Developing Bayesian Networks based Prognostic Radiomics Model for Clear Cell Renal Cell Carcinoma Patients

Abstract

Clear cell renal cell carcinoma (ccRCC) is one of the most aggressive histologic subtype of RCC. In this study, we developed and evaluated a Bayesian network as a prognostic model using computed tomography (CT) radiomic features and clinical information to predict the risk of death within 5 years for ccRCC patients. Seventy patients who had abdominal CT scans with delayed post-contrast phase and outcome data were enrolled. 3D volumes of interest (VOIs) covering the whole tumor on CT images were manually delineated. Image preprocessing techniques including, wavelet, Laplacian of Gaussian, and resampling of the intensity values to 32, 64 and 128 bin levels were applied on all VOIs. Different radiomic features, including shape, first-order, and texture features were extracted from the VOIs. For features selection, we first used the z-score method to normalize all image features, then the relevant features were selected based on mutual information (MI) criteria. The patients were divided into a low- and high-risk group based on survival or death at 5 years after surgery, respectively. Bayesian networks were used as a classifier for risk stratification. The model was evaluated using the area under the curve (AUC), sensitivity, specificity, and accuracy by 1000 bootstra resampling. The Bayesian model with Laplacian of Gaussian (LOG) filter showed the best predictive performance in this cohort with an AUC, sensitivity, specificity, and accuracy of 0.94, 85 %, 94%, and 89%, respectively. The results of the current study indicated that prognostic models based on radiomic features are very promising tools for risk stratification for ccRCC patients.