Accuracy and interrater reliability for the diagnosis of Barrett's neoplasia among users of a novel, portable high-resolution microendoscope.

The high-resolution microendoscope (HRME) is a novel imaging modality that may be useful in the surveillance of Barrett's esophagus in low-resource or community-based settings. In order to assess accuracy and interrater reliability of microendoscopists in identifying Barrett's-associated neoplasia using HRME images, we recruited 20 gastroenterologists with no microendoscopic experience and three expert microendoscopists in a large academic hospital in New York City to interpret HRME images. They prospectively reviewed 40 HRME images from 28 consecutive patients undergoing surveillance for metaplasia and low-grade dysplasia and/or evaluation for high-grade dysplasia or cancer. Images were reviewed in a blinded fashion, after a 4-minute training with 11 representative images. All imaged sites were biopsied and interpreted by an expert pathologist. Sensitivity of all endoscopists for identification of high-grade dysplasia or cancer was 0.90 (95% confidence interval [CI]: 0.88-0.92) and specificity was 0.82 (95% CI: 0.79-0.85). Positive and negative predictive values were 0.72 (95% CI: 0.68-0.77) and 0.94 (95% CI: 0.92-0.96), respectively. No significant differences in accuracy were observed between experts and novices (0.90 vs. 0.84). The kappa statistic for all raters was 0.56 (95% CI: 0.54-0.58), and the difference between groups was not significant (0.64 vs. 0.55). These data suggest that gastroenterologists can diagnose Barrett's-related neoplasia on HRME images with high sensitivity and specificity, without the aid of prior microendoscopy experience.

High resolution microendoscopy for classification of colorectal polyps.

BACKGROUND AND STUDY AIMS: It can be difficult to distinguish adenomas from benign polyps during routine colonoscopy. High resolution microendoscopy (HRME) is a novel method for imaging colorectal mucosa with subcellular detail. HRME criteria for the classification of colorectal neoplasia have not been previously described. Study goals were to develop criteria to characterize HRME images of colorectal mucosa (normal, hyperplastic polyps, adenomas, cancer) and to determine the accuracy and interobserver variability for the discrimination of neoplastic from non-neoplastic polyps when these criteria were applied by novice and expert microendoscopists. METHODS: Two expert pathologists created consensus HRME image criteria using images from 68 patients with polyps who had undergone colonoscopy plus HRME. Using these criteria, HRME expert and novice microendoscopists were shown a set of training images and then tested to determine accuracy and interobserver variability. RESULTS: Expert microendoscopists identified neoplasia with sensitivity, specificity, and accuracy of 67 % (95 % confidence interval [CI] 58 % - 75 %), 97 % (94 % - 100 %), and 87 %, respectively. Nonexperts achieved sensitivity, specificity, and accuracy of 73 % (66 % - 80 %), 91 % (80 % - 100 %), and 85 %, respectively. Overall, neoplasia were identified with sensitivity 70 % (65 % - 76 %), specificity 94 % (87 % - 100 %), and accuracy 85 %. Kappa values were: experts 0.86; nonexperts 0.72; and overall 0.78. CONCLUSIONS: Using the new criteria, observers achieved high specificity and substantial interobserver agreement for distinguishing benign polyps from neoplasia. Increased expertise in HRME imaging improves accuracy. This low-cost microendoscopic platform may be an alternative to confocal microendoscopy in lower-resource or community-based settings.

Confocal microscopy and molecular-specific optical contrast agents for the detection of oral neoplasia.

Using current clinical diagnostic techniques, it is difficult to visualize tumor morphology and architecture at the cellular level, which is necessary for diagnostic localization of pathologic lesions. Optical imaging techniques have the potential to address this clinical need by providing real-time, sub-cellular resolution images. This paper describes the use of dual mode confocal microscopy and optical molecular-specific contrast agents to image tissue architecture, cellular morphology, and sub-cellular molecular features of normal and neoplastic oral tissues. Fresh tissue slices were prepared from 33 biopsies of clinically normal and abnormal oral mucosa obtained from 14 patients. Reflectance confocal images were acquired after the application of 6% acetic acid, and fluorescence confocal images were acquired after the application of a fluorescence contrast agent targeting the epidermal growth factor receptor (EGFR). The dual imaging modes provided images similar to light microscopy of hematoxylin and eosin and immunohistochemistry staining, but from thick fresh tissue slices. Reflectance images provided information on the architecture of the tissue and the cellular morphology. The nuclear-to-cytoplasmic (N/C) ratio from the reflectance images was at least 7.5 times greater for the carcinoma than the corresponding normal samples, except for one case of highly keratinized carcinoma. Separation of carcinoma from normal and mild dysplasia was achieved using this ratio (p<0.01). Fluorescence images of EGFR expression yielded a mean fluorescence labeling intensity (FLI) that was at least 2.7 times higher for severe dysplasia and carcinoma samples than for the corresponding normal sample, and could be used to distinguish carcinoma from normal and mild dysplasia (p<0.01). Analyzed together, the N/C ratio and the mean FLI may improve the ability to distinguish carcinoma from normal squamous epithelium.

