Dynamic multimodal radiomic model for survival prediction in cervical cancer: a multi-cohort study

Accurate survival prediction is critical for guiding individualized radiotherapy in locally advanced cervical cancer (LACC), yet most existing models rely solely on pre-treatment imaging and overlook tumor changes during therapy. In this study, we developed and validated a multivariable Cox model to predict disease-free survival (DFS) in 971 LACC patients undergoing concurrent chemoradiotherapy. The model incorporated clinical variables, radiomics features from pre- and post-radiotherapy CT scans, and their changes during treatment, thereby quantifying treatment-associated imaging changes. It achieved C-indices of 0.713 and 0.693 in internal and external validation cohorts, with AUCs up to 0.859, showing improved overall performance compared with single-timepoint imaging models in most evaluation settings. By integrating dynamic imaging and clinical data, the model improved prognostic accuracy, enhanced risk stratification, and showed strong potential for supporting personalized radiotherapy planning.