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

In-vitro fertilization (IVF) provides an effective infertility treatment. However, the success of IVF heavily depends on manual morphological assessment of embryos, a process that is both time-consuming and labor-intensive. While artificial intelligence (AI) enables automated assessment, its reliance on centralized large-scale data training raises privacy concerns.

Here, we develop a distributed AI system, termed 'FedEmbryo', tailored for personalized embryo selection while preserving data privacy. Within FedEmbryo, we introduce a federated task-adaptive learning (FTAL) approach with a hierarchical dynamic weighting adaptation (HDWA) mechanism. The FTAL integrates multitask learning (MTL) with federated learning (FL) by proposing a unified multitask architecture that consists of shared layers and task-specific layers to accommodate the single- and multi-task learning within each client. The HDWA mechanism considers the learning feedback (loss ratios) from the tasks and clients, facilitating a dynamic balance between task attention and client aggregation.

We conduct extensive experiments in different scenarios of the IVF cycle to evaluate the effectiveness of FedEmbryo for personalized embryo selection. The observer study validates that FedEmbryo achieves superior performance in both morphological valuation and prediction of live-birth outcomes compared to the locally trained model, as well as state-of-the-art FL methods.

We present FedEmbryo, an AI-powered system designed to improve IVF outcomes through privacy-preserving, decentralized training across multiple clinical sites. FedEmbryo demonstrates superior performance in capturing stage-specific morphological features of embryos and achieves accurate predictions for key IVF-related tasks. We hope that FedEmbryo will serve as a practical tool for enhancing clinical decision-making in IVF.