Biophotonics technologies for the detection of VOCs in healthcare applications: Are we there yet? Open Access

Volatile organic compounds (VOCs) are organic compounds that have a high vapor pressure at room temperature and a low boiling point. They can be found in many aspects of our daily lives, such as in scents, food, and pollutants.¹ VOCs can be biogenic, primarily emitted by plants, or artificial, mainly from fossil fuel use and production, solvents, compressed aerosol products, biofuel use, and biomass combustion. In the context of biomedical and healthcare applications, most of these VOCs can have harmful effects on human health and may have long-term chronic health effects. For example, human breath contains hundreds of VOCs, which can serve as biomarkers for diseases.^2–6 In addition to inhalation and exhalation, other pathways for VOCs in the human body include metabolism and excretion. The carrier media of VOCs can include blood, urine, feces, sweat, and skin.^7,8

Figure 1(a) shows that the detection of VOCs has gained a great deal of attention in recent years, as evidenced by the increased number of journal publications on the subject, which has grown 2.6 times from 2012 to 2022. Figure 1(b) shows that VOCs are commonly quantified and identified using chemical technologies, accounting for approximately 47% of all detection technologies. The first major technology is gas chromatography (GC), which is used for separating and analyzing compounds in a gaseous or liquid mixture that can be vaporized.^9–11 The second major technology is mass spectroscopy (MS), which measures the mass-to-charge ratio of ions and determines the masses of molecules or other compounds.^9,12,13 The combined technology GC-MS is the gold standard for accurately quantifying VOCs.^14,15 However, these methods have limited clinical use due to their cost and bulkiness. In contrast, optical technologies offer advantages such as portability, cost-effectiveness, and the ability to integrate multiple technologies into a single, multi-modality device. Specifically, optical technologies (or biophotonics technologies) account for 25% of detection technologies, while electrical and mechanical technologies account for 25% and 3%, respectively.

There have been several recent review articles on similar topics. Optical technologies focused review articles include Xing et al. for optical spectroscopy,¹⁶ Lelevic et al. for GC combined with vacuum ultraviolet (VUV),¹⁷ Meher et al. for fluorescence spectroscopy (FS),¹⁸ Usman et al. for plasmonic sensors,¹⁹ and Kazzy et al. for surface plasmon resonance (SPR).²⁰ Application-focused review articles on the detection of breath VOCs include Vaittinen et al.,²¹ Henderson et al.,²² and Selvaraj et al.²³ for laser absorption spectroscopy (LAS), Li et al.²⁴ and Wilson et al.²⁵ for electronic noses, Chisanga et al.²⁶ for surface-enhanced Raman spectroscopy (SERS), Buszewski et al.²⁷ and Chow et al.²⁸ for various other technologies. Figure 1(c) illustrates that biomedical science accounts for roughly 8% of the applications of VOC detection using optical technologies. Within biomedical applications, research on respiratory systems is the largest, taking up 60% of the total, followed by oncology (20%), gastroenterology and hepatology (13%), and infectious diseases (7%), as shown in Fig. 1(d). Here, we present a comprehensive review of the detection of VOCs using various optical technologies in the last decade, with a focus on biomedical and healthcare applications. We also provide an insight into the challenges faced in adopting these technologies in clinical settings.

This review is organized as follows: First, an overview of the optical technologies, including SERS, FS, LAS, photoacoustic spectroscopy (PAS), and SPR, is provided. Second, the detection of VOCs for various biomedical applications is discussed. Figure 2 illustrates the relationship between VOCs and the medical conditions covered in this paper, showing that a VOC can often serve as a biomarker for one or multiple medical conditions. For example, acetone has been linked to lung cancer, gastric disease, diabetes, and COVID-19.^29–33 Conversely, a medical condition may have one or a group of VOC biomarkers, such as dimethyl disulfide for bacterial infection, and methane, ethylbenzene (EBz), hexanal, ethylene, ethane, acetylene, etc., for lung disease. Additionally, a table summarizing the limit of detection (LOD) of all the VOCs and the related optical technologies is presented for reference, serving as a useful starting point for evaluating their performance. Papers with demonstrated experiments using clinical samples were specifically pointed out. Finally, future perspectives are provided, focusing on identifying the gaps in current technology development that need to be improved for implementation in the healthcare environment, integrating machine learning (ML) and deep learning (DL) into data analytics to improve detection accuracy and efficiency, and developing devices capable of multimodal measurements for a complete characterization of target VOC biomarkers.

Raman spectroscopy is a promising optical technology that has been gaining increasing attention due to its exceptional molecular specificity and high sensitivity. The inelastic scattering of incident light radiation from the sample is measured to obtain a unique fingerprint Raman spectrum of the molecule of interest.³⁴ However, the real-life application of Raman spectroscopy is limited due to the intrinsically weak Raman scattering, which only occurs for one in every 10⁶ – 10⁸ scattered photons.³⁵ 

To overcome this limitation and improve the sensitivity of the Raman technology, a plasmonic active surface (e.g., silver or gold-based nanoparticles or other planar nanostructures) is introduced, and upon which shining the incident light creates electromagnetic hotspots in the nanocavity region. This allows the scattering to be significantly enhanced by 10⁸ times or higher, a phenomenon known as SERS.³⁶ However, it is difficult to capture molecules in the hotspot regions on a SERS substrate.³⁷ Hence, various approaches have been proposed to solve the issue, such as improving the fabrication of plasmonic SERS substrates with specific analyte-capturing moiety and modification of SERS gas sensors with the incorporation of nanopillars/particles or metal–organic framework (MOF) structures.³⁸ Another trend is the combination of SERS technology with ML analytical tools for the detection of VOCs in exhaled human breath due to their complexity and low concentration.³⁹ Overall, SERS is an excellent spectroscopy analyzing technology that is sensitive, cost-effective, and simple to use, along with the rise of high-end portable Raman spectrometers.

Recent advancements in optical technology have broadened the applications of FS for the identification of biochemical analytes due to its high sensitivity, specificity, rapid response, and its portability for real-time monitoring.⁴⁰ FS provides information on the structure and dynamic properties of biomolecules and biomolecular complexes. Typically, ultraviolet (UV) or visible light is utilized to excite electrons in a molecule from its ground state (S₀) to an excited state (S₁), which will then relax back to its ground state by emitting photons radiatively or non-radiatively.⁴¹ Radiative transition relaxation happens when fluorescence occurs at specific emission wavelengths, while non-radiative relaxation happens due to thermal generation.

Generally, organic molecules that absorb or emit fluorescence, known as fluorophores, are characterized by their quantum yield and lifetime. Quantum yield refers to the ratio of the number of emitted photons to the number of photons absorbed, and lifetime refers to the average time a molecule remains in the excited state, whereby a typical fluorescence lifetime is 10⁻¹⁰–10⁻⁷ s.⁴² Moreover, various factors such as solvents, temperature, pH, ionic strength, and the presence of other substances can affect fluorescence. As such, these factors could reduce fluorescence intensity, known as quenching, due to intramolecular or intermolecular interactions.

The flexible synthesis, remarkable biocompatibility, and reproducibility of organic fluorophores make fluorescence a good candidate for identifying a wide range of analytes.^18,43 Hence, multiple sensing materials have been constructed to identify VOCs by observing fluorescence response on a variety of platforms.

In LAS, the wavelength ranges used for the detection of VOCs in the infrared region are from 0.78 to 2.5 μm (or 4000–12 820 cm⁻¹) in the near-infrared (NIR) range for acetylene, hydrogen cyanide, dimethyl disulfide, etc., and from 2.5 to 25 μm (or 400–4000 cm⁻¹) in the mid-infrared (MIR) range for acetone, ethane, methane, and isoprene.^44,45 The traditional LAS systems with a single pass cell that measures how much incident light is absorbed by the sample have undergone several modifications over the decade to improve their sensitivity, speed, specificity, and convenience. For example, cavity-enhanced absorption spectroscopy (CEAS) uses highly reflective mirrors to increase the optical path length of the cell and, thus, improve the absorption rate of the sample. When the incident light is introduced in an off-axis (OA) manner, it is called OA-CEAS. There are several types of variations of CEAS, such as cavity ringdown spectroscopy (CRDS), cavity leak-out spectroscopy (CALOS), and integrated cavity output spectroscopy (ICOS). In addition to modifications in the cavity, different laser sources have been introduced. For example, tunable diode laser absorption spectroscopy (TDLAS) uses a tunable laser, and it includes wavelength modulation spectroscopy (WMS) and frequency modulation spectroscopy (FMS). In particular, WMS is more widely used. Other types of lasers include edge-emitting diode lasers such as quantum cascade lasers (QCL), interband cascade lasers (ICL), and surface-emitting diode lasers such as vertical cavity surface emitting lasers (VECSEL). QCLs are faster than ICLs but require a higher input power. VECSELs are advantageous due to their ease of production, increased output power, high-temperature environment tolerance, and are known to have low speckle noise. Although laser sources have been used for detecting VOCs, they still have the drawback of limited tunability for detecting multiple VOCs. To address this issue, frequency comb lasers were introduced, which generate a broadband discrete spectrum. This technique is called cavity-enhanced direct frequency comb spectroscopy (CE-DFCS).^46,47 Infrared (IR) spectroscopy in combination with GC has also been studied for detecting multiple organic molecules. This involves separating gas molecules by GC and passing them through a cavity or hollow fiber where they interact with a QCL to produce emission spectra.^48–50 However, the application of this technology in VOC detection for medical applications, such as breath analysis, is yet to be investigated.

Studies have been conducted on LAS in the UV region from 0.2 to 0.4 μm (or 25 000–50 000 cm⁻¹) and the vacuum UV (VUV) region from 0.01 to 0.2 μm (or 50 000–1 000 000 cm⁻¹).^17,51 VOC detection in the UV region has a significant advantage over IR because it has very little overlap with water vapor and can be used to detect small volume. The setup involves using deuterium lamps for high-intensity UV and a simple gas cell for adsorption, which is detected by a spectrometer. One modification to this setup is the use of a hollow-core fiber instead of a gas cell for better sensitivity. However, these techniques are usually used in conjunction with GC for the separation of VOCs from the samples and UV absorption spectroscopy for detecting the separated VOCs.^52,53

PAS utilizes the photoacoustic effect in which light pulses are absorbed and the excited energy molecules are converted into kinetic energy via molecular collision. This causes localized heating and expansion, which produces the acoustic signal that can be detected using microphones, micromachined cantilevers, quartz tuning forks, and other microelectromechanical systems (MEMS) devices.^54,55 This technology is known to be efficient in the accurate detection of trace gases for biological and medical applications.⁵⁶ However, it is a challenge to detect VOCs using PAS because the absorption of light for complex molecules produces a wide spectral range of absorption lines.⁵⁷ Hence, the detection of VOCs requires a wide tunability in wavelength, which needs high-power lasers such as optical parametric oscillators (OPO), external cavity quantum cascade lasers (EC-QCL), and CO₂ lasers.¹⁶ In addition, PAS has the capability of providing highly precise results using small sample volumes. Techniques like quartz-enhanced photoacoustic spectroscopy (QEPAS) and cantilever-enhanced photoacoustic spectroscopy (CEPAS) have demonstrated sub-parts per trillion (ppt) level detection limits and concentrations.^58,59 As a result, PAS is an attractive technique to combine with GC detectors to achieve high detection sensitivity for gas mixtures.

SPR is a noninvasive optical technology that analyzes molecular interactions in real time. SPR is a phenomenon where photons of a polarized light excite electrons on a metal surface layer, typically made of gold.⁶⁰ At a particular angle, energy from the incident light will interact with electrons on the surface layer between two media of different refractive indices and light will travel along the surface. Molecules that bind to the surface layer could result in a shift in SPR angle, hence, monitoring of the reflected light intensity or resonance angle shifts can determine the analyte amount. With increasing interest in VOCs and advancements in SPR biosensors, the development of PoC testing methods can potentially be used for disease diagnosis and patient treatment.

Lung cancer (LC) is the uncontrollable growth of cells in the lung that can spread to adjacent tissues and expand to other parts of the body. Continuous exposure to carcinogens (e.g., cigarette smoke) may cause genetic mutations, which alter protein synthesis, causing carcinogenesis in the lungs.⁶¹ It has been reported to be one of the main causes of cancer mortality worldwide, with an estimated amount of 1.79 × 10⁶ deaths in 2020.⁶² VOCs detected in the exhaled breath of individuals suffering from LC were found to be significantly different compared to a healthy person.⁶³ Therefore, analysis of breath VOCs has shown an increasing potential in providing a noninvasive and highly informative diagnostic tool that could pave the way for lower mortality and improved quality of life.

