Spatio-temporal characterizations of non-linear changes in intracranial activities prior to human temporal lobe seizures M. Le Van Quyen, 1 C. Adam, 1,2 J. Martinerie, 1 M. Baulac, 1,2 S. Cle Âmenceau 1,2 and F. Varela 1 1 Laboratoire de Neurosciences Cognitives et Imagerie Ce Âre Âbrale, CNRS UPR 640, University of Paris VI, Ho Ãpital de la Salpe Ãtrie Áre, 47 Blvd. de l'Ho Ãpital, 75651 Paris cedex 13, France 2 Unite  d'Epileptologie, Ho Ãpital de la Salpe Ãtrie Áre, 47 Blvd. de l'Ho Ãpital, 75651 Paris cedex 13, France Keywords: long-term non-stationarity, non-linear analysis, pre-ictal phase, seizure anticipation, temporal lobe epilepsy Abstract Recent studies have shown that non-linear analysis of intracranial activities can detect a `pre-ictal phase' preceding the epileptic seizure. Nevertheless, the dynamical nature of the underlying neuronal process and the spatial extension of this pre-ictal phase still remain unknown. In this paper, we address these aspects using a new non-linear measure of dynamic similarity between different parts of intracranial recordings of nine patients with medial temporal lobe epilepsy recorded during transitions to seizure. Our results con®rm that non-linear changes in neuronal dynamics allow, in most cases (16 out of 17), a seizure anticipation several minutes in advance. Furthermore, we show that the spatial distribution of pre-ictal changes often involves an extended network projecting beyond the limits of the epileptogenic region. Finally, the pre-ictal phase could frequently (13 out of 17) be characterized with a marked shift toward slower frequencies in upper delta or theta frequency range. Introduction Human focal epilepsy can be viewed as a dynamic process in space and time with several more or less distinct phases, including the interictal and ictal states. The processes mediating the transitions from the interictal to the ictal state remain poorly understood in human epilepsy. When recorded by intracranial electrodes, the transition to seizure is characterized by sudden changes, e.g. high- amplitude periodic spikes or low-voltage fast activity (Engel, 1989). Such patterns are identi®ed by the visual inspection of intracranial tracings or traditional (frequency-based) signal analysis several seconds before clinical manifestations, and there is no other known clue in advance predicting the occurrence of a seizure (Rogowski etal., 1981; Alarcon etal., 1995; Osorio etal., 1998). In particular, the quanti®cation of the spike rate prior to seizure onset was investigated with negative results (Katz etal., 1991). The purpose of the present study is to further investigate the intracranial recordings with a new non-linear method to characterize more hidden changes in neuronal activity preceding the visually discernible seizure onset. Indeed, two recent reports (Martinerie etal., 1998; Lehnertz & Elger, 1998) indicate that non-linear time series analyses are more sensitive in detecting changes in neuronal activity than methods referring to spectral parameters only. These studies reported that 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 re¯ects a transition from high to low complexity of the neuronal dynamics. Nevertheless, low dimensionality has been shown to be dif®cult to identify in EEG signals (Jeong etal., 1999), even during the seizure (Theiler, 1995). Consequently, the dynamic nature of the pre-ictal transition still remains problematic. Furthermore, previous studies have exclusively focused their attention on the epileptogenic zone (Martinerie etal., 1998) or on time-independent measures (e.g. the `neuronal com- plexity loss' in Lehnertz & Elger, 1995). A more complete picture in time and space of the pre-ictal changes should be determined, and in particular, compared with the location and extent of the epileptogenic zone based on the clinical investigations. To answer some of these questions, we have recently proposed a new non-linear methodology to track long-term qualitative changes in the neuronal dynamics (Le Van Quyen etal., 1999b). Indeed, our previous study (Martinerie etal., 1998) supports the view that pre- ictal changes can be interpreted in terms of a slow dynamic which does not vary signi®cantly over a shorter time scales of a few seconds, but exhibits variations over a larger time scale. To obtain a detailed picture about this long-time non-stationarity, our proposed strategy consists of ®rst dividing the recording into windows of tens of seconds that can be regarded as quasi-stationary, and then quantifying the extent to which the underlying dynamics differ between (distant) pairs of windows using a measure of dynamic similarity (Fig.1). This idea to use relative measures between segments of a long sequence for non-stationarity testing has been brought up in a number of recent theoretical works (Manuca & Savit, 1996; Casdagli, 1997), showing a greater discriminatory power than previous non-linear techniques (Schreiber, 1997). In the present study, we apply this method for the study of a group of patients with medial temporal lobe epilepsy recorded with multicontact intracranial electrodes during transitions to seizure Correspondence: Dr M. Le Van Quyen, as above. E-mail: lenalm@ext.jussieu.fr Received 14 October 1999, revised 25 January 2000, accepted 29 February 2000 European Journal of Neuroscience, Vol. 12, pp. 2124±2134, 2000 Ó European Neuroscience Association