Optimizing modulation frequency for structured illumination in a fiber-optic microendoscope to image nuclear morphometry in columnar epithelium.

Fiber-optic microendoscopes have shown promise to image the changes in nuclear morphometry that accompany the development of precancerous lesions in tissue with squamous epithelium such as in the oral mucosa and cervix. However, fiber-optic microendoscopy image contrast is limited by out-of-focus light generated by scattering within tissue. The scattering coefficient of tissues with columnar epithelium can be greater than that of squamous epithelium resulting in decreased image quality. To address this challenge, we present a small and portable microendoscope system capable of performing optical sectioning using structured illumination (SI) in real-time. Several optical phantoms were developed and used to quantify the sectioning capabilities of the system. Columnar epithelium from cervical tissue specimens was then imaged ex vivo, and we demonstrate that the addition of SI achieves higher image contrast, enabling visualization of nuclear morphology.

Detection of Entamoeba histolytica by Recombinase Polymerase Amplification.

Amebiasis is an important cause of diarrheal disease worldwide and has been associated with childhood malnutrition. Traditional microscopy approaches are neither sensitive nor specific for Entamoeba histolytica. Antigen assays are more specific, but many cases are missed unless tested by molecular methods. Although polymerase chain reaction (PCR) is effective, the need for sophisticated, expensive equipment, infrastructure, and trained personnel limits its usefulness, especially in the resource-limited, endemic areas. Here, we report development of a recombinase polymerase amplification (RPA) method to detect E. histolytica specifically. Using visual detection by lateral flow (LF), the test was highly sensitive and specific and could be performed without additional equipment. The availability of this inexpensive, sensitive, and field-applicable diagnostic test could facilitate rapid diagnosis and treatment of amebiasis in endemic regions.

Cervical human papillomavirus infection and intraepithelial neoplasia: a review.

Cervical human papillomavirus (HPV) infections and intraepithelial neoplasias are precursors to cervical cancer, the second most common cancer in women worldwide. HPV satisfies the epidemiologic criteria for causality; the role of other cofactors is under study. Natural history studies show that most low-grade lesions (productive HPV infections) regress or persist, whereas high-grade lesions (those with integrated HPV DNA) progress. Immunobiologic studies demonstrate that infection peaks in the early 20s, leading to a 10- to 20-year period of persistent infection, before finally progressing to a preinvasive or invasive lesion. Papanicolaou (Pap) screening has lowered the morbidity and mortality from cervical cancer in every country in which screening programs have been introduced. The diagnostic strategy for an abnormal Pap smear includes colposcopy; the role of HPV DNA testing in screening or diagnosis remains unclear. Patients are treated with cervical ablation, cone biopsy, or chemopreventive agents. Efforts to strengthen screening and prevention, as well as new directions for research, are needed.

Biomarker modulation in a nonhuman rhesus primate model for ovarian cancer chemoprevention.

