Biparametric MRI-based radiomics for differentiating clinically significant prostate cancer among prostate-specific antigen level of gray zone.

PURPOSE

This study was intended to evaluate the performance of biparametric MRI (bpMRI) radiomics for detecting clinically significant prostate cancer (csPCa) in men with prostate-specific antigen (PSA) of 4-10 ng/mL.

METHOD

We retrospectively included 287 patients with PSA levels of 4-10 ng/mL. Radiomics features were extracted from two MRI protocols of T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI, with b-values of 0, 1000, and 2000 s/mm²), and then selected with the least absolute shrinkage and selection operator (LASSO) regression method. The apparent diffusion coefficient (ADC) maps were calculated from these images and used for analysis. The radiomics signature (Radscore) based on the most useful radiomics features was calculated with the logistic regression method. MRI/US fusion targeted biopsy results were used as the reference standard. Diagnostic performance was decided using the area under the receiver operating characteristic (ROC) curve (AUC), and compared with Delong's test. Finally, a model integrating radiomics features and Prostate Imaging Reporting and Data System (PI-RADS) was constructed.

RESULTS

A total of 15 T2WI radiomics features and 12 from DWI features were retained after selection with LASSO regression. On the test set, radiomics outperformed PI-RADS, with an AUC of 0.928 (95% CI 0.868-0.988) vs. 0.807 (95% CI 0.705-0.908; P=0.04). Additionally, the combined nomogram generated higher diagnostic accuracy (AUC 0.955, 95% CI 0.905-1.00), significantly outperforming both PI-RADS (P=0.002) and radiomics alone (P=0.02).

CONCLUSION

bpMRI-based radiomics exhibited promising diagnostic accuracy for the detection of csPCa, significantly outperforming either PI-RADS or PSAD among patients with PSA of 4-10 ng/mL. Furthermore, the developed nomogram integrating radiomics and PI-RADS could further enhance diagnostic performance.