Gait Pattern Recognition Based on Supervised Contrastive Learning Between EEG and EMG.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Rehabilitation Technology

Background:

• Electroencephalography (EEG) and lower-limb electromyography (EMG) are crucial for lower-limb kinematic tasks.
• EMG offers superior accuracy but faces challenges like fatigue and signal acquisition difficulties.
• EEG signals are stable and easier to obtain, presenting an alternative for gait analysis.

Purpose of the Study:

• To develop a multimodal training strategy leveraging supervised contrastive learning.
• To enhance lower-limb kinematic classification and regression using EEG signals.
• To improve gait analysis accuracy by utilizing EMG as a guide during training.

Main Methods:

• Proposed a multimodal training strategy employing supervised contrastive learning.
• Utilized EMG signals to guide the model's training phase for gait analysis.
• Employed EEG signals exclusively during the testing phase for evaluation.

Main Results:

• The multimodal strategy demonstrated superior performance compared to single-modal EEG training.
• Models trained with the proposed strategy achieved higher accuracy in lower-limb kinematic tasks.
• The Pearson's Correlation Coefficient exceeded that of all baseline models, validating the approach.

Conclusions:

• The proposed multimodal training strategy effectively utilizes EMG's guiding role for EEG-based gait analysis.
• This approach overcomes EMG acquisition limitations while achieving robust kinematic prediction.
• The findings suggest a promising direction for improving non-invasive gait monitoring systems.