Histopathology has remained the gold standard for surgical margin assessment for decades. However, routine pathological examination based on formalin-fixed and paraffin-embedded (FFPE) tissues is laborious and time-consuming, failing to guide surgeons intraoperatively. Here, we propose a rapid, label-free, and non-destructive histological imaging method, termed microscopy with ultraviolet single-plane illumination (MUSI). With intrinsic fluorescence from deep ultraviolet excitation, MUSI enables both ex vivo and in vivo imaging of fresh and unprocessed tissues at the subcellular level with an imaging speed of 0.5 mm2/s, generating high-quality optically sectioned surface images from irregular surgical tissues with a long depth-of-field. We demonstrate that MUSI could differentiate between different subtypes of human lung adenocarcinomas (e.g., lepidic, acinar, papillary, and micropapillary), revealing diagnostically important features that are comparable to the gold standard FFPE histology. As an assistive imaging platform, MUSI can provide immediate feedback to surgeons and pathologists for intraoperative decision-making, holding great promise to revolutionize the current clinical practice in surgical pathology.

Histopathology has remained the gold standard for surgical margin assessment for decades. However, routine pathological examination based on formalin-fixed and paraffin-embedded (FFPE) tissues is laborious and time-consuming, failing to provide immediate on-site feedback to surgeons for intraoperative decision-making. Although frozen section can serve as a rapid alternative to FFPE, it still requires a turnaround time of 20–30 min during surgery.1 In addition, the frozen section is subjected to an inadequate sampling of resection margins, and freezing artifacts are inevitable when dealing with lipid-rich tissues. All these issues will affect the histopathological interpretation and diagnostic accuracy.2 

Recent advances in optical microscopy enable rapid and slide-free imaging of thick and unprocessed fresh tissues, greatly streamlining the current practice in FFPE histology.3 Imaging modalities based on exogenous fluorophores, including fluorescence confocal microscopy,4–6 light-sheet microscopy,7,8 structured illumination microscopy,9,10 and microscopy with ultraviolet (UV) surface excitation,11,12 can provide a sufficient sampling of large resection margins with highly specific cellular features within a point-of-care timeframe. However, the use of fluorescent contrast agents is usually prohibited during intraoperative procedures due to safety reasons. In addition, the staining process may interfere with the subsequent molecular assays, such as fluorescence in situ hybridization and DNA/RNA sequencing.8 In contrast, label-free imaging modalities based on intrinsic contrast mechanisms are highly desired in clinical settings. Reflectance-based methods, such as optical coherence tomography13,14 and reflectance confocal microscopy,15,16 have successfully been used in ophthalmology and dermatology for years. However, they reveal very limited cellular contents within internal organs, thus causing an inconsistency with histochemical staining in routine pathology. The same problem occurs in contrast from autofluorescence excited in the long wavelength range.17,18 A recently proposed label-free imaging technique greatly strengthens the details of cellular features by leveraging the contrast from dark-field reflectance.19 However, it is subject to tissue-dependent optical sectioning thickness, which is not ideal for thick tissue imaging. Other imaging platforms, such as UV-excited photoacoustic microscopy20–22 and nonlinear microscopy23–26 (e.g., coherent Raman scattering, multiphoton absorption, and harmonic generation), can also achieve promising results in label-free tissue assessment. However, these systems usually require sequential scanning of a tightly focused beam during image acquisition, which limits the imaging speed and impedes their applications in the rapid screening of large-area tissues. The comparisons between different imaging modalities are listed in Fig. S1.

Our recently proposed method, termed computational high-throughput autofluorescence microscopy by pattern illumination (CHAMP),27 can achieve rapid and label-free imaging of thick and unprocessed tissues with enhanced nucleic features via deep-UV excitation. However, the depth-of-field (DOF) of CHAMP is restricted to 80 μm, which is insufficient to tolerate large surface irregularities presented in grossly cut tissues, causing an out-of-focus blur in the reconstructed images. In addition, CHAMP achieves optical sectioning by leveraging the shallow penetration depth of deep-UV light, which varies between different types of tissues.12 This causes slight deviations between CHAMP and slide-based FFPE histology. To overcome these limitations, here we propose a new imaging method, termed microscopy with ultraviolet single-plane illumination (MUSI). Similar to light-sheet microscopy, MUSI rejects the out-of-focus fluorescence by illuminating a specimen with a thin sheet of light and detects the resulting signal from an orthogonal direction. The dual-axis configuration of MUSI decouples the illumination from detection beam paths, overcoming the inherent trade-off between long DOF and high spatial resolution in a conventional epi-illumination setup. In addition, with enhanced cellular features revealed by deep-UV-excited autofluorescence, MUSI enables label-free and non-destructive imaging of fresh and unprocessed tissues at an imaging speed of 0.5 mm2/s, achieving a spatial resolution of 2 μm (lateral) and 2.8 μm (axial) with a long DOF up to ∼200 μm. This guarantees that high-quality optically sectioned surface images can be obtained from irregular surgical tissues. We demonstrate that MUSI can differentiate between different subtypes of human lung adenocarcinomas (n = 15), revealing diagnostically important features that are comparable to the gold standard histology. Our results suggested that MUSI has great potential as an assistive diagnostic tool that can be used by surgeons and pathologists for intraoperative decision-making.

For ex vivo imaging, internal mouse organs, including the liver, brain, kidney, lung, spleen, skin, muscle, and tongue, were harvested immediately after the mice (C57BL/6 type) were sacrificed. After extraction, these organs were either kept intact or manually cut into 3- to 5-mm-thick tissue slabs for imaging without any further processing. For in vivo imaging, the mice were supinated on the sample holder and anesthetized with 3% isoflurane during experiments. To expose the targeted brain and kidney tissues for imaging, a 5-mm by 5-mm cranial window was made on the skull and a small incision was made on the left abdomen. For human lung tissues, the specimens were obtained from lung cancer patients who underwent curative lung cancer surgery at the Queen Mary Hospital. Following lung lobectomy, the cancer tissues were cut with a scalpel from the resected lobe and, subsequently, formalin-fixed and transported to the lab for imaging. After imaging, all the specimens were histologically processed with a standard protocol to obtain the hematoxylin and eosin (H&E)-stained images. Specifically, the specimens were fixed in 4% neutral-buffered formalin at room temperature for 24 h and processed for dehydration and infiltration using a tissue processor (Revos, Thermo Fisher Scientific, Inc.) for 12 h. After that, the specimens were paraffin-embedded and, subsequently, sectioned into 4-μm-thick tissue slices using a microtome (RM2235, Leica Microsystems, Inc.). Finally, the sectioned thin tissue slices were mounted on glass slides, stained by H&E, and imaged using a digital slide scanner (NanoZoomer-SQ, Hamamatsu Photonics K.K.) to generate the corresponding histological images. All animal experiments were carried out in conformity with the guidelines and protocols approved by the Health, Safety and Environment Office of the Hong Kong University of Science and Technology (HKUST) (license number: AH18038). All human experiments were carried out in conformity with a clinical research ethics review approved by the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (reference number: UW 20-335). Informed consent was obtained from all lung cancer tissue donors.

1. 4T1 cell culture

The mouse breast cancer 4T1 cell line was purchased from American Type Culture Collection (ATCC, Manassas, VA, USA). The cells were grown in Roswell Park Memorial Institute 1640 medium supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin solution and cultured at 37 °C in a 5% carbon dioxide incubator. Subconfluent 4T1 cells were trypsinized and resuspended in phosphate-buffered saline (PBS). 10 μl of the resuspension was mixed with trypan blue, and the number of 4T1 cells was counted using an automated cell counter (Countess™ 3 FL Automated Cell Counter, Thermo Fisher Scientific, Inc.). A cell resuspension with a final cell density of 1 × 106 cells/ml was prepared in PBS.

