A comprehensive scoring system integrating clinical and radiological variables for the detection of clinically significant prostate cancer on bi-parameter MRI: multi-center comparison with multi-parametric MRI.

PURPOSE: To develop and validate a scoring system that combines clinical and radiological variables to predict the likelihood of clinically significant prostate cancer (csPCa, which is defined as Gleason Grade group ≥ 2) before biopsy and stratify patients by predicted risk.

METHODS

This retrospective study enrolled 788 patients. Data were stratified into a derivation cohort and two validation cohorts by institutions. Imaging evaluation included: Prostate Imaging Reporting and Data System v2.1 (PI-RADS v2.1), Simplified PI-RADS [S-PI-RADS, incorporating qualitative and quantitative assessment of diffusion restriction degree and lesion volume (LV) on DWI/ADC sequences], prostate volume (PV), LV, and the longest to shortest diameter ratio (LD/SD). %fPSA was the ratio of free prostate-specific antigen (PSA) to total PSA (tPSA). The adjusted PSA (aPSA) was derived as: aPSA = tPSA×(LV/PV). Independent csPCa predictors were determined through multivariate regression analysis and transformed into the scoring system. Diagnostic performance of PI-RADS, S-PI-RADS, and the scoring system were compared using ROC analysis. The scoring system was stratified into four risk categories based on total scores.

RESULTS

The scoring system-integrating %fPSA, S-PI-RADS, aPSA, and LD/SD-demonstrated robust predictive performance for csPCa across all derivation and validation cohorts (AUC: 0.891, 0.875, and 0.897, respectively), comparable to PI-RADS (AUC: 0.861, 0.880, and 0.865; all P > 0.05). It achieved the highest PPV among three systems in all cohorts (0.780, 0.832, and 0.722). Median predicted probabilities for the low (0-2 points), intermediate-low (3-7), intermediate-high (8-12), and high-risk groups (13-17) were 3.8%, 14.4%, 67.8%, and 90.3%, respectively, aligning with observed risks. While inter-reader agreement for PI-RADS was suboptimal between trained and untrained radiologists (κ = 0.643), the scoring system showed stronger consensus (κ = 0.808).

CONCLUSION

This scoring system demonstrating comparable diagnostic performance to PI-RADS and improving PPV, highlighting its potential clinical practicability.