AI model predicts prostate cancer

An artificial intelligence (AI) model using diffusion basis spectrum imaging (DBSI) can accurately predict clinically significant prostate cancer (csPCa) on biopsy and even reduce unnecessary prostate biopsies when combined with Prostate Imaging Reporting and Data System (PI-RADS), suggests a study.

Some 241 patients underwent MRI between February 2020 and March 2024, including conventional and DBSI-specific sequences prior to prostate biopsy. AI models were used with DBSI metrics as input classifiers and biopsy pathology as the ground truth.

The investigators then compared the DBSI-based model with available biomarkers (PSA, PSA density, and PI-RADS) for risk discrimination of csPCa defined as Gleason score >7.

The DBSI-based model independently predicted csPCA (odds ratio [OR], 2.04, 95 percent confidence interval [CI], 1.52–2.73; p<0.01), as did PSA density (OR, 2.02, 95 percent CI, 1.21–3.35; p=0.01) and PI-RADS classification (PI-RADS 3: OR, 4.00, 95 percent CI, 1.37–11.6; p=0.01; PI-RADS 4-5: OR, 9.67, 95 percent CI, 2.89–32.7; p<0.01), adjusting for age, race, and family history.

Furthermore, the DBSI-based model alone showed comparable performance to PSA density plus PI-RADS (AUC 0.863 vs 0.859; p=0.89), while the DBSI-based model plus PI-RADS combination demonstrated the highest risk discrimination of csPCa (AUC 0.894; p<0.01).

“A clinical strategy using the DBSI-based model for patients with PI-RADS 1-3 could have reduced biopsies by 27 percent while missing 2 percent of csPCa compared with biopsy for all,” the investigators said.