JMIR Publications

ABSTRACT

Background:

Upper extremity (UE) impairment affects up to 75% of stroke survivors and accounts for most portion of the rehabilitation after the hospital's release. Compensation, commonly employed by stroke survivors during the UE rehabilitation is applied to adapt to the loss of motor function and may impede the rehabilitation process in the long term and lead to new orthopedic problems. Intensive monitoring of compensatory movements is critical for improving functional outcomes during rehabilitation.

Objective:

This review is to analyze how sensors and artificial intelligence (AI) have been applied to assess and detect compensation during stroke upper extremity (UE) rehabilitation.

Methods:

A wide database search was performed. All studies were screened by two reviewers independently, with a third reviewer involved to resolve discrepancies. The final included studies were rated for their level of clinical evidence based on the correlation with clinical scales (with the same tasks and/or the same evaluation criteria). One reviewer extracted data on publication, demographic information, compensation types, sensors used for compensation assessment, compensation measurement and statistical/AI methods. Accuracy was checked by another reviewer. A descriptive summary of findings was conducted. Data was synthesized and tabulated based on each research question.

Results:

Seventy two studies were included in the review. Two types of compensation were identified: 1) the disuse of the affected upper limb, and 2) the awkward use of the affected upper limb to adjust for limited strength, mobility, and motor control. There are various models applied to describe the two types of compensation. A large variety of quantitative measurements were proposed to characterize these compensation models. The body-worn technology (24 studies) was the most used sensor technology to assess compensation, followed by the marker-based motion capture system (22 studies) and the maker-free vision sensor technology (17 studies). Most studies used statistical methods (56 studies) for compensation assessment, while heterogeneous machine learning algorithms (15 studies) were also applied for automatic detection of compensatory movements and postures.

Conclusions:

This systematic review provides insights for future research in realizing automatic real-time compensation assessment and detection during stroke UE rehabilitation. It is advised that open data together with deep learning algorithms and multi-label classification algorithms could benefit to automatic real-time compensation detection. It is convinced that robot-assisted devices and technology-based compensation assessment and detection have the capacity to augment rehabilitation independent from the constant care of therapists. Wearable sensors, marker-free vision sensors, and ambient sensors have the potential to be used for compensation assessment and detection in home-setting therapy. It is also recommended to explore technology-based compensation prediction.