Rapid pseudo-H&E imaging using a fluorescence-inbuilt optical coherence microscopic imaging system

: A technique using Linnik-based optical coherence microscopy (OCM), with built-in fluorescence microscopy (FM), is demonstrated here to describe cellular-level morphology for fresh porcine and biobank tissue specimens. The proposed method utilizes color-coding to generate digital pseudo-H&E (p-H&E) images. Using the same camera, colocalized FM images are merged with corresponding morphological OCM images using a 24-bit RGB composition process to generate position-matched p-H&E images. From receipt of dissected fresh tissue piece to generation of stitched images, the total processing time is < 15 min for a 1-cm 2 specimen, which is on average two times faster than frozen-section H&E process for fatty or water-rich fresh tissue specimens. This technique was successfully used to scan human and animal fresh tissue pieces, demonstrating its applicability for both biobank and veterinary purposes. We provide an in-depth comparison between p-H&E and human frozen-section H&E images acquired from the same metastatic sentinel lymph node slice ( ∼ 10 µm thick), and show the differences, like elastic fibers of a tiny blood vessel and cytoplasm of tumor cells. This optical sectioning technique provides histopathologists with a convenient assessment method that outputs large-field H&E-like images of fresh tissue pieces without requiring any physical embedment. with their frozen-sectioned H&E-stained counterparts for the same original tissue pieces. In the two p-H&E images of thyroid tissue specimens, normal follicles and aggregate tumor cells could be easily identified.


Introduction
Optical coherence tomography (OCT) evolved from low-coherence white-light interferometry [1]. In clinical practice, optical-path-length concept of OCT was first used for in vivo ocular measurements [2]. OCT technology [3], such as frequency-domain OCT [4], has been used to acquire in vivo tomographic retinal images. Owing to its optical sectioning property, OCT is applied in oncology [5] for rapid intraoperative diagnosis. Although spectral-domain OCT [6] can be used for rapid diagnosis, it lacks the required lateral resolution and nuclear information for histological diagnosis. Full-field OCT (FF-OCT) [7], a type of time-domain OCT, reconstructs axially stacked en-face optical coherence microscopy (OCM) [8] images to generate a series of continuous adjacent tomographic images acquired commonly using a broadband tungsten halogen lamp [9], high numerical aperture objective lens, and CCD/CMOS camera; FF-OCT can capture the cellular-level morphological images of biological tissues [10,11] at submicron spatial resolution. FF-OCT has been adopted in pathological studies to investigate the feasibility [31]. Microscopy with ultraviolet surface excitation (MUSE), which is generally used for the surface scan of formalin-fixed tissue, was applied on fresh tissue pieces by using ATTO 655 NHS ester (cytoplasmic stain) and SYBR Gold (nucleus stain) [32], with a scan speed of 0.2 cm 2 /min; however, MUSE using 285-nm excitation UV light still showed a tissue penetration of less than 25 µm [33]. An open-top light-sheet (OTLS) [34] microscope showed a high scan speed of 1.5 cm 2 /min when the same fluorescent dyes, namely ATTO 655 NHS ester and SYBR Gold, were used. The image processing time (15 min for 1-cm 2 image data) was fast, but image quality was not sufficient for high-resolution assessment. Each of the aforementioned imaging modalities has its own advantages and limitations [35].
In this study, we proposed a method that performs optical p-H&E sectioning through fluorescence-inbuilt OCM (f-OCM) [36]; this method was applied to a fresh porcine cerebrum tissue piece. The proposed approach can generate corresponding p-H&E images after a step-by-step en-face scan in real time [37] for further stitching a large-field p-H&E image. The quality of the image of the fresh porcine kidney piece acquired using f-OCM was higher than that acquired using optical sectioning SIM [31]; our p-H&E image provides various morphological details of the cellular structure. Moreover, fresh porcine and human (provided by a biobank) tissue pieces were used to evaluate the image quality of this system. Finally, a comprehensive comparison of p-H&E and traditional H&E images obtained from the scans of the same 10µm-thick frozen-sectioned metastatic sentinel lymph node (SLN) slice was performed, and the images showed high correlation and similarity. f-OCM with a scan speed of 0.25 cm 2 /min can be used to assess fresh, small-sized pieces, such as a brain tumor or breast core-needle biopsy tissue, for surgical evaluation.

