Basic Graphic Shape Decoding for EEG-based Brain-Computer Interfaces.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Image decoding from electroencephalogram (EEG) is an emerging area in brain-computer interface (BCI) research.
• Prior studies focused on complex objects, leaving the impact of simple image variations on EEG signals less understood.

Purpose of the Study:

• To investigate the decodability of simple visual stimuli with varying spatial patterns from EEG signals.
• To explore the efficacy of combining Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM) for EEG-based image decoding.

Main Methods:

• Utilized a single white bar with eight distinct spatial patterns as visual stimuli.
• Employed a hybrid CNN-LSTM model to analyze and decode the corresponding EEG data.
• Recruited four human subjects for data acquisition.

Main Results:

• Achieved high binary classification accuracies, reaching up to 97.2% for one subject.
• Demonstrated substantial 4-class classification accuracy, exceeding 70% for most subjects.
• Indicated that basic graphic shapes elicit decodable EEG signal patterns.

Conclusions:

• Basic graphic shapes can be successfully decoded from EEG signals.
• The CNN-LSTM model shows significant potential for image decoding in EEG-based BCIs.
• This research paves the way for more sophisticated visual decoding applications in BCI technology.