2. 4T1 orthotopic allograft

6–8 week-old female BALB/c mice were used to generate the 4T1 cell allograft mouse model. Hair around the injection site, the left abdominal mammary gland, was removed using hair clippers and was sterilized with three alternating swabs of betadine and 70% ethanol. 1 ml tuberculin syringe with a 27 gauge needle was used to inject 200 μl of 4T1 cells (2 × 105 cells) subcutaneously. A palpable primary tumor was developed and observed after one week. The mice injected with the cells were euthanized by carbon dioxide asphyxiation three weeks after the injection of 4T1 cells. The organs and tissues were then harvested.

The setup of the MUSI system shares similarities with the previously reported open-top light-sheet configurations,7,28,29 which offer great flexibility to manipulate large and thick samples without the use of immersion objectives. As shown in Fig. 1(a), the system consists of two orthogonal beam paths under the specimen that sits at 45° with respect to the optical axes. A customized 20-mm-diameter UV-transparent quartz hemisphere (n = 1.45, Worldhawk Optics Ltd.) is held beneath a quartz-bottom sample holder (1-mm thickness) to mitigate imaging aberrations induced by oblique detection at the air–holder interface.7 Both illumination and detection paths focus perpendicularly through the hemisphere and conjugate at the sample holder. A thin layer of UV-transparent oil (No. 56821, Sigma-Aldrich Corp.) is applied in the space between the holder and the bottom surface of the tissue to achieve optimal index matching. For the illumination arm, a 266-nm laser is used as the excitation source (WEDGE HF 266 nm, Bright Solutions Srl.), which is spectrally filtered using a bandpass filter, F1 (FF01-300/SP-25, Semrock, Inc.), and expanded using a pair of lenses, L1 and L2 (LA4647-UV and LA4874-UV, Thorlabs, Inc.). Then, the beam is propagated through an adjustable slit aperture, SA (VA100C, Thorlabs, Inc.), outputting a beam with a size of 2 × 6 mm2, which is subsequently focused using a UV cylindrical lens, CL1 (LJ4395RM, f = 100 mm, Thorlabs, Inc.), and generates a 2-mm-wide Gaussian light sheet with a waist radius (w0) of 2.8 μm and a DOF (2ZR) of 190 μm (Fig. S2). For the detection arm, the excited intrinsic fluorescence is collected using an achromatic UV objective lens, OL (LMU-5X-NUV, NA = 0.12, Thorlabs, Inc.), filtered using a long pass filter, F2 (BLP01-325R-25, Semrock, Inc.), subsequently refocused using an infinity-corrected tube lens, TL (TTL180-A, Thorlabs, Inc.), transmitted through an additional low-power cylindrical lens, CL2 (f = 2000 mm, Worldhawk Optics Ltd.), and finally imaged using a scientific complementary metal–oxide–semiconductor camera (pco.edge 4.2, 2048 × 2048 pixels, PCO, Inc.). The specimen is raster scanned using a three-axis high-speed motorized stage (L-509.20SD00, PI miCos GmbH) with a maximum traveling range of 5 cm. The specimen is scanned through a static light sheet at a constant velocity of 0.25 mm/s along the primary scanning direction [x axis in Fig. 1(b)], and the images are recorded at 250 frames/s with a sampling pitch of 1 μm/pixel on the camera plane. The image height can be adjusted according to the surface irregularities of the imaged specimen, with a maximum tolerance of ∼200 μm in the current design. Followed by the primary scanning, the specimen was translated laterally [y axis in Fig. 1(b)] at an interval of 1.8 mm, causing a 10% overlap between adjacent image strips for stitching a final image with a large field-of-view (FOV). The raw images were stored at 16-bit depth and transferred through a Camera Link interface at a streaming rate of 2000 pixels (w) × 200 pixels (h) × 2 bytes × 250 frames/s = 0.2 GB/s to a local workstation equipped with high-speed solid-state disks (970 EVO Plus, Samsung Electronics Co., Ltd.). The image acquisition and stage scanning were synchronized using our lab-designed LabVIEW software (National Instruments Corp.). The design of MUSI allows label-free and slide-free imaging of fresh and unprocessed tissues at an imaging speed of 0.5 mm2/s with an in-plane resolution of ∼2 μm (Fig. S3), producing histological images with morphological features that can be easily interpreted by pathologists for intraoperative decision-making.

FIG. 1.

Overview of MUSI imaging. (a) Schematic of the MUSI system. (b) Image processing pipeline for MUSI imaging. F: filter; L: lens; SA: slit aperture; CL: cylindrical lens; OL: objective lens; and TL: tube lens.

FIG. 1.

Overview of MUSI imaging. (a) Schematic of the MUSI system. (b) Image processing pipeline for MUSI imaging. F: filter; L: lens; SA: slit aperture; CL: cylindrical lens; OL: objective lens; and TL: tube lens.

Close modal

Figure 1(b) demonstrates the image processing pipeline for MUSI. The recorded images within an image strip are loaded as a stack (data volume) into MATLAB (MathWorks, Inc.) for processing. Note that the tissue geometry reconstructed by the raw data volume will be distorted since the light sheet images are recorded at 45° with respect to the tissue surface. To correct this distortion, the raw data volume is sheared by 45° in the xz plane to create a trapezoidal data volume.7,8 After that, an extended DOF algorithm is implemented on the sheared data volume through a Fiji plugin30 to extract the intact tissue surface of this image strip. This procedure is repeated until all the image strips are processed. Then, the extracted surfaces of adjacent image strips are registered and stitched by the grid-stitching plugin in Fiji31 to produce a large-area tissue image. The processing is jointly completed in MATLAB and Fiji through micro scripts. The algorithm is run on a workstation with a Core i9-10980XE CPU @ 8 × 32 GB RAM and 4 NVIDIA GeForce RTX 3090 GPUs, which takes ∼15 min to process 1-cm2 tissue area with 200-μm surface irregularity (∼40-GB data size). To extract the distributions of nuclear features, the MUSI images are segmented and binarized to acquire the cross-sectional area and centroid of each cell nucleus. With the localized center positions of cell nuclei, the intercellular distance is calculated to be the shortest adjacent distance to a neighboring cell nucleus. To generate virtually stained MUSI images for pathological evaluation, a weakly supervised deep generative network (WeCrest32) is used. The weights of the generators and discriminators are initialized by Kaiming initialization and normal distribution, respectively. All bias is initialized as zero, and the model is optimized using an Adam optimizer. The learning rate is fixed as 0.0002 for epochs of 0 to 40 and then linearly decayed to zeros for epochs of 41 to 80. The batch size is set as 2.

The spatial resolution of MUSI is measured by imaging agarose-embedded sub-diffraction fluorescence beads (B200, 200-nm-diameter, λem = 445 nm, Thermo Fisher Scientific, Inc.) across a volume of 0.5 × 1 × 0.05 mm3 [Fig. S3(a)]. The full-width at half-maximum (FWHM) of the measured point-spread functions (PSFs) are varied with spatial positions. It is shown that the PSF will deteriorate along the propagation of the light sheet [Figs. S3(c) and S3(d)], and this is more evident at the edge of the imaging FOV [Fig. S3(e)]. The FWHM of the PSF located at the beam waist and center of the imaging FOV [PSFi, Fig. S3(c)] is measured to be ∼2 μm (lateral) and ∼3 μm (axial), which is sufficient to reveal subcellular features (e.g., cell nuclei) and produce an optical section instead of a physical section in slide-based FFPE histology.