Qiao et al. developed a SERS sensor to enhance the Raman signal of p-ethylbenzaldehyde (p-EBZA) molecules, which is mostly found in large amounts in patients diagnosed with non-small-cell LC.⁶⁴ They designed a hollow Co-Ni layered double hydroxide nanocage on an Ag nanowire to trap p-EBZA molecules in a whirlpool-like pattern, allowing efficient adsorption of these molecules on the device with the help of p-aminothiophenol (p-ATP). The composition and concentration of the biomarker analyzed by the SERS data gave excellent reproducibility with a standard deviation of 7.8% and an LOD of 1.9 parts per billion (ppb). The SERS sensor was able to accurately differentiate seven species of aldehyde mixed with mimetic gas using both principal component analysis (PCA) and hierarchical cluster analysis (HCA). The composition, concentration, and varying aldehyde concentration were converted into a barcode by incorporating R, G, and B values of the Raman intensity images that were read by smartphones or handheld scanners. In a related study, Qiao et al. developed another SERS approach that enhanced 4-ethylbenzaldehyde (1-EBZA) adsorption on SERS substrate.⁶⁵ Note that p-EBZA and 4-EBZA are equivalent. To improve the biomarker adsorption onto the SERS substrate, the gas flow rate was reduced by using a specific core-shell 3D structure. It was comprised of an 8-layer coated gold superparticles, which act as the SERS hotspots. Additionally, a nucleophilic addition reaction using p-ATP attached to the gold superparticles can selectively capture 4-EBZA molecules in mixed mimetic exhalation gas, providing an LOD of 10 ppb. The two studies exhibit a promising technology that provides a user-friendly, accurate, and quick diagnosis of LC, which have a great potential to be applied in future clinical applications using real breath samples.

Zhou et al. developed a SERS nose consisting of a sponge-like Cu-doping NiO (NiO_x/Cu)-SnO₂ p–n semiconductor heterostructure (SnO-NiO/Cu-CuPc) structure; the aldehyde gas went through a cavity-vortex effect that improved the analyte and substrate surface interaction along with the nucleophilic addition reaction of p-ATP. It significantly enhanced the analyte signal, which provided an LOD of 2 ppb, with an excellent linearity range of 2 ppb to 20 parts per million (ppm). The device showed a good selectivity of benzaldehyde (BA) detection compared to other LC biomarkers, which gave an intensity ratio of near zero. Hence, its specificity and sensitivity in detecting BA molecules can be used in future clinical studies on LC patients. In another study, Xia et al. developed a dual-modal sensing system that can utilize both SERS and FS to detect and quantify BA.⁶⁷ The overall fabrication and process of the dual-modal system can be seen in Fig. 3. They created a vapor generation paper-based thin-film microextraction (VG-PTFM) method using carboxyl capped quantum dots (QD) and embedded 4-mercaptonoaniline-modified core-shell gold nanorod (GNRs-QDs@NU-901) onto a MOF structure, where they were assembled through electrostatic interaction. The introduction of the BA molecule interacted with 4-mercaptonoaniline and disassembled the GNRsQDs@NU-901, resulting in an enhanced fluorescence and Raman signal. The porous MOF structure also provided a “cavity-diffusion” effect that allowed high specificity of BA as it can be distinguished from other aldehyde molecules. By exposing to different concentration of gaseous BA, the Raman intensity ratio revealed a good linearity from 10⁻⁴ to 10⁻² ppm, along with an increased fluorescence signal at 543 nm. Additionally, by immobilizing CuPc that could detect different VOCs simultaneously.⁶⁶ The SERS nose provided a strong enhancement factor of 1.46 × 10¹⁰ with high selectivity to detect various VOCs (4-EBZA, 2-napthalenethiol (2-NT) and pyrene (PYR)) found in the exhaled breath from early-stage LC patients. Using a mixed mimetic exhalation gas, the authors reported that the SERS intensity ratio provided excellent linearity from 10 ppb to 1000 ppm with an LOD of 7.6–8.3 for PYR, 5.3–5.9 for 2-NT, and 4.1–4.6 ppb for 4-EBZA, which was much lower than the values acquired by GC-MS and FS.^66,67 The system also displayed decent stability within 3 months along with reproducibility of a relative standard deviation below 3.18%. The R, G, and B values from Raman intensity images were converted into barcodes for different VOC concentrations that can be read by smartphones or handheld scanners. Exhaled breath samples from 15 LC patients and 15 healthy subjects were tested. This allowed for simultaneous detection of VOC markers, providing improved reliability and reduced false positive results. Although the SERS device has excellent potential for use in biomarker screening, further verification is necessary. For example, a larger sample size should be used to provide a more representative analysis.

Zhang et al. developed a dendritic Ag nanocrystal substrate by taking inspiration from the features of the moth's antennae to detect the BA biomarker.⁶⁸ Due to the countless holes in the dendritic dynamic light scattering analysis of GNRs-QDs, assembly showed that fluorescence recovery could be adjusted from the amount of BA added. It is important to note that BA was detected with an LOD of 0.1 and 1.2 ppb for Raman and fluorescence spectroscopy, respectively. To assess the practicality of real-time detection of BA for the proposed method, breath samples were analyzed (10 LC patients and 10 healthy volunteers). The fluorescence and Raman assays matched with GC-MS assays. Further validation was performed using PCA, where LC and healthy controls were separated into distinct clusters, demonstrating that the GNRs-QDs@NU-901 substrate can differentiate between LC patients and healthy subjects through breath analysis. However, the correlation between disease stage and biomarker concentration remains ambiguous due to the small sample size. To clarify this relationship, it would be beneficial to conduct a larger scale study for future research.

Yang et al. developed a new method to detect isopropanol (IPA) based on a ratiometric fluorescent probe and a color recognition phone application through a smartphone platform.⁶⁹ The device provided real-time sensing of breath IPA, along with direct visualization and semiquantitative detection. The fluorescent probe was chemically synthesized by functionally modifying nicotinamide adenine dinucleotide (NAD+) onto a fluorescence internal standard, red carbon dots (CDs). Upon the addition of IPA, secondary alcohol dehydrogenase (S-ADH) reduced NAD+ to NADH. This electron transfer caused an increase in the blue fluorescence of NADH, resulting in the colored solution changing from red to blue. To determine system specificity, eight different organic solvents were used. They have reported that the turn-on fluorescence was observed for 1.0 µM of IPA, but it showed a slight increase in fluorescence intensity for other solvents. However, light blue color was observed in the presence of IPA, but color changes were not displayed in other solvents. As minor color changes could be hard to view with the naked eye, a color recognition smartphone application was made for semiquantitative determination for standard IPA, achieving an LOD of 0.75 ppb through mathematical calculations. Additionally, breath samples from 14 volunteers (4 healthy controls and 10 LC patients) were tested, whereby the latter showed significant fluorescence color change. Although the proposed method could be highly useful for diagnosing LC with good selectivity and sensitivity, the device should be tested on a larger sample size to be a home medical device.

Saalberg et al. reported six recurrent breath VOCs as LC biomarkers, which are EBz, hexanal, styrene, 2-butanone, isoprene, and 1-propanol.⁷⁰ To detect these biomarkers, several groups have employed MOF-based FS due to its high specific area and tunable physical-chemical properties. Zhang et al. studied a zirconium UiO-66 based framework, which was post-synthetically modified with methylamine (MA), ethyldiamine (ED), and N,N′dimethylethylenediamine (MMEN) as different amine functional groups.⁷¹ In comparison with UiO-66 material, the modified UiO-66 samples showed a significant increase in fluorescence intensity upon exposure to hexanal, styrene, and isoprene, while exposure to 1propanol and EBz resulted in a reduction in intensity along with frequency shifts. The sensitivity of modified UiO-66 samples toward hexanal was also investigated, and the LOD was calculated using Stern–Volmer equation, giving 12, 22, 20, and 31 ppm for UiO-66-MA, UiO-66-ED, UiO-66-MMEN, and UiO-66, respectively. In another study, Xia et al. also explored MOFs by being the first to employ chemically and thermally stable zeolite imidazole framework (ZIF-8) as a fluorescent probe.⁷² ZIF-8 samples were postmodified using MMEN and N,N-dimethylaminoethylamine (MAEA). As a result, ZIF-MMEN and ZIF-MAEA showed turn-on effects toward 1-propanol and turn-off effects and emission frequency shift for hexanal and EBz. However, fluorescence was not observed when modified ZIF-8 (ZIF8@amine) samples were exposed to styrene and isoprene. With varying concentrations of hexanal, the LOD of ZIF-8 samples was found to be as low as 1 ppb with high efficiency, sensitivity, and good reproducibility. The two studies display great potential in detecting selected breath VOC biomarkers using MOFs. However, further practical research on breath samples should be done to determine its applicability in healthcare.

Mitrayana et al. used PAS technology to measure breath acetone obtained from 11 LC patients, 9 patients with other lung diseases, and 10 healthy volunteers.³⁰ The system was based on a CO₂ laser. High photoacoustic signal intensities were obtained at many CO₂ laser lines for acetone, along with ethylene and ammonia, which showed significant absorption values. Using various concentrations of acetone and ethylene artificially calibrated samples in laboratory setting, the LOD for ethylene and acetone was obtained at 6 and 11 parts per billion by volume (ppbv), respectively. Using the exhaled breath samples, the concentration range for acetone measured from the exhaled breath of the three groups was 200–450 ppbv range for healthy volunteers, 400–720 ppbv range for LC patients and 222–487 ppbv for patients with other lung diseases. In contrast, the ethylene concentration from the three groups fell in the range 39–201 ppbv. Therefore, it can be inferred that ethylene showed no substantial difference among the groups (p-value > 0.1), while acetone showed a significant difference between patients with LC and without LC (p-value < 0.01). Although this shows that acetone could be a potential biomarker for LC diagnosis, further research could be done to validate the reliability of the data by using a commercial device to compare with the PAS system.

The salt imbalance in the secretory cells found in patients with cystic fibrosis (CF) causes excessive sweating and swelling of the pancreatic duct where the dehydrated, thick mucus, and bacteria lead to cough and breathlessness.⁷³ It is estimated that 162 428 people are affected by CF across 94 countries.⁷⁴ Moreover, children suffering from CF are more likely to develop complications. Therefore, it is crucial to diagnose Pseudomonas aeruginosa (PA), which is the major cause of the condition. Hydrogen cyanide (HCN) was stated as a potential biomarker for PA, and HCN has been found in the oral cavity of adults.⁷⁵ Lauridsen et al. developed a SERS substrate comprised of silicon nanopillars coated with gold to detect HCN.⁷⁶ It was subjected to HCN (5 ppm) in N₂ gas and a serial dilution of potassium cyanide from 10 nM to 1 mM. The Raman signal was obtained using a dispersive smart Raman spectrometer. They reported that it was possible to differentiate samples with varying potassium cyanide concentrations down to 1 μM, and the detected LOD was 18 ppb. In another study, Lauridsen et al. detected HCN in exhaled breath using clinical samples obtained from the early and late stages of PA isolated from five patients.⁷⁷ The SERS signal showed a peak intensity at 2135 cm⁻¹, but only part of the strains with the lasR transcriptional gene could be detected. It suggested that the HCN biomarker is hard to detect in patients with late-stage PA infection because the mutation of the lasR gene in late-stage CF causes the HCN expression to be altered. However, although this technology has potentially proven to be a noninvasive method of monitoring early-stage PA infection in CF patients, more research using a larger sample size should be conducted to further validate the data.

Mastrigt et al. utilized a broadband QCL and a multipass cell to assess the repeatability of the measurements for exhaled breath VOC profiling and analyze its usability to differentiate 35 healthy subjects, 39 asthmatic patients, and 15 children with CF.⁷⁸ The QCL covered a wavenumber range of 832–1262.55 cm⁻¹ with a resolution of 0.05 cm⁻¹, focusing on the detection of hydrocarbons in breath. The reflected light was detected by a mercury cadmium telluride (MCT) detector. The repeatability test was conducted by finding the correlation between the first, second, third, and fourth measurements performed at time 0, after 30 min, 24 h, and 1 week, respectively, and they concluded there was poor repeatability with Spearman's rho values of less than 0.5 for all cases, which could be due to daily lifestyle-based activities. Despite this, it was possible to differentiate classes of children with CF by analyzing the spectral profiles using PCA. A group of VOCs was identified between healthy children and children with asthma in wavenumber ranges 1181.80–1182.55  cm⁻¹ and 1261.40–1262.05 cm⁻¹ and between healthy children and children with CF in 1260.70 and 1261.65 cm⁻¹. The authors claimed to be the first to use LAS to detect VOC profiles in exhaled breath samples. Although this setup achieved sensitivity in the range of parts per million by volume (ppmv), the specificity obtained by GC-MS and e-nose is higher. However, they used this technique for patient classification using PCA over the entire absorption spectra, which does not require high specificity. They also pointed out that the QCL-based method is relatively easy and fast for investigating the diagnostic and prognostic potential of volatiles in exhaled breath.

The outbreak of coronavirus disease 2019 (COVID-19) has spread worldwide, creating devastating consequences for humans and the economy.⁷⁹ COVID-19 is an enveloped, positive single-stranded RNA virus that is in the subfamily of Orthocoronavirinae, where its surface consists of “crown-like” spikes.⁸⁰ The pandemic has led to the need for fast and accurate mass screening tools, such as the rapid antigen test and the polymerase chain reaction (PCR) test, to identify infected patients to control their spread.⁸¹ However, these tests may cause discomfort to the user as it requires swabbing of the throat or nose. Hence, non-contact breath analysis based on the detection of VOCs in exhaled breath is promising.

Leong et al. developed a portable SERS-based breathalyzer that can identify COVID-19 in 5 min.⁸² The detection process and results are summarized in Fig. 4.