Experimental setup and imaging methods
Figure 1(a) shows the schematic of the f-OCM setup, which includes OCM and FM. The p-H&E images combine cytoplasmic morphology and nuclei-related details obtained from OCM and FM, respectively. Figure 1(b) shows the movable device hardware (AcuOnPath, AcuSolutions Inc.) including the scan machine and control box, which were designed for research use and preclinical testing. In this system, a broadband light source serves as a low-coherence light source, and the measured spectrum and measured interferometric signal of which are shown in Fig. 1(c) and 1(d), respectively. In contrast to traditional FF-OCT, which operates at near-infrared wavelengths, the OCM part of the f-OCM system utilizes a visible-range broadband light source to obtain high lateral and axial resolutions for facilitating the observation of fine cellular features. OCM detects only ultra-weak signals from nucleus reflection. To address this issue, an FM excitation path was incorporated into the sample arm of the OCM structure to enable f-OCM. In contrast to traditional dual-modality FM and FF-OCM [23] that merge two spectral light sources before a beam splitter, our method implementing FM can reflect and reject most of the excitation light using the same dichroic mirror, resulting in decreased loss of intensity from nuclei-emitted signals because signals do not have to pass through the beamsplitter twice. Table 1 shows theoretical and measured resolutions along axial and lateral directions for OCM, FM, and pixel. Here, the pixel resolution along the lateral dimension is determined as the equivalent lateral size of the focal plane of a sample projected from the single pixel of CMOS camera. It is the capability of spatial division restricted by CMOS camera pixel size. Data of Table 1 will be brought out in Secs. 2.1 and 2.2.

OCM
A collimated beam from the broadband light (BL) source (Thorlabs, #SOLIS-2C) was used as the low-coherence light source ( Fig. 1(a)), with the spectral spike at 450 nm blocked by a long-pass filter. The output spectrum is shown in Fig. 1(c), where the maximal incident power onto the specimen (S) was approximately 5 mW. The output of BL light beam was divided into reference   [18]) was 2.58 µm (NA: 0.8).
The reflectance between the N-BK7 glass plate (GP, thickness: 500 µm) and air in the reference arm was approximately 4%, which was appropriate for obtaining satisfactory interference efficiency for general fresh tissues. For the sample beam, all optical components are identical to those used in the reference arm. The specimen (S) was placed into a container (C) on a 3D linear stage (3D-LS; Thorlabs, #MLS203-1/MZS500-E), which included a biaxial DC-servo linear stage and a z-axial piezoelectric stage. Both reference and sample beams were reflected, combined, and then projected onto a monochrome CMOS camera through a tube lens (TL; Thorlabs, #TL200-CLS2, effective focal length: 20 cm). Raw images were then captured and processed. As shown in Fig. 1(a), a tiny displacement (equivalent to five interferometric carrier fringes, approximately 1.07 µm in tissue) along the z-axial direction (controlled by the 3D-LS) generated one en-face OCM image.
To enable focal and coherent planes be coplanar, the retarder (R) was adjusted through the projected images of the CMOS camera. Considering both confocal-gate (depth-of-focus) and coherent-gate (coherent-length) effects, the theoretical axial resolution R a in the tissue, equal to [(0.44λ 0 2 /n tissue ∆λ) −1 + (2λ 0 n tissue /NA 2 ) −1 ] −1 [18], was approximately 0.60 µm; however, the measured R a in the tissue based on the envelope FWHM of interferometric carrier wave shown in Fig. 1(d) was approximately 0.81 µm. By contrast, the estimated theoretical lateral resolution R l in the tissue was 0.45 µm (resolution power equal to 0.61λ 0 /NA), whereas the measured R l determined using a knife-edge test [18] was 0.60 µm near the surface of the GP. These measurements were due to the equivalent coating bandwidth of all optical components and the total effect of dispersion on light transmitted by GPs in both arms. On the basis of the imaging properties of O and TL, the equivalent pixel lateral resolution R l was determined to be 0.37 µm, whereas the real pixel lateral resolution R l measured according to the x-axial position of 3D-LS was 0.38 µm. In AcuOnPath, the field of view (FOV) calculated from the CMOS camera with 2390 (width) × 1936 (height) pixels was 908 µm (width) × 736 µm (height) in the tissue.
The CMOS camera has a maximum frame rate of approximately 160 frame/s. Raw images collected within 1 s were averaged to generate 20 new images with a higher SNR. These images enabled five reconstructed OCM images using a four-step phase shifting algorithm (phase: 0°, 90°, 180°, and 270°) [18]. Then, these five continuous single-wave images were averaged again to create a final high-quality en-face OCM image. OCM had an optimized depth scanning speed of 1.07 µm/s for an en-face OCM image, depending on the refractive index of the fresh tissue being tested. In other words, one OCM en-face image was output by averaging 20 raw images. The duration of each en-face OCM scan performed using the large-field stitched mode was set as 1 s, where the integral carrier fringe number was 5 and the equivalent scan depth was 1.07 µm in the tissue.

