A novel machine-learning framework based on early embryo morphokinetics identifies a feature signature associated...

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Embryology

Background:

• Time-lapse technology generates extensive morphodynamic data in embryology.
• Machine Learning (ML) offers objective predictive models for human embryo assessment.
• Developing automated tools is crucial for standardizing embryo evaluation.

Purpose of the Study:

• To create a novel ML-based strategy for predicting blastocyst development on day 5.
• To identify key patterns in embryo morphokinetics linked to successful blastocyst formation.
• To develop an objective tool for early embryo developmental assessment.

Main Methods:

• Retrospective analysis of 575 embryos from 80 IVF patients.
• Embryo morphokinetics recorded using the Geri plus® time-lapse system.
• Combined 30 clinical, morphological, and morphokinetic variables for analysis.

Main Results:

• Developed the EmbryoMLSelection framework through four steps: Feature Selection, Rules Extraction, Rules Selection, and Rules Evaluation.
• Selected six rules comprising 8 variables with high predictive power.
• Achieved an Area Under the Curve (AUC) of 0.84 and 81% accuracy.

Conclusions:

• A novel feature-signature effectively identifies embryos with high developmental competence for day 5 blastocyst formation.
• The developed tool provides clear, clinically relevant cut-offs for objective early embryo assessment.
• This ML-based approach enhances the prediction of embryo developmental potential in IVF.