CT-derived extracellular volume fraction and AI-assisted body composition for predicting neoadjuvant therapy response and prognosis in colorectal cancer.

OBJECTIVES: To develop a combined model integrating clinical parameters, quantitative parameters from contrast-enhanced computed tomography (CECT), and Artificial Intelligence-assisted body composition metrics for predicting treatment response and prognosis in colorectal cancer (CRC) patients receiving neoadjuvant therapy (NAT). MATERIALS AND METHODS: This retrospective study included CRC patients received NAT and surgery between January 2016 and June 2025. Quantitative CECT parameters (extracellular volume fraction [ECV] and arterial enhancement fraction), and body composition parameters (skeletal muscle index [SMI], skeletal muscle attenuation, and visceral-to-subcutaneous adipose tissue ratio) were measured before and after NAT, and their changes were calculated. Logistic regression analysis was used to identify independent predictors of treatment response. A multivariable logistic regression model was constructed, and evaluated using the area under the receiver operating characteristic curve (AUC). Patients were stratified into high- and low-risk groups based on risk score produced from the model. Cox regression analysis was used to assess association between risk score and prognosis. RESULTS: A total of 163 patients (age, 58.50 ± 10.55 years) were included. A combined model incorporating post-NAT carcinoembryonic antigen, ECV and SMI achieved an AUC of 0.836 (95% CI: 0.754-0.915) and 0.784 (95% CI: 0.583-0.947) in the training and validation cohorts. The risk score was significantly associated with prognosis. High-risk patients had shorter overall survival (Hazard ratio [HR]: 3.745, 95% CI: 1.014-13.836, P = 0.048) and disease-free survival (HR: 2.801, 95% CI: 1.050-7.475, P = 0.040). CONCLUSION: The combined model may help predict treatment response and survival outcomes in patients with CRC undergoing NAT.