FM
As shown in Fig. 1(a), another collimated beam of UV light (λ 0 = 385 nm; Thorlabs, #SOLIS-385C) was used to excite Hoechst 33342 to allow nuclei imaging. The central wavelength λ 0 of the emission spectrum of Hoechst 33342 was approximately 500 nm. This excitation beam was reflected by a long-pass dichroic mirror (DM; Semrock, #LP02-407RU-25) before being focused onto S through O. The incident excitation UV light focused onto the sample was approximately 15 mW. Fluorescent light emitted from the nuclei of S was collected by O. Most residual excitation light reflected and scattered by S was blocked by the dichroic mirror, and the emitted fluorescence light was projected onto the CMOS camera through TL.
The focal plane is adjusted to penetrate deeper into the sample. The theoretical R l (0.38 µm, equal to 0.61λ 0 /NA, where λ 0 is 500 nm) of FM, which is caused by the decreased dispersion of single transmission trip, is better than that of OCM, for which the measured R l was 0.45 µm. However, the R a of FM was lower (depending on the depth of field) than that of OCM. The theoretical (depth of focus, equal to 2λ 0 n tissue /NA 2 , where λ 0 is 500 nm) and measured (by measuring FWHM of 50-µm-diameter fluorescent nanobeads) R a values were 2.19 and 3.03 µm, respectively. In addition, to obtain the highest frame rate for acquiring fluorescent images, the exposure time was set as 0.6 ms, which was consistent with the 160 frame/s of the CMOS camera. After completing the OCM process, eight FM images were collected and averaged to form a single high-quality image. For the CMOS camera, a high-quality image obtained through frame averaging was better than that obtained using longer integration time because switching integration time between OCM and FM modes require additional time that can affect the scan speed of a large-field stitched image. Finally, the output frame rate of the CMOS camera is 20 frame/s for FM, identical to that for OCM.
In contrast to confocal microscopy, this wide-field FM has a certain amount of background noise resulting from excited back-scattering fluorescence, which affects the contrast of nucleus images. Here, acquired FM images were obtained after background subtraction to enhance the contrast. A background subtraction algorithm [39] was used to enhance the nucleus image contrast, where the parameter of rolling ball radius was set to 16 before magnifying the subtracted nucleus image.

P-H&E, the composite image of OCM and FM
A simple color-coding method similar to that proposed by Gareau et al. [25] is shown in Fig. 2. The 8-bit gray OCM images were converted into 24-bit RGB images with dark-green color and black backgrounds; 8-bit gray FM images were converted into 24-bit RGB images with yellow-green color and black backgrounds. Subsequently, these two transformed images were combined (summed) and inverted to create 24-bit RGB format p-H&E images. Because of the tissue inhomogeneous property, induced nuclear background noise needs to be substantially eliminated from the image of FM for obtaining a composite high-quality p-H&E image. The FM image needs to be processed after background subtraction before the color-coding process. Otherwise, the obtained composite image would be blurred.

6
In contrast to confocal microscopy, this wide-field FM has a certain amount of background noise resulting from excited back-scattering fluorescence, which affects the contrast of nucleus images. Here, acquired FM images were obtained after background subtraction to enhance the contrast. A background subtraction algorithm [39] was used to enhance the nucleus image contrast, where the parameter of rolling ball radius was set to 16 before magnifying the subtracted nucleus image.

