Content uploaded by Klaus Lehnertz
Author content
All content in this area was uploaded by Klaus Lehnertz on Apr 26, 2019
Content may be subject to copyright.
Is it possible to anticipate seizure onset by non-linear analysis of intracerebral EEG in human partial epilepsies?
Abstract
Detection of electrophysiological features preceding and indicative of imminent seizures in patients with epilepsy would be a major breakthrough with immense scientific and clinical implications. The definition of a "pre-ictal state" several minutes prior to seizure onset would open a new time window for studying mechanisms of seizure generation as well as for possible therapeutic interventions. In this review we present recent findings from nonlinear time series analysis of intracranially recorded EEG that may allow to forecast seizure onset in patients with partial epilepsy.
... Since the 1990s more successful attempts have been made to apply techniques from nonlinear dynamics to characterize, detect and anticipate imminent seizure activity in electrophysiological recordings (e.g. Iasemidis et al, 1990;Casdagli et al, 1996;Elger and Lehnertz, 1998;Hively et al, 1999;Le Van Quyen et al, 1998Lehnertz et al, 1998Lehnertz et al, , 1999Andrzejak, 2001b;Jerger et al, 2001;Savit et al, 2001;Van Drongelen et al 2003a). Although these studies suggest the feasibility of p. 4 of 60 09/15/15 seizure detection and prediction, it is also clear that the applied methodology has limitations. ...
... The latter study indicates that a drop in correlation dimension may occur up to 25 min prior to seizure onset. An overview of this group's work can be found in Lehnertz et al (1999Lehnertz et al ( , 2001. More recently, this group has become interested in measures of nonlinearity and nonstationarity (Andrzejak et al 2001a(Andrzejak et al , 2001bRieke et al, 2003). ...
... These mostly included phase synchronization methods, generalized synchronized methods, and phase lag index. [67][68][69][70] To understand epileptogenesis better, Aubert et al., [24] developed the concept of epileptogenic index, which is essentially the propensity of the brain to generate high frequency oscillations and the time for this area to get involved in the seizure. He found that only in 1/3 rd of cases with epilepsy there is a single epileptogenic zone, while in the rest cases; there was evidence of more than one area, which generated epilepsy. ...
... In fact, the epileptic neurons are now recognized to have a bi-stable nature during epileptogenesis and during interictal period. [5,23,24,36,37,[69][70][71] Thus, a small area of neurons, which epileptogenic stimulate more distant neurons, and through reentrant pathways forms a reverberating circuit (also called a ''node''). Such circuits progressively recruit more distant circuits forming larger and larger networks over a period of time [ Figure 2]. ...
A multidisciplinary approach is required to understand the complex intricacies of drug-resistant epilepsy (DRE). A challenge that neurosurgeons across the world face is accurate localization of epileptogenic zone. A significant number of patients who have undergone resective brain surgery for epilepsy still continue to have seizures. The reason behind this therapy resistance still eludes us. Thus to develop a cure for the difficult to treat epilepsy, we need to comprehensively study epileptogenesis. Till date, most of the studies on DRE is focused on undermining the abnormal functioning of receptors involved in synaptic transmission and reduced levels of antiepileptic drugs around there targets. But recent advances in imaging and electrophysiological techniques have suggested the role epileptogenic networks in the process of epileptogenesis. According to this hypothesis, the local neurons recruit distant neurons through complex oscillatory circuits, which further recruit more distant neurons, thereby generating a hypersynchronus neuronal activity. The epileptogenic networks may be confined to the lesion or could propagate to distant focus. The success of surgery depends on the precision by which the epileptogenic network is determined while planning a surgical intervention. Here, we summarize various modalities of electrophysiological and imaging techniques to determine the functionally active epileptogenic networks. We also review evidence pertaining to the proposed role of epileptogenic network in abnormal synaptic transmission which is one of the major causes of epileptiform activity. Elucidation of current concepts in regulation of synaptic transmission by networks will help develop therapies for epilepsy cases that cannot be managed pharmacologically.
