz images scanned by a broadband system, we generally look for features in each pixel that contain a full waveform. This is different from the classification for visible pictures, which is mainly based on the spatial patterns.

FIG. 29.

Flowchart of the common approach to data analysis for THz biomedical applications.

FIG. 29.

Flowchart of the common approach to data analysis for THz biomedical applications.

Close modal
TABLE II.

Methods and algorithms used for different biomedical applications.

Data acquisitionFeature extractionClassificationApplicationYear, Reference
THz-TDS magnitude spectra ⋯ SVM, DNN Glucose/lactose classification 2020, Li et al.344  
THz-TDS refractive index ⋯ Morphological dilation + thresholding Breast carcinoma delineation 2021, Cassar et al.348  
VNA features of different domains ⋯ SVM, kNN, decision tree Water content in leafs 2019, Zahid et al.347  
THz-TDS absorption spectra PPCA ⋯ Exhaled air from diabetes mellitus patients 2020, Kistenev et al.349  
THz-TDS absorption spectra PPCA ⋯ Melanoma and nevus tissue 2020, Kistenev et al.346  
THz-TDS absorption spectra PPCA SVM Prostate cancer grading 2020, Knyazkova et al.350  
THz-TDS transmission spectra PPCA SVR BSA concentration 2018, Sun et al.343  
THz-TDS ESER PPCA SVM, kNN, ensemble Breast invasive ductal carcinoma 2020, Liu et al.351  
THz-TDS absorption spectra PPCA + t-test K-means, SVM Human gastric normal and tumor tissues 2014, Hou et al.295  
THz-TDS17 parameters PPCA + decision tree NN, SVM Colon cancer diagnosis 2013, Eadie et al.352  
THz-TDS refractive index and absorption spectra MIC + PPCA RF, Adaboost Mouse liver injury 2019, Huang et al.353  
THz-TDS absorption spectra PPCA, t-SNE SVM, RF, NBM, XGBoost Protein conformation 2020, Cao et al.354  
THz-TDS absorption spectra HMFA ⋯ Gastric cancer 2018, Cao et al.345  
THz-TDS complex spectra SIT ⋯ Metastatic lymph nodes 2017, Park et al.355  
CW THz imaging intensity ReliefF SVM, kNN, RF Traumatic brain injury 2018, Shi et al.61  
Data acquisitionFeature extractionClassificationApplicationYear, Reference
THz-TDS magnitude spectra ⋯ SVM, DNN Glucose/lactose classification 2020, Li et al.344  
THz-TDS refractive index ⋯ Morphological dilation + thresholding Breast carcinoma delineation 2021, Cassar et al.348  
VNA features of different domains ⋯ SVM, kNN, decision tree Water content in leafs 2019, Zahid et al.347  
THz-TDS absorption spectra PPCA ⋯ Exhaled air from diabetes mellitus patients 2020, Kistenev et al.349  
THz-TDS absorption spectra PPCA ⋯ Melanoma and nevus tissue 2020, Kistenev et al.346  
THz-TDS absorption spectra PPCA SVM Prostate cancer grading 2020, Knyazkova et al.350  
THz-TDS transmission spectra PPCA SVR BSA concentration 2018, Sun et al.343  
THz-TDS ESER PPCA SVM, kNN, ensemble Breast invasive ductal carcinoma 2020, Liu et al.351  
THz-TDS absorption spectra PPCA + t-test K-means, SVM Human gastric normal and tumor tissues 2014, Hou et al.295  
THz-TDS17 parameters PPCA + decision tree NN, SVM Colon cancer diagnosis 2013, Eadie et al.352  
THz-TDS refractive index and absorption spectra MIC + PPCA RF, Adaboost Mouse liver injury 2019, Huang et al.353  
THz-TDS absorption spectra PPCA, t-SNE SVM, RF, NBM, XGBoost Protein conformation 2020, Cao et al.354  
THz-TDS absorption spectra HMFA ⋯ Gastric cancer 2018, Cao et al.345  
THz-TDS complex spectra SIT ⋯ Metastatic lymph nodes 2017, Park et al.355  
CW THz imaging intensity ReliefF SVM, kNN, RF Traumatic brain injury 2018, Shi et al.61  

The goal of the second step is to extract features that can maximize the difference between two/multiple targeted classes. This step can be skipped by directly sending some measured values or sample properties to the classifier without further extraction. For example, Li et al. used the spectral magnitudes in the whole effective bandwidth to classify glucose and lactose spectra under various imperfect conditions, in order to remove the influence from the environment.344 Cassar et al. used the refractive index at 0.55 THz to differentiate between benign and malignant breast carcinoma, making the classification straightforward by directly thresholding the refractive index with morphological dilation.348 Zahid et al. compared the classifying accuracy for features selected from the time domain, frequency domain, and time–frequency domain, respectively, showing that they all provided a good accuracy for water content estimation in living plant leaves.347 In these studies, directly taking the raw features had the advantage of a clear physical meaning. For example, if two classes can be differentiated by the transmission, the contrast is known to be from a difference in their transparency. However, sending large dimensional data for classification has a poor efficiency, which complicates the algorithm and increases the computational difficulties. Reducing the dimension by selecting some of the values requires prior knowledge on the origin of contrast and may miss other useful features. A more widely used strategy is to perform an additional feature extraction process to the raw data. This step extracts features with the greatest contrast and reduces the number of parameters. Principle component analysis (PPCA, to be distinguished from PCA referring to photoconductive antenna) is the most widely used method. In PPCA, a collection of input features is decomposed by a few orthogonal vectors (i.e., principle components) to describe the major variations of the data. By projecting the input data to the linearly uncorrelated principle components, the dimension can be efficiently reduced while the most informative features are kept. For example, applying PPCA to the absorption coefficient or transmission spectra extracts the key attenuation features. Examples can be found in the classification of exhaled air from patients with diabetes mellitus and healthy volunteers,349 as well as classifying tissues of melanoma and nevus,346 normal and prostate cancer,350 and different concentrations of bovine serum albumin (BSA) solutions.343 Other measured values can also be dimensionally reduced by PPCA. For example, Li et al. introduced wavelet transformation as the alternative to the Fourier transform due to its merit of multi-scale resolution for nonstationary signals.351 They adopted the energy to Shannon entropy ratios (ESERs) from the wavelet packet transform for PPCA. In some experiments, PPCA may still offer too many parameters for the classification. To further reduce the dimension, Hou et al. used a t-test method to evaluate the significance of each individual principle component, and found the first four important components provide the best performance.295 Similarly, Eadie et al. applied the decision tree method, which was originally designed for classification, to evaluate the importance of the features extracted from PPCA.352 They finally selected eight parameters for the classifications. Huang et al. used another method of maximal information coefficient (MIC) to select the frequency points that provide the best contrast first, before applying PPCA.353 MIC is a kind of the method which can find the relationship between two variants.356 They showed a higher accuracy compared to the results without applying MIC. Apart from PPCA, features can also be extracted by other methods. For example, Cao et al. compared the PPCA with t-distributed stochastic neighbor embedding (t-SNE).354 They claimed that the t-SNE method is better at dimensional reduction for spectral data. The results showed the outperformed accuracy in classifying the protein conformation states by t-SNE. Modified hard modeling factor analysis (HMFA) is an unmixing method decomposing the absorption spectrum into combinations of multiple weighted peak functions, thus the broadband absorption is represented by the weights of the discrete peaks.357 This method was applied to extract the absorption features of the gastric cancer samples.345 Park et al. used spectroscopic integration technique (SIT) such that each signal is represented by a single feature value. They showed an enhanced contrast in different levels of metastatic lymph nodes.355 Shi et al. imaged different degrees of traumatic brain injury by a CW THz system. The gray histogram of the region of interest was taken from the measurement, and the features were extracted by ReliefF algorithm for further classifications.61 

