An Assessment of PI-RADS v2.1: Inter-Reader Agreement, Diagnostic Accuracy, and the Added Value of PSA Density in Prostate Cancer Prediction

Abstract

Introduction: Multiparametric MRI (mpMRI) interpreted using Prostate Imaging Reporting and Data System (PI-RADS) v2.1 is central to prostate cancer (PCa) assessment, but inter-reader variability remains a concern. This study evaluated inter-reader agreement, diagnostic performance, the utility of prostate-specific antigen density (PSA-D), and imaging predictors of biopsy-detected PCa. Methods: We conducted a retrospective study of 102 patients who underwent prostate mpMRI followed by transrectal ultrasound (TRUS)-guided biopsy. Three experienced radiologists independently scored examinations using PI-RADS v2.1. Agreement was assessed using intraclass correlation coefficients (ICC) and weighted κ statistics. Diagnostic performance was evaluated at a PI-RADS ≥3 threshold. Patients were stratified by PSA-D, and logistic regression identified imaging predictors of biopsy outcomes. Reader confidence was scored on a 10-point scale. Results: PCa was detected in 60/102 patients (58.8%) and csPCa in 46/102 (45.1%). Overall agreement was moderate to good (ICC, 0.749; κ, 0.748), with higher agreement in the transitional compared to the peripheral zone (ICC, 0.822 vs. 0.662). Averaged PI-RADS scores yielded 100% sensitivity for both PCa and csPCa, with specificities of 47.6% and 35.7%, respectively. PSA-D stratification showed higher PPVs at PSA-D ≥0.15 ng/mL2, while sensitivity and NPV were 100% in both strata; specificity remained relatively low. Reader confidence correlated with PI-RADS category and interpretation accuracy. In multivariable patient-level analyses, DWI hyperintensity was independently associated with biopsy-detected PCa and csPCa. Conclusions: PI-RADS v2.1 showed moderate to good inter-reader agreement and high sensitivity but modest specificity for PCa and csPCa. Agreement was higher for transitional-zone lesions, and DWI hyperintensity was independently associated with PCa and csPCa. PSA-D subgroup estimates were imprecise, and its incremental value requires confirmation in larger prospective cohorts.