OBJECTIVE: The objective of this study was to explore whether a nonhuman primate model could be developed to test drugs for the prevention of ovarian cancer. METHODS: Nineteen adult female Rhesus macaques were given fenretinide (4HPR), oral contraceptives (OCP), the combination (4HPR + OCP), or no medication for 3 months. Exploratory laparotomy was done pre- and postdrug to assess intermediary biomarkers of neoplastic phenotype, proliferation, response pathways, and growth-regulatory and metabolic markers. Fluorescence emission spectra were plotted for each group pre- and postdrug and means were overlaid on these plots and normalized. Fluorescence intensities were compared using the 2-tailed Student t test, (P = 0.1-0.01). RESULTS: All monkeys tolerated drugs and surgeries without difficulty. Histochemical markers showed no significant trend. However, fluorescence spectroscopy showed increased intensity at 450 nm excitation, 550 nm emission correlating with increased FAD presence. The 4HPR group (P = 0.01) showed higher intensity than the OCP group (P = 0.05-0.07) when compared with the controls. Decreased emission was seen at 350 nm excitation, 450 nm emission correlating with decreased NAD(P)H presence. The OCP group showed the largest change (P < 0.01), and the control group showed the smallest change. CONCLUSIONS: The nonhuman primate is an excellent model to test drug effect on the ovarian surface epithelium and merits additional study. Fluorescence spectroscopy was the most sensitive marker for drug activity and the apparent increase in NAD and FAD in the 4HPR group is consistent with the effect of 4HPR observed in cell culture. The differences between the OCP and the 4HPR groups suggest a different mechanism of activity of these drugs.

A pulsed finite-difference time-domain (FDTD) method for calculating light scattering from biological cells over broad wavelength ranges.

We combine the finite-difference time-domain method with pulse response techniques in order to calculate the light scattering properties of biological cells over a range of wavelengths simultaneously. The method we describe can be used to compute the scattering patterns of cells containing multiple heterogeneous organelles, providing greater geometric flexibility than Mie theory solutions. Using a desktop computer, we calculate the scattering patterns for common homogeneous models of biological cells and also for more complex representations of cellular morphology. We find that the geometry chosen significantly impacts scattering properties, emphasizing the need for careful consideration of appropriate theoretical models of cellular scattering and for accurate microscopic determination of optical properties.

Reflectance spectroscopy with polarized light: is it sensitive to cellular and nuclear morphology.

We present a method for selective detection of size-dependent scattering characteristics of epithelial cells in vivo based on polarized illumination and polarization sensitive detection of scattered light. We illustrate the method using phantoms designed to simulate squamous epithelial tissue and progressing to epithelial tissue in vitro and in vivo. Elastic light scattering spectroscopy with polarized illumination/detection dramatically reduces background signals due to both diffuse stromal scattering and hemoglobin absorption. Resulting spectra can be described as a linear combination of forward and backscattering components determined from Mie theory. Nuclear sizes and refractive indices extracted by fitting experimental spectra to this model agree well with previous measurements. Reflectance spectroscopy with polarized light can provide quantitative morphological information which could potentially be used for non-invasive detection of neoplastic changes.

Fiber confocal reflectance microscope (FCRM) for in-vivo imaging.

In-vivo imaging can be achieved with a coherent-fiber-bundle based confocal reflectance microscope. Such a microscope could provide the means to detect pre-cancerous lesions in the cervix by characterizing cells' nuclear-to-cytoplasmic ratio. In this paper we present the design of such a fiber confocal reflectance microscope, with an emphasis on its optical sub-systems. The optical sub-systems consist of a commercially available microscope objective and custom designed telescope, scan lens, and coupling lens systems. The performance of the fiber confocal reflectance microscope was evaluated by imaging a resolution bar target and human cervical biopsy tissues. The results presented in this paper demonstrate a lateral resolution of 2 microm and axial resolution of 6 microm. The sensitivity of the system defined by the smallest refractive-index mismatch that can be detected is approximately Delta n 0.05.

Low Temperature Fluorescence Imaging of Freeze-trapped Human Cervical Tissues.

We characterized the fluorescence intensity distribution within the epithelia and stroma of frozen human cervical tissues at the following excitation-emission wavelength pairs: 440, 525 nm and 365, 460 nm. The intensities at both excitation-emission wavelength pairs are significantly lower in the epithelia of severely dysplastic tissues, relative to that in normal and inflammatory tissues. Furthermore, there are small differences in (1) the epithelial intensity of severe dysplasia and mild dysplasia at 440, 525 nm and (2) the stromal intensity of inflammatory and severely dysplastic tissues at 365, 460 nm. A comparison of the ratio of intensities at 440, 525 nm and 365, 460 nm between the epithelia of each tissue type indicates that this ratio is lowest in severely dysplastic tissues. It is interesting to note that the epithelial and stromal intensities are comparable at 365, 460 nm; however, at 440, 525 nm, the epithelial intensity is more than a factor of two less that that of the stroma for all tissue types.