... In the 1990s, independent laboratories found indications for the existence of a preictal state from nonlinear EEG analyses several minutes prior to clinical symptoms in patients implanted with intracranial electrodes during evaluation for epilepsy surgery in the temporal lobes (Iasemidis et al., 1990(Iasemidis et al., , 1997Elger andLehnertz, 1994, 1998;Lehnertz and Elger, 1998;Martinerie et al., 1998;Lehnertz et al., 1999;Le Van Quyen et al., 1999b, 2000Moser et al., 1999). These findings were further supported by studies indicating detectability of preictal states in neocortical epilepsies Navarro et al., 2002) and in animal models of epilepsy (Geva and Kerem, 1998;Widman et al., 1999;Lian et al., 2001;Sunderam et al., 2001) as well as by studies demonstrating seizure anticipation from scalp-EEG recordings (Iasemidis et al., 1997;Le Van Quyen et al., 2001a;Protopopescu et al., 2001). ...
... Since these increases are only detected by the non-linear association analysis, but not by linear GC analysis, it is likely that these increases in communication, take place in a non-linear fashion. Indeed epileptic seizures are often regarded to be of non-linear nature and signal analyses derived from the theory/mathematics of non-linear dynamics are proposed to be of particular value for the understanding of seizure generation mechanism (Le Van Quyen et al., 1999;Lehnertz et al., 1999;Litt and Echauz, 2002;Lopes da Silva et al., 2003b;Stefan and Lopes da Silva, 2013). Linear coupling changes were also detected pre-ictally: at 1.2 s prior to FCTS a phasic decoupling between the caudal RTN and layer 4 of the somatosensory cortex and rostral RTN and layer 5 of the somatosensory cortex, was observed, shortly followed by a phasic decoupling between cRTN and Po and cRTN and VPM. ...
Genetic rat models for childhood absence epilepsy have become instrumental in developing theories on the origin of absence epilepsy, the evaluation of new and experimental treatments, as well as in developing new methods for automatic seizure detection, prediction and or interference of seizures.
Various methods for automated off and on-line analyses of ECoG in rodent models are reviewed, as well as data on how to interfere with the spike-wave discharges by different types of invasive and non-invasive electrical, magnetic and optical brain stimulation. Also a new method for seizure prediction is proposed.
Many selective and specific methods for off- and on-line spike-wave discharge detection seem excellent, with possibilities to overcome the issue of individual differences. Moreover, electrical deep brain stimulation is rather effective in interrupting ongoing spike-wave discharges with low stimulation intensity. A network based method is proposed for absence seizures prediction with a high sensitivity but a low selectivity. Solutions that prevent false alarms, integrated in a closed loop brain stimulation system open the ways for experimental seizure control.
The presence of preictal cursor activity detected with state of the art time frequency and network analyses shows that spike-wave discharges are not caused by sudden and abrupt transitions but that there are detectable dynamic events. Moreover, they are preceded by precursors. These detectable dynamical changes in time-space-frequency characteristics might yield new options for seizure prediction and seizure control.
Copyright © 2015. Published by Elsevier B.V.
... Indeed epileptic seizures and possibly their pre-ictal preparation processes are often regarded to be of non-linear nature. Signal analyses derived from the theory/mathematics of non-linear dynamical systems are proposed to be of particular value for the understanding of seizure generation mechanism, potential preictal seizure preparation phases, as well as to possibilities of seizure prediction (Lehnertz et al., 1999;Le Van Quyen et al., 1999;Litt and Echauz, 2002;Lopes Da Silva et al., 2003a,b;Stefan and Lopes Da Silva, 2013). In line with this, non-linear changes have been reported for a different kind of seizure: Martinerie et al. (1998) reported changes in non-linear coupling 2-6 min before seizure onset between amygdala and posterior hippocampus in patients with temporal lobe epilepsy. ...
