Measuring Functional Arm Movement after Stroke Using a Single Wrist-Worn Sensor and Machine Learning.

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Machine Learning in Healthcare

Background:

• Objective efficacy endpoints are crucial for stroke rehabilitation trials.
• Real-world upper extremity (UE) functional use is a key metric.
• Current methods for assessing UE use are limited.

Purpose of the Study:

• To present a novel, inexpensive, and feasible method for differentiating functional UE use from nonfunctional movement post-stroke.
• To utilize a single wrist-worn accelerometer for this assessment.

Main Methods:

• Ten controls and 10 stroke survivors wore wrist accelerometers during activities.
• Video data served as ground truth to label sensor data as functional or nonfunctional UE use.
• A Random Forest machine learning model was trained and tested on the annotated sensor data.

Main Results:

• The method achieved high accuracy in classifying UE use: 94.80% in controls and 88.38% in stroke subjects (intrasubject).
• Intersubject testing showed 91.53% accuracy for controls and 70.18% for stroke subjects.
• Machine learning on raw sensor data proved feasible for remote UE use determination.

Conclusions:

• The accelerometer-based method offers a promising, cost-effective, and objective way to quantify functional UE use in hemiparesis.
• It can aid in assessing the impact of UE treatments and facilitate remote monitoring.
• An intrasubject test/train approach is particularly accurate for restorative treatment trials.