2.1.

Pathology OCT Scanner

The imaging device used in this study is a custom-designed system based on spectral-domain OCT [Fig. 1(a)]. The light source was a multiplexed superluminescent diode (IPSDW0825, InPhenix) with a center wavelength of 850 nm and a bandwidth of 100 nm (coherence length of $7.25\mu \mathrm{m}$). In the fiber-based OCT system, a 50/50 fiber optic coupler (TW850R2A2, Thorlabs) was used to separate the light into the sample and reference arms. Two triplet collimators (TC12APC-780, Thorlabs) were used to collimate the light in both arms. Polarization controllers (FPC030, Thorlabs) maintained the polarization state of both arms. In the sample arm, a two-axis galvanometer scanner (GVS002, Thorlabs) was used to scan the beam prompted by a digital-to-analog connector block (BNC-2110, National Instruments) controlled by a data acquisition (DAQ) card (PCI-6255, National Instruments). A low-NA scan lens ($\mathrm{NA}=0.055$, LSM03-BB, Thorlabs) was used as the imaging lens. In the reference arm, a glass block was used for dispersion compensation (LSM03DC, Thorlabs). The reflected light was relayed to a high-speed spectrometer (Cobra-S 800, Wasatch Photonics). The spectrometer detected a spectral range covering 750 to 930 nm and acquired A-lines at a rate of 250 kHz with 1024 A-lines per B-scan. Each B-scan was composed of multiple A-lines acquired across a transverse plane. The imaging range of the spectrometer was 3.00 mm. The 2048-pixel camera (OctoPlus, E2V) was controlled by a CameraLink framegrabber (PCIe-1437, National Instruments). The reference signal was acquired using a diaphragm shutter (SHB1T, Thorlabs) that is controlled by the digital-to-analog connector block.

Fig. 1

Schematic of the pathology imaging system. (a) UHS–OCT interferometer splits light from the broadband light source into the sample and reference arms and then relays light back to the spectrometer. Imaging is controlled by a GUI, which allows for rapid visualization and processing of OCT images. (b) Zoomed-in view of tissue block in pathology cassette in cassette holder. (c) Snake-scan imaging pattern allows the full specimen to be imaged. (d) Resulting stitched OCT represented in an en face projection that shows entire tissue surface structure. $\text{Scale bar}=3\mathrm{mm}$.

Our system scans across an area of up to $10\mathrm{cm}\times 10\mathrm{cm}$ using a motorized two-dimensional stage that automatically moves across the region of interest and stitches together sequential volumetric datasets (DDS100, Thorlabs). The stages were actuated by DC motor controllers (KBD101, Thorlabs). A customized cassette holder was machined from aluminum to place the tissue cassettes on the 2D scanning stage [Fig. 1(b)]. Tissue samples were placed on the lower right corner of the cassette, and the software was calibrated to begin snake-scanning from that corner [Figs. 1(c) and 1(d)]. The large field of view enabled by our system allowed us to visualize structures at both the macroscopic and microscopic levels. The user can operate the system by placing the tissue cassette in a holder on the 2D stage. The user-friendly acquisition software snake-scans the entire surface of the tissue. A schematic of the system is shown in Fig. 1. Based on the 250 kHz A-line rate, optical imaging of a typically-sized ($20\mathrm{mm}\times 20\mathrm{mm}$) pathology tissue cassette takes 40 s. Image quality only suffers minorly as the imaging speed increases (Fig. S1 in the Supplemental Material). We showcase our imaging results in different planes. The macroscopic visualization of the 3D data is visualized using $Z$-projections. An intensity projection along the $Z$-axis enables analysis of a stack by “flattening” it into a 2D representation (Fig. 2). Individual features can be seen in the $Z$-projections and directly compared with H&E histology. The cross-sectional features of the tissue can be seen in the $B$-scans, which provide further information on the tissue architecture.

Fig. 2

Different types of intensity projections along the depth ($z$-axis) of the tissue. Median, sum, maximum, average, and standard deviation projections highlight different features of this breast tissue with cystic change and IDC. The maximum projection shows the external features of the tissue, whereas other projection operations are more suited for examining the internal features of the tissue. The maximum shows a clear topological architecture, whereas standard deviation, median, and average show clearer tissue contrast. $\mathrm{Scale}\mathrm{bar}=3\mathrm{mm}$.

The system achieves $3.7\text{-}\mu \mathrm{m}$ axial resolution in tissue ($5.52\mu \mathrm{m}$ in the air), $3.7\text{-}\mu \mathrm{m}$ lateral resolution in tissue ($5.52\mu \mathrm{m}$ in air), and 1.1 mm 6-dB sensitivity fall-off. The power on the sample is 2 mW, the images are not averaged, and the average SNR is 95 dB at 250 kHz.

