Daily Activity Recognition and Assessment System for Stroke Rehabilitation

Abstract

<h3>Research Objectives</h3> To test an initial algorithm for recognizing and classifying daily activities of individuals with stroke in an unstructured home environment using a depth sensor. <h3>Design</h3> This study was a developmental, iterative design, completed in three phases. <h3>Setting</h3> In Phase 1, the algorithm was trained in a kitchen. In Phase 2, the trained algorithm was tested on 5 participants data collected from natural home kitchen environments. In Phase 3, the algorithm was trained and tested in natural home kitchen environments (individuals post-stroke). <h3>Participants</h3> In Phase 1, the system was trained with 11 individuals (age 18-40) without any disabilities. In Phase 2, the trained algorithm was tested on 5 participants with and without disabilities. In Phase 3, the algorithm was trained and tested on 15 adults post-stroke. <h3>Interventions</h3> In Phase 1, each participant completed 3 sets of prescribed activities. In Phase 2, participants completed meal preparation tasks and all data were recorded by the VicoVR and the Foresite sensor.All data (depth images and skeletal data) from Phase I were used to trained the Convolutional-deconvolutional neural network (CNN). <h3>Main Outcome Measures</h3> Per-frame accuracies and per-action accuracies were calculated for all algorithm tests. <h3>Results</h3> For the mock kitchen in Phase 1, the per-frame accuracy was 85.1% and the per-action accuracy was 92.1%. In Phase 2, the per-frame accuracies ranged from 52.0% to 89.7% and the per-action accuracies ranged from 54.3% to 88.5%. Phase 3 is currently ongoing. Our current tests of the algorithm show per-frame accuracies of 51.5% to 67.3% and per-action accuracies from 55.6% to 77.9%. <h3>Conclusions</h3> The per-action and per-frame accuracies indicate that the algorithm requires further training and testing. This study is the first of its kind to develop an unobtrusive assessment of daily activity. <h3>Author(s) Disclosures</h3> Dr. Skubic holds equity in Foresite Healthcare, the company that produces the sensor used in Phases 2 and 3 of this study. Dr. Proffitt has nothing to declare. Ms. Ma has nothing to declare.