Using radiomics model for predicting extraprostatic extension with PSMA PET/CT studies: a comparative study with the Mehralivand grading system.

Abstract

Purpose: This study aimed to evaluate the effectiveness of using a radiomics model to predict extraprostatic extension (EPE) in prostate cancer from PSMA PET/CT, and to directly compare its performance with the Mehralivand Grading System, an MRI-based method for EPE assessment.

Methods: A total of 206 patients who underwent radical prostatectomy were included in this study. Radiomics features were extracted from PSMA PET/CT images to construct predictive models using Support Vector Machine (SVM) and Random Forest algorithms. In addition, among the 63 patients who underwent both PSMA PET/CT and multiparametric MRI (mpMRI), the performance of the radiomics model was compared with that of the Mehralivand Grading System. Key performance metrics, including the area under the curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), were reported.

Results: Among the 63 patients who underwent both PSMA PET/CT and multiparametric MRI (mpMRI), the radiomics model achieved an AUC of 76.8% (95% CI: 64.4-86.5%), sensitivity of 72.0%, specificity of 81.5%, PPV of 72.0%, and NPV of 81.6%. In comparison, the Mehralivand Grading System yielded AUCs of 66.8%, 63.5%, and 60.2% from three independent readers. DeLong's test showed that the radiomics model significantly outperformed all three readers in terms of AUC (p = 0.013, 0.003, and 0.001, respectively).

Conclusion: The radiomics model derived from PSMA PET/CT can better capture features associated with EPE and shows promise for aiding preoperative assessment in prostate cancer. However, further validation in larger, independent cohorts is necessary to confirm its stability and clinical utility.