Multi-Modal Sensing for Propulsion Estimation in People Post-Stroke Across Speeds

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Machine Learning in Healthcare

Background:

• Gait rehabilitation is crucial for locomotor recovery after neuromotor injuries.
• Community-based rehabilitation requires accurate real-world gait monitoring.
• Single wearable sensors (IMUs, pressure insoles) have limitations in achieving clinical accuracy for gait analysis.

Purpose of the Study:

• To investigate the benefits of multi-modal sensing by integrating IMU and insole data.
• To develop individualized machine learning models for estimating gait propulsion in post-stroke individuals.
• To assess the feasibility of real-world propulsion estimation during community-based gait rehabilitation.

Main Methods:

• Integrated data from inertial measurement units (IMUs) and pressure insoles.
• Developed individualized machine learning models to estimate propulsion, a key biomechanical variable.
• Validated models in laboratory settings and applied them to track propulsion in real-world scenarios with variable-speed walking and gait interventions.

Main Results:

• IMU + Insole models significantly improved propulsion estimation accuracy compared to single-modality models in the lab (RMSE of 0.80 %BW).
• Achieved clinically relevant RMSEs for peak paretic propulsion (0.71%BW) and propulsion impulse (0.19%BW s).
• Observed consistent changes in estimated propulsion in real-world settings, mirroring lab findings across different walking speeds and interventions.

Conclusions:

• Multi-modal sensing with IMUs and insoles offers a promising approach for accurate gait propulsion estimation in post-stroke individuals.
• This technology can support community-based rehabilitation by enabling reliable real-world gait monitoring.
• The developed machine learning models show feasibility for tracking gait changes and supporting motor recovery beyond clinical settings.