Abstract
PURPOSE: This study evaluated inter-/intra-reader agreement with the Prostate Imaging Reporting and Data System (PI-RADS) version 2.1 to improve the detection rate of prostate cancer. MATERIALS AND METHODS: We enrolled 210 patients who underwent multiparametric magnetic resonance imaging (mpMRI) for clinically suspected or diagnosed prostate cancer. Four readers, including two urologists, viewed patients' mpMRI and scored PI-RADS between two sessions, including the time for feedback and training after the first reading session. Inter- and intra-reader agreements were evaluated using Fleiss' kappa coefficient (κ), agreement coefficient 1 (AC1), and percentage of agreement (PA). RESULTS: The overall inter-reader agreement between all readers was moderate (κ=0.466, AC1=0.522, and PA=0.610). The overall inter-reader agreement improved in the second session. The agreement for peripheral zone (PZ) lesions was higher than that for transitional zone (TZ) lesions. At a PI-RADS cut-off of 4, the agreement for PZ lesions was almost perfect (PA=0.888) and higher than that for TZ lesions. The inter-reader agreement for lesions with a PI-RADS ≥4 and Gleason score ≥7 was almost perfect (AC1=0.960 and PA=0.964). The intra-reader agreement for lesions overall and PI-RADS ≥4 lesions were substantial (AC1=0.601) and almost perfect (PA=0.876), respectively. CONCLUSIONS: Readers achieved moderate agreement for PI-RADS version 2.1 and benefitted from training sessions. Feedback, training, and multidisciplinary discussions also improved inter-reader agreement. Our study can provide guidance, updates, and further steps for the standardization and improvement of PI-RADS scoring.