Accuracy and Interobserver Agreement for Prostate Imaging Reporting and Data System, Version 2, for the Characterization of Lesions Identified on Multiparametric MRI of the Prostate.

Abstract

The objective of this study was to measure the accuracy and interobserver agreement of the Prostate Imaging Reporting and Data System, version 2 (PI-RADSv2), for the characterization of prostate lesions on multiparametric MRI. This retrospective study included 170 men examined at a single institution between August 2014 and February 2015 on a 3-T MRI scanner. Study patients were found to have lesions concerning for prostate cancer that were targeted for MRI/transrectal ultrasound fusion biopsy. Two experienced readers independently assigned a PI-RADSv2 assessment category to the dominant lesion in each patient. The AUC was calculated to determine reader accuracy for the detection of clinically significant prostate cancer (Gleason score ≥ 3 + 4). The Cohen kappa statistic was used to quantify interobserver agreement. The prevalence of clinically significant prostate cancer was 0.36 (61/170 patients). The AUCs for readers 1 and 2 were 0.871 and 0.882, respectively. The AUCs were greater for peripheral zone lesions than for transition zone lesions. When a PI-RADSv2 assessment category ≥ 3 was considered positive, the agreement between readers was good overall (κ = 0.63) and was fair for transition zone lesions (κ = 0.53). When a PI-RADSv2 assessment category ≥ 4 was considered positive, the agreement was excellent overall (κ = 0.91) and was excellent for both peripheral zone lesions (κ = 0.91) and transition zone lesions (κ = 0.87). Two experienced readers were able to accurately identify patients with clinically significant prostate cancer using PI-RADSv2 with good interobserver agreement overall.