Anatomy-specific Performance of CT Angiography-based AI for Anterior Circulation Occlusion: Systematic Review and...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Large-vessel occlusions (LVOs) in the anterior circulation are critical for stroke diagnosis and treatment.
• CT angiography (CTA) is a primary imaging modality for detecting these occlusions.
• Artificial intelligence (AI) is emerging as a tool to aid in CTA interpretation.

Purpose of the Study:

• To evaluate the anatomy-specific diagnostic performance of AI algorithms for detecting anterior circulation occlusions on CTA.
• To assess the implications of negative AI findings for distal occlusions.

Main Methods:

• A systematic review and diagnostic meta-analysis adhering to PRISMA-DTA guidelines.
• Searched PubMed, Embase, and Web of Science for AI-based CTA studies from January 2019 to March 2026.
• Pooled sensitivity and specificity were calculated for global LVOs, ICA/M1 occlusions, and M2/M3 occlusions using a bivariate random-effects model.

Main Results:

• Thirty-one studies involving 15,708 patients were analyzed.
• Pooled sensitivity/specificity: 80.1%/91.9% for global LVOs, 91.6%/92.7% for ICA/M1, and 52.7%/94.8% for M2/M3 occlusions.
• AI demonstrated higher sensitivity for ICA/M1 occlusions compared to global LVOs, but significantly lower sensitivity for M2/M3 occlusions.

Conclusions:

• The diagnostic performance of CTA-based AI varies significantly depending on the anatomical location of the occlusion.
• Lower sensitivity for distal M2/M3 occlusions reduces the reliability of negative AI results for these specific findings.
• Further research is needed to improve AI performance in detecting distal occlusions.