Near real time confocal microscopy of amelanotic tissue: dynamics of aceto-whitening enable nuclear segmentation.

High resolution, in vivo confocal imaging of amelanotic epithelial tissue may offer a clinically useful adjunct to standard histopathologic techniques. Application of acetic acid has been shown to enhance contrast in confocal images of these tissues. In this study, we record the time course of aceto-whitening at the cellular level and determine whether the contrast provided enables quantitative feature analysis. Confocal images and videos of cervical specimens were obtained throughout the epithelium before, during and post-acetic acid after the application of 6% acetic acid. Aceto-whitening occurs within seconds after the application. The confocal imaging system resolved sub-cellular detail throughout the entire epithelial thickness and provided sufficient contrast to enable quantitative feature analysis.

Optimal fluorescence excitation wavelengths for detection of squamous intra-epithelial neoplasia: results from an animal model.

Using the hamster cheek pouch carcinogenesis model, we explore which fluorescence excitation wavelengths are useful for the detection of neoplasia. 42 hamsters were treated with DMBA to induce carcinogenesis, and 20 control animals were treated only with mineral oil. Fluorescence excitation emission matrices were measured from the cheek pouches of the hamsters weekly. Results showed increased fluorescence near 350-370 nm and 410 nm excitation and decreased fluorescence near 450-470 nm excitation with neoplasia. The optimal diagnostic excitation wavelengths identified using this model - 350-370 nm excitation and 400-450 nm excitation - are similar to those identified for detection of human oral cavity neoplasia.

A comparison of C/B ratios from studies using receiver operating characteristic curve analysis.

In receiver operating characteristic (ROC) curve analysis, the optimal cutoff value for a diagnostic test can be found on the ROC curve where the slope of the curve is equal to (C/B) x (1-p[D])/p[D], where p[D] is the disease prevalence and C/B is the ratio of net costs of treating nondiseased individuals to net benefits of treating diseased individuals. We conducted a structured review of the medical literature to examine C/B ratios found in ROC curve analysis. Only two studies were found in which a C/B ratio was explicitly calculated; in another 11 studies, a C/B ratio was based on a so-called holistic estimate, an all-encompassing educated estimate of the relative costs and benefits relevant to the clinical situation. The C/B ratios ranged from 0.0025 (tuberculosis screening) to 2.7 (teeth restoration for carious lesions). Clinical scenarios that are directly life threatening but curable had C/B ratios of less than 0.05. This analysis led us to construct a table of ordered C/B ratios that may be used by investigators to approximate C/B ratios for other clinical situations in order to establish cutpoints for new diagnostic tests.

Fluorescence spectroscopy for diagnosis of squamous intraepithelial lesions of the cervix.

OBJECTIVE: To calculate receiver operating characteristic (ROC) curves for fluorescence spectroscopy in order to measure its performance in the diagnosis of squamous intraepithelial lesions (SILs) and to compare these curves with those for other diagnostic methods: colposcopy, cervicography, speculoscopy, Papanicolaou smear screening, and human papillomavirus (HPV) testing. DATA SOURCES: Data from our previous clinical study were used to calculate ROC curves for fluorescence spectroscopy. Curves for other techniques were calculated from other investigators' reports. To identify these, a MEDLINE search for articles published from 1966 to 1996 was carried out, using the search terms "colposcopy," "cervicoscopy," "cervicography," "speculoscopy," "Papanicolaou smear," "HPV testing," "fluorescence spectroscopy," and "polar probe" in conjunction with the terms "diagnosis," "positive predictive value," "negative predictive value," and "receiver operating characteristic curve." METHODS OF STUDY SELECTION: We found 270 articles, from which articles were selected if they reported results of studies involving high-disease-prevalence populations, reported findings of studies in which colposcopically directed biopsy was the criterion standard, and included sufficient data for recalculation of the reported sensitivities and specificities. TABULATION, INTEGRATION, AND RESULTS: We calculated ROC curves for fluorescence spectroscopy using Bayesian and neural net algorithms. A meta-analytic approach was used to calculate ROC curves for the other techniques. Areas under the curves were calculated. Fluorescence spectroscopy using the neural net algorithm had the highest area under the ROC curve, followed by fluorescence spectroscopy using the Bayesian algorithm, followed by colposcopy, the standard diagnostic technique. Cervicography, Papanicolaou smear screening, and HPV testing performed comparably with each other but not as well as fluorescence spectroscopy and colposcopy. CONCLUSION: Fluorescence spectroscopy performs better than colposcopy and other techniques in the diagnosis of SILs. Because it also permits real-time diagnosis and has the potential of being used by inexperienced health care personnel, this technology holds bright promise.

