Relaxing the PI-RADS dominant sequence rule improves the characterization of high-grade prostate cancer on multiparametric MRI.

[PURPOSE] The Prostate Imaging-Reporting and Data System 2.0 (PI-RADSv2.0) and 2.1 (PI-RADSv2.1) scores are deduced from the pulse sequence categories using the "dominant sequence" scoring rule. The purpose of this study was to develop and evaluate a new scoring rule that makes better use of non-dominant pulse sequence findings.

[MATERIAL AND METHODS] The new scoring rule was developed using a single-center database of 1627 patients who underwent prostate multiparametric MRI and prostate biopsy. The combinations of PI-RADSv2.0 pulse sequence categories observed at sextant level were ranked based on their rate of high-grade (grade group ≥ 2) prostate cancer and assigned to one of the five levels of the new score. Then, a hidden evaluation dataset of 240 MRI lesions to which 21 readers of varying experience had assigned PI-RADSv2.1 pulse sequence categories was used. For each reader, the PI-RADSv2.1 score of the lesions (PI-RADSv2.1 dominant sequence rule) and the new score (scoring rule defined in the development cohort) were computed. The scores were compared using areas under the curve (AUC), sensitivities, specificities, reproducibility, and clinical utility.

[RESULTS] Across all readers, the mean AUC of the new score (0.78; 95 % confidence interval [CI]: 0.73-0.83) was significantly greater than that of the PI-RADSv2.1 score (0.76; 95 % CI: 0.71-0.81; P < 0.01). The new score showed lower sensitivity, higher specificity and better inter-reader agreement in all reader experience subgroups. Across all readers, for a ≥ 3 dichotomization, it provided a higher net benefit than the PIRADSv2.1 score for risk thresholds > 0.15.

[CONCLUSION] The new scoring rule outperformed the dominant sequence rule in characterizing high-grade prostate cancer regardless of reader experience.