Using Unlabeled Information of Embryo Siblings from the Same Cohort Cycle to Enhance In Vitro Fertilization...

Area of Science:

• Reproductive medicine and artificial intelligence
• Embryology and machine learning

Background:

• Machine learning for in vitro fertilization (IVF) embryo assessment is advancing reproductive medicine.
• Current IVF implantation prediction models often overlook data from sibling embryos within the same cohort, despite their potential relevance.

Purpose of the Study:

• To investigate the contribution of sibling embryo information to machine learning-based implantation prediction accuracy.
• To determine the extent to which unlabeled sibling embryo data can enhance predictive models for IVF success.

Main Methods:

• Utilized high-content time-lapse embryo imaging data from IVF cohorts.
• Applied machine learning algorithms to analyze features derived from sibling embryos.
• Evaluated the impact of incorporating unlabeled sibling data on implantation prediction performance.

Main Results:

• Implantation outcomes were found to be correlated with attributes derived from sibling embryos.
• The inclusion of unlabeled sibling embryo data significantly boosted the performance of implantation prediction models.
• Specific cohort properties were identified as key drivers of prediction accuracy, particularly in correcting erroneous predictions.

Conclusions:

• Unlabeled data from sibling embryos in IVF cohorts can be a valuable, underutilized resource for improving implantation prediction.
• Leveraging sibling embryo information can reduce inherent noise associated with individual embryos, leading to more robust and accurate predictive models for IVF success.