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

Wilms’ tumor is the most common pediatric renal tumor in children and is known for its frequent recurrence. The usual treatment approaches are surgery, chemotherapy, or radiother- apy. In North America and Europe, they frequently use preop- erative chemotherapy. In this study, we built a computer-aided prediction system for Wilms’ tumor response to preoperative chemotherapy based on contrast-enhanced CT scans through six steps: (i) delineate the images of the tumor, (ii) use first- order and second-order texture features to describe the tu- mors’ texture, (iii) use spherical harmonics, sphericity, and elongation to extract shape features, (iv) describe the tumors’ functionality by detecting the intensity changes in the contrast phases, (v) apply feature fusion based on extracted features, (vi) find the prediction results via support vector machine classifier as responsive/non-responsive. The model achieved an overall accuracy of 95.24%, with 95.66% sensitivity, and 94.12% specificity. In addition, imaging markers were used to predict early Wilms’ tumor response to chemotherapy.