AI outperforms radiologists in detecting clinically significant prostate cancer

Editor's Note

In a recent study, an artificial intelligence (AI) system detected more clinically significant prostate cancers and fewer indolent cancers than human radiologists reading MRIs, MedPage today reported June 13.

The MedPage report covers a study published in Lancet Oncology that, according to researchers, “provided evidence that AI systems, when adequately trained and validated for a target population with thousands of patient cases, could potentially support the diagnostic pathway of prostate cancer management. A clinical trial is required to determine if such a system translates to improvements in workflow efficiency, healthcare equity, and patient outcomes."

In a subset of 400 cases, the AI system achieved an area under the receiver operating characteristic curve (AUROC) of 0.91, compared to 0.86 for 62 radiologists using the Prostate Imaging-Reporting and Data System version 2.1 (PI-RADS 2.1), MedPage reports. This result confirmed the AI system's noninferiority and superiority in case-level diagnosis.

At the mean PI-RADS 3 or greater operating point of all readers, the AI system detected 6.8% more cases of Gleason grade group 2 or greater cancers at the same specificity (57.7%), while reducing false positives by 50.4% and lowering the detection of Gleason grade group 1 cancers by 20% at the same sensitivity (89.4%).

Noninferiority was not confirmed among all 1,000 testing cases of AI versus radiologist readings. Among this larger group, the AI system showed similar sensitivity (96.1%) but slightly lower specificity (68.9% vs. 69%) compared to radiologists in practice. Researchers reportedly attributed the performance gap to radiologists having access to patient history and peer consultations.

The full MedPage report offers further context on related research and the limitations of this study.