Understanding optical reflectance contrast for real‐time characterization of epithelial precursor lesions

Abstract Detecting early‐stage epithelial cancers and their precursor lesions are challenging as lesions could be subtle and focally or heterogeneously distributed over large mucosal areas. Optical coherence tomography (OCT) that enables wide‐field imaging of subsurface microstructures in vivo is a promising screening tool for epithelial diseases. However, its diagnostic capability has not been fully appreciated since the optical reflectance contrast is poorly understood. We investigated the back‐scattered intensities from clustered or packed nanometer scale intracellular scatterers using finite‐difference time‐domain method and 1‐μm resolution form of OCT, and uncovered that there existed correlations between the reflectance contrasts and the ultrastructural clustering or packing states of these scatterers, which allows us to interpret the physiological state of the cells. Specifically, both polarized goblet cells and foveolar cells exhibited asymmetric reflectance contrast, but they could be differentiated by the optical intensity of the mucin cup due to the different ultrastructural make‐ups of the mucin granules; keratinocytes could demonstrate varied cytoplasmic intensity and their cytoplasmic contrast was closely correlated with the packing state of keratin filaments. Further preliminary study demonstrated that these new understandings of OCT image contrast enables the characterization of precancerous lesions, which could complement the current morphology‐based criteria in realizing “virtual histology” and would have a profound impact for the screening and surveillance of epithelial cancers.


| INTRODUCTION
Epithelial cancers rank in the leading causes of mortality worldwide. [1][2][3] Although detecting cancers at their early or precancerous stages is associated with favorable prognosis, it remains a major challenge in routine clinical practice. In most cases, these early-stage lesions are difficult to be recognized by the eye [4][5][6][7] and can be focally and heterogeneously distributed over a large mucosal area. 5,6,8,9 Therefore, biopsies often have to be randomly performed on multiple portions of the mucosa, with the hope of sampling changes of particular clinical relevance, which unfortunately is frequently missed. 5,6,10 Histological analysis of stained, thin sections from resected materials remains the gold standard for a definitive diagnosis, where contrast agents are adopted to specifically highlight structures of interest with clear understandings on the underlying mechanistic chemistry of staining.
However, this workup is time-consuming and labor-intensive which limits the capability of clinicians to immediately characterize the lesions, possibly leading to unnecessary biopsies or the need for repeated biopsies. In addition, concerns on the costs from the histological assessment have been raised when lesions of limited clinical importance are increasingly found as is the case in diminutive colorectal polyps. 11,12 Optical imaging modalities have attracted significant interests for "virtual histology" where resolving cellular or subcellular information in situ is possible without the need of histological assessment. Confocal fluorescence endomicroscopy provides satisfactory cellular details by use of contrast agents that could selectively stain the nucleus, cytoplasm, or extracellular matrix. 13,14 However, the limited field of view (~400 × 400 μm) makes it difficult for wide-field imaging. 13,14 Optical coherence tomography (OCT) is capable of imaging subsurface microstructures at cellular resolution across a large mucosal area. [15][16][17][18] The reflectance (back-scattering) contrast is provided by the refractive index differences between microstructures. Whereas, unlike histological methods where the image contrasts have been fully understood and well matched with the physiological events in tissues, opticalreflectance image interpretation has ever relied on morphological or architectural similarities to histology, leaving the biological bases underlying the contrast poorly understood. 16,17,[19][20][21] This knowledge gap has relegated these modalities to nonspecific morphometric tools where reflectance signals that may carry diagnostically important information are missed, precluding their applications in a broader research or clinical arena.
Epithelial cells are typically specialized with variations in their intracellular inclusions, suited to the particular task in a specific organ. 22 The light scattering from cells has been investigated both experimentally and theoretically. While cell nuclei have a quasiuniform distribution of its compositions and consistently present low back-scattering in the core, 21,23,24 the cytoplasm is the predominant site reflecting the specialization-mediated intracellular variations and corresponding alterations in their back-scattered intensity. Chen et al. measured the back-scattered intensity of the nuclear cores and cytoplasm in the stratified squamous epithelium to be in the order of 10 −7 and 10 −6 , respectively, when they are normalized to a perfect reflector. 23 Saidi et al. used Mie theory to approximate skin tissue scattering but by assuming cells are homogeneous spheres of a single size. 25 Dunn and Drezek et al. used the finite-difference time-domain (FDTD) method to model light scattering from cells containing multiple organelles of arbitrary shape. 26,27 Their studies suggest that small organelles whose size is comparable to the wavelength of light play a more important role than the nucleus in scattering from a cell, 26,27 such as mitochondria 28 and melanin granules. 29 However, most of the previous efforts were focused on the scattering contributions from micrometer-scale quasi-spherical scatterers, and little is understood on the back-scattering (reflectance) from nanometer scale scatterers, in particular, clustered mucin granules and packed keratin filaments and microvilli. Furthermore, the effects of their biological variations on the back-scattered intensity of cells under different physiological states are largely unknown. In this study, we investigated the micro-and ultra-structural bases underlying the cytoplasmic optical reflectance contrast in a wide variety of specialized epithelia both theoretically using the FDTD method and experimentally using a cellular-resolution OCT (μOCT). We further validate the feasibility of the improved understandings on the optical reflectance contrast of epithelial cells for real-time characterization of epithelial cancers and their precursors.