Breath samples were gathered from 501 volunteers (74 COVID positives and 427 COVID negatives). The exhaled gas interacted with the Ag nanocubes with 4-mercaptobenzoate (MBA), 4-mercaptopyridine (MPY), and 4-aminothiophenol (ATP) on the SERS substrate through hydrogen bonding, ion-dipole interactions, or π-π interactions. They reported that there was an increase in aldehydes (ethanal, heptanal, and octanal) and a decrease in methanol and acetone levels for COVID-19 positive patients, creating a distinct fingerprint profile to differentiate between individuals that are COVID-19 positive and negative. Additionally, a 3D PCA score plot of the SERS spectra was obtained from the absence and presence of the heptanal, methanol, and acetone biomarkers, producing the LODs 9, 190, and 90 ppb, respectively. By constructing a confusion matrix with the averaged classification outcomes across 50 model iterations, the sensor achieved a sensitivity of 96.2% and a specificity of 99.9%. Moreover, a binary classification model using partial least squares discriminant analysis (PLS-DA) showed a false negative rate much higher than rapid antigen tests and equal to PCR tests. Hence, this technology has a high potential for mass screening processes with its rapid, accurate, and portable capabilities.

Popa et al. investigated the respiratory issues of wearing a mask during the COVID-19 pandemic.⁸³ They measured ethylene and CO₂ concentration in breath using the PAS technology. The samples were collected from four healthy volunteers, and each sample was confined into an H-type resonant cylindrical cell to detect acoustic waves. Using a CO₂ laser, signals were first detected and normalized through four Knowles electret EK-3033 miniature microphones, where the acoustic signal was converted into an electrical signal and analyzed by a lock-in amplifier. It was reported that the photoacoustic signals for ethylene levels increased by 1.5% (0.018–0.027 ppm) after the volunteers wore the surgical masks for an hour. The authors pointed out that wearing a mask for a prolonged period may increase ethylene levels and above a certain threshold, there could be a possibility of exhaustion, loss of focus, and headaches. Therefore, further research should be conducted to verify the claim such as using other established methods as comparison along with a larger sample size.

Ethyl butanoate has been shown to be a potential biomarker for COVID-19 detection.⁸⁴ Alrowaili et al. theoretically investigated using one-dimensional photonic crystals (1D PCs) based on Tamm plasmon resonance (TPR) to detect ethyl butanoate.⁸⁵ The TPR structure was a cavity layer sandwiched between the metallic Au layer and 1D PCs. The incident light was passed through a dispersion device and a polarizer, and then it was directed onto a lateral face of the prism of the sensor structure, i.e., prism/Au/cavity/1D-PC. The reflected light was collected by a photodetector spectrometer. The proposed sensor was exposed to various concentrations of ethyl butanoate (ranging from 0 to 100 ppm) using simulations, resulting in an LOD of 3 ppm through mathematical calculations. Although this study presents theoretical investigations, it is important to address research gaps in future practical research.

Liang et al. recently proposed a new noninvasive method for detecting COVID-19 using breath samples, called CE-DFCS.⁴⁶ This method uses an optical cavity similar to CEAS systems, where a broadband MIR frequency comb laser is irradiated onto the sample to generate a broadband molecular spectrum, which was then detected by an FTIR spectrometer as described in a previous study.⁴⁷ The spectra were processed for binary diagnosis using two ML techniques: pattern-based recognition and molecule-based recognition. The spectra of 83 positive and 87 negative subjects, recruited based on PCR tests, were classified using ML, resulting in a high mutual agreement of 0.849 area under the curve (AUC) for the pattern-based classification. The authors concluded that this technology was robust and ultra-sensitive. With the enrichment of the molecular database, the ML approach could be more interpretable and applied to other breath diagnosis.

Chronic obstructive pulmonary disease (COPD), also known as chronic bronchitis or emphysema, describes the airway and lung disease that leads to airflow blockage and respiratory problems.⁸⁶ COPD may cause lung damage, airways being blocked by mucus or inflammation and swelling of the airway lining. COPD is the third leading cause of death worldwide, and people who are 40 years and above are mostly affected.⁸⁷ To diagnose COPD, bronchoalveolar lavage, sputum analysis, or transbronchial biopsy are widely used. However, such methods are often expensive and may pose considerable risks.⁸⁸ Hence, breath analysis of VOC biomarkers provides a noninvasive, low-cost, and safer alternative.

Kistenev et al. utilized two types of gas analyzers based on IR laser PAS (LPAS).⁸⁹ It is either combined with a CO₂ laser, referred to as LGA-2 or an OPO, referred to as LaserBreeze with a range of 2.5–10.7 μm. Using a mixed gaseous sample, the LOD was measured at 1 ppb for COPD-related VOCs such as acetone, acetylene, butane, ethane, ethanol, ethyl acetate, ethylene, methane, pentane, and propane. The exhaled breath of 11 healthy nonsmokers and seven patients with COPD was analyzed using LGA-2 and LaserBreeze LPAS gas analyzers to obtain the absorbed spectra range. Using PCA, the best informative subranges were found to be 9.2–9.8, 2.59–2.817, and 3.272–3.725 μm. This offered a promising base reference for future studies to further profile the absorption spectra of patients as a “fingerprint” through an exhaled breath.

Tuberculosis (TB) is an infectious lung disease caused by mycobacterium tuberculosis, primarily affecting the lungs but also damaging other body parts such as kidneys, intestines, the brain, or the spine.^90,91 Diagnosis of TB in children is especially challenging.⁹² Furthermore, the disease is more common in developing countries due to poor ventilation and overcrowding living conditions. Hence, there is an urgent need to develop affordable and real-time devices for the diagnosis of TB to prevent delayed patient treatment. As infections can cause changes in the host metabolism, distinct VOCs from mycobacterium tuberculosis can be detected in breath.^93,94

Bhattacharyya et al. used stable colloidal suspension of cadmium selenide (CdSe) QDs and CDs as a photoluminescent platform and FS to detect methyl nicotinate as a biomarker for TB.⁹⁵  Figure 5(a) shows the overall mechanism of the proposed method. Breath samples were collected and liberated into the sensing solution (QDs or CDs) mixed with an equal amount of 10 mM methyl nicotinate solution. Nitrogen gas was introduced to mimic the breath VOCs. The resulting samples in a vapor form were mixed with the sensing solution. The fluorescent signals were measured over time. A significant difference in emission and absorbance was observed upon mixing with methyl nicotinate, as shown in Figs. 5(b) and 5(c). The emission peak of pristine QDs was blue-shifted ∼60 nm after mixing, and the absorption band edge was blue-shifted 100 nm. Note that the same experiments were also performed for QDs mixed with pyridine to avoid the detection of other nitrogen-bearing aromatic compounds. Figure 5(d) shows the photographs of different solutions under visible light and UV light illumination. The color of the QD-nicotinate solution appeared to be dark brown under visible light and displayed different emission in comparison to pristine QDs upon UV light illumination. The authors have also found that the mixing of methyl nicotinate in the CDs solution resulted in quenching. In conclusion, the proposed platform displayed a novel device with tunable excitation and emission properties to selectively detect TB biomarkers. The development of an algorithm to correlate the concentration and measurable spectral features can potentially pave the way for rapid disease diagnosis.

As ethane is an important biomarker for oxidative stress, the prognosis for LC and other lung disorders is possible with the detection of ethane in breath.^96,97 Krzempek et al. developed a trace-gas sensor to measure ethane concentrations in a cylinder containing ethane nitrogen mixture using a distributed feedback (DFB) diode laser based on the principle of TDLAS and WMS.⁹⁸ A thermoelectrically cooled continuous wave laser was used with a wavelength of 3.36 μm (2976.8 cm⁻¹), housing the absorption line for ethane. A 630 nm visible laser for optical alignment was combined and inputted into a compact spherical shape multipass cell of 12.5 cm to produce an effective optical path length of 57.6 m, which was detected by an MCT detector. A sensor control board controlled the laser input current and temperature and produced a continuous sawtooth ramp spectrum at 8 Hz. Overall, the proposed system gave an LOD of 740 parts per trillion by volume (pptv) with an average time of 1 s based on HITRAN database and calibration using a lock-in amplifier.

Acetylene is widely analyzed as a hydrocarbon in industrial processes, but it is also used as a biomarker in breath to detect lung diseases. It can be detected based on a person's exposure to smoke within 4 hours but cannot be used as an indicator for passive smoking.⁹⁹ 

Marchenko et al. presented the development of a compact NIR distributed Bragg reflector (DBR) laser-based spectrometer using the principle of OA-ICOS for the simultaneous detection of acetylene and CO₂ at a wavelength of 1529.2 nm, and HCN at 1533.5 nm.¹⁰⁰ The laser was temperature controlled by a Peltier maintained at 25 °C and operated between 1527 and 1564 nm (6394 and 6548 cm⁻¹). The optical cavity had a finesse of ∼1560, a length of 25 cm, a volume of 310 ml, and an effective optical path length of 150 m. Additionally, the custom-made coupling system included an aluminum holder for the optical fiber and a collimating aspheric lens to allow precise and flexible adjustment of incident light parameters, such as the focal length, angle of incidence (max 6° off-axis angle), and beam position. As wavenumbers of CO₂ and acetylene are close at 6539.51 and 6539.59 cm⁻¹, respectively, they claimed that a single laser scan could be done to identify physiologically nonfunctional lung space while breathing. The detector used was an indium gallium arsenide (InGaAs) IR detector. The device sensitivity was 2.1 × 10⁻⁹ cm⁻¹ Hz^−1/2 and produced an LOD of 8 and 1.5 ppbv with an averaging time of 1 and 128 s, respectively. The calculated and the measured acetylene levels fit linearly; hence, it shows that the different peaks for the compounds can be detected by the system. It was mentioned that the coupling system allows robust optical alignment. Hence, it was suitable for field campaigns or hospital trials. Zou et al. developed a novel acetylene sensor by using a triple-row circular multipass cell (CMPC) to keep an optimum effective path length along with low cavity volume.¹⁰¹ The new CMPC had a minimum volume of 100.1 ml and an optical path length of 21.9 m, representing a high path-to-volume ratio value. The sensor introduced a DFB diode laser operating at 1.5316 μm into a multipass cell with WMS and used an InGaAs detector. To reduce data fluctuation and time required for a stable output, the distance between two minima in the second harmonic was used to normalize the maximum of it. Experimented with an artificially calibrated sample of acetylene of 60.1 ppm at a pressure of 1 atm for 0.9 h, a signal-to-noise ratio (SNR) of 56, a noise equivalent absorption sensitivity of 8.8 × 10⁻¹⁰ cm⁻¹ Hz^−1/2, and an LOD of 76.75 ppb were reported. As the proposed device is compact, rapid, and sensitive, this could be useful in breath diagnosis for smoking tests.

As of 2019, asthma was found in an estimated 262 ×  10⁶ people, with death tallying up to 455 000.¹⁰² Detection of fractioned exhaled nitric oxide has been the most common method used to diagnose asthma for breath analysis, where only nitric oxide is detected.¹⁰³ It was also reported that asthmatic individuals have low levels of glucose due to hypoxia-induced changes causing the breakdown of lipids like very low-density lipoprotein and low-density lipoprotein.¹⁰⁴ This could potentially increase acetone levels in asthmatic patients.¹⁰⁵ Acetone is a VOC that can be detected in exhaled breath. Nidheesh et al. designed a highly sensitive PAS system to analyze and detect acetone in ppb levels.^106, Figure 6(a) shows the setup of the PAS system where the 266 nm pulsed Nd-YAG laser was used as the excitation source and the energy was controlled by a polarizer attenuator with a reduced laser diameter using a lens combination as it traveled to the photoacoustic cell. There was a sample insertion module to evacuate the breath sample after every measurement. The photoacoustic signal was detected by a lock-in amplifier with a trigger input from the laser. Through Fourier-transformation, photoacoustic signals were converted into the frequency domain and categorized using PCA to distinguish breath samples from 15 asthma patients and 20 healthy volunteers as seen in Figs. 6(b) and 6(c). Using calibrated asthma samples, a Match/No Match analysis was done using Mahalanobis distance and sum-of-squared differences of stimulated and actual breath signals, which provided a sensitivity of 89% and a specificity of 93%. Using artificially calibrated samples of five concentrations of acetone, an LOD of 6.85 ppb was reported. Thus, this is a promising approach in detecting acetone and potentially other applications such as COVID-19 and pulmonary diseases. Smokers have found alternatives to reduce their nicotine addiction, such as using nicotine gums, patches, or e-cigarettes. Compared to tobacco cigarettes, e-cigarettes do not create smoke or tar and are be considered a safer alternative.¹⁰⁷ However, there are still many safety concerns regarding e-cigarettes as the smoke consists of toxic components such as ethylene oxide, which can cause damage to lung tissues.¹⁰⁸ Popa et al. investigated the ethylene content in exhaled breath of traditional smokers and e-cigarette smokers using the PAS technology.¹⁰⁹ A pulsed CO₂ laser was used as the light source. Photoacoustic signals were detected by four microphones and a lock-in amplifier. Breath samples from two smokers and one healthy subject were first measured to give a baseline for ethylene. For 15-min intervals, subjects inhaled traditional cigarettes followed by e-cigarettes. They reported that the mean concentration of ethylene decreased by 35% when traditional cigarettes were swapped with e-cigarettes. However, a 50% increase was observed when e-cigarettes were switched to traditional cigarettes. As ethylene concentration was much lower in subjects that inhaled e-cigarettes, it shows that it can potentially substitute traditional cigarettes. Still, future studies are required to validate this claim, such as testing with a larger sample.

