DOI: 10.29188/2222-8543-2026-19-2-36-42
For citation:
Avanesyan I.O., Stroganov R.V., Gritskov I.O., Afanaseva V.S., Gligorich A.S., Vasyutin I.A., Panin A.A., Kasyan G.R., Pushkar D.Yu. The use of machine learning methods in the interpretation of uroflowmetry results. Experimental and Clinical Urology. 2026;19(2):36-42; https://doi.org/10.29188/2222-8543-2026-19-2-36-42
Avanesyan I.O., Stroganov R.V., Gritskov I.O., Afanaseva V.S., Gligorich A.S., Vasyutin I.A., Panin A.A., Kasyan G.R., Pushkar D.Yu.
Information about authors:
- Avanesyan I.O. – student, Russian University of Medicine, Moscow, Russia; RSCI Author ID: 1268300, https://orcid.org/0009-0009-8542-4243
- Stroganov R.V. – PhD, Associate Department of Urology, Russian University of Medicine; Leading Specialist, Research Institute for Healthcare Organization and Medical Management of Moscow Healthcare Department, Urologist, Department of Urology №66, S.P. Botkin City Clinical Hospital, Moscow, Russia; RSCI Author ID: 1074224, https://orcid.org/0000-0002-5529-1787
- Gritskov I.O. – junior researcher, S.P. Botkin City Clinical Hospital, Lecturer, Department of Medical Equipment, Russian Medical Academy of Continuous Professional Education, Moscow, Russia; RSCI Author ID: 1175027, https://orcid.org/0000-0002-4708-1683
- Afanaseva V.S. – student, Russian University of Medicine, Moscow, Russia; https://orcid.org/0009-0005-7160-2062
- Gligorich A.S. – student, Russian University of Medicine, Moscow, Russia; https://orcid.org/0009-0005-9342-6434
- Vasyutin I.A. – Master’s degree student, Department of Industrial Programming, Institute of Advanced Technologies and Industrial Programming, MIREA – Russian Technological University, Moscow, Russia; https://orcid.org/0009-0003-0878-2430
- Panin A.A. – Master’s degree student, Department of Industrial Programming, Institute of Advanced Technologies and Industrial Programming, MIREA – Russian Technological University, Moscow, Russia; https://orcid.org/0009-0003-5745-4142
- Kasyan G.R. – Dr. Sci., Professor, Department of Urology Russian University of Medicine, Moscow, Russia; RSCI Author ID: 686514, https://orcid.org/0000-0001-7919-2217
- Pushkar D.Yu. – MD, Dr. Sci. (Med.), Professor, Academy of the RAS, Head, Department of Urology, Russian University of Medicine, Moscow, Russia; RSCI Author ID: 417122, https://orcid.org/0000-0002-6096-5723
45
Introduction. Uroflowmetry (UFM) is a key noninvasive method for assessing lower urinary tract function; however, its interpretation is subjective and depends on the physician's experience, leading to varying results and complicating the diagnosis of lower urinary tract disorders associated with age, injuries, and neurological diseases.
The aim is to demonstrate the possibility of applying machine learning in the interpretation of UFM, as well as to propose optimized algorithms based on combined data and evaluate them in prospective studies.
Material and methods. Three approaches to time series classification were considered: random forest on numerical features, a one-dimensional convolutional neural network (ResNet-like architecture), and a two-dimensional convolutional network with transformation into a Gram matrix (Inception-like architecture). Training sample: a total of 160 records, test sample – 40 (20%). Five nosologies were classified: normal urination, benign prostatic hyperplasia, urethral stricture, overactive bladder, hyposensory bladder.
Results. For random forest, classification accuracy: for the one-dimensional network – 77.5% and 87.5%, for the two-dimensional – 82.5%. The one-dimensional network demonstrated the greatest advantage in identifying UFM curve patterns. The models had difficulty differentiating benign prostatic hyperplasia and urethral stricture due to the similarity of obstructive patterns.
Conclusion. Three classification methods were evaluated: random forest on numerical features (classification accuracy 77.5%), one-dimensional convolutional ResNet-like neural network (87.5%), and two-dimensional with transformation into a Gram matrix (82.5%), using an Inception-like architecture. A comparative analysis of the methods' effectiveness was conducted, confirming the advantage of one-dimensional convolutional networks in identifying UFM curve patterns, with recommendations for optimization in clinical practice and prospective studies.
