Automated Prediction Of TNM Stage For Clear Cell Renal Cell Carcinoma Disease By Analyzing CT Images of Primary Tumors

Abstract

TNM Staging is important for prognosis, treatment planning and research of Clear Cell Renal Cell Carcinoma (CCRCC). The goal of this study was to investigate if CT images alone can be used to predict stages. Overall 191 patient data (TCGA-KIRC) were used and the number of images for stages one to four was 92, 19, 50 and 30 respectively. Tumors were manually defined by an expert radiologist on single slices. Open-source software was used to extract 136 features from ROI. Normalization and data balancing were performed. The feature number was reduced to 10 after the feature selection process. Classification accuracy was found 85.4% (KNN with random space model). Accuracies were distributed among stages 1-4 as 79%, 92%, 84%, 91%. CT images can be potentially used to automatically predict the TNM stage of CCRCC patients. A higher number of CT images with standard acquisition protocols may further increase accuracy.