A high-throughput tumor location system with deep learning for colorectal cancer histopathology image

Colorectal cancer is one of the major causes of morbidity and mortality worldwide, however, when discovered at an early stage, it is highly treatable. As the number of specimens increases every year, there has been a boost in the diagnostic workload on pathologists in recent years. In parallel to the development of digital pathology, deep learning has demonstrated its strong capability in feature extraction and interpretation in a variety of medical applications. In this paper, we propose a high-throughput whole-slide image (WSI) analysis system to localize tumor regions accurately with a patch-based convolutional neural network (CNN). We employ Monte Carlo adaptive sampling for a fast detection of tumors at slide level and a conditional random field (CRF) model to integrate spatial correlation for better classification accuracy. We use three datasets of colorectal cancer from The Cancer Genome Atlas (TCGA) for performance evaluation. Compared with the regular WSI analysis, the experimental benchmark shows an obvious decrease in processing time while a noticeable improvement in classification accuracy.