Enhancing predictive models for egg donation: time to blastocyst hatching and machine learning insights.

Area of Science:

• Reproductive Medicine
• Data Science in Healthcare
• Embryology

Background:

• Artificial intelligence and data science are increasingly utilized in assisted reproduction.
• Time-lapse technology incubators provide valuable data for embryo assessment.
• Improving in vitro fertilization (IVF) clinical outcomes is a key goal.

Purpose of the Study:

• To compare and identify the most predictive machine learning algorithms for embryo implantation.
• To analyze morphokinetic and morphological variables using a known implantation database.
• To recognize the most predictive embryo parameters for enhancing IVF treatment success.

Main Methods:

• A multicenter retrospective cohort study involving 378 egg donor recipients undergoing fresh single embryo transfer.
• Embryos were cultured in Geri® time-lapse incubators and analyzed for morphokinetic events.
• Ten machine learning algorithms, including Random Forest and AdaBoost, were applied and optimized.

Main Results:

• Random Forest showed the highest predictive power for implantation (AUC=0.725).
• AdaBoost classification trees were most predictive for live birth (AUC=0.749).
• Time to hatching and other developmental timings (pronuclei, cleavage, compaction) were significant predictors.

Conclusions:

• Random Forest and AdaBoost are effective machine learning models for predicting implantation and live birth in egg donation programs.
• Time to blastocyst hatching is a crucial parameter for predictive models.
• Embryonic developmental processes like syngamy, genomic imprinting, and compaction are vital for successful implantation and live birth.