[Evaluation of the possibility of using neural networks for automatic diagnostics of obstructive urination].

Introduction: Obstructive type of urination requires accurate and timely diagnosis to prevent complications and improve the quality of life of patients. Traditional diagnostic methods such as uroflowmetry, although they remain the standard, have their limitations. In this context, videography of urine stream followed by image analysis is a more cost-effective and promising approach that allows for a more detailed picture of urination, accessible not only to urologists, but also to patients.

Aim: To establish the possibility of recognizing and classifying graphs of the urination using neural network and machine learning technologies.

Materials and methods: This retrospective study involved 152 male patients aged 19 to 87 years who underwent examination and treatment at the MC Medassist clinic from June 2024 to January 2025. There were 43 patients (28%) with obstructive type of urination, 39 patients with benign prostatic hyperplasia, 4 patients with urethral stricture, and 109 patients (72%) with normal urination. The diagnostics algorithm included a general urinalysis, kidney and bladder ultrasound and/or MRI of the prostate, as well as uroflowmetry. The neural network architecture was designed based on the Keras framework of the Python programming language.

Results: Three studies with obtained data were carried out, which differed in the architecture of the neural network and the methods of preparing the initial data. The average area under the ROC curve for a network with random image feed averaged 0.5 for both the training and test samples. For a network with linear feed of the entire data set, it was 1 for the training and test samples. A neural network with three inputs differing in two-threshold binarization ranges showed a result of 0.9 for the training and 0.7 for the test sample.

Discussion: An important aspect of this study is the possibility of using neural networks to process large amounts of video data. Automating the analysis of images of urine stream allows not only to reduce the time required for diagnosis, but also to identify hidden patterns that may be overlooked during visual assessment by a specialist. The study of methods for analyzing video recordings of urination based on artificial neural network technologies and machine learning algorithms can become the basis for creating new diagnostic tools that will increase the speed of diagnosis, accelerate drug research, and monitor patients with chronic diseases.

Conclusion: Despite the current limitations, this study confirms that the use of neural networks and machine learning in urology has significant potential and can become the basis for the development of new diagnostic tools that can improve the efficiency of medical care, and thereby improve the quality of life of patients.