Wednesday, December 3 | 10:10 a.m.-10:20 a.m. | SSER02-5 | Room N228
French researchers have found that AI-driven triage alerts serve as "effective safety nets," improving the detection of missed critical findings in emergency departments (EDs).
Presenter Guillaume Herpe, MD, PhD, of University Hospital Center Poitiers in France, will share results from a retrospective, multicenter study of AI-based detection algorithms deployed for brain trauma intracranial hemorrhage (ICH), pneumothorax on chest x-ray, and incidental pulmonary embolism (IPE) across 2,336 CT scans and 119 chest x-rays.
The study involved 35 radiologists and one tertiary and two secondary emergency radiology centers. For ICH detection alone, AI flagged 133 positives (19.5%, 133/682), including 46 false positives (6.7%, 46/682) and 11 false negatives (1.6%, 11/682) among 682 head CT scans, the group reported.
A key finding was that while there were 1.5 false-positive alerts every day, AI-driven triage tools still identified a life-threatening condition (ICH, IPE, or pneumothorax) missed by a radiologist every five days. The false-positive rate was manageable but required radiologist validation, according to the group.
"Given the acceptable number of alerts, their integration remains valuable, provided that false positives are managed to maintain workflow efficiency and prevent alert fatigue," Herpe and colleagues wrote in their research abstract.
Join the session to hear what conditions radiologists missed the most and how AI increased the detected prevalence.
