A pilot study towards a smart-health framework to collect and analyze biomarkers with low-cost and flexible wearables at a smart and connected community

Abstract Artificial intelligence-enabled applications on edge devices have the potential to revolutionize disease detection and monitoring with smart health (sHealth) applications. The major challenges are user-friendly and flexible sensors for seamless collection of physiological data for long hours, online and on-device processing of sensitive medical data to facilitate privacy protection, reliable extraction of disease-related biomarkers, and implementation of lightweight artificially intelligent algorithms for inference at the edge without degrading the system performance. In this pilot project, we conducted a yearlong field study with 9 participants conducting 480 data collection sessions in the “living lab” environment. We used smartphones as the edge computing device and implemented pre-trained machine learning algorithms in the smartphone app for computing disease-related Events-of-Interest (EoI). We considered real-time data processing on the smartphone itself without sharing raw data with the cloud or any other computing facility to minimize privacy concerns, and network bandwidth requirements. We used a commercial smart band and a custom-designed zero-power inkjet-printed sensor for physiological sensing and capturing health biomarkers such as heart rate variability (HRV) and core body temperature. The extracted HRV feature values are within the 95% confidence interval of normative values. On top of that, the extracted HRV that shows some informative trends i.e. hammock pattern for healthy subjects which may be helpful in subsequent research studies. Moreover, we used core body temperature with user-reported outcomes for estimating flu-related symptoms severity and visualizing the spatiotemporal trend in a cloud-server to facilitate personalized as well as community-wide health monitoring. Inference at the edge provided a data reduction of 3 order while the runtime latency, power consumption, memory requirement, and storage size of the smartphone app were 500 ms, 51.90 mAH, 9.4 MB, and 2.4 MB, respectively. Our developed framework of sHealth enables automated community-wide monitoring of symptoms severity in addition to personalized monitoring which paves the way for early monitoring of a disease outbreak for a smart and connected community.