CT-based radiomics for interpretable outcome prediction after aneurysmal subarachnoid hemorrhage: a multicenter cohort study.
Authors
Affiliations (11)
Affiliations (11)
- Department of Neurosurgery, The Second Affiliated Hospital, Jiangxi Medical School, Nanchang University, Nanchang, Jiangxi, 330006, China.
- Jiangxi Engineering Research Center for Intelligent Translational Medicine of Cerebrovascular Diseases, Nanchang, Jiangxi, 330006, China.
- Jiangxi Health Commission Key Laboratory of Neurological Medicine, Nanchang, Jiangxi, 330006, China.
- Institute of Neuroscience, Nanchang University, Nanchang, Jiangxi, 330006, China.
- Department of Neurosurgery, The 908th Hospital of the Chinese PLA Joint Logistic Support Force, Nanchang, Jiangxi, 330002, China.
- Department of Neurosurgery, Nanchang County People's Hospital, Nanchang, Jiangxi, 330200, China.
- Department of Neurosurgery, The People's Hospital of Yichun City, Yichun, Jiangxi, 336000, China.
- Department of Neurosurgery, The Second Affiliated Hospital, Jiangxi Medical School, Nanchang University, Nanchang, Jiangxi, 330006, China. [email protected].
- Jiangxi Engineering Research Center for Intelligent Translational Medicine of Cerebrovascular Diseases, Nanchang, Jiangxi, 330006, China. [email protected].
- Jiangxi Health Commission Key Laboratory of Neurological Medicine, Nanchang, Jiangxi, 330006, China. [email protected].
- Institute of Neuroscience, Nanchang University, Nanchang, Jiangxi, 330006, China. [email protected].
Abstract
Prognostic assessment after aneurysmal subarachnoid hemorrhage (aSAH) mainly relies on clinical variables and computed tomography (CT) findings, but current models often lack biological interpretability. We developed and validated a clinical-radiomics prognostic model and subsequently performed exploratory cerebrospinal fluid (CSF) proteomic characterization of model-defined risk phenotypes. This multicenter cohort from the PROSAH-MPC registry included 879 patients across four hospitals. Patients from the primary center were split into training (n = 637) and internal validation (n = 160) cohorts, while patients from three other centers formed an external validation cohort (n = 82). Admission CT hemorrhagic lesions were automatically segmented using the validated BrainHemoAI platform, and radiomic features were extracted with PyRadiomics. Clinical, radiomics, and combined clinical-radiomics XGBoost models were developed and interpreted using SHapley Additive exPlanations (SHAP). Risk scores from the clinical-radiomics model were applied to an independent prospective enrichment, gene setCSF proteomic cohort to define combined model-based high- and low-risk groups. Gene Ontology (GO)/ Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment, gene set enrichment analysis (GSEA), and protein-protein interaction (PPI) analyses were performed. The combined model yielded AUCs of 0.892 and 0.820 in the internal test and external validation cohorts, respectively. Risk reclassification analyses suggested that radiomic features may provide complementary prognostic information. SHAP identified radiomic heterogeneity and key clinical variables as major predictors. Meanwhile, exploratory CSF proteomic analysis identified candidate differentially expressed proteins and associated pathway signals that differed between groups defined by high and low combined clinical-radiomics risk scores. This clinical-radiomics prognostic model enables accurate outcome prediction and provides exploratory biological insight into unfavorable recovery after aSAH, supporting refined risk stratification and potential targeted therapeutic development. PROSAH-MPC. NCT05738083. Registered 16 November 2022 - Retrospectively registered, https://clinicaltrials.gov/study/NCT05738083 .