With the features from the measured values or the extraction methods, it is possible to do classification manually by comparing the tested sample with the statistical results. However, this is usually very difficult and subjective for high-dimensional data. A classifier is generally required for an automatic and accurate classification. Machine-learning (ML) based methods have become very popular in the recent decade, due to their successful application in many other fields especially for image recognition. ML can be classified into two types: supervised and unsupervised learning methods. The former type requires a manually labeled dataset to train or “supervise” the algorithm, such that the algorithm is built to maximize the difference between different targeted classes. This method is more commonly used in THz applications since a prior knowledge about the class of the measured sample is usually available. The major concern for supervised learning is that the sample size should be large enough for a generalized and accurate training and testing. Unsupervised learning automatically clusters the unlabeled data without human involvement. The best merit is that training is not required. However, unsupervised learning methods output an unknown number of classes, and the algorithm is usually more complicated.

Among various supervised ML methods, support vector machine (SVM) is most widely used. SVM maps the low-dimensional input data into a high-dimensional space where an optimal hyperplane boundary separating the two classes can be constructed. Successful classifications of glucose and lactose,344 human gastric normal and tumor tissues295 and traumatic brain injury,61 etc., can be found in Table II. Most of these applications show accuracies higher than 90%, demonstrating reliability suitable for THz biomedical applications. From the simplicity perspective, the method of k-nearest neighbor (kNN) could be the best choice which has also been widely adapted. It classifies a tested sample by a plurality vote of its k-nearest neighbors in the training set and decides its class by the most common class observed. The comparison with SVM shows similar performance in various applications.61,347,351 Random forest (RF) is another supervised ML method by using multiple decision trees.358 The class of a tested sample is decided by the majority class of the multiple trees. Therefore, it usually provides a better accuracy than a single decision tree. Applications in mouse liver tissues,353 traumatic brain injury,61 and protein conformation evaluation354 have all shown good accuracy. Many other methods are available, such as deep neural network (DNN) or neural network (NN),344,352 support vector regression (SVR),343 Adaboost,353 decision tree,347 naive Bayes model (NBM) and extreme gradient boosting (XGBoost),354 Ensemble,351 etc. They have all demonstrated high precision in different applications.

The examples discussed present the promising applications of advanced data analysis, especially using ML methods in THz biomedical applications. They provide a solution to extract the contrast between two targeted classes by considering high-dimensional features. However, a few improvements should be made in the future. First, a larger dataset should be constructed for ML. Currently, many of the datasets are rather small which is not very convincing for generalization and application. On the other hand, a larger dataset will be more comprehensive to represent the features of the data, benefiting the downstream classifier on improving the performance. Second, the numerous feature extraction methods and classifiers applied in different applications make it confusing for people to decide which one should be selected. A topical study on the technical comparisons of these methods and algorithms is needed to provide general guidance on their usages in THz biomedical studies. A recent review on ML techniques for THz applications has been reported by Park et al.,359 which provides technical details and comparisons between different ML methods. In addition, data-driven methods like DNN or convolutional neural network which map the input directly to the output could be offered as a solution without the need of feature extraction. Third, although most of the classifiers achieve relatively high accuracy in classifications, an even better performance is needed for clinical applications. Some of the studies outlined above have compared multiple classifiers in a task to decide which one performs best. Actually, the performance can be further improved by combining two or multiple methods, with the decision made according to the majority. In any case, the decision of ML is made by a set of purely mathematical calculations. The algorithm performs a bit like a black box so care must be taken, particularly for biomedical applications, so that interpretable results are produced. For example, differentiating samples simply by their different absorptions can be explained by their different water concentration. However, differentiating them by a set of features with complicated rules is difficult to link to their physical properties. Therefore, being able to provide physical data interpretation is important to push it forward to clinical usage.

Despite significant progress being made on both the techniques and applications aspects of THz biophotonics, new challenges are emerging with the expansion of applications. In this section, we summarize four major challenges that remain for THz biophotonics: the issue of high water-absorption, the low flexibility of THz systems, the slow imaging speed, and the lack of methodology standards.

The high THz absorption by water is a big obstacle for THz technology to be used in biomedical applications. The penetration depth of THz light in skin (which is approximately 70% water by volume) is limited to a wavelength comparable level, namely, less than 500 μm at 1 THz.17 For comparison, traditional biophotonic techniques, such as OCT and diffuse optical spectroscopic imaging (DOSI),360 have a penetration depth of around 5 mm in skin and eyes. Increasing the SNR by using a brighter source can improve the detectable depth. Pickwell–MacPherson compared the penetration depth of skin, water, and adipose as a function of SNR and showed a near 1-mm penetration depth for skin when a SNR of 100 000 can be achieved.361 Using an intense source to alleviate the absorption issue for liquid samples has been demonstrated by Havenith's group. In their work, they used a p-Ge laser with a sufficiently high power to maximize the penetration distance and hence improve the accuracy. This motivates development of high-power THz sources to enhance the capability for absorptive nonliving samples. Indeed, solid-state devices and THz QCLs have both been improving their output powers at the same time as extending their higher/lower operating frequency limits. However, for in vivo applications thermal effects by intense sources can produce unwanted side effects in living biosystems, see Sec. IV.

