Sunday, November 30 | 2:40 p.m.-2:50 p.m. | S5-SSGU02-2 | Room E353B
An AI algorithm can increase the accuracy of nonexpert radiologists and decrease interreader variability, according to this scientific presentation.
In a multireader study, researchers led by presenter Guillaume Herpe, MD, PhD, of the University Hospital of Poitiers in France enlisted six certified radiologists to read 201 prostate MRI cases from the PROSTATEx Challenge. Of the six radiologists, five had previously read 250-500 prostate MRI cases and were considered to be nonexpert radiologists. The sixth radiologist was an expert, having read more than 1,000 prostate MRI exams.
All participants read each case twice in two reading blocks with at least four weeks of separation, both without and then with AI assistance. In each session, the radiologists classified and located lesions based on the PIRADS 2.1 scoring system. Biopsy results were then used to determine ground truth.
On their own, the nonexpert radiologists had an accuracy of 55% for detecting clinically significant prostate cancer, improving to 65% with AI assistance. The AI software independently yielded an accuracy of 73%. But the expert radiologist had the best performance of them all, producing 79% accuracy.
“This AI software increased the diagnostic accuracy of nonexpert radiologists and reduced interreader variability, and therefore has the potential to support the MRI process for biopsy decision-making and target identification,” the authors wrote.
Attend this talk on Sunday afternoon to learn more.
