Epileptic Seizure Detection in EEG Signals Using a Unified Temporal-Spectral Squeeze-and-Excitation Network

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epileptic seizure detection using electro-encephalogram (EEG) signals is crucial for patient care.
• Current deep learning models often overlook the combined spectral and temporal characteristics of EEG, limiting their effectiveness.
• Nonstationary and nonlinear properties in epileptic EEGs can be missed by existing methods, leading to suboptimal performance.

Purpose of the Study:

• To propose an advanced deep learning framework for accurate and automatic epileptic seizure detection.
• To address the limitations of existing models in simultaneously analyzing spectral and temporal EEG data.
• To mitigate overfitting issues common in seizure detection due to data scarcity.

Main Methods:

• Development of a novel channel-embedding spectral-temporal squeeze-and-excitation network (CE-stSENet).
• Integration of multi-level spectral and multi-scale temporal analysis within the CE-stSENet architecture.
• Utilization of a maximum mean discrepancy-based information maximizing loss function to combat overfitting.

Main Results:

• The proposed CE-stSENet framework demonstrated superior performance in recognizing epileptic EEGs across three datasets.
• The model effectively captured hierarchical multi-domain representations by integrating spectral and temporal features.
• Experimental results showed competitive performance against state-of-the-art methods in automatic seizure detection.

Conclusions:

• The developed CE-stSENet framework offers a powerful and effective solution for automatic epileptic seizure detection.
• Simultaneous analysis of spectral and temporal domains significantly enhances the recognition of epileptic EEG signals.
• The proposed loss function effectively addresses data scarcity and overfitting challenges in seizure detection.