A machine learning approach to detect renal calculi by studying the physical characteristics of urine

Abstract

Renal calculi (also called kidney stones) are hard deposits made of minerals and salts that form inside the kidneys. A kidney stone is usually asymptomatic until it moves around within the kidney or passes into one of the ureters, after which one may experience a sharp pain similar to a muscle strain. The current methods of detecting kidney stones (like CT imaging) cannot be employed until the onset of the first symptoms. Our approach provides a solution to detect kidney stones before the appearance of any symptoms. Using existing data on the physical characteristics of urine, we trained a machine learning model using various classifiers – logistic regression, decision tree and Support Vector Machine – to detect kidney stones and calculate their accuracies. Using one-way ANOVA test to compare the accuracies yielded by each approach of these classifiers, we found that there was no statistically significant difference between using one classifier over another. We plotted a confusion matrix and calculated the F1 scores for each classifier, in order to evaluate the performance of the algorithms. All the classifiers reported an accuracy of >0.80 and an F1 score of >0.85. Thus, our findings suggest that it is possible to detect kidney stones before the onset of symptoms by analyzing the physical characteristics of urine using machine learning classifiers.