Feasibility of transperineal prostate biopsy guided by fusion of [18F]F-PSMA-1007 PET/MRI with real-time transrectal ultrasound. Prostate cancer classification using imaging biomarkers and radiomics in PIRADS 3 lesions.

OBJECTIVE

To assess the diagnostic feasibility of transperineal biopsy guided by fusion of PET/MRI with [18F]F-PSMA-1007 and real-time transrectal ultrasound (BP PET/MR PSMA + TRUS) in patients with PIRADS 3 lesions. To analyze imaging biomarkers and radiomic features for differentiating between patients with negative biopsy, clinically non-significant prostate cancer (cnsPCa), and clinically significant prostate cancer (csPCa).

METHODS

A prospective study was conducted in 20 patients with PIRADS 3 lesions and PSA ≥ 4 ng/mL, with no concurrent suspicious lesions. All patients underwent a PET/MRI with [18F]F-PSMA-1007, followed by targeted biopsy using real-time ultrasound fusion. Diagnostic accuracy for prostate cancer (PCa) and the proportion of csPCa and cnsPCa diagnoses were calculated. Imaging biomarkers and machine learning models based on radiomic features were analyzed.

RESULTS

The technique was feasible in 100% of cases. The overall detection rate of PCa was 80%, with 100% sensitivity and 36% specificity for csPCa. SUVpeak was the only biomarker that showed significant differences between patients with negative histology and PCa (AUC 0.81; sensitivity 75%; specificity 80%; cutoff value 3.5). No biomarker successfully differentiated between csPCa and cnsPCa. Both supervised classification models demonstrated high diagnostic performance (AUC > 0.95): Multimodal regression performed slightly better for binary classification (negative vs. PCa). Random Forest outperformed in three-class classification (negative vs. cnsPCa vs. csPCa).

CONCLUSIONS

BP PET/MRI PSMA + TRUS is a feasible, safe, and potentially superior technique compared to MRI-only targeted biopsy, especially in PIRADS 3 lesions. Radiomic analysis improves discriminative ability over conventional imaging biomarkers, particularly for distinguishing between cnsPCa and csPCa.