|Home | About | Journals | Submit | Contact Us | Français|
Histological interpretation of disease is currently performed with static images of excised tissues, and is limited by processing artifact, sampling error, and interpretive variability. To demonstrate use of functional optical imaging of viable mucosa for quantitative evaluation of colonic neoplasia in real time.
Fluorescein (5 mg/ml) was topically administered in (n=54) human subjects undergoing screening colonoscopy. Fluorescence images were collected with 488 nm excitation at 12 frames/second with the confocal microendoscopy system. Movement of fluorescein in the transient period (<5 sec) and the lamina propria:crypt contrast ratio in the steady state phase (>5 sec) were quantified.
Normal mucosa showed circular crypts with uniform size, hyperplasia revealed proliferative glands with serrated lumens, and adenomas displayed distorted, elongated glands. For t<5 sec, fluorescein passed through normal epithelium with a peak speed of 1.14±0.09 μm/sec at t=0.5 sec, and accumulated into lamina propria as points-of-fluorescence that moved through the interglandular space with an average speed of 41.7±3.4 μm/sec. Passage of fluorescein through adenomatous mucosa was substantially delayed. For t>5 sec, high sensitivity, specificity, and accuracy was achieved using a discriminant function to evaluate the contrast ratio to distinguish normal from lesional mucosa (91%, 87%, and 89%, respectively; p<0.001), hyperplasia from adenoma (97%, 96%, and 96%, respectively; p<0.001), and tubular from villous adenoma (100%, 92%, and 93%, respectively; p<0.001).
Confocal imaging can be performed in vivo to assess the functional behavior of tissue in real time for providing pathological interpretation, representing a new method for histological evaluation.
Confocal microscopy provides clear images from thick layers of tissue using the principle of optical sectioning, and can achieve sufficient resolution to provide fine details about cells and sub-cellular structures.1 Recent advances in instrument miniaturization2–5 have allowed for the development of flexible, fibered confocal microscopes that have a size compatible with medical endoscopes and speed to image without motion artifact, enabling in vivo imaging in hollow organs to detect disease.6–10 These instruments can collect either backscattered light to provide details about tissue micro-architecture, or fluorescence images to provide insights into mechanisms of mucosal function, and thus provide an unprecedented opportunity to assess viable tissue in real time. Previous studies with fibered confocal microscopes demonstrated static images of mucosal microanatomy for comparison to histology.11–13 This investigation uses a novel confocal instrument to acquire dynamic images of tissue function for assessment of pathology.
We aim to observe functional differences in the uptake and distribution of topically applied fluorescein by colonic crypts to distinguish among normal mucosa, hyperplasia, and adenoma. Crypts are collections of epithelial cells organized into glands, whose lumens are exposed to the mucosal surface.14 The tight junctions in between neighboring cells form a barrier to luminal contents. Passage of fluid and solutes from the lumen into the lamina propria, a loose matrix of connective tissue that contains vasculature and lymphatics, is regulated by the epithelial layer. Specifically, crypt transport is mediated by transport proteins located on the apical (luminal) and basolateral (lamina propria) surfaces of the epithelial cell membrane. Solutes also can pass along a ‘paracellular’ pathway through regulated permeability of the intercellular tight junctions. Here, we assess the uptake of topically administered fluorescein, an FDA approved drug used clinically to detect epithelial defects in cornea,15 as a substrate for organic anion transport (OATP)16,17 and multidrug resistance (MRP) proteins,18,19 and as a marker of solute movement via the paracellular pathway.20,21 Effort was made to distinguish normal colonic mucosa from that exhibiting hyperplasia or adenomatous changes. Hyperplasia is a non-neoplastic proliferation of colonic mucosa that results from reduced exfoliation of normal epithelium, and adenoma is a pre-malignant condition that arises from unregulated epithelial growth. These lesions are commonly found on routine screening colonoscopy.
The confocal imaging system, Cellvizio®-GI (Mauna Kea Technologies, Paris, France), consists of a flexible (1.5 mm diameter) miniprobe, control unit, and processing software. A 488 nm (peak absorption of fluorescein) semiconductor laser (Coherent, Inc, Santa Clara, CA) delivers the excitation beam to a 4 kHz oscillating mirror for horizontal scanning (lines) and then to a 12 Hz galvo mirror for vertical scanning (frames). The mirrors raster scan the beam across the proximal face of an imaging bundle (Fujikura Ltd, Tokyo, Japan) that contains over ten thousand optical fibers (1.9 μm core diameter, 3.3 μm average intercore spacing). A gradient index (GRIN) microlens located at the distal end focuses the beam into the tissue (0 μm working distance) with a lateral and depth resolution of 3.5 and 15 μm, respectively, measured by a standard imaging target. Images are collected in a horizontal plane (en face) at 12 frames per second with a field of view of 600×500 μm2. Fluorescence is collected by the same lens, and refocused back into the illumination fiber. The cores of the fiber act as collection pinholes for rejecting out of focus light to perform optical sectioning. A long pass filter rejects the excitation light, and fluorescence is detected with an avalanche photodiode. Image processing performed includes subtraction of fiber autofluorescence and calibration of individual fiber transmission efficiencies.
