A machine learning study of 351 breast cancer patients found that MRI radiomics features drawn from both the tumor and peritumoral region improved early prediction of pathological complete response to neoadjuvant chemotherapy. The results suggest that peritumoral imaging carries additional predictive signal beyond tumor-only models. For clinical development, the key operational takeaway is feature selection: models may need broader spatial coverage and consistent radiomics pipelines to generalize across centers. The work supports continued investment in radiomics-based companion diagnostics, particularly in settings where treatment selection depends on response probability before surgery.