This work develops, through close participatory design with rehabilitation clinicians, an end-to-end data-acquisition pipeline for building algorithms that automate the Action Research Arm Test (ARAT) in upper-limb stroke rehabilitation. Its novel contribution is a clinician-operable four-RGB-camera capture rig with a tablet calibration/administration interface (yielding 106 in-clinic ARAT sessions), a custom standardized video-segmentation tool, and a standardized rating tool, jointly engineered to remove the multi-view, segmentation-consistency, and rating-reliability limitations that had blocked robust automated assessment. The paper positions participatory design itself as a transferable method for encoding expert human-movement knowledge into AI tools so they stay compatible with expert clinical practice.
This human-computer-interaction paper reports a participatory-design process, carried out with rehabilitation clinicians, for acquiring high-quality data to build automated assessment of the Action R…