MRI-Radiomics Deep Learning Model for Post-Gamma Knife Survival in Brain Metastases: Multicenter Retrospective Study

In a multicenter retrospective study of 875 brain metastasis patients, a multimodal MRI-radiomics deep learning model predicted 1-year survival with AUCs up to 0.870 (training) and 0.740-0.788 (external validation), but C-indices were moderate (0.649-0.655), indicating limited d…

• Multicenter retrospective design (875 patients from 3 centers) with internal and two external validation cohorts.
• Model-derived risk score independently associated with overall survival across all cohorts on multivariable Cox regression.
• Retrospective design; requires prospective validation, recalibration, and integration of comprehensive clinical variables before clinical use.

RadPigeon summaries are original and for information only. They are not clinical advice.