Authors: Parul Tyagi, Dr. Brij Mohan Singh
Abstract: Kidney disease is a growing global health challenge requiring early, accurate, and automated diagnostic solutions. This paper introduces KidneyNet, a deep learning framework designed for automated kidney disease detection and classification from Computed Tomography (CT) scan images. KidneyNet leverages the power of transfer learning through ResNet50, enhanced with custom classification layers and advanced data augmentation strategies, to classify kidney CT images into four categories: cyst, normal, stone, and tumor. The proposed system is compared against two baseline architectures — Artificial Neural Network (ANN) and Convolutional Neural Network (CNN) — using a publicly available dataset of 12,446 kidney CT images. Experimental results demonstrate that KidneyNet (ResNet50) achieves superior performance with an accuracy of 92%, precision of 91.44%, recall of 92%, and an F1-score of 91.72%, outperforming both ANN (86% accuracy) and CNN (89% accuracy). These findings confirm the effectiveness of deep residual transfer learning as a reliable computer-aided diagnostic tool for kidney disease classification.
