Prediction of surgery type for uterine fibroids using machine learning algorithms and hormone values

Area of Science:

• Reproductive Medicine and Surgery
• Artificial Intelligence in Healthcare
• Gynecologic Oncology

Background:

• Uterine fibroids (UFs) affect a significant portion of women, leading to varied surgical interventions like hysterectomy and myomectomy.
• Surgical classification for UFs traditionally relies on clinical judgment, fibroid characteristics, and patient factors.
• The role of female sex hormone profiles in surgical decision-making for UFs is complex and warrants further quantitative investigation.

Purpose of the Study:

• To develop and externally validate machine learning (ML) models for characterizing surgical classification patterns between hysterectomy and myomectomy.
• To assess the utility of combining fibroid characteristics with female sex hormone profiles for predicting surgical outcomes.
• To evaluate the performance of various ML algorithms in classifying surgical decisions based on integrated patient data.

Main Methods:

• A multicenter study involving 600 women with UFs, divided into hysterectomy (60.3%) and myomectomy (39.7%) groups.
• Development of ML models using five classification algorithms (SVM, decision trees, random forests, k-NN, logistic regression) with inputs including fibroid characteristics and female sex hormones.
• External validation of the best-performing model on an independent cohort of 30 cases.

Main Results:

• Hysterectomy patients exhibited significantly higher age, FSH, LH, UF number/volume, uterine volume, disease duration, gravidity, parity, and PRL levels.
• Estradiol and anti-Müllerian hormone levels were significantly lower in the hysterectomy group compared to the myomectomy group.
• Models combining sex hormone profiles and UF characteristics achieved 100% accuracy in 2012 of 2555 combinations; UF number alone reached 96% accuracy. Algorithmic predictions agreed with surgical decisions in 97% of cases.

Conclusions:

• ML models integrating hormone profiles and fibroid characteristics can effectively reproduce established surgical classification patterns for UFs.
• The models demonstrate strong baseline separability driven by age- and menopause-associated hormonal profiles, with consistent external validation.
• These findings support the feasibility of quantitative modeling in clinical decision-making for UF surgery, suggesting potential for broader application after further validation in complex cases.