The clinical-grade system is easily transportable and compact and was engineered to be easy to use in the pathology clinical setting. The imaging device, accompanying optics, electronic controllers, DAQ system, and computer were fitted on a custom medical device cart (AFC Industries) (Fig. S3 in the Supplemental Material).

2.2.

OCT Software

We designed an acquisition software in Labview (National Instruments) that was implemented in a state machine to efficiently enable the application to perform a variety of complex tasks. Figure S2 in the Supplemental Material shows the graphical user interface (GUI). The user can adjust the acquisition parameters including A-lines/B-scan, B-scans/s (live view), B-scans/s (acquisition mode), B-scans/volume, camera exposure time, number of B-scans to average, and surface area scanned per volume through the galvanometer controls ($X$ and $Y$ voltage amplitudes). Dispersion compensation and spectral-shaping can be turned on and off. When imaging large areas, the user can input the $X$ and $Y$ dimensions to the snake-scan. The user inputs the lateral size ($x,y$) of the sample, and the motorized stage moves and acquires subsequent volumes in a snake-scan pattern across the sample with a small overlap between samples. There was no $Z$-translation; each of the volumes was automatically taken; the spacing between volumes was known and input into the acquisition parameters; and there was no overlap between volumes, so blending was not needed.

The OCT processing software was developed in the C programming language. The image processing sequence includes five steps: reference light intensity background subtraction, linear-$k$ interpolation, apodization, digital dispersion compensation, and Fourier transform. The method for calibrating the vectors for $k$-linearization and dispersion compensation has been previously described.^16–18 The average processing time is $0.08\mathrm{s}/\mathrm{B}$-scan.

When imaging large areas, the user can input the $X$ and $Y$ dimensions to the snake-scan (Fig. 1). The data are subsequently stitched in MATLAB.

2.3.

OCT Imaging within the Anatomic Pathology Standard of Care

In Table 1, we illustrate how our ultra-high speed-OCT (UHS-OCT) system meets the CAP IVM functional requirements. The requirements include (1) a portable imaging system; (2) simple (one-step) specimen preparation that must not have adverse effects on formalin-fixed paraffin-embedded (FFPE) histology, immunohistochemistry (IHC), in situ hybridization (ISH), or genomic testing; (3) an imaging time of under 5 min; (4) a 1 to $5{\mathrm{cm}}^2$ imaging field of view; (5) cellular-level resolution; and (6) easy use such that a technician would be able to operate the device without trouble.

Table 1

How we meet CAP IVM requirements. CAP IVM functional requirements for pathology applications of ex vivo microscopy (EVM).19

We performed all imaging in the Division of Anatomic Pathology at Columbia University Irving Medical Center. This study was approved by Columbia University Institutional Review Board (IRB) AAAS6938. We examined specimens from seven patients: two patients who underwent breast excisions (18 cassettes), two patients who underwent prostate excisions (eight cassettes), two patients who underwent lung excisions (six cassettes), and one patient who underwent pancreatic excision (four cassettes). Following surgical excision, the resected specimens were processed as per standard of care as outlined by the College of American Pathologists. Different organs have specific grossing recommendations. For instance, breast tissue requires submission of all gross lesions, random sections, and margins. Prostate tissue requires sections taken from apical anterior and posterior bilateral regions as well as seminal vesicles. The lung requires submission of mass lesions nearest pleural surfaces, adjacent lung parenchyma, bronchial margins, and lymph nodes. The pancreas requires submission of mass lesions and anterior surface and posterior margin, pancreatic and bile duct margin, lymph nodes, and duodenum (if applicable). However, they broadly followed this general protocol: slices were grossly inspected via visual inspection and palpation. In keeping with standard pathology workflow, tissues representing areas of interest and additional random sections were placed into cassettes (number and type dependent on organ). Cassettes were fixed in 10% formalin for 6 to 72 h (occasionally fixation followed serial slice sectioning). Within this fixation window, OCT imaging of tissues in cassettes was performed. Figure 3 illustrates the protocol for imaging. Following specimen orientation, inking, and serial sectioning into 3- to 5-mm slices, the physician assistants selected and submitted tissue into cassettes per standard of care. We imaged tissue samples in cassettes after they were formalin-fixed for a minimum of 6 h and a maximum of 72 h.

Fig. 3

OCT imaging within the pathology workflow. After surgical specimens are serially sectioned by physician assistants, they grossly identify blocks via visual inspection and palpation. OCT imaging of the block allows for real-time histology evaluation and determination of regions of interest.

2.4.

Pathology Examination

After OCT imaging, the tissues were processed and embedded into paraffin blocks, $4\text{-}\mu \mathrm{m}$ tissue sections were cut and stained, and H&E slides were reviewed by a pathologist. The pathology slide characterization was done by a pathologist without knowledge or view of the OCT images. This study did not alter the standard of care process of sampling, we evaluated samples by OCT as well as histopathology once chosen by the grosser.

