A Nomogram Based on a Multiparametric Ultrasound Radiomics Model for Discrimination Between Malignant and Benign Prostate Lesions

Objectives: To evaluate the potential of a clinical-based model, a multiparametric ultrasound-based radiomics model, and a clinical-radiomics combined model for predicting prostate cancer (PCa). Methods: A total of 67 patients with prostate lesions were included in this retrospective study. Among them, 36 patients had no prostate cancer detected by biopsy and 31 patients had prostate cancer. Clinical risk factors related to PCa (age, prostate volume, serum PSA, etc.) were collected in all patients. Prior to surgery, patients received transrectal ultrasound (TRUS), shear-wave elastography (SWE) and TRUS-guided prostate biopsy. We used the ratio of 7:3 to establish training and validation sets. The images were manually delineated and registered. All modes of ultrasound radiomics were retrieved. Machine learning used the co-registered histopathology of systematic biopsy as a reference to draw the benign and malignant regions of interest (ROI) through the application of LASSO regression. Three models were developed to predict the PCa: a clinical model, a multiparametric ultrasound-based radiomics model and a clinical-radiomics combined model. The diagnostic performance and clinical net benefit of each model were compared by receiver operating characteristic curve (ROC) analysis and decision curve. Results: The multiparametric ultrasound reached a region-wise area under the receiver operating characteristics curve (ROC-AUC)of validation set of 0.86 for PCa, meanwhile, AUC of B-mode radiomics and SWE radiomics were 0.75 and 0.82, respectively. Both the multiparametric ultrasound-based radiomics model (AUC value of 0.86) and the clinical-radiomics combined group model (AUC value of 0.92) have a higher predictive effect than the clinical model (AUC: 0.82). Conclusions: Radiomics-based machine learning models can improve the accuracy of PCa predictions both in terms of diagnostic performance and clinical net benefit, compared with evaluating only clinical risk factors associated with PCa.