P-H&E, the composite image of OCM and FM
A simple color-coding method similar to that proposed by Gareau et al. [25] is shown in Fig.  2. The 8-bit gray OCM images were converted into 24-bit RGB images with dark-green color and black backgrounds; 8-bit gray FM images were converted into 24-bit RGB images with yellow-green color and black backgrounds. Subsequently, these two transformed images were combined (summed) and inverted to create 24-bit RGB format p-H&E images. Because of the tissue inhomogeneous property, induced nuclear background noise needs to be substantially eliminated from the image of FM for obtaining a composite high-quality p-H&E image. The FM image needs to be processed after background subtraction before the colorcoding process. Otherwise, the obtained composite image would be blurred. (R, G, B) is the 24-bit RGB gain set of an 8-bit gray-level image, where R, G, and B are all between 0 and 1. The color-coding process is demonstrated for OCM and FM images in Fig. 3: a gray-level scale cytoplasmic OCM image ICy(xi,yj) ( Fig. 3(a)) was multiplied by (R, G, B)Cy = (1, 0, 0) to obtain a red-level image with a black background, and a gray-level nucleus FM image INu(xi,yj) ( Fig. 3(b)) was multiplied by (R, G, B)Nu = (0, 1, 0) to obtain a green-level image with a black background. Subsequently, OCM and FM processing paths were combined to produce a red-and-green image with a black background (Fig. 3(c)). Typically for INu(xi,yj) had a yellow-and-deep-green color scheme with a black background. After being inverted, it became a pink-and-blue-violet color scheme with a white background, which is traditionally associated with H&E images [24]. (R, G, B)Cy and (R, G, B)Nu were respectively fine-tuned to be (0, 0.60, 0.05) and (0.80, 1, 0.21) by a histopathologist through comparison of numerous f-OCM scan images with many formalinfixed paraffin-embedded H&E images of original tissues. These two gain sets were used to output Fig. 3(d). Fig. 3 represents the images of the fresh porcine cerebrum tissue piece scanned using the f-OCM instrument after soaking in AcuSolutions pretreatment medium (PM) #2 for 5 min following process I shown in Fig. 4. The images reveal the details of neural and glial cells within the cerebral cortex, with nuclei clearly observed. (R, G, B) is the 24-bit RGB gain set of an 8-bit gray-level image, where R, G, and B are all between 0 and 1. The color-coding process is demonstrated for OCM and FM images in Fig. 3: a gray-level scale cytoplasmic OCM image I Cy (x i ,y j ) ( Fig. 3(a)) was multiplied by (R, G, B) Cy = (1, 0, 0) to obtain a red-level image with a black background, and a gray-level nucleus FM image I Nu (x i ,y j ) ( Fig. 3(b)) was multiplied by (R, G, B) Nu = (0, 1, 0) to obtain a green-level image with a black background. Subsequently, OCM and FM processing paths were combined to produce a red-and-green image with a black background (Fig. 3(c)). Typically 3) for I Nu (x i ,y j ) had a yellow-and-deep-green color scheme with a black background. After being inverted, it became a pink-and-blue-violet color scheme with a white background, which is traditionally associated with H&E images [24]. (R, G, B) Cy and (R, G, B) Nu were respectively fine-tuned to be (0, 0.60, 0.05) and (0.80, 1, 0.21) by a histopathologist through comparison of numerous f-OCM scan images with many formalin-fixed paraffin-embedded H&E images of original tissues. These two gain sets were used to output Fig. 3(d). Figure 3 represents the images of the fresh porcine cerebrum tissue piece scanned using the f-OCM instrument after soaking in AcuSolutions pretreatment medium (PM) #2 for 5 min following process I shown in Fig. 4. The images reveal the details of neural and glial cells within the cerebral cortex, with nuclei clearly observed.   4. Two processing methods of fresh tissue samples after dissection by using a surgical blade. Process I shows the material preparation of fresh tissue pieces for producing p-H&E images. Process II shows the material preparation of frozen-sectioned slices for image validation purposes, and comprehensive comparison between p-H&E and H&E-stained images.

