Multicenter-derived machine learning model for individualized rupture-risk prediction in intracranial aneurysms: study protocol and preliminary validation.

Background: Intracranial aneurysms (IAs) account for approximately 85% of all spontaneous subarachnoid hemorrhage (SAH) cases and are associated with poor outcomes, including death, in up to 25-50% of patients. Rupture risk is influenced by a variety of factors and exhibits substantial inter-patient variability. In this study, we aimed to develop a prediction model for IAs rupture risk by integrating clinical, computed tomographic angiography (CTA) morphological, and radiomics features, and to evaluate the consistency of predictive performance across multiple machine learning (ML) algorithms. Methods: In this retrospective multicenter study, 756 consecutive patients with 877 IAs were enrolled from three centers between January 2020 and November 2024. Clinical, morphological, and radiomics features were extracted. Independent risk factors for IAs rupture were identified, and six predictive models were constructed using logistic regression (LR). A nomogram incorporating significant clinical, morphological, and radiomics score (Radscore) predictors was developed. To evaluate model robustness, seven additional ML algorithms were applied to the clinical-morphological-radiomics model (Model_C&M&R) features for comparative analysis. Results: Age, admission blood glucose, aneurysm diameter, location, parent artery average diameter, and ten radiomics features were identified as independent predictors of IAs rupture (P<0.05). Model_C&M&R achieved the highest predictive performance in both the training set [area under the curve (AUC) =0.887] and the internal validation set (AUC =0.910). It also demonstrated consistent generalizability in external validation sets I (AUC =0.773) and II (AUC =0.735). Among the seven ML algorithms applied to the Model_C&M&R feature set, five models exhibited strong performance in the internal validation set. Conclusions: The integrated Model_C&M&R provides optimal predictive accuracy for IAs rupture, and the prediction models developed utilizing various ML algorithms demonstrate consistently excellent performance.