Can MRI-Based AI Have an Impact in Reducing Unnecessary Prostate Biopsies?

Emerging research from a recently published study suggest that artificial intelligence (AI) can provide significant benefit in reducing uncertainty with initial PI-RADS 3 evaluations on prostate MRI.

For the retrospective multicenter study, recently published in European Radiology, researchers evaluated the utility of a three-tier risk classification (low, medium and high risk) with the AI software DeepHealth Prostate Suite in 787 men (median age of 70) who had prostate MRI scans and histopathologic verification. The study authors noted that 380 men in the cohort had clinically significant prostate cancer (csPCa).

Does AI Enhance Biopsy Triaging with Prostate MRI?

The researchers found that the AI software led to upgraded assessments in 93.1 percent of prostate lesions csPCa-positive lesions originally deemed as PI-RADS 3 presentations and noted low risk for 24.1 percent of lesions garnering initial PI-RADS 3 classification. The AI software combined a 98.4 percent sensitivity with an 18.9 percent reduction in unnecessary prostate biopsies, according to the study authors.

In a recent interview with Diagnostic Imaging, Francesco Giganti, MD, PhD, a co-author of the study, shared his perspective on the key findings.

“I definitely think that biopsy triaging is maybe the clearest opportunity for AI to have a role in prostate MRI at the moment. If we, for example, add the risk information that we get from the three-point scale that we have just published, this can help … the radiologist in determining if the lesion has to be biopsied or not,” maintained Dr. Giganti, a professor of radiology at University College London.

How Does AI Stack Up with Lesion-Level Sensitivity for Prostate MRI?

However, while the study revealed that the sensitivity of the AI software was non-inferior to the PI-RADS classification system (97.6 percent vs. 92.6 percent), the researchers also noted 10 percent lower sensitivity for AI at the lesion level (78.8 percent vs. 88.8 percent).

“This means that basically, it's not picking up all the cancers within the prostate, which is something that we have seen in the AI literature,” added Dr. Giganti. “ … I think the clinical message that we tried to convey with this study is that AI is supporting the radiologist. The clinical message for us is to combine AI, the information that we can get with this three-point scale, with the interpretation from experts. Radiologists are, in the end, responsible for lesion mapping, but (it can be done) with the aid of AI.”