Aim: Prostate cancer (PCa) is one of the most common neoplasms in men. However, the value of ultrasound-based radiomics for diagnosing PCa remains uncertain.
Material and methods: We retrospectively analyzed ultrasonic and clinical data from 373 patients. Patients were divided into two groups according to the pathological results. Radiomics features wereextracted from TRUS, and we screened the optimal features to construct radiomics models. Relationships between clinical characteristics and prostate lesions were identified by univariate and multivariate logistic regression analysis. Finally, a clinical-radiomics model was developed, and then visualized in the form of a nomogram.
Results: Of the 373 patients, 178 had benign disease and 195 had malignant disease. The support vector machine (SVM) classification model showed the best performance, while the diagnostic performance of the clinical model was poorer than that of the radiomics model (p<0.05) or the combined (clinical-radiomics) model (p<0.05). In general, the combined model demonstrated the highest AUC and proved to be more advantageous.
Conclusion: The prediction model we constructed based on TRUS predicted PCa preoperatively with high efficiency. In addition, combining radiomics with clinical factors improved diagnostic accuracy.