Use of artificial intelligenceapplications in prediction of success and complications in the surgical treatment of urinary stone disease

Objective: In this study, we aimed to use explainable artificial intelligence methods to predict the treatment success and complications of patients who underwent ureterorenoscopy, retrograde intrarenal surgery and percutaneous nephrolithotomy for urinary system stone disease. Material and Method: This study was approved by Adıyaman University NonInvasive Clinical Research Ethics Committee (2022/3-15). The preoperative, preoperative and postoperative variables of 917 patients who underwent URS, RIRC and PNL for urinary stone disease (kidney and ureteral stones) between January 2018 and January 2022 in the Urology Clinic of Adıyaman University Faculty of Medicine Training and Research Hospital were evaluated retrospectively. The patients were divided into 3 groups according to the type of surgery performed. Separately recorded features for each group were classified with different classification algorithms to both predict success and predict the development of complications. Confusion matrix was used as model performance metric. LIME and SHAP algorithms were used to explain the model. Classifier models were created by using Hold-out and 10-fold crossvalidation methods. The SMOTE algorithm was used to resolve the imbalance in the dataset. Results: XGBoost (%89.13) had the highest accuracy in predicting success in PNL patients, and Random Forest (%93.81) in predicting the development of complications. The LIME algorithm calculated the probability that a patient's surgery would be successful or unsuccessful and whether complications would develop. According to LIME and SHAP algorithms; high stone size, high hydronephrosis degree, male gender, high stone-to-skin distance contribute to failure. High mean platelet volume values contribute to success. According to the same algorithms; pelvis localization of the stone, high Hounsfield Unit value, high stone size, low leukocyte values and high neutrophil values increase the possibility of developing complications. Multi Layer Perceptron (%85.09) had the highest accuracy in predicting success in URS patients, and Random Forest (%96.35) algorithm in predicting the development of complications. The LIME algorithm calculated the probability that a patient's surgery would be successful or unsuccessful and whether complications XVII would develop. According to LIME and SHAP algorithms; proximal ureter localization decreases success, low mean platelet volume values and male gender increase success. High neutrophil/lymphocyte ratio and monocyte/lymphocyte ratio values between 0.2-0.35 increase the possibility of complications, distal ureter localization reduces the possibility of complications. Multi Layer Perceptron (%91.17) algorithms achieved the highest accuracy in predicting success in RIRC patients, and Multi-Layer Perceptron and Random Forest (%97.67) algorithms in predicting the development of complications. The LIME algorithm calculated the probability that a patient's surgery would be successful or unsuccessful and whether complications would develop. According to LIME and SHAP algorithms; nitrite positivity, localization of the stone at the upper calyx and ureteropelvic junction, high stone size, high erythrocyte distribution width and high monocyte count are the factors that negatively affect success. The growth of rare microorganisms in the preoperative urine culture, high preoperative fever, high hydronephrosis degree and high monocyte count increase the possibility of developing complications. Conclusion: Studies on the factors affecting success and complications in urinary system stone disease are lacking. The use of artificial intelligence in this sense allows us to clarify these factors more, predict success and complication development, and thus provide more effective information to the patient about the surgery. Keywords: urinary system stone disease, artificial intelligence, prediction of success, prediction of complication development