Stepped-Wedge Cluster Randomized Trial of AI-Assisted CTA Detection for Intracranial Aneurysms in Regional Hospitals

A locked, independently validated deep learning model was used to assist radiologists in interpreting head CTA scans. The model was trained on 16,546 CTA cases and externally validated on an independent set of 900 DSA-verified CTA cases, achieving a patient-level sensitivity of 0.943 and an average of 0.187 false positives per case.

Head CTA interpretation performed by radiologists using local routine diagnostic workflows without AI support.