The formalin-fixed mouse brain [Figs. S4(a)–S4(d)] and kidney tissues [Figs. S4(e)–S4(g)] are imaged to validate the performance of MUSI initially (Fig. S4), since the fixed tissues are less deformed than fresh and soft tissues during imaging. The tissues are fixed at room temperature for 24 h and manually sectioned into 3- to 5-mm-thick tissue slabs for imaging and, subsequently, histologically processed to obtain the corresponding H&E-stained images for comparison. The cell nuclei located at the hippocampus [Fig. S4(b)], olfactory area [Fig. S4(c)], and isocortex [Fig. S4(d)] in the mouse brain are visualized with a negative contrast in the MUSI images, showing high accordance with the corresponding H&E-stained images. In addition, MUSI provides well-characterized structures in the mouse kidney, such as glomerular capsules [Fig. S4(f)] and renal tubules [Fig. S4(g)], in which densely packed cell nuclei can be clearly identified. The functional layers of the cortex, outer medulla, and inner medulla in the kidney are well recognized by MUSI based on intensity variations.

Figure 2 further demonstrates the potential of MUSI for label-free and non-destructive imaging of fresh and hydrated tissues. The freshly excised mouse tissues, including the liver [Figs. 2(a)2(c)], brain [Figs. 2(e)2(g)], and kidney [Figs. 2(h)2(j)], are manually sectioned into thick tissue slabs for imaging, after which the tissues are histologically processed to generate the corresponding H&E-stained images for comparison. Note that the FFPE thin slice is not able to exactly replicate the surface imaged by MUSI due to the difference in imaging thickness and tissue orientation due to tissue processing. During experiments, an image height (h) of 100 μm is sufficient to accommodate irregular surfaces presented on fresh tissues. We observed that the UV penetration depth typically varies between 5 and 30 μm, depending on different tissue architectures and degrees of tissue scattering. For instance, UV penetrates differently in regions with densely packed cell nuclei (e.g., primary tumors), lipid droplets (e.g., subcutaneous tissues), or large intercellular spaces (e.g., lung alveoli). The morphology of hepatocytes in mouse liver is characterized at a depth of 10 μm below the tissue surface, showing great accordance with the H&E-stained images although the sinusoidal capillaries are less visualized by MUSI [Fig. 2(b)]. The nuclear features, such as cross-sectional areas and intercellular distances, are extracted from both MUSI and FFPE histology for comparison. The statistical results [Fig. 2(d)], which are calculated from 50 hepatocytes selected from both MUSI and H&E-stained images in Fig. 2(c), suggest that the cellular features extracted by MUSI agree fairly well with the clinical standard method. In the mouse cerebellum, a clear separation between the molecular layer and the granular layer with the intermediate Purkinje cells can be visualized in both MUSI and H&E-stained images [Figs. 2(f) and 2(g)]. Similarly, the features of renal tubules [Fig. 2(i)] and glomerulus [Fig. 2(j)] in the kidney revealed by MUSI are remarkably similar to that by conventional FFPE histology, although Bowman’s space in renal corpuscles is less visible in the MUSI image.

FIG. 2.

Ex vivo MUSI imaging of fresh mouse tissues. (a) MUSI image of a fresh mouse liver; the inset at the bottom left shows the photograph of the specimen. (b) and (c) Zoomed-in MUSI and the corresponding H&E-stained images of the orange and green solid regions marked in (a), respectively. (d) Distributions of nuclear features extracted from (c). Wilcoxon rank-sum testing is carried out across groups with n = 50 for each distribution. The significance is defined as p* ≤ 0.05 in all cases. (e) MUSI image of a fresh mouse brain; the inset at the bottom left shows the photograph of the specimen. (f) and (g) Zoomed-in MUSI and the corresponding H&E-stained images of the orange and green solid regions marked in (e), respectively. (h), MUSI image of a fresh mouse kidney; the inset at the bottom left shows the photograph of the specimen. (i) and (j) Zoomed-in MUSI and the corresponding H&E-stained images of the orange and green solid regions marked in (h), respectively. Scale bars: 1 mm [(a), (e), and (h)] and 100 μm (the remaining). GL: granule layer; ML: molecular layer; and WM: white matter.

FIG. 2.

Ex vivo MUSI imaging of fresh mouse tissues. (a) MUSI image of a fresh mouse liver; the inset at the bottom left shows the photograph of the specimen. (b) and (c) Zoomed-in MUSI and the corresponding H&E-stained images of the orange and green solid regions marked in (a), respectively. (d) Distributions of nuclear features extracted from (c). Wilcoxon rank-sum testing is carried out across groups with n = 50 for each distribution. The significance is defined as p* ≤ 0.05 in all cases. (e) MUSI image of a fresh mouse brain; the inset at the bottom left shows the photograph of the specimen. (f) and (g) Zoomed-in MUSI and the corresponding H&E-stained images of the orange and green solid regions marked in (e), respectively. (h), MUSI image of a fresh mouse kidney; the inset at the bottom left shows the photograph of the specimen. (i) and (j) Zoomed-in MUSI and the corresponding H&E-stained images of the orange and green solid regions marked in (h), respectively. Scale bars: 1 mm [(a), (e), and (h)] and 100 μm (the remaining). GL: granule layer; ML: molecular layer; and WM: white matter.

Close modal

To further explore the potential of MUSI for rapid screening of cancerous tissues, freshly excised mouse tissues with metastatic human breast cancer, including mouse skin [Figs. 3(a)3(c)], lung [Figs. 3(d)3(g)], and spleen [Figs. 3(h)3(j)], are imaged by MUSI for validation. The MUSI image of a primary tumor [Figs. 3(a)3(c)], which is developed through subcutaneous injection and obtained in the skin, is in good accordance with the H&E-stained images in which the densely packed cancer cells are spread across the whole slide. The sarcomatoid feature shown in Fig. 3(b) indicates the invasiveness of the cancer cell line. In addition, it is observed that the mouse lung is invaded by a large number of poorly differentiated breast tumor cells [Fig. 3(e)], and the alveolar spaces are spatially compressed by the infiltrating lymphocytes, which will obstruct the normal gaseous exchange in the lung [Fig. 3(f)]. Figure 3(g) shows a mixture of metastatic tumor cells trying to invade and spread to other parts of the body through the periphery of blood vessels, and immune cells that are involved in responding to invasion are also observed. In addition, extensive fibrosis can be found in the mouse spleen with splenomegaly (i.e., enlargement of the spleen) [Fig. 3(h)], which may be related to neoplasm or increased immunologic activity. Disruption of the functional parenchyma, namely red pulp and white pulp, is shown in Fig. 3(i), where the morphologically distinctive compartments are indistinguishable and without the presence of follicles and germinal centers.

FIG. 3.

Ex vivo MUSI imaging of cancerous mouse tissues. (a) MUSI and clinical standard images of a primary tumor obtained in mouse skin; the inset at the bottom left shows the photograph of the specimen. (b) and (c) Zoomed-in MUSI and the corresponding H&E-stained images of the orange and green solid regions marked in (a), respectively. (d) MUSI and clinical standard images of a fresh mouse lung with cancer metastasis; the inset at the bottom left shows the photograph of the specimen. (e)–(g) Zoomed-in MUSI and the corresponding H&E-stained images of the orange, green, and yellow solid regions marked in (d), respectively. (h) MUSI and clinical standard images of a fresh mouse spleen with splenomegaly; the inset at the bottom left shows the photograph of the specimen. (i) Zoomed-in MUSI and the corresponding H&E-stained images of the orange solid region marked in (h). Scale bars: 1 mm [(a), (d), and (h)], 200 μm [(b), (c), (e)–(g), and (i)], and 100 μm (the remaining).