Network mechanisms relevant for the generation, maintenance and termination of spike-wave discharges (SWD), the neurophysiological hallmark of absence epilepsy, are still enigmatic and widely discussed. Within the last years, however, improvements in signal analytical techniques, applied to both animal and human fMRI, EEG, MEG, and ECoG data, greatly increased our understanding and challenged several, dogmatic concepts of SWD. This review will summarize these recent data, demonstrating that SWD are not primary generalized, are not sudden and unpredictable events. It will disentangle different functional contributions of structures within the cortico-thalamo-cortical system, relevant for the generation, generalization, maintenance, and termination of SWD and will present a new “network based” scenario for these oscillations. Similarities and differences between rodent and human data are presented demonstrating that in both species a local cortical onset zone of SWD exists, although with different locations; that in both some forms of cortical and thalamic precursor activity can be found, and that SWD occur through repetitive cyclic activity between cortex and thalamus. The focal onset zone in human data could differ between patients with varying spatial and temporal dynamics; in rats the latter is still poorly investigated.
Epilepsy is the second most common serious neurological disease after stroke. This disease affects approximately 50 million people worldwide and 50–70 cases per 100,000 in the developed countries. In approximately 40% of patients with so-called partial seizures, current medications are unable to control their symptoms. One of the most devastating aspects of epilepsy is the anxiety and apprehension associated with the inability to predict when a seizure will occur. The inability to predict the time of seizure onset also implies the need for continuous medication therapy with the associated continuous side effects. For a number of years, investigators and commercial interest groups have sought methods for early detection and anticipation of seizures so that ‘discontinuous’ therapies could be introduced (e.g., [1]). At the heart of most predictive efforts is the description and analysis of the cerebral electrical activity reflected in the electroencephalogram (EEG). The brain electrical activity of a patient with epilepsy shows abnormal and often rhythmic discharges during the seizure. This activity pattern is called an electrographic seizure. Between such electrographic seizures, short discharges (spikes) are also frequently observed in the EEG of these patients. Identification of these activity patterns in clinical practice has typically been a subjective process. The introduction of computer based instrumentation and analysis to the field of electroencephalography made evaluation of automated spike and seizure detection techniques possible (e.g., [2–4]). During the 1980s, the EEG during seizure activity was characterized using more complex measures such as those derived from chaos theory (e.g., [5, 6]). There were a number of ‘early’ reports of the successful application of frequency-domain template analysis and auto-regressive models to the problem of seizure prediction (e.g., [7, 8]). Unfortunately, the performance of these methods was either difficult to evaluate or the average anticipation time was only a few seconds. An interval of several seconds could fall within the uncertainty bounds of the clinical judgment against which the predictions were compared. Since the 1990s more successful attempts have been made to apply techniques from nonlinear dynamics to characterize, detect and anticipate imminent seizure activity in electrophysiological recordings (e.g., [9–14]; Lehnertz et al, 1998, 1999; [15–18]). Although these studies suggest the feasibility of seizure detection and prediction, it is also clear that the applied methodology has limitations. One of these limitations is that we do not yet know the underlying processes against which prediction algorithms should be validated. In spite of a vast amount of electrophysiological studies in the field of epilepsy, the exact mechanisms responsible for initiating or stopping seizures are unknown, meaning there is no generally accepted ‘gold-standard’ for the detection of the pre-seizure state. Therefore, prediction algorithms explore electrophysiological data sets that include seizures, and the seizure-prediction capability is assigned to a particular algorithm a posteriori, if it recognizes a change in the electrical activity prior to seizure onset. Despite the difficulties, the huge potential benefits of a practical and reliable seizure predictor for the lives of epilepsy sufferers have attracted many researchers to the problem, and much progress has been made. Here we summarize the strategies that have been taken to predict seizures via processing of EEG, evaluate the progress so far, and lay out what we consider the most notable challenges in the field.
