Prognostic Value of Intratumoral Habitat and Peritumoral CT Radiomics After Pancreatic Ductal Adenocarcinoma Resection: A Retrospective Cohort Study

Background and Objectives: Pancreatic ductal adenocarcinoma (PDAC) is spatially heterogeneous and may extend beyond the tumour margin visible on CT. We compared intratumoral habitat radiomics with whole-tumour and 1-, 3-, and 5-mm peritumoral radiomics for overall survival after resection. Materials and Methods: This single-centre retrospective study included 315 patients assigned to training (n = 221) and internal test (n = 94) cohorts. Feature selection, model fitting, and cutoff definition were confined to training data. Clinicopathological, imaging, and integrated Cox models were evaluated with C-indices, censoring-weighted time-dependent AUCs, Kaplan–Meier analysis, and calibration. Additional analyses examined geometric partitions, reader repeatability, and scanner subgroups. Results: Habitat C-indices were 0.781 (95% CI 0.744–0.812) in training and 0.744 (0.700–0.784) in testing, compared with 0.647 and 0.619 for whole-tumour radiomics. Test AUCs at 1, 2, and 3 years were 0.828, 0.854, and 0.867 for habitat and 0.813, 0.852, and 0.890 for the integrated model; all paired difference intervals included zero. Habitat C-indices were 0.760 and 0.757 in the two scanner subgroups. Time-varying effects did not consistently improve test calibration. Conclusions: Habitat radiomics showed the strongest discrimination among the original imaging models, without evidence of improvement from integration. The results support further evaluation of postoperative risk stratification; external validation is needed before clinical application.