2.1.

Optical Coherence Microscopy Imaging System

This study used a portable OCM prototype system designed for high-speed imaging at $1060\mathrm{nm}$ . A detailed description of the system is described in Ref. 28. The light source was a compact Nd:Glass femtosecond laser that was spectrally broadened in a high numerical aperture optical fiber to a bandwidth of $>200\mathrm{nm}$ at $1060\mathrm{nm}$ . High-speed phase modulation with an electro-optic waveguide modulator produced a heterodyne frequency of $1\mathrm{MHz}$ , which enabled fast raster scan imaging. Using special techniques to compensate chromatic dispersion and wavelength-dependent source polarization, a coherence gated axial resolution of $<4\mu \mathrm{m}$ was achieved. This corresponds to optical image slices thinner than traditional histologic sections. En face imaging was performed using a confocal microscope with fast galvanometer scanners and a $40\times$ water immersion objective lens (Zeiss Achroplan 440095). The measured transverse resolution was $<2\mu \mathrm{m}$ , and the axial point spread function width was $\sim 19\mu \mathrm{m}$ . Images were acquired at $2\text{frames}∕\text{second}$ over a field of view of $400\mu \mathrm{m}\times 400\mu \mathrm{m}$ with $500\times 750\text{pixels}$ . The detection sensitivity was $-98\mathrm{dB}$ with $10\mathrm{mW}$ of incident power. The system used a custom autofocusing technique to ensure that the coherence gate position was matched to the confocal gate position in the OCM images. Automated image processing after acquisition consisted of digital demodulation, pixel resampling to remove scanner hysteresis and correct aspect ratio, mild spatial filtering with a $3\times 3$ triangular kernel, square-root compression of signal values, and contrast enhancement. Images are displayed on an inverse grayscale color map, where black represents increased reflectivity.

An ultrahigh-resolution OCT system operating at $1060\mathrm{nm}$ was combined with the OCM system to perform co-registered OCT and OCM imaging. The OCT system was similar to that used in a previous in vitro imaging study.²⁹ The OCT and OCM systems shared the same optics, except for the objective lens, which was turret mounted to allow rapid interchange between high and low magnifications. Cross-sectional images measuring $3\mathrm{mm}$ transverse by $1.3\mathrm{mm}$ deep were acquired with $1344\times 1000\text{pixels}$ at $1\text{frame}∕\text{second}$ . The OCT system had a $14\text{-}\mu \mathrm{m}$ transverse resolution and $<3\text{-}\mu \mathrm{m}$ axial resolution. The OCT signal was demodulated and logarithmically compressed using an analog circuit before analog to digital conversion. Image processing after acquisition consisted of pixel resampling to the correct aspect ratio followed by contrast enhancement. Again, images are displayed on an inverse grayscale color map. The compact OCT and OCM systems were integrated into portable instrument carts and transported to the hospital for the imaging studies.

2.2.

Study Design and Imaging Protocol

Imaging was performed on freshly excised surgical specimens in the pathology laboratory at the Beth Israel Deaconess Medical Center and on freshly excised tissue acquired during upper and lower gastrointestinal endoscopy procedures at the endoscopy suite of the VA Boston Healthcare System. Imaging in the pathology laboratory has the advantage that it provides access to normal specimens, as well as larger and more intact specimens, compared with pinch biopsy samples obtained in the endoscopy suite. In contrast, biopsy imaging provided access to smaller, more diagnostically sensitive specimens that were not readily available in the pathology laboratory.

Imaging protocols were approved by the institutional review boards at the Beth Israel Deaconess Medical Center, the Massachusetts Institute of Technology, Harvard Medical School, and the Boston VA Healthcare System. Fresh surgical specimens were selected based on the presence of pathology upon gross examination, prompt arrival to the pathology laboratory, and large specimen size, which allowed normal and pathologic tissues to be collected from each specimen without interfering with routine diagnostic procedures. The study used excess tissue from surgical specimens deemed unnecessary for diagnosis by supervising pathologists. Imaging was performed within $\sim 2\text{to}3\mathrm{h}$ of excision.

Imaging studies in the endoscopy suite occurred as part of ongoing studies investigating endoscopic OCT. Subjects enrolled under the upper GI protocol were selected from those undergoing surveillance endoscopy and biopsy based upon a history of previously diagnosed Barrett’s esophagus, dysplasia, or adenocarcinoma. Subjects enrolled under the lower GI protocol were selected from those undergoing routine screening colonoscopy. Specimens consisted of biopsy samples as well as samples from snare polypectomy procedures.