THz spectroscopy is an important tool for exploring and understanding many fundamental molecular dynamics. It is most ideal to characterize the properties of the samples by mimicking their solvent environment; hence, the biggest challenge comes from the high absorption of water that makes the tiny changes induced by the molecules, such as the folding or conformational changes of proteins, difficult to measure precisely. The same issue was found for most reported biosensors, which have high-Q resonances in measuring low-loss samples but suffer significantly from the damping effect for absorptive samples. As a result, MM, resonant waveguides and THz SPPs usually only work as refractive index sensors. Designing configurations sensitive to high absorption samples could be another approach. We have shown the good sensitivity of the Si-ATR configuration to water in Sec. II B 1.76,77 A similar configuration has also been adopted in the study of hydration shells of molecules of different weights.216 Resonant sensors specifically designed for absorptive samples would also be very useful for detecting trace amounts of aqueous samples. Improving the accuracy of measuring highly absorptive liquids is a long-term but important topic that will have a great impact on many applications.

The high absorption by water means that measurements of in vivo tissues need to be performed in a reflection geometry. Most biological samples are “smooth” compared to the long wavelength of THz waves; thus, they behave similar to mirrors in that there is only specular reflection. The drawback is that unlike optical cameras that can easily capture the diffuse light without a precise illumination source, THz waves need to be measured accurately in the direction of the reflected beam and the detected signal will be very sensitive to small alignment variations. For in vivo measurements, a pre-aligned window making good contact with the measured area is usually used to mitigate against this, as it flattens the area being investigated and ensures there are no phase errors. As a result, most in vivo measurements have been limited to soft and hairless skin regions, such as the volar forearm, hands, and feet because they can be measured noninvasively with good contact with the imaging window. However, although there have been a few experiments of animal intraoperative measurements which are also based on a window-contact modality,154,309 diseases are seldom in body areas that can be flattened and contacted by a window. This has made it nearly impossible to investigate using present THz imaging systems for cancer detection. Therefore, configurations with high flexibility that can quickly adjust the alignment according to the measured surface profile are urgently needed. For example, the configurations introduced in Sec. II B 4 were developed for this purpose. Further improvements using advanced sensors, mechanical components, and algorithms are on the horizon. With increased flexibility, the areas accessible for THz in vivo measurements will be greatly extended. This is a necessary step to progress THz technology from laboratory-based investigations to clinical applications. Another potential solution is to use passive imaging techniques and detect the THz emission from the body itself. At the moment, such technology is tailored to security screening to image large targets with a low sensitivity, as the main purpose is to distinguish hidden metal objects. Technically it is difficult to achieve medical-level sensitivity because the thermal noise at THz frequencies generates a lot of electrical noise. Such systems are also bulky and have limited frequency outputs. For example, the TS4-SC by Thruvision weighs 24 kg with a size of 61 × 61 × 25 cm3 and measures the intensity at 0.25 THz only. This is currently far away from the capabilities needed to perform studies investigating the origin of THz contrast in diseased tissues. However, depending on the results from broadband studies with other systems, passive technology could potentially be tailored for biomedical applications in the future.

Developing fast THz-imaging technologies is an essential step before large-scale clinical studies can be performed. For example, close to real-time interpretation would be needed to enable the determination of cancer margins. This is because it is difficult to keep a patient sufficiently still for long periods of time and there is also a need to remove any time-induced variation on an image, such as occlusion. The acquisition time for certain parameters is longer than for others and also depends on the approach used to acquire the data. For example, if spectroscopic information is needed from a TDS source, it will take longer than if acquired from a CW source, but in the research phase, we do not know which frequencies will give the best contrast. Measuring a broadband response at this stage will help determine and optimize which sources and detectors can be used in the future. Recently, bolometer-array THz cameras working at room temperature over a broadband frequency have been combined with digital holography to provide amplitude and phase information,64,362 with a frame rate of 5 and 3.5 Hz, respectively. The ultimate limit of frame rate is set by the camera, which can reach a video rate for current commercial products, while in practice the speed and field of view are further restricted by the power of the source. This detector technology combined with a frequency selective source has the potential of being used in hospital settings. As a source, THz QCLs are potentially viable since they emit a single frequency that is tunable at discrete frequencies with suitable powers (>10 mW), with recent developments increasing the working temperature (see Sec. II A 2). However, it should be noted that the operating frequency of QCLs is usually too high (>2 THz); hence, the penetration depth is too small for biomedical applications. Commercial difference frequency photomixers emit powers of around 50 μW,363 which is too weak for such bolometer arrays.

Another possible technique is to use a single-pixel camera combined with a pulsed time-domain THz system (see Sec. II A 1).364 Although this technique obtains the image information in serial as opposed to in parallel, compared with detector arrays, Ref. 364 has achieved six frames-per-second real-time imaging of a 32 × 32 sample plotting the max amplitude. A follow-up publication showed that they can obtain a 32 × 32 image over a 10 ps time-interval in 2.6 s of acquisition,365 which is around ten times faster than current commercial THz time-domain imaging systems. In the optical regime, a single-pixel camera was able to display a 101 × 103 video with 72 frames-per-second in real-time;366 hence, this technology can certainly be fast enough with sufficient signal and a fast modulator. In this imaging modality, the fundamental acquisition speed is limited by the response time of the spatial modulator, and a recent overview of spatial THz modulators for single-pixel cameras is given in Ref. 98.

It is worth mentioning here that a new technology for CW single-wavelength imaging proposed by Weatherill's group.367,368 They utilized alkali atomic vapors in the Rydberg state, which refers the atoms being excited to very a high energy level where the energy levels are closely spaced, such that their difference could be equal to the energy of a THz photon. In their recent work,368 the detection is based on stimulating the emission of green light with THz photons. The most promising aspect of this technology is the very high efficiency that each THz photon has a 52.4% chance of being converted to a 535 nm photon. This provides the ability to achieve an ultrafast THz frame rate up to 3 kHz by using a low-power THz source of 17 μW. However, the emission mechanism means that the limitation of this technology is that it is only responsible for a narrow THz bandwidth for a specific Rydberg state.

