An aneurysm detection algorithm from Aidoc found 55 brain aneurysms that radiologists missed during routine care, according to a prospective study across Northwell Health.
The trial covered 3,856 CT angiography exams. Aidoc’s aiOS ran in parallel with normal workflows, and radiologists were kept blind to its output while reading the scans, so the results reflect real practice rather than a controlled comparison.
The two readers agreed on more than 96% of exams. On the disagreements, sensitivity favored the software: 84.6% of aneurysms caught versus 71.8% by clinicians alone. Clinicians were more often right when they flagged one, with a positive predictive value of 92.7% against 78.2% for the algorithm.
That combination produced a 39% relative lift in detection. Each side caught aneurysms the other missed, 55 for the software and 30 for the clinicians, which the authors read as complementary strength rather than grounds for replacement.
Payoff varied by setting. Inpatient scans produced 18 additional aneurysms against seven false-positive alerts, and emergency department performance was similarly favorable. Outpatient imaging was weaker, with four additional detections and more false positives than true ones.
Lead author Shlomit Stein, a professor of radiology at the Zucker School of Medicine at Hofstra/Northwell, said the software mostly surfaced small aneurysms that clinicians overlooked. She noted that size is only one input: shape, location and a patient’s other risk factors matter too.
