We developed a novel deep-learning based algorithm for a mobile Detection of Oral Cancer (mDOC) platform that captures white light and auto-fluorescence images of the oral cavity. The algorithm first segments images and subsequently identifies suspicious lesions in need of further review by an expert clinician. Preliminary results show a dice score accuracy between ground truth annotated and the network produced segmentation to be higher than 0.9 for the network architectures we tested. This fully automated pipeline enables a data-driven approach with the potential to aid faster diagnosis in the clinic and earlier detection of oral lesions that can ultimately improve patient outcomes.
SignificanceDespite recent advances in multimodal optical imaging, oral imaging systems often do not provide real-time actionable guidance to the clinician who is making biopsy and treatment decisions.AimWe demonstrate a low-cost, portable active biopsy guidance system (ABGS) that uses multimodal optical imaging with deep learning to directly project cancer risk and biopsy guidance maps onto oral mucosa in real time.ApproachCancer risk maps are generated based on widefield autofluorescence images and projected onto the at-risk tissue using a digital light projector. Microendoscopy images are obtained from at-risk areas, and multimodal image data are used to calculate a biopsy guidance map, which is projected onto tissue.ResultsRepresentative patient examples highlight clinically actionable visualizations provided in real time during an imaging procedure. Results show multimodal imaging with cancer risk and biopsy guidance map projection offers a versatile, quantitative, and precise tool to guide biopsy site selection and improve early detection of oral cancers.ConclusionsThe ABGS provides direct visible guidance to identify early lesions and locate appropriate sites to biopsy within those lesions. This represents an opportunity to translate multimodal imaging into real-time clinically actionable visualizations to help improve patient outcomes.
Purpose:In vivo optical imaging technologies like high-resolution microendoscopy (HRME) can image nuclei of the oral epithelium. In principle, automated algorithms can then calculate nuclear features to distinguish neoplastic from benign tissue. However, images frequently contain regions without visible nuclei, due to biological and technical factors, decreasing the data available to and accuracy of image analysis algorithms.
Approach: We developed the nuclear density-confidence interval (ND-CI) algorithm to determine if an HRME image contains sufficient nuclei for classification, or if a better image is required. The algorithm uses a convolutional neural network to exclude image regions without visible nuclei. Then the remaining regions are used to estimate a confidence interval (CI) for the number of abnormal nuclei per mm2, a feature used by a previously developed algorithm (called the ND algorithm), to classify images as benign or neoplastic. The range of the CI determines whether the ND-CI algorithm can classify an image with confidence, and if so, the predicted category. The ND and ND-CI algorithm were compared by calculating their positive predictive value (PPV) and negative predictive value (NPV) on 82 oral biopsies with histopathologically confirmed diagnoses.
Results: After excluding the images that could not be classified with confidence, the ND-CI algorithm had higher PPV (65% versus 59%) and NPV (78% versus 75%) than the ND algorithm.
Conclusions: The ND-CI algorithm could improve the real-time classification of HRME images of the oral epithelium by informing the user if an improved image is required for diagnosis.
Oral premalignant lesions (OPLs), such as leukoplakia, are at risk of malignant transformation to oral cancer. Clinicians can elect to biopsy OPLs and assess them for dysplasia, a marker of increased risk. However, it is challenging to decide which OPLs need a biopsy and to select a biopsy site. We developed a multimodal optical imaging system (MMIS) that fully integrates the acquisition, display, and analysis of macroscopic white-light (WL), autofluorescence (AF), and high-resolution microendoscopy (HRME) images to noninvasively evaluate OPLs. WL and AF images identify suspicious regions with high sensitivity, which are explored at higher resolution with the HRME to improve specificity. Key features include a heat map that delineates suspicious regions according to AF images, and real-time image analysis algorithms that predict pathologic diagnosis at imaged sites. Representative examples from ongoing studies of the MMIS demonstrate its ability to identify high-grade dysplasia in OPLs that are not clinically suspicious, and to avoid unnecessary biopsies of benign OPLs that are clinically suspicious. The MMIS successfully integrates optical imaging approaches (WL, AF, and HRME) at multiple scales for the noninvasive evaluation of OPLs.
We developed an automated frame selection algorithm for high-resolution microendoscopy video sequences. The algorithm rapidly selects a representative frame with minimal motion artifact from a short video sequence, enabling fully automated image analysis at the point-of-care. The algorithm was evaluated by quantitative comparison of diagnostically relevant image features and diagnostic classification results obtained using automated frame selection versus manual frame selection. A data set consisting of video sequences collected in vivo from 100 oral sites and 167 esophageal sites was used in the analysis. The area under the receiver operating characteristic curve was 0.78 (automated selection) versus 0.82 (manual selection) for oral sites, and 0.93 (automated selection) versus 0.92 (manual selection) for esophageal sites. The implementation of fully automated high-resolution microendoscopy at the point-of-care has the potential to reduce the number of biopsies needed for accurate diagnosis of precancer and cancer in low-resource settings where there may be limited infrastructure and personnel for standard histologic analysis.