Experimental results: p-H&E images of fresh tissue pieces and frozen-sectioned slice
As shown in Fig. 4, two additional processes were introduced for fresh tissue specimens obtained after dissection by using a surgical blade. In process I, the tested tissue specimen was soaked in AcuSolutions pretreatment medium (PM) #1 or #2 for 5 min, where PMs #1 and #2 will be described later. Subsequently, the soaked tissue specimen was carefully placed on the supporting material (soft sponge) of the container. The acceptable tissue size was less than 2 cm (W) × 2 cm (H) × 0.5 cm (T); thus, p-H&E imaging can be applied on thick tissue specimens. Then, a cover glass was carefully pressed and attached to the soaked tissue specimen after dropping PM #1 or #2 (using the same PM of previous soak) on the surface of the soaked tissue specimen. Finally, this container was positioned inside the AcuOnPath instrument for p-H&E scanned. In process II, the tested tissue surrounded by cryostat gel was placed on the aluminum chuck and then cooled down to −20°C for obtaining a frozen-sectioned slice. Then, a 10-µm-thick slice was produced and affixed on the 1-mm-thick glass substrate (GS, or glass slide). The specimen prepared using process II was obtained only for performing a comprehensive co-localized comparison between p-H&E and frozen-sectioned H&E images for image validation. The frozen-sectioned slice was completely submerged in PM #2 for 5 min. Then, a glass plate was slightly placed on the soaked frozen-sectioned slice and glass substrate for further p-H&E scanning in AcuOnPath.
The concentration of fluorescent dyes in PM #1 was 50 µg/mL (dissolved in 1× PBS); this concentration exerted a minimal effect on the section in case the soaked tissue specimen was required for follow-up RNA extraction at the biobank. The concentration of fluorescent dyes in PM #2 was 33 µg/mL; the dyes were dissolved in a glycol-compound buffer solution with 1.3% formaldehyde. Similarly, this process did not preclude the soaked tissue specimen from being used in further protocols, such as DNA extraction, immunohistochemistry (IHC), or formalin-fixed paraffin-embedded H&E staining.
The AcuOnPath instrument produces p-H&E images by stitching multiple side-by-side fields at a depth beneath surface of the fresh tissue specimen. Through the stitching method, a large-field p-H&E image of the dissected fresh tissue specimen was obtained, as indicated from the snapshot image of the real fresh porcine kidney tissue specimen in the container, where the scanned field could be cropped using our in-house developed AcuViewer interface. The scanned p-H&E images of the flatly dissected tissue specimens collected from different fresh porcine organs could be obtained for examining various histological structures. In addition, two thyroid and two fresh breast tissue specimens were tested before storing them in −80°C refrigerator for biobank application. Finally, to compare p-H&E images with traditional H&E images, a frozen-sectioned slice with a thickness of 10 µm was prepared and scanned using f-OCM. After acquiring the p-H&E image, this frozen-sectioned slice was stained with H&E dyes to validate the p-H&E imaging method.

Large-field p-H&E image of the normal fresh tissue piece
According to process I shown in Fig. 4(a) dissected fresh porcine kidney specimen was soaked in PM #2 for 5 min and then directly placed into a container for obtaining a large-field p-H&E image; the top-view snapshot of the container is shown in the top-right inset of Fig. 5(a). The cropped rectangular region (small green box) with 7 × 7 FOVs in the top-view snapshot of the container localized the real position of the stitched large-field p-H&E image shown in Fig. 5(a) (large green box) for the fresh tissue piece. The cropped area in the top-view snapshot of the container was limited by the window size of the glass plate (2 × 2 cm 2 ). Because of the degradation of the intact tissue specimen, many glomeruli in the fresh porcine kidney specimen were easily dropped after being cut. Figure 5(b) and 5(c) respectively shows the low-and high-magnified images from the red and blue boxes of Fig. 5(a). Figure 5(b) shows the low-resolution image of glomeruli and renal tubules from the red box of Fig. 5(a). Furthermore, Fig. 5(c) magnifies a single glomerulus and its structure in detail from the blue box of Fig. 5(a). In Fig. 5(c), basic structures in the glomerulus and renal tubule of the normal porcine kidney [40], such as Bowman's capsule, parietal epithelial cells, capillary loop, red blood cells (typically in the hollow position of the capillary loop), and basement membrane, are distinctly observed. The full-size scan of this dissected fresh porcine kidney specimen could be selectively cropped using the AcuViewer interface.