FIG. 3.

Ex vivo MUSI imaging of cancerous mouse tissues. (a) MUSI and clinical standard images of a primary tumor obtained in mouse skin; the inset at the bottom left shows the photograph of the specimen. (b) and (c) Zoomed-in MUSI and the corresponding H&E-stained images of the orange and green solid regions marked in (a), respectively. (d) MUSI and clinical standard images of a fresh mouse lung with cancer metastasis; the inset at the bottom left shows the photograph of the specimen. (e)–(g) Zoomed-in MUSI and the corresponding H&E-stained images of the orange, green, and yellow solid regions marked in (d), respectively. (h) MUSI and clinical standard images of a fresh mouse spleen with splenomegaly; the inset at the bottom left shows the photograph of the specimen. (i) Zoomed-in MUSI and the corresponding H&E-stained images of the orange solid region marked in (h). Scale bars: 1 mm [(a), (d), and (h)], 200 μm [(b), (c), (e)–(g), and (i)], and 100 μm (the remaining).

Close modal

Adenocarcinoma is the most common type of non-small cell lung cancer, which typically evolves from mucosal glands and occurs in the lung periphery. Lung adenocarcinomas are usually developed with a mixture of histologic subtypes, including lepidic, acinar, papillary, micropapillary, and solid. The lung cancer specimens (n = 15) were obtained from patients who were treated with lung lobectomy with informed consent. The tissues are manually sectioned and then imaged by MUSI, and subsequently histologically processed to generate the clinical standard images for comparison. Multiple H&E-stained slices are prepared for each specimen tissue, and the slice that is the most representative of the MUSI image is presented. A partial collection of the imaged specimen is shown in Fig. S5. Figure 4 demonstrates several representative cases of different adenocarcinoma subtypes. Figure 4(a) showcases a specimen with acinar-predominant adenocarcinoma, where the irregular-shaped glands in a fibrotic stroma can be visualized by MUSI [Figs. 4(b) and 4(c)]. It can be observed that some glands are arranged as solid clusters of tumor cells with a less recognizable lumen [Fig. 4(c)]. Tissue fragments generated through tumor cell breakup can be easily distinguished by MUSI with an increased fluorescence intensity [Fig. 4(d)]. In comparison, the growth patterns of papillary-type are different from that in acinar-type. Figure 4(e) shows a papillary-predominant lung adenocarcinoma specimen with a positive margin that outlines a clear interface between the normal and cancer regions [denoted by the dotted lines in Fig. 4(e)]. The nuclei of tumor cells are clearly identified at a depth of 10 μm below the tissue surface [Fig. 4(f)]. The finger-like papillary architecture with tumor cells lining the surface of branching fibrovascular cores is revealed in both MUSI and H&E-stained images [Fig. 4(g)]. In comparison, pulmonary alveolus with large air spaces is well characterized in the normal lung tissue [Fig. 4(h)]. Figure 4(i) shows a case of micropapillary-predominant adenocarcinoma. Figure 4(k) showcases a region with a large number of tumor-infiltrating lymphocytes and anthracosis pigments. The micropapillary clusters are found floating and dissociative within alveolar spaces with a lack of fibrovascular cores [Fig. 4(l)].

FIG. 4.

Ex vivo imaging of human lung adenocarcinoma specimens. (a) MUSI and clinical standard images of a lung specimen with acinar adenocarcinoma; the inset at the bottom left shows the photograph of the specimen. (b)–(d) Zoomed-in MUSI and the corresponding H&E-stained images of the orange, green, and yellow solid regions marked in (a), respectively. (e) MUSI and clinical standard images of a lung specimen with papillary adenocarcinoma; the inset at the bottom left shows the photograph of the specimen. (f)–(h) Zoomed-in MUSI and the corresponding H&E-stained images of the orange, green, and yellow solid regions marked in (e), respectively. (i) MUSI and clinical standard images of a lung specimen with micropapillary adenocarcinoma; the inset at the bottom left shows the photograph of the specimen. (j)–(l) Zoomed-in MUSI and the corresponding H&E-stained images of the orange, green, and yellow solid regions marked in (i), respectively. Scale bars: 1 mm [(a), (e), and (i)], 100 μm [(b)–(d), (f)–(h), (j)–(l)], and 50 μm (the remaining). TN: tumor cell nuclei; LN: lymphocyte nuclei; FC: fibrovascular core; AS: alveolar space; MC: micropapillary cluster; and AP: anthracosis pigment.

FIG. 4.

Ex vivo imaging of human lung adenocarcinoma specimens. (a) MUSI and clinical standard images of a lung specimen with acinar adenocarcinoma; the inset at the bottom left shows the photograph of the specimen. (b)–(d) Zoomed-in MUSI and the corresponding H&E-stained images of the orange, green, and yellow solid regions marked in (a), respectively. (e) MUSI and clinical standard images of a lung specimen with papillary adenocarcinoma; the inset at the bottom left shows the photograph of the specimen. (f)–(h) Zoomed-in MUSI and the corresponding H&E-stained images of the orange, green, and yellow solid regions marked in (e), respectively. (i) MUSI and clinical standard images of a lung specimen with micropapillary adenocarcinoma; the inset at the bottom left shows the photograph of the specimen. (j)–(l) Zoomed-in MUSI and the corresponding H&E-stained images of the orange, green, and yellow solid regions marked in (i), respectively. Scale bars: 1 mm [(a), (e), and (i)], 100 μm [(b)–(d), (f)–(h), (j)–(l)], and 50 μm (the remaining). TN: tumor cell nuclei; LN: lymphocyte nuclei; FC: fibrovascular core; AS: alveolar space; MC: micropapillary cluster; and AP: anthracosis pigment.

Close modal

With enhanced nucleic features by deep-UV excitation, our images can be virtually stained to mimic the appearance of the H&E-stained images to ensure clinical adoption. For demonstration, we employ WeCrest,32 a weakly supervised deep generative network, to generate virtually stained MUSI images for pathological evaluation [Figs. S6 and Fig. 5 (Multimedia view)]. Both virtually stained MUSI and H&E-stained images show a predominantly lepidic growth pattern with a proliferation of pneumocytes along the surface of alveolar septa [Figs. S6(e) and S6(f)]. This case also shows smaller foci of acinar architectures at the lower part of the specimen [Figs. S6(h) and S6(i)], where irregularly shaped glands are present in a desmoplastic stroma. This phenomenon is common in lung adenocarcinomas, which are usually developed with a mixture of histologic subtypes. Overall, the morphological patterns depicted in the virtually stained MUSI images are sufficient to reach a diagnosis of lepidic adenocarcinoma by a pathologist with minimal difficulty, which agrees well with the standard H&E histology. However, due to the lack of strictly paired training data in the weakly supervised network, artifacts may be generated when testing the model across specimens with different histologic subtypes. For simplicity, we only demonstrate one case here. We believe that the advances in deep neural networks could improve the model generalization to different types of lung cancers.

FIG. 5.

Virtual staining of MUSI images for pathological evaluation. Zoomed-in MUSI image, the corresponding H&E-stained image, and the virtually stained MUSI image are indicated by yellow, blue, and magenta boxes, respectively. (Multimedia available online).

FIG. 5.