IRB approval was granted by Stanford University Medical Center and the Veterans Administration Palo Alto Health Care Systems. Patients undergoing routine screening colonoscopy were recruited into the study, and informed consent was obtained. Those with coagulation abnormalities, poor preparation, or low tolerance for colonoscopy were excluded. A total of 54 subjects with mean age of 65±10 years were included. When lesional mucosa, including hyperplasia and adenomatous lesions, was seen on the conventional white light image, a standard spray catheter was inserted through the instrument channel, and 3 ml of 5% acetic acid was topically applied to the mucosa to remove mucus followed by 3 ml of 5 mg/ml fluorescein sodium (Akorn, Inc, Buffalo Grove, IL). No rinse was applied. As shown in Fig. 1a, the confocal microscope (blue light from laser) was passed through the instrument channel (3.7 mm diameter) of the colonoscope (Olympus CFQ-160), whose objective lens collects the standard video image produced by reflection of white light. The distal end of the confocal miniprobe was then placed gently and directly onto the lesional mucosa and the immediately adjacent normal mucosa using the endoscopic image for guidance, until the contrast reached the focal plane of the instrument, as shown in Fig. 1b. The site imaged was centered in the field of view of the endoscope to avoid sites of fluorescein accumulation. The total amount of time needed to obtain the measurements per site varied from 2 to 3 minutes, including passing the spray catheter and miniprobe. A pinch biopsy was obtained from all sites imaged, and processed for routine histology which was used as the standard to distinguish ‘normal’ from ‘hyperplastic’ and ‘adenomatous’ mucosa (stained with hematoxylin and eosin, H&E).
The captured video sequences were exported into individual images for evaluation. Images of normal and neoplastic mucosa from the same subject were grouped together. Those that displayed discrete glands with centrally located lumens and well-defined space in between the glands (lamina propria) were included in the analysis. The image immediately preceding the appearance of fluorescence signal in the running video record determined the start time (t=0). The transient (t<5 sec) behavior of fluorescein uptake and movement was measured using Volocity software ver 4.0.1 (Improvision, Coventry, England) by identifying regions of interest (ROI) using an intensity threshold of 35% of the maximum in each image, a level sufficient to reject spurious ROI’s that arise from noise. Image functionality was quantified by measuring the steady state fluorescence contrast ratio (CR) defined by a ratio of the mean intensity from the lamina propria to that of a crypt using an average of 3 sites chosen randomly from each region with an area of 35×35 μm2, sufficient to span several cells and to account for image variability. The identity of the imaged structures were established by rigorous validation of image scale and image plane, so as to ensure satisfactory mapping of the image to the histologic samples obtained from the same site. The intensity measurements were made without knowledge of the histology results; likewise, histology was evaluated in a blinded fashion, without knowledge of either the endoscopic white light or fluorescent image findings.
All results are expressed as mean ± standard deviation, and significance is determined using a paired t-test for ‘normal versus lesional’ and two-sided t-test with unequal variance for the others. The performance of the fluorescence contrast ratio to predict mucosal pathology was evaluated using a leave-one-out cross-validation method where one data point was removed from the set, and the remaining data were used to generate a linear discriminant function with STATA, ver 9.2 (College Station, TX).22 The data point removed was then evaluated by this discrimant function, and mucosal sites were designated as positive for disease if the fluorescence contrast ratio was less than the discriminant function (negative if above). This process was repeated for all of the data points in each set, and the results were compared to histology as the standard. Bootstrapping was performed to increase the number of data points in each set by random sampling with replacement. Diagnostic performance is defined by sensitivity = TP/(TP + FN), specificity = TN/(TN + FP), positive predictive value = TP/(TP + FP), negative predictive value = TN/(TN + FN), and total accuracy = (TP + TN)/(TP + TN + FP + FN), given the number of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN).23
Real time confocal fluorescence images were collected in vivo from colonic mucosa following topical administration of fluorescein in n = 54 human subjects, and no toxicity associated with contrast administration was reported. Movement of fluorescein in the direction from the crypt lumen to the lamina propria was observed, and is shown in Fig. 2, scale bar 20 μm. Contrast can first be appreciated in the crypt at t=0.08 sec, shown in Fig. 2a, where it then moves in a radial direction from the lumen toward basolateral aspect of the epithelial layer, and accumulates into focused intracellular points-of-fluorescence (dotted red circle) at t = 0.67 sec, shown in Fig. 2b, presumably intracellular vesicles. The fluorescence then enters the lamina propria at t = 0.75 sec, and remains as a point-of-fluorescence with travels with an average speed of 41.7±3.4 μm/s, shown in Fig. 2c. The fluorescence intensity in the crypt decreases, concurrent with the appearance of multiple points-of-fluorescence (mean surface area of 0.54×103 ± 0.02×103 μm2) in the lamina propria at t = 4.5 sec, shown in Fig. 2d. The radial speed of fluorescein travel within the crypt is shown in Fig. 2e for the transient phase (t< 5 sec), and a peak speed of 1.14±0.09 μm/s is achieved at t = 0.5 sec. The movement of two separate points-of-fluorescence within the lamina propria is shown in Fig. 2f, where the center of each is plotted in the horizontal plane at 0.083 sec intervals.
