REI-Net: A Reference Electrode Standardization Interpolation Technique Based 3D CNN for Motor Imagery Classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• High-quality scalp electroencephalography (EEG) datasets are crucial for motor imagery (MI) analysis.
• Channel information loss in EEG datasets due to electrode size and montage presents a significant challenge for MI decoding.
• Existing 2D and 3D EEG representations have limitations in retaining both temporal and spatial information while mitigating noise and channel loss.

Purpose of the Study:

• To propose a novel framework, Reference Electrode Standardization Interpolation Network (REI-Net), for improved motor imagery decoding.
• To enhance spatial resolution in 2D scalp EEG data while preserving temporal information.
• To improve the robustness of MI decoding against individual data variability using transfer learning.

Main Methods:

• Development of the Reference Electrode Standardization Interpolation Network (REI-Net) framework.
• Utilizing a 3D EEG representation with interpolation to retain temporal information and improve spatial resolution.
• Application of transfer learning to address data variability and enhance decoding robustness.

Main Results:

• REI-Net achieved promising performance on two widely-recognized MI datasets.
• Achieved an accuracy of 77.99% on the BCI-C IV-2a dataset.
• Achieved an accuracy of 63.94% on the Kaya2018 dataset, outperforming state-of-the-art methods.

Conclusions:

• The proposed REI-Net framework effectively improves spatial resolution and retains temporal information in scalp EEG for MI analysis.
• Transfer learning enhances the robustness of MI decoding, leading to more accurate results.
• REI-Net demonstrates superior performance compared to existing methods, offering a more accurate and robust solution for motor imagery decoding.