Pretreatment ET radiomics based on primary-tumor HR status for assessing the risk of local progression after GKRS in BCBM

Background To determine whether hormone receptor (HR) status modifies the association between pretreatment enhancing-tumor (ET) MRI radiomics and 12-month lesion-level local progression after Gamma Knife radiosurgery (GKRS) for breast cancer brain metastases (BCBM). Methods We screened 61 patients and locked a cohort of 49 patients with 105 lesions, including 33 progressed and 72 non-progressed lesions. Enhancing tumor (ET) was manually delineated on pretreatment contrast-enhanced T1-weighted MRI. The prespecified model combined ET Rad-score, hormone receptor (HR) status and the ET Rad-score × HR interaction. Performance was evaluated using patient-grouped repeated nested cross-validation. Results The locked model achieved lesion-level ROC-AUC 0.750, average precision (AP) 0.567 and Brier score 0.204. Patient-balanced evaluation yielded ROC-AUC 0.786, AP 0.621 and Brier score 0.194. The ET Rad-score × HR interaction was positive in 98 of 100 outer folds. HR-positive lesions had higher ROC-AUC and AP and a lower Brier score (ROC-AUC 0.827, AP 0.664, Brier score 0.174). Patient-level maximum predicted risk achieved ROC-AUC 0.803 and AP 0.694. Descriptive median-risk stratification separated lesions with progression rates of 50.9% versus 11.5%. Conclusion Pretreatment ET radiomics showed an HR-dependent association with 12-month lesion-level local progression after GKRS in BCBM, supporting further investigation of HR-informed radiomic risk stratification. Public datasets provided construct support; endpoint-matched external validation is needed before clinical application.