Advances in Kidney Disease Classification: Medical Imaging, Deep Learning, and Performance Comparison and Challenges

Abstract

This study presents an overview of the progress of kidney disease classification using deep learning and medical imaging techniques. Given the increasing prevalence of kidney-related ailments such as chronic kidney disease (CKD), renal tumors, and nephrolithiasis, there is a critical need for accurate and automated diagnostic tools. This work systematically examines existing literature focusing on various deep learning models, including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and hybrid architectures applied to kidney disease classification. The study further explored the use of 3 datasets like the KiTS (Kidney Tumor Segmentation), the Chronic Kidney Disease Dataset from the UCI Machine Learning Repository, and the Kidney Lesion Image Collection from The Cancer Imaging Archive. The review explores into the preprocessing techniques, model architectures, and evaluation metrics used across different studies, highlighting the advances in image segmentation, feature extraction, and classification accuracy. Key challenges, such as the heterogeneity of kidney diseases, the need for good and generalizable models, and the impact of limited labeled datasets, are highlighted. The review concludes by presenting Performance Analysis of Deep Learning models in recent years, emphasizing the significance of deep learning in enhancing kidney disease diagnosis, achieving high accuracy rates in the classification of complex kidney conditions, and ultimately improving patient outcomes.