Les contractions utérines sont contrôlées par deux phénomènes physiologiques: l'excitabilité cellulaire et la propagation de l'activité électrique utérine probablement liées aux hautes et basses fréquences de l‟electrohysterograme (EHG) respectivement. Toutes les études précédentes ont porté sur l'extraction de paramètres de la partie haute fréquence et n'ont pas montré un potentiel satisfait pour l'application clinique. L'objectif de cette thèse est l'analyse de propagation de l'EHG pendant la grossesse et le travail dans la vue de l'extraction des outils pour une application clinique. Une des nouveautés de la thèse est l‟enregistrement multicanaux à l'aide d‟une matrice d'électrodes 4x4 posée sur l'abdomen de la femme. Analyse monovariés visait à étudier les caractéristiques non linéaires des signaux EHG, analyses bivariées et multivariées ont été effectuées pour analyser la propagation des signaux EHG par la détection de la connectivité entre les signaux. Une augmentation de la non- linéarité associée par une synchronisation en amplitude et de désynchronisation en phase a été détectée. Les résultats indiquent plus de propagation au cours du travail que la grossesse et une augmentation de cette propagation avec les semaines de gestations. Les résultats montrent le potentiel élevé de paramètres de propagation dans le point de vue clinique tel que la détection du travail et de prédiction du travail prématuré. Finalement, nous avons proposé une nouvelle combinaison entre Séparation Aveugles de Sources et la Décomposition en Modes Empiriques pour débruiter les signaux EHG monopolaires comme un moyen possible d'augmenter le taux de classification de signaux grossesse et l'accouchement.
We present potential applications of nonlinear time series analysis techniques to electroencephalographic recordings (EEG) derived from epilepsy patients. Apart from diagnostically oriented topics including localization of epileptic foci in different anatomical locations during the seizure-free interval we discuss possibilities for seizure anticipation which is one of the most challenging aspects in epileptology.
We discuss key theoretical and practical issues related to the identification of transitions in dynamical systems in real-time. We focus on the choice of candidate measures and optimization of decision thresholds for candidate measures. To illustrate these issues we develop and test a procedure for identifying one particular transition, the end-point of seizures in two-channel electro-encephalographic data recorded during generalized tonic–clonic seizures. Data from twenty-eight seizures were available and used to develop and test the procedure in terms of the agreement between the computationally identified seizure end-point compared against the ratings of an expert clinical electroencephalographer. Generalizations to multivariate seizure onset detection and to seizure prediction are described.
Ziel: Unterscheidung der Low risk von normalen Neugeborenen. Methode: 7 Low risk Neugeborenen wurden mit 6 normalen Neugeborenen anhand der Herzfrequenzfluktuationen (HFF) und Atembewegungen (AB) sowie deren Koordinationen verglichen. Ergebnisse: Bei den Low risk Neugeborenen fanden wir signifikant höhere Kohärenz der HFF und AB, höhere Komplexität der AB in den beiden Schlafstadien sowie der HFF im aktiven Schlaf sowie höhere Komplexität der Kopplung respirokardialer Koordinationen. Ziel: Beschreibung der Propofol-Sedierungsstadien durch Analyse der linearen und Komplexitäts-Eigenschaften der thalamo-kortikalen, kortikothalamischen und retikulo-thalamischen Koordinationen. Methode: Am Tiermodell von 13 juvenilen Schweinen wurden das Veränderngen von Elektrocortico- und Elektrothalamogramm (Nucl. reticularis thalami) unter tiefer und moderater Propofolsedierung untersucht. Ergebnisse: Komplexe Kopplungseigenschaften der retikulothalamo-kortikalen Koordination zeigten höhere Kopplungsstärke unter tiefer Sedierung; lineare Kopplungseigenschaften der intrakortikalen Koordination zeigen höhere Kopplungsstärke unter moderater Sedierung. Diskussion: Komplexität respirokardialer Koordinationen erlaubt eine Unterscheidung normaler und Low risk Neugeborener. Perinataler Streß und Hypotrophie steigern Komplexität respirokardialer Koordinationen. Lineare und Komplexitätsparameter der Koordination beschreiben unterschiedliche Eigenschaften der untersuchten Systeme.
