EVALUATION OF METHODS FOR DETERMINING A PRE-ICTAL STATE IN SCALP-EEG MEASUREMENTS
W. De Clercq*, P. Lemmerling*, S. Van Huffel*, W. Van Paesschen**
* Department of Electrical Engineering, ESAT-SCD(SISTA),Katholieke Universiteit Leuven, Leuven, Belgium
** Department of Neurology, University Hospital Gasthuisberg, Leuven, Belgium
Abstract: Anticipation of epileptic seizures from non- linear changes in standard EEG recordings has been reported in 25 of 26 recordings (96%) of mesial temporal lobe complex partial seizures (mTLCPS) [3]. We were not able to replicate these findings and did not detect a pre-ictal state using nonlinear time series analysis of scalp-EEG recordings in 12 patients (0%) with mTLCPS. Possible reasons for the discrepancies between our findings and those reported in the literature are discussed.
Keywords: Epilepsy, Nonlinear Analysis, Pre-ictal state, Scalp-EEG
Introduction
The evolution of a seizure involves not just two states – interictal and ictal – but also a pre-ictal transitional phase of several minutes that differs dynamically from the other two. It has been widely conjectured that this pre-ictal process reflects a transition from high to low complexity of the neuronal dynamics. The goal of this research was to determine whether scalp-EEG measure- ments can be used to identify a pre-ictal state in pa- tients suffering from mTLCP epilepsy. It has been shown previously that nonlinear analysis techniques (correlation dimension [1], Lyapunov exponent [2], sim- ilarity measure [3]) can be used to determine a pre-ictal state in intracranial EEG measurements. Recent experi- ments on scalp-EEG measurements indicate that the similarity measure is able to determine a pre-ictal state of TLCPS and neocortical partial seizures (NPS) [3,4].
Our goal was to replicate the work of Le Van Quyen and colleagues [3,4] and to detect a pre-ictal state of mTLCPS using nonlinear time series analysis of scalp- EEG recordings.
Materials
We studied a sample of 13 patients who required contin- uous scalp-EEG recording and video monitoring to lo- calize seizure onsets. Twelve consecutive patients had mTLCPS associated with hippocampal sclerosis, who were operated on, and became seizure free for more than two years. The EEG were recorded on 22-channel OSG EEG recorders. The electrodes were placed ac- cording to the extended international 10-20 System. The scalp potentials were sampled at 250 Hz. To avoid changes induced by variation of arousal states only seizures in which the patients were awake for the entire recording were studied. All the long-term recordings in- cluded the 60 minutes before the clinical seizure. The start of the clinical seizure was noted by the epileptolo- gist electroencephalographer. The thirteenth patient was a patient with NPS and had simultaneous scalp and in- tracranial EEG recordings. The data of this patient was provided to us by the authors of [4].
Methods
We use two different nonlinear methods to try to iden- tify the dynamical changes towards a seizure:
correlation dimension [1]: A reliable estimation of the correlation dimension requires a large number of data points. However, a compromise has to be achieved be- tween the requirements for a sufficiently long EEG win- dow and stationarity. For the latter it can be assumed that an EEG window of a duration of tens of seconds can be regarded as quasi-stationary, depending on the patient’s behavioural state [5]. We divided the EEG recording in successive windows of 30 seconds. After reconstructing the dynamics of each EEG window with the method of delays [6], the correlation integral [7] was calculated and the correlation dimension was estimated for each window. The temporal evolution of the correla- tion dimension was monitored by plotting the correla- tion dimension versus the window number.
similarity measure [3]: The recording was segmented into evaluation windows of 30 sec. For each window we made a phase space representation using the time inter- vals between positive-going crossings. The reference dynamics of the non-seizure state was chosen approxi- mately one hour before the seizure. The duration of the reference EEG recording is 5 min and this reference contains all common features of interictal activity. The dynamics of the subsequent evaluation windows are compared to the dynamics of the reference window us- ing the similarity measure, which is based on the cross- correlation integral [3]. If the EEG is stationary, the similarity measure has a value close to 1. If changes in the dynamical state occur, the similarity measure de- creases below 1. The similarity measure between the reference window and the test window was calculated and the similarity profile was plotted.
