Mapping Scalp to Intracranial EEG using Generative Adversarial Networks for Automatically Detecting Interictal...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Scalp electroencephalograms (EEGs) are crucial for diagnosing brain disorders but suffer from skull attenuation and artifacts.
• Intracranial EEGs (iEEGs) offer superior signal quality, free from attenuation and artifacts, providing a clearer view of brain activity.
• The discrepancy between sEEG and iEEG limits diagnostic accuracy and requires improved signal processing methods.

Purpose of the Study:

• To enhance the performance of scalp EEGs (sEEGs) by developing a method to map sEEG signals to intracranial EEG (iEEG) data.
• To improve the detection of interictal epileptiform discharges (IEDs) by leveraging the strengths of both sEEG and iEEG.
• To introduce a computationally efficient deep learning approach for EEG signal enhancement.

Main Methods:

• Development of a deep neural network utilizing a generative adversarial network (GAN) architecture.
• Training the GAN to estimate sEEG signals from simultaneously recorded iEEG data.
• Application of the trained model to real-world sEEG and iEEG data from epileptic patients for IED detection.

Main Results:

• The proposed GAN-based method successfully estimates sEEG from iEEG signals.
• The method achieves 76% accuracy in detecting interictal epileptiform discharges (IEDs), surpassing current state-of-the-art techniques.
• The developed approach demonstrates a computational complexity at least twelve times lower than existing methods.

Conclusions:

• Deep learning, specifically GANs, can effectively bridge the gap between sEEG and iEEG, enhancing diagnostic capabilities.
• The proposed method offers a significant advancement in the accurate and efficient detection of IEDs.
• This technique holds promise for improving non-invasive brain disorder diagnosis and reducing computational burden in clinical settings.