Deep learning classification of ex vivo human colon tissues using spectroscopic OCT

Abstract

Screening programs for colorectal cancer (CRC) have had a profound impact on the morbidity and mortality of this disease by detecting and removing early cancers and precancerous adenomas with colonoscopy. However, CRC continues to be the third leading cause of cancer-related mortality in both men and woman, partly because of limitations in colonoscopy-based screening. Thus, novel strategies to improve the efficiency and effectiveness of screening colonoscopy are urgently needed. Here, we propose to address this need using an optical biopsy technique based on spectroscopic optical coherence tomography (OCT). The depth resolved images obtained with OCT are analyzed as a function of wavelength to measure optical tissue properties. The optical properties can be used as input to machine learning algorithms as a means to classify adenomatous tissue in the colon. In this study, biopsied tissue samples from the colonic epithelium are analyzed ex vivo using spectroscopic OCT and tissue classifications are generated using a novel deep learning architecture, informed by machine learning methods including LSTM and KNN. The overall classification accuracy obtained was 88.9%, 76.0% and 97.9% in discriminating tissue type for these methods. Further, we apply an approach using false coloring of en face OCT images based on SOCT parameters and deep learning predictions to enable visual identification of tissue type. This study advances the spectroscopic OCT towards clinical utility for analyzing colonic epithelium for signs of adenoma.