Non-invasive CT-based model predicts tumor immune environment in CRC

A deep learning radiomics model utilizing preoperative CT scans shows promise for non-invasively predicting the tumor immune microenvironment (TIME) in colorectal cancer patients. The Densenet-169 model achieved an AUC of 0.892 for tumor-stroma ratio (TSR), while the Densenet-121 model reached AUCs of 0.851 for tumor-infiltrating lymphocytes and 0.852 for immune score. These models could facilitate personalized immunotherapy strategies, significantly enhancing CRC management.