In this research for prostate cancer detection, a team at Vall Hebron Hospital in Barcelona tested whether adding radiomics and clinical variables to PI-RADS could improve prediction of clinically significant prostate cancer (Gleason ≥7) on MRI. The study analyzed 1497 MRI scans from 1395 men, using QP-Prostate for automatic prostate segmentation and QP-Insights for radiomic feature extraction, both developed by Quibim.
Published in Insights into Imaging, the results show that radiomics alone performed similarly to PI-RADS (AUC 0.838 vs. 0.833, no significant difference). Combining radiomics, PI-RADS, and six clinical variables, including age, PSA, and prostate volume, raised the AUC to 0.891, a statistically significant improvement over every other model tested.
At a sensitivity matched to the PI-RADS ≥3 threshold, the combined model avoided 18.15% of biopsies compared with 15.18% for PI-RADS alone, without missing additional cases of significant cancer. The findings support integrating radiomics and clinical data into prostate MRI workflows to reduce unnecessary biopsies while preserving diagnostic sensitivity.