Predicting Disease-Free Survival in Rectal Cancer Patients via Radiomics Signatures from Multi-Modality MRI Images: a multi-center study

Motivation: Accurate prediction of disease-free survival (DFS) of rectal cancer patients has substantial influences on treatment planning. Goal(s): To explore the value of ADC and T2W images in predicting DFS for rectal cancer patients. Approach: Radiomics models using features extracted from ADC and T2W were built and evaluated. Results: Models built with ADC features and clinical variables achieved C-Index values of 0.675 and 0.768 over the internal and external test cohorts, respectively. Models built with T2W features and clinical variables achieved C-Index values of 0.724 and 0.747 over the internal and external test cohorts, respectively. Impact: Both the shape features extracted from ADC maps and the first-order features extracted from T2W images demonstrated strong predictive power for DFS estimation, implying the potential of combination anatomical images with diffusion models to predict DFS in rectal cancer patients.