CELL Ultracomputing Platform for Metabolic and Cardiovascular Health Monitoring using Wearables

Current processing systems are limited in their ability to provide holistic view of personalized health from wrist wearables and smart clothing, embedded with a mix of low and high fidelity - optical, MEMS, RF and other electronic biosensors. The mathematics of combining data from disparate sensors with asynchronous arrival events, episodic and continuous sampling, and varying performance due to confounders and physiological variability injects unique challenges in building predictive models and providing actionable, engaging information over long duration to the users. As wearables become more sophisticated, with new analytics added on a regular basis, the information processing challenge becomes harder. Our ultra-scalable CELL analytical platform is inspired by the scale of cellular networks found in nature - distributed information pathways and triggers in living organisms and creatures. CELL is a virtual machine implementing the data-flow architecture where “the network is the computer”. It integrates the communication and processing resources into a coherent model where computing is generally distributed between the wearable, smart phone and cloud. The paper will describe our CELL architecture in detail and discuss departures from traditional and emerging processing containers (such as Confluent, Kubernetes etc.) applied to analyzing health data from wearables. Our implementation and results using the CELL will be presented from multi-sensor devices with metabolic, cardiovascular and impedance biosensor examples for wellness and clinically relevant metrics; including implications for scalability and real-time fused analytics that combine data from multiple sensors.