The lack of commonly accepted measurement standards has resulted in diversity of the methodology, including the systems, configurations, experimental protocols, and processing methods. Even for the relatively standard THz-TDS transmission measurement of solid samples, divergence among different groups around the world was found in an international comparison study.369 Biomedical samples and applications are further complicated by having more variables than solid-phase materials, which undoubtedly makes it more difficult to establish common standards. Therefore, focused research on specific applications is needed to reduce the divergence gradually. For example, in skin studies there is no agreement in the modeling of the skin, and structures of single-layer, bilayer, bilayer with continuous water-content or birefringence have been applied. There are also control variables affecting the THz response. For sensing applications, a standardization of the techniques used to detect the quantities of interest is also needed. Some work has been done to address these issues. For in vivo skin measurements, there have been investigations into the optical model of the skin as well as how different control variables affect the THz response of skin.9,110 Similarly, the topical review on ML provides a subjective comparison among numerous ML algorithms.359 These efforts pave the way to improve the data accuracy as well as the comparability between different studies.

Finally, any bioeffects due to exposure to THz radiation at different powers will have to be considered. While THz photons are nonionizing, they do have effects on biological matter as discussed in Sec. IV. However, the exact mechanism of how THz radiation interacts with living biosystems is currently not experimentally established. The theoretical framework is that a Fröhlich condensate state is formed, and this perturbs the biochemical kinetics; however, there is only a single experimental study from 2015 that is thought to have observed such a Fröhlich condensate in single isolated lysozyme protein,191 which is far from a living organism. Without an adequate understanding of how THz waves affect biomatter, any attempt to establish safety protocols will be a precarious empirical challenge.

In this review, we have given a detailed overview regarding aspects relating to the fundamental instrumentation, interaction theories, radiation concerns, biomedical applications, and advanced data analysis. We emphasize the importance of having a comprehensive understanding of these aspects in order to gain a clear perspective of this interdisciplinary area. THz systems have been developed into various configurations to study samples ranging from molecules to cells and tissues and been applied in areas ranging from fundamental physical and chemical investigations to pathological diagnosis. THz instrumentation is becoming more widespread and different techniques have been developed and applied for specific applications. This has also promoted investigations into the biological effects of THz radiation. New challenges and difficulties have arisen from these applications and they push forward the further improvement of the techniques. As we summarized in Sec. VII, although we have seen tremendous progress being made to improve different aspects over the years, there are still several significant obstacles in between THz technology and its mature application to medicine. Thus, there is still a long way to go, but we envisage that further developments in biophotonics will gradually shorten the distance between research and clinical practice.

This work was partially supported by the Research Grants Council of Hong Kong (Project No. 14206717), the Engineering and Physical Sciences Research Council (EPSRC) (No. EP/S021442/1) and the Royal Society Wolfson Merit Award (EPM), and the National Natural Science Foundation of China (No. 61988102).

The authors have no conflicts to disclose.