| RESULTS
We firstly developed back-scattering models of mucin granules (Supporting Information Figure 1), microvilli (Supporting Information Figure 2), and keratin filaments (Supporting Information Figure 3) using the FDTD method according to transmission electron microscopic (TEM) images and previously published data ( Table 1).
The results show that these clustered or packed nanometer-scale scatterers may contribute significantly or even dominantly to cell back-scattering, depending on their clustering or packing states. The corresponding experimental realizations were carried out using μOCT on intact mammalian epithelia including simple columnar epithelia from swine stomach, colon and small intestine, and stratified squamous epithelia (SSE) from swine skin (orthokeratinized), esophagus (parakeratinized), and floor of month (nonkeratinized), respectively. 23 On one hand, we validated our theoretical predictions by correlating the cytoplasmic reflectance contrasts with the micro-or ultra-structural features of the above mentioned cytoplasmic inclusions pertaining to the physiological states of specialized epithelial cells; on the other, our theoretical analysis provide meaningful explanations of the mechanisms underlying the observed back-scattering phenomena. Thereafter, we conducted a preliminary study to evaluate the feasibility of the new understandings on the image contrast for the interpretation of back-scattering features in precancerous lesions using specimens from the mice esophageal and human gastrointestinal tracts.    heterogeneously filled with nanometer-scale, discrete, electron-dense granules circled by relatively low-electron-density surroundings; in contrast, those of goblet cells were homogenous and full of fused, lowelectron-density mucin granules (Figures 1d and 2d). We modeled mucin granules in foveolar cells as nanoscale spheres suspended in the cytosol. The refractive index data of mucin granules is not available so the refractive index range of cell organelles was used as the best estimate (Table 1). 30 The simulated normalized back-scattered intensity ranges from (6.7-18) × 10 −6 , which is one of the brightest intracellular structures investigated in this study (Supporting Information Table 1). We predicted that the back-scattering intensity from mucin granules of goblet cells were insignificant since there was no noticeable electron density change in the TEM images and thus simulation was not conducted.
In μOCT images, the asymmetric optical contrasts of mucinsecreting cells well reproduced the polarization feature of the cells (Figures 1a and 2a). In agreement with our simulation results, foveolar cells demonstrated a high reflectance contrast in the apical cytoplasm relative to the low intensity in the basal portion where nuclei resided ( Figure 1a,b, white arrows). These bright signals lining gastric mucosa signified well-polarized foveolar cells, a finding that has unfortunately long been overlooked 20,32 or misunderstood. 33,34 In contrast, goblet cells could be identified by the low reflectance contrast of their barrel-shaped apical cytoplasm relative to adjacent enterocytes whose apical portion was relatively bright (Figure 2a,b, yellow arrows). These back-scattering features of goblet cells agreed with previous reports in human and rat. 34,35 The normalized back-scattered intensity of the mucin cups of foveolar cells was measured to be (2.74 ± 0.09) × 10 −6 , which was significantly higher than that of goblet cells (0.28 ± 0.02) × 10 −6 ( Figure 8a; independent-samples t test; p < .01), which agrees well with their ultrastructural differences (Figures 1d and 2d) and the theoretical predictions.