In total, 75 samples were imaged from 39 patients. Sample sizes from surgical specimens typically measured approximately $1\times 1\times 0.6\mathrm{cm}$ . The size of pinch biopsies typically measured $0.3\text{to}0.4\mathrm{cm}$ in greatest dimension, while resected polyps ranged from $0.3\text{to}0.7\mathrm{cm}$ in maximum dimension. The high imaging speed of the OCM system enabled extensive survey of multiple sites in each specimen. In total, thousands of images were collected for review. Normal specimen subtypes included esophageal squamous mucosa (10), glandular mucosa of the stomach (3), small intestine (5), colon (11), and pancreas (2). Pathology specimen subtypes included columnar-lined esophagus (11), esophageal adenocarcinoma (1), celiac disease (1), inflammatory bowel disease (3), acute inflammation (2), chronic colitis (2), melanosis coli (1), tubular adenoma (11), hyperplastic polyp (5), colorectal adenocarcinoma (4), cholecystitis (2), and chronic pancreatitis (1). Specimens were classified based on histologic diagnosis by an experienced gastrointestinal pathologist.

To prevent tissue dehydration, specimens were immersed in isotonic phosphate-buffered saline. Imaging was performed through a coverslip, with a thin $\sim 100\text{-}\mu \mathrm{m}$ layer of transparent ultrasound gel between the coverslip and tissue surface to prevent distortion of surface architecture. Imaging was first performed using ultrahigh-resolution OCT over a $3\text{-}\mathrm{mm}$ field of view. The OCT and OCM images were acquired through the same microscope with variable, collinear objective lenses, allowing co-registration of the center of the OCM images to the center of the OCT data set. Once regions of interest were located on OCT images, OCM was performed of those areas. Registration of OCM and OCT images was approximate in real-time imaging, but was improved in post-processing of the data by using tissue landmark features to correlate the images. After imaging, the sample was prepared for histologic sectioning. Discarded samples from the pathology laboratory were inked to mark the OCT imaging plane before formalin fixation and histologic processing. Sections were cut in both cross-sectional and en face planes relative to the luminal surface to allow comparison to both OCT and OCM images. Specimens from the endoscopy clinic were placed in formalin in accordance with standard clinical protocol and processed for histology. Tissue sections were stained with hematoxylin and eosin, and digital photomicrographs of histology were recorded.

2.3.

Data Analysis

A selection of normal and pathologic conditions from this feasibility study is presented in this paper. Data analysis focused on qualitative image interpretation and correlation with histology. The image database and histology slide set were first reviewed, and representative specimen data selected for further evaluation based on several factors, including relative correlation with histology, overall image quality, and accurate representation of the larger data set. Representative photomicrographs of histologic sections were then made with the best effort attempt to provide correlation among features. OCT images proved more useful than OCM for histologic correlations, because the larger field of view allowed appreciation of architectural features visible in low-power fields. For presentation, OCT and OCM still frames were selected from the individual specimen data set to most accurately correlate with histologic features. Image correlation with histology was nonblinded since the purpose of this investigation was to establish correspondence between OCM and histology and understand which features could be visualized by OCM in a cross section of GI normal tissues and pathologies.

3.1.

Upper Gastrointestinal Tract

Figures 1 and 1 present representative ultrahigh-resolution OCT (UHR OCT) and histology images of normal esophageal squamous mucosa. The OCT image in Fig. 1 clearly delineates the layered cross-sectional architecture of the esophageal mucosa, including the cellular epithelial layer and underlying lamina propria and submucosa. UHR OCT cannot, however, identify cellular epithelial features. Images acquired with en face OCM of the same region of the same specimen are presented in Figs. 1 and 1. The OCM images have a smaller field of view than OCT, but the much higher transverse resolution enables visualization of squamous cells and their nuclei. Figures 1 and 1 show images at depths of $30\mu \mathrm{m}$ and $125\mu \mathrm{m}$ , respectively. In Fig. 1, nuclei and cell membranes can be readily visualized as highly scattering relative to the cytoplasm. The transition region between the stratified squamous epithelium and the underlying lamina propria can be seen in Fig. 1, demarcated by arrows. Small cells evident on the top portion of this image point out the change in cell size expected in going from the surface to the basal layer in a stratified squamous epithelium. The lamina propria appears highly scattering and disorganized relative to the cellular epithelium. Representative en face histology is shown for comparison, with Fig. 1 corresponding approximately to OCM image 1 and Fig. 1 to the image in 1.

Fig. 1

Normal squamous esophagus. (a) Ultrahigh-resolution cross-sectional OCT image. The layered mucosal structure including the epithelium (e), lamina propria (lp), and submucosa (sm) is visualized. No muscularis mucosa layer is seen in this cross section. Scale bar, $500\mu \mathrm{m}$ . (b) Corresponding cross-sectional histology, hematoxylin and eosin $4\times$ ; scale bar, $500\mu \mathrm{m}$ . (c) OCM en face image of squamous epithelial cells. Nuclei (n) and cell membranes are readily visible. Image depth is $30\mu \mathrm{m}$ . (d) OCM image near the basement membrane. The transition between cellular epithelium and more highly scattering loose connective tissue in the lamina propria is visible (arrows). Image depth is $125\mu \mathrm{m}$ . (e) and (f) Corresponding en face hematoxylin and eosin histology $(20\times )$ to (c) and (d), respectively. Scale bars (c) to (f), $100\mu \mathrm{m}$ .