Diabetes is a complicated long-term illness that requires continuous medical care and lifestyle adjustments to prevent the progression of acute complications. The current gold standard for diagnosing diabetes is through blood glucose tests, which require trained personnel and are time-consuming and invasive.¹¹⁰ Thus, studies have been focused on developing noninvasive methods for screening diabetes; however, many lack accuracy and specificity due to movement, skin irritation, or sweating.^111–113 With increased focus on breath biomarkers, the characteristic of breath was correlated with various diseases. For example, a noticeable acidic smell shows signs of diabetes. This is the cause of increased ketone production in the liver, which transports to the blood and is hence seen in the breath matrix. Breath acetone levels have also been reported to be at elevated concentrations for diabetes patients.^114–117 

For LAS technology, it was often measured in the infrared wavenumber range of 1150–1300 and around 1216 cm⁻¹ as this range is not significantly affected by the presence of other endogenous molecules like water, carbon dioxide, or methane. Reyes-Reyes et al. studied type 1 diabetes (T1D) by detecting acetone in breath using a QCL operating at the spectral range of 850–1250 cm⁻¹ at room temperature, a multipass cell of 50 cm length, and two MCT detectors placed on the same side as the QCL.¹¹⁸ This setup was previously proven to detect acetone in ppbv levels from 1150 to 1250 cm⁻¹. The spectral acetone concentrations were measured without the overlap with water in the same absorbance range by removal of the water peaks using the HITRAN database reference. They also calibrated the system for different concentrations of acetone and devised a linear fit model. From results obtained with one adult and two minor T1D subjects in comparison with one healthy control, it was found that acetone concentrations in T1D subjects are higher than those in healthy adults, but this was not always true for T1D minors. Hence, it was suggested that personalized studies for minor subjects are important. In addition, blood glucose and ketone concentrations were measured simultaneously. The authors claimed that the device could be a noninvasive, robust, and simpler alternative to detect ketosis in children for early diagnosis of T1D. They pointed out that it is possible to build a fast and cheap device for diabetes diagnosis in low-income countries. However, the system's specificity could be improved by using acetone traps, which can help reduce misclassification of diabetes.¹¹⁹ Schwarm et al. presented a calibration-free acetone sensor based on a DFB-QCL operating at 8.2 μm.¹²⁰ With WMS paired with a spectral fitting approach, the sensor was able to detect acetone in the presence of methane and water vapor. The laser was directed to a multipass cell generating an effective optical path length of 76 m in 0.5l volume. The reflected signal was detected by a photovoltaic detector. An LOD of 0.11 ppm for acetone detection using test reference gases was achieved. The authors also showed its application in the tracking of acetone levels during ketosis for human subjects implementing a ketogenic diet for 12 days associated with different lifestyle and dietary changes, which could be helpful in clinical studies of diabetes.

Ciaffoni et al. demonstrated a CEAS using a continuous-wave distributed feedback QCL (CW-DFBQCL) with a radiation frequency in the range 1215–1223 cm⁻¹ and obtained the spectra of acetone at different pressures and compared it with the results using Fourier transform infrared (FTIR) technology and MS.¹²¹ The optical cavity was a 10 cm long glass cell filled with a gaseous sample of high-performance liquid chromatography-grade acetone. The detector used was an MCT detector. Compared to FTIR spectrum, the effect of pressure broadening on the acetone spectra was negligible except at 1220 cm⁻¹ for pure acetone spectra at 6.9 Torr and acetone buffered with air at 700 Torr. They used a simulated breath sample with methane, water, and acetone to conclude that acetone spectra were overlapped majorly by the water vapor constituent of human breath as the system was used at 1216.5 cm⁻¹. To verify the accuracy of results, a water vapor trap was used and an LOD of 0.51 ppm was obtained. The study explored using the exhaled breath from a healthy volunteer to attain acetone concentrations similar to the measurements using an ion molecule reaction MS. It was mentioned that with further improvement in sensitivity and reproducibility, this could be a sensitive, robust, and compact device for acetone detection. Centeno et al. pointed out the interference of ethanol in acetone measurements and proposed a correction method.¹²² Their setup included a broadly tunable external cavity QCL (EC-QCL) in combination with OA-CEAS measurements. Figure 7(a) shows the EC-QCL configuration consisting of a QCL chip, a diffraction grating, reflective mirrors, collimating lenses, an optical cavity, and a photovoltaic detector. The EC-QCL was operated between 1120 and 1450 cm⁻¹ with an output power of 200 mW. The cavity length was 30 cm, with an effective optical path length of 1 km. The system was calibrated with a reference gas mixture containing 17 ± 0.06 ppmv of acetone in nitrogen to obtain a correlation of 0.95 against the calculated acetone as shown in Fig. 7(b). The authors compared the simulated absorption profiles of ethanol and acetone and concluded that the ethanol interference could result in an error of up to 39%. The suggested correction steps were to measure the reference signals for both calibrated acetone and ethanol mixtures before the actual sample analysis. The signals were fitted to the recorded signal for each wavelength scan. Since the relative concentrations of both were known, the absolute concentrations were calculated and represented as in Fig. 7(c), where a linear fit model gives the desired output for reference gases. Figure 7(d) shows a comparison between acetone measured without (black) and with ethanol correction (red), suggesting accurate measurement in the latter case. These results were validated with proton transfer–mass spectrometry. Similarly, Nadeem et al. developed a single mode widely tunable EC-QCL operating at around 8 μm with WMS.¹²³ The entire vibrational band of acetone at 1300–1420 cm⁻¹ was characterized. The LOD improved by two orders of magnitude than that of direct absorption spectroscopy with a value of ∼15 ppbv in 10 s and an NEAS of 1.9 × 10⁻⁸ cm⁻¹ Hz^−1/2. There were no clinical samples being tested, so the sensitivity and specificity of detecting acetone in an actual breath sample are unknown.

Non-QCL-based LAS technologies were also used for the detection of VOCs in breath. Hancock et al. developed a diode laser-based spectroscopy method along with a robust sample handling protocol for the detection of acetone in the breath as a cost-effective alternative to QCL-based spectroscopy.¹²⁴ The laser was operating at 1699 nm with an output power of 10 mW, and it was connected to the cavity through a single-mode fiber. The cavity was 45 cm long with a volume of 115 cm³, capable of detecting a minimum absorption coefficient on the order 10⁻⁹ cm⁻¹ with an LOD greater than 250 ppbv in a 2 ppmv acetone in N₂ sample. They also tested the device with breath gas sample collected in aluminized breath bags, cooled at 20 °C to remove the most amount of water, and the signal was measured. The cavity was then evacuated, and CO₂ and methane were introduced into it through a small molecular sieve, and the background was measured. These steps help in adequate sample handling removing the interference of water, methane, and CO₂ from breath in acetone measurements forming the absorption baseline. The acetone concentrations obtained were shown to be linearly correlated with the readings from MS. It was, however, pointed out that the interference of isoprene remained a challenge and could be corrected using a zeolite trap for acetone.

Revalde et al. used a modification of the CEAS principle with the development of a portable CRDS setup with real-time functionality and detected acetone concentrations in breath.¹²⁵ An Nd:YAG pulsed laser operated at wavelength 266 nm was incident on a 50 cm long absorption cell where the pulse decayed at a specific rate. The transmitted signal was measured by a photomultiplier tube (PMT) and an oscilloscope. First, acetone was calibrated in N₂ samples and an LOD of 10 ppb was derived. It was later tested on 40 individuals above the age of 50. Although the results in this study proved the functionality of the setup, its application on clinical breath samples needs a detailed study of the interference by other breath compounds. Xia et al. demonstrated sensitive measurements of acetone using a distributed feedback interband cascade laser (DFB-ICL) operating at around 3.37 μm with WMS features.¹²⁶ They used calibrated acetone mixtures in N₂ for measurements. The output of the DFB-ICL was split into two beams. The first beam was introduced into a Herriott cell with a cavity length of 44 cm and an effective path length of 15.8 m. The other beam was used for the tuning of IR laser parameter settings. The beam was detected by MCT detectors. An LOD of 0.58 ppm with 1 s averaging time was achieved, showing that the setup can be used in breath analysis as a pre-processing step to reduce the effect of methane and water. Tuzson et al. used a broadly tunable MIR VECSEL for the spectroscopic analysis of acetone in human breath.¹²⁷ The laser operated at the wavelength of 3.38 μm with an optimum chosen scanning rate of 500 Hz. It covered a spectral range of 2950–2980 cm⁻¹. The incident light was introduced to a low volume astigmatic Herriott multipass cell with an optical interaction length of 36 m requiring only 100 ml of breath sample, which was smaller than the volume of one human breath. The reflected light was detected in real time, and a multispectral analysis was carried out. The authors demonstrated a new technique to account for the power variation between each pulse of the laser. The technique was based on pulse normalization using a reference optical path by splitting the input laser to the reference and the sample cell and calculating the difference between them. It was shown that acetone could be detected at different water vapor levels up to 6% without any pretreatment. The LOD obtained using artificially calibrated samples with 1s averaging for these VOCs are 241 ppbv for acetone, 152 ppbv for isoprene, 63 ppbv for methane, 257 ppbv for ethanol, and 274 ppbv for methanol, with 5-min averaging time for acetone at 13 s, isoprene at 6.5 s, methane at 3 s, ethanol at 40 s, and methanol at 13 s. It was observed that acetone concentrations did not change with different concentrations of water, methane, and isoprene. Despite the overlap of these components, the proposed setup could measure acetone accurately without pretreatment. The authors conducted a pilot clinical study using the system in a follow-up study.¹²⁸ The changes in the concentration of acetone as a function of the negative energy balance of lifestyle changes of the subjects were studied. The results showed that breath acetone can indicate lifestyle changes.

While acetone detection has been studied extensively using LAS, SPR is one of the growing fields for high-sensitivity VOC detection. He et al. claimed to be the first to employ a localized surface plasmon resonance (LSPR) sensing platform using an MOF film, named HKUST-1, as a novel sensor to detect acetone, ethanol, and methanol.¹²⁹ The sensor was built by using a chemisorption process, where the tip of a multimode optical fiber was coated with gold nanoparticles (40 nm) and functionalized with a HKUST-1 MOF film through a layer-by-layer process. For optimal sensor performance, various coating cycles were experimented, whereby the 120-cycle-coated MOF sensor displayed sensitivity toward acetone, ethanol, and methanol of 13.7 nm/% (R² = 0.951), 15.5 nm/% (R² = 0.996), and 6.6 nm/% (R² = 0.998). Additionally, through mathematical calculations, the sensor exhibited an LOD of 0.005% (50 ppm), 0.003% (30 ppm), and 0.011% (110 ppm), respectively. Although the proposed sensor has the potential in detecting multiple VOCs, breath samples should be experimented for future biomedical applications. Usman et al. proposed the modeling and simulation of an SPR sensor toward detecting breath acetone for screening and tracking diabetes using a tunable polyaniline (PANI)-doped graphene composite.¹³⁰ It has largely negative real and close to zero imaginary dielectric constants, displaying promising plasmonic properties for biosensing applications. The simulation results showed a significant shift in the SPR angle (from 45° to 50°). Though the modeled composite SPR sensor performed better than conventional SPR sensors, thorough research on the performance validation for the exhaled breath of diabetic patients should be carried out to determine the LOD, sensitivity, and efficacy before implementing it in relevant biomedical applications.

Gastric cancer occurs when cells in the stomach proliferate at an abnormal rate. In 2020, there was an estimation of 1.1  ×  10⁶ new cases and 770 000 deaths, accounting for gastric cancer as one of the most commonly diagnosed cancers and one of the top four causes of cancer mortality globally.^131,132 It is a challenge to detect gastric cancer as traditional diagnosis methods such as gastroscopy or biopsies are often invasive and inconvenient.¹³³ Hence, it is vital to implement a rapid and noninvasive technique to diagnose the disease in its early stage.¹³⁴ 

Chen et al. developed a SERS sensor to distinguish different VOC biomarkers (acetone, isoprene, hexane, 2–3-dimethylpentane, 3-methylhexane, pivalic acid, 2-methylpentane, 2-methylhexane, 3-methylpentane, hexanol, phenylacetate, methanol, tetradecane, and dodecane) using simulated and actual breath samples from 56 healthy volunteers, 55 early gastric cancer patients, and 89 advanced gastric cancer patients.¹³⁵ The SERS sensor consists of a reduced graphene oxide incorporated onto an Au film for selective adsorption to detect the biomarkers. To provide the reacting sites, hydrazine vapor was adsorbed in the graphene oxide, and it served as a stabilizing agent. The biomarkers were obtained from the simulated breath (a combination of fluid biomarkers from GC-MS analysis), detected using the SERS sensor and processed using a statistical package for the social sciences software to establish 14 different bands as the fingerprint to differentiate healthy and cancerous patients. PCA was then used to compare the simulated and actual breath samples, giving a specificity of >92% and sensitivity of >83%, which implies that the sensor has excellent potential to diagnose gastric cancer effectively. Future work should focus on optimizing the SERS sensor and developing methods to increase its portability for clinical applications.

