International Journal of Science, Engineering and TechnologyAuthors: Parul Tyagi, Dr. Brij Mohan Singh
Abstract: Kidney stone disease — clinically referred to as nephrolithiasis — remains one of the most painful and widely encountered urological conditions worldwide. Catching it early and getting the diagnosis right can dramatically change a patient's care pathway and reduce the financial strain on healthcare systems. In this study, we take a close, side-by-side look at five machine-learning techniques that have shown promise for automated kidney-stone detection: the Multilayer Perceptron trained with Back Propagation (MLP-BPA), Radial Basis Function (RBF) networks, Learning Vector Quantization (LVQ), Support Vector Machines (SVM), and Deep Convolutional Neural Networks (CNN). All experiments are run on a standardised clinical dataset using WEKA 3.7.5 and Python, with each model assessed on accuracy, sensitivity, specificity, and F1-score. The features fed into every model include creatinine and BUN levels, CT-scan findings, kidney size and contour, and several urinary markers. Among the classical approaches, SVM came out on top with 93.6% accuracy, while MLP-BPA was close behind at — a CNN — pushed accuracy to 96.1% when adequate training images were available. Beyond the raw numbers, we discuss what each architecture actually trades off in practice: how hard it is to train, how transparently it reaches its decisions, and whether a busy nephrology clinic could realistically deploy it. Our hope is that this comparison gives clinicians and AI researchers a clear, honest basis for choosing the right tool for kidney stone diagnosis.