| Reflectance contrast of brush border
Enterocytes are epithelial cells fulfilling the function of absorption, whose apical surfaces are characterized by the presence of microvilli driven by polarization. The tips of microvilli are tightly and evenly packed, and the surface formed by their tips is equivalent to a high grade optical surface (Supporting Information Table 2). We  (Table 1). 30  Supporting Information Table 2).
In μOCT images of the small intestine, we did identify high-

| Stratified squamous epithelia
In μOCT images of the squamocolumnar junction of albino rat cervix in vivo, keratinocytes presented high-scattering cytoplasm, in striking contrast with the adjacent low-scattering columnar cells (Figure 4a1,a2).
These observations indicated that the high cytoplasmic intensity was correlated with keratinization, which could also be supported by the evidences from albino rat vagina ex vivo whose epithelium could alternate between mucification and keratinization provoked by progesterone and estrogen respectively during estrous cycle ( In particular, in the swine esophageal epithelium, we could see clearly the variation of the packing state as the cell evolved from the prickle layer to the intermediate layer (Figure 6d and e). We developed FDTD models of the two packing states of keratin filaments (Supporting Information Figure 3), and the results show that the normalized backscattered intensity of tonofilaments in closely packed state was 5.46 × 10 −6 , which was larger than those in the loosely dispersed state 3.08 × 10 −6 (Supporting Information Table 3).
In μOCT images of swine SSEs with different degree of keratinization, we found that the cytoplasmic optical intensity of keratinocytes was indeed associated with the packing states of intracellular keratin filaments: cells presenting bright cytoplasm had filaments that were closely packed into bundles; those demonstrating relatively low intensity contained filaments that were loosely dispersed within cytoplasm ( Figures 5-7). Note that the basal layer of nonpigmented SSEs sometimes seemed to be low scattering although filaments in basal cells were also packed into electron dense bundles (Figures 5c, 6c, and 7c). It is possibly owing to the fact that these cells have large nucleocytoplasmic ratio and are generally crowded to each other so that the scattering feature at this region is dominated by that of nuclei. Besides, imaging artifacts such as light attenuation may also contribute to its low optical intensity relative to the overlying epithelial layers. Also, the keratinized layer in orthokeratinized SSE could demonstrate either high intensity or low intensity ( Figure 5a). The low intensity was possibly due to the fact that these terminally keratinized cells were highly dehydrated, compact, and filled only with keratin filaments, rendering a homogenous distribution of refractive index (Figure 5f).
Those cells rich in densely packed tonofilaments presented much higher normalized back-scattered intensity than those with loosely dispersed tonofilaments (Figure 8b

