An AI-Based CAP Framework for Wilms' Tumor Preoperative Chemotherapy Susceptibility

Abstract

In the field of pediatric oncology, Wilms' tumor is a common occurrence and is known for its high rate of recurrence. The study's purpose was to create a computer-based prediction system for the response of Wilms' tumor to preoperative chemotherapy. The system was developed based on contrast-enhanced CT scans using six methods. Firstly, the tumor images were delineated, followed by the characterization of the tumor's form using a 3D histogram of oriented gradients. Shape features were then extracted using spherical harmonics, sphericity, and elongation. The tumors' functionality was also demonstrated by determining the intensity changes in the contrast phases. Feature fusion was applied to the extracted features, and the responsive/non-responsive results were found using the classifier support vector machine. The system demonstrated an accuracy rate of 96.83% in total, detecting 97.83% of sensitivity and accurately identifying 94.12% specificity. Additionally, imaging markers were used to predict the early Wilms' tumor response to chemotherapy.

Keywords: Features Integration, Machine Learning, Preoperative Chemotherapy, Treatment Response, Wilms' Tumor