[Development and validation of an integrated intratumoral-peritumoral habitat radiomics model on multiparametric MRI...

Objective: To develop and validate an MRI-based habitat radiomics model (EPM) integrating intra-and peritumoral features, and to build a combined clinical habitat model (CHPM) for predicting short-term biochemical recurrence (BCR) after radical prostatectomy (RP). Methods: A total of 429 patients who underwent RP were retrospectively enrolled from 2 institutions (Center 1: the First Affiliated Hospital of Soochow University, January 2015 to December 2018, n=335; Center 2: Zhangjiagang Hospital Affiliated with Soochow University, January 2015 to December 2023, n=94). Center 1 was split in a 7∶3 ratio into training (n=234) and internal test (n=101) sets by stratified random sampling based on BCR status; Center 2 served as the external test set (n=94). Follow-up data were retrospectively collected from inpatient and outpatient medical records, with follow-up censored at BCR or the last visit, and only patients with≥1 year of postoperative follow-up were included. Using multi-parametric magnetic resonance imaging (mpMRI), GMM clustering divided intra-and peritumoral (3 mm) regions into K habitat subregions for feature extraction. intratumor habitat (ITH), peritumoral habitat (PTH), and the combined EPM model were built. CHPM integrated ITH/PTH risk scores with independent clinical predictors selected by Cox regression. Model risk stratification was assessed by Kaplan-Meier curves. CHPM performance was evaluated using C-index, calibration curves, decision curve analysis (DCA), integrated discrimination improvement (IDI), and net reclassification improvement (NRI), and compared with CAPRA-S and EPM, with between-model C-index differences compared using Bootstrap resampling (1 000 resamples). Results: All 3 models stratified high-and low-risk patients in both test sets (all P<0.05). CHPM showed a higher C-index than CAPRA-S (0.782 vs 0.672, P=0.036; 0.780 vs 0.689, P=0.042). C-index differences between EPM and CAPRA-S, and between EPM and CHPM, were not significant (all P>0.05). EPM and CHPM both outperformed CAPRA-S in calibration, net benefit, IDI, and NRI. Conclusions: EPM and CHPM had comparable C-index values for predicting short-term BCR after RP, and both outperformed CAPRA-S in calibration, net benefit, IDI, and NRI. CHPM showed the best overall performance, suggesting its potential as a noninvasive preoperative risk assessment tool.