Development and validation of machine learning models for predicting blastocyst yield in IVF cycles

Area of Science:

• Reproductive Medicine
• In Vitro Fertilization (IVF)
• Machine Learning in Healthcare

Background:

• Predicting blastocyst formation is crucial for IVF clinical decisions, yet quantitative prediction of blastocyst yield remains challenging.
• Existing research often focuses on the presence of any blastocyst, not the quantity, limiting precise decision-making for extended embryo culture.

Purpose of the Study:

• To develop and validate a quantitative predictive tool for estimating blastocyst yields in IVF cycles.
• To compare the performance of machine learning models against traditional regression for blastocyst prediction.
• To identify key predictors influencing blastocyst formation for improved clinical insights.

Main Methods:

• Employed and compared Support Vector Machine (SVM), LightGBM, and XGBoost machine learning models against linear regression.
• Validated model performance using R-squared and Mean Absolute Error for regression tasks.
• Assessed discriminative performance using multi-classification accuracy and kappa coefficients, stratifying predictions and actual yields.

Main Results:

• Machine learning models (R²: 0.673-0.676) outperformed linear regression (R²: 0.587).
• LightGBM was selected as the optimal model due to its efficiency (8 features) and interpretability.
• The model achieved robust accuracy (0.675-0.71) in multi-classification tasks, with key predictors including extended culture embryos, Day 3 mean cell number, and proportion of 8-cell embryos.

Conclusions:

• Machine learning, particularly LightGBM, offers a powerful and interpretable tool for predicting blastocyst yields in IVF.
• The developed model provides valuable quantitative insights to support individualized clinical decisions regarding extended embryo culture.
• Identifying key predictors enhances understanding of factors influencing blastocyst development, paving the way for optimized IVF protocols.