Decoding Subject-Driven Cognitive States from EEG Signals for Cognitive Brain-Computer Interface.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Differentiating cognitive states using brain signals is crucial for advanced brain-computer interfaces (BCIs).
• Electroencephalogram (EEG) offers a non-invasive method for monitoring brain activity.
• Current methods often struggle with the complexity and variability of cognitive states.

Purpose of the Study:

• To assess the feasibility of using EEG signals to distinguish between four distinct cognitive states: resting state, narrative memory, music, and subtraction tasks.
• To develop and evaluate a deep learning model for automated cognitive state classification from EEG data.
• To explore the generalizability of the proposed model across different EEG signal lengths.

Main Methods:

• Collected EEG data from seven healthy male participants performing four cognitive tasks.
• Transformed raw EEG signals into time-frequency maps using continuous wavelet transform.
• Developed a convolutional neural network with channel and frequency attention (TF-CNN-CFA) for classification.

Main Results:

• The TF-CNN-CFA model achieved an average classification accuracy of 76.14% for the four cognitive states.
• The model significantly outperformed traditional EEG processing methods and classical image classification algorithms.
• Consistent performance across varying EEG signal window lengths indicated strong generalization capability.

Conclusions:

• EEG signals can effectively differentiate between higher cognitive states.
• The developed TF-CNN-CFA model demonstrates a promising approach for cognitive state classification.
• This research could pave the way for novel BCI paradigms leveraging EEG-based cognitive state detection.