Normal fresh tissue pieces of animal organs
The p-H&E images of six fresh porcine organs (lymph node, liver, skeletal muscle, salivary gland, lung, and esophagus) are presented in Figs. 6 and 7. Each organ was obtained through slaughter, with the organs transported at a temperature of 4°C in 12 h. Before performing f-OCM scans, each fresh porcine tissue specimen was soaked in PM #2 for 5 min following process I shown in Fig. 4. Figures 6(b), 6(d), 6(f), 7(b), 7(d), and 7(f) represent blue-bounded regions shown in Figs. 6(a), 6(c), 6(e), 7(a), 7(c), and 7(e) respectively, magnified by a factor of 2. Figure 6(a) shows the p-H&E image of lymph nodes, revealing the population of nuclei, which was similar to the morphological property of tissue inflammation described in Fig. 6(b) in detail. In Fig. 6(b), the reticular fiber [41] could easily be observed because of the strong reflection of OCM; however, its observation through H&E staining appeared to be difficult [42]. In Fig. 6(c), showing the liver, the cell border between two adjacent hepatocytes could be readily identified, and the central vein and sinusoid [43] were well spaced, as shown in Fig. 6(d) in detail. In addition, as shown in Fig. 6(d), some hepatocytes were already autolytic and degraded. Figure 6(e) demonstrates the high spatial resolution of images by resolving the striations of muscular sarcomeres from the skeletal muscle.
Typically, the A-bands of the porcine skeletal muscle are fixed, with the measured sarcomere length representing the summation of A-and I-bands [44] at 2.2 µm. As shown in Fig. 6(f), the average sarcomere length was estimated to be 2.3 µm. Compared with alternative techniques used for imaging skeletal striations [45,46], including confocal and nonlinear microscopy, our method provides a convenient approach based on H&E-like images. In addition, the distributed heterochromatin of the single nucleus was discriminated. As shown in Fig. 7(a), the aggregations of mucous acini, serous acini, and sectioned intralobular ducts [47] were well arranged and are further indicated in Fig. 7(b), with the salivary gland showing similar signs of degradation as the liver, as shown in Fig. 6(d). Figure 7(b) reveals the proximity of intralobular ducts to the arteriole. Figure 7(c) shows the balloon-like structure in the p-H&E image of the lung [48], and the magnified image ( Fig. 7(d)) represents the optically sectioned contour of the alveoli and arteriole clearly. Figure 7(e) shows the squamous epithelium and lamina propria regions of the mucosa of the esophagus tissue distinguished by a basal layer [49], and Fig. 7(f) describes the regions in detail. To obtain high-quality p-H&E images of fresh lung and esophagus specimens, the cutting flatness of the attached side and soft attachment is crucial. A high-sharpness (small bevel angle) surgical blade can help improve the cutting flatness.

Biobank-provided thyroid and breast tissue pieces
Two human intraoperative fresh thyroid tissue specimens (one specimen was normal and the other one was cancerous) were prepared for scanning following process I (soaked in PM #1 for 5 min) shown in Fig. 4. These specimens were then placed in a freezer at −80°C for biobank storage. The normal and cancerous tissue specimens of the thyroid were scanned using the AcuOnPath system, producing p-H&E images shown in Fig. 8(a) and (c), respectively. Subsequently, both original tissue pieces were frozen and sectioned as 4-µm-thick slices to acquire traditional H&E-stained images for comparison, as shown in Fig. 8(b) and (d). In Fig. 8(a) and (b), characteristic follicles that comprise the ultrastructure of the thyroid gland are easily identifiable. The abnormal structures of tumor-invaded tissue regions (Fig. 8(c) and (d)) are enclosed by brown lines. In   this case, it is capable to determine the area percentage of tumor for tissue quality check before warehouse entry. However, the boundary seems much less clear in the p-H&E image. It may be due to tissue over press onto the glass plate (see Fig. 4), which is easy to induce blur morphology for boundary identification. In addition, two intraoperative fresh breast tissue specimens, of which one was normal and the other cancerous, were prepared for scanning. Similar to the comparison of Fig. 8(a) and (c), Fig. 8(e) and (g) shows the p-H&E images of the normal and cancerous breast tissue specimens acquired using this system. Figure 8(f) and (h) shows the corresponding frozen H&E images of the 4-µm-thick slice obtained from the original tissue specimens in Fig. 8(e) and (g). In Fig. 8(e), the morphology of adipocytes is more complete than that in Fig. 8(f) because the fatty tissue easily shrinks and is thus missed in a frozen section. In Fig. 8(g) and (h), the clusters of tumor regions are represented, and one of the clusters is segmented by a brown line as the cluster boundary. The use of all these biobank fresh tissue specimens was approved by the Institutional Review Board of Taipei Medical University (No. N202003041).

