Abstract / Summary
To develop and validate an MRI-based radiomics-habitat model integrating intratumoral and peritumoral features to noninvasively predict MammaPrint risk. This retrospective dual-center study included 156 patients who underwent pretreatment dynamic contrast-enhanced MRI and MammaPrint testing. Radiomic features were extracted from intratumoral and peritumoral regions, and habitat analysis was used to characterize intratumoral heterogeneity. Multiple machine learning models were developed and evaluated using receiver operating characteristic analysis, calibration analysis, and decision curve analysis. Independent clinical predictors were mass margin (odds ratio [OR], 0.854; P < 0.05) and tumor size (OR, 1.014; P < 0.05). The ExtraTrees classifier performed best. The combined model achieved areas under the receiver operating characteristic curve (AUCs) of 0.931 in the training set, 0.896 in the internal validation set, and 0.877 in the external test set. Calibration was good (Hosmer–Lemeshow P = 0.267, 0.491, and 0.892), and decision curve analysis confirmed clinical utility. This MRI-based radiomics-habitat model shows promise for noninvasive prediction of MammaPrint risk in hormone receptor-positive, human epidermal growth factor receptor 2-negative (HR+/HER2−) breast cancer, offering a potential imaging surrogate for genomic risk stratification.