The histological processing of breast specimens followed standard clinical procedures and was not altered for the OCT imaging study. All surgical specimens were evaluated by pathologists (H.H., H.G.) based on the guidelines of the College of American Pathologists. All key histologic features were identified. In situ and invasive carcinomas were graded and, when appropriate, staged, and margins were determined.

Photographs of the tissue blocks were taken prior to OCT imaging and tissue processing. Corresponding OCT volumes of the blocks were captured, H&E images of the tissue blocks were digitized, and all three were commonly oriented and juxtaposed. H&E slides were reviewed for interesting histologic features, and corresponding regions were identified in OCT.

3.1.

Imaging Breast Tissues

The following are various lesions of the breast as imaged by OCT and compared with H&E histology. These include normal tissue, non-neoplastic, and neoplastic changes of the breast.

OCT imaging of a prototypical example of normal breast tissue includes fibroadipose tissue, ducts, lobules (breast acini), and blood vessels (Fig. 4). The capacity to reliably identify normal tissue can be utilized to exclude areas with these features from tissue processing, enhancing the efficiency of the breast diagnostic pathology process.

Fig. 4

BREAST lumpectomy block containing normal breast parenchyma. (a) Photograph of specimen. (b) Average projection along $z$-axis. (c) H&E histology slide of block. (d) OCT en-face representation of mildly dilated duct; (e) corresponding H&E histology. (f) OCT en-face representation of fibroadenomatoid hyperplasia; (g) corresponding H&E histology. (h) OCT en-face representation of fibrotic region of breast with regions of collagen surrounded by adipose tissue. Inset represents breast acini that are a part of a lobule; (i) corresponding H&E histology. (j) OCT en-face representation of blood vessel; (k) corresponding H&E histology. (a)–(c) $\text{Scale bar}=3\mathrm{mm}$; (d)–(k) $500\mu \mathrm{m}$.

Furthermore, Fig. 5 demonstrates the capacity of OCT to identify nongrossly apparent heterogenous changes in tissue. The section includes non-neoplastic cystic changes, as well as invasive ductal carcinoma (IDC). An enlarged en-face image shows a cystic structure with a clearly defined outer boundary that is typical of cystic and apocrine metaplasia, a non-neoplastic breast lesion [Figs. 5(d) and 5(e)]. An OCT B-scan view of the lesion is shown in Fig. 5(h). Figures 5(f) and 5(g) show that OCT can identify IDC among cystic change, and an OCT B-scan (in a perpendicular plane) is shown in Fig. 5(i).

Fig. 5

BREAST lumpectomy block showing cystic changes and IDC. (a) Photograph of specimen, (b) standard deviation projection along $z$-axis, and (c) H&E histology slide of block. (d) OCT en-face representation of cystic and papillary apocrine metaplasia; (e) corresponding H&E histology. (f) OCT en-face representation of IDC; (g) corresponding H&E histology. (h) OCT B-scan through cystic and papillary apocrine metaplasia (arrow). (i) OCT B-scan through IDC (arrow). (a–c), (h), and (i) $\text{Scale bar}=3\mathrm{mm}$; (d)–(g) $500\mu \mathrm{m}$.

In another case, the gross photograph of a tissue block [Fig. 6(a)] shows what appears to be mature adipose tissue, exclusively. OCT and H&E scans of the tissue block [Figs. 6(b) and 6(c)], however, show that, in addition to adipose tissue, there is fibrous tissue admixed with cancer. Figures 6(d) and 6(e) show a higher power en-face representation of the admixed tissue, and Fig. 6(f) shows a B-scan through the admixed tissue. This illustrates how OCT achieves near-H&E detection capability of lesions. In Fig. 6(d), it can be observed that the left shows an area with texture that represent neovascularity and mild chronic inflammation. The inset shows the region in more detail, with more homogenous texture and less reflectivity. However, on the right side of the same fibrous area, the tissue appears brighter and more heterogenous with a small interspersed dark spot within the carcinoma. This region corresponds to a well-differentiated desmoplastic invasive carcinoma with reactive stroma interspersed toward the far-right side. Tumors that are desmoplastic (containing a mixture of myofibroblasts, collagen, and neovascularity) reflect stronger than epithelial-rich tumors. The differentiability of these two tissue types is confirmed by the OCT B-scan [Fig. 6(f)], which shows stronger penetration depth on the left side of the fat, and less penetration on the right side.

