A Self-Supervised Learning Based Channel Attention MLP-Mixer Network for Motor Imagery Decoding.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Convolutional Neural Networks (CNNs) are standard for Electroencephalogram (EEG) motor-imagery (MI) decoding but struggle with small sample sizes.
• Segmenting EEG trials for data augmentation risks losing crucial long-range temporal dependencies.

Purpose of the Study:

• To develop a novel self-supervised learning (SSL) network, S-CAMLP-Net, for improved EEG-based MI decoding.
• To address the limitations of existing methods in capturing long-range temporal and spatial information in EEG signals.

Main Methods:

• Proposed a self-supervised learning (SSL) based channel attention MLP-Mixer network (S-CAMLP-Net).
• Introduced a novel EEG slice prediction pretext task to capture long-range temporal dependencies.
• Utilized an MLP-Mixer adapted for signal classification and integrated an attention mechanism for spatial feature learning.

Main Results:

• S-CAMLP-Net demonstrated superior classification performance on the MI-2 and BCI Competition IV Dataset 2A.
• The method effectively learned long-range temporal information and global spatial features from EEG signals.
• Achieved better results compared to all other evaluated algorithms.

Conclusions:

• The proposed S-CAMLP-Net effectively overcomes the limitations of small sample sizes in EEG-based MI decoding.
• The integration of SSL, MLP-Mixer, and channel attention enhances the model's ability to capture complex EEG signal characteristics.
• S-CAMLP-Net offers a promising advancement for brain-computer interfaces and neurological studies.