Related Experiment Video
Updated: Oct 29, 2026
Minimally Invasive Thumb-sized Pterional Craniotomy for Surgical Clip Ligation of Unruptured Anterior Circulation Aneurysms
Published on: August 11, 2015
Artificial Intelligence for Rupture Risk Stratification of Small Unruptured Intracranial Aneurysms (≤ 7 mm): A
Neda Kamandi1, Saeed Mohammadzadeh2, Moein Mirzai2
1Sussex Partnership NHS Foundation Trust, Brighton, East Sussex, UK.
Abstract:
Small unruptured intracranial aneurysms (UIAs; ≤ 7 mm) carry collectively significant rupture risk, yet conventional scoring tools (PHASES, ELAPSS) perform near chance-level in this subgroup. This systematic review and meta-analysis evaluated the pooled diagnostic accuracy of AI-based models for rupture risk prediction in small UIAs and compared performance across algorithmic architectures. Prospectively registered with PROSPERO (CRD420251142915) and conducted per PRISMA-DTA/PRISMA-S standards. MEDLINE, Scopus, and Web of Science were searched in October 2025 and updated on June 2026. Eligible studies applied machine learning (ML), deep learning (DL), radiomics, or computational fluid dynamics-integrated models to small UIAs. Area under the receiver operating characteristic curve (AUC) was pooled via a three-level random-effects model with restricted maximum likelihood estimation and Knapp-Hartung correction; heterogeneity was explored through multivariable meta-regression. Bias and reporting quality were assessed using QUADAS-3 and METRICS, respectively. Ten studies (2019-2025) encompassing 5409 patients, 7057 aneurysms, and 62 models were included. Pooled AUC was 0.793 (95% CI 0.754-0.827), with sensitivity of 0.713 and specificity of 0.780. Conventional ML achieved the highest AUC (0.827), followed by statistical models (0.779) and DL (0.741). Meta-regression explained 76.2% of variance; statistical model family was the sole significant predictor of inferior performance (p = 0.002). Nine of ten studies carried high risk of bias. No publication bias was detected (p = 0.186). AI-based models, particularly conventional ML ensemble architectures, demonstrate moderate-to-good discriminative accuracy for small UIA rupture risk stratification, meaningfully surpassing conventional scoring tools. Multicenter prospective registries, pre-rupture imaging protocols, and explainable AI frameworks are essential to support transparent neurosurgical decision-making.
