Moving from Manual to Automated Sleep Staging with Federated Learning

This study explored Federated Learning (FL) for automated sleep stage classification, addressing challenges in traditional manual scoring and privacy-sensitive data sharing. By implementing FL on the ODIN platform, an artificial intelligence model was trained using distributed patient data consisting of 148 sleep recordings from 117 patients. Model validation achieved an accuracy of 74.2%, which is comparable to manual scoring and centralized training methods.Clinical Relevance- The proposed FL framework demonstrates a scalable and privacy-compliant solution for sleep stage classification, reducing the workload of clinicians and enabling broader collaboration across institutions. Its integration into a user-friendly inference engine facilitates immediate clinical application for diagnosing sleep disorders.