Segmentation of oral optical coherence tomography with deep learning

Diagnosis of oral cancer involves collecting multiple biopsies to increase the likelihood of sampling the most pathologic site within a lesion. Optical coherence tomography (OCT) allows for examination of subsurface morphology, and has shown potential in biopsy guidance. OCT captures changes in tissue stratification related to depth, topology, and presence of the stromal-epithelial boundary which are structural biomarkers for pre-invasive and invasive oral cancer. This thesis presents a four-part neural network pipeline to simplify OCT interpretation by providing en face maps of epithelial depth and stratification. U-nets models are employed to segment the stromal-epithelial boundary, and supporting convolutional neural networks are used for identification of the imaging field and artifacts. Training was conducted on a variety of non-cancerous and cancerous pathologies across the oral cavity to promote generalizability. Predictions demonstrate as-good-as or better agreement than inter-rater agreement, suggesting strong predictive power.