We review several aspects of the analysis of time sequences, and concentrate on recent methods using concepts from the theory of nonlinear dynamical systems. In particular, we discuss problems in estimating attractor dimensions, entropies, and Lyapunov exponents, in reducing noise and in forecasting. For completeness and since we want to stress connections to more traditional (mostly spectrum-based) methods, we also give a short review of spectral methods.
We evaluate the capability of nonlinear time series analysis to extract features from brain electrical activity (EEG) predictive of epileptic seizures. Time-resolved analysis of the EEG recorded in 16 patients from within the seizure-generating area of the brain indicate marked changes in nonlinear characteristics for up to several minutes prior to seizures as compared to other states or recording sites. If interpreted as a loss of complexity in brain electrical activity these changes could reflect the hypothesized continuous increase of synchronization between pathologically discharging neurons and allow one to study seizure-generating mechanisms in humans.
Electrocorticograms (ECoG's) from 16 of 68 chronically implanted subdural electrodes, placed over the right temporal cortex in a patient with a right medial temporal focus, were analyzed using methods from nonlinear dynamics. A time series provides information about a large number of pertinent variables, which may be used to explore and characterize the system's dynamics. These variables and their evolution in time produce the phase portrait of the system. The phase spaces for each of 16 electrodes were constructed and from these the largest average Lyapunov exponents (L's), measures of chaoticity of the system (the larger the L, the more chaotic the system is), were estimated over time for every electrode before, in and after the epileptic seizure for three seizures of the same patient. The start of the seizure corresponds to a simultaneous drop in L values obtained at the electrodes nearest the focus. L values for the rest of the electrodes follow. The mean values of L for all electrodes in the postictal state are larger than the ones in the preictal state, denoting a more chaotic state postictally. The lowest values of L occur during the seizure but they are still positive denoting the presence of a chaotic attractor. Based on the procedure for the estimation of L we were able to develop a methodology for detecting prominent spikes in the ECoG. These measures (L*) calculated over a period of time (10 minutes before to 10 minutes after the seizure outburst) revealed a remarkable coherence of the abrupt transient drops of L* for the electrodes that showed the initial ictal onset. The L* values for the electrodes away from the focus exhibited less abrupt transient drops. These results indicate that the largest average Lyapunov exponent L can be useful in seizure detection as well as a discriminatory factor for focus localization in multielectrode analysis.
In 12 epileptic patients suffering from “absences” 8-channel EEG was recorded by telemetry. The autoregressive model was applied to the signal and the prediction coefficients being the basis for calculation of the poles of the predictor. The location of the poles in thez- ands-planes was described as a function of time for 0.1 s steps along the pre-seizure EEG. In 10 of the 12 patients, and in 25 of the 28 recorded seizures this presentation of the poles of the predictor showed specific pattern linked with the occurrence of the siezure. The trajectory of the “most mobile pole” during the pre-seizure period could aid in the prediction of the seizure by several seconds.
Analysis methods derived from the theory of non-linear dynamics have been shown to provide new information about the complex spatio-temporal behaviour of neuronal networks involved in temporal lobe epilepsy. To test whether day to day alterations in neuronal complexity are influenced by changes in serum level of carbamazepine (CBZ), a moving-window correlation dimension analysis was applied to electrocorticographic and stereoelectroencephalographic recordings of 10 patients with unilateral temporal lobe epilepsy. Data sets (n = 78) were obtained from interictal states at subsequent days during the presurgical evaluation with strongly variant CBZ serum levels. The so-called neuronal complexity loss L* was used to quantify the change of dimensionality in brain electrical activity recorded under different levels of medication. We found a significant inverse relationship between L* and CBZ serum level spatially restricted to the primary epileptogenic area. This finding can be assumed to reflect the mechanism of action of CBZ attributed to an inhibition of sustained high-frequency firing of bursting neurons.
