Upfront Surgery or Preoperative Chemotherapy for Wilms Tumor: Challenging the Dogma Using Artificial Intelligence
Ahmed Abdelhalim, M.D.1, Ahmed Nashat, M.D.2, Rasha T. Abouelkheir, M.D.2, Israa Sharaby, B.Sc.3, Hossam M. Balaha, M.Sc.3, Ahmed Alksas, M.Sc.3, Mohamed Shehata, Ph.D.3, Mallorie Gayhart, B.Sc.4, Ali Mahmoud, Ph.D.3, Nora S. Alghamdi, Ph.D.5, Mohammed Ghazal, Ph.D.6, Ashraf Khalil, Ph.D.7, Ayman El-Baz, Ph.D.3, Ahmed Mosbah, M.D.2.
1West Virginia University, Morgantown, WV, USA, 2Mansoura Urology and Nephrology Center, Mansoura, Egypt, 3University of Louisville, Louisville, KY, USA, 4Berea College, Berea, KY, USA, 5Princess Nourah Bint Abdulrahman University, Riyadh, Saudi Arabia, 6Abu Dhabi University, Abu Dhabi, United Arab Emirates, 7Zayed University, Abu Dhabi, United Arab Emirates.
BACKGROUND: Childhood renal tumors (CRT) don’t respond similarly to preoperative chemotherapy (PC). We sought to create a computer-aided prediction system to help identify tumors that are less likely to respond to PC based on pre-therapy contrast-enhanced CT (CECT).
Methods: A tertiary center database was reviewed for children <18 years diagnosed with Wilms’ tumor and received PC between 2000-2021. Patients were excluded if pre- and post-PC CECT could not be retrieved. According to the Response Evaluation Criteria in Solid Tumors, >30% tumor size reduction was considered a favorable response. To create the prediction model, the following steps were followed: tumor delineation in the three contrast phases; extraction of shape, texture and functionality-based features; integration (fusion) of the features extracted and selection of the prediction model with the highest diagnostic performance. K-fold cross-validation allowed all data parts to be presented in the training and testing phases.
Results: : A total of 63 tumors with a mean patient age of 4.31±2.82 years were used to train and test the prediction model. Patients were treated with 6-8 weeks of vincristine/actinomycin-D combination. Favorable response to PC was observed in 46 (75.4%) tumors. Among other machine learning classifiers, support vector machine had the best diagnostic performance with 95.24% overall accuracy, 95.65% sensitivity of and 94.12% specificity.
Conclusions: Based on pre-therapy CECT, computer-aided prediction systems can help identify CRT that are less likely to respond to PC with excellent accuracy. These tumors can be offered upfront surgery, and subsequently avoid the disadvantages of PC and individualize treatment plans.
Back to 2023 Abstracts