Phase-Based Gait Prediction after Botulinum Toxin Treatment Using Deep Learning.

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Gait disorders in neurological diseases often involve spasticity, which can be treated with Botulinum Toxin Type A (BTX-A) injections.
• Optimizing BTX-A treatment requires accurate prediction of post-treatment outcomes, balancing benefits and risks.
• Predicting kinematic changes is essential for effective gait rehabilitation.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for predicting knee and ankle kinematics after BTX-A treatment.
• To assess the effectiveness of incorporating pre-treatment kinematics and detailed treatment information into prediction models.
• To compare different deep learning architectures for their accuracy in predicting gait kinematics post-treatment.

Main Methods:

• A Bidirectional Long Short-Term Memory (Bi-LSTM) deep learning architecture was employed.
• Two models were designed: one integrating treatment data into hidden layers, and another using a gating mechanism.
• The models were trained and validated using pre-treatment kinematics and specific BTX-A treatment data, analyzing interactions between injected muscles.

Main Results:

• The deep learning model with a gating mechanism achieved superior prediction accuracy.
• Average prediction root mean squared error for knee kinematics was 2.99° (R2 = 0.85).
• Average prediction root mean squared error for ankle kinematics was 2.21° (R2 = 0.84).

Conclusions:

• The proposed Bi-LSTM approach, particularly with the gating mechanism, significantly outperforms existing methods for predicting post-BTX-A gait kinematics.
• This novel method offers enhanced accuracy in forecasting knee and ankle movement, aiding in personalized spasticity management.
• The findings support the use of advanced AI for optimizing therapeutic interventions in neurological gait disorders.