Wednesday, December 3 | 9:50 a.m.-10:00 a.m. | SSER02-3 | Room N228
Although AI can reduce the number of missed intracranial hemorrhages (ICH) in emergencies, radiologists remain the gold standard for diagnosis, researchers have found.
AI, while highly sensitive, produces 92-fold more false positives and necessitates radiologist oversight, according to presenter Anna Khoruzhaya of the Moscow Health Care Department in Russia. During this session, Khoruzhaya and colleagues will explain this and other findings from their multicenter study involving radiologists across 67 hospitals.
The study compared three diagnostic AI tools designed to detect ICH on head CT scans. AI-1, AI-2, and AI-3 all performed differently, varying most in accuracy and specificity, according to the group. Ground-truth expert verification included 3,409 CT scans from 439,542 total.
Radiologists outperformed all AI systems, the authors found. However, they suggested two models, in particular, show clinical potential -- if optimized for specificity. They also said training radiologists to interpret AI output is critical.
"A theoretical combined AI-radiologist approach could achieve 99.5% sensitivity, though overlapping errors limit practical implementation," Khoruzhaya and colleagues wrote in their research abstract.
The group emphasized that while AI can reduce missed ICH in emergencies, it requires rigorous validation. Importantly, they said integration should prioritize AI as a screening tool, with mandatory radiologist review of AI-positive cases.
"Synergy between AI and human expertise may enhance diagnostic safety if protocols address false positives," the group wrote.
Drop in to review performance metrics of all three ICH detection services..
