EEG decoding method based on multi-feature information fusion for spinal cord injury
Fangzhou Xu1, Jincheng Li2, Gege Dong2
1School of Optoelectronic Engineering International, Qilu University of Technology (Shandong Academy of Sciences), Jinan 250353, China.
Summary
This study introduces a novel deep learning framework using modified graph convolutional networks (M-GCN) and modified S-transform (MST) for enhanced brain-computer interface (BCI) systems. The M-GCN method significantly improves motor imagery (MI) recognition accuracy in spinal cord injury (SCI) patients.
Area of Science:
- Neuroscience and Biomedical Engineering
- Artificial Intelligence in Healthcare
Background:
- Electroencephalography (EEG) is crucial for brain-computer interface (BCI) systems, measuring neuronal activity.
- Existing EEG-based motor imagery (MI) studies often underutilize brain network topology.
- There is a need for advanced signal processing and deep learning techniques to improve BCI performance, especially for individuals with spinal cord injury (SCI).
Purpose of the Study:
- To develop an efficient BCI system by improving the decoding performance of EEG signals for motor imagery (MI) recognition.
- To integrate temporal-frequency processing with brain network topology using a novel deep learning framework.
- To enhance rehabilitation training for SCI patients by accurately decoding their motor intentions.
Main Methods:
- Proposed a deep learning framework based on a modified graph convolution neural network (M-GCN).
- Employed modified S-transform (MST) for temporal-frequency processing of EEG data, aligning with electrode spatial relationships.
- Fused temporal-frequency-spatial features and utilized M-GCN to decode relation features from EEG signals of SCI patients.
Main Results:
- The M-GCN framework demonstrated superior performance compared to existing methods in MI recognition.
- Achieved a classification accuracy of 87.456% for identifying MI tasks.
- 10-fold cross-validation confirmed the algorithm's reliability and stability, with an average accuracy of 87.442%.
Conclusions:
- The proposed M-GCN framework effectively enhances BCI performance by integrating advanced signal processing and network topology analysis.
- This method offers a reliable and stable approach for decoding motor intentions from EEG signals, particularly beneficial for SCI patients.
- The developed BCI system holds potential for effective rehabilitation training, aiding in the partial restoration of motor function in SCI individuals.
