Machine Learning-Based Prediction of Urolithiasis Recurrence Using Patient’S Clinical Data, Demography, and Ct Findings Publisher Pubmed



Homayoun H ; Mousavirad SJ ; Baghdadabad LZ ; Bahri RA ; Oskouie IM ; Mohammadi A ; Aghamir SMK
Source: Urology Journal Published:2026

Abstract

Purpose: Urolithiasis is the condition of forming stones inside urinary tract with diverse shape, size, and location. The sooner urolithiasis is diagnosed, the easier it is to treat and prevent complication. This study aims to propose a method for predicting urolithiasis recurrence based on machine learning methods. Materials and Methods: The proposed method uses clinical data, demographics, and CT findings of 4246 patients who were referred to the clinic once or multiple times within three years. The proposed method has three main phases of data engineering and pre-processing, machine learning prediction model development, and performance evaluation. In addition, the performance of six machine learning-based classifiers is evaluated by performance metric calculation, ROC curve analysis, calibration analysis, and decision curve analysis. Results: The results of 10 independent repeats of the proposed method using a train/test split evaluation strategy reveal that the best-performing classifier is random forest with the area under the ROC curve, sensitivity, and positive predictive value of 0.64, 0.87, and 0.84, respectively. On the other hand, k-fold cross-validation: A comma is needed after hand and before k-fold evaluation strategy reveals that the best-performing classifier again is RF, with the area under the ROC curve, sensitivity, and positive predictive value of 0.63, 0.90, and 0.83, respectively. Moreover, the brier score of 0.18 shows that this classifier is well-calibrated among other evaluated classifiers. Conclusion: This study presents a practical application of predictive machine learning methods for predicting urolithiasis recurrence with clinically acceptable accuracy compared to traditional scoring systems. To select the best classifier, six different predictive ML models have been evaluated using different performance metrics and analysis tools. © 2025 Urology and Nephrology Research Centre.All rights reserved