Fig. 6

BREAST lumpectomy block diagnosed with focus of invasive carcinoma. (a) Photograph of specimen. (b) Average projection along $z$-axis. (c) H&E histology slide of block. (d) OCT en-face representation of nodule containing reactive stroma with neovascularity and mild chronic inflammation (left) and well-differentiated IDC with reactive stroma (right). Insets show that reactive stroma is more heterogenous and lowly reflective in OCT compared with bright, homogenous cancer. (e) Corresponding H&E histology. (f) OCT B-scan through nodule showing that stroma shows heterogeneous texture and deeper depth penetration (left) compared with highly reflective and lower depth penetration cancer (right). (a)–(c) $\text{Scale bar}=3\mathrm{mm}$; (d)–(f) $500\mu \mathrm{m}$.

Figure S4 in the Supplemental Material shows an additional IDC instance.

3.2.

Imaging Prostate Tissues

It is well-documented that prostatic carcinoma is difficult to recognize grossly.²⁰ Here, we highlight tissues from prostatectomies demonstrating some typical pathologies. Figure 7 shows a block with clearly defined nodular, partly-cystic benign prostatic hyperplasia (BPH). The OCT en-face projection shows both macroscopic and microscopic features of the tissue specimen. Although the photograph in Fig. 7(a) shows what appears to be a large white mass, the region of interest is actually on the midleft side of the specimen with a focus of carcinoma [Figs. 7(b) and 7(c)]. Upon further inspection, a focus of cancer is visible in Figs. 7(f) and 7(g), where cancer is surrounded by non-neoplastic prostatic tissue, which is confirmed with the corresponding H&E histology slide.

Fig. 7

PROSTATE block containing nodular partly cystic BPH, prostatic carcinoma, and fibromuscular stroma. (a) Photograph of specimen. (b) Average projection along $z$-axis. (c) H&E histology slide of block. (d) OCT en-face representation of focus of prostatic carcinoma; (e) corresponding H&E histology. (f) OCT en-face representation of nodular BPH; (g) corresponding H&E histology. (a)–(c) $\text{Scale bar}=3\mathrm{mm}$; (d) and (e) $500\mu \mathrm{m}$.

Given its predominance in prostate tissue, we show further instances of BPH in Figs. S5–S7 in the Supplemental Material.

3.3.

Imaging Pancreas Tissues

Gross examination of the pancreas frequently highlights the inability to distinguish between atrophic and inflamed tissue from cancer. A block of pancreatic tissue from a Whipple procedure is shown in Fig. 8. The gross view [Fig. 8(c)] shows pancreatic tissue with a nonclassical lobulated architecture but without a clearly defined mass. Figures 8(d) and 8(e) show non-neoplastic, noninvoluted/residual exocrine pancreas. Figures 8(f)–8(h) show a zoomed-in view of lobulated clusters of non-neoplastic acinar glands. Figures 8(i)–8(k) show glands of pancreatic carcinoma.

Fig. 8

PANCREAS block of pancreatic adenocarcinoma. (a) Photograph of specimen. (b) Average projection along $z$-axis. (c) H&E histology slide of block. (d) OCT en-face representation of normal exocrine pancreas, which represents residual normal pancreas. (e) Corresponding H&E histology. (f) H&E histology of pancreatic acini representation from (d). (g) OCT en-face representation of pancreatic acini. (h) OCT B-scan through acini. (i) H&E histology of pancreatic adenocarcinoma glands with a lumen. (j) OCT en-face representation of adenocarcinoma. (k) OCT B-scan through carcinoma. (a)–(c) $\text{Scale bar}=3\mathrm{mm}$; (d)–(k) $500\mu \mathrm{m}$.

3.4.

Imaging Lung Tissues

Lung tissue specimens may also demonstrate poor correlation between gross and microscopic evaluation. In Fig. 9, we demonstrate a block of emphysematous tissue from a bilateral lung transplant. The texture of the tissue in the OCT en-face projection [Fig. 9(a)] shows the abnormal representation of air spaces within the lung, along with areas of fibrosis. The highlighted area in Figs. 9(d) and 9(e) shows a fibrotic focus, which is more homogeneously textured than the emphysematous component showing multiple cystic spaces in Figs. 9(f) and 9(g). There are also corresponding stitched B-scans in Fig. 9(h) which represent the smooth normal fibrotic tissue, and Fig. 9(i) shows the more heterogenous and lowly reflective emphysematous tissue with very clearly defined textural features.

Fig. 9

LUNG block with emphysema. (a) Photograph of specimen. (b) Average projection along $z$-axis. (c) H&E histology slide of block. (d) OCT en-face representation of normal fibrotic tissue pancreas; (e) corresponding H&E histology. (f) OCT en-face representation of emphysematous tissue; (g) corresponding H&E histology. (h) OCT B-scan of normal fibrotic tissue. (i) OCT B-scan of emphysematous tissue. (a)–(c) $\text{Scale bar}=3\mathrm{mm}$ and (d)–(i) $500\mu \mathrm{m}$.

An example of lung adenocarcinoma is demonstrated in Fig. S8 in the Supplemental Material.