Dept. of Biomedical Engineering
Oregon Health & Science University
A rapidly growing proportion of older adults in the populations of the US, EU, and Japan, combined with an increasing prevalence of conditions associated with aging, and further exacerbated by behavioral issues, comprise significant challenges to healthcare delivery. The development of cost-effective, proactive, and preventive approaches, focused on quality-of-life is therefore of utmost societal importance. Replacing institution-centered (clinic-centered) reactive approaches with user-centered sensor and computer-aided care is emerging as a potential solution that would allow early detection and intervention through continuous monitoring and assessment. In contrast to well controlled, in-clinic measurements, however, the context of the user-centered behavioral observations is typically unknown, and the quantities measured by the sensors are usually remote from the quantities of interest. These aspects pose significant challenges for the development of robust algorithms. The only way to mitigate these impediments is the development of computational models that relate the observable quantities to meaningful behavioral metrics. The focus of this thesis is the development of several algorithmic techniques based on computational models of behaviors. These serve as examples of approaches that would enable inference of the patient state from unobtrusive, but continuous monitoring of everyday, real-life behaviors. I examine three techniques for performing in-home monitoring of cognitive performance. The first technique monitors an older adult’s walking speed unobtrusively in the home using motion sensors placed on the ceiling or wall. The second technique monitors the older adult’s cognitive performance by observing how the older adult plays a specially designed computer game. The third technique monitors the older adult’s cognitive performance by observing how the older adult uses a computer in the course of everyday, real-life computer usage. In the case of the first technique, the walking speed serves as immediately behavioral metric, while in the cases of the second and third techniques, we define the behavioral metrics as the parameters of computational models of the behavior of playing the game or using the computer, respectively. The model-based inferences in the second and third techniques characterize the older adults’ ability to utilize the component cognitive functions in order to carry out the task of playing the game or using the computer. The computational models that are used in the second and third techniques provide descriptions of how the behaviors are carried out physically, and the associated behavioral metrics characterize the older adults’ physical performance of specific aspects of the behavior being measured. In this thesis, I demonstrate that the unobtrusive performance measurements combined with the computational model can provide estimates of cognitive functionality similar to that obtained in the controlled environment of a clinical assessment using standard neuropsychological tests – the walking speed, and the Trail-Making Test. In addition, I argue that the observations of older adults playing the computer game and going about everyday, real-life computer usage support a computational model of the Trail-Making Test in which set-switching is performed as a dual-task with movement during Trail-Making Test Part B.
School of Medicine
Hagler, Stuart, "Computational modeling of cognitive processes for continuous in-home assessment of cognitive performance" (2014). Scholar Archive. 3559.