Learning Spatiotemporal Graph Representations for Visual Perception Using EEG Signals.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Object perception is crucial for environmental interaction.
• Brain-computer interfaces (BCI) offer intuitive interaction but face challenges in decoding visual perception signals.
• Current BCI decoding performance for visual perception is insufficient for real-world applications.

Purpose of the Study:

• To classify electroencephalography (EEG) signals evoked by visual stimuli into semantic object categories.
• To develop an advanced BCI system for recognizing user intentions based on visual perception.
• To improve the accuracy of decoding brain signals for object recognition.

Main Methods:

• Proposed a two-stream convolutional neural network (CNN) architecture.
• Incorporated a spatial stream utilizing graph convolutional neural networks (GCNN).
• Included a temporal stream employing channel-wise convolutional neural networks (CNN).

Main Results:

• Achieved high classification accuracies on two public datasets: 54.28 ± 7.89% for SU DB (6-class) and 84.40 ± 8.03% for MPI DB (2-class).
• The proposed two-stream CNN outperformed existing state-of-the-art methods.
• Demonstrated effective classification of single-trial EEG signals based on visual stimuli.

Conclusions:

• The developed two-stream CNN significantly enhances the classification performance of EEG signals for visual perception.
• This advancement paves the way for more intuitive and practical BCI systems.
• The findings contribute to the application of BCI in real-world scenarios requiring object recognition.