Anatomy-specific Performance of CTA-based AI for Anterior Circulation Occlusion: Systematic Review and Meta-analysis.
Yike Liu, Chaofan Li, Zepeng Ren et al.
✦ AI-curated · Sources linked
In 30 seconds
This prospectively registered systematic review and diagnostic meta-analysis assessed CTA-based artificial intelligence for detecting anterior-circulation occlusions, using 31 reports involving 15 708 patients. Pooled sensitivity and specificity differed by territory: for ICA/M1 occlusions, 91.6% and 92.7%; for M2/M3 occlusions, 52.7% and 94.8%; and for global LVOs, 80.1% and 91.9%. A negative AI result was less reassuring for distal M2/M3 occlusions, with a negative likelihood ratio of 0.50 (95% CI: 0.28, 0.73).
Key findings
- Pooled sensitivity for ICA/M1 occlusions was 91.6%, compared with 52.7% for M2/M3 occlusions.
- Pooled specificity was 92.7% for ICA/M1 and 94.8% for M2/M3 occlusions.
- For M2/M3 occlusions, a negative AI result corresponded to posttest probabilities of 17.6% and 33.3% at 30% and 50% pretest probabilities.
- Evidence certainty was moderate for ICA/M1 sensitivity and very low for M2/M3 sensitivity.
Why it matters
AI-assisted CTA interpretation may perform differently across occlusion territories. The substantially lower sensitivity for M2/M3 occlusions means that a negative result does not reliably exclude distal occlusion, particularly when clinical or imaging suspicion is appreciable. Understanding this variation is important when interpreting AI outputs as part of diagnostic assessment rather than treating them as uniformly informative across vascular territories.
What to keep in mind
Evidence certainty was very low for M2/M3 sensitivity, limiting confidence in estimates for distal occlusions.