Taking inspiration from the thermal desorption tube typically used in GC-MS, Huang et al. created a tubular SERS gas sensor to detect and enhance the adsorption of ketone and aldehyde analytes in breath samples as seen in Fig. 8(a).¹³⁶ The sensor consists of a silver particle core in a core-shell composite covered with a uniform ZIF-67 shell. It is anchored with 4-ATP to increase the SERS signal with enhanced VOCs adsorption. Breath samples from 57 gastric cancer patients and 61 healthy subjects were collected and pumped into the capillary tube, which acts as the SERS sensor. The data were then produced as a barcode to be read by smartphone devices as depicted in Fig. 8(b). The averaged SERS spectra obtained from the respective volunteers were plotted into the graph as seen in Fig. 8(d). The sensor performance was evaluated using different BA concentrations, which gave a 9.82% relative standard deviation and an LOD of 3 ppb that was calculated using a triple standard deviation of the signal obtained from blank samples. Soft independent modeling by class analogy software was utilized to produce an orthogonal partial least squares discriminant analysis (OPLS-DA) model as shown in Fig. 8(e), depicting data points of the respective volunteer with a small overlap. A receiver operating characteristic (ROC) curve with an area under the curve at 0.9715 is depicted in Fig. 8(f). Additionally, the system gave a sensitivity of 91.23% and specificity of 88.52% with a total accuracy of 89.83% derived from sevenfold cross-validation. OPLS-DA was also used to differentiate the VOC biomarkers (methylglyoxal, cyclohexanone, butanone, octanal, decanal, hexanal, pentanal) where peaks predominantly exist in 1100–1250 and 1350–1650 cm⁻¹ regions. Moreover, it was observed that gastric cancer patients present higher Raman peak intensities. This could potentially be due to greater concentration of aldehyde, ketones, and other VOC molecules adsorbed onto the Ag@ZIF-67/4ATP composite. Figure 8(c) shows the respective barcodes converted from the SERS spectra, which could differentiate between healthy individuals and cancerous patients. This assisted in the data collection and analysis, along with system storage for clinical applications. Overall, this opens opportunities for future clinical breath analysis, aiding noninvasive, rapid, and user-friendly diagnosis.

Bacteria metabolites reveal information about their growth, development, and interaction with their environment.¹³⁷ Monitoring metabolites is essential to regulate bacterial growth and improve disease diagnosis. Various strategies have been implemented for the detection of bacterial metabolites. In this context, Wang et al. utilized the SERS platform to measure the VOC signals from bacterial metabolites to diagnose infection.¹³⁸ They used a low-cost plasmonic SERS substrate using anchored silver nanoparticles inside a Petri dish cover to capture the volatile metabolites produced from bacterial cultivation. Additionally, variable virus titers (lytic bacteriophage Phi6) were added to the Pseudomonas syringae bacteria culture to diagnose viral infection and monitor bacterial growth. Figure 9 depicts the sensing platform to capture volatile and nonvolatile metabolites using different SERS substrates that produce their respective spectra and the bacterial optical density (OD₆₀₀). A volatile metabolite, dimethyl disulfide, was detected from P. syringae with a strong peak at ∼680 cm⁻¹. By introducing lytic bacteriophage Phi6, a reduction in bacterial metabolic activity with a relative decrease in SERS intensity was observed. Using PCA-SVM confusion matrix, infection was predicted with a sensitivity of 92.8% and a specificity of 93.3%. Therefore, this method provides an alternative way to detect viable viruses using a portable Raman spectrometer, with the potential to track environmental conditions that influence microbial growth. Similarly, Kelly et al. detected dimethyl sulfide from six bacterial species (E. coli DH5α, E. coli K12 WT, S. aureus Cowan I, E. faecalis ATCC 15041, P. aeruginosa OA1, and B. fragilis NCTC 9343) cultured in a growth vial.¹³⁹ The SERS substrate (fabricated with either gold or silver nanoparticles) was installed onto the growth vial cap, and the volatile metabolite was detected using a hand-held Raman spectrometer. For rapid bacterial detection, Au-based substrates were found to be a better fit as they displayed a sensitivity limit for E. coli DH5 at 1.5 × 10⁷ CFUml⁻¹. They have also used an antibiotic, gentamicin, to treat the samples, proving that antibiotic treatment was effective in killing E. coli bacteria. Hence, this is a potential method for the rapid detection and screening of bacterial infection and treatment efficacy with antibiotics.

The current gold standard for the diagnosis of bladder cancer is cystoscopy and biopsy, which are often invasive, expensive, and time-consuming.¹⁴⁰ A wide range of human samples such as blood, tissue, and skin have been used to detect various VOCs for disease diagnosis. Compared to these samples, urine has the advantage of being easy and inexpensive to obtain. Moreover, several studies have shown the potential for urinary VOCs in cancer patients to provide insight into the urogenital tract and its systemic metabolism.^141–143 

Zhu et al. demonstrated a fluorescence sensor array for the early detection of urinary bladder cancer (UBC) VOC biomarkers (EBz, hexanal, lauric aldehyde, and nonanoyl chloride) by observing the changes in their fluorescence spectra.^144,145 They built the sensor array by spotting sensitive materials on a polyvinylidene fluoride (PVDF) film to detect potential VOC biomarkers. To validate the sensor as an automatic real-time system, five urine samples from UBC patients were analyzed by PLS-DA to give an overall sensitivity of 77.75% and specificity of 93.25%. Due to the trace levels of VOCs present in human urine, a PLS-DA plot of varying concentrations of nonanoyl chloride vapors (200 ppb, 100 ppb, 5 ppb, and 200 ppt) was used to test the sensor sensitivity. The proposed device was able to discriminate at ppb levels, but it had difficulties in distinguishing signal from noise at 200 ppt. Thus, it can be inferred that the LOD of the device is ∼5 ppb. However, due to the small sample population, the data obtained were insufficient to produce a PLS-DA prediction model. Due to the limited amount of sample size in the previous report, Zhu et al. carried out a follow-up study to establish a fluorometric optical sensor array to detect UBC patients with frank hematuria by sensing the same set of VOCs in real time.¹⁴⁴ They constructed the sensor using VOC-sensitive materials on PVDF films with excitation lights at UV (365 nm), blue (450 nm), and green (532 nm). To determine the sensor array performance, 38 UBC patients were recruited. Fluorescence signals of each VOC element were measured and analyzed by PLS-DA. The algorithm could accurately classify 68 out of 79 urine samples correctly with a sensitivity and specificity of 84.21% and 87.80%, respectively. Furthermore, the sensor array showed potential in distinguishing between high-grade and low-grade bladder cancer patients with a sensitivity and specificity of 66.67% and 75.00%, respectively. Though the proposed fluorescence sensor could successfully detect UBC VOC biomarkers, the limitation of the study was that only patients who were recently diagnosed with UBC were recruited. Using other analytical instruments in conjunction with a fluorescence array could potentially enhance the efficacy of the approach. It is also recommended to expand the study by recruiting patients from a larger multi-center cohort.

Bhattacharyya et al. investigated the detection of EBz in urine samples of UBC patients using a portable fluorometric sensor, NABIL. The device was fabricated using sensitive fluorophores such as Nile red (NR), Eosin Y (EY), and Rose bengal (RB).¹⁴⁶  Figure 10(a) depicts a schematic diagram of VOC collection from urine samples where a sensor strip was placed at the top of the urine sample holder via the proposed NABIL device. The NABIL is a multi-wavelength spectroscopic device with four UV LEDs each at 395 nm. They conducted the performance validation of the device by comparing the data from a clinical trial of 21 healthy and 52 UBC patients. Upon exposure to urine samples, UBC patients displayed multiple absorption maxima at ∼150–280 nm, with a characteristic absorption peak at ∼268 nm as seen in Fig. 10(b). The UV-Vis absorption spectra indicate the presence of unique organic compounds that are absent in control urine samples. Additionally, the NABIL displayed a significant increase in EY and RB fluorescence intensity as seen in Figs. 10(c) and 10(e). The blueshift emission in EY was observed as it acts as a photoredox catalyst for the bromination of EBz through a chain reaction as seen in Fig. 10(d). Figure 10(f) displays the potential mechanism for enhanced intensity in RB due to the dye acting as a photosensitizer, producing ion pairs. However, quenching was observed for NR as depicted in Fig. 10(g). This phenomenon could be explained in Fig. 10(h), which depicts the NR molecules favoring charge transfer state through a non-radiative deactivation channel. Through a calibration curve of specific EBz concentration, the NABIL device produced an LOD of ∼80 ppm with a sensitivity and specificity of 87% and 86%, respectively.

Other than cancer diagnosis, many studies have been focused on developing novel sensing mechanisms for the detection of various VOCs. A study by Iitani et al. developed a human “skin-gas cam” for real-time imaging of transcutaneous blood VOCs, ethanol, and acetaldehyde after alcohol consumption.¹⁴⁷ Ethanol concentration distribution was obtained by applying gaseous ethanol onto NAD-wetted alcohol dehydrogenase (ADH)-immobilized mesh to give a reduced form of NAD, i.e., NADH, as seen in Fig. 11(a). UV-LEDs were mounted on a camera lens to simultaneously excite and provide fluorescence imaging, causing NADH to release fluorescence, as depicted in Fig. 11(b). To accommodate for the irregularities on the skin and body surface, a two-dimensional pipe array (2D Mako) was utilized to equalize the distance between the body surface and ADH-immobilized mesh, as shown in Fig. 11(c). Figure 11(d) shows the changes in fluorescence rate when applying 500 ppm of ethanol onto the ADH mesh via the 2D structure. Moreover, changes in the rate of fluorescence can be observed at varying concentrations in Fig. 11(f). Additionally, Fig. 11(e) indicates the temporal changes with varying ethanol concentration, whereby an increase in fluorescence rate was observed with increasing concentration. The quantitative characterization of the system is presented in Fig. 11(g). Through spatiotemporal imaging, the ear of human subjects was ideal for metabolic monitoring as it was not affected by sweat. Overall, the system could be utilized for the investigation of the release mechanism of transdermal VOCs in local skin locations. The system was also capable of detecting ethanol and acetaldehyde from the human body after alcohol intake and was able to quantify 10 ppb of gaseous acetaldehyde determined through a calibration equation.

The differences between normal skin and melanoma are perceptible visually and through olfaction, which shows the presence of discriminable VOCs. Among other compounds present in different quantities in these skin samples, dimethyl disulfide was produced only by melanoma cells and, hence, can be used as a potential VOC for screening skin cancer.^148,149 Wang et al. used a CRDS setup to detect dimethyl disulfide as a biomarker to study melanoma or skin cancer.¹⁵⁰ They introduced a 266 nm laser with a repetition frequency of 1 kHz into a 43 cm long cavity. After passing through the sample gas, the transmitted signal was collected by a PMT on the other end, and the ringdown time was calculated. Dimethyl disulfide vapor was mixed with various concentrations of nitrogen to give a linear CRDS response. The proposed method displayed an LOD of 9 ppb at a temperature and pressure of 300 K and 100 Torr, respectively. Based on the laser wavelength and absorption cross section, the theoretical dynamic range of detection was from ppm to ppt levels. Hence, the compact design and rapid detection here shown with an artificially calibrated sample indicate a potential for diagnosing melanoma in a clinical setting.

Methane is known to exist in the intestine, and it increases with age. Therefore, breath methane has been correlated with bowel diseases such as diverticulosis and constipation.^151–153 Keppler et al. presented a CRDS setup to study the presence and concentration of methane in the human body.¹⁵⁴ Additionally, the stable isotope composition of methane in the exhaled breath was also studied to relate it to the physiological activity leading to methane concentrations. It was found that methane values were higher in exhaled breath than in the inhaled breath (surrounding environment) of 112 volunteers within the age of 1–80 years. The results had shown that the methane release varies tremendously between individuals in a range between 26 and 40 948 ppbv, with 20% of the subjects being high emitters of methane (>3 ppmv). They used Drierite and Ascarite II traps to remove the interference of water and CO₂. Though it has been established that methane emissions by the human body are due to microbial activity in the gastrointestinal tract, in this study, they hypothesized based on the emission concentration and carbon isotope signatures that methane is also produced by human cells endogenously.

Globally, breast cancer (BC) is the second most common prevalent cancer among women.¹⁵⁵ Typically, BC is detected via mammography, ultrasonography, and magnetic resonance imaging, but these detection methods give a high false positive result.^156,157 Hence, screening frequencies will be increased, and more time and resources will be spent on diagnosis. Therefore, it is vital to develop a noninvasive technique that can diagnose BC with high sensitivity and specificity.

