Predictive modeling of pregnancy outcomes utilizing multiple machine learning techniques for in vitro...

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Biostatistics

Background:

• In vitro fertilization and embryo transfer (IVF-ET) is a key assisted reproductive technology.
• Predicting pregnancy outcomes in IVF-ET remains challenging, impacting treatment efficacy and patient counseling.
• Accurate prediction models are needed to optimize IVF-ET procedures and improve success rates.

Purpose of the Study:

• To investigate factors influencing pregnancy outcomes in IVF-ET.
• To construct and validate predictive models for IVF-ET success.
• To identify high-accuracy models for potential clinical implementation.

Main Methods:

• Utilized clinical data from 2625 women undergoing IVF-ET (2016-2022).
• Developed and compared predictive models (e.g., XGBoost, LightGBM) for clinical pregnancy and live birth.
• Employed ROC curve analysis and AUC calculation to assess model performance.

Main Results:

• The XGBoost model achieved an AUC of 0.999 for predicting pregnancy.
• The LightGBM model demonstrated an AUC of 0.913 for predicting live births.
• Both models showed high accuracy in predicting IVF-ET outcomes.

Conclusions:

• XGBoost model accurately predicts pregnancy likelihood in IVF-ET (0.999 AUC).
• LightGBM model effectively predicts live birth possibility in IVF-ET (0.913 AUC).
• These AI-driven models offer valuable tools for enhancing IVF-ET clinical practice.