Colposcopy for the diagnosis of squamous intraepithelial lesions: a meta-analysis.

OBJECTIVE: To quantify by meta-analysis the performance of colposcopy to set a standard against which new technologies can be compared. DATA SOURCES: MEDLINE was searched for articles on colposcopy for diagnosis of squamous intraepithelial lesions (SIL). The search selected articles from 1960 to 1996 combining the key word "colposcopy" with key words "diagnosis," "positive predictive value," "negative predictive value," "likelihood ratio," and "receiver operating characteristic (ROC) curve." METHODS OF STUDY SELECTION: Articles were selected if the authors studied a population of patients with abnormal screening Papanicolaou smears and presented raw data showing for each cervical lesion type the number of patients judged positive and negative by colposcopic impression versus the standard of colposcopic biopsy results. Nine of 86 studies met these criteria. TABULATION, INTEGRATION, AND RESULTS: Biopsies had been categorized as normal, atypia, cervical intraepithelial neoplasia (CIN) I, CIN II, CIN III, carcinoma in situ, and invasive cancer; we recalculated performance measures using the Bethesda system. Overall sensitivity, specificity, likelihood ratios, ROC curves, and the corresponding areas under the curves were calculated. The average weighted sensitivity of diagnostic colposcopy for the threshold normal compared with all cervix abnormalities (atypia, low-grade SIL, high-grade SIL, cancer) was 96% and the average weighted specificity 48%. For the threshold normal cervix and low-grade SIL compared with high-grade SIL and cancer, average weighted sensitivity was 85% and average weighted specificity 69%. Likelihood ratios generated small but important changes in probability for distinguishing normal cervix and low-grade SIL from high-grade SIL and cancer. Areas under the ROC curve were 0.80 for the threshold normal cervix compared with all abnormalities and 0.82 for the threshold normal cervix and low-grade SIL compared with high-grade SIL and cancer. CONCLUSION: Colposcopy compares favorably with other medical diagnostic tests in terms of sensitivity, specificity, and area under the ROC curve. New diagnostic methods for the cervix can be compared with colposcopy using these quantified values.

Screening for squamous intraepithelial lesions with fluorescence spectroscopy.

OBJECTIVE: To evaluate the accuracy of fluorescence spectroscopy in screening for squamous intraepithelial lesions (SILs) and to compare its performance with that of Papanicolaou smear screening, colposcopy, cervicoscopy, cervicography, and human papillomavirus (HPV) testing. DATA SOURCES: Receiver operating characteristic (ROC) curve analysis was used to analyze performance by fluorescence spectroscopy (primary data) and other methods (secondary data). METHODS OF STUDY SELECTION: In our search, 275 articles were identified in MEDLINE (1966-1996). Articles were included if the investigators had studied a population in whom low disease prevalence was expected; used either Papanicolaou smear screening and colposcopy or colposcopically directed biopsy as a standard against which the screening technique was measured, and included enough data for recalculation of reported sensitivities and specificities. TABULATION, INTEGRATION, AND RESULTS: Receiver operating characteristic curves for fluorescence spectroscopy were calculated using a Bayesian algorithm, and ROC curves for the other screening methods were constructed using metaanalytic techniques. Areas under the ROC curves and Q points were calculated. Screening colposcopy had the highest area under the curve (0.95), followed by screening cervicography (0.90), HPV testing (0.88), cervicoscopy (0.85), fluorescence spectroscopy (0.76), and Papanicolaou smear screening (0.70). CONCLUSION: In terms of screening for SILs, fluorescence spectroscopy performed better than the standard technique, Papanicolaou smear screening, and less well than screening colposcopy, cervicography, HPV testing, and cervicoscopy. The promise of this research technique warrants further investigation.

Cost-effectiveness analysis of diagnosis and management of cervical squamous intraepithelial lesions.

