Commercial Artificial Intelligence for Intracranial Aneurysm Detection on CT Angiography: A Systematic Review

Abstract Background Intracranial aneurysms may go undetected or incompletely characterized on CT angiography (CTA) in clinical practice. Unrecognized lesions carry the risk of subsequent hemorrhage. Commercially available AI software for intracranial aneurysm detection on CTA has entered clinical deployment, but evidence for its diagnostic performance remains fragmented across products, study settings, and aneurysm size thresholds. Purpose To systematically evaluate the diagnostic accuracy and clinical implementation evidence for commercially available AI software for intracranial aneurysm detection on CTA. Methods This systematic review was conducted in accordance with PRISMA 2020 guidelines and registered prospectively with PROSPERO (CRD420261403585). PubMed, Scopus, and Web of Science were searched. Eligible studies were original peer-reviewed publications evaluating commercially available, FDA-cleared, CE-marked, or clinically deployed AI software for intracranial aneurysm detection on CTA in adult patients. Studies evaluating non-commercial research algorithms, review articles, conference abstracts, and phantom-only studies were excluded. Risk of bias was assessed using QUADAS-2. Due to heterogeneity in study design, populations, reference standards, and outcome reporting, findings were synthesized narratively. Results Eight studies met the inclusion criteria, evaluating five commercial AI platforms: RAPID Aneurysm (iSchemaView), Viz.ai Aneurysm/Viz ANEURYSM (Viz.ai), Aidoc (Aidoc Medical), CerebralDoc (Shukun Technology), and Dr. Wise-AADS (Deepwise). Patient-level sensitivity and specificity varied widely across studies and were strongly dependent on aneurysm size and study population. Most platforms demonstrated higher performance for aneurysms ≥ 4–5 mm than for smaller lesions. Multiple studies (n = 5) reported AI detection of aneurysms absent from original radiology reports, including centers with subspecialty neuroradiology coverage. AI-assisted reading improved reader sensitivity and reduced CTA interpretation time in two studies. False positive burden varied considerably across platforms and study settings. Conclusions Commercial AI tools for intracranial aneurysm detection on CTA show variable but consistently size-dependent diagnostic performance. Most platforms performed more reliably for aneurysms ≥ 4–5 mm. Evidence of missed-aneurysm detection across multiple studies suggests a potential safety-net role. However, heterogeneity in study design, reference standards, and outcome reporting limits cross-platform comparison and generalizability. Prospective real-world evaluation with standardized outcome reporting is needed before broad clinical adoption.