Data-Driven Techniques for Estimating Energy Expenditure in Wheelchair Users.

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Wearable Technology

Background:

• Obesity prevention is a significant challenge for manual wheelchair users (MWU).
• Accurate feedback on energy expenditure (EE) can support weight management strategies.
• Existing methods for EE estimation may not be suitable for the unique physiological demands of MWU.

Purpose of the Study:

• To develop and validate data-driven algorithms for activity classification and EE estimation in MWU.
• To investigate the influence of personal characteristics on EE estimation accuracy.
• To determine optimal sensor placement for wearable EE monitoring devices.

Main Methods:

• Collected data from 40 participants (20 MWU, 20 controls) during various activities (lying, sitting, wheelchair propulsion).
• Utilized heart rate, inertial measurement units (IMU), and personal characteristics.
• Employed machine learning models including Support Vector Machines, Gaussian Processes, Random Forest, XGBoost, and Neural Networks.

Main Results:

• Achieved high accuracy in activity classification with minimal confusion between activities.
• Demonstrated strong generalization capabilities of EE estimation models on unseen participants.
• Identified the wrist as the optimal IMU placement location, balancing accuracy and wearability.
• Found no significant improvement in EE estimation accuracy for MWU when including data from non-disabled individuals.

Conclusions:

• Data-driven algorithms using wearable sensors and personal characteristics are effective for activity classification and EE estimation in MWU.
• Personalization of algorithms is essential for ecological validity in daily life settings.
• Wearable sensor technology offers a viable approach to support obesity prevention efforts in the MWU population.