Steady state (t > 5 sec) images of normal colonic mucosa, hyperplasia, tubular adenoma and villous adenoma, along with the corresponding histology (H&E) are shown in Fig. 3, scale bar 40 μm. For normal mucosa, shown in Fig. 3a, individual colonocytes (c) can be seen surrounding the lumen (l) of the crypt, and the apical (ap) and basolateral (bl) epithelial cell membranes can be clearly distinguished. Contrast enhances the lamina propria (lp) in between the glands. The crypts are circular in shape, approximately uniform in size, and have oval lumens. For hyperplasia, shown in Fig. 3b, proliferative colonocytes (c) line an irregularly shaped lumen (l) that displays inward buckling in a serrated or saw-toothed appearance. The crypts are circular in shape but significantly larger in size than that of normal mucosa. Contrast can be seen in the lamina propria (lp) beyond the border of the gland. For the tubular adenoma, shown in Fig. 3c, distorted colonocytes (c) surround an elongated lumen (l), and the crypts appear eccentric in shape and are slightly larger in size than that of normal mucosa. Contrast appears to accumulate along the basolateral (bl) border of the epithelial cells rather than the apical (ap) surface. Moreover, the fluorescence intensity in the lamina propria (lp) appears reduced. For the villous adenoma, shown in Fig. 3d, enlarged colonocytes (c) surround a significantly elongated lumen (l). Contrast accumulates within the colonocytes in a punctuate fashion with multiple filling defects. The adenomatous mucosa exhibits a papillary appearance with elliptical mucosal profiles which are significantly larger in size than that of normal mucosa. Moreover, the fluorescence intensity in the lamina propria (lp) of these papillary profiles is significantly reduced.
The lamina propria:crypt fluorescence contrast ratio for normal mucosa (n=54), hyperplasia (n=24), tubular adenoma (n=25), and villous adenoma (n=5) are shown in the scatter plot in Fig. 4. For all subjects, a greater value was found for normal than for lesional mucosa, results shown in separate columns. On average, the fluorescence contrast ratios (mean±std dev) for normal, hyperplasia, tubular adenoma, and villous adenoma are 1.29±0.24 (range 0.85 to 1.74), 0.92±0.10 (range 0.69 to 1.08), 0.60±0.10 (range 0.43 to 0.81), and 0.41±0.03 (range 0.38 to 0.44), respectively, and are represented by the horizontal lines. The type I error rate (p-value) was < 0.001 for 1) normal versus lesional mucosa (including hyperplasia, tubular adenoma, and villous adenoma), 2) hyperplasia versus adenoma (including tubular and villous), and 3) tubular versus villous adenoma.
By defining contrast ratios below the discriminant function as positive for lesional mucosa (above as negative) the tradeoffs among sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy for distinguishing normal from neoplastic mucosa can be determined, shown in Fig. 5a. The optimal tradeoff between sensitivity and specificity is found from the receiver-operator characteristic (ROC) curve, shown in Fig. 5b, where the discriminant function of 0.0 results in a sensitivity, specificity and accuracy of 91%, 87% and 89%, respectively. Moreover, the area-under-the-curve (AUC) of 0.97 reflects the quality of this test (AUC = 1 is ideal). Similarly, use of the contrast ratio to distinguish hyperplasia from adenoma (dysplasia) is shown in Fig. 5c, where values below the discriminant function are defined as positive for dysplasia (above as negative). The corresponding ROC curve is shown in Fig. 5d and has an AUC = 0.99. A discriminant function of 0.5 corresponds to a sensitivity, specificity and accuracy of 97%, 96%, and 96%, respectively. Finally, the ability to distinguish tubular from villous adenomas is shown in Fig. 5e, where values below the discriminant function are defined as positive for villous (above as negative). The corresponding ROC curve is shown in Fig. 5f and has an AUC = 0.98. A discriminant function of −1.0 achieves a sensitivity, specificity and accuracy of 100%, 92%, and 93%, respectively. Bootstrapping the data with 200 points in each group was performed, but did not reveal any significant differences in the results.
