E2SGAN: EEG-to-SEEG translation with generative adversarial networks

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Stereoelectroencephalography (SEEG) provides high-quality brain data crucial for epilepsy presurgical assessments.
• SEEG is invasive, limiting its application to select epilepsy patients.
• Non-invasive Electroencephalography (EEG) offers a more accessible alternative but lacks SEEG's spatial resolution.

Purpose of the Study:

• To develop a framework for synthesizing SEEG signals from non-invasive EEG signals.
• To improve presurgical assessment for epilepsy by enabling SEEG data generation from EEG.
• To enhance the accuracy and applicability of brain signal analysis for epilepsy diagnosis.

Main Methods:

• A channel matching strategy considering signal correlation and spatial distance.
• Development of the EEG-to-SEEG generative adversarial network (E2SGAN) for signal synthesis.
• Introduction of instantaneous frequency spectra and Correlative Spectral Attention (CSA) for enhanced signal representation and discrimination.
• Implementation of Weighted Patch Prediction (WPP) for temporal robustness.

Main Results:

• E2SGAN demonstrated superior performance compared to baseline methods in both temporal and frequency domains on real patient data.
• The synthesized SEEG signals showed potential in capturing abnormal discharges preceding epileptic seizures.
• The framework effectively translates non-invasive EEG data into high-fidelity SEEG signal representations.

Conclusions:

• The proposed E2SGAN framework offers a viable method for synthesizing SEEG signals from EEG.
• This advancement could expand the utility of non-invasive EEG in epilepsy presurgical planning.
• The synthesized data holds promise for improving the detection of pre-epileptic seizure activity.