Results
The similarity measure was not able to detect pre-ictal changes in the scalp-EEG recordings of all 12 patients with mTLPS (figure 1). However, in both intracranial and scalp-EEG recordings of the patient with NPS the similarity measure could identify a pre-ictal state (figure 2). The correlation dimension was unable to detect a pre-ictal state in all 13 patients.
Figure 1: Analysis of a scalp-EEG channel before a mTLCPS. The seizure lasted from window number 180 to 187 (S). The reference state was chosen from win-
dows 1 to 10 (R). The similarity measure stayed high until the start of the epileptic seizure. At the start of the epileptic seizure the similarity measure suddenly dropped. No sustained deviation beyond the statistical threshold (horizontal line) up to the onset of the seizure was recorded. A pre-ictal state could not be defined.
Figure 2: Analysis of a scalp-EEG channel before a NPS. Different seizures occurred between window num- bers 186-211 (S). The reference state was chosen here from windows 1 to 10 (R). A sustained deviation be- yond the statistical threshold (horizontal line) up to the onset of the seizure was recorded. A pre-ictal state could be defined beginning 21 minutes before seizure onset.
Discussion
The detection of a pre-ictal state in the patient with NPS ensures us that the similarity measure was correctly implemented and that the similarity measure has the potential for determining a pre-ictal state in scalp EEG- recordings. Nevertheless, we were unable to detect a pre-ictal state in mTLCPS using nonlinear time series analysis of scalp-EEG recordings in 12 patients (0%).
Possible reasons for the discrepancies between our findings and those reported in the literature [3] could be:
1) the evolution towards a seizure does not always contain a pre-ictal transitional phase, 2) technical differences may exist between EEG recorders, 3) the scalp-EEG recordings need additional preprocessing.
Conclusions
We were unable to detect a pre-ictal state in mTLCPS using nonlinear time series analysis of scalp-EEG recordings. We were not able to replicate the robust findings of Le Van Quyen and colleagues
Acknowledgments
WDC and PL are supported by F.W.O. Research spon- sored by Belgian Programme IUAP V-10-29, Flemish GOA project MEFISTO-666, KU Leuven- project IDO/99/03, FWO projects G.0269.02 and G.0407.02. I also want to thank the authors of [4] for providing us the data of the patient with NPS.
REFERENCES
[1] C.E. Elger, G. Widman, R.Andrzeja et al. Nonlinear EEG analysis and its potential role in epileptology, Epilepsia, 41, Suppl.3, pp. 34-38, 2000.
[2] L.D. Iasemidis, J.C. Sackellares, The evolution with time of the spatial distribution of the largest Lyapunov exponent of the human epileptic cortex, In Duke D., Pritchard W. (eds.), Measuring chaos in the brain, pp.
49-82. World Scientific, Singapore, 1991.
[3] M. Le Van Quyen, J. Martinerie , V. Navarro, et al.
Anticipation of epileptic seizures from standard EEG recordings, Lancet, 357, pp. 183-188, 2001.
[4] V. Navarro, J. Martinerie, M. Le Van Quyen et al.
Seizure anticipation in human neocortical partial epilepsy, Brain (2002), 125, pp. 640-655.
[5] F.H. Lopes da Silva, EEG analysis: theory and prac- tice. In : E. Niedermayer and F.H. Lopes da Silva (eds.), Electroencephalography, Baxic Principles, Clinical Ap- plications and related fields. Urban and Swarzenberg, Baltimore, MD, pp. 871-897, 1987.
[6] F. Takens, Detecting strange attractors in turbulence In: D.A. Rand and L.S. Young (eds.), Lecture notes on Mathematics 898. Springer, Berlin, Heidelberg, New York, pp. 366-381.
[7] P. Grassberger, I. Procaccia, Characterisation of strange attractors, Phys.Rev.Lett., 50, pp.346-349, 1983.
