Using feature optimization and LightGBM algorithm to predict the clinical pregnancy outcomes after in vitro...

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Biostatistics

Background:

• Infertility affects approximately 17.5% of adults globally, necessitating improved prediction methods for assisted reproductive technologies.
• Machine learning (ML) models are increasingly explored to enhance the prediction of clinical pregnancy outcomes in in vitro fertilization (IVF) cycles.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting pregnancy outcomes following IVF.
• To provide a tool that assists clinicians in patient counseling and treatment strategy adjustment.

Main Methods:

• A retrospective analysis of data from Chinese reproductive centers (March 2020-March 2021) was performed.
• Six ML algorithms (XGBoost, LightGBM, KNN, Naïve Bayes, Random Forest, Decision Tree) were applied using 13 selected features.
• Model performance was assessed using precision, recall, F1-score, accuracy, and AUC via five-fold cross-validation.

Main Results:

• The LightGBM model demonstrated superior performance with 92.31% accuracy, 87.80% recall, 90.00% F1-score, and 90.41% AUC.
• Key predictive features identified were estrogen concentration at HCG injection (etwo), endometrium thickness (EM TNK), years of infertility (Years), and body mass index (BMI).

Conclusions:

• The LightGBM model offers the best predictive capability for IVF pregnancy outcomes.
• Estrogen concentration (etwo) emerged as the most significant predictor of successful IVF.
• This ML approach can support fertility specialists in optimizing patient counseling and treatment plans.