Decoding Brain Signals to Classify Gait Direction Anticipation.

Area of Science:

• Neuroscience
• Rehabilitation Engineering
• Biomedical Signal Processing

Background:

• Brain-computer interface (BCI) technology offers a promising avenue for motor rehabilitation.
• Electroencephalography (EEG) signals are utilized in BCIs to control assistive devices like lower-limb exoskeletons.
• Existing research primarily focuses on predicting gait intention and variations, with limited exploration of gait direction prediction.

Purpose of the Study:

• To predict anticipated gait direction from electroencephalography (EEG) signals.
• To develop and evaluate a BCI system for real-time gait direction prediction.
• To explore the potential of BCI in enhancing lower-limb exoskeleton-assisted rehabilitation.

Main Methods:

• Collected EEG data from three healthy participants performing directional gait tasks.
• Processed EEG epochs in the time-frequency domain using event-related synchronization (ERS) and desynchronization (ERD).
• Classified gait direction using logistic regression (LR), support vector machine (SVM), and convolutional neural network (CNN) with ten-fold cross-validation.

Main Results:

• The Convolutional Neural Network (CNN) classifier demonstrated superior performance compared to LR and SVM.
• The CNN achieved an accuracy of 0.75 in predicting gait direction from EEG signals.
• Event-related synchronization and desynchronization patterns were effectively utilized for classification.

Conclusions:

• Predicting gait direction from EEG signals is feasible and can be achieved with high accuracy using advanced machine learning models like CNNs.
• This research provides a foundation for developing more intuitive and responsive control systems for lower-limb exoskeletons in rehabilitation.
• The findings have significant implications for improving neuroplasticity and aiding individuals with lower-limb motor function disabilities through advanced BCI-driven therapies.