Trusted federated learning for Internet of Medical Things: solutions and challenges

The clinical adoption of state-of-the-art artificial intelligence (AI) applications for disease diagnosis with digital healthcare technologies will ease the burden on medical practitioners, ensure timely interventions, and aid patients’ welfare. The digital innovations in medical data acquisition have made it possible to collect multi-modal data through Internet of Medical Things (IoMT) but the privacy-preserving nature and compliance with medical data regulations often preclude data sharing across organizational and geographical boundaries. An AI model requires high quality and large-scale training datasets collected from diverse sources to mitigate bias and aid better prediction accuracy of the unseen data. With secure and trusted federated learning, the data of an organization can stay local and yet contribute to AI models’ training for building better trained models with improved accuracy and generalization. Thus, the trusted federated learning approaches facilitate sharing the trained AI models across different participating healthcare organizations by breaking down barriers, increasing trust, and preserving privacy for better disease predictions and diagnoses that can increase deployment of AI models in clinical practice. In this chapter, we provide the current state-of-the-art solutions, adoption challenges, and future research directions, and a framework for trusted federated learning in healthcare applications.