Decoding covert visual attention of electroencephalography signals using continuous wavelet transform and deep...

Area of Science:

• Cognitive Neuroscience
• Brain-Computer Interfaces
• Machine Learning

Background:

• Decoding covert visual attention from electroencephalography (EEG) signals is crucial for cognitive neuroscience and brain-computer interface (BCI) applications.
• Conventional methods often require manual feature extraction, limiting their scalability and generalizability.

Purpose of the Study:

• To develop and evaluate a deep learning framework for end-to-end classification of covert attention states using EEG.
• To investigate the effectiveness of time-frequency representations, specifically Continuous Wavelet Transform (CWT), in enhancing attention decoding.

Main Methods:

• EEG data were collected from ten healthy participants engaged in spatial and feature-based attention tasks.
• A deep learning approach integrating CWT with neural networks (ShallowConvNet, EEGNet) was employed for classification.
• Performance was evaluated on binary and four-class attention decoding scenarios.

Main Results:

• ShallowConvNet achieved 100% accuracy in binary classification and over 90% in four-class conditions.
• EEGNet demonstrated competitive performance, exceeding 97% and 88% accuracy in two- and four-class tasks, respectively.
• The CWT-integrated deep learning models significantly outperformed conventional raw-signal approaches.

Conclusions:

• Integrating CWT with deep neural networks offers a scalable and efficient solution for decoding covert attention from EEG signals.
• This approach enhances decoding performance, paving the way for improved real-time attention monitoring in BCI and neuroscience research.