| Human esophagus (nonkeratinized) with intraepithelial neoplasia
In the μOCT images acquired from the Lugol's-positive clean margin of human esophageal specimen, we could read out characteristic architectural and cytologic information of the normal nonkeratinized epithelium (Figure 11c1,c2). Keratinocytes at the intermediate layer presented relatively low-intensity cytoplasm similar to those observed in nonkeratinized floor of month (Figure 11c1). By contrast, in the images obtained from Lugol's-negative area that was confirmed to be severe dysplasia pathologically, the optical intensity differences divided the epithelium into two portions: the surface high-intensity portion and the underlying low-intensity portion (Figure 11d1,d2). Barret's esophagus and stomach, 42,43 which is challenging to be targeted with traditional video-imaging modalities even when a rigorous "systematic biopsy" protocol is used due to inadequate contrast between IM and surrounding epithelium. 5,19 Besides, as cellular information could be differentiated based on intrinsic contrast, it would possibly eliminate labor-intensive staining procedures and concerns on the cytotoxicity of contrast agents. 13,14 In addition, although human foveolar cells contain exclusively of neutral mucins which is different from those of swine, their apical portion also present high optical intensity. This agreement in the back-scattering features of foveolar cells is possibly owing to that they share a similar ultrastructural feature in the mucin granules. 44 The characterization of colon polyps exemplifies another clinical benefit to understand the subcellular reflectance contrast. With the expanding implementation of colonoscopy for cancer screening, colorectal polyps are increasingly detected. 12 However, since more than 90% polyps are small with approximately 50% of them being nonneoplastic, 12,45,46 pathologic assessment is in most cases used only to determine surveillance intervals, 12,46 which give rise to significant healthcare costs. 12,47 Based on the understanding of reflectance contrasts established in this study, we show that pathology hallmarks like decreased goblet cells, pseudostratified nuclei and irregular crypt patterns in adenomatous polyps can be well recognized in μOCT images.
This capability offers the possibility to differentiate adenomatous polyps from non-neoplastic polys in real time, which would possibly save substantial pathology costs in colon cancer prevention. 12,47 The current study also provides us a comprehensive understanding of the reflectance signals from nonpigmented keratinocytes. We disclosed that the cytoplasmic reflectance intensity was correlated with the packing state of keratin filaments. With the biological bases underlying subcellular reflectance contrast of keratinocytes clarified, we demonstrate the capability to distinguish between differentiated cells and dysplastic cells in both the orthokeratinized and nonkeratinized SSEs. Since the packing state of keratin filament is associated with the level of keratinization, this knowledge may also be useful for grading cancer cells in the squamous cell carcinoma because the synthesis as well as the packing state of keratin filaments varied between well-differentiated and poorly-differentiated neoplastic cells. 48 Additionally, the reflectance contrast difference between the squamous and columnar epithelium at the squamocolumnar junction suggests a possibility to detect squamous metaplasia in the simple columnar epithelium, which is a preneoplastic change in organs like cervix and lung. 49 While few tools are available in optimally evaluating brush border except for histology, we demonstrate the possibility to evaluate this polarization-derived subcellular structure using μOCT noninvasively and in real-time. This ability may benefit the research on a number of microvilli related abnormalities, such as celiac disease 50 and microvillus inclusion disease. 51 F I G U R E 9 Subcellular reflectance contrast of human gastric mucosa with intestinal metaplasia imaged by μOCT ex vivo. (a1 and a2) OCT cross-sectional (a1) and en face (a2) images of a specimen with normal mucosa and the white arrows indicates foveolar cells. (b1-b4) OCT cross-sectional (b1 and b2) and en face images (b3) of another specimen from the region with intestinal metaplasia (red arrowheads): White arrows in (b1-b3) indicate foveolar cells and yellow arrows in (b1 and b2) represent "ectopic" goblet cells featured with low-scattering apical cytoplasm; corresponding pathology (b4) with PAS-AB mucin staining to highlight metaplastic area (red arrowheads). Scale bars, 50 μm Back-scattered contrast depends on complex processes of Mie and Rayleigh scattering, involving relative refractive indices, shape and orientation and clustering of structures, etc. The contrast seen in optical reflectance images is due to an ensemble of scattering events-those in the optical section that are modulated by those in the tissue layers above and below. Our numerical models were developed based on multiple assumptions in the shape, orientation, distribution, and refractive index of the clustered and packed scatterers. These simplification strategies are necessary since some of these data is not available or measurable. In our study, this short- 6 | MATERIALS AND METHODS

