A neural network for predicting occurrence of uterine fibroids in women of reproductive age

Aim : to create a model for predicting emergence of uterine leiomyoma (UL) using neural network analysis of risk factors and to evaluate its prognostic characteristics. Materials and Methods . A retrospective case-control study with 209 patients aged 20–47 years was performed covering the years from 2022 to 2024. Two groups of patients were identified: 1 – 106 women with UL, 2 – 103 patients without UL. Preliminary data processing was carried out, followed by a quantitatively analyzed relationship between risk factors and UL development using neural network analysis. The multilayer perceptron method was used to create a prognostic model for predicting UL emergence. Results . During the study, there were selected 12 model-based factors showing statistically significant inter-group differences: body mass index (BMI), age at menarche, number of abortions and spontaneous abortions, age at first birth, presence of arterial hypertension (AH), benign ovarian tumors, history of in vitro fertilization, level of anti-Müllerian hormone, number of pregnancies, serum cholesterol and glucose levels. The prediction accuracy for the developed model was 92.3 %, sensitivity – 90.6 %, specificity – 94.2 %. The predictive value was confirmed using ROC analysis – the area under the curve was 0.93 (95 % confidence interval = 0.91–0.94; p < 0.001), which proves the promise of this method for clinical practice. Modifiable and potentially modifiable factors included increased BMI, AH, benign ovarian tumors, cholesterol and glucose levels. Such factors are considered as most relevant, due to an opportunity to be directly or indirectly affected, which proves an importance for preventive approach to this disease. Conclusion . The developed model is an effective tool for predicting UL emergence (accuracy 92.3%), the use of which in clinical practice will allow shifting from the established paradigm of radical treatment to a preventive approach.