Radiomics prediction of the pathological grade of bladder cancer based on multi-phase CT images

Abstract

Abstract Background The pathological grade of bladder cancer(BCa)is a critical determinant for the follow-up clinical decision and treatment of patients. The authors investigated a radiomic-clinical model in predicting the pathological grade of BCa. Objective This study explored the feasibility of the radiomics based on multi-phase thick-slice CT images combined with clinical risk factors in predicting of the pathological grade of BCa. Methods Patients with BCa who underwent CT scan and surgical treatment from January 2019 to December 2021 were analyzed retrospectively, with 104 cases of high-grade BCa and 100 cases of low-grade BCa included. Radiomics features were extracted from tumor volume in the images of the plain scan, corticomedullary phase, and parenchymal phase, respectively. Logistic Regression model, SVM model, and Random Forest model were established, and the model with higher diagnostic efficiency was chosen. Additionally, a radiomics-clinical model was conducted by selected independent predictors according to logistic regression analysis. Then the performance of the model was assessed. Results Among the 204 patients enrolled, the training cohort was consisted of 142 patients and the validation cohort was made up of 62 patients. The Logistic Regression model proved to be the most effective one among the three models. The radiomics-clinical model consisted of 2 independent predictors, patient age and Rad-Score, with an AUC of 0.904(95%CI 0.857–0.951) and 0.906༈95%CI 0.837–0.975༉in the training and validation cohorts, respectively. The diagnostic accuracy, sensitivity, and specificity of the validation cohort were 0.790, 0.813, and 0.767 respectively. Conclusion The radiomics-clinical model possesses great potential in predicting the pathological grade of BCa.