A prospective diagnostic accuracy study of Prostate Imaging Reporting and Data System version 2 on 3 Tesla multiparametric multiparametric magnetic resonance imaging in detecting prostate cancer with whole-mount pathology
Little data is available on the performance of the Prostate Imaging Reporting and Data System version 2 (PI-RADS v2) on 3 Tesla multiparametric magnetic resonance imaging (mpMRI) in detecting prostate cancer (PCa) on definitive pathology after radical prostatectomy. In this study, we assessed the accuracy of PI-RADS v2 in detecting any PCa and clinically significant PCa (csPCa) on 3 Tesla mpMRI using whole-mount histology as the standard of reference.
Between May 2016 and February 2017 we prospectively enrolled patients with biopsy-proven PCa who underwent 3 Tesla mpMRI before radical prostatectomy. Three radiologists with 8, 7 and 2 years of mpMRI experience, who were blinded to final pathology, independently mapped and scored imaging findings according to PI-RADS v2 criteria. One experienced uropathologist processed prostate specimens using whole-mount protocol, and mapped all cancers according to the PI-RADS v2 39-sector scheme. Based on a lesion-by-lesion mpMRI-to-whole-mount-pathology matching, we calculated sensitivity and specificity in detecting PCa and csPCa defined according to Epstein criteria, using PI-RADS ≥3 and ≥4 score as thresholds. We also assessed inter-reader agreement using Cohen&[prime]s weighted-kappa statistic.
Included were 48 patients (median age 68 years, median PSA 7.2 ng/ml) with 71 cancers on final histology (median size 16 mm, Gleason score ≥7 in 46 cases, stage ≥pT3 in 22 cases). Lesions were in the peripheral zone, transition zone or both in 58, 9 and 4 cases, respectively. On a per-lesion basis, sensitivity was slightly higher with PI-RADS ≥3 vs. ≥4 threshold (range 0.55-0.62 vs. 0.48-0.60 for any PCa, and 0.67-0.74 vs. 0.61-0.72 for csPCa, respectively) at the expense of lower specificity (range 0.11-0.30 vs. 0.56-0.71 for any PCa, and 0.28-0.30 vs. 0.55-0.71 for csPCa, respectively). Accuracy improved on a per-patient basis, with 0.64-0.77 sensitivity and 0.73-0.83 specificity for csPCa with a PI-RADS ≥4 threshold. Inter-reader agreement was moderate to substantial (k=0.47-0.72 vs. 0.51-0.71 for PI-RADS thresholds ≥3 and ≥4, respectively).
PI-RADS v2 showed good diagnostic performance in detecting csPCa, with acceptable inter-reader agreement. PI-RADS ≥4 threshold offered a better trade-off between sensitivity and specificity.