Long short-term memory-based Gait Phase Prediction Using Heel Acceleration in People with Gait Disorders.

Area of Science:

• Biomechanics
• Computational Neuroscience
• Rehabilitation Engineering

Background:

• Gait disorders result in slower, unstable walking compared to healthy individuals.
• Accurate gait phase prediction is crucial for developing effective gait-assistive orthoses.
• Current prediction models may lack personalization, potentially limiting accuracy in patient populations.

Purpose of the Study:

• To develop and evaluate an improved gait phase prediction model for individuals with gait disorders.
• To enhance prediction accuracy by personalizing the model to individual gait characteristics.
• To assess the potential application of the prediction model in gait-assistive orthoses.

Main Methods:

• Gait data from healthy volunteers and individuals with gait disorders were collected using motion capture technology.
• A long short-term memory (LSTM) neural network was employed to construct a gait phase prediction model.
• Heel acceleration data served as input, with the model predicting gait phase 0.1 seconds in advance.
• A novel approach involved creating individual models and adjusting input data length to each participant's gait cycle.

Main Results:

• The conventional prediction model achieved an 84% accuracy in gait phase prediction.
• The proposed individualized model demonstrated a higher prediction accuracy of 89%.
• Personalization significantly improved the model's performance for participants with gait disorders.

Conclusions:

• Individualized gait phase prediction models, utilizing LSTM and adjusted input data, enhance accuracy for gait disorder patients.
• This personalized approach shows promise for improving the functionality and effectiveness of gait-assistive orthoses.
• The developed model offers a pathway toward more responsive and adaptive assistive technologies for mobility impairments.