Hydronephrosis Detection

This study developed deep learning models to automatically detect hydronephrosis from ultrasound images, aiming to reduce misdiagnosis of renal colic. Using a dataset of 523 kidney images for training and 200 for testing, AlexNet achieved the highest validation accuracy of 98.5%, while segmentation networks (deeplabv3_resnet50/101) reached dice scores around 94%. The results demonstrate that automated ultrasound analysis can accurately support kidney disease diagnosis and has potential for clinical use in managing acute renal failure.