Machine learning analysis of CT images for the prediction of extracorporeal shock wave lithotripsy efficacy

Abstract

Objective The study aimed to evaluate the use decision support analysis for the prediction of extracorporeal shock wave lithotripsy (ESWL) efficacy and to analyze the factors influencing outcomes in patients who underwent ESWL using machine learning (ML) methods. Methods This retrospective study analyzed the clinical data, including preoperative CT images, of 302 patients who received a single ESWL session treatment for urinary tract stone (UTS) between May and October 2022 in the Department of Urology. The data was preprocessed and incorporated into an ML model, and the dataset was validated at a ratio of 4:1. The AUC and the confusion matrix were used to evaluate the predictive efficacy of the model. Results The CT image-based ML model predicting ESWL efficacy for UTS removal achieved an AUC of 0.86, precision of 88.33%, F1 score of 86.57%, sensitivity of 82.86%, and specificity of 88.89%. The model showed increased predictive accuracty for stones in different locations, with an AUC of 0.95 for kidney stones, 95.45% precision, 96% F1 score, 100% sensitivity, and 90% specificity. The AUC value for upper ureteral stones was 0.89, with 89.14% precision, 88.05% F1 score, 83.33% sensitivity, and 94.51% specificity, while that for mid-ureteral stones was 0.85, with 82.93% precision, 84.09% F1 score, 74% sensitivity, and 96.88% specificity, and the success rate of ESWL for lower ureteral stones was 100%, with an AUC of 1. Conclusions ML analysis was used to predict outcomes following ESWL treatment for UTS. The ML-based model was found to be approximately 86% accurate. The use of ML algorithms can provide matched insights to domain knowledge on effective and influential factors for the prediction of ESWL outcomes.