Virtual staining of MUSI images for pathological evaluation. Zoomed-in MUSI image, the corresponding H&E-stained image, and the virtually stained MUSI image are indicated by yellow, blue, and magenta boxes, respectively. (Multimedia available online).

Close modal

The wide applicability of MUSI is also validated by imaging other types of tissues, such as the human skin [Figs. S7(a)–S7(d)] and human brain [Figs. S7(e)–S7(g)]. These tissues are leftover tissue, i.e., a portion of a collected specimen that is not needed for the diagnosis and treatment of the patient. It can be shown that a variety of histologic features can be simultaneously visualized by MUSI with deep-UV excitation. For instance, anatomic structures in human skin, including the sudoriferous gland [Fig. S7(b)], erythrocyte-filled arterial lumen [Fig. S7(c)], and adipose tissue [Fig. S7(d)], are well characterized by MUSI in a label-free and non-destructive manner. These features share great similarities with that in slide-based FFPE histology. In addition, cerebellar granule cells are resolved individually in the human brain tissue [Fig. S7(f)], and neuroglial cells can be visualized within a depth range of 30 μm [Fig. S7(g)].

In addition to ex vivo imaging of resected tissues, we further explore the potential of MUSI for rapid in vivo imaging of intact tissues. First, we imaged several freshly excised organs without cutting open the specimens to verify that histologic features can be extracted from the surface of intact tissues (Fig. S8). Slight pressure is applied from the top of these samples to make sure that they are in contact with the sample holder during imaging. After imaging, multiple H&E-stained thin slices are prepared for each specimen, and the slice that is the most representative of the imaged surface is shown for comparison. We found that a variety of anatomic structures, including vessels and cell nuclei in the brain cortex [Figs. S8(a) and S8(e)], renal tubules in the kidney [Figs. S8(b) and S8(f)], muscle fibers in the anterolateral thigh [Figs. S8(c) and S8(g)], and filiform papillae in dorsum tongue [Figs. S8(d) and S8(h)], are clearly revealed and are in good accordance with the conventional slide-based FFPE histology.

Figure 6 shows the MUSI’s capacity in label-free in vivo imaging. We used the mouse brain [Figs. 6(a) and 6(b)] and kidney tissues [Figs. 6(h) and 6(i)] for the idea demonstration. With a rapid acquisition speed of 250 frames/s, anatomic features can be captured without obvious breath-induced motion artifacts. The energy fluence measured at the imaging plane is ∼2 mJ/cm2, which is below the UV safety threshold of 3 mJ/cm2 regulated by the American National Standards Institute.33, Figure 6(c) shows an area in the brain that contains histologic features from both the cerebral cortex and pia mater. The cell nuclei located at the surface of the brain cortex can be resolved individually with a negative contrast [Fig. 6(d)]. The pia mater, which is the innermost layer of the meninges that clings tightly to the brain, can also be identified [Fig. 6(e)]. We can observe from both MUSI and H&E-stained images that the pia mater is richly vascularized and consists of a thin layer of squamous epithelial cell. Figure 6(f) shows a surrounding area in the dorsum ear that is filled with epidermal keratinocytes [Fig. 6(g)]. While in the mouse kidney, due to the highly fluorescent cytoplasmic lipofuscin and urinary cast material, the renal tubules at the outermost cortex can be visualized with a high signal-to-noise ratio (SNR) [Figs. 6(j) and 6(k)]. In addition, other morphological structures, such as adipose tissue in perinephric fat [Fig. 6(l)] and connective tissues in the lumbodorsal fascia [Fig. 6(m)], can also be identified. Although MUSI cannot provide 1:1 spatial mapping with conventional histology due to the difference in imaging depth and tissue orientation, the tissue morphology remains remarkably similar from the gross to the subcellular levels. The ability of MUSI to image the intact tissues non-destructively can potentially find applications that preclude invasive biopsy procedures. Although MUSI has limited access to deep tissue information compared with other label-free in vivo imaging platforms such as MediSCAPE17 and LC-OCT,34 it significantly strengthens the contrast between nucleic acid and extracellular matrix in internal organs with deep-UV excitation, thus showing a higher consistency with the histochemical staining in routine pathological practice.

FIG. 6.

In vivo imaging of intact mouse tissues. (a) and (b) Photographs of the imaged brain with parietal craniotomy. (c) Zoomed-in MUSI image of the orange solid region marked in (b). Pial vessels are depicted by red dotted lines. (d) and (e) Zoomed-in MUSI and the corresponding H&E-stained images of the blue solid and dashed regions marked in (c), respectively. (f) Zoomed-in MUSI image of the green solid region marked in (b). (g) Zoomed-in MUSI and the corresponding H&E-stained images of the blue dashed region marked in (f). (h) and (i) Photographs of the imaged kidney with lumbar nephrectomy. (j) Zoomed-in MUSI image of the orange solid region marked in (i). (k) Zoomed-in MUSI and the corresponding H&E-stained images of the blue dashed regions marked in (j). (l) and (m) Zoomed-in MUSI and the corresponding H&E-stained images of the green and yellow solid regions marked in (i), respectively. Scale bars: 500 μm [(c), (f), and (j)], 200 μm [(l) and (m)], and 100 μm (the remaining).

FIG. 6.

In vivo imaging of intact mouse tissues. (a) and (b) Photographs of the imaged brain with parietal craniotomy. (c) Zoomed-in MUSI image of the orange solid region marked in (b). Pial vessels are depicted by red dotted lines. (d) and (e) Zoomed-in MUSI and the corresponding H&E-stained images of the blue solid and dashed regions marked in (c), respectively. (f) Zoomed-in MUSI image of the green solid region marked in (b). (g) Zoomed-in MUSI and the corresponding H&E-stained images of the blue dashed region marked in (f). (h) and (i) Photographs of the imaged kidney with lumbar nephrectomy. (j) Zoomed-in MUSI image of the orange solid region marked in (i). (k) Zoomed-in MUSI and the corresponding H&E-stained images of the blue dashed regions marked in (j). (l) and (m) Zoomed-in MUSI and the corresponding H&E-stained images of the green and yellow solid regions marked in (i), respectively. Scale bars: 500 μm [(c), (f), and (j)], 200 μm [(l) and (m)], and 100 μm (the remaining).

Close modal

MUSI enables rapid ex vivo or in vivo assessment of fresh and unprocessed tissues in a label-free and non-destructive manner, holding great promise to revolutionize the current clinical practice in tissue histopathology. However, there are still some deviations between the MUSI and conventional FFPE histology. First, anatomic structures, such as sinusoidal capillaries in mouse liver and Bowman’s space in mouse kidney, are well identified by FFPE histology but hardly visualized by MUSI [Figs. 2(b) and 2(j)]. This is probably because the grossly cut soft tissues would be deformed when being flattened and placed on the sample holder. Second, fibrous structures, such as fibrotic stroma in human lung [Figs. S6(d) and S6(g)] and elastic membrane in the artery [Fig. S7(c)], are better visualized in MUSI than that in H&E-stained images. This is because these structural proteins present a high quantum yield with deep-UV excitation, while eosin exhibits a similar affinity across the cytoplasm.