Here, we demonstrate a novel method of functional optical imaging to perform accurate pathological evaluation of living tissue, using a fibered confocal microscope by observing crypt uptake and distribution of topically applied fluorescein. In the transient period (t < 5 sec), we observed contrast enter through the crypt lumen, travel across the epithelium, and accumulate in basolateral intracellular vesicles that dissipate in the lamina propria. Subsequently, the lamina propria:crypt fluorescence intensity ratio was used to assess pathology during steady state (t > 5 sec). This approach is quantitative, thus may improve our ability to make a definitive evaluation of disease and improve patient outcomes in situations where conventional pathology has shown limitations. For example, although pathologists are readily able to distinguish ‘adenomatous’ (dysplastic) polyps obtained during screening colonoscopy from normal and hyperplastic mucosa in the colon, the assessment of dysplasia in the setting of ulcerative colitis24 and Barrett’s esophagus25 is plagued by lack of diagnostic consistency, as evidenced by human interpretive differences, and significant inter- and intra-observer variability.26,27
In this study, dynamic observation of the behavior of topical fluorescein provides immediate ability to distinguish dysplasia from hyperplastic and normal mucosa. Indeed, the result found for normal mucosa (CR > 1) suggests that crypt transport mechanisms play a more significant role than paracellular permeability for maintaining fluorescein homeostasis, and the finding for lesional mucosa (CR < 1) implies that transport is impaired in transformed mucosa, resulting in a potentially very accurate method for discriminating between normal and lesional mucosa in situ. Furthermore, the results for distinguishing hyperplasia from adenoma and tubular from villous adenoma are promising, and suggest that this method may be a useful clinical tool in that regard as well. While the diagnostic performance of this method was quite good in this study, the results may differ for a new validation sample or in clinical practice. This study does not permit anatomic identification of the points-of-fluorescence that accumulate in the lamina propria, but it seems reasonable to note that the lamina propria is a heterogeneous tissue compartment so that sequestration of the fluorescent solute may well be occurring.
The fibered confocal microscope used in this study has the size and flexibility to pass through the instrument channel, and to be accurately placed onto the mucosa with guidance from the white light image. Furthermore, topical administration of contrast is safe and adequate for visualizing epithelial function. Optical sections (working distance 0 μm, depth resolution 15 μm) were collected near the mucosal surface to best visualize luminal uptake of contrast. Previous studies of endoscopic confocal images used intravenously administered fluorescein to demonstrate instantaneous snapshots of tissue that exhibit a striking resemblance to subsequent histology.11–13 However, observation of functional mucosal physiology, such as contrast uptake, would be challenging at the imaging speeds (0.8 frames per second) used. Furthermore, while separate optics are used for collecting the white light and confocal images, the components are packaged together in the distal end of the endoscope, a feature that limits use of the white light image to guide placement of the confocal window.
This study represents a first demonstration of the use of physiological properties of viable tissue to perform assessment of pathology. Future applications may include targeted detection with topically administered peptides,28 antibodies,29 and small molecules.30 Functional imaging is commonly performed in clinical radiology for the study of solid organs. For example, positron emission tomography (PET) assesses tissue metabolism of intravenously administered 18fluorodeoxyglucose (FDG) to distinguish benign from malignant tumors.31 The continued development of functional methods with optical imaging has great importance for the detection and evaluation of pre-malignant lesions in the epithelium of hollow organs, where greater than 80% all cancer originate and early detection may be preventative.32 Furthermore, with use of medical endoscopes, fibered instruments can be placed in close proximity to tissue and can collect images with sub-cellular resolution in real time, a level of performance that far exceeds that of other imaging modalities. Finally, methods of functional optical imaging can be generalized to evaluate pathology in other hollow organs, such as bladder, breast, cervix, colon, lung, oropharynx, pancreas, skin and stomach.
The authors would like to acknowledge funding support from the National Institutes of Health, including NIDDK K08 DK067618 (TDW) and NCI U54 CA105296, and the Doris Duke Charitable Foundation (TDW), and thank Mauna Kea Technologies for providing use of the Cellvizio®-GI imaging system for these studies.
Grant Support: NIH/NIDDK K08 DK067618 (TDW), NIH/NCI U54 CA105296, Doris Duke Charitable Foundation (TDW)
Conflicts of Interest: No conflicts of interest exist. There are no investigator conflicts of interest that have been disclosed to study participants.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.