Extraction and Visualization of Functional Brain Connectivity Networks from Eeg and Fmri Data

We study the extraction and visualization of brain connectivity networks from EEG and fMRI data. The method is based upon the construction of functional unit maps. It is indicated how this representation can be used for comparing brain networks in the original network representation. INTRODUCTION Various types of brain connectivity are distinguished. Structural or anatomical connectivity refers to the macroscopic neural pathways or structural fiber tracts linking together different parts of the brain. Functional connectivity on the other hand, looks at statistical dependencies (correlation, coherence) between different parts of the brain during the performing of a task or during rest. This is often measured using functional MRI, but can for example also be obtained using EEG, MEG or PET. Finally, effective connectivity concerns causal interactions between distinct units within a nervous system (Friston, 1994). EEG COHERENCE NETWORK VISUALIZATION Electrical potentials generated within the brain can be measured with electrodes at the scalp during an EEG recording. For EEG analysis, one often studies activity in various frequency bands, such as alpha, beta, theta or delta bands. Nowadays, the number of electrodes can be as large as 128 or even 512; in that case one speaks of multichannel or high-density EEG. Brain networks based on EEG coherence can be modelled in terms of functional units and their interrelations. A functional unit (FU) represents a set of electrodes that are spatially close and record pairwise significantly coherent signals (ten Caat et al, 2008). Such FU maps can be constructed for different frequency bands or for different subjects. Figure 1 (left) shows an example where brain responses were collected from three subjects using 119 scalp electrodes during a so-called P300 experiment in which participants had to count target tones of 2000Hz (probability 0.15), alternated with standard tones of 1000Hz (probability 0.85) which were to be ignored. BRAIN NETWORK COMPARISON The FU method has been used to compare different EEG networks in the original network representation, see Crippa et al. (2011). The approach is based on graph averaging for attributed relational graphs and results in an average FU-map and a dissimilarity measure for individual FU-maps. Figure 1 (right) shows an example. The approach was extended for resting state fMRI data by Crippa and Roerdink (2011). The FU map representation can be Porto/Portugal, 22-27 July 2012 2 used to study multiple subjects under different conditions, such as mental fatigue or brain disease. Figure 1 Left: FU maps for multichannel EEG coherence visualization. To each electrode a Voronoi cell is associated and all cells belonging to an FU have a corresponding color. Lines connect FU centers if the inter-FU coherence exceeds a significance threshold. The color of a line depends on the inter-FU coherence. Shown are FU maps with FUs larger than 5 cells, for the 1-3Hz EEG frequency band (top row) and for 13-20Hz (bottom row), for three datasets. Right: Two FU-maps, A and B, and their average FU-map C. Spatial clusters of coloured cells correspond to FUs, white cells do not belong to any FU. Circles represent the barycenters of the FUs and are connected by edges whose colour indicates their inter-node coherence. In C, colour saturation is proportional to the multiplicity of a cell (electrode) in a graph node, and the size of the nodes reflects their occurrence in the input graphs. Only statistically significant edges are included. Dissimilarities between A/B and C are shown.