Investigating the role of multiparametric and biparametric MRI based on PI-RADS v2.1 and machine learning models in prostate cancer diagnosis

Purpose

We compared the diagnostic performance of multiparametric MRI (mpMRI) and biparametric MRI (bpMRI) in detecting clinically significant prostate cancer (csPCa) using the Prostate Imaging Reporting and Data System (PI-RADS) version 2.1. Additionally, we constructed multiple machine learning (ML) models for detecting csPCa using PI-RADS scores and clinical parameters.

Materials and Methods

We enrolled 583 patients with 594 lesions who underwent mpMRI before MRI/transrectal ultrasound (MRI-TRUS) fusion-targeted biopsy and systematic biopsy. The diagnostic performance of bpMRI and mpMRI was analyzed by the area under the curve (AUC). We built multiple ML models for detecting csPCa. The input parameters were: PI-RADS scores in bpMRI or mpMRI, age, prostate-specific antigen (PSA), MRI-defined PSA density (PSAD), and prostate volume (PV). Training and test cohorts included 475 and 119 lesions, respectively.

Results

The AUCs of bpMRI and mpMRI for the diagnosis of csPCa in total lesions were 0.88 and 0.90, respectively (p<0.05). mpMRI had higher sensitivity (93.1%) but lower specificity (77.3%) compared to bpMRI (sensitivity: 79.2%; specificity: 86.2%). All the ML models exhibited the high AUC in detecting csPCa (0.93–0.96 based on mpMRI models and 0.91–0.94 based on bpMRI models). There were no statistically significant differences in the AUC values between the two groups of ML models in test sets.

Conclusions

Compared to bpMRI, the AUC of mpMRI based on PI-RADS v2.1 to detect csPCa was higher. The diagnostic performance of ML models for detecting csPCa using PI-RADS scores and clinical parameters was excellent and comparable between mpMRI and bpMRI.