Cross-Channel Specific-Mutual Feature Transfer Learning for Motor Imagery EEG Signals Decoding

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Deep learning frameworks are increasingly used in brain-computer interface (BCI) research for decoding motor imagery (MI) electroencephalogram (EEG) signals.
• Current methods often embed mixed neural signals into a single feature space, neglecting distinct regional characteristics and reducing feature expressiveness.

Purpose of the Study:

• To propose a novel deep learning model, the cross-channel specific-mutual feature transfer learning (CCSM-FT) network.
• To address the challenge of mixed neural activities recorded by electrodes in BCI.
• To enhance the accuracy of motor imagery decoding by effectively utilizing specific and mutual neural features.

Main Methods:

• Developed a multibranch network to extract specific and mutual features from multiregion brain signals.
• Employed specific training techniques to maximize feature distinction and improve algorithm effectiveness.
• Utilized feature transfer learning and an auxiliary dataset to enhance identification performance.

Main Results:

• The CCSM-FT network demonstrated superior classification performance on benchmark datasets (BCI Competition IV-2a and HGD).
• The model effectively distinguished between specific and mutual neural features.
• Feature transfer learning enhanced the expressive power of the extracted features.

Conclusions:

• The proposed CCSM-FT network offers a significant advancement in motor imagery decoding for BCI applications.
• Separating and transferring specific and mutual neural features improves the accuracy and robustness of brain activity analysis.
• This approach holds promise for developing more effective BCI systems.