Abstract: Heart disease and stroke are becoming the leading causes of death worldwide. Electrocardiography monitoring devices (ECG) are the only tool that helps physicians diagnose cardiac abnormalities. Although the design of ECGs has followed closely the electronics miniaturization evolution over the years, existing wearable ECGs have limited accuracy and rely on external resources to analyze the signals and evaluate heart activity. In this paper, we work towards empowering the wearable device with processing capabilities to locally analyze the signal and identify abnormal behaviour. The ability to differentiate between normal and abnormal heart activity significantly reduces (a) the need to store the signals, (b) the data transmitted to the cloud, (c) the overall power consumption and (d) the confidentiality of private data. Based on this concept, the HEART system presented in this work combines wearable embedded devices, mobile edge devices, and cloud services to provide on-the-spot, reliable, accurate, and instant heart monitoring. The wearable device is remotely trained by a physician to learn to accurately identify critical events related to each particular patient. Following this training session, the wearable device becomes capable of interpreting a large number of heart abnormalities without relying on cloud services and edge resources, when the medical doctor is not present. The Fog computing approach extends the cloud computing paradigm by migrating data-processing closer to the production site, thus accelerating the system’s responsiveness to events. The HEART system’s performance concerning the accuracy of detecting abnormal events and the power consumption of the wearable device is evaluated. Results indicate that a very high success rate can be achieved in terms of event detection ratio and the battery is able to sustain operation up to a full week without the need for a recharge.
Keywords: Internet of things, machine learning, user-centric design, system design, experimental evaluation