OBJECTIVE: To compare five strategies for the diagnosis and treatment of cervical squamous intraepithelial lesions (SILs), including those that incorporate colposcopy and a new technology, fluorescence spectroscopy. METHODS: On the basis of a health care perspective, we performed a cost-effectiveness analysis using a decision-analytic model for the diagnosis and management of SILs. We compared the five strategies based on the expected costs and number of cases that were treated appropriately, missed, treated inappropriately, and appropriately not treated in a hypothetical cohort of 100 patients referred after an abnormal Papanicolaou smear. Data on prevalence and operating characteristics were derived from the medical literature. Costs were adjusted from hospital charge data. RESULTS: A see-and-treat strategy based on fluorescence spectroscopy was the least expensive but least effective strategy, costing $160,479 to detect 31.55 cases of cervical precancer accurately in 100 patients. The most expensive strategy was colposcopically directed biopsy, at $311,808 to find 45.78 cases; however, when both tests were used in a see-and-treat modality, slightly more cases were found (46.05) at a lower cost ($285,133). Other strategies were dominated in the base case. The incremental cost-effectiveness of the joint strategy compared with the spectroscopy-only strategy was $8596 per case of cervical precancer detected. Sensitivity analysis showed that the analysis was sensitive to the cost of the new technology of fluorescence spectroscopy. CONCLUSION: Fluorescence spectroscopy should be considered an important innovation in the diagnosis of SILs as demonstrated by its efficacy and economic advantages.

Near real time confocal microscopy of cultured amelanotic cells: sources of signal, contrast agents and limits of contrast.

The use of high resolution, in vivo confocal imaging for noninvasive assessment of tissue pathology may offer a clinically important adjunct to standard histopathological techniques. To augment the present understanding of both the capabilities and limitations of in vivo confocal imaging, we investigated cellular sources of image contrast in amelanotic tissues, how contrast can be enhanced with external agents and how contrast is degraded by the scattering of overlying cells. A high-resolution reflected light confocal microscope was constructed and used to obtain images of various types of unstained amelanotic cells in suspension in real time before and after the addition of contrast agents. Reflectance images were compared to phase contrast images and electron micrographs to identify morphology visible with real time reflected light confocal microscopy. Mechanisms which decrease image contrast, including interference effects and scattering in overlying layers of cells, were considered. In amelanotic epithelial cells, fluctuations in the nuclear index of refraction provide signal which can be imaged even under several overlying cell layers. Acetic acid is an external contrast agent which can enhance this nuclear backscattering. Image contrast is degraded by the presence of multiple scattering in overlying cell layers. The degradation of image contrast by cell scattering depends on the scattering phase function; in vitro models which use polystyrene microspheres to approximate tissue underestimate the actual degradation caused by cell scattering. The loss in contrast can be explained using a finite difference time domain model of cellular scattering. We conclude that near real time reflected light confocal microscopy can be used to study cell morphology in vivo. Contrast degradation due to overlying tissue is a concern and cannot adequately be modeled using conventional tissue phantoms; however, acetic acid may be used to substantially increase intrinsic contrast, allowing imaging at significant depths despite distortion from overlying layers. © 1998 Society of Photo-Optical Instrumentation Engineers.

Detection and diagnosis of oral neoplasia with an optical coherence microscope.

The use of high resolution, in vivo optical imaging may offer a clinically useful adjunct to standard histopathologic techniques. A pilot study was performed to investigate the diagnostic capabilities of optical coherence microscopy (OCM) to discriminate between normal and abnormal oral tissue. Our objective is to determine whether OCM, a technique combining the subcellular resolution of confocal microscopy with the coherence gating and heterodyne detection of optical coherence tomography, has the same ability as confocal microscopy to detect morphological changes present in precancers of the epithelium while providing superior penetration depths. We report our results using OCM to characterize the features of normal and neoplastic oral mucosa excised from 13 subjects. Specifically, we use optical coherence and confocal microscopic images obtained from human oral biopsy specimens at various depths from the mucosal surface to examine the optical properties that distinguish normal and neoplastic oral mucosa. An analysis of penetration depths achieved by the OCM and its associated confocal arm found that the OCM consistently imaged more deeply. Extraction of scattering coefficients from reflected nuclear intensity is successful in nonhyperkeratotic layers and shows differentiation between scattering properties of normal and dysplastic epithelium and invasive cancer.