We developed an automated frame selection algorithm for high resolution microendoscope images. The algorithm
rapidly selects a representative frame with minimal motion artifact from a short video sequence, enabling fully
automated image analysis at the point-of-care. The performance of the algorithm was evaluated by comparing
automatically selected frames to manually selected frames using quantitative image parameters. The implementation of fully automated high-resolution microendoscopy at the point-of-care has the potential to reduce the
number of biopsies needed for accurate diagnosis of precancer and cancer in low-resource settings, where there
may be limited infrastructure and personnel for standard histologic analysis.
In this longitudinal study, a mouse model of 4-nitroquinoline 1-oxide chemically induced tongue carcinogenesis was used to assess the ability of optical imaging with exogenous and endogenous contrast to detect neoplastic lesions in a heterogeneous mucosal surface. Widefield autofluorescence and fluorescence images of intact 2-NBDG-stained and proflavine-stained tissues were acquired at multiple time points in the carcinogenesis process. Confocal fluorescence images of transverse fresh tissue slices from the same specimens were acquired to investigate how changes in tissue microarchitecture affect widefield fluorescence images of intact tissue. Widefield images were analyzed to develop and evaluate an algorithm to delineate areas of dysplasia and cancer. A classification algorithm for the presence of neoplasia based on the mean fluorescence intensity of 2-NBDG staining and the standard deviation of the fluorescence intensity of proflavine staining was found to separate moderate dysplasia, severe dysplasia, and cancer from non-neoplastic regions of interest with 91% sensitivity and specificity. Results suggest this combination of noninvasive optical imaging modalities can be used in vivo to discriminate non-neoplastic from neoplastic tissue in this model with the potential to translate this technology to the clinic.
Dysplastic and cancerous alterations in oral tissue can be detected noninvasively in vivo using optical techniques
including autofluorescence imaging, high-resolution imaging, and spectroscopy. Interim results are presented from a
longitudinal study in which optical imaging and spectroscopy were used to evaluate the progression of lesions over time
in patients at high risk for development of oral cancer. Over 100 patients with oral potentially malignant disorders have
been enrolled in the study to date. Areas of concern in the oral cavity are measured using widefield autofluorescence
imaging and depth-sensitive optical spectroscopy during successive clinical visits. Autofluorescence intensity patterns
and autofluorescence spectra are tracked over time and correlated with clinical observations. Patients whose lesions
progress and who undergo surgery are also measured in the operating room immediately prior to surgery using
autofluorescence imaging and spectroscopy, with the addition of intraoperative high-resolution imaging to characterize
nuclear size, nuclear crowding, and tissue architecture at selected sites. Optical measurements are compared to
histopathology results from biopsies and surgical specimens collected from the measured sites. Autofluorescence
imaging and spectroscopy measurements are continued during post-surgery followup visits. We examined correlations
between clinical impression and optical classification over time with an average followup period of 4 months. The data
collected to date suggest that multimodal optical techniques may aid in noninvasive monitoring of the progression of oral
premalignant lesions, biopsy site selection, and accurate delineation of lesion extent during surgery.
Wide-filed autofluorescence examination is currently considered as a standard of care for screening and
diagnostic evaluation of early neoplastic changes of the skin, cervix, lung, bladder, gastrointestinal tract and
oral cavity. Naturally occurring fluorophores within the tissue absorb UV and visible light and can re-emit
some of this light at longer wavelengths in the form of fluorescence. This non-invasive tissue
autofluorescence imaging is used in optical diagnostics, especially in the early detection of cancer. Usually,
malignant transformation is associated with thickening of the epithelium, enhanced cellular density due to
increased nuclear cytoplasmic ratio which may attenuate the excitation leading to a decrease in collagen
autofluorescence. Hence, dysplastic and cancerous tissues often exhibit decreased blue-green
autofluorescence and appear darker compared to uninvolved mucosa. Currently, there are three commercially
available devices to examine tissue autofluorescence in the oral cavity. In this study we used the oral cancer
screening device IdentafiTM 3000 to examine the tissue reflectance and autofluorescence of PML and
confounding lesions of the oral cavity. Wide-field autofluorescence imaging enables rapid inspection of large
mucosal surfaces, to aid in recognition of suspicious lesions and may also help in discriminate the PML (class
1) from some of the confounding lesions (class II). However, the presence of inflammation or pigments is
also associated with loss of stromal autofluorescence, and may give rise to false-positive results with widefield
fluorescence imaging. Clinicians who use these autofluorescence based oral cancer screening devices
should be aware about the benign oral mucosal lesions that may give false positivity so that unnecessary
patient's anxiety and the need for scalpel biopsy can be eliminated.
Optical techniques including widefield autofluorescence and reflectance imaging, depth-sensitive optical spectroscopy,
and high-resolution imaging can be used to noninvasively detect dysplastic and cancerous alterations in oral tissue. The
diagnostic performance of depth-sensitive optical spectroscopy with respect to histopathology is examined. A compact,
portable spectroscopy device for clinical use is described. Practical considerations for the comparison of optical
measurements to histopathologic diagnoses are outlined. Important considerations for comparison to histopathology
include the physical correspondence of the measured region to the biopsy or specimen; data collection and processing
procedures; and data analysis procedures. Multimodal combinations of widefield imaging, point spectroscopy, and highresolution
imaging may enhance the ability of clinicians to accurately assess the margins of neoplastic oral lesions in
vivo.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.