Brain Aneurysm Detection with AI Triage Assistance: A Retrospective Quality Improvement Study

Purpose: Modern radiologists are faced with the difficult task of reading a high volume of studies quickly and accurately. Delays in treatment due to large workloads can result in negative patient outcomes for a variety of pathologies. Brain aneurysms (BAs) can be fatal if not evaluated, monitored, and treated as necessary. Artificial intelligence (AI) enabled workflow triage tools can assist radiologists in flagging scans suspected of BA for more rapid diagnosis and attention. This study analyzes the improvement in the detection of BA cases in a health system after implementing an AI triage tool.

Methods/Materials: A retrospective analysis was performed on 2422 head CT scans from Feb. 16, 2024, to July 31, 2024. Cases were analyzed for brain aneurysm presence via an image-based AI triage workflow tool (Aidoc, Tel Aviv, Israel). Radiologist findings for BA were assessed by applying an NLP to the finalized reports. Radiologists identified 65 positive cases in this time frame (2.68% prevalence). In cases of disagreement between the AI and radiologist findings, the images were sent to radiologists for ground truth adjudication. Enhanced detection rate (EDR) was calculated based on the percentage of cases with AI+ and radiologist negative findings. A ground-truth corrected miss rate was calculated following adjudication. Patients with newly discovered findings were contacted for clinical follow-up.

Results: The AI module labeled 16 unique positive patients that radiologists had labeled negative, yielding a 24.6% enhanced detection rate. A single false positive case resulted in a miss rate of 23.1%. Of the 15 true positive new findings, 4 resulted in a recommendation of angiogram and surgical intervention. Over a year with similar case volume, the AI triage tool could assist with the detection of approximately 33 brain aneurysms than without the tool.

Conclusions: Image-based AI radiology workflow tools can improve the detection of time-sensitive pathologies, leading to improved patient outcomes.