CT4CMS: Preoperative Computed Tomography-Based Consensus Molecular Subtyping Prediction in Colorectal Cancer Using Interpretable Deep Learning

CT4CMS is a noninvasive deep learning framework that predicts consensus molecular subtypes of colorectal cancer directly from preoperative computed tomography scans. The study included 2,444 patients across a discovery cohort and internal and external validation cohorts. A self-supervised 3D representation network and an attention-based multi-instance learning classifier achieved a cross-validation AUC of 0.867. CT-predicted CMS4 was associated with poorer disease-free survival and greater benefit from adjuvant chemotherapy, supporting interpretable imaging-based molecular stratification before surgery.