The Accuracy in Rupture Risk Prediction of Intracranial Aneurysms by Artificial Intelligence Algorithms Using Imaging
Ruixuan Zhang1,2, Ruibo Liu1, He Ma1
1The College of Medicine and Biological Information Engineering, Northeastern University, Shenyang, China.
Abstract:
Ruptured intracranial aneurysms (IAs) are the leading cause of aSAH. There are limitations in combining traditional imaging methods (CTA and DSA) and clinical scores (PHASES) to predict IAs rupture risk, whereas artificial intelligence (AI) algorithms show potential. This meta-analysis evaluated AI algorithm performance for predicting IAs rupture risk based on CTA and DSA. As of February 2025, we searched Web of Science, PubMed, Scopus, and Embase, extracting TP, FP, FN, and TN from included studies. The combined sensitivity, specificity, and AUC were synthesised with a bivariate random-effects model. Subgroup analyses were performed. PROSPERO: CRD420251008866. Twenty studies (13,232 patients, 14,344 IAs) reported pooled sensitivity 0.84 (95% CI: 0.80-0.87), specificity 0.82 (95% CI: 0.78-0.86), and AUC 0.90 (95% CI: 0.87-0.92) with substantial heterogeneity. Subgroup analyses showed DOR in the DSA versus CTA groups (DSA 23.55, CTA 22.21) with persistent heterogeneity. The clinical-morphological-radiomics group had DOR 18.76 without heterogeneity. By publication year, 2021 group had a lower DOR (12.99) versus 2022 (23.03) versus 2023 (26.98), with low heterogeneity. AI algorithms predicting IAs rupture risk based on CTA and DSA demonstrate high diagnostic accuracy and have potential to advance the field.
