EEGformer: A transformer-based brain activity classification method using EEG signal.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Effective analysis of steady-state visual evoked potential (SSVEP) signals is crucial for early glaucoma diagnosis.
• Current methods often adapt existing techniques for SSVEP-based brain-computer interface (BCI) tasks, rather than developing domain-specific solutions.

Purpose of the Study:

• To propose a novel transformer-based model, EEGformer, for unified analysis of electroencephalogram (EEG) signals.
• To capture the temporal, regional, and synchronous characteristics of brain activity within EEG signals.

Main Methods:

• Utilized a one-dimensional convolution neural network (1DCNN) for automatic EEG-channel-wise feature extraction.
• Developed the EEGformer model, comprising regional, synchronous, and temporal transformers.
• Validated performance on the BETA benchmark database for SSVEP-BCI and compared against state-of-the-art models on the SEED and DepEEG datasets.

Main Results:

• EEGformer achieved superior classification performance across all three tested EEG datasets.
• The model's architecture and unified approach to learning EEG characteristics enhance classification accuracy.

Conclusions:

• EEGformer demonstrates strong generalization capabilities across diverse EEG datasets.
• The approach holds potential for accurate brain activity classification in various applications, including SSVEP-based glaucoma diagnosis, emotion recognition, and depression discrimination.