JMIR Publications

ABSTRACT

Background:

Stroke therapy is essential to reduce impairments and improve motor movements by engaging autogenous neuroplasticity. Utilizing machine learning and therapy session kinematic measurements promises to have a central role in rehabilitation decision-making in determining if patient therapy is improving.

Objective:

This study uses supervised learning methods to address a clinician’s autonomous classification of stroke residual severity labeled data towards improving in-home robotics-assisted stroke rehabilitation.

Methods:

Thirty-three stroke patients participate in in-home therapy sessions using the Motus Nova robotics rehabilitation technology to capture upper and lower body motion. The therapy session summary data is based on high-resolution movement and assistance data and clinician-informed discrete stroke residual severity labels. This arises from a final processed dataset of 32,902 patient sessions based on the maximum score per patient per session. Four machine learning algorithms are used to classify stroke residual severity: light gradient boosting, extra trees, deep neural networks, and classical logistic regression. Their performance measures are evaluated to identify which method maximizes stroke residual severity classification accuracy.

Results:

We demonstrate that the light gradient boosting method provides the most reliable autonomous detection of stroke severity.

Conclusions:

We show how objectively measured rehabilitation training paired with machine learning methods can be used to identify the residual stroke severity class with efforts to enhance in-home self-guided, individualized stroke rehabilitation. As data from rehabilitation practices are often of comparable size and nature to the data collected in our study, this suggests that the light gradient boosting method should be considered a standard, more efficient tool for this analysis.