Fast and peer-to-peer vital signal learning system for cloud-based healthcare

Abstract

Wearable devices in the Internet of Things (IoT) make home-based personal healthcare systems popular and affordable. With an increasing number of patients, such healthcare systems are challenged to store and process enormous volumes of data. Some medical institutions employ Cloud services to meet requirements of analyzing big data without considering sharing their own knowledge which could increase diagnostic accuracy. In order to obtain such collaborative healthcare community in the Cloud environment, we propose a peer-to-peer (p2p) learning system which is fast, robust and learning-efficient. Our proposed system continuously collects vital biosignals from wearable devices of users (e.g., chronic patients living alone at home) and analyzes the biosignals in real-time with Extreme Learning Machine (ELM). The traditional centralized learning models suffer in having huge communication costs to share massive amounts of personal vital biosignal data among the institutions for the training purpose. Our proposed p2p learning model can overcome this limitation by allowing every institution to maintain its own raw data while also being updated by other institutions’ shared knowledge a.k.a semi-model which is lightweight output during the training process, as well as being smaller than raw data. The extensive experimental analysis demonstrates that our proposed p2p learning model is efficient in learning and sharing for patient diagnosis. We also show the potential impact under different network topologies, network sizes and the number of learning peers.