Abstract
Renal cancer accounts for considerable amounts of new cases and deaths worldwide. In recent years, increment in the use of imaging technology like computed tomography (CT) and magnetic resonance imaging (MRI) has significantly im-proved the detection of renal cancer. The approach for man-aging this disease can vary according to the properties of the tumor and the status of the patients. In this paper, we studied the effectiveness of deep learning-based methods for guiding in management of renal cancer using CT and clinical infor-mation through two tasks: Task 1 of categorizing whether the patients are eligible for adjuvant therapy and Task 2 of clas-sifying patients into one of the five risk groups as suggested by American urological association (AUA). We evaluated our methods on the "Kidney clinical Notes and Imaging to Guide and Help personalize Treatment and biomarkers discovery (KNIGHT) challenge" database with 300 cases for training and validation and an additional 100 cases for testing. We performed various experiments using only CT scans, only clinical data, and combining CT and clinical data. The best results we achieved is an area under the curve (AUC) of 0.813 on Task 1 and 0.626 on Task 2 through the TabNet model, which utilizes only clinical data. These promising results demonstrate the potential of a deep learning-based method to direct renal cancer management from CT and clinical data.