Classifying Unimanual and Bimanual Upper Extremity Tasks in Individuals Post-Stroke

Area of Science:

• Neuroscience
• Rehabilitation Engineering
• Biomedical Signal Processing

Background:

• Stroke often leads to maladaptive compensatory motions, hindering motor recovery.
• Tracking bimanual motions using sensor data can indicate reduced reliance on compensatory strategies.
• Individual variations in motor strategies create noisy sensor data, posing challenges for analysis.

Purpose of the Study:

• To develop classifiers distinguishing unimanual, bimanual asymmetric, and bimanual symmetric gestures using wearable sensor data.
• To evaluate the efficacy of artificial neural networks (ANNs) and echo state neural networks (ESNNs) for gesture classification in post-stroke individuals.
• To establish a novel method for monitoring and potentially correcting compensatory motion after stroke.

Main Methods:

• Collected wearable sensor data from 20 participants post-stroke and 20 age-matched controls performing specific tasks.
• Developed and compared gesture classifiers using ANNs and ESNNs.
• Analyzed sensor data to differentiate between unimanual, bimanual asymmetric, and bimanual symmetric movements.

Main Results:

• The ESNN classifier achieved higher testing accuracy compared to ANNs for both control (91.3%) and post-stroke (80.3%) participants.
• The ESNN demonstrated improved performance in classifying gestures from potentially chaotic sensor data.
• Accurate gesture classification was achieved in individuals with varying motor strategies post-stroke.

Conclusions:

• ESNNs offer a promising approach for accurately classifying gestures using wearable sensor data in post-stroke populations.
• This classification method can serve as a biomarker for motor recovery and the reduction of compensatory motions.
• The developed classifiers may enable longitudinal monitoring and intervention for improving motor function after stroke.