Patch-Based Coupled Attention Network to Predict MSI Status in Colon Cancer

Identifying and diagnosing key markers from WSI images is the key to accurate diagnosis and treatment of colon cancer (CC). However, how to extract marker-related features from huge scale WSI images is a challenge faced by AI models. In this study, we propose a patch-based coupled attention neural network (CovAttnNet) designed to predict Microsatellite Instability (MSI) status from WSI images. CovAttnNet consists of a transformer based backbone network and a neural network based on convolutional operations. A global and local feature attention module based on patch coupling is proposed to extract and fuse key features. We validated the performance of the model on the publicly available dataset TCGA-COAD, and the experimental results demonstrated the superior ability of CovAttnNet in predicting the status of colon cancer MSI status. This study provides a new method for deep learning in marker prediction research.