Frequency Band Personalization for Seizure Network Analysis in Multifocal Patients

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Stereo-electroencephalography (SEEG) is crucial for pre-surgical evaluation in multifocal epilepsy patients.
• Responsive neurostimulation (RNS) therapy requires precise seizure detection for optimal implantation and adjustment.
• Individualized ictal signatures from SEEG data are vital for seizure network modeling and early detection.

Purpose of the Study:

• To propose a data-driven methodology for personalizing frequency band selection in SEEG-based seizure network modeling.
• To improve the accuracy and robustness of seizure detection for responsive neurostimulation therapy.
• To identify optimal frequency bands that best distinguish seizure onset zones in multifocal epilepsy.

Main Methods:

• Developed a directed seizure network using SEEG data with spectral edges characterized by directed transfer function.
• Applied surrogate data analysis to ensure the statistical significance of network estimations.
• Utilized subgraph density to identify discriminative frequency ranges by maximizing differences between seizure onset zones and other brain regions.

Main Results:

• The personalized frequency band selection method achieved an Area-Under score of 0.94 in seizure classification.
• Outperformed standard frequency bands in differentiating ictal from pre-ictal states.
• Demonstrated consistent, type-specific spectral patterns across different frequency-domain network metrics.

Conclusions:

• The proposed data-driven approach effectively personalizes frequency band selection for seizure network modeling.
• This method enhances the accuracy of seizure detection, aiding in responsive neurostimulation therapy.
• Individualized spectral patterns offer valuable insights into seizure network dynamics in multifocal epilepsy.