Intrinsic fluorescence with deep-UV excitation naturally forms a contrast mechanism for label-free tissue imaging. However, the fluorescence properties of endogenous fluorophores, such as emission maximum and quantum yields, are highly related to tissue phenotypes and disease status, causing obvious intensity variations in the MUSI images. For instance, the spleen, the largest mass of lymphoid organs, usually presents the least fluorescence among other types of tissues. In cancerous human lung, solid adenocarcinoma with a lack of fibrotic stroma shows a relatively weak intensity compared with other histologic subtypes. In addition, we observed that the overall intensity will be increased with continuous UV radiation. The combination of MUSI imaging and autofluorescence spectroscopy holds great promise to further increase the accuracy in the diagnosis of lung cancer, as some metabolic enzymes (e.g., NADH and flavin adenine dinucleotide) are highly associated with cellular changes and can effectively serve as a tumor-specific biomarker.35 

MUSI is currently in an early stage of development. The system can be further optimized to achieve a higher imaging performance for wide clinical applications. The current system can achieve an imaging speed of 0.5 mm2/s, which is predominantly restricted by the weak intrinsic fluorescence of tissues. For each raw image in the sequence of primary scanning, an integration time of 4 ms is required to maintain an acceptable SNR under an illumination power of 1 mW such that the camera is only allowed to operate at a frame rate of 250 fps. The imaging speed is expected to increase by another order of magnitude with a high-power UV illumination source that could reduce the exposure time to a few hundred microseconds (camera-limited and not photon-limited speed in this case). However, it is not applicable for in vivo applications since the strong UV radiation may cause damage to internal organs. Currently, the processing time is linearly scaled with tissue surface geometry, which is a common issue in thick tissue imaging. For instance, with 2D imaging techniques, it is also necessary to process a series of axially refocused images to generate a single in-focus image of the intact tissue surface.11,36 In this study, we use a model-based deconvolution algorithm30 for extending depth-of-field, which is time-consuming and requires a heavy computational load. For further acceleration in data processing, the recently proposed deep learning-based image reconstruction strategies hold great promise.37,38 With MUSI’s dual-axis configuration, imaging parameters, such as DOF and lateral/axial resolution, can be optimized separately. In the illumination arm, the trade-off between the length of a Gaussian light sheet (i.e., DOF and 2ZR) and its thickness (i.e., 2w0) is inevitable due to the diffraction of light. In practice, the DOF could be further extended without the compromise of axial resolution by using the light sheet generated through rapid scanning of non-diffracting beams.39,40 This can further improve the system tolerance for large irregular surfaces. In clinical practice, pathologists first pan through the slide at a relatively low resolution (2.5× to 5×) to rapidly identify suspicious structures and then zoom in to a higher resolution (20× to 40×) for selected regions to interrogate diagnostic details such as cancer-associated nuclear atypia.41,42,43 This efficient multi-scale imaging workflow can be achieved by using a switchable set of objective lenses with different resolution scales in the detection arm.28 However, nuclear structures, such as chromatin and nucleoli, are less recognizable in the MUSI images even with a sub-micron resolution (Fig. S8), which is likely because the autofluorescence property of cell nuclei is not chemically identical to the histological stains. In addition, the development from a benchtop to a probe-based handheld MUSI device can greatly increase the flexibility of operation, promoting the applications of intraprocedural biopsy guidance.

In summary, MUSI enables label-free and non-destructive imaging of fresh and unprocessed tissues, allowing surgical specimens with large irregular surfaces to be assessed at the subcellular level within a point-of-care timeframe. We experimentally demonstrated that MUSI can provide diagnostically important features to differentiate between different subtypes of human lung adenocarcinomas, holding great promise to streamline the current workflow in surgical pathology. As a proof-of-concept study, this work is limited by a small number of specimens (n = 15). Large-scale clinical trials should be carried out as follow-up work to quantify the diagnostic metrics (i.e., sensitivity and specificity) and study the inter-patient variations. In addition, computer-aided diagnosis can also be incorporated with MUSI to further improve the standard of care in cancer diagnosis.

See the supplementary material for the comparisons of state-of-the-art slide-free imaging modalities; experimental profile of the illumination light sheet; experimental characterization of the point-spread functions of the MUSI system; ex vivo MUSI imaging of formalin-fixed and unprocessed mouse tissues; collection of human lung cancer specimens involved in this study; virtual staining of MUSI images for pathological evaluation; ex vivo MUSI imaging of human skin and brain tissues; ex vivo MUSI imaging of intact mouse organs; and multi-scale imaging of hippocampal structure in a formalin-fixed thick mouse brain (PDF).

The Translational and Advanced Bioimaging Laboratory (TAB-Lab) at HKUST acknowledges the support of the Hong Kong Innovation and Technology Commission (Grant No. MRP/012/20), the Research Grants Council of the Hong Kong Special Administrative Region (Grant No. 16208620), and the Hong Kong University of Science and Technology startup (Grant No. R9421).

Y.Z., B.H., and T.T.W.W. have applied for a patent (US Provisional Patent Application No. 63/417, 682) related to the work reported in this manuscript.

Y.Z. and T.T.W.W. conceived of the study. Y.Z., B.H., and L.K. built the imaging system. Y.Z., B.H., and V.T.C.T. prepared the specimens involved in this study. Y.Z., B.H., and J.W. performed imaging experiments. C.T.K.L. performed histological staining. Y.Z. processed the data. Y.Z. and T.T.W.W. wrote the manuscript. T.T.W.W. supervised the whole study.

Yan Zhang: Investigation (lead); Methodology (lead); Validation (lead); Writing – original draft (lead). Bingxin Huang: Project administration (supporting); Resources (supporting). Weixing Dai: Data curation (supporting); Methodology (supporting). Lei Kang: Resources (supporting); Software (supporting). Victor T. C. Tsang: Project administration (supporting); Resources (supporting). Jiajie Wu: Project administration (supporting); Resources (supporting). Claudia T. K. Lo: Project administration (supporting); Resources (supporting). Terence T. W. Wong: Funding acquisition (lead); Supervision (lead).

All data involved in this work, including raw/processed images provided in this paper, are available from the corresponding author upon request.

