A Combined Virtual Electrode-Based ESA and CNN Method for MI-EEG Signal Feature Extraction and Classification

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Brain-Computer Interfaces (BCIs) facilitate communication between the brain and external devices by decoding Electroencephalography (EEG) signals.
• Motor Imagery (MI) is a key BCI paradigm involving imagined movements to generate EEG signals, crucial for computer-aided diagnosis and rehabilitation.
• Existing MI-BCI systems face challenges including individual differences, low signal-to-noise ratio, and poor online performance.

Purpose of the Study:

• To address the limitations of current MI-BCI systems.
• To develop an improved method for feature extraction and classification of MI-EEG signals.
• To enhance the performance and adaptability of online MI-BCI systems.

Main Methods:

• A combined approach using virtual electrode-based EEG Source Analysis (ESA) and Convolutional Neural Networks (CNNs) was proposed.
• This method focuses on feature extraction and classification of MI-EEG signals.
• The system was evaluated for its online performance and cross-subject generalization capabilities.

Main Results:

• The developed online MI-BCI system demonstrated improved decoding of multi-task MI-EEG signals post-training.
• The method successfully learned generalized features across subjects, showing adaptability to individual differences.
• The system achieved online decoding of EEG intent, enabling brain control of an intelligent cart.

Conclusions:

• The proposed ESA and CNN method offers a new approach for online MI-BCI systems.
• This technique enhances the decoding ability and adaptability of MI-BCI systems.
• The findings provide a promising direction for the future research and application of MI-BCI technology.