Multimodal radiomics based on fluorine-18 prostate-specific membrane antigen positron emission tomography and multiparametric magnetic resonance imaging in predicting persistent prostate-specific antigen after radical prostatectomy.

Background: Persistent prostate-specific antigen (PSA) after radical prostatectomy (RP) is associated with increased metastasis and mortality. However, the value of the radiomics for predicting persistent PSA is unclear. Our study aimed to evaluate the diagnostic performance of F-PSMA-1007 positron emission tomography (PET) and multiparametric magnetic resonance imaging (mpMRI) radiomics for the prediction of persistent PSA after RP.

Methods: Retrospective analysis was performed on 141 patients with prostate cancer (PCa) who had undergone F-prostate-specific membrane antigen (PSMA)-1007 PET and mpMRI scans before RP. Patients were placed into two groups according to PSA levels examined within 4-8 weeks after surgery: a nonpersistent PSA group and a persistent PSA group. PET-derived and mpMRI-derived radiomics features were used to develop radiomics models. Age and initial PSA were incorporated into the clinical model. Individual models and their various combinations were developed and their performance evaluated.

Results: All radiomics models consistently outperformed the clinical model [C model: area under curve (AUC) =0.744]. The best-performing radiomics model was the PET- and mpMRI-derived model (PM model) created by combining the radiomics features of PET and mpMRI, which yielded an AUC of 0.849 in the validation cohort, and was superior to the other radiomics models, including the PET-derived model (P model: AUC =0.794) and the mpMRI-derived model (M model: AUC =0.815). The combined model, integrating the clinical variables and the best-performing radiomics model, demonstrated the highest performance (AUC =0.903) and significantly outperformed the C model (P<0.05). Decision curve analysis indicated that the combined model provided greater net benefits than did the C model and PM model.

Conclusions: The combined radiomics-clinical model was the best-performing model and outperformed both clinical and radiomics models in predicting persistent PSA, indicating that clinical variables can complement PSMA-PET and mpMRI radiomics for early risk stratification following RP.