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

1.
H.
Yada
,
M.
Nagai
, and
K.
Tanaka
, “
Origin of the fast relaxation component of water and heavy water revealed by terahertz time-domain attenuated total reflection spectroscopy
,”
Chem. Phys. Lett.
464
,
166
170
(
2008
).
2.
B. B.
Hu
and
M. C.
Nuss
, “
Imaging with terahertz waves
,”
Opt. Lett.
20
,
1716
(
1995
).
3.
B. M.
Fischer
,
M.
Walther
, and
P. U.
Jepsen
, “
Far-infrared vibrational modes of DNA components studied by terahertz time-domain spectroscopy
,”
Phys. Med. Biol.
47
,
3807
3814
(
2002
).
4.
Y.
Ueno
,
R.
Rungsawang
,
I.
Tomita
, and
K.
Ajito
, “
Quantitative measurements of amino acids by terahertz time-domain transmission spectroscopy
,”
Anal. Chem.
78
,
5424
5428
(
2006
).
5.
C.
Yu
,
S.
Fan
,
Y.
Sun
, and
E.
Pickwell-MacPherson
, “
The potential of terahertz imaging for cancer diagnosis: A review of investigations to date
,”
Quant. Imaging Med. Surg.
2
,
33
45
(
2012
).
6.
Y.
Peng
,
C.
Shi
,
X.
Wu
,
Y.
Zhu
, and
S.
Zhuang
, “
Terahertz imaging and spectroscopy in cancer diagnostics: A technical review
,”
BME Front.
2020
,
2547609
.
7.
K. I.
Zaytsev
,
I. N.
Dolganova
,
N. V.
Chernomyrdin
,
G. M.
Katyba
,
A. A.
Gavdush
,
O. P.
Cherkasova
,
G. A.
Komandin
,
M. A.
Shchedrina
,
A. N.
Khodan
,
D. S.
Ponomarev
,
I. V.
Reshetov
,
V. E.
Karasik
,
M.
Skorobogatiy
,
V. N.
Kurlov
, and
V. V.
Tuchin
, “
The progress and perspectives of terahertz technology for diagnosis of neoplasms: A review
,”
J. Opt.
22
,
013001
(
2020
).
8.
A. I.
Nikitkina
,
P. Y.
Bikmulina
,
E. R.
Gafarova
,
N. V.
Kosheleva
,
Y. M.
Efremov
,
E. A.
Bezrukov
,
D. V.
Butnaru
,
I. N.
Dolganova
,
N. V.
Chernomyrdin
,
O. P.
Cherkasova
,
A. A.
Gavdush
, and
P. S.
Timashev
, “
Terahertz radiation and the skin: A review
,”
J. Biomed. Opt.
26
,
043005
(
2021
).
9.
J.
Wang
,
H.
Lindley-hatcher
,
X.
Chen
, and
E.
PickwellâMacPherson
, “
THz sensing of human skin: A review of skin modeling approaches
,”
Sensors
21
,
3624
(
2021
).
10.
S.
Fan
,
Y.
He
,
B. S.
Ung
, and
E.
Pickwell-Macpherson
, “
The growth of biomedical terahertz research
,”
J. Phys. D
47
,
374009
(
2014
).
11.
X.
Yang
,
X.
Zhao
,
K.
Yang
,
Y.
Liu
,
Y.
Liu
,
W.
Fu
, and
Y.
Luo
, “
Biomedical applications of terahertz spectroscopy and imaging
,”
Trends Biotechnol.
34
,
810
824
(
2016
).
12.
Q.
Sun
,
Y.
He
,
K.
Liu
,
S.
Fan
,
E. P. J.
Parrott
, and
E.
Pickwell-MacPherson
, “
Recent advances in terahertz technology for biomedical applications
,”
Quant. Imaging Med. Surg.
7
,
345
355
(
2017
).
13.
M.
Tonouchi
, “
Cutting-edge terahertz technology
,”
Nat. Photonics
1
,
97
105
(
2007
).
14.
X.-C.
Zhang
and
J.
Xu
,
Introduction to THz Wave Photonics
(
Springer
,
New York
,
2010
).
15.
D. H.
Auston
and
K. P.
Cheung
, “
Coherent time-domain far-infrared spectroscopy
,”
J. Opt. Soc. Am. B
2
,
606
612
(
1985
).
16.
Y.
Hirota
,
R.
Hattori
,
M.
Tani
, and
M.
Hangyo
, “
Polarization modulation of terahertz wave by multipole type photoconductive antenna
,”
Opt. Express
14
,
4486
(
2006
).
17.
E. P. J.
Parrott
,
S. M. Y.
Sy
,
T.
Blu
,
V. P.
Wallace
, and
E.
Pickwell-MacPherson
, “
Terahertz pulsed imaging in vivo: Measurements and processing methods
,”
J. Biomed. Opt.
16
,
106010
(
2011
).
18.
Z. B.
Harris
,
M. E.
Khani
, and
M.
Hassan Arbab
, “
Terahertz portable handheld spectral reflection (PHASR) scanner
,”
IEEE Access
8
,
228024
228031
(
2020
).
19.
Y. B.
Ji
,
E. S.
Lee
,
S.-H.
Kim
,
J.-H.
Son
, and
T.-I.
Jeon
, “
A miniaturized fiber-coupled terahertz endoscope system
,”
Opt. Express
17
,
17082
(
2009
).
20.
J.
Wang
,
H.
Lindley-Hatcher
,
K.
Liu
, and
E.
Pickwell-MacPherson
, “
Evaluation of transdermal drug delivery using terahertz pulsed imaging
,”
Biomed. Opt. Express
11
,
4484
(
2020
).
21.
X.
Chen
,
E. P. J.
Parrott
,
Z.
Huang
,
H.-P.
Chan
, and
E.
Pickwell-MacPherson
, “
Robust and accurate terahertz time-domain spectroscopic ellipsometry
,”
Photonics Res.
6
,
768
775
(
2018
).
22.
K.-E.
Peiponon
,
J. A.
Zeitler
, and
M.
Kuwata-Gonokami
,
Terahertz Spectroscopy and Imaging
(
Springer
,
2012
), Vol.
171
.
23.
A.
Bonvalet
,
M.
Joffre
,
J. L.
Martin
, and
A.
Migus
, “
Generation of ultrabroadband femtosecond pulses in the mid-infrared by optical rectification of 15 fs light pulses at 100 MHz repetition rate
,”
Appl. Phys. Lett.
67
,
2907
(
1995
).
24.
B.
Braaf
,
M. G. O.
Gräfe
,
N.
Uribe-Patarroyo
,
B. E.
Bouma
,
B. J.
Vakoc
,
J. F.
de Boer
,
S.
Donner
, and
J.
Weichsel
,
High Resolution Imaging in Microscopy and Ophthalmology
(
Springer
,
2019
), pp.
161
179
.
25.
J. A.
Fülöp
,
S.
Tzortzakis
, and
T.
Kampfrath
, “
Laser-driven strong-field terahertz sources
,”
Adv. Opt. Mater.
8
,
1900681
(
2020
).
26.
J.
Zhao
,
E.
Yiwen
,
K.
Williams
,
X. C.
