2.1.1.

Cross-polarized imaging

In CPI, tissue is illuminated with linearly polarized white light, and images are acquired through a second linear polarizer oriented to transmit only reflected light, with its axis of polarization oriented orthogonally to that of the illumination light. This serves both to reject specular reflection from the tissue surface and to enhance the appearance of subsurface vasculature. The top row of Fig. 1 shows the module designed to adapt the standard endoscope for CPI imaging.

To achieve CPI, white light from the endoscope light guides pass through linear polarizers ($2.15\times 3{\mathrm{mm}}^2$) contained in the cap, one in front of each end of the bifurcated light guide (indicated by the stars in the CPI filter module image of Fig. 1); this is done to accommodate the illumination channels that exist in the Pentax video endoscope. The module holds another linear polarizer ($4.6\times 4.8{\mathrm{mm}}^2$) in front of the CCD camera; the axis of this polarizer is oriented orthogonally to that of the illumination polarizers. These filters are fixed in the module using optically clear epoxy (Epo-Tek, Billerica, Massachusetts). The filters are cut from HN22 polarizing sheets (Knight Optical Ltd., UK); the thickness of the film is 0.75 mm, and the maximum transmission of two orthogonally oriented sheets is 0.001%. With the CPI module in place, the surface irradiance at the center of the FOV is approximately $3.6\mathrm{mW}/{\mathrm{cm}}^2$ at 5 mm and approximately $1.9\mathrm{mW}/{\mathrm{cm}}^2$ at 10 mm; both are typical working distances for imaging.

2.1.2.

Vital-dye fluorescence imaging

In VFI mode, tissue is illuminated with quasi-monochromatic light designed to excite fluorescence; a long-pass filter is placed in front of the CCD camera to reject specular reflection at the excitation wavelength and to allow for the collection of resulting fluorescence. The VFI module was designed specifically to image fluorescence following topical application of the fluorescent antiseptic proflavine hemisulfate, which has been shown to localize in cell nuclei. The agent has an absorption maximum at 450 nm and an emission maximum at 515 nm. The bottom row of Fig. 1 shows the module designed to adapt the standard endoscope for VFI imaging.

To provide illumination during VFI, a 455-nm laser diode (Nichia Corporation, Tokyo, Japan) was installed in the Pentax EPK-i; light from the blue laser diode was coupled to the bifurcating light guide used during white light imaging, and a mechanical control on the processor was used to switch between the two illumination modes. A driver (Wavelength Electronics, Bozeman, Montana) controls the input current to the laser diode and thus allows variation of the illumination intensity. The module holds a custom-designed long-pass filter ($4.6\times 4.8{\mathrm{mm}}^2$) passing wavelengths of more than 500 nm in front of the CCD for the collection of proflavine fluorescence. The filters were cut from a custom-coated colored glass filter (Schott North America, Inc., Duryea, Pennsylvania); the thickness of the filter is 0.8 mm, and the optical density (OD) at the illumination wavelength is 3.8. The filter was designed to have out-of-band rejection of at least OD 3 for light ranging from normal incidence up to 70 deg incidence, accommodating the angular FOV of the endoscope. The edges of the filter were coated black to minimize the effect of stray light on the image. The filters were installed using a medical-grade, optically clear epoxy (#301-2FL, Epo-tek, Billerica, Massachusetts). With the module and the blue laser diode in place, the surface irradiance at the center of the FOV is approximately $14.9\mathrm{mW}/{\mathrm{cm}}^2$ at 5 mm and approximately $7.5\mathrm{mW}/{\mathrm{cm}}^2$ at 10 mm; both are typical working distances for imaging.

2.1.3.

