Research Article
The Toronto Rehab Stroke Pose Dataset to Detect Compensation during Stroke Rehabilitation Therapy
@INPROCEEDINGS{10.1145/3154862.3154925, author={Elham Dolatabadi and Ying Zhi and Bing Ye and Giorgia Lupinacci and Alex Mihailidis and Rosalie Wang and Babak Taati}, title={The Toronto Rehab Stroke Pose Dataset to Detect Compensation during Stroke Rehabilitation Therapy}, proceedings={11th EAI International Conference on Pervasive Computing Technologies for Healthcare}, publisher={ACM}, proceedings_a={PERVASIVEHEALTH}, year={2018}, month={1}, keywords={automated coaching upper body motion benchmarking kinect stroke rehabilitation compensatory movements}, doi={10.1145/3154862.3154925} }- Elham Dolatabadi
Ying Zhi
Bing Ye
Giorgia Lupinacci
Alex Mihailidis
Rosalie Wang
Babak Taati
Year: 2018
The Toronto Rehab Stroke Pose Dataset to Detect Compensation during Stroke Rehabilitation Therapy
PERVASIVEHEALTH
ACM
DOI: 10.1145/3154862.3154925
Abstract
Stroke often leads to upper limb movement impairments. To accommodate new constraints, movement patterns are sometimes altered by stroke survivors to use stronger or unaffected joints and muscles. If used during rehabilitation exercises, however, such compensatory motions may result in ineffective outcomes. A system that can automatically detect compensatory motions would be useful in coaching stroke survivors to use proper positioning. Towards the development of such an automated tool, we present a dataset of clinically relevant motions during robotic rehabilitation exercises. The dataset is captured with a Microsoft Kinect sensor and contains two groups of participants – 10 healthy and 9 stroke survivors – performing a series of seated motions using an upper-limb rehabilitation robot. Healthy participants performed additional sets of scripted motions to simulate common post-stroke compensatory movements. The dataset also includes common clinical assessment scores. Compensatory motions of both healthy and stroke participants were annotated by two experts and are included in the dataset. We also present a preliminary evaluation of the dataset in terms of its sensitivity and specificity in detecting compensatory movements for selected tasks. This dataset is valuable because it includes clinically relevant motions in a clinical setting using a cost-effective, portable, and convenient sensor.
