Classifying tumor heterogeneity of human esophageal cancer biopsies by dynamic contrast OCT with deep learning

Abstract

Tumor heterogeneity is one of the greatest obstacles standing in the way of successful cancer therapy. Cancer in a single patient is not a single disease, but is a host of related diseases, all of which need to respond to a single treatment regimen. We have completed the first human clinical trial in esophageal cancer using dynamic-contrast OCT (DC-OCT) based on full-frame digital holography to assess the spatial heterogeneity of biopsy response to platinum-based chemotherapy. A deep twin neural network successfully identified biopsy sub-phenotypes in the dynamic tissue response that enabled accurate prediction of patient treatment success.