High-resolution microendoscopy

The MVE was designed to acquire widefield images in both CPI and VFI modes; in addition, it was designed to be compatible with a high-resolution microendoscope (HRME) capable of resolving subcellular detail in areas identified as suspicious during widefield imaging. The HRME has been previously described^20–22 and has been used during in vivo endoscopy to visualize gastrointestinal pathology using proflavine contrast.²³ In this system, the light from a 455-nm LED is coupled to a fiber-optic bundle for tissue illumination. The fluorescence from proflavine stained tissue is collected by the bundle while in contact with the tissue and is delivered to a CCD camera through a 550-nm bandpass filter with a cut-on wavelength of approximately $500\mathrm{nm}$. The field of view of the HRME is 720 μm in diameter, and the spatial resolution is 4.5 μm.

2.1.4.

Image capture

Image collection is achieved through a custom designed interface (Labview 2010, National Instruments, Austin, Texas). The program displays the HD signal ($1280\times 1024\text{pixels}$) via a DVI-to-USB capture card (Epiphan Systems, Ottawa, Ontario, Canada) and allows for the collection of both digital video and image frames. The interface also allows the user to adjust brightness and contrast where appropriate and to store information such as the date and time of the procedure.

2.1.5.

Instrument performance

The system resolution was determined by capturing an image of a U.S. Air Force resolution target (Newport Corp., Irvine, California). To assess background autofluorescence, images were acquired in VFI mode from a 2-in. diameter nonfluorescent frosted quartz disk (which approximates tissue reflectance) with settings that matched or exceeded those used for proflavine stained tissue. The measurements were used to verify the absence of excitation light leakage and to assess the performance of the emission filter.

2.2.1.

Ex vivo imaging procedure

The MVE was used first to acquire images of an esophagectomy specimen. Immediately following surgical resection, the mucosal surface of the specimen was rinsed with saline and imaged with a digital single-lens reflex (SLR) camera under white light illumination. Visually abnormal and normal sites were identified by the study pathologist based on appearance; borders of these regions were marked on the white light image.

Areas identified by the pathologist as grossly normal and abnormal were imaged using all three widefield imaging modalities; images were first acquired using WLI, then using CPI, and then in VFI mode. The specimen was imaged ex vivo at a fixed working distance of 5 mm to allow for optimal correlation between image sites and subsequent pathological assessment. Before VFI imaging, proflavine hemisulfate (0.01% w/v) was applied to the mucosal surface of the sample using a sterile cotton tip applicator. Excess proflavine was removed with dry gauze. Finally, high-resolution images were acquired using the HRME from clinically normal and abnormal areas.

In order to ensure the same FOV was imaged with each imaging modality, the distance from the edge of the specimen, landmarks such as vessels, islands of squamous or columnar tissue, and the perimeter of the squamo-columnar junction were tracked as each widefield modality was used. To track high-resolution imaging, widefield images were taken of the probe in contact with the mucosa, verifying that the images were obtained from the same FOV. In an effort to reduce sampling error, black or blue ink was placed at each area imaged, fixed with acetic acid, and photographed. Since the ink spread to approximately 2 to 4 mm in diameter, the photograph guided the approximation of image sites on the resected specimens necessary to facilitate registration between widefield imaging, high-resolution imaging, and subsequent histopathologic evaluation.

The specimen was then fixed in formalin and submitted for standard histopathologic analysis; vertical cross-sections were examined to grade and verify the presence of disease. The study pathologist, blinded to the image results, assigned diagnoses to histologic sections of inked areas using standard histologic criteria.

2.2.2.

In vivo imaging procedure

The MVE was next used in vivo to acquire images during endoscopic surveillance using three high-definition imaging modalities: WLI, CPI, and VFI. After endoscopic surveillance using WLI, the scope was removed, the module for CPI was placed on the distal tip, and the scope was reinserted. The scope was removed again after CPI, the VFI module was installed, and the scope was inserted a third time. Following the third insertion, proflavine (5 to 10 mL) was administered via spray catheter (Olympus America, Center Valley, Pennsylvania) on the epithelial surface, and VFI images were acquired. Finally, the endoscopist introduced the HRME probe via the instrument channel of the endoscope and placed the probe in gentle contact with the tissue surface. HRME images were obtained from sites considered clinically abnormal using widefield imaging. Additional HRME images were obtained from sites that were considered clinically normal by widefield modalities. Widefield imaging in all three modes and high-resolution imaging were performed by a single endoscopist.

