KEY POINTS
- This multicenter retrospective study included 875 patients with brain metastases treated with Gamma Knife radiosurgery across three centers: 354 training, 146 internal validation, 174 external validation 1, and 201 external validation 2.
- The proposed model integrated baseline contrast-enhanced T1-weighted magnetic resonance imaging, grid-based magnetic resonance imaging patches, radiomics, and consistently available clinical variables to generate a patient-level overall survival risk score.
- One-year area under the curve values were 0.870 in training, 0.755 in internal validation, 0.740 in external validation 1, and 0.788 in external validation 2.
- Concordance indices were more modest: 0.766 in training, 0.655 in internal validation, 0.653 in external validation 1, and 0.649 in external validation 2.
- The model-derived risk score remained independently associated with overall survival after clinical adjustment across all cohorts, but external calibration varied and prospective validation is still required.
CLINICAL TAKEAWAY
This model may add prognostic information after Gamma Knife radiosurgery for brain metastases, especially for post-treatment risk stratification and follow-up planning. The key limitation is that performance was only moderate in external validation, and the clinical branch used a limited set of variables because richer data were not consistently available. This is technically relevant progress, not a deployable clinical decision tool yet.