Full Breakdown
AI Algorithm Boosts Detection of Brain Aneurysms on CT Scans
By Drooid · · How we work
Core Study Findings
A prospective evaluation of 3,856 CT-angiography (CTA) examinations performed across the Northwell Health system tested an Food and Drug Administration (FDA)-cleared AI algorithm (Aidoc’s aiOS) designed to detect intracranial aneurysms. The AI identified 55 aneurysms that radiologists initially missed, while radiologists identified 30 aneurysms the AI did not flag. Overall agreement between radiologists and the AI exceeded 96 % of examinations. Compared with radiologist-only interpretation, the AI-assisted workflow achieved a 39 % relative increase in detection rate. Sensitivity rose to 84.6 % for the AI versus 71.8 % for radiologists, whereas positive predictive value was higher for radiologists (92.7 %) than for the AI (78.2 %). Of the 101 findings flagged only by the AI, 46 were later classified as false positives.
Complementary Strengths of AI and Radiologists
The AI proved especially adept at spotting small aneurysms that radiologists often overlook, a factor that can influence clinical risk assessment because aneurysm danger depends on size, shape, location, and patient-specific factors. Performance varied by care setting: in inpatient scans the AI added 18 true detections with seven false positives; in emergency-department scans it added a similar benefit; in outpatient scans the AI’s net gain was modest, yielding four additional detections but more false positives than true positives. Radiologists, meanwhile, maintained higher accuracy when confirming the presence of an aneurysm.
Official Statements & Responses
She noted that the findings support a collaborative model rather than replacement of radiologists.
Verbatim Quotes
- “The AI tool demonstrated high sensitivity for aneurysm detection which in fact surpassed that of radiologists, implying a potentially high Radiologist-AI collaborative detection rate,” — Shlomit Stein, lead study author
Implications for Clinical Practice
The enhanced sensitivity for small, potentially high-risk aneurysms suggests that AI-assisted interpretation could enable earlier risk stratification and timely intervention, potentially reducing the incidence of subarachnoid hemorrhage. However, the noted false-positive rate and setting-specific variability underscore the need for careful integration of AI outputs into radiologists’ workflow and for ongoing validation before broader adoption.