UCSF

Abstract: Background and Objective: Positive predictive value of PI-RADS for clinically significant prostate cancer (csPCa, grade group [GG]≥2) varies widely between institutions and radiologists. The Restriction Spectrum Imaging restriction score (RSIrs) is a metric derived from diffusion MRI that could be an objectively interpretable biomarker for csPCa. Methods: In patients scanned for suspected or known csPCa at 7 centers, we calculated patient-level csPCa probability based on maximum RSIrs in the prostate, without relying on subjectively defined lesions. We used area under the ROC curve (AUC) to compare patient-level csPCa detection for RSIrs, ADC, and PI-RADS. Finally, we combined RSIrs with clinical risk factors via multivariable regression, training in a single-center cohort and testing in an independent, multi-center dataset. Key Findings and Limitations: Among all patients (n=1892), probability of csPCa increased with higher RSIrs . GG≥4 csPCa was most common in patients with very high RSIrs. Among biopsy-naïve patients (n=877), AUCs for GG≥2 vs. non-csPCa were 0.73 (0.69-0.76), 0.54 (0.50-0.57), and 0.75 (0.71-0.78) for RSIrs, ADC, and PI-RADS, respectively. RSIrs significantly outperformed ADC (p<0.01) and was comparable to PI-RADS (p=0.31). The combination of RSIrs and PI-RADS outperformed either alone. Combining RSIrs with PI-RADS, age, and PSA density in a multivariable model achieved the best discrimination of csPCa. Conclusions and Clinical Implications: RSIrs is an accurate and reliable quantitative biomarker that performs better than conventional ADC and comparably to expert-defined PI-RADS for patient-level detection of csPCa. RSIrs provides objective estimates of probability of csPCa that do not require radiology expertise.