Wearable Sensors for Precision Medicine Through Personalized, Holistic and Context-Aware Analytics
Abstract
Recently, a growing percentage of healthcare takes place outside the traditional walls of clinical care and is tightly coupled with daily experiences. The field of remote and decentralized patient care and precision medicine, catalyzed by the COVID-19 pandemic, benefits from recent developments of wearable sensors and internet of things. Although these devices cannot replace clinical diagnosis, they are capable of complementing clinical care by predicting the onset of disorders that could trigger medical tests and assessing effectiveness of therapeutics. This research builds a suite of algorithms to facilitate deployment of remote health monitoring and precision medicine with the objective of disorder prediction and therapeutic assessment with wearable sensors in day-to-day life. To accomplish these objectives, we need to track minor changes in continuous physiological and behavioral data collected by wearables. However, identifying physiological changes that are not the result of external stimuli, such as daily activities, is challenging due to the imperfection of physiological sensing with wearables in uncontrolled environments. Hence, there is a need for identifying surrounding contexts, e.g., activities, to enable apple-to-apple comparison of physiological parameters. Moreover, inter-subject variabilities and personal baselines should be considered in the process of tracking changes in physiological parameters. Finally, large-scale data collection and processing with wearables is required to assess their effectiveness in real-world applications.
In this research, the problem of context-aware sensing is specifically addressed in the field of activity recognition. The proposed novel extension to existing motion-based methods enables understanding of users’ environment through freely available nearables. Experimental results show that leveraging contextual information improves the detection of complex activities that are challenging to be detected by merely motion sensors. A personalization framework is also designed for activity recognition models with a novel uncertainty quantification algorithm to maximize personalization performance while minimizing users’ burden. Lastly, to investigate feasibility of using wearables for disease monitoring, a large-scale real-world study with smartwatches and smart rings was conducted. A novel machine learning model is designed to identify pre-symptoms of the disease before diagnosis. The findings validate the potential of wearable sensing for continuous health monitoring in day-to-day life.
Citation
Akbari, Ali (2021). Wearable Sensors for Precision Medicine Through Personalized, Holistic and Context-Aware Analytics. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /196305.