Influence of Time-Series Normalization, Number of Nodes, Connectivity and Graph Measure Selection on Seizure-Onset Zone Localization from Intracranial EEG

Pieter Van Mierlo,Octavian Lie,Serge Vulliémoz,Ana Coito,Willeke Staljanssens

doi:10.1007/s10548-018-0646-7

Abstract

We investigated the influence of processing steps in the estimation of multivariate directed functional connectivity during seizures recorded with intracranial EEG (iEEG) on seizure-onset zone (SOZ) localization. We studied the effect of (i) the number of nodes, (ii) time-series normalization, (iii) the choice of multivariate time-varying connectivity measure: Adaptive Directed Transfer Function (ADTF) or Adaptive Partial Directed Coherence (APDC) and (iv) graph theory measure: outdegree or shortest path length. First, simulations were performed to quantify the influence of the various processing steps on the accuracy to localize the SOZ. Afterwards, the SOZ was estimated from a 113-electrodes iEEG seizure recording and compared with the resection that rendered the patient seizure-free. The simulations revealed that ADTF is preferred over APDC to localize the SOZ from ictal iEEG recordings. Normalizing the time series before analysis resulted in an increase of 25–35% of correctly localized SOZ, while adding more nodes to the connectivity analysis led to a moderate decrease of 10%, when comparing 128 with 32 input nodes. The real-seizure connectivity estimates localized the SOZ inside the resection area using the ADTF coupled to outdegree or shortest path length. Our study showed that normalizing the time-series is an important pre-processing step, while adding nodes to the analysis did only marginally affect the SOZ localization. The study shows that directed multivariate Granger-based connectivity analysis is feasible with many input nodes (> 100) and that normalization of the time-series before connectivity analysis is preferred.

Highlights

In approximately 15–25% of the epilepsy patients in the presurgical evaluation, intracranial EEG monitoring is necessary to obtain additional localization information about the seizure-onset zone (SOZ) and eloquent cortex (Carrette et al 2010)
The statistical analysis of the area under the curve (AUC) values revealed that the normalization, the channel selection, the signal-to-noise ratio (SNR) and the chosen connectivity measures all significantly influence the connectivity analysis
The Adaptive Directed Transfer Function (ADTF) measures have a higher AUC than the Adaptive Partial Directed Coherence (APDC) measures

Summary

Introduction

In approximately 15–25% of the epilepsy patients in the presurgical evaluation, intracranial EEG (iEEG) monitoring is necessary to obtain additional localization information about the seizure-onset zone (SOZ) and eloquent cortex (Carrette et al 2010). Because of the limited spatial sampling of iEEG, a clear hypothesis about the EZ must be available prior to electrode implantation. The identification of the SOZ from iEEG is done visually by the epileptologist. This is a time consuming, labor intensive task that requires much expertise and suffers from interpreter dependency. Functional brain connectivity is defined as the study of temporal correlations between spatially distinct neurophysiological events (Friston et al 1993).

Methods

Results

Conclusion