Wednesday, December 2 | 9:30 a.m.-10:30 a.m. | W3-SSNR10-6 | Room E350
Although an AI algorithm flagged far more cerebral microbleeds on brain MRI than a radiologist did, it also raised more false alarms, researchers will report in this presentation.
Presenter Adolfo Manuel Delgado Brito, MD, of Clínica Universidad de Navarra in Pamplona, Spain, and colleagues evaluated whether an AI algorithm could match or outperform a radiologist in detecting cerebral microbleeds on susceptibility-weighted imaging (SWI) MRI.
In the retrospective study of 146 patients, the AI algorithm and an experienced radiologist independently reviewed each SWI scan. The radiologist then performed a second-look review that incorporated the AI findings, creating a combined reference standard.
At the patient level, AI had much higher sensitivity than the radiologist (97% vs. 65%) but lower specificity (54% vs. 100%). The radiologist tended to undercount microbleeds, while AI tended to overcount them. The researchers noted that because the reference standard incorporated AI results, it may favor the algorithm.
“AI improves sensitivity for microbleed detection, reducing missed lesions, while radiologists maintain perfect specificity,” the researchers wrote. “Their complementary error patterns support a combined diagnostic approach. AI-assisted detection of cerebral microbleeds enhances diagnostic sensitivity while preserving clinical accuracy through radiologist validation, supporting its role as a second-reading tool within routine MRI workflows.”
For the full story, attend this session on Wednesday morning.
