Abstracts

Rationale: There is currently an increasing interest in developing refined analysis methods that can help to understand the relationship between structure and function in complex systems. Recent studies show that new insights into normal and disturbed brain functioning can be achieved by considering the brain as a complex network of interacting dynamical systems. A number of analysis methods already allow one to characterize statistical and spectral properties of functionally (e.g. via EEG or MEG recordings) defined networks, and the mean path length L and the cluster coefficient C have been widely used to characterize complex networks. We here investigated whether these network characteristics can help to distinguish between normal and disturbed brain functioning. Specifically, we address the question whether these network characteristics differ in patients with focal epilepsies and in age-matched controls. Methods: We recorded non-invasively the EEG from patients suffering from mesial temporal lobe epilepsies (MTLE; 11 patients) or neocortical lesional epilepsies (NLE; 10 patients) and from 21 age-matched controls during different behavioral conditions (eyes open vs. eyes closed; 15 min each). Using a moving-window approach (duration of each window: 16 s corresponding to 4096 data points; no overlap) we calculated, for each channel combination, the mean phase coherence as a measure for phase synchronization. Phases were obtained by applying either the Hilbert-transform (broad-band signals) or the wavelet transform, which allowed us to specifically concentrate on interactions in the δ, θ, α, and β-band. Using these measures we constructed, for each window, a network and calculated the averaged mean path length L and the cluster coefficient C for each condition. Results: Depending on the type of the constructed network (weighted or unweighted) we observed clear-cut difference in statistical network characteristics for different behavioral conditions and for the control and the patient group. The differentiation of patients and controls was most pronounced for weighted functional networks, and best discrimination could be achieved in the δ and β-band. L was smaller while C was larger for the patient group. Differentiation between different behavioral states was most pronounced for unweighted functional network and was present in all analyzed frequency-bands, except the θ-band. Both L and C were larger during the eyes closed condition. Conclusions: Our findings indicate that functionally defined weighted and unweighted networks reflect different aspects of normal and disturbed brain functioning. Different behavioral states are characterized by different topological properties of the networks, while normal and disturbed brain functioning are characterized by both an increased local and global connectedness. Statistical properties such as the mean path length or the cluster coefficient can be helpful in characterizing epileptic brain networks, since they allow an efficient compression of the complex information content in multichannel EEG recordings.