1.
S. J.
Maygarden
,
F. C.
Detterbeck
, and
W. K.
Funkhouser
, “
Bronchial margins in lung cancer resection specimens: Utility of frozen section and gross evaluation
,”
Mod. Pathol.
17
(
9
),
1080
1086
(
2004
).
2.
A.
Preeti
,
G.
Sameer
,
S.
Kulranjan
,
S. A.
Abhinav
,
R.
Preeti
,
Y.
Sunita
, and
G. M.
Mati
, “
Intra-operative frozen sections: Experience at a tertiary care centre
,”
Asian Pac. J. Cancer Prev.
17
(
12
),
5057
(
2016
).
3.
Y.
Zhang
,
B.
Huang
,
J.
Wu
, and
T. T. W.
Wong
, “
Advances in optical microscopy revolutionize the practice of surgical pathology with rapid and non-destructive tissue assessment
,”
Eur. Phys. J. Spec. Top.
231
,
763
779
(
2022
).
4.
J.
Dobbs
,
S.
Krishnamurthy
,
M.
Kyrish
,
A. P.
Benveniste
,
W.
Yang
, and
R.
Richards-Kortum
, “
Confocal fluorescence microscopy for rapid evaluation of invasive tumor cellularity of inflammatory breast carcinoma core needle biopsies
,”
Breast Cancer Res. Treat.
149
(
1
),
303
310
(
2014
).
5.
S.
Krishnamurthy
,
K.
Ban
,
K.
Shaw
,
G.
Mills
,
R.
Sheth
,
A.
Tam
,
S.
Gupta
, and
S.
Sabir
, “
Confocal fluorescence microscopy platform suitable for rapid evaluation of small fragments of tissue in surgical pathology practice
,”
Arch. Pathol. Lab. Med.
143
(
3
),
305
313
(
2019
).
6.
M.
Ragazzi
,
S.
Piana
,
C.
Longo
,
F.
Castagnetti
,
M.
Foroni
,
G.
Ferrari
,
G.
Gardini
, and
G.
Pellacani
, “
Fluorescence confocal microscopy for pathologists
,”
Mod. Pathol.
27
(
3
),
460
471
(
2014
).
7.
A. K.
Glaser
,
N. P.
Reder
,
Y.
Chen
,
E. F.
McCarty
,
C.
Yin
,
L.
Wei
,
Y.
Wang
,
L. D.
True
, and
J. T. C.
Liu
, “
Light-sheet microscopy for slide-free non-destructive pathology of large clinical specimens
,”
Nat. Biomed. Eng.
1
(
7
),
0084
(
2017
).
8.
Y.
Chen
,
W.
Xie
,
A. K.
Glaser
,
N. P.
Reder
,
C.
Mao
,
S. M.
Dintzis
,
J. C.
Vaughan
, and
J. T. C.
Liu
, “
Rapid pathology of lumpectomy margins with open-top light-sheet (OTLS) microscopy
,”
Biomed. Opt. Express
10
(
3
),
1257
(
2019
).
9.
M.
Wang
,
H. Z.
Kimbrell
,
A. B.
Sholl
,
D. B.
Tulman
,
K. N.
Elfer
,
T. C.
Schlichenmeyer
,
B. R.
Lee
,
M.
Lacey
, and
J. Q.
Brown
, “
High-resolution rapid diagnostic imaging of whole prostate biopsies using video-rate fluorescence structured illumination microscopy
,”
Cancer Res.
75
(
19
),
4032
4041
(
2015
).
10.
M.
Wang
,
D. B.
Tulman
,
A. B.
Sholl
,
H. Z.
Kimbrell
,
S. H.
Mandava
,
K. N.
Elfer
,
S.
Luethy
,
M. M.
Maddox
,
W.
Lai
,
B. R.
Lee
, and
J. Q.
Brown
, “
Gigapixel surface imaging of radical prostatectomy specimens for comprehensive detection of cancer-positive surgical margins using structured illumination microscopy
,”
Sci. Rep.
6
(
1
),
27419
(
2016
).
11.
F.
Fereidouni
,
Z. T.
Harmany
,
M.
Tian
,
A.
Todd
,
J. A.
Kintner
,
J. D.
McPherson
,
A. D.
Borowsky
,
J.
Bishop
,
M.
Lechpammer
,
S. G.
Demos
, and
R.
Levenson
, “
Microscopy with ultraviolet surface excitation for rapid slide-free histology
,”
Nat. Biomed. Eng.
1
(
12
),
957
966
(
2017
).
12.
T.
Yoshitake
,
M. G.
Giacomelli
,
L. M.
Quintana
,
H.
Vardeh
,
L. C.
Cahill
,
B. E.
Faulkner-Jones
,
J. L.
Connolly
,
D.
Do
, and
J. G.
Fujimoto
, “
Rapid histopathological imaging of skin and breast cancer surgical specimens using immersion microscopy with ultraviolet surface excitation
,”
Sci. Rep.
8
(
1
),
4476
(
2018
).
13.
O.
Assayag
,
M.
Antoine
,
B.
Sigal-Zafrani
,
M.
Riben
,
F.
Harms
,
A.
Burcheri
,
K.
Grieve
,
E.
Dalimier
,
B. L. C.
de Poly
, and
C.
Boccara
, “
Large field, high resolution full-field optical coherence tomography: A pre-clinical study of human breast tissue and cancer assessment
,”
Technol. Cancer Res. Treat.
13
(
5
),
455
(
2014
).
14.
O.
Assayag
,
K.
Grieve
,
B.
Devaux
,
F.
Harms
,
J.
Pallud
,
F.
Chretien
,
C.
Boccara
, and
P.
Varlet
, “
Imaging of non-tumorous and tumorous human brain tissues with full-field optical coherence tomography
,”
NeuroImage: Clin.
2
(
1
),
549
557
(
2013
).
15.
V. Q.
Chung
,
P. J.
Dwyer
,
K. S.
Nehal
,
M.
Rajadhyaksha
,
G. M.
Menaker
,
C.
Charles
, and
S. B.
Jiang
, “
Use of ex vivo confocal scanning laser microscopy during Mohs surgery for nonmelanoma skin cancers
,”
Dermatol. Surg.
30
(
12p1
),
1470
1478
(
2004
).
16.
M.
Rajadhyaksha
,
A.
Marghoob
,
A.
Rossi
,
A. C.
Halpern
, and
K. S.
Nehal
, “
Reflectance confocal microscopy of skin in vivo: From bench to bedside
,”
Lasers Surg. Med.
49
(
1
),
7
19
(
2017
).
17.
K. B.
Patel
,
W.
Liang
,
M. J.
Casper
,
V.
Voleti
,
W.
Li
,
A. J.
Yagielski
,
H. T.
Zhao
,
C.
Perez Campos
,
G. S.
Lee
,
J. M.
Liu
,
E.
Philipone
,
A. J.
Yoon
,
K. P.
Olive
,
S. M.
Coley
, and
E. M. C.
Hillman
, “
High-speed light-sheet microscopy for the in-situ acquisition of volumetric histological images of living tissue
,”
Nat. Biomed. Eng.
6
(
5
),
569
583
(
2022
).
18.
S.
Kretschmer
,
M.
Pieper
,
G.
Hüttmann
,
T.
Bölke
,
B.
Wollenberg
,
L. M.
Marsh
,
H.
Garn
, and
P.
König
, “
Autofluorescence multiphoton microscopy for visualization of tissue morphology and cellular dynamics in murine and human airways
,”
Lab. Invest.
96
(
8
),
918
931
(
2016
).
19.
S.
Ye
,
J.
Zou
,
C.
Huang
,
F.
Xiang
,
Z.
Wen
,
N.
Wang
,
J.
Yu
,
Y.
He
,
P.
Liu
,
X.
Mei
,
H.
Li
,
L.
Niu
,
P.
Gong
, and
W.
Zheng
, “
Rapid and label-free histological imaging of unprocessed surgical tissues via dark-field reflectance ultraviolet microscopy
,”
iScience
26
(
1
),
105849
(
2023
).
20.
T. T. W.
Wong
,
R.
Zhang
,
P.
Hai
,
C.
Zhang
,
M. A.
Pleitez
,
R. L.
Aft
,
D. V.
Novack
, and
L. V.
Wang
, “
Fast label-free multilayered histology-like imaging of human breast cancer by photoacoustic microscopy
,”
Sci. Adv.
3
(
5
),
e1602168
(
2017
).
21.
J. W.
Baik
,
H.
Kim
,
M.
Son
,
J.
Choi
,
K. G.
Kim
,
J. H.
Baek
,
Y. H.
Park
,
J.
An
,
H. Y.
