Heuristic Detection of Recovery Progress using Robotic Data

Abstract

Assessment methods for rehabilitation and recovery have recently been the focal point of research for medical professionals and engineers alike. Current assessment protocols rely on historical ordinal metrics which have been disputed despite their inter-rater reliability. Contemporary kinematic measures have allowed for new approaches to assess recovery progress. However, the abundance of data has deterred medical professionals from adopting these new protocols. This paper presents a method, based on the RMSE-LWSS (Longest Warping Subsequence) score, to distinguish outliers from systemic change for updating the personalized exercise path for users. By treating change detection as a classification problem, the incorporation of a compromised path based on the user's current capability is possible. Experiments were conducted to verify the efficacy of the method, comparing against statistical techniques for change detection and classification of pre-determined paths. The paper highlights how readily available data, rather than complex sensor systems, can be utilized to improve the robustness of personalization capabilities for robotic rehabilitation systems.