Federated Learning Lifecycle Management for Distributed Medical Artificial Intelligence Applications: A Case Study on Post-Transcatheter Aortic Valve Replacement Complication Prediction Solution

The evolution of artificial intelligence (AI) has unveiled considerable prospects for delivering efficacious solutions in the medical domain. Nevertheless, existing legal frameworks and concerns regarding data privacy associated with medical information impose substantial constraints on implementing AI solutions in this domain. Federated learning is a paradigm that enables the training of machine learning models in a decentralized manner without transferring data to a central repository, allowing model development while preserving data privacy across medical and other industries. This study provided a comprehensive framework for applying federated learning to AI solutions in the medical domain. It advocates a sustainable learning ecosystem by overseeing federated learning servers and clients and evaluating performance by managing the federated learning lifecycle. To enhance its practical relevance, this framework includes a detailed process for continuous lifecycle management, involving model deployment, aggregation, testing, evaluation, versioning, and real-time monitoring through the FedOps platform, supporting a sustainable solution. In this study, the feasibility of the proposed methodology was verified using a post-transcatheter aortic valve replacement (TAVR) complication–prediction framework. The performance of the solution after transitioning to a federated learning approach was compared with that of an existing centralized solution. The findings indicated no statistically significant difference in performance between the two methodologies. This implies that federated learning can augment data usability and facilitate the integration of AI technologies into the medical domain, where the preservation of data privacy is critically important.