Zhang
, and
R. W.
Boyd
, “
Spatial sampling of terahertz fields with sub-wavelength accuracy via probe-beam encoding
,”
Light
8
,
55
(
2019
).
27.
L.
Olivieri
,
J. S.
Totero Gongora
,
A.
Pasquazi
, and
M.
Peccianti
, “
Time-resolved nonlinear ghost imaging
,”
ACS Photonics
5
,
3379
3388
(
2018
).
28.
L.
Olivieri
,
J. S.
Gongora
,
L.
Peters
,
V.
Cecconi
,
A.
Cutrona
,
J.
Tunesi
,
R.
Tucker
,
A.
Pasquazi
, and
M.
Peccianti
, “
Hyperspectral terahertz microscopy via nonlinear ghost imaging
,”
Optica
7
,
186
191
(
2020
).
29.
T.
Kampfrath
,
M.
Battiato
,
P.
Maldonado
,
G.
Eilers
,
J.
Nötzold
,
S.
Mährlein
,
V.
Zbarsky
,
F.
Freimuth
,
Y.
Mokrousov
,
S.
Blügel
,
M.
Wolf
,
I.
Radu
,
P. M.
Oppeneer
, and
M.
Münzenberg
, “
Terahertz spin current pulses controlled by magnetic heterostructures
,”
Nat. Nanotechnol.
8
,
256
260
(
2013
).
30.
E. T.
Papaioannou
and
R.
Beigang
, “
THz spintronic emitters: A review on achievements and future challenges
,”
Nanophotonics
10
,
1243
1257
(
2021
).
31.
T.
Seifert
,
S.
Jaiswal
,
U.
Martens
,
J.
Hannegan
,
L.
Braun
,
P.
Maldonado
,
F.
Freimuth
,
A.
Kronenberg
,
J.
Henrizi
,
I.
Radu
,
E.
Beaurepaire
,
Y.
Mokrousov
,
P. M.
Oppeneer
,
M.
Jourdan
,
G.
Jakob
,
D.
Turchinovich
,
L. M.
Hayden
,
M.
Wolf
,
M.
Münzenberg
,
M.
Kläui
, and
T.
Kampfrath
, “
Efficient metallic spintronic emitters of ultrabroadband terahertz radiation
,”
Nat. Photonics
10
,
483
488
(
2016
).
32.
S. C.
Chen
,
Z.
Feng
,
J.
Li
,
W.
Tan
,
L. H.
Du
,
J.
Cai
,
Y.
Ma
,
K.
He
,
H.
Ding
,
Z. H.
Zhai
,
Z. R.
Li
,
C. W.
Qiu
,
X. C.
Zhang
, and
L. G.
Zhu
, “
Ghost spintronic THz-emitter-array microscope
,”
Light
9
,
99
(
2020
).
33.
Y.
Wu
,
M.
Elyasi
,
X.
Qiu
,
M.
Chen
,
Y.
Liu
,
L.
Ke
, and
H.
Yang
, “
High-performance THz emitters based on ferromagnetic/nonmagnetic heterostructures
,”
Adv. Mater.
29
,
1603031
(
2017
).
34.
T.
Seifert
,
S.
Jaiswal
,
M.
Sajadi
,
G.
Jakob
,
S.
Winnerl
,
M.
Wolf
,
M.
Kläui
, and
T.
Kampfrath
, “
Ultrabroadband single-cycle terahertz pulses with peak fields of 300 kV cm-1 from a metallic spintronic emitter
,”
Appl. Phys. Lett.
110
,
252402
(
2017
).
35.
I.
Mehdi
,
J. V.
Siles
,
C.
Lee
, and
E.
Schlecht
, “
THz diode technology: Status, prospects, and applications
,”
Proc. IEEE
105
,
990
1007
(
2017
).
36.
M.
Naftaly
,
R. G.
Clarke
,
D. A.
Humphreys
, and
N. M.
Ridler
, “
Metrology state-of-the-art and challenges in broadband phase-sensitive terahertz measurements
,”
Proc. IEEE
105
,
1151
1165
(
2017
).
37.
P.
Tan
,
J.
Huang
,
K. F.
Liu
,
Y. Q.
Xiong
, and
M. W.
Fan
, “
Terahertz radiation sources based on free electron lasers and their applications
,”
Sci. China Inf. Sci.
55
,
107–125
(
2012
).
38.
G. N.
Kulipanov
,
E. G.
Bagryanskaya
,
E. N.
Chesnokov
,
Y. Y.
Choporova
,
V. V.
Gerasimov
,
Y. V.
Getmanov
,
S. L.
Kiselev
,
B. A.
Knyazev
,
V. V.
Kubarev
,
S. E.
Peltek
,
V. M.
Popik
,
T. V.
Salikova
,
M. A.
Scheglov
,
S. S.
Seredniakov
,
O. A.
Shevchenko
,
A. N.
Skrinsky
,
S. L.
Veber
, and
N. A.
Vinokurov
, “
Novosibirsk free electron laser-facility description and recent experiments
,”
IEEE Trans. Terahertz Sci. Technol.
5
,
798
809
(
2015
).
39.
A.
Doria
,
G. P.
Gallerano
,
E.
Giovenale
,
G.
Messina
,
A.
Lai
,
A.
Ramundo-Orlando
,
V.
Sposato
,
M.
D'Arienzo
,
A.
Perrotta
,
M.
Romanò
,
M.
Sarti
,
M. R.
Scarfì
,
I.
Spassovsky
, and
O.
Zeni
, “
THz radiation studies on biological systems at the ENEA FEL facility
,”
Infrared Phys. Technol.
45
,
339
347
(
2004
).
40.
S.
Yamazaki
,
M.
Harata
,
Y.
Ueno
,
M.
Tsubouchi
,
K.
Konagaya
,
Y.
Ogawa
,
G.
Isoyama
,
C.
Otani
, and
H.
Hoshina
, “
Propagation of THz irradiation energy through aqueous layers: Demolition of actin filaments in living cells
,”
Sci. Rep.
10
,
9008
(
2020
).
41.
Y. U.
Jeong
,
G. M.
Kazakevitch
,
H. J.
Cha
,
S. H.
Park
, and
B. C.
Lee
, “
Demonstration of a wide-band compact free electron laser to the THz imaging of bio samples
,”
Nucl. Instrum. Methods Phys. Res., Sect. A
575
,
58
62
(
2007
).
42.
P.
Chevalier
,
A.
Amirzhan
,
F.
Wang
,
M.
Piccardo
,
S. G.
Johnson
,
F.
Capasso
, and
H. O.
Everitt
, “
Widely tunable compact terahertz gas lasers
,”
Science
366
,
856
860
(
2019
).
43.
A.
Khalatpour
,
A. K.
Paulsen
,
C.
Deimert
,
Z. R.
Wasilewski
, and
Q.
Hu
, “
High-power portable terahertz laser systems
,”
Nat. Photonics
15
,
16
20
(
2021
).
44.
K.
Murate
and
K.
Kawase
, “
Perspective: Terahertz wave parametric generator and its applications
,”
J. Appl. Phys.
124
,
160901
(
2018
).
45.
A. D.
Rakić
,
T.
Taimre
,
K.
Bertling
,
Y. L.
Lim
,
P.
Dean
,
A.
Valavanis
, and
D.
Indjin
, “
Sensing and imaging using laser feedback interferometry with quantum cascade lasers
,”
Appl. Phys. Rev.
6
,
021320
(
2019
).
46.
C.
Baker
,
I. S.
Gregory
,
M. J.
Evans
,
W. R.