We developed an optimized algorithm that allows computation of auto- and cross-correlation integrals from single channel time series without restricting the range of hypersphere radii and embedding dimensions. Optimization was achieved by eliminating any multiple computation of subsets entering the distance function in the time-delay reconstruction of the phase space repeatedly for increasing embedding dimensions; this is most effective when using the maximum norm. Compared to more naive implementations an improvement of 2–3 was achieved, depending on the type of workstation, the operating system, and the compiler. An additional optimization for 80×86 assembly language allows torun the algorithm on a standard personal computer as fast as on a workstation. In contrast to other implementations, the execution speed of this algorithm is nearly unaffected by the type of underlying data. Thus, it is optimal for real-time analyses.
We studied 10 patients with intractable epilepsy being evaluated for epilepsy surgery for preictal changes in spiking. All patients were implanted with intracranial electrodes and underwent continuous EEG/audiovisual monitoring. Interictal spikes were detected and recorded continuously by a dedicated computerized system. Edited spikes were counted during 0-5, 5-10, and 0-60 min epochs before each seizure, during epochs of unvarying state of arousal (awake or sleep stage II). When comparing by repeated measures, 1-way ANOVA, total spiking (in all recording channels) did not differ among the different preictal epochs (0-5, 5-10, 0-60 min) in 45 seizures (F = 0.88, P = 0.40, using the Geisser-Greenhouse adjustment--GGA). Likewise, no significant differences were obtained during those same epochs when comparing spiking originating from the channel of seizure onset in 5 patients with 28 seizures of localized onset (F = 1.19, P = 0.38 using the GGA). Our findings indicate that in patients with intractable epilepsy, no changes in spiking occur in the 5 min prior to seizures, when compared to more distant preictal epochs.
EEG signals have been considered to result either from random processes or to be generated by non-linear dynamic systems exhibiting chaotic behaviour. In the latter case, the system may behave as a deterministic chaotic attractor. The complexity of the attractor can be characterized by the correlation dimension that can be computed from one signal generated by the system. A new procedure was developed and applied in order to test whether the correlation dimension, calculated from an EEG epoch, may correspond to a chaotic attractor or to a random process. This procedure was applied to EEG signals recorded from different sites of the limbic cortex of the rat during different states: wakeful rest, locomotion and in the course of an epileptic seizure induced by kindling. The signals recorded during the first two states had high dimensions and could not be distinguished from random noise. However, during an epileptic seizure the correlation dimension became low (between 2 and 4) indicating that in this state the networks behave as chaotic systems. A low correlation dimension appeared at different times and brain sites during an epileptic seizure. These results show that the computation of the correlation dimension may be useful in order to obtain insight into the dynamics of the propagation of an epileptic seizure in the brain.
1. 1. Extracellular recordings of the spontaneous firing patterns of cortical neurons were obtained in patients undergoing craniotomy for the surgical excision of an epileptogenic focus. 2. 2. Because many kinds of experimental "epileptic" foci in animals exhibit cells with high frequency (200-500/sec) bursts of action potentials, lasting 5-50 msec and recurring many times per second, we explored the electrocorticographically defined epileptogenic focus in humans in search of bursting firing patterns. 3. 3. In addition to normal firing patterns, many cells near the focus exhibited "epileptic" bursting firing patterns. Sometimes, normal and bursting cells could be recorded simultaneously, indicating that cells in proximity to one another are not uniform in such firing properties. 4. 4. Such high frequency bursts are not typically the result of artifact such as micro-electrode pressure upon a cell, or heartbeat or respiratory movements of the cortex. Most bursting cells were obtained under operating conditions involving local anesthesia, but similar results were seen under general anesthesia. 5. 5. Bursts were not necessarily synchronized with the EEG sharp waves, nor with bursts from other simultaneously recorded neurons. 6. 6. Attempts were made to modify the burst patterns by arousing a sleeping patient and by electrical stimulation of the adjacent cortical surface. While some modifications in the rate of recurrence of the bursts could be obtained, the timing patterns of the first few spikes within a burst did not change readily. 