In a proof-of-concept study, Kim et al. developed an SPR biosensor based on a highly sensitive and enhanced evanescent wave technique to detect formaldehyde, a biomarker for BC.¹⁵⁸ The detection depends on an improved Kretschmann configuration to obtain an enhanced signal transducer algorithm using a unique microfluidic gas channel to achieve a highly sensitive quantification, as seen in Fig. 12(a). The detection involved polyethyleneimine (PEI)-modified TiO₂ nanoparticles with amine ligands that were selective to formaldehyde, along with an Au thin film with TiO₂ nanoparticles, which amplified the signal. To simulate the exhalation, they investigated the sensor performance upon direct exposure to standard formaldehyde gas molecules. The SPR biosensor was able to respond to different formaldehyde concentrations in terms of the change in the reflectance (ΔR) within less than 10 s, and the LOD was proved to be 0.2 ppm, as shown in Fig. 12(b). A calibration curve was established between ΔR and gas concentrations, as shown in Fig. 12(c). The authors mentioned that the proposed SPR biosensor was 5–15 times more sensitive compared to other reported methods for formaldehyde detection. However, to establish the practical feasibility of biomedical applications, breath samples from BC patients should be studied in detail.

The liver is a crucial organ responsible for breaking down food and eliminating toxic substances. However, it can be susceptible to dysfunction caused by obesity, alcohol abuse, viral exposure, and genetic factors. Increased isoprene in exhaled breath can potentially be related to liver fibrosis, which is caused by various liver diseases such as viral hepatitis or chronic alcoholism.^159,160 It is important to diagnose such diseases at their early stages to avoid scarring of the tissue, leading to liver damage and liver failure, which can be fatal.

In a theoretical investigation, Mehaney et al. claimed to be the first to utilize 1D PCs toward sensing trace isoprene levels in exhaled breath based on TPR.¹⁶¹ The 1D PC gas sensor comprises GaN/SiO₂, with a thin layer of Au and an air cavity between exhaled breath to pass through and a prism to enhance reflection. An overall mechanism of the proposed method is depicted in Fig. 13(a). The plot of effective refractive index (RI) in exhaled breath as a function of isoprene concentration from 0 to 100 ppm can be observed in Figs. 13(b) and 13(c), whereby a relatively small increment in effective RI was observed, and it poses a significant concern in sensing small changes in gas concentration. This leads to further investigation of the device's capability in monitoring isoprene in ppm and ppb levels. In Fig. 13(d), the increase in isoprene concentrations from 0 to 100 ppm showed that the TPR was shifted toward a longer wavelength. This gives a great demonstration of the sensor sensitivity of 0.322 nm/ppm, which is better compared to other PC counterparts. Furthermore, the sensor was simulated with various ppb levels of isoprene. A sensitivity of 0.000 36 nm/ppb and LOD was determined to be 80 ppb through mathematical calculations.

Although this theoretical study showed potential in detecting small changes in isoprene at ppm and ppb levels, a thorough study using patient breath samples could be explored to validate the efficacy of the proposed method.

Schizophrenia (SCZ) is a severe mental illness that causes distortion of reality and how a person feels, thinks, and behaves, and it affects 1% of the population worldwide.¹⁶² Many of the previous studies on the detection of SCZ have employed invasive methods requiring the extraction of blood samples or cerebrospinal fluid.¹⁶³ There is an unmet need to develop a noninvasive and cheap method for diagnosis. Hence, breath analysis can be potentially used to detect the oxidative stress markers from the exhaled breath of SCZ patients as it is noninvasive and convenient with no patient discomfort.

Popa et al. developed a noninvasive LPAS method to identify and compare the oxidative stress markers, ethylene and ammonia.¹⁶⁴ They studied the exhaled breath gathered from 19 healthy volunteers and 15 patients with SCZ prior to and after treatment using levomepromazine (an antipsychotic medication). The photoacoustic waves were detected by four Knowles electrets miniature microphones, and the signal was amplified through a dual-phase lock-in amplifier. A modular software architecture was used to collect and analyze the data generated. In contrast to the values reported in other literature, they claimed that their PAS system has a high responsivity (405 cmV/W) and an LOD of 0.9 ppbv by using gas mixtures consisting of ethylene diluted in N₂ gas, which is more sensitive compared to others that could only detect 3.8 ppbv. The average ethylene concentration from SCZ patients was 0.07 ppm, which is higher compared to healthy subjects with 0.008 ppm. In addition, SCZ patients were found to have a lower average ethylene concentration (0.06 ppm) after 30 min of levomepromazine treatment, but it was still higher than the healthy subjects. Therefore, with the higher ethylene concentration obtained from the exhaled breath of SCZ patients compared to healthy individuals, this indicates the link to oxidative stress index in individuals with SCZ. It is an interesting approach to detect ethylene at much lower levels, as it provides a noninvasive and rapid measurement for clinical applications. However, further research should be conducted to explore other types of treatments, along with a larger sample size, to validate the accuracy of the data.

Renal disease is a condition where the kidneys are damaged with loss of function, which leads to the build-up of wastes and fluid in the body. The decreasing kidney functionality may inhibit acetone release in the kidney, causing a rise in its concentration in the blood, which is then released in the lungs.²⁹ It is a progressive disease that impacts more than 10% of the general population, which adds up to more than 800  ×  10⁶ people globally.¹⁶⁵ Previous reports indicate that oxidative stress is present in chronic kidney disease and even in the early stages, it is linked to loss of renal function requiring therapy.^166,167

Mitrayana et al. investigated the potential of ethylene and acetone as biomarkers using a multicomponent lab-built CO₂ laser-based PAS system.²⁹ The acoustic wave was analyzed through three Knowles EK 3033 microphones in the H-type cylinder photoacoustic cell that produces a signal, and it is detected by a lock-in amplifier. By using standard ethylene and acetone gases, they have reported that the LOD was obtained for ethylene at 8 ppbv and acetone at 11 ppbv. In addition, in a clinical study, a significant increase in acetone for 10 renal disease patients (0.628–0.941 ppmv) was observed compared to 30 healthy subjects (0.128–0.819 ppmv). However, no significant difference in ethylene was measured between the patients (0.167–0.213 ppmv) and healthy subjects (0.046–0.245 ppmv). This study indicates that acetone is a prospective biomarker for renal disease, but thorough future studies should be conducted to validate the data using other established methods and a larger sample size.

Among benzene-based VOCs, Basak et al. developed a quinoxaline-based fluorophore probe as a chemosensor to detect mesitylene, specifically in an aqueous medium.¹⁶⁸ The fluorophore probe was synthesized with a yield of 85%. The design introduced the electron-withdrawing-NO₂ group into quinoxaline, which enhances the charge transfer and probe sensitivity. The performance of the fluorophore probe was investigated using various benzene-based VOCs in water. The addition of mesitylene in an aqueous medium resulted in an 18-fold enhancement of fluorescence emission intensity. Furthermore, the probe possesses a 1:1 stoichiometric binding with mesitylene with a binding constant of 16.67 × 10⁸ M⁻¹, a quantum yield of 0.08 and an LOD of 2.66 ppm, determined through mathematical calculations. To prove the usability of the probe, river and lake water samples were analyzed, resulting in a mesitylene emission maximum of 481 nm. Applications for the detection of potential cancer biomarkers were also experimented by analyzing urine, intestinal body, and gastric fluid to indicate similar “turn-on” fluorescence enhancements. Overall, the probe was able to detect mesitylene in an aqueous medium and other biological fluids. This low-cost and simple-to-use technology could potentially be employed for practical mesitylene detection in biofluidic samples. However, further practical research should be done for healthcare applications. Arakawa et al. built a novel “sniffer-cam” as a 2D fluorometric imaging system that detects ethanol from human breath and skin.¹⁶⁹ The imaging system comprised a mesh substrate, multiple UV-LED excitation sheets, and a CCD camera. Due to S-ADH, an enzymatic reaction occurred on the mesh producing NADH. Using a 340 nm UV-LED for excitation, NADH will then produce a fluorescence signal at 490 nm, which was used to recognize ethanol vapor. By loading standard ethanol vapor at different concentrations, the calibration curve of the proposed device produced an LOD of 0.5 ppm. Moreover, imaging of transdermal ethanol vapor from the palm of a subject was observed after 15 and 45 min of alcohol consumption. Pixels from the palm image reflected the distribution of ethanol concentration based on alcohol metabolism. Therefore, this study offered a noninvasive ethanol imaging platform with high sensitivity that could be useful in monitoring diseases and analyzing metabolism. Due to the versatility of the NAD enzyme used in this system, the detection of other VOCs could be vastly explored for biomedical applications.

Using a similar sensing mechanism, Chien et al. fabricated a biochemical gas sensor (bio-sniffer) to sense breath IPA concentration by measuring the fluorescence intensity of oxidized NAD.¹⁷⁰ With many studies reporting the presence of IPA in several diseases, this system could serve as a noninvasive method for clinical diagnosis.^171–173 The bio-sniffer mainly comprises UV-LED for excitation and a PMT for detection. The gas sensing region consists of an optical fiber probe employed with a flow cell and an enzyme-immobilized membrane. Ten different VOCs that are structurally similar to IPA at 600 ppb were used to determine the sensor specificity, where a limited response was observed for primary alcohols and a higher response was observed toward racemic 2-butanol and (S)–(+)-2-butanol. To prove the usability of the device, breath samples from 30 healthy volunteers were measured to give a sensitivity of 1277.8 counts ppb⁻¹ and an LOD of 0.75 ppb, which was mathematically calculated. Therefore, the proposed device shows potential in being a noninvasive sensor that could be effective in detecting IPA in human breath for disease diagnosis.

Tabalina et al. showed an experimental setup based on a tunable QCL in the range of 5.3–12.8 μm, which covers the signature peaks of diagnostic biomarkers in the spectral range of 800–1800 cm⁻¹, such as ethane, pentane, and acetone.¹⁷⁴ A multi-pass gas cell with a 76 m long effective optical length was used to reduce the laser and optical noise and increase sensitivity. The output beam is collected with an MCT detector. Here, a database of 20 biomarkers was considered. Based on the known spectra of acetone, methanol, and ethanol, their concentrations were calculated using the Bayes estimator, parametric optimization, and double parametric optimization algorithms to deal with the problem of determination of multi-component concentrations in multispectral measurements of gas mixtures. They concluded that most accurate determination was obtained by Bayes estimator, which gave an LOD of sub-ppm levels. They used a membrane dryer to remove the water vapor in the artificially calibrated breath samples and claimed that real-time breath analysis could be done without much preparation. It was mentioned that the setup could be used for disease diagnosis, such as diabetes (acetone above 1.7 ppm), ethanol poisoning (ethanol above 130–650 ppm) or methanol poisoning (above 2 ppm), and kidney diseases (ammonia within 1–4.5 ppm). The biological fluid analysis was not for VOC detection and, hence, will not be covered in this review. Zhu et al. presented a plasma-enhanced infrared absorption spectroscopy (PEIRA) assisted with ML for the detection of multiple VOCs in the range of 2–25 μm.⁴⁵ The vibrational modes of VOCs, such as methanol, ethanol, and acetone, are present in the MIR region, as shown in Fig. 14(a). This region has low sensing accuracy due to the weak coupling to incident light. The coupling between the molecules and the plasma generated by the strong electric field greatly enhances the absorption of infrared light, also strengthening the fingerprint absorption peaks, hence, increasing the detection sensitivity and specificity. PCA was performed on the raw mid-IR data, based on its eigenvalue decomposition for visualizing the data of three VOCs, i.e., ethanol, methanol, and acetone, at various concentrations and in a mixed environment. A 3D space PCA proved to represent more information than a 2D space. The results showed clear distinctions between the three VOC clusters with a sensitivity in the range of hundreds of ppm. The system was further demonstrated for the detection of alcohol (ethanol) and diagnosis of diabetes using acetone detection, as shown in Figs. 14(b) and 14(e). Figures 14(b) and 14(c) show the representative spectrum from a healthy and a drunk subject, which is greatly disturbed by the CO₂ peak at 2400 cm⁻¹. This is absent in Figs. 14(d) and 14(e) because it is from an airflow mimic and not from the subject. The ethanol peak present at ∼2950 cm⁻¹ is shown in Fig. 14(c), indicating a drunk state. Similarly, the acetone peak at ∼1800 cm⁻¹ was observed only in Fig. 14(e), which shows an indication of a diabetic state. Figure 14(f) shows the overall summary of ML assisted PEIRA platform, indicating the feasibility for early and accurate detection of VOCs in healthcare diagnosis. The need to detect multiple VOCs with high sensitivity was studied by Liang et al. with the introduction of CE-DFCS in breath analysis.⁴⁷ The sample was introduced into a high finesse cavity and interacted with the frequency comb laser in the MIR region (3.4–3.6 μm). They measured different analytes separately and obtained a multi-element LOD in the ppt range for formaldehyde, ethane, methane, methanol, and carbonyl sulfide, and in the ppb range for ethylene. They tested the system with a breath sample collected in a Tedlar bag from one volunteer and fit the spectra with the HITRAN database to extract the concentration of different breath analytes. They detected 12.39 ppm of methanol and 8.52 ppm of methane and stated that the system can be used to detect other VOCs with more breath samples. Due to the high finesse cavity, they achieved a high sensitivity of 5 × 10⁻¹¹ cm⁻¹ Hz^−1/2 and high spectral resolution due to the use of a frequency comb laser.