Tribe
,
E. H.
Linfield
, and
M.
Missous
, “
All-optoelectronic terahertz system using low-temperature-grown InGaAs photomixers
,”
Opt. Express
13
,
9639
(
2005
).
47.
M.
Wan
,
J. J.
Healy
, and
J. T.
Sheridan
, “
Terahertz phase imaging and biomedical applications
,”
Opt. Laser Technol.
122
,
105859
(
2020
).
48.
R.
Guo
,
K.
Akiyama
,
H.
Minamide
, and
H.
Ito
, “
Frequency-agile terahertz-wave spectrometer for high-resolution gas sensing
,”
Appl. Phys. Lett.
90
,
121127
(
2007
).
49.
T.
Hagelschuer
,
M.
Wienold
,
H.
Richter
,
L.
Schrottke
,
H. T.
Grahn
, and
H.-W.
Hübers
, “
Real-time gas sensing based on optical feedback in a terahertz quantum-cascade laser
,”
Opt. Express
25
,
30203
(
2017
).
50.
J. A.
Zeitler
,
P. F.
Taday
,
D. A.
Newnham
,
M.
Pepper
,
K. C.
Gordon
, and
T.
Rades
, “
Terahertz pulsed spectroscopy and imaging in the pharmaceutical setting—A review
,”
J. Pharm. Pharmacol.
59
,
209
223
(
2007
).
51.
F.
Keilmann
and
R.
Hillenbrand
, “
Near-field microscopy by elastic light scattering from a tip
,”
Philos. Trans. R. Soc. London, Ser. A
362
,
787
805
(
2004
).
52.
H.-G.
von Ribbeck
,
M.
Brehm
,
D. W.
van der Weide
,
S.
Winnerl
,
O.
Drachenko
,
M.
Helm
, and
F.
Keilmann
, “
Spectroscopic THz near-field microscope
,”
Opt. Express
16
,
3430
(
2008
).
53.
B.
Li
,
D.
Wang
,
L.
Rong
,
C.
Zhai
,
Y.
Wang
, and
J.
Zhao
, “
Application of continuous-wave terahertz computed tomography for the analysis of chicken bone structure
,”
Opt. Eng.
57
,
023105
(
2018
).
54.
A.
Dobroiu
,
M.
Yamashita
,
Y. N.
Ohshima
,
Y.
Morita
,
C.
Otani
, and
K.
Kawase
, “
Terahertz imaging system based on a backward-wave oscillator
,”
Appl. Opt.
43
,
5637
(
2004
).
55.
J. Y.
Lu
,
C. C.
Kuo
,
C. M.
Chiu
,
H. W.
Chenc
,
C. L.
Pan
, and
C. K.
Sun
, “
THz interferometric imaging using subwavelength plastic fiber based THz endoscopes
,”
Opt. Express
16
,
2494
2501
(
2008
).
56.
H.
Chen
,
S.
Ma
,
X.
Wu
,
W.
Yang
, and
T.
Zhao
, “
Diagnose human colonic tissues by terahertz near-field imaging
,”
J. Biomed. Opt.
20
,
036017
(
2015
).
57.
Z. D.
Taylor
,
J.
Garritano
,
S.
Sung
,
N.
Bajwa
,
D. B.
Bennett
,
B.
Nowroozi
,
P.
Tewari
,
J. W.
Sayre
,
J. P.
Hubschman
,
S. X.
Deng
,
E. R.
Brown
, and
W. S.
Grundfest
, “
THz and mm-wave sensing of corneal tissue water content: In vivo sensing and imaging results
,”
IEEE Trans. Terahertz Sci. Technol.
5
,
184
196
(
2015
).
58.
S.
Sung
,
S.
Selvin
,
N.
Bajwa
,
S.
Chantra
,
B.
Nowroozi
,
J.
Garritano
,
J.
Goell
,
A. D.
Li
,
S. X.
Deng
,
E. R.
Brown
,
W. S.
Grundfest
, and
Z. D.
Taylor
, “
THz imaging system for in vivo human cornea
,”
IEEE Trans. Terahertz Sci. Technol.
8
,
27
37
(
2018
).
59.
L.
Rong
,
T.
Latychevskaia
,
C.
Chen
,
D.
Wang
,
Z.
Yu
,
X.
Zhou
,
Z.
Li
,
H.
Huang
,
Y.
Wang
, and
Z.
Zhou
, “
Terahertz in-line digital holography of human hepatocellular carcinoma tissue
,”
Sci. Rep.
5
,
8445
(
2015
).
60.
P.
Doradla
,
K.
Alavi
,
C.
Joseph
, and
R.
Giles
, “
Detection of colon cancer by continuous-wave terahertz polarization imaging technique
,”
J. Biomed. Opt.
18
,
090504
(
2013
).
61.
J.
Shi
,
Y.
Wang
,
T.
Chen
,
D.
Xu
,
H.
Zhao
,
L.
Chen
,
C.
Yan
,
L.
Tang
,
Y.
He
,
H.
Feng
, and
J.
Yao
, “
Automatic evaluation of traumatic brain injury based on terahertz imaging with machine learning
,”
Opt. Express
26
,
6371
6381
(
2018
).
62.
P. H.
Siegel
and
R. J.
Dengler
, “
Terahertz heterodyne imager for biomedical applications
,”
Proc. SPIE
5354
,
1
9
(
2004
).
63.
S. M.
Kim
,
F.
Hatami
,
J. S.
Harris
,
A. W.
Kurian
,
J.
Ford
,
D.
King
,
G.
Scalari
,
M.
Giovannini
,
N.
Hoyler
,
J.
Faist
, and
G.
Harris
, “
Biomedical terahertz imaging with a quantum cascade laser
,”
Appl. Phys. Lett.
88
,
153903
(
2006
).
64.
M.
Locatelli
,
M.
Ravaro
,
S.
Bartalini
,
L.
Consolino
,
M. S.
Vitiello
,
R.
Cicchi
,
F.
Pavone
, and
P. D.
Natale
, “
Real-time terahertz digital holography with a quantum cascade laser
,”
Sci. Rep.
5
,
13566
(
2015
).
65.
Y. L.
Lim
,
T.
Taimre
,
K.
Bertling
,
P.
Dean
,
D.
Indjin
,
A.
Valavanis
,
S. P.
Khanna
,
M.
Lachab
,
H.
Schaider
,
T. W.
Prow
,
H.
Peter Soyer
,
S. J.
Wilson
,
E. H.
Linfield
,
A. G.
Davies
, and
A. D.
Rakić
, “
High-contrast coherent terahertz imaging of porcine tissue via swept-frequency feedback interferometry
,”
Biomed. Opt. Express
5
,
3981
(
2014
).
66.
P.
Knobloch
,
C.
Schildknecht
,
T.
Kleine-Ostmann
,
M.
Koch
,
S.
Hoffmann
,
M.
Hofmann
,
E.
Rehberg
,
M.
Sperling
,
K.
Donhuijsen
,
G.
Hein
, and
K.
Pierz
, “
Medical THz imaging: An investigation of histo-pathological samples
,”
Phys. Med. Biol.
47
,
3875
3884
(
2002
).
67.
K. J.
Siebert
,
T.
Löffler
,
H.
Quast
,
M.
Thomson
,
T.
Bauer
,
R.
Leonhardt
,
S.
Czasch
, and
H. G.
Roskos
, “
All-optoelectronic continuous wave THz imaging for biomedical applications
