Federated task-adaptive learning for personalized selection of human IVF-derived embryos

Area of Science:

• Reproductive Medicine
• Artificial Intelligence
• Data Privacy

Background:

• In-vitro fertilization (IVF) success relies on manual embryo assessment, which is time-consuming and labor-intensive.
• Artificial intelligence (AI) offers automated assessment but faces privacy challenges due to centralized data training.
• Developing privacy-preserving AI for IVF is crucial for improving treatment efficacy and patient confidentiality.

Purpose of the Study:

• To develop a distributed AI system, FedEmbryo, for personalized embryo selection in IVF.
• To address data privacy concerns associated with centralized AI training in fertility treatments.
• To enhance the accuracy and efficiency of embryo assessment for improved IVF outcomes.

Main Methods:

• Developed FedEmbryo, a distributed AI system utilizing federated learning (FL) for privacy-preserving training.
• Introduced a federated task-adaptive learning (FTAL) approach with hierarchical dynamic weighting adaptation (HDWA).
• Integrated multitask learning (MTL) within FL to accommodate diverse learning tasks across clinical sites.

Main Results:

• FedEmbryo demonstrated superior performance in morphological valuation of embryos compared to local models.
• The system achieved higher accuracy in predicting live-birth outcomes in various IVF scenarios.
• FedEmbryo outperformed state-of-the-art federated learning methods in experimental evaluations.

Conclusions:

• FedEmbryo offers a privacy-preserving, decentralized AI solution to enhance IVF clinical decision-making.
• The system excels at capturing stage-specific embryonic features and predicting key IVF outcomes.
• FedEmbryo represents a practical advancement for improving IVF success rates through AI.