The high resolution of OCM well below the tissue surface is further highlighted in Figs. 2, 2, 2, 2 , which present sequential OCM images taken at depths between $30\mu \mathrm{m}$ and $210\mu \mathrm{m}$ . Surface cells are larger and with lower relative nuclear to cytoplasm (N/C) ratio compared to cells at deeper levels. In this specimen, the papillae projections in the squamous mucosa can be visualized, with squamous cells appearing to swirl around them. At greater depths, the cellular epithelium disappears earlier over the ridges, as evident in Figs. 2 and 2. Figures 2, 2, 2, 2 show $3\times$ zoom views of the boxes identified in 2, 2, 2, 2. A clear decrease in the cell size, as well as an increase in the relative N/C ratio can be appreciated.

Fig. 2

(a) to (d) OCM imaging of squamous cellular progression in depth. Decrease in cell size and increased relative nuclear-to-cytoplasm ratio is evident in the images. Image depths for (a) to (d) are $30\mu \mathrm{m}$ , $90\mu \mathrm{m}$ , $180\mu \mathrm{m}$ , and $210\mu \mathrm{m}$ , respectively. Scale bars, $100\mu \mathrm{m}$ . (e) to (h) Zoom views $(3\times )$ of the regions highlighted in (a) to (d). Scale bars, $20\mu \mathrm{m}$ .

Figure 3 demonstrates the squamo-columnar junction marking the transition between squamous esophagus and gastric mucosa. The cross-sectional OCT image in Fig. 3 shows a fairly homogeneous squamous epithelium as well as the gastric pit architecture, as verified in the histology of Fig. 3. The en face OCM images in Figs. 3 and 3 show squamous cells immediately adjacent to the gastric pits. The image depth is approximately $100\mu \mathrm{m}$ . Normal epithelial gastric mucous cells can be identified surrounding the lumens of the pits.

Fig. 3

Squamo-columnar junction. (a) UHR OCT image. Esophageal squamous (s) and gastric (g) mucosa are highlighted, with gastric pit architecture identified (arrow). Scale bar, $100\mu \mathrm{m}$ . (b) Histology, hematoxylin and eosin, $10\times$ . (c) to (d) OCM images of the squamo-columnar junction. Squamous cells (s) and gastric glands (g) are visible with individual mucous cells lining the gastric pits (arrows). Scale bars, $100\mu \mathrm{m}$ .

Figure 4 shows OCT and OCM images of Barrett’s esophagus and corresponding histology acquired from a pinch biopsy specimen. The UHR OCT image in Fig. 4 shows a distinct architecture for columnar mucosa that differs from either the esophageal squamous or gastric glandular morphology. Structural heterogeneity results from glandular features at and beneath the mucosal surface. Representative cross-sectional histology is presented in Fig. 4. An OCM image from this same specimen is presented in Fig. 4 along with higher magnification histology in Fig. 4. Goblet cells, a marker of Barrett’s epithelium, appear on OCM as distinct, nonscattering inclusions within the epithelium. The OCM image also shows heterogeneity in scattering properties of the glandular epithelium. Contrast is generated within and between the adjacent columnar cells, which can be appreciated in parts of the epithelium by the striated appearance radiating away from the lumen.

Fig. 4

Barrett’s esophagus. (a) UHR OCT image of a biopsy specimen. Barrett’s glands (arrows) can be identified in the heterogeneous epithelium. Scale bar, $500\mu \mathrm{m}$ . (b) Histology, hematoxylin and eosin, $4\times$ . Scale bar, $500\mu \mathrm{m}$ . (c) OCM image of Barrett’s epithelium. Individual goblet-like cells (arrows) are identified surrounding the crypt lumen. Scale bar, $100\mu \mathrm{m}$ . (d) Histology, hematoxylin and eosin, $20\times$ . Scale bar, $100\mu \mathrm{m}$ .

Figure 5 shows UHR OCT and OCM images of invasive esophageal adenocarcinoma. The UHR OCT image in 5 demonstrates fine tissue heterogeneity and relatively low image penetration depth compared to squamous mucosa, as well as an absence of larger glandular structures generally seen on OCT images of Barrett’s esophagus. Histology in Fig. 5 confirms invasive adenocarcinoma. The OCM image of the tumor in Fig. 5 demonstrates profound structural heterogeneity consistent with the histology of Fig. 5. Darkly staining nuclei of malignant cells evident on histology appear as highly scattering spots on OCM. In addition, gland forming entities devoid of OCM signal are present throughout the tumor and chords of highly scattering stroma can be seen penetrating the tumor.

