Stroke is the leading cause of serious, long-term disability in the United States. Every year, approximately 800,000 people experience a new or recurrent stroke. Due to advances in acute neurological care, nearly 85% survive and many live with the long-term chronic effects on motor and cognitive function. After initial recovery and rehabilitation, life for people with stroke occurs outside of a rehabilitation clinic or hospital in day-to-day activities. Those that return to work and the community after a stroke strive to live daily life to the fullest. Several recent studies have demonstrated the benefit of activity and exercise for people with stroke. Those individuals that do not exercise or engage in regular activity are at a 30% increased risk of experiencing a recurrent stroke.
Unfortunately, participation in exercise and activity post-stroke has several barriers. For the nearly 50% of individuals who experience hemiparesis post-stroke, even in the chronic phase (> 6 months post), one-on-one rehabilitation sessions are frequently limited in number due to insurance regulations . A number of researchers have identified factors such as fatigue and a lack of motivation that may prevent people with stroke from initiating and/or maintaining a structured exercise program . People with stroke may also avoid certain activities altogether after a stroke, for fear of failure or insecurity with having a disability. Game-based approaches have been shown to increase motivation to complete home exercise and activity programs but are limited in capturing data on performance and participation in everyday activities. Other researchers have attempted to use wearable sensors to track and monitor activity in the home setting in people with stroke as well as other rehabilitation populations with moderate success. All of these sensors are limited in scope and cannot discern between various activities and contexts.
In our prior work, we have utilized ambient depth sensors (Microsoft Kinect) in the home setting to detect falls and monitor in-home gait patterns of older adults. The older adults found this technology acceptable. Other researchers have used depth sensors to detect and discern activities in “natural” settings. Most studies have been short and in laboratories or mock home environments. Additionally, most research has been conducted in populations without disabilities. We have developed an initial set of algorithms for tracking daily kitchen activities using depth sensors and tested this method with a well population and one individual post-stroke.
The goal of this proposal is to develop, test, and refine a home-based, ambient clinical tool called the Daily Activity Recognition and Assessment System (DARAS). The system will provide robust and accurate measurement of the amount and type of activity performed at home by people with stroke.
Specific Aim 1: Develop, test, and refine an ambient, in-home monitoring system for activity recognition and performance assessment of people with stroke.
We will develop and refine the DARAS algorithms for recognizing activities of people with stroke in the kitchen environment using the Foresite Healthcare depth sensor. These algorithms will be trained using real-world data and we will refine the Convolutional Neural Network-based algorithm (CNN) for accurately segmenting and recognizing activities from continuous capture of depth videos. The DARAS will be deployed in the homes of 20 individuals with stroke in 4 successive cohorts. We will label ground truth for a portion of the data (~30%) and train and test the algorithm. The system will be considered successful when 89% accuracy is achieved in the home environment. Assessment metrics (efficiency, smoothness, gross range of motion) will be calculated once the threshold is reached. Longitudinal performance of assessment metrics will also be calculated.
Specific Aim 2: Evaluate the acceptability of the system for use in the home by people with stroke.
We will hold focus groups with the study participants to determine the acceptability of the data (e.g., depth images recorded) and the impact on daily life. General themes related to acceptability across all groups will be the expected outcome for this aim.
Successful development of the DARAS will be the start of several lines of research with the goal of adding a robust and accurate rehabilitation clinical measure. The DARAS will be subsequently validated against existing gold standard clinical outcome assessments. The DARAS can be combined with other in-home or community-based monitoring devices (e.g., proximity sensors in rooms). Eventually, the DARAS will provide a novel outcome assessment for a variety of post-stroke interventions and provide occupational therapists the ability to detect early declines in performance and deliver tailored interventions.