Adaptive Algorithms for Personalized Health Monitoring
Engelhard, Matthew, Systems Engineering - School of Engineering and Applied Science, University of Virginia
Patek, Stephen, Department of Systems and Information Engineering, University of Virginia
Health monitoring has entered the era of precision medicine. With support from President Obama's Precision Medicine Initiative, the National Institutes of Health have renewed their focus on prevention, management, and treatment strategies that are tailored to individual patients. The initiative relies heavily on genomics and bioinformatics, but clinical informatics and mobile health technologies have been consistently emphasized, including the real-time monitoring of physiologic data. A patient's heart rate can now be measured with a wristband and transmitted wirelessly to their clinician; skin impedance can be sampled every second to develop a personalized stress profile. As illustrated by these examples, the most distinctive feature of monitoring is the repeated, sequential collection of physiologic measurements by a computing platform. Consequently, monitoring systems have capabilities not found elsewhere in precision medicine: they can interact with the patient and adapt in real time.
In this work, we develop monitoring algorithms uniquely suited for health care applications. They are designed for different monitoring scenarios with distinct data types, but each one adapts its outputs or decisions to a growing history of observations. Our central hypothesis is that health monitoring benefits immensely from an adaptive approach -- more so than other monitoring applications -- due to fundamental differences between engineered and biological systems. These differences pertain not only to variability between persons, but also to our knowledge of physiology, our access to salient system parameters, and the importance of subjective experiences in health care.
Many of our objectives have been motivated by a target application, walking ability in multiple sclerosis (MS). MS is a disease of the central nervous system which can produce almost any neurological sign or symptom, making adaptation and personalization all the more critical. After presenting a case study in MS, we formulate adaptive algorithms for three common health monitoring scenarios. The first is a physiologic signal monitoring algorithm designed for signals that vary substantially between persons or over time. The second, adaptive symptom reporting, personalizes its queries to accurately track disability while reducing the burden placed on the patient. The third, active health event identification, learns to classify events from the patient's perspective by requesting event labels at opportune times. Each algorithm is validated with data from persons with MS.
PHD (Doctor of Philosophy)
adaptive algorithms, personalized algorithms, health monitoring, multiple sclerosis, machine learning, artificial intelligence, gait analysis, hidden Markov model