Iwata et al. discussed an adaptation of UV for LAS for the detection of isoprene in breath samples.⁵¹ They used a laser-driven UV lamp instead of the conventional deuterium lamps to obtain high SNR. The light was focused into an aluminum film-coated silica-glass capillary based 3 m long hollow optical fiber. They obtained an LOD of 50 ppb using a nitric oxide calibrated sample. With the proposed system, breath measurements of several volunteers were carried out. In the resulting spectra, water vapor, isoprene, and ozone-converted oxygen were observed as the main components in the 200–300 nm region. The concentration of isoprene was obtained using multiple linear regression, and they concluded that the system can be used for simple, cost-effective, and real-time isoprene detection, which plays a primary role in cholesterol metabolism and indicates hyperlipidemia.

To improve the specificity of detecting target molecules in a mixed environment, Friberg et al. introduced the integration of GC and UV spectrophotometry. The VOCs and non-VOCs from breath samples were extracted and concentrated with a solid-phase microextraction (SPME) system and then separated using GC.⁵² This technique not only provides high sensitivity but also specificity with a range of analytes such as aliphatic aldehydes, ketones, pyridines, benzene, naphthalenes, and higher aromatics. Hence, it shows promise in potential clinical applications.

Gruber et al. used a similar technology to identify VOCs in breath, but they employed a 2D GC approach with improved sensitivity and selectivity. They compared their results with those obtained using 2D GC-time-off-light MS (GC-TOFMS) measurements.⁵³ They used needle trap microextraction (NTME) for VOC extraction in an off-line sampling of breath from a male volunteer. They tabulated the LODs of 12 VOCs present in exhaled breath using their reference compounds to obtain values that were 100 times better than TOFMS. Therefore, they concluded that the present technology could be useful in clinical breath analysis due to its highly selective and sensitive nature, as well as the possibility of miniaturization, compared to traditional GC-MS vacuum tubes. In a similar study, Zanella et al. evaluated the “Peppermint Initiative,” which is a standard for sample handling and analysis of exhaled breath analytics still in its preliminary phase.¹⁷⁵ They then followed up to prove that both testing technologies follow the guidelines of this standard. They concluded that, with this study and the previous study by Gruber et al., 2D GC-VUV has great potential in applications for clinical breath analysis with further research.⁵³ In another interesting study, Anthony et al. combined 2D GC with MS and VUV, as shown in Fig. 15(a).¹⁷⁶ They used samples such as alkanes, isomers, and perfumes. The GC effluents underwent VUV spectroscopy in the 125–240 nm detection range. They used a background spectrum collected in the absence of VUV light and a reference spectrum with the light on to get the final sample spectrum. The exhaust transfer line of VUV then entered the MS, which gave spectrometric readings based on electron ionization (EI). They observed that these detection techniques were complementary in nature, with alkanes like heptane, octane, and pentane identifiable using MS but not with VUV, as shown in Fig. 15(b), and isomers of dichlorobenzene (DCB) identifiable with VUV but not with MS, as shown in Fig. 15(c). Hence, by combining both, they obtained the best identification accuracy with coupling. Although this results in fewer false predictions, the design conditions, such as conductance efficiency for transfer from GC to VUV cell and then to EI-MS, are complicated. The cost and bulkiness of the system also increase with such analysis.

An interesting study was conducted by Popa et al. on the comparison between nasal and mouth breath analysis of common biomarkers of diseases, such as ethylene, methanol, ammonia, ethanol, and CO₂, done using LPAS.¹⁷⁷ Here, ethylene, methanol, and ethanol will be discussed. Breath samples from 15 healthy volunteers were introduced into a resonant photoacoustic cell and CO₂ laser, which was modulated by a chopper. This excites the gas molecules and gives rise to heat and pressure variation, producing an acoustic wave that was detected by four microphones and measured by a lock-in amplifier. Interestingly, in contrast to exhaled breath from the nose, breath samples from the mouth gave a higher average level of 6.3% for methanol, 8% for ethanol, and 6.9% for ethylene. This indicates that the VOCs are slightly higher in the mouth compared to the alveolar interface. Nevertheless, future studies can be done to make a comparison with more volunteers and keep the current research as a reference. Popa et al. investigated the potential of stress responses by analyzing ethylene from the exhaled breath of cancer patients during radiotherapy using LPAS.¹⁷⁸ They utilized a CO₂ laser and collected the output from a lock-in amplifier and a data acquisition module. Exhaled breath samples from nine cancer patients were taken into a sampling bag and analyzed by the LPAS to measure ethylene trace amounts before, immediately after, and at 15 min during radiotherapy. The increase in ethylene levels in patients undergoing radiation treatment established the presence of an oxidative attack. Hence, the ethylene biomarker for oxidative attack can be utilized to monitor stress in the human body and has the potential to evaluate the efficiency of radiotherapy. However, more comprehensive studies should be conducted to prove the biomarker correlation using a larger sample size. Tomberg et al. have presented a novel technique that utilizes a CEPAS detector combined with GC and a widely tunable EC-QCL, packed into a portable compact device, to identify and detect analytes in the optical fingerprint region.¹⁷⁹ In a mixed solution of alcohols in cyclohexane, the sample was injected into the GC, which passed through the sample loop using a carrier gas. A fraction of the column effluent flows into the CEPAS in a quasi-online manner. The periodic heating from the laser beam converts it into periodic pressure, which produces the PA signal that is detected by the microphone and collected by the lock-in detector. In the proof-of-concept experiment, they produced a model library prepared from an internal calibration standard in which the spectra of eight VOCs (cyclohexane, methanol, 1-propanol, ethanol, isobutanol, 1pentanol, 1-hexanol, and 2-ethyl-1-hexanol) matched the Pacific Northwest National Laboratory (PNNL) database results. Additionally, using N₂ as the makeup and flush gas, they obtained LODs for all VOCs: cyclohexane at 902 ppb, methanol at 57 ppb, 1-propanol at 73 ppb, ethanol at 64 ppb, isobutanol at 44 ppb, 1-pentanol at 109 ppb, 1-hexanol at 109 ppb, and 2-ethyl-1-hexanol at 161 ppb. Hence, due to its capability to provide a simpler interpretation of results in contrast to MS, coupled with its portability, proves to be a suitable choice as field instruments for non-experts. The identified challenges, such as the restricted operation at low temperatures and slow gas-exchange cycles, offer possibilities for future studies aimed at overcoming these limitations. Such studies could explore novel approaches and strategies to enhance the system's performance and extend its application range.

LSPR is a modified SPR that uses nanoparticles instead of a continuous metal film. LSPR-based biosensors provide several advantages, such as high linearity, simple instrumentation compared to SPR, and higher sensitivity toward changes in RI.¹⁸⁰ Sohi et al. modeled and simulated a wavelength-interrogated plasmonic biosensor with tunable sensitivity and selectivity for exhaled breath VOC biomarkers (cyclopentane, methanethiol, methanol, 2-pentanone, acetone, 2-butanone, octane, cyclohexane) for different types of cancers.¹⁸¹ The biggest challenge in plasmonic sensing is the narrow detection range of RI, as most SPR biosensors have a RI detection range of 1.32–1.41 in visible NIR spectrometry.^180,181 However, the highest RI of 1.41 is less than RIs of many VOC biomarkers. Thus, by employing perforated thin films, the benefit of both localized and delocalized surface plasmons could be explored. Additionally, to improve the detection range of biomarker RIs, variation in thickness and configuration of the gold layer was explored. As the gold layer thickness was decreased, SPR peak blue shifted. Moreover, the sensor with a 20 nm gold layer displayed the highest RI with an LOD of 1.426 toward the cyclohexane biomarker. As each breath VOC biomarker displays unique RIs, the 20 nm thick gold layer device was able to display resonance wavelength. In this study, the proposed plasmonic biosensor showed an optimized device with high LSPR peaks in the visible spectrum for a wide range of RIs of up to 1.436. With further research on patients, the LSPR biosensor could be a promising device in the biomedical and healthcare industry.

A comprehensive overview of the advancement in optical technologies for the detection of VOCs in various biomedical applications has been presented. The recent COVID-19 pandemic has led to a renewed interest in portable and noninvasive PoC devices for early-stage respiratory disease diagnosis. Optical technologies provide efficient, cost-effective, and sensitive detection of VOCs. In particular, the last decade has seen significant developments in SERS, FS, LAS, PAS, and SPR technologies, which have been extensively discussed. Below are the key highlights from each technology that have demonstrated low LODs supported by clinical validation and showed promising potential for commercialization. Note that all LODs were obtained using artificially calibrated samples in a laboratory setting or theoretical modeling.

Most established applications that monitor VOCs as biomarkers are for lung-related diseases. Qiao et al. developed a SERS sensor to detect p-EBZA molecules and showed an LOD of 1.9 ppb.^64,65 Zhou et al. developed a SERS nose consisting of sponge-like Cu-doped heterostructures that can detect LC VOCs such as PYR, 2-NT, and 4-EBZA below 10 ppb, which is ten times lower than the detection limit using the gold standard method of GC-MS.⁶⁶ Xia et al. have developed a SERS and FS dual-modality sensing system to detect and quantify BA with an LOD of 0.1 ppb.⁶⁷ One of the most promising clinical demonstrations of real-time detection using SERS technology was for rapid COVID-19 diagnosis by Leong et al.⁸² They developed a portable, hand-held SERS-based breathalyzer that can detect COVID-19 infections within 5 mins. The device was clinically tested by 501 volunteers. It was found that there is an increase in aldehydes (ethanal, heptanal, and octanal) and a decrease in methanol and acetone levels for COVID-19 positive patients. A sensitivity of 96.2%, a specificity of 99.9%, and LODs in the range of low ppb levels were demonstrated.

In one of the promising studies, Bhattacharyya et al. used FS to detect methyl nicotinate as a VOC biomarker for the early diagnosis of TB.⁹⁵ They employed a stable colloidal suspension of CdSe QDs and CDs as a photoluminescent platform and developed algorithms to correlate VOC concentration and FS spectral features. As a portable sensing platform, Yang et al. explored the use of a smartphone-based fluorescent detection method for IPA as a VOC biomarker for LC. Although high sensitivity up to 0.5 ppb was achieved in their proof-of-concept study, the specificity for IPA compared to other organic solvents was poor. Nevertheless, it has significant potential for translation to detect other VOCs.⁶⁹ Zhu et al. established a fluorescent sensor array for the early detection of UBC using VOC biomarkers (EBz, hexanal, lauric aldehyde, and nonanoyl chloride).^144,145 They accurately classified 68 out of 79 urine samples with a sensitivity and specificity of 84.21% and 87.80%, respectively. Basak et al. developed a quinoxaline-based fluorescent probe as a chemosensor to specifically detect mesitylene in an aqueous medium in exhaled breath and achieved an LOD of 2.66 ppm.¹⁶⁸ 

The presence of acetone in exhaled breath is an indicator of ketogenesis, a metabolic process in which the body burns stored fats for energy. Studies have found that the acetone concentration in the breath of diabetes patients is greater than 900 ppb.¹⁸² Ciaffoni et al. used a CW-DFB-QCL system operating at 1216.5 cm⁻¹ to detect acetone and reported an LOD of 0.51 ppm.¹²¹ Centeno et al. pointed out the interference of ethanol in acetone measurements and proposed a correction method.¹²² Their study utilized a broadly tunable EC-QCL in combination with OA-CEAS measurements. Revalde et al. developed a portable CALOS setup with real-time monitoring capabilities with an LOD of several 10 ppb and tested it on 40 individuals above the age of 50. Acetone levels in these patients were in the range of 0.1–80 ppm, which is suitable for human breath analysis.¹²⁵ Reyes-Reyes et al. conducted a clinical study in which they detected acetone in the exhaled breath of both control subjects and patients using a spectral range of 850–1250 cm⁻¹ at room temperature.¹¹⁸ The results showed that control subjects had acetone levels ranging from 0.39 to 1.09 ppm, while patients with type 1 diabetes exhibited acetone concentrations at the upper end of the healthy range. Xia et al. demonstrated for the first time the use of WMS combined with a DFB-ICL for the detection of acetone and achieved an LOD of 0.12 ppm with an averaging time of 60 s.¹²⁶ Wang et al. successfully reported the detection of skin cancer using dimethyl disulfide as a marker for melanoma, with an LOD of 9 ppb at a temperature of 300 K and pressure of 100 Torr.¹⁵⁰ However, the stringent experimental conditions may pose a limitation for its application as a PoC device. Methane has been correlated with bowel diseases such as diverticulosis and constipation. Keppler et al. used a CRDS setup to study the concentration of methane in human breath.¹⁵⁴ This study of 112 individuals, aged 1–80 years, found that methane concentration varies greatly between 26 and 40 948 ppbv, with 20% of the test subjects being high emitters of methane (>3 ppmv). Liang et al. utilized CE-DFCS, an advanced laser spectroscopic method that can collect a vast amount of molecular absorption features in real time. They tested exhaled breath samples from 170 patients to report the system's excellent discrimination capability for detecting COVID-19 infection. Their system demonstrated exceptional sensitivity in detecting volumes at parts-per-trillion levels using a pattern recognition approach.⁴⁶ Researchers have explored a promising multimodal approach where GC-MS, considered the gold standard in trace chemical analysis for separating VOCs through chromatographic processes, was combined with VUV spectroscopy that provides a greater level of specificity and sensitivity in detecting a wider range of analytes with high accuracy. Anthony et al. combined 2D GC, MS, and VUV measurements to provide a comprehensive analysis of VOCs in a mixed environment with high sensitivity and specificity because MS and VUV provide complementary information.¹⁷⁶ 

