Rédiger un PREreview

Background/Objectives: Wilms tumor is the most common pediatric renal malignancy, and delayed or inaccurate diagnosis can significantly affect clinical outcomes. This study aimed to evaluate whether integrating traditional machine-learning and deep-learning models with computed tomography (CT) imaging could improve the accuracy of Wilms tumor detection. Methods: A large CT image dataset consisting of 18,205 kidney scans, including both normal and Wilms tumor cases, collected from publicly available medical sources. Images were preprocessed and resized to standardized dimensions before model training. Four supervised learning approaches: ResNet50, VGG16, XGBoost, and Random Forest, were developed and evaluated. The dataset was split into training (14,055 images) and independent testing (4,150 images) subsets. Model performance was assessed using accuracy, precision, recall, F1-score, and confusion matrix analysis. Results: Among the evaluated models, VGG16 demonstrated superior performance, achieving an accuracy of 99.98%, precision of 99.92%, recall of 100%, and an F1-score of 99.96%, indicating excellent sensitivity and overall classification reliability. The remaining models also performed robustly, with accuracies exceeding 94% and recall values above 90%. Conclusions: These findings suggest that deep-learning-based image classification, particularly using VGG16, can substantially enhance non-invasive detection of Wilms tumor from CT scans. The proposed approach has the potential to support clinical decision-making, reduce diagnostic delays, and improve early detection in pediatric oncology settings.