CT-based radiomics for interpretable outcome prediction after aneurysmal subarachnoid hemorrhage: a multicenter cohort study

Background: Prognostic assessment after aneurysmal subarachnoid haemorrhage (aSAH) mainly relies on clinical variables and CT findings, but current models often lack biological interpretability. We integrated CT radiomics with cerebrospinal fluid proteomics to develop an interpretable prognostic model and explore mechanisms underlying unfavourable outcomes. Methods: This multicentre cohort from the PROSAH-MPC registry included 879 patients across four hospitals. Patients from the primary centre were split into training (n=637) and internal validation (n=160) cohorts, while patients from three other centres formed an external validation cohort (n=82). Admission CT haemorrhagic lesions were automatically segmented using the validated BrainHemoAI platform, and radiomic features were extracted with PyRadiomics. Three XGBoost-based models were developed and interpreted using SHapley Additive exPlanations (SHAP). Risk scores from the best-performing model were applied to an independent prospective cerebrospinal fluid proteomic cohort to define high- and low-risk groups. GO/KEGG enrichment, gene set enrichment analysis, and protein–protein interaction analyses were performed. Results: The combined radiomics-clinical model achieved the best performance, with AUCs of 0.892 and 0.820 in the internal and external validation cohorts, respectively. SHAP identified radiomic heterogeneity and key clinical variables as major predictors. Proteomic analyses showed that high-risk patients were characterized by neurovascular barrier injury, thromboinflammatory activation, apoptosis, stress-related metabolic adaptation, AKT1 downregulation, and RHOA upregulation, whereas low-risk patients showed enrichment of extracellular matrix remodelling, collagen biosynthesis, and wound healing pathways. Conclusions: This multi-omics prognostic model enables accurate outcome prediction and provides biological insight into unfavourable recovery after aSAH, supporting refined risk stratification and potential targeted therapeutic development.