Motion biomarkers for early detection of dementia-related agitation

Abstract

Agitation in dementia poses a major health risk for both the patients and their caregivers and induces a huge caregiving burden. Early detection of agitation can facilitate timely intervention and prevent escalation of critical episodes. Sensing behavioral patterns for detecting health critical events is a challenging task. Wearable sensors are often employed for sensing physiological signals, but extracting possible biomarkers for confident detection of early agitation is still an open research. In this paper, we employ an ongoing iterative study to explore the motion biomarkers related to agitation in community-dwelling persons with dementia (PWD). This study uses accelerometers in smart watches to capture PWD behavioral patterns unobtrusively. Analysis of the feature space is performed using data from multiple subjects to discriminate among epochs of onset, preset, and offset of agitation while considering inter-person variability in real deployments. This paper shows the prospect of feature space analysis of the motion data for developing early agitation detection models to deploy in the wild.