Fig. 5

Esophageal adenocarcinoma. (a) UHR OCT image. Scale bar, $500\mu \mathrm{m}$ . (b) Histology, hematoxylin and eosin, $4\times$ . Scale bar, $500\mu \mathrm{m}$ . (c) OCM image of adenocarcinoma. Malignant gland architecture is evident (black arrows). Highly scattering bands of stroma can also be identified (black arrow heads). In addition, highly scattering inclusions (white arrows) likely represent darkly staining malignant cell nuclei visible on histology. Image depth is $\sim 50\mu \mathrm{m}$ . Scale bar, $100\mu \mathrm{m}$ . (d) Histology, hematoxylin and eosin, $20\times$ . Scale bar, $100\mu \mathrm{m}$ .

3.2.

Lower Gastrointestinal Tract

Figure 6 shows UHR OCT and OCM images of normal colon in the region of the cecum. Crypt architecture can be appreciated by OCT with good correspondence to histology. OCM visualizes the individual crypt morphology with high resolution. Crypt lumens show high contrast relative to the epithelium, and individual goblet cells can be clearly identified in correspondence to histology.

Fig. 6

Normal colon. (a) UHR OCT image. Crypt epithelial architecture as well as the underlying muscularis layer is identified. Scale bar, $500\mu \mathrm{m}$ . (b) Histology, hematoxylin and eosin, $4\times$ . Scale bar, $500\mu \mathrm{m}$ . (c) OCM image of individual colonic crypts. Single goblet cells are identified within the crypt epithelium (arrows). Image depth is $\sim 75\mu \mathrm{m}$ . Scale bar, $100\mu \mathrm{m}$ . (d) Histology, hematoxylin and eosin, $10\times$ . Scale bar, $100\mu \mathrm{m}$ .

Figure 7 presents images and corresponding histology from a tubular adenoma of the colon. Histology in Figs. 7 and 7 confirms the presence of long, parallel crypts with enlarged, cigar-shaped nuclei typical of tubular adenomas. On UHR OCT, the crypt lumens can be identified, and the high axial resolution enables delineation of the columnar epithelial layer lining the crypts. OCM en face images show the crypts in the transverse plane and compare well to the representative histology in Fig. 7. Notably, the elongated, eccentric crypts with varying alignment differ markedly from the uniform field of smaller crypts of normal colon. Moreover, OCM visualizes striations in the epithelial layer that likely represent pseudostratified nuclei in the crypt epithelium (arrows).

Fig. 7

Tubular adenoma. (a) UHR OCT image. Crypt architecture with distinct epithelial lining is identified. Scale bar, $500\mu \mathrm{m}$ . (b) Histology, hematoxylin and eosin, $4\times$ . Scale bar, $500\mu \mathrm{m}$ . (c) OCM image of adenomatous crypts. Elongated, irregular crypts are visible with striated highly scattering epithelium representative of dysplastic nuclei (arrows). Image depth is $90\mu \mathrm{m}$ . Scale bar, $100\mu \mathrm{m}$ . (d) Histology, hematoxylin and eosin, $10\times$ . Scale bar, $100\mu \mathrm{m}$ .

Compared to the images of normal and adenomatous colon, OCT and OCM images of adenocarcinoma of the colon display a prominent loss of crypt architecture, as demonstrated in Fig. 8 . Similar features are visualized as in esophageal adenocarcinoma. UHR OCT images show relatively low image penetration and heterogeneous fine tissue architecture consistent with densely packed malignant glands, as seen in the histology in Fig. 8. The en face OCM image in Fig. 8 provides a high-resolution view of the tissue architecture and identifies individual gland-forming units within the tumor. In addition, highly scattering malignant nuclei are also prominent in the OCM images. Corresponding histology is provided in Fig. 8.

Fig. 8

Adenocarcinoma of the colon. (a) UHR OCT image. Fine scattering heterogeneity with interspersed malignant gland formation can be recognized on OCT (arrow). Scale bar, $500\mu \mathrm{m}$ . (b) Histology, hematoxylin and eosin, $4\times$ . Scale bar, $500\mu \mathrm{m}$ . (c) OCM image of tumor microstructure. Gland formation (black arrows) as well as highly scattering malignant nuclei (white arrows) are highlighted. Image depth is $80\mu \mathrm{m}$ . Scale bar, $100\mu \mathrm{m}$ . (d) Histology, hematoxylin and eosin, $20\times$ . Scale bar, $100\mu \mathrm{m}$ .