Peritumoral habitat radiomics predicts axillary lymph node metastasis in breast cancer.

Accurate preoperative prediction of axillary lymph node metastasis (ALNM) is critical for personalized management of breast cancer. Here, we developed a habitat-aware, multiscale radiomics framework based on dynamic contrast-enhanced MRI to characterize intratumoral heterogeneity and peritumoral microenvironmental changes. Intratumoral habitats were delineated using superpixel segmentation and clustering, and radiomic features from habitats, concentric peritumoral regions, and clinical variables were integrated to construct predictive models. A fusion model combining the habitat signature, 1-mm peritumoral signature, and clinical factors achieved the best performance, consistently outperforming conventional radiomics approaches, with AUCs of 0.940, 0.905, 0.857, and 0.884 in the training, validation, and two independent test cohorts, respectively. These findings demonstrate that habitat-guided multiscale radiomics enables accurate, interpretable, and noninvasive assessment of nodal metastasis risk and may support individualized surgical and therapeutic decision-making in breast cancer.