In order to ensure the images in all modalities were taken at the same site, three pieces of information were recorded during the procedure: clinical landmarks indicated by the endoscopist (such as Barrett’s borders, islands, ulceration, or bleeding), endoscope depth and quadrant, and time stamps during the procedure (upper right hand corner of every video). Post-procedure, videos were evaluated frame by frame to ensure images from the same site were extracted from each modality. All extracted image frames associated with each image site were reviewed by researchers (NT, ML, SA, RRK).

2.2.3.

Endoscopic image criteria

The entire Barrett’s segment was scanned for apparent abnormalities with each image modality, using image criteria developed from previous ex vivo imaging studies.^19,20 CPI images were considered suspicious if they demonstrated increased and abnormal vascularization, with crowding and branching of blood vessels, or if there were areas of glandular effacement. VFI images were considered suspicious if they demonstrated alterations in glandular architecture or glandular effacement. HRME images were considered abnormal if they exhibited enlarged, crowded, and pleomorphic nuclei. Images were also considered abnormal if they revealed overlapping glands that were heterogeneous in size and shape with irregular luminal spacing. Many abnormal areas also exhibited loss of overall glandular architecture.^19,20

After imaging, biopsies were obtained from areas deemed suspicious in any of the imaging modes. Then, standard four-quadrant biopsies were taken every 1 to 2 cm of the BE segment. Images acquired from suspicious sites were compared to the histologic evaluation of the biopsy from the same site. In addition, biopsies of nonneoplastic regions containing Barrett’s metaplasia were obtained as controls.

3.2.1.

Ex vivo: MVE images of esophagectomy specimen

The MVE was used to acquire images from five sites on a single esophagectomy specimen. Histopathology results were available from all five sites; two were diagnosed as non-neoplastic Barrett’s, two were diagnosed as adenocarcinoma, and one was diagnosed as carditis. Figures 3 and 4 show representative images of neoplastic and non-neoplastic tissue from this specimen. Figure 3 compares images of BE (top row) and adenocarcinoma (bottom row) in WLI and CPI modes. As expected, CPI images show reduced specular reflection (black arrows) and enhanced vascular contrast. In images of Barrett’s metaplasia, vessels are better visualized in the cross-polarized image when compared to the same areas in the white light image (white boxes). In the images showing adenocarcinoma, not only are the vessels better visualized, but the increase in vasculature and vessel branching associated with neoplasia is more readily apparent (white box).

Fig. 3

The top row shows (a) white light image and (b) cross-polarized images of Barrett’s metaplasia. The bottom row shows (c) white light image and (d) cross-polarized image of intramucosal carcinoma. Black arrows show areas where specular reflection appears reduced in the CPI image. White boxes in the CPI image indicate areas where vessel branching is enhanced when compared with the same areas in the WLI. Images were acquired from an ex vivo esophageal specimen.

Fig. 4

The top row shows the (a) white light image, (b) vital-dye fluorescence image, and (c) high-resolution fluorescence image of proflavine-stained Barrett’s metaplasia. The white box in (b) indicates a region where glandular architecture is clearly visible. The bottom row shows the (e) white light image, (f) vital-dye fluorescence image, and (g) high-resolution fluorescence image of proflavine-stained intramucosal carcinoma. The yellow box in (f) indicates partial glandular effacement, and white boxes indicate complete effacement. The yellow arrows in (g) indicate small, irregularly shaped glands, and the yellow box indicates a gland with disrupted edges. The white box indicates an area of nuclear crowding. Corresponding histologic cross-sections are shown (d) and (f). Note the ink on the surface of each histologic section, verifying that the section was taken from the imaged area. Images were acquired from an ex vivo esophageal specimen.

