Toward community-based wheelchair evaluation with machine learning methods.

Area of Science:

• Biomechanics
• Rehabilitation Engineering
• Machine Learning

Background:

• Manual wheelchair users experience upper extremity pain, impacting function and quality of life.
• Chronic overuse from wheelchair propulsion is a key factor in upper limb injuries.
• Current methods for tracking wheelchair propulsion in community settings are limited.

Purpose of the Study:

• To investigate the feasibility of measuring wheelchair propulsion patterns using wearable accelerometers and machine learning.
• To develop and test a subject-specific machine learning model for accurate propulsion tracking.

Main Methods:

• Two participants (manual wheelchair user, able-bodied individual) wore arm-based accelerometers.
• Wheelchair propulsion patterns were recorded on a roller system and overground.
• Machine learning models (support vector machine) were trained and tested on acceleration and rotation data.

Main Results:

• A support vector machine model accurately predicted overground wheelchair propulsion patterns with 99.7% accuracy using data from a roller system.
• Incorporating rotation data improved prediction accuracy (F1 score = 0.968).

Conclusions:

• Tracking wheelchair propulsion in community settings is possible using wearable sensors and machine learning.
• This technology holds promise for future research and clinical applications in manual wheelchair use.