Epileptic network identification: insights from dynamic mode decomposition of sEEG data

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Medically-refractory epilepsy necessitates precise identification of the epileptogenic zone (EZ) for surgical intervention.
• Stereoelectroencephalography (sEEG) provides intracranial recordings but its data quantification and interpretation pose clinical and computational challenges.
• Data-driven approaches offer novel methods for pattern identification in complex, high-dimensional sEEG data.

Purpose of the Study:

• To apply an unsupervised data-driven algorithm, dynamic mode decomposition (DMD), to sEEG recordings for improved EZ identification.
• To develop and validate novel visualization tools (dynamic modal maps - DMMs, and higher-frequency mode-based norm index - MNI) for sEEG data analysis.
• To assess the concordance of DMD-derived metrics with clinical sEEG findings and surgical outcomes in epilepsy patients.

Main Methods:

• Dynamic Mode Decomposition (DMD) was employed to approximate nonlinear sEEG data dynamics, extracting coherent modes representing signal features.
• DMD was adapted to generate dynamic modal maps (DMMs) across frequency sub-bands to visualize epileptiform dynamics.
• A static EZ localization metric, the higher-frequency mode-based norm index (MNI), was developed and DMM/MNI maps were validated against clinical data.

Main Results:

• DMD proved most effective in higher frequency bands (gamma, beta), successfully identifying EZ contacts.
• Combined interpretation of DMM and MNI maps accurately captured the spatiotemporal evolution of seizure networks.
• The DMD-based method showed strong concordance with clinical sEEG results and post-surgical seizure-freedom across all five patients.

Conclusions:

• This study demonstrates the first application of DMD for sEEG data analysis in epilepsy.
• The integration of DMD, DMMs, and MNI offers a powerful data-driven approach to enhance sEEG interpretation.
• This neuroengineering and machine learning methodology supports traditional workflows for epilepsy surgical decision-making.