Establishing a model for predicting the cases requiring surgery in congenital hydronephrosis with artificial intelligence algorithms

Congenital hydronephrosis is the most common urinary system pathology in the antenatal period and requires close follow-up in the postnatal period. In this study, it is aimed to develop a machine learning method for estimating the necessity of surgery in cases that underwent upper urinary system and lower urinary system surgery by using Artificial Intelligence (AI) algorithms. The data of the patients who had pyeloplasty due to ureteropelvic junction obstruction (group 1, n=129), a surgical intervention performed due to lower urinary system pathologies (group 2, n=271) and follow up without surgical intervention due to hydronephrosis (group 3, n=96) were included into the study contacted by department of Pediatric Surgery between June 2012 and January 2021 in Eskişehir Osmangazi University Faculty Of Medicine. Data regarding demographic characteristics, clinical findings, radiological and scintigraphy examinations of the patients were evaluated. Two separate estimation sets were created to compare group 1 and 3 to predict the need for surgery in the upper urinary system, and to compare group 2 and 3 for the lower urinary system. Cases requiring surgery in the upper and lower urinary tracts were predicted using the Machine Learning Method (MLM). According to the results, it was found that MLM's Random Forest Classifier algorithm was the most appropriate algorithm in estimating with 97.83% accuracy in upper urinary tract pathologies and 92.11% in lower urinary tract pathologies. In the future AI studies, it was thought that it would be possible to make more accurate predictions by enlarging the data set and including clinical and other evaluation criteria in studies evaluating the need of surgery or the other related topics.