Permutation Statistics for Connectivity Analysis between Regions of Interest in EEG and MEG Data

Fahimeh Mamashli,Jyrki Ahveninen,Tal Kenet,Matti Hämäläinen,Sheraz Khan

doi:10.1038/s41598-019-44403-z

Abstract

Connectivity estimates based on electroencephalography (EEG) and magnetoencephalography (MEG) are unique in their ability to provide neurophysiologically meaningful spectral and temporal information non-invasively. This multi-dimensional aspect of the MEG/EEG based connectivity increases the challenges of the analysis and interpretation of the data. Many MEG/EEG studies address this complexity by using a hypothesis-driven approach, which focuses on particular regions of interest (ROI). However, if an effect is distributed unevenly over a large ROI and variable across subjects, it may not be detectable using conventional methods. Here, we propose a novel approach, which enhances the statistical power for weak and spatially discontinuous effects. This results in the ability to identify statistically significant connectivity patterns with spectral, temporal, and spatial specificity while correcting for multiple comparisons using nonparametric permutation methods. We call this new approach the Permutation Statistics for Connectivity Analysis between ROI (PeSCAR). We demonstrate the processing steps with simulated and real human data. The open-source Matlab code implementing PeSCAR are provided online.

Highlights

If the activation inside each regions of interest (ROI) is distributed uniformly across the entire ROI, one could average the activity inside the ROI and conduct the functional connectivity analysis on the averaged time series
We demonstrate that when the SNR is low, Permutation Statistics for Connectivity Analysis between ROI (PeSCAR) offers more statistical power than alternative conventional averaging approach, in which time series is averaged over vertices across ROI and the connectivity is estimated on the averaged time series and the cluster statistics in time and frequency is used for contrast
A generally accepted method for ROI-based functional connectivity analysis of MEG/EEG data has so far been lacking, and previous studies have been limited to using tailored heuristic approaches

Summary

Introduction

This approach relies on a robust estimate of activity in the ROIs and may not be applicable when weak cognitive effects are investigated. In many situations the ROIs are selected using existing anatomically and functionally defined parcellations These parcellations are agnostic to variations of the effects of interest within the ROI. For discontinuous effects, sub-division of ROIs would increase the spatial specificity and statistical power. To correct for multiple comparisons across the sub-ROIs, we employed a non-parametric permutation method[26], which has been successfully applied before[15,27,28,29,30,31,32,33,34,35,36] We name this method PeSCAR, which stands for Permutation Statistics for Connectivity Analysis between ROI. Using simulation and real data set we show the processing steps

Methods

Results

Conclusion