Figure 4 shows images acquired from BE (top row) and adenocarcinoma (bottom row) in WLI and VFI modes. The VFI image of Barrett’s metaplasia allows for visualization of regular-appearing glandular architecture (white box). The edges of the glands are discernible, and the pattern appears consistent throughout the region. The corresponding HRME image shows nuclear staining primarily at the gland edges, indicating polarized nuclei, which is characteristic of non-neoplastic Barrett’s metaplasia.^19,20 This feature can also be seen in the corresponding vertical histology cross-section, which shows Barrett’s metaplasia.

The VFI image of adenocarcinoma allows visualization of partly effaced (yellow box) and completely effaced (white box) glandular architecture. The gland edges are no longer clearly discernible. The HRME image shows nuclear crowding, small and irregularly shaped glands (yellow arrows), loss of nuclear polarity within the glands (yellow box), and absence of glands in some areas (white box). These features are mirrored in the corresponding vertical histology cross-section, which shows adenocarcinoma.

3.2.2.

In vivo: MVE images obtained during endoscopy

In vivo imaging was conducted on one patient. During the endoscopic procedure using the MVE, three sites were imaged (using all four modalities) and biopsied. Two sites were diagnosed as non-neoplastic BE, and one was diagnosed as adenocarcinoma.

Figure 5 shows representative images from a site diagnosed as BE. In Fig. 5(a), the white light image shows glandular architecture with some vascular detail. The CPI image in Fig. 5(b) again shows enhanced vascular contrast. Indeed, both larger vessels and smaller vessels that were not visible during WLI become clear (white box). In the VFI image shown in Fig. 5(c), glands are present throughout the examined area. Gland borders appear wide, and there is little interruption between the border edges. High-resolution imaging with the HRME through the instrument channel of the endoscope in Fig. 5(d) allows for a magnified view of the features seen in VFI. Gland border edges appear regular (yellow boxes) and similar throughout the FOV. Nuclei within these regions appear polarized toward the gland edges. These features, along with intestinal-type goblet cells, are apparent in the corresponding vertical histology cross-section in Fig. 5(e), showing Barrett’s metaplasia, which is negative for dysplasia.

Fig. 5

In vivo images of an area of Barrett’s metaplasia in (a) white light, (b) cross-polarized, (c) vital-dye fluorescence, and (d) high-resolution imaging modes. The white boxes in (b) indicate areas where the appearance of vessel branching is enhanced. The yellow boxes in (d) show regular gland edges and indicate gland borders where nuclei are primarily polarized toward the edge. The corresponding histopathology cross-section is shown in (e).

Figure 6 shows images from a site diagnosed as adenocarcinoma. In Fig. 6(a), the WLI shows a flat, nonulcerated lesion with both vascular and glandular abnormalities. In the CPI image of the same area, shown in Fig. 6(b), the pattern of vessel branching is easier to assess when compared with WLI (white box). A VFI image of the same area, shown in Fig. 6(c), shows glands that appear partly effaced in some regions (white boxes) and completely effaced in others (yellow boxes). Proflavine staining is heterogeneous throughout the area. The HRME image shown in Fig. 6(d) reveals crowded glandular structures with irregular gland borders (yellow boxes), along with nuclear crowding both within the glands and outside of the glands (white box). Nuclei also appear pleomorphic. The histologic section presented in Fig. 6(e) was diagnosed as adenocarcinoma.

Fig. 6

In vivo images of an area of adenocarcinoma in (a) white light, (b) cross-polarized, (c) vital-dye fluorescence, and (d) high-resolution imaging modes. The white boxes in (b) indicate areas where appearance of vessel branching is enhanced when compared to (a). The white boxes in (c) indicate partial glandular effacement, and the yellow boxes indicate complete effacement. The yellow boxes in (d) show irregular gland edges, and the white box indicates a border where nuclei appear crowded. The corresponding histopathology cross-section is shown in (e), where black boxes indicate areas of LGD, HGD, and ACA.