Application Study of Clinical Features Combined With Ultrasound-based Radiomics Nomogram For Grading Diagnosis of Bladder Urothelial Carcinoma.

UNLABELLED: RATIONALE AND OBJECTIVE: This study aims to develop a prediction model using multimodal ultrasound (US) images and clinical features to evaluate its effectiveness in predicting bladder cancer grading. MATERIALS AND METHODS: We selected 190 patients with bladder urothelial carcinoma; each patient underwent 2D US and contrast-enhanced US examinations. Based on pathological results, the patients were categorized into either a low-grade urothelial carcinoma group or a high-grade group. We extracted 1240 radiomic features from each patient's region of interest. Using logistic regression, we established the US radiomics model, the contrast (C) radiomics model, and the US-C radiomics model. Diagnostic accuracy was evaluated using receiver operating characteristic curve analysis. Based on the rad score obtained from the US-C radiomics model and three independent predictive factors, we constructed a clinical imaging nomogram. RESULTS: The area under the curve (AUC) values for the US radiomics model, the C radiomics model, and the US-C radiomics model were 0.772, 0.868, and 0.917, respectively, in the training groups. In the validation groups, the AUC values were 0.690, 0.715, and 0.735, respectively. The predictive power of the clinical model was 0.766 in the training group and 0.763 in the validation group. The predictive performance of the clinical imaging nomogram was 0.939 in the training group and 0.807 in the validation group. CONCLUSION: Among the three models, the US-C radiomics model demonstrated the highest predictive performance. The calibration curve indicates that the nomogram model provides a superior fit for predicting the grading of bladder urothelial carcinoma.