,”
Phys. Med. Biol.
47
,
3743
3748
(
2002
).
68.
K.
Kawase
,
Y.
Ogawa
,
Y.
Watanabe
, and
H.
Inoue
, “
Non-destructive terahertz imaging of illicit drugs using spectral fingerprints
,”
Opt. Express
11
,
2549
(
2003
).
69.
Y.
Wang
,
H.
Minamide
,
M.
Tang
,
T.
Notake
, and
H.
Ito
, “
Study of water concentration measurement in thin tissues with terahertz-wave parametric source
,”
Opt. Express
18
,
15504
(
2010
).
70.
D. B.
Bennett
,
W.
Li
,
Z. D.
Taylor
,
W. S.
Grundfest
, and
E. R.
Brown
, “
Stratified media model for terahertz reflectometry of the skin
,”
IEEE Sens. J.
11
,
1253
1262
(
2010
).
71.
W.
Zhang
,
E. R.
Brown
,
L.
Viveros
,
K. P.
Burris
, and
C. N.
Stewart
, “
Narrow terahertz attenuation signatures in Bacillus thuringiensis
,”
J. Biophotonics
7
,
818
824
(
2014
).
72.
I.
Ozheredov
,
M.
Prokopchuk
,
M.
Mischenko
,
T.
Safonova
,
P.
Solyankin
,
A.
Larichev
,
A.
Angeluts
,
A.
Balakin
, and
A.
Shkurinov
, “
In vivo THz sensing of the cornea of the eye
,”
Laser Phys. Lett.
15
,
055601
(
2018
).
73.
A.
Tamminen
,
M.
Baggio
,
I.
Nefedova
,
Q.
Sun
,
S.
Presnyakov
,
J.
Ala-laurinaho
,
E.
Brown
,
V.
Wallace
,
E.
Macpherson
,
T.
Maloney
,
N.
Kravchenko
,
M.
Salkola
,
S.
Deng
, and
Z.
Taylor
, “
Extraction of thickness and water content gradients in hydrogel-based, water-backed corneal phantoms via submillimeter wave reflectometry
,”
IEEE Trans. Terahertz Sci. Technol.
11
(
6
),
647–659
(
2021
).
74.
A.
Pashkin
,
M.
Kempa
,
H.
Němec
,
F.
Kadlec
, and
P.
Kužel
, “
Phase-sensitive time-domain terahertz reflection spectroscopy
,”
Rev. Sci. Instrum.
74
,
4711
4717
(
2003
).
75.
X.
Chen
,
E. P.
Parrott
,
B. S.
Ung
, and
E.
Pickwell-Macpherson
, “
A robust baseline and reference modification and acquisition algorithm for accurate THz imaging
,”
IEEE Trans. Terahertz Sci. Technol.
7
,
493
501
(
2017
).
76.
U.
Møller
, “
Terahertz spectroscopy applied to food model systems
,” Ph.D. thesis (
Tedhnical University of Denmark
,
2009
).
77.
P. U.
Jepsen
,
D. G.
Cooke
, and
M.
Koch
, “
Terahertz spectroscopy and imaging—Modern techniques and applications
,”
Laser Photonics Rev.
5
,
124
166
(
2011
).
78.
Q.
Sun
,
K.
Liu
,
X.
Chen
,
X.
Liu
,
A. I.
Hernandez-Serrano
, and
E.
Pickwell-MacPherson
, “
Utilizing multilayer structures to enhance terahertz characterization of thin films ranging from aqueous solutions to histology slides
,”
Opt. Lett.
44
,
2149
2152
(
2019
).
79.
H.
Fujiwara
,
Spectroscopic Ellipsometry Principles and Applications
(
John Wiley & Sons
,
2007
).
80.
T.
Hofmann
,
C. M.
Herzinger
,
A.
Boosalis
,
T. E.
Tiwald
,
J. a.
Woollam
, and
M.
Schubert
, “
Variable-wavelength frequency-domain terahertz ellipsometry
,”
Rev. Sci. Instrum.
81
,
023101
(
2010
).
81.
M.
Yamashita
,
H.
Takahashi
,
T.
Ouchi
, and
C.
Otani
, “
Ultra-broadband terahertz time-domain ellipsometric spectroscopy utilizing GaP and GaSe emitters and an epitaxial layer transferred photoconductive detector
,”
Appl. Phys. Lett.
104
,
051103
(
2014
).
82.
N.
Matsumoto
,
T.
Hosokura
,
T.
Nagashima
, and
M.
Hangyo
, “
Measurement of the dielectric constant of thin films by terahertz time-domain spectroscopic ellipsometry
,”
Opt. Lett.
36
,
265
267
(
2011
).
83.
M.
Neshat
and
N.
Armitage
, “
Terahertz time-domain spectroscopic ellipsometry: Instrumentation and calibration
,”
Opt. Express
20
,
29063
29075
(
2012
).
84.
X.
Chen
,
Q.
Sun
,
J.
Wang
,
H.
Lindley-Hatcher
, and
E.
PickwellâMacPherson
, “
Exploiting complementary terahertz ellipsometry configurations to probe the hydration and cellular structure of skin in vivo
,”
Adv. Photonics Res.
2
,
2000024
(
2021
).
85.
M.
Eisele
,
T. L.
Cocker
,
M. A.
Huber
,
M.
Plankl
,
L.
Viti
,
D.
Ercolani
,
L.
Sorba
,
M. S.
Vitiello
, and
R.
Huber
, “
Ultrafast multi-terahertz nano-spectroscopy with sub-cycle temporal resolution
,”
Nat. Photonics
8
,
841
845
(
2014
).
86.
K.
Okada
,
K.
Serita
,
Q.
Cassar
,
H.
Murakami
,
G.
MacGrogan
,
J.-P.
Guillet
,
P.
Mounaix
, and
M.
Tonouchi
, “
Terahertz near-field microscopy of ductal carcinoma in situ (DCIS) of the breast
,”
J. Phys.
2
,
044008
(
2020
).
87.
F.
Blanchard
,
A.
Doi
,
T.
Tanaka
,
H.
Hirori
,
H.
Tanaka
,
Y.
Kadoya
, and
K.
Tanaka
, “
Real-time terahertz near-field microscope
,”
Opt. Express
19
,
8277
(
2011
).
88.
L. E.
Barr
,
P.
Karlsen
,
S. M.
Hornett
,
I. R.
Hooper
,
M.
Mrnka
,
C. R.
Lawrence
,
D. B.
Phillips
, and
E.
Hendry
, “
Super-resolution imaging for sub-IR frequencies based on total internal reflection
,”
Optica
8
,
88
(
2021
).
89.
R. I.
Stantchev
,
J. C.
Mansfield
,
R. S.
Edginton
,
P.
Hobson
,
F.
Palombo
, and
E.
Hendry
, “
Subwavelength hyperspectral THz studies of articular cartilage
,”
Sci. Rep.
8
,
6924
(
2018
).
90.
C.-M.
Chiu
,
H.-W.