SSGCNet: A Sparse Spectra Graph Convolutional Network for Epileptic EEG Signal Classification.

Area of Science:

• Neurology
• Computer Science
• Machine Learning

Background:

• Epileptic seizures are often diagnosed using electroencephalogram (EEG) signals.
• Accurate and efficient classification of EEG signals is crucial for diagnosis and treatment.
• Existing deep learning models can be computationally intensive.

Purpose of the Study:

• To propose a lightweight deep learning model for epileptic EEG signal classification.
• To maintain high classification accuracy while reducing model complexity.
• To introduce a novel graph representation for EEG signals.

Main Methods:

• Developed a weighted neighborhood field graph (WNFG) to represent EEG signals, reducing graph complexity.
• Implemented a sparse spectra graph convolutional network (SSGCNet) using WNFG.
• Utilized sparse weight pruning and the alternating direction method of multipliers (ADMM) for model optimization.

Main Results:

• The SSGCNet achieved comparable classification accuracy to state-of-the-art methods on public and clinical datasets.
• The proposed WNFG representation significantly reduced graph generation time and memory usage.
• The model demonstrated high performance even with a tenfold reduction in connection rate.

Conclusions:

• SSGCNet offers an efficient and accurate solution for epileptic EEG signal classification.
• The WNFG approach effectively represents EEG data, leading to computational savings.
• This lightweight model holds promise for real-time EEG analysis and clinical applications.