Seizure-Onset Mapping Based on Time-Variant Multivariate Functional Connectivity Analysis of High-Dimensional...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Visual interpretation of intracranial EEG (iEEG) is standard for epilepsy surgery but is time-consuming and subjective.
• Multivariate functional connectivity using adaptive autoregressive (AR) modeling with Kalman filters can localize seizure onsets but is computationally expensive.
• Current methods are limited to analyzing fewer than 60 iEEG time-series due to computational demands.

Purpose of the Study:

• To evaluate two Kalman filter implementations of AR models for high-dimensional iEEG connectivity analysis.
• To assess the efficiency and accuracy of a simplified AR model compared to a known one for iEEG data.
• To determine the feasibility of data-driven multivariate connectivity estimation on large iEEG datasets.

Main Methods:

• Compared a standard multivariate adaptive AR model with a computationally efficient variant using simulated and real iEEG data.
• Utilized partial directed coherence as a multivariate connectivity estimator.
• Employed graph-theory index (outdegree) to map seizure onsets in patient data.

Main Results:

• Both AR models demonstrated high mapping accuracy on simulated seizures with low-to-moderate noise.
• The simplified Kalman filter model showed comparable performance to the established model.
• Both models successfully mapped real seizure onsets to the surgical resection volume in epilepsy patients.

Conclusions:

• The Kalman filter approach enables accurate, data-driven multivariate connectivity analysis of high-dimensional iEEG data.
• The computationally efficient model holds promise for clinical application in epilepsy surgery planning.
• This study supports the use of advanced signal processing for improved seizure localization.