Application of Deep Learning Algorithm to Monitor Upper Extremity Task Practice.

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Neuroscience

Background:

• Upper extremity hemiplegia significantly impacts post-stroke survivors' quality of life.
• Effective motor recovery necessitates high-repetition, task-specific practice.
• Current home-based rehabilitation adherence and quality are often suboptimal, limiting recovery.

Purpose of the Study:

• To develop algorithms for objective identification of task type and movement quality during upper extremity exercises.
• To assess the feasibility of using IMU sensors and LSTM classifiers for this purpose.

Main Methods:

• Twenty neurotypical participants performed four distinct upper extremity tasks with varying movement qualities.
• Inertial Measurement Unit (IMU) sensors were worn on the wrist.
• Long Short-Term Memory (LSTM) classifiers were trained to differentiate tasks and movement qualities.

Main Results:

• The LSTM models achieved 90.8% accuracy in identifying the task type.
• Movement quality classification accuracy varied across tasks: 84.9%, 81.1%, 58.4%, and 73.2%.
• Models demonstrated effectiveness on unseen participants.

Conclusions:

• Objective algorithms can accurately classify upper extremity task type and movement quality.
• Further research is needed to validate performance in stroke survivors.
• Objective monitoring and feedback hold potential for enhancing home exercise adherence and motor recovery.