Classification of epilepsy types through global network analysis of scalp electroencephalograms

UnCheol Lee, Seunghwan Kim, and Ki-Young Jung

Phys. Rev. E 73, 041920 – Published 17 April 2006

Abstract

Epilepsy is a dynamic disease in which self-organization and emergent structures occur dynamically at multiple levels of neuronal integration. Therefore, the transient relationship within multichannel electroencephalograms (EEGs) is crucial for understanding epileptic processes. In this paper, we show that the global relationship within multichannel EEGs provides us with more useful information in classifying two different epilepsy types than pairwise relationships such as cross correlation. To demonstrate this, we determine the global network structure within channels of the scalp EEG based on the minimum spanning tree method. The topological dissimilarity of the network structures from different types of temporal lobe epilepsy is described in the form of the divergence rate and is computed for 11 patients with left (LTLE) and right temporal lobe epilepsy (RTLE). We find that patients with LTLE and RTLE exhibit different large scale network structures, which emerge at the epoch immediately before the seizure onset, not in the preceding epochs. Our results suggest that patients with the two different epilepsy types display distinct large scale dynamical networks with characteristic epileptic network structures.

Received 17 May 2005

DOI:https://doi.org/10.1103/PhysRevE.73.041920

Authors & Affiliations

UnCheol Lee^* and Seunghwan Kim^†

Asia Pacific Center for Theoretical Physics & Nonlinear Complex Systems Laboratory, National Core Research Center on System Biodynamics, Department of Physics, Pohang University of Science and Technology, Pohang, 790-784, Korea

Ki-Young Jung^‡

Department of Neurology, Samsung Medical Center, Sungkyungkwan University School of Medicine, 50 Ilwon-dong, Gangnam-Gu, Seoul, 135-710 Korea

^*Electronic address: uclee@postech.ac.kr
^†Electronic address: swan@postech.ac.kr
^‡Electronic address: kyjung@smc.samsung.ac.kr

Article Text (Subscription Required)

Click to Expand

References (Subscription Required)

Click to Expand

Issue

Vol. 73, Iss. 4 — April 2006

Reuse & Permissions

Access Options

Author publication services for translation and copyediting assistance advertisement

Images

Figure 1
(Color online) The mean cross correlations between channels, $\bar{C (i, j)}$ , and their standard deviations are calculated for two types of epilepsy patients (six with LTLE and five with RTLE), which are averaged over the five segmented EEG periods. (a) The time evolution of the mean cross correlations in two epilepsy groups (six with LTLE and five with RTLE) with the standard deviation bar. (b), (c) The average difference of $\bar{C (i, j)}$ between window 4 and window 5, in LTLE and RTLE groups, respectively. The white indicates the pair of channels above the upper 10%, while the black indicates those below the lower 10%. [LF (left frontal): FP1,F7,F3,AF7,F5,FC5,FT7,F9,FT9. RF (right frontal): Fp2,F8,F4,AF8,F6,FC6,FT8,F10,FT10. LT (left temporal): T7,TP7,T9,TP9. RT (right temporal): T7,TP8,T10,TP10. LC (left central): C3,C5,CP5. RC (right central): C4,C6,CP6. LP (left parietal): P7,P3,P9. RP (right parietal): P4,P8,P10. O (occipital): O1,O2. Z(zero line): Cz,Pz].Reuse & Permissions
Figure 2
(a) MST of a LTLE patient. (b) MST of a RTLE patient. $D_{(X, Y)}$ quantifies how far the neighborhood channels TP7, FT7, FC5, and FT9 of the T7 in the MST of a LTLE are located in the MST of a RTLE patient. The subplots are the dendrograms of the 43 EEG channels. The bold lines indicate the distance between the neighborhood channels and the reference channel. The MST gives information on the linkage of channels, while the dendrogram gives information on the distances between the channels.Reuse & Permissions
Figure 3
Dendrograms of the $40 \min$ scalp EEGs are segmented into five EEG periods of $8 \min$ . The dendrograms are constructed with 11 MSTs of all epilepsy patients at each segmented EEG period, respectively. (a)–(e) are dendrograms constructed with $D_{(X, Y)}$ as a measure of dissimilarity. (a) The first EEG epoch (from 32 to $40 \min$ ), (b) the second EEG epoch (from 24 to $32 \min$ ), (c) the third EEG epoch (from 16 to $24 \min$ ), (d) the fourth EEG epoch (from 8 to $16 \min$ ), (e) the fifth EEG epoch (from 0 to $8 \min$ before seizure onset), and (f) the dendrogram constructed with the survival ratio as a measure of dissimilarity, at the fifth EEG epoch.Reuse & Permissions
Figure 4
(a)–(f) Dendrograms of the fifth EEG epoch segmented into six EEG periods of $10 s$ . Each dendrogram is constructed on the segmented EEG duration which begins at (a) 60, (b) 50, (c) 40, (d) 30, (e) 20, and (f) $10 s$ before the seizure onset.Reuse & Permissions
Figure 5
The discriminative weights (DWs) of two epilepsy types based on two different dissimilarity measures are presented at different perturbation levels. The perturbation level $R (k)$ is given from 5% to 100% in proportion to the mean standard deviation of all normalized channel EEGs.Reuse & Permissions

Physical Review E

covering statistical, nonlinear, biological, and soft matter physics