Monday, November 30 | 9:10 a.m.- 9:20 p.m. | M1-SSGU02-8 | Room E353B
AI-assisted prostate MRI interpretation appears to improve exam readers' cancer detection and to reduce variability in lesion delineation, according to research to be presented Monday morning.
A group led by presenter Brittaney Everitt, clinical data scientist at DeepHealth in Stamford, CT, conducted a study that included 247 prostate MRI exams (138 with a Gleason score ≥6, 109 cancer negative, and 90% from U.S. sites) interpreted by 12 readers. The readers reviewed each exam with and without the help of AI, and assessed prostate cancer detection performance using area under the localization receiver operating characteristic curve (AUC-LROC) stratified across patient, imaging and reader subgroups.
Everitt's team found that the AI algorithm improved readers' ability to detect clinically significant prostate cancer by 10.4%. It also reported that the algorithm improved disease detection by age (15.4%); race (White men, 17.8% and Black men, 7.4%); by scanner (GE HealthCare, 13.1%; Philips, 21.5%; Siemens Healthineers, 14.5%); by field strength (1.5-tesla, 18.5% and 3-tesla by 15.6%); and by reader experience (≤9 years: 14.8%; >9 years: 16.9%).
The takeaway? "Automated AI segmentations could enable more accurate lesion localization and streamline workflows through consistent contouring and compatibility with fusion biopsy platforms," the group concluded.
Find out more by attending this talk on Monday.
