Introduction. Choosing the optimal method of surgical treatment of benign prostatic hyperplasia (BPH) remains one of the most difficult clinical tasks in modern urology. The presence of several effective interventions – transurethral resection of the prostate (TURP), transurethral enucleation of the prostate (TUEP) and embolization of the prostatic arteries (EPA) increases the variability of medical decisions and increases the risk of subjective choice.
The purpose of the study. To evaluate the possibility of using a neural network model based on nonlinear logistic regression to support the choice of the optimal method of surgical treatment of BPH at the stage of preoperative examination.
Materials and methods. The study included 105 patients with BPH who were treated in 2019–2025. The patients were divided into three groups depending on the performed intervention: TURP (n=35), TUEP (n=35), EPA (n=35). To build the model, clinical and functional parameters were used: PSA level, uroflowmetry parameters (Qmax, Qav), prostate volume, residual urine volume and anaesthetic risk according to ASA. Based on these data, a neural network model has been developed, trained to predict the preferred method of operation. The quality assessment was carried out using ROC analysis.
Results. The coincidence of the recommendations of the neural network with the expert decision of surgeons was 83.8%. The area under the ROC curve (AUC) reached 0.88, which corresponds to a high level of discrimination ability. The highest accuracy was noted when predicting TURP (91%), slightly lower when choosing TUE (85,7%) and EPA (81.2%). In a number of clinical cases, the recommendations of the model turned out to be safer than the actual intervention performed.
Conclusion. The use of a neural network model based on standard clinical and functional indicators makes it possible to increase the objectivity of the choice of surgical treatment for BPH and can serve as an effective tool for supporting clinical decisions.