Choi
,
S. Y.
Ryu
,
J. Y.
Kim
,
K.
Byun
, and
C.
Kim
, “
Intraoperative label-free photoacoustic histopathology of clinical specimens
,”
Laser Photonics Rev.
15
,
2100124
(
2021
).
22.
M. T.
Martell
,
N. J. M.
Haven
,
B. D.
Cikaluk
,
B. S.
Restall
,
E. A.
McAlister
,
R.
Mittal
,
B. A.
Adam
,
N.
Giannakopoulos
,
L.
Peiris
,
S.
Silverman
,
J.
Deschenes
,
X.
Li
, and
R. J.
Zemp
, “
Deep learning-enabled realistic virtual histology with ultraviolet photoacoustic remote sensing microscopy
,”
Nat. Commun.
14
(
1
),
5967
(
2023
).
23.
Y. K.
Tao
,
D.
Shen
,
Y.
Sheikine
,
O. O.
Ahsen
,
H. H.
Wang
,
D. B.
Schmolze
,
N. B.
Johnson
,
J. S.
Brooker
,
A. E.
Cable
,
J. L.
Connolly
, and
J. G.
Fujimoto
, “
Assessment of breast pathologies using nonlinear microscopy
,”
Proc. Natl. Acad. Sci. U. S. A.
111
(
43
),
15304
15309
(
2014
).
24.
H.
Tu
,
Y.
Liu
,
D.
Turchinovich
,
M.
Marjanovic
,
J. K.
Lyngsø
,
J.
Lægsgaard
,
E. J.
Chaney
,
Y.
Zhao
,
S.
You
,
W. L.
Wilson
,
B.
Xu
,
M.
Dantus
, and
S. A.
Boppart
, “
Stain-free histopathology by programmable supercontinuum pulses
,”
Nat. Photonics
10
(
8
),
534
540
(
2016
).
25.
D. A.
Orringer
,
B.
Pandian
,
Y. S.
Niknafs
,
T. C.
Hollon
,
J.
Boyle
,
S.
Lewis
,
M.
Garrard
,
S. L.
Hervey-Jumper
,
H. J. L.
Garton
,
C. O.
Maher
,
J. A.
Heth
,
O.
Sagher
,
D. A.
Wilkinson
,
M.
Snuderl
,
S.
Venneti
,
S. H.
Ramkissoon
,
K. A.
McFadden
,
A.
Fisher-Hubbard
,
A. P.
Lieberman
,
T. D.
Johnson
,
X. S.
Xie
,
J. K.
Trautman
,
C. W.
Freudiger
, and
S.
Camelo-Piragua
, “
Rapid intraoperative histology of unprocessed surgical specimens via fibre-laser-based stimulated Raman scattering microscopy
,”
Nat. Biomed. Eng.
1
(
2
),
0027
(
2017
).
26.
X.
Chen
,
O.
Nadiarynkh
,
S.
Plotnikov
, and
P. J.
Campagnola
, “
Second harmonic generation microscopy for quantitative analysis of collagen fibrillar structure
,”
Nat. Protoc.
7
(
4
),
654
669
(
2012
).
27.
Y.
Zhang
,
L.
Kang
,
I. H. M.
Wong
,
W.
Dai
,
X.
Li
,
R. C. K.
Chan
,
M. K. Y.
Hsin
, and
T. T. W.
Wong
, “
High-throughput, label-free and slide-free histological imaging by computational microscopy and unsupervised learning
,”
Adv. Sci.
9
,
2102358
(
2021
).
28.
L. A.
Barner
,
A. K.
Glaser
,
H.
Huang
,
L. D.
True
, and
J. T. C.
Liu
, “
Multi-resolution open-top light-sheet microscopy to enable efficient 3D pathology workflows
,”
Biomed. Opt. Express
11
(
11
),
6605
(
2020
).
29.
R.
McGorty
,
H.
Liu
,
D.
Kamiyama
,
Z.
Dong
,
S.
Guo
, and
B.
Huang
, “
Open-top selective plane illumination microscope for conventionally mounted specimens
,”
Opt. Express
23
(
12
),
16142
(
2015
).
30.
F.
Aguet
,
D.
Van De Ville
, and
M.
Unser
, “
Model-based 2.5-D deconvolution for extended depth of field in brightfield microscopy
,”
IEEE Trans. Image Process.
17
(
7
),
1144
1153
(
2008
).
31.
S.
Preibisch
,
S.
Saalfeld
, and
P.
Tomancak
, “
Globally optimal stitching of tiled 3D microscopic image acquisitions
,”
Bioinformatics
25
(
11
),
1463
1465
(
2009
).
32.
W.
Dai
,
I. H. M.
Wong
, and
T. T. W.
Wong
, “
A weakly supervised deep generative model for complex image restoration and style transformation
,” https://doi.org/10.36227/techrxiv.19747057.v1 (
2022
).
33.
American National Standard for Safe Use of Lasers-ANSI Z136.1-2022.
34.
C.
Ruini
,
S.
Schuh
,
C.
Gust
,
B.
Kendziora
,
L.
Frommherz
,
L. E.
French
,
D.
Hartmann
,
J.
Welzel
, and
E.
Sattler
, “
Line-field optical coherence tomography: In vivo diagnosis of basal cell carcinoma subtypes compared with histopathology
,”
Clin. Exp. Dermatol.
46
(
8
),
1471
1481
(
2021
).
35.
M.
Monici
, “
Cell and tissue autofluorescence research and diagnostic applications
,”
Biotechnol. Annu. Rev.
28
,
227
256
(
2005
).
36.
Y.
Zhang
,
L.
Kang
,
C. T. K.
Lo
,
V. T. C.
Tsang
, and
T. T. W.
Wong
, “
Rapid slide-free and non-destructive histological imaging using wide-field optical-sectioning microscopy
,”
Biomed. Opt. Express
13
(
5
),
2782
2796
(
2022
).
37.
V.
Sitzmann
,
S.
Diamond
,
Y.
Peng
,
X.
Dun
,
S.
Boyd
,
W.
Heidrich
,
F.
Heide
, and
G.
Wetzstein
, “
End-to-end optimization of optics and image processing for achromatic extended depth of field and super-resolution imaging
,”
ACM Trans. Graphics
37
(
4
),
1
(
2018
).
38.
L.
Jin
,
Y.
Tang
,
Y.
Wu
,
J. B.
Coole
,
M. T.
Tan
,
X.
Zhao
,
H.
Badaoui
,
J. T.
Robinson
,
M. D.
Williams
,
A. M.
Gillenwater
,
R. R.
Richards-Kortum
, and
A.
Veeraraghavan
, “
Deep learning extended depth-of-field microscope for fast and slide-free histology
,”
Proc. Natl. Acad. Sci. U. S. A.
117
(
52
),
33051
33060
(
2020
).
39.
L.
Gao
,
L.
Shao
,
B. C.
Chen
, and
E.
Betzig
, “
3D live fluorescence imaging of cellular dynamics using Bessel beam plane illumination microscopy
,”
Nat. Protoc.
9
(
5
),
1083
1101
(
2014
).
40.
T. A.
Planchon
,
L.
Gao
,
D. E.
Milkie
,
M. W.
Davidson
,
J. A.
Galbraith
,
C. G.
Galbraith
, and
E.
Betzig
, “
Rapid three-dimensional isotropic imaging of living cells using Bessel beam plane illumination
,”
Nat. Methods
8
(
5
),
417
423
(
2011
).
41.
M. N.
Gurcan
,
L. E.
Boucheron
,
A.
Can
,
A.
Madabhushi
,
N. M.
Rajpoot
, and
B.
Yener
, “
Histopathological image analysis: A review
,”
IEEE Rev. Biomed. Eng.
2
,
147
171
(
2009
).
42.
T. T.
Brunyé
,
E.
Mercan
,
D. L.
Weaver
, and
J. G.
Elmore
, “
Accuracy is in the eyes of the pathologist: The visual interpretive process and diagnostic accuracy with digital whole slide images
,”
J. Biomed. Inf.
66
,
171
179
(
2017
).
43.
L.
Privitera
,
I.
Paraboschi
,
D.
Dixit
,
O. J.
Arthurs
, and
S.
Giuliani
, “
Image-guided surgery and novel intraoperative devices for enhanced visualisation in general and paediatric surgery: A review
,”
Innovative Surg. Sci.
6
(
4
),
161
172
(
2022
).

Supplementary Material