GDNet-EEG: An attention-aware deep neural network based on group depth-wise convolution for SSVEP stimulation...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Early glaucoma diagnosis relies on accurate processing of steady-state visually evoked potentials (SSVEPs).
• Deep learning models are crucial for effective stimulation frequency recognition in SSVEPs.
• Electroencephalography (EEG) data processing presents challenges for precise diagnostic outcomes.

Purpose of the Study:

• To propose a novel deep learning model, GDNet-EEG, for SSVEPs-based stimulation frequency recognition.
• To tailor an EEG-oriented deep learning model for capturing regional and network brain activity characteristics.
• To enhance the accuracy of early glaucoma diagnosis through improved SSVEPs analysis.

Main Methods:

• Developed Group Depth-wise Convolution (GDNet-EEG) to extract temporal and spectral EEG features.
• Implemented EEG attention mechanisms (channel-wise and network-wise) to identify critical brain regions and functional networks.
• Validated the model on two public SSVEPs datasets (benchmark and BETA) and their combination.

Main Results:

• GDNet-EEG achieved average classification accuracies of 84.11%, 85.93%, and 93.35% on the benchmark, BETA, and combined datasets, respectively.
• The model demonstrated significant improvements over baseline methods, with accuracy increases ranging from 1.96% to 18.2% on the combined dataset.
• High classification accuracy was achieved with a signal length of 1 second.

Conclusions:

• The GDNet-EEG model shows potential for accurate SSVEP stimulation frequency recognition.
• This approach could be valuable for early glaucoma diagnosis.
• The developed deep learning model effectively analyzes EEG data for neurological applications.