The study of early diagnosis of COPD using VOC biomarkers has been explored by Kistenev et al.⁸⁹ In the study, they employed an infrared LPAS system, the LaserBreeze gas analyzer, which operates in the 2.5–10.7 μm range and allows for the rapid detection of multiple VOCs in seconds. This technology has the potential to be used for disease screening in hospitals. Another study by Popa et al. developed a CO₂ laser-based LPAS system to detect markers of oxidative stress, such as ethylene in human breath, which is quite promising.¹⁶⁴ The study included 19 healthy subjects and 15 patients with SCZ and found that the mean ethylene level of SCZ patients was higher (0.07 ppm) than that of healthy subjects (0.008 ppm). The LOD of the system was 0.9 ppbv. Clinical detection of acetone using this platform can reach up to ppb levels.^29,30 Tomberg et al. improved detection specificity by combining GC and PAS, which provides a simpler interpretation of results compared to MS. The portability of this method makes it a suitable choice as field instruments, especially for non-experts.^59,179

Kim et al. developed an SPR biosensor to detect formaldehyde in human breath, a biomarker for BC.¹⁵⁸ The sensor achieved an LOD of 0.2 ppm in less than 10 s. However, it is important to note that while SPR technology is highly sensitive and versatile, it may face challenges in PoC diagnostics due to the interference of nonspecific bindings. Moreover, data obtained from biomolecular interactions, such as kinetic rate, may have limited significance in PoC analysis, as highlighted by Nguyen et al.⁶⁰ 

Table I summarizes the LODs of detection of VOCs for biomedical applications reported by various optical technologies in the references. Table II provides a list of the abbreviations used in this review paper. All the LODs were obtained using artificially calibrated samples in a laboratory setting or by theoretical modeling. The information of the type of samples used was also included (55 with clinical samples, 58 with non-clinical samples, and 3 based on theoretical study). It is encouraging that nearly half of the papers have been tested in a clinical environment indicating a strong potential for the clinical adoption of biophotonics technologies. The SERS technology has demonstrated the ability to achieve detection levels in the ppb range under clinical conditions,⁷² with a lowest LOD of 0.1 ppb for BA, a biomarker for LC.⁶⁷ The FS technology detected up to 1 ppb for hexanal, another LC biomarker.⁷² The PAS technology achieved an LOD of 1 ppb for most VOCs, whereas it could detect in the 10 ppb range of acetaldehyde in a clinical environment.⁵⁵ LAS technology achieved an LOD of 0.7 ppb for ethane, a biomarker for asthma.⁹⁸ The SPR technology reported a modest LOD in the sub-ppm range for formaldehyde, a biomarker for breast cancer.¹⁵⁸ It is clear that the technological advancements over the years have pushed the LODs of optical technologies close to the threshold for clinical disease diagnosis. In particular, SERS and FS technologies have shown promising results toward commercialization, with portable, robust, and sensitive devices for real-time PoC diagnosis. However, other optical technology platforms, such as PAS and LAS still have room for improvement, such as miniaturizing the systems and increasing the sensitivity for SPR.

The detection specificity of target molecules is a crucial aspect of breath analysis. To overcome this inherent challenge, there have been numerous promising studies, particularly those using SERS technology. For instance, Qiao et al. developed a novel technique utilizing a hollow Co–Ni layered double hydroxide nanocage on an Ag nanowire to trap gaseous benzaldehydes with high specificity by selective enrichment molecule p-ATP.⁶⁴ Zhang et al. created a dendritic Ag nanocrystal substrate that demonstrated high selectivity for detecting aldehydes through a p-ATP layer.⁶⁸ The porous MOF structure of the substrate allowed for a “cavity-diffusion” effect, enabling high specificity in distinguishing the target VOC from other benzaldehyde molecules. Similarly, Huang et al. employed a tubular SERS gas sensor with a core-shell composite covered with a uniform ZIF-67 shell with p-ATP to enhance the adsorption of benzaldehydes in breath samples, demonstrating high sensitivity, specificity, and accuracy.¹³⁶ In the case of FS, acetaldehyde and benzaldehyde were detected up to 10 and 1.2 ppb levels using QDs, respectively, in a clinical environment.^67,147 Additionally, Hancock et al. developed a cost-effective diode LAS method and a robust sample handling protocol by employing a molecular sieve to enrich and detect acetone in breath samples with high selectivity.¹²⁴ These experimental techniques show promise in improving the specificity of detecting particular VOCs in breath samples, which could have important implications for clinical breath analysis. Recent developments include multimodality approaches that achieve high specificity and selectivity in a clinical environment, such as GC combined with VUV spectroscopy^52,53,176 and GC combined with PAS.¹⁷⁹ 

Other studies have not used any specific enrichment steps but have utilized methods and data analytics to ensure the detection and quantification of the VOCs under study in a clinical environment. For example, Mitrayana et al. used PAS technology to measure breath acetone concentrations in healthy volunteers, patients with LC, and patients with other lung diseases.³⁰ Instead of using the 9R42 CO₂ laser line due to the strong absorption coefficients of ethanol and methanol at that line, and the relatively low power, they used the 10P20 line of the CO₂ laser, which is the strongest line, for detecting acetone. Ciaffoni et al. used a CEAS with CW-DFBQCL-based LAS study to obtain the absorption spectra of acetone within the 1215–1222 cm⁻¹ range at different pressures and compared the results with FTIR technology and MS.¹²¹ Xia et al. developed a FS-based detection method to detect gaseous BA.⁶⁷ They used PATP-modified GNRs in a porous MOF shell to produce a Schiff base complex, which led to an increase in the fluorescence signals of the hybrid system. Chien et al. created a biosniffer that exhibited good selectivity when tested with ten volatile organic compounds with a similar chemical structure to IPA at 600 ppb levels.¹⁷⁰ They observed only slight reactions to primary alcohols, such as ethanol, 1-propanol, and 1-butanol, while not significantly responding to acetaldehyde and certain ketones. Based on these findings, various conditions were studied and optimized, effectively demonstrating the bio-sniffer's performance. Currently, SPR lags in specificity in selectively enriching specific VOCs in complex exhaled breath samples, as they are all either theoretical or modeling-based on the reviewed work.

One of the critical challenges in realizing a practically relevant photonics-based VOC sensing device is the lack of sensitive yet inexpensive and portable optical components such as lasers and detectors. Specifically, the detection of VOCs using optical spectroscopy in the NIR and MIR range requires expensive detectors such as InGaAs detectors, MCT detectors, and PMTs.^101,150 Similarly, Raman and SERS technologies demand expensive spectrometers and Raman-stabilized lasers, which significantly limits their translation to cost-effective PoC devices.¹⁸⁴ Therefore, in-depth research into the development of miniaturized and sensitive optical components is needed to shift the paradigm for healthcare applications. For PAS, highly sensitive VOC detection requires high power pulsed and widely tunable lasers, such as OPOs, EC-QCLs, and CO₂ lasers.¹⁶ However, this is not suitable for translating the technology into a user-friendly and inexpensive PoC diagnosis platform. In this context, recent advancements in light-emitting diode-based photoacoustic imaging systems may provide a more realistic direction in implementing photoacoustic systems with cheap light sources to detect VOCs.¹⁸⁵ However, the sensitivity and LOD of such a system will have to be thoroughly investigated (Table I).

Another aspect that needs concerted effort is developing planar SERS substrates and colloidal metallic nanoparticles to achieve reproducible signal intensity in SERS and LSPR-based sensors. The development of SERS substrates with high Raman signal enhancement and low spatial signal variation (high reproducibility) is always a big challenge.¹⁸⁶ In this regard, advancement in realizing cheaper but reliable SERS substrate fabrication protocols and identifying novel nanomaterials that can be adopted to develop robust and reproducible SERS substrates should be studied in detail.¹⁸⁴ Photonic crystal fiber (PCF)-based SERS platforms may overcome some of these limitations. PCF-based SERS sensing offers high sensitivity due to the increased interaction length when the analyte is incorporated into the microstructured holes that run along the length of the fiber.¹⁸⁷ Beffara et al. recently demonstrated that the PCF-SERS sensing platform could offer the best reproducibility and repeatability in measurements with a relative standard deviation of 1.5% and 4.6%, respectively.¹⁸⁸ Such novel PCF designs could be explored for Raman and SERS-based detection of VOCs. Similarly, LSPR-based biosensors need significant improvement in the reproducibility of nanostructures in terms of size and shape.

Since human breath contains hundreds of VOCs in varying concentration, the detection of a specific VOC from a mixture can be quite challenging. Therefore, advancements in optical sensing techniques are imperative. Conventional absorption based optical techniques generally lack the specificity, which is essential to establish a particular VOC as a diagnostic biomarker. In this context, among the optical techniques, Raman spectroscopy is quite promising as it generates fingerprint spectra unique to a particular VOC, making it straightforward to differentiate biochemicals from a mixture. FS sensors could also be promising, as they generate unique spectra with specific stokes shift. However, to achieve improved specificity using other optical techniques such as LAS, SPR, or PAS, it is necessary to optimize the sensing surfaces that can selectively enrich VOCs. This includes developing specific chemical “filtering” layers that can partition VOCs based on their size or affinity, or by employing chemical capturing moieties to target a specific VOC or group of VOCs. A thorough investigation is required in realizing such “smart” sensing layers. For FS-based sensors, developing fluorophores with high quantum efficiency, high photostability, and specificity is crucial for achieving sensitive and specific VOC detection. One way to enhance sensitivity is by synthesizing organic fluorophores and small molecules that can selectively capture and enrich VOCs on the sensing surface, such as porous organic frameworks. In addition, easier and more flexible synthesis of these fluorophores will aid in the translation of the technology.

The analysis and unmixing of spectral data from breath samples for disease diagnosis demand for novel data analytics approaches. ML and DL techniques can be used to analyze such datasets to make disease diagnosis more robust. Spectral information obtained using various optical modalities can be unmixed into phenotypic descriptors to improve the diagnosis accuracy in patients with multiple diseases. This implies that large sets of data should be collected from each clinical study to make ML- and DL-based analytics robust. PoC devices integrated with ML and DL functions could offer opportunities for reducing hospital resources through telemedicine approaches.

Multimodal and multiplexed detection is another important trend, as a VOC can be a biomarker for various diseases, and a disease can have multiple VOC biomarkers. For example, acetone is detected in breath for LC, diabetes, liver diseases, and GC.^{30,135,171,189} This requires a detailed study of the pathobiology of various VOCs in breath to establish them as unique biomarkers. In this context, integrated multimodal approaches will add a new dimension to the accurate analysis of VOCs in clinical settings. For example, SERS, SPR, and FS can be integrated relatively easily due to their similar instrumentation requirements. The recent advancements of combining GC-MS with LAS is noteworthy, and it can be a game changer in VOC detection for clinical applications.¹⁷⁶ Such an approach will provide a sensitive detection platform with high specificity, which is crucial for detecting VOCs in breath. The simultaneous detection of multiple VOCs could also help to produce foolproof diagnostic results. Due to the inherent ability to differentiate biochemicals using fingerprint vibrational spectra, Raman spectroscopy is an ideal technique for multiplexed detection in a label-free manner.

From a practical point of view, to translate the technology for clinical use, researchers in the biophotonics field should work closely with clinicians from the beginning of instrument development. A multidisciplinary approach where researchers from various science and engineering domains work together is also necessary. The development of technology should be driven by addressing unmet clinical needs in detecting VOC biomarkers for a particular disease or group of diseases. The one-size-fits-all approach, where one modality is used to detect all VOCs with a relevant LOD, should be avoided. Instead, the focus should be on developing technology that is a “must-have” rather than a “nice-to-have” in order to achieve the ultimate goal of a cost-effective and commercially viable PoC sensor.

In conclusion, this review has highlighted the superiority of biophotonics technologies for the detection of VOCs. Each technology offers unique advantages, but the critical roadblocks toward clinical translation, such as sensitivity, cost-effectiveness, portability, and ease of use, can be overcome by relevant engineering and technical advancements in the respective field. We envision that such sensors will be accessible in every household in the near future and will lead to a paradigm shift in managing respiratory disorders and other diseases where fool proof VOC-based biomarkers can be established.

This research was supported by the Agency for Science, Technology and Research (A^*STAR), Singapore, ‘Translational Biophotonics Innovation Platform’ IAF-PP Grant No. H19H6a0025 and A^*STAR BMRC UIBR Grant.

The authors have no conflicts to disclose.

Keertana Vinod Ram, Ran Zhi Tong Chua and Jodie Chuan Yue Foo contributed equally to this work.

Shuyan Zhang: Conceptualization (equal); Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Keertana Vinod Ram: Conceptualization (equal); Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Ran Chua: Conceptualization (equal); Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Jodie Chuan Yue Foo: Conceptualization (equal); Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Jayakumar Perumal: Conceptualization (lead); Investigation (lead); Supervision (equal); Writing – original draft (lead); Writing – review & editing (lead). Dinish U.S.: Conceptualization (lead); Investigation (lead); Supervision (equal); Writing – original draft (lead); Writing – review & editing (lead). Malini Olivo: Conceptualization (equal); Supervision (lead).

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