Image validation: Comprehensive comparison between p-H&E and H&E-stained images of the same frozen-sectioned slice
First, a dissected fresh metastatic SLN tissue with breast invasive ductal carcinoma (IDC) was prepared for frozen sectioning by surrounding it with cryostat gel (−20°C), and process II was followed, as shown in Fig. 4. Then, a 10-µm-thick frozen-sectioned slice was affixed to the surface of a 1-mm-thick GS. Next, the GS-mounted slice was soaked by dropping PM #2 (containing 1.3% formaldehyde for better attachment) for 5 min, and the glass plate was slightly placed on this GS-mounted slice for further scanning and stitching p-H&E images. To stain this prefixed frozen slice, the glass plate was removed before H&E staining. Finally, the H&E-stained slice was sealed using mounting medium. Images acquired using p-H&E and H&E-stained methods for the same 10-µm-thick frozen-sectioned slice are shown in Fig. 9(a) and 9(b), respectively Upon inspection of Fig. 9(a), the metastatic lesion in the SLN tissue could be readily identified in the light pink area of the lymphoid tissue. The same lesion in Fig. 9(b) appears purple. The micrometastatic lesion (200 µm < diameter < 2 mm) in the area enclosed by red lines can be observed in Fig. 9(c); a cluster of isolated tumor cells (ITCs; diameter < 200 µm, enclosed by green lines) is shown in Fig. 9(e). In Fig. 9(g), three single cancer cells are observed within a blood vessel (enclosed by blue lines). Figure 9(b) represents the co-localized H&E-stained image corresponding to the p-H&E image in Fig. 9(a). In Fig. 9(c) and 9(d), multiple clusters of IDC micrometastasis can be observed. In Fig. 9(e) and 9(f), cytoplasmic vacuolization, a characteristic feature of metastatic cancer, is evident. Comparing Fig. 9(g) with Fig. 9(h), single cancer cells in the tiny blood vessel surrounded by elastic fibers are resolved less clearly and are therefore more difficult to identify in the H&E-stained image.

Discussion
Because of system dispersion resulting from optical components, measured axial and lateral resolutions of f-OCM are not as good as theoretical resolutions. In this system, the types of dispersion affecting resolution performance mainly come from optical component mismatch and incident angle dispersion. Optical component mismatch, resulted from unbalanced spectrums of reference and sample arms, will form worse autocorrelation FWHM of interferometric signal, so the axial resolution will become poor. Incident angle dispersion, resulted from using thick GPs and their thickness difference between reference and sample arms, will become poor especially when high NA objective is used, so the lateral resolution will become poor. But, based on the real submicron spatial resolution of f-OCM and LEDs properties, p-H&E images obtained using f-OCM were still able to exhibit high feature identification for cellular structures. LEDs used in f-OCM have the following properties: (1) a broadband near-Gaussian spectrum for OCM that prevents a ghost image, thus providing a high-quality cytoplasmic image [50]; (2) stable illumination intensity for FM that helps the p-H&E image to be unified and reproduced [51]; (3) long illumination lifetime that results in less light source replacement [52]; (4) small space requirement resulting in a compact imaging system [52]; and (5) cost-effectiveness [53]. Because this system utilizes visible-range broadband light, it could enable pigment-absorbed imaging of hemoglobin (in red blood cells) [54,55] and melanin (in melanocytes) [56], like adding orthogonal polarization spectral imaging [57] or sidestream dark field imaging [58] for additional colors of rapid H&E images.
In contrast to using separate cameras for FM and OCM, this f-OCM method combines co-localized functional and morphological features in a p-H&E image. In addition, the f-OCM system achieves co-localization by compensating the z-axial difference between coherent and focal planes. This can improve the positional correctness of morphological features for a p-H&E image in the scanned tissue specimen. In contrast to traditional en-face scans that rely on the movement of the reflector in the reference arm, this f-OCM system can eliminate the dust/scratches (on the reflector surface of the reference mirror) pattern from interference images generated using a four-step phase shifting algorithm through moving the z-axial piezoelectric stage incrementally because of no lateral movement of the reference mirror; this is particularly effective for a high-NA objective lens. In theory, the use of a higher NA objective lens [14] and SIM [31] can improve the axial resolution of FM considerably, further improving the nucleus quality of a p-H&E image. However, SIM can easily induce a grid line artifact after the calculation of background subtraction when the modulation speed of a Ronchi ruling is unstable.
Porcine tissues have become the general surgical model for both training and research globally because their anatomical and physiological characteristics [59] are similar to those of humans. In this study, eight types of normal fresh tissue samples of porcine organs, namely the cerebrum, kidney, lymph node, liver, skeletal muscle, salivary gland, lung, and esophagus, were scanned to produce several kinds of high-quality p-H&E images for the proposed system validation, where most cellular features could be easily observed for histological identification. This f-OCM system provided high resolution power for large-field p-H&E scans, which was sufficiently demonstrated by muscular striations and nucleus heterochromatins of the fresh porcine skeletal muscle. Thus, the porcine p-H&E images of neural axons in the cerebrum, Bowman's capsule and renal tubules in the kidney, high cell-density lymphocytes in lymph nodes, border-resolved hepatocytes in the liver, mucous and serous acini distinguishable in the salivary gland, complete contours of alveoli in the lungs, and squamous epithelium and lamina propria distinguished by a basal layer in the esophagus were all clearly represented.
In terms of biobank samples, two fresh thyroid and two fresh breast tissue specimens were scanned using the f-OCM system, processed as p-H&E images, and compared with their frozensectioned H&E-stained counterparts for the same original tissue pieces. In the two p-H&E images of thyroid tissue specimens, normal follicles and aggregate tumor cells could be easily identified.
For fresh breast tissue pieces, tumor clusters of the IDC specimen were easily identified compared with the normal ones because normal tissues have simple structures including collagen fibers, adipocytes, connective tissues, and breast ducts. The toxicity of PM #1 was low and exerted a negligible effect on cell culture in vitro (data not shown) because it has minor-groove binding ligands and lower membrane permeability. Currently, before warehouse storage in a biobank, tissue specimens are frozen-sectioned using the H&E method; however, the average yield rate is still low (∼70%). With the low toxicity of PM and the high quality of p-H&E images, f-OCM scans can be useful for improving the tissue quality of fresh human tissue specimens before they are stored into the biobank. In addition, the typical size of fresh tissue specimens stored in a biobank is comparatively small when compared to intraoperative ones, thus requiring less time to scan; this is beneficial for the f-OCM system. In general, the total tissue processing time was less than 15 min before warehouse storage; thus, this system could effectively evaluate biobank tissue specimens.
To validate image similarity, we performed an in-depth comprehensive comparison between p-H&E and frozen-sectioned H&E images acquired from the same thick frozen-sectioned slices of fresh SLN tissue specimens that had been metastasized by an IDC. In general, the overall images of p-H&E and frozen-sectioned H&E images were quite similar. The key indicators of metastasis, including the presence of single cancer cells in a tiny blood vessel of the SNL, were resolved clearly in p-H&E images. Ultimately, p-H&E images revealed ultrastructural features in comparable details to traditional H&E-stained images. In addition, tiny blood vessel structures surrounded by elastic fibers and cytoplasmic vacuolization were clear in p-H&E images. The colors of p-H&E and H&E-stained images were not identical because this system used color space that was fine tuned by a previous default referring to paraffin-sectioned H&E images. Appropriate R, G, and B components to obtain high color similarity for frozen-sectioned H&E images were also available. Regarding the nuclei of cancer cells, some nucleoli were not represented in p-H&E images; these were identifiable in co-localized H&E-stained images because the RNA of nucleoli cannot be stained.
For rapid diagnosis using histopathology, the frozen-sectioned method is limited by many artifacts related to sample preparation (e.g., those resulting from fatty tissues and ice crystals). By contrast, our system, which uses Hoechst-stained fresh tissue specimens with 5-min soaking, provides high-quality p-H&E images at a depth of less than 50 µm beneath the surface of the attached fresh tissue specimen. The tissue-processing time of frozen-sectioned H&E slices starting from receiving the fatty or water-rich fresh tissue specimens is typically over 30 min on average. From a clinical perspective, compared with traditional frozen-sectioned H&E images, p-H&E images acquired using f-OCM could significantly shorten tissue-processing time to less than 15 min (1 × 1 cm 2 ) from receiving the fatty or water-rich fresh tissue specimens to delivering a pathological diagnosis. The tissue-processing time of the 1-cm 2 fresh tissue piece includes soaking the specimen (∼5 min), attachment and autofocus of the specimen (<5 min), and scanning of the specimen (∼4 min). Although the scan speed could be improved using a low magnification objective to increase the field of view, spatial resolution becomes poor.

Conclusion
Owing to its depth-scanning capability, the f-OCM system can capture continuous p-H&E images without any physical wastage of the tissue specimen. In total, the processing time from receiving an approximately 1-cm 2 fresh tissue specimen for scanning to generating the entire en-face p-H&E image was less than 15 min, which is two times faster than the time required for the typical frozen-sectioned H&E process for fatty or water-rich tissue specimens on average. In contrast to digital H&E visualization, image data generated by two independent optical sectioning methods provide more in-depth information that can be used to observe additional morphological features of individual gray-level channels [60], and the data can be used to train AI-based image segmentation methods to distinguish cancer tissues from normal tissues [61]. In short, the p-H&E imaging technique has high applicability for histopathology (including intraoperative histology and biopsies for brain and breast tumor diagnoses) and offers a viable alternative to the frozen-sectioning method. This method may assist veterinary and human surgeons in surgery-related treatment planning. In addition, digital p-H&E imaging can be used to realize telepathology in the field of rapid image diagnosis.