Leveraging federated learning for boosting data privacy and performance in IVF embryo selection

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Data Science

Background:

• In vitro fertilization (IVF) relies on accurate embryo evaluation for successful outcomes.
• Traditional data analysis methods face challenges with data privacy and distributed datasets.
• Federated learning offers a solution for collaborative model training without centralizing sensitive patient data.

Purpose of the Study:

• To evaluate the effectiveness of federated learning (FL) for embryo evaluation tasks in IVF.
• To assess the impact of FL on data privacy and security in multi-institutional IVF studies.
• To compare the performance of FL models against traditional models in predicting embryo ploidy and clinical pregnancy.

Main Methods:

• Retrospective cohort analysis using two large datasets: ploidy status (10,065 embryos) and clinical pregnancy (4,495 embryos).
• Data from multiple hospitals (5 for ploidy, 4 for pregnancy) were utilized.
• Federated learning and gradient boosting decision tree algorithms were employed for model development.

Main Results:

• Federated learning models showed an average increase of 2.5% in Area Under the ROC Curve (AUC) for ploidy status prediction across 5 hospitals.
• For clinical pregnancy prediction, FL models demonstrated an average AUC improvement of 3.08% across 4 hospitals.
• These results indicate enhanced predictive performance when leveraging multi-institutional data via FL.

Conclusions:

• Federated learning effectively improves embryo selection task performance in IVF by enabling secure, multi-source data utilization.
• FL enhances data privacy and security, crucial for sensitive reproductive health data.
• The study highlights the significant potential of federated learning for future applications in assisted reproductive technologies and embryo evaluation.