The Role of Artificial Intelligence in Estimating Stroke Events in Moyamoya Patients: A Systematic Review and Meta-Analysis of Diagnostic Test Accuracy.
Authors
Affiliations (6)
Affiliations (6)
- School of Public Health and Preventive Medicine, Monash University, Melbourne, Victoria, Australia.
- Department of Neurosurgery, Cedars-Sinai Medical Center, Los Angeles, CA, USA.
- Department of Neurosurgery, Rutgers New Jersey Medical School, Rutgers University, NJ, USA.
- Department of Neurosurgery, Robert Wood Johnson University Hospital, Robert Wood Johnson Medical School, Rutgers University, NJ, USA.
- School of Medicine and Health Sciences, Universidad del Rosario, Bogotá, Colombia.
- College of Medicine, University of Sharjah, Sharjah, UAE.
Abstract
Moyamoya disease (MMD) and syndrome (MMS) are rare cerebrovascular arteriopathies marked by progressive internal carotid stenosis, fragile collateral networks, and a five-year stroke risk near 10% despite optimal care. Artificial-intelligence (AI) models integrating angiographic, perfusion, and clinical data show promise for risk stratification, but their diagnostic accuracy and clinical readiness remain uncertain. We conducted a systematic review of AI algorithms for predicting ischemic or hemorrhagic stroke events in angiographically or magnetic resonance imaging (MRI)-confirmed MMD/MMS. PubMed, EMBASE, and Scopus were searched through September 3, 2025, for English-language studies employing machine-learning, deep-learning, or radiomics models. We extracted sensitivity, specificity, and area under the curve (AUC) metrics and assessed study quality with Radiomics Quality Score and CLEAR checklists. Pooled estimates and summary receiver-operating characteristic curves were generated; decision-curve analysis evaluated clinical net benefit. Seven retrospective cohorts (n = 4,795) met inclusion criteria. The pooled sensitivity was 0.65 (95% CI 0.50-0.79) and specificity 0.85 (95% CI 0.82-0.89). The summary AUC was 0.85. Decision-curve analysis demonstrated that AI predictions improved net benefit over "treat-all" or "treat-none" strategies across relevant risk thresholds. Tree-based classifiers (XGBoost, random forest) showed more stable external performance than deep-learning networks. Explainability tools enhanced model interpretability. AI models achieve moderate-to-high accuracy for stroke prediction in MMD/MMS and offer potential for individualized risk stratification. However, small single-center datasets, heterogeneous imaging protocols, and opaque modeling limit clinical adoption. Prospective multicenter validation, standardized data pipelines, and robust explainability frameworks are essential for integrating AI into routine neurovascular care.