| OCT system
We used a μOCT system described in one of our previous works. 23 The

| Modeling of scatterers using FDTD method
Modeling of clustered or packed nanometer scale scatterers was conducted using Lumerical software (Lumerical FDTD Solutions, Vancouver, Canada). We followed the method of modeling reported in our previous study. 55 Simulating the reflectance imaging system of μOCT was conducted using MatLAB (MathWorks, MA) based on the scalar diffraction theory and the optical parameters of the sample arm optics, which are described in detail previously. 55 The spatial and optical parameters of models were obtained from TEM images and published data, and are provided in the Supporting Information Tables 1-3.

| Normalized back-scattering intensity calculation and measurement
The normalized back-scattered intensity is defined as the ratio between the back-scattered light intensity from the scatterers under investigation and the back-reflected intensity from a goldcoated mirror. In μOCT-based measurements, we firstly measured the back-reflected intensity from a gold-coated mirror, and then normalize all the back-scattered intensity to it. By doing so, the measured values are independent of the imaging system, and therefore, can be compared between different experiments and with the simulated values. We also conducted μOCT imaging of above-mentioned tissues except for the oral mucosa in another two white micropigs (female, 12 months) in vivo. We anesthetized the animals and surgically opened the lumen to access the epithelial surface of internal organs. 6.7 | Mouse esophagus and human specimens with epithelial precancerous lesions C57BL/6 mice (female, 6 weeks) treated with 4-nitroquinoline 1-oxide (4NQO) at 100 μg/mL in the drinking water were used to develop esophageal cancer. 31 The mice were sacrificed every 2 weeks starting in week 14 and the esophagus was harvested, opened longitudinally. μOCT images were sequentially acquired from 6-10 locations from the proximal to the distal end. Regions with visible tumors or intratumor vessels were excluded. Following OCT imaging, specimens were fixed for pathological analysis. OCT datasets with landmarks such as blood vessels and epithelial rete pegs that well-matched with the corresponding histopathology were selected for analysis. We acquired approval from the IACUC of NTU (ARF-SBS/NIE-A0319).

| Transmission electron microscopy
We also obtained images from human endoscopically resected specimens. Four esophageal specimens with pathologically confirmed severe dysplasia or superficial cancer were used and μOCT images were acquired from both the Lugol's-positive margin and the Lugol'snegative region (Supporting Information Figure 4a). In stomach, three specimens were included including two without epithelial lesions and one with IM. In the one with IM, μOCT images were obtained from the smaller depressed lesion which were confirmed with monofocal IM pathologically (Supporting Information Figure 4b; red arrow).
Eleven colorectal polyps including six nonadenomatous polyps and five adenomatous polyps were included, and μOCT images were acquired from 2 to 4 random locations for each specimen. Following image acquisition, all the specimens were fixed overnight and subject to a standard surficial specimen processing protocol. The specimen was sectioned at the interval of 2 mm in width and embedded in its entirety for histopathological diagnosis. Written informed consent was obtained from participants prior to the study and the use of human tissues was approved by the IRB at Renmin Hospital of Wuhan University (2017K-C053).

| Statistical analysis
A quantitative analysis of the cytoplasmic optical intensity were conducted among cells in normal swine epithelial tissues. We manually measured the averaging intensity from 50 randomly selected locations within an area of 10 × 8 pixels (width × height) for each location.
The data for each epithelial type were from the 1,024 images of one 3D OCT dataset. Independent-samples t test and one-way ANOVA was adopted to compare optical intensity in-between mucus-secreting cells and among keratinocytes at different level of maturation, respectively. Further Bonferroni post hoc test was performed if statistical differences were detected. A p < .05 was regarded as a statistically significant difference. All numerical values were presented as mean ± SE. All statistical analyses were conducted using SPSS software (IBM SPSS Statistics 23.0).