Pediatric Renal Ultrasound Abnormalities

This study developed a deep learning–based computer-aided diagnostic model using a ResNet-50 architecture to classify pediatric renal ultrasound images as normal or abnormal. Trained on 1,599 images (330 normal and 1,269 abnormal), the model demonstrated strong performance, achieving an overall accuracy of 92.9% and an AUC of 0.959, with high diagnostic accuracy across multiple abnormal categories including stones, cysts, and hydronephrosis. The results support the feasibility of AI-assisted ultrasound screening to aid early detection of pediatric kidney diseases in clinical practice.