Comparison of PI-RADS v1 and v2 for multiparametric MRI detection of prostate cancer with whole-mount histological workup as reference standard

Purpose: The aim of this study was to compare Prostate Imaging Reporting and Data System (PI-RADS) versions v1 and v2 for the detection of prostate cancer (PCa) in multiparametric MRI (mpMRI) using whole-mount histological workup as reference standard.

Material and methods: MRI data of 40 patients with positive transrectal ultrasound-guided biopsy were analyzed retrospectively by two blinded readers (5 and 4 years' experience) with PI-RADS v1 and v2 for cancer-suspicious lesions. Prior to radical prostatectomy, patients had undergone IRB-approved mpMRI at 3 T according to PI-RADS recommendations: T2-weighted (T2w), diffusion-weighted (DWI) and dynamic contrast-enhanced (DCE) imaging. The reference standard was provided by whole-mount sections of the prostatectomy specimens. Versions v1 and v2 were compared with respect to sensitivity and positive predictive value (PPV) per lesion. Subgroups stratified by tumor location (peripheral vs. transition zone) and aggressiveness (high vs. low grade) were also analyzed. We also evaluated the concordance of the dominant MRI sequence in v2 (DWI or T2w) and the highest individual score under v1. Interobserver agreement for PI-RADS v1 and v2 was assessed by Cohen's kappa statistics.

Results: Reader 1 (R1) described 66 and Reader 2 (R2) 72 MRI lesions. The average Gleason score of 58 PCa lesions was 6.5 (range: 6 = 3 + 3 to 8 = 4 + 4), most of them (65.5%) located in the peripheral zone. PI-RADS v2 showed a trend towards lower sensitivities, but differences were not significant for both readers: R1 72.4% (v1) vs. 63.8% (v2) (P = 0.426) and R2 77.6% (v1) vs. 69.0% (v2) (P = 0.402). The trends were more pronounced in the transition zone and for low-grade cancers but remained insignificant (p-values from 0.313 to 0.691). Likewise, the apparent PPV differences, overall as well as in each zone, were not significant. Agreement between high-score v1 and dominant v2 sequence was 48% for R1 and 53% for R2. Cohen's κ of PCa detection for two readers was 0.48 for both v1 and v2.

Conclusion: Our findings indicate that the simplified, zone-specific approach of PI-RADS v2 (2015) for MRI assessment of prostate cancer may not necessarily be better than the original v1 criteria (2012). In specific cases, a strict interpretation of v2 criteria may even lead to false-negative findings. Therefore, the current PI-RADS criteria should be reconsidered, despite the low statistical evidence here.