EEG classification based on visual stimuli via adversarial learning.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Visual brain decoding aims to interpret brain activity in response to visual stimuli.
• Electroencephalogram (EEG) signals offer a non-invasive method for capturing neural activity.
• Classifying EEG signals based on visual stimuli is crucial for advanced brain-computer interfaces.

Purpose of the Study:

• To develop a dual path deep learning architecture for visual brain decoding.
• To classify electroencephalogram (EEG) signals based on evoked image categories.
• To improve subject-invariant feature learning and channel selection for enhanced decoding performance.

Main Methods:

• A dual path deep learning architecture combining Convolutional Neural Networks (CNNs) on time and channel axes.
• Utilizing a Gradient Reversal Layer (GRL) for learning subject-invariant features.
• Employing guided back-propagation for informative EEG channel selection.

Main Results:

• The proposed dual path architecture successfully decodes image categories from EEG signals.
• The inclusion of GRL significantly boosted the system's performance.
• Channel reduction using guided back-propagation maintained high performance, comparable to using all channels.

Conclusions:

• The developed deep learning model offers an effective approach for visual brain decoding using EEG.
• Subject-invariant feature learning and optimized channel selection are key to robust EEG-based visual decoding.
• This work contributes to advancing brain-computer interface technology through improved EEG signal interpretation.