Individualized Learning-Based Ground Reaction Force Estimation in People Post-Stroke Using Pressure Insoles.

Area of Science:

• Clinical Biomechanics
• Rehabilitation Engineering
• Wearable Sensor Technology

Background:

• Stroke is a primary cause of gait disability, impacting independence and quality of life.
• A gap exists between biomechanical analysis tools and clinical practice, particularly in gait assessment.
• Traditional 3D ground reaction force (3D GRF) measurement requires lab-based equipment like force plates.

Purpose of the Study:

• To evaluate subject-specific machine learning approaches for estimating 3D GRFs using pressure insoles in post-stroke individuals.
• To assess the performance of these methods across varying walking speeds in a heterogeneous population.
• To determine the accuracy of estimated GRFs and their correlation with clinical gait metrics.

Main Methods:

• Utilized pressure insoles to collect data from post-stroke individuals during gait.
• Applied three subject-specific machine learning models to estimate 3D GRFs.
• Validated the estimated GRFs against ground truth measurements from force plates.
• Analyzed estimation errors and correlations for medio-lateral, antero-posterior, and vertical GRF components.

Main Results:

• A Convolutional Neural Network (CNN)-based approach demonstrated the lowest estimation errors for all GRF components.
• Estimated GRF components showed strong correlations with ground truth measurements.
• High accuracy was achieved for estimating clinically relevant point metrics on the paretic limb.

Conclusions:

• Subject-specific machine learning models, particularly CNNs, show promise for accurate 3D GRF estimation using wearable pressure insoles in post-stroke gait.
• This approach has the potential to bridge the gap between lab-based biomechanics and real-world clinical applications for stroke rehabilitation.
• Individualized machine learning offers a pathway to personalized gait analysis and improved rehabilitation strategies for stroke survivors.