Localizing seizure onset zone by convolutional transfer entropy from iEEG

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Signal Processing

Background:

• Automatic seizure onset zone (SOZ) localization aids epilepsy therapy by providing objective results.
• Transfer entropy is a common information-theory measure for SOZ localization but can yield temporally variable results.
• Human interpretation is often required to identify the most reliable SOZ from transfer entropy outputs.

Purpose of the Study:

• To develop a novel method for objective and stable SOZ localization over extended periods.
• To reduce the number of potential SOZ results, simplifying clinical decision-making.
• To enhance the accuracy and reliability of SOZ identification in epilepsy patients.

Main Methods:

• Utilized traditional transfer entropy to identify potential SOZ regions.
• Applied a 3D convolution method to integrate spatial and temporal information from electroencephalogram (iEEG) signals.
• Employed a connected component analysis to extract stable SOZ blocks.

Main Results:

• Achieved 100% sensitivity in SOZ localization during preliminary experiments on iEEG data.
• Reported a false positive rate of 1.79% for the proposed SOZ localization method.
• Demonstrated the method's ability to output a limited number of stable SOZ results.

Conclusions:

• The proposed 3D convolution and connected component method offers a robust solution for SOZ localization.
• This technique enhances the objectivity and stability of SOZ identification, aiding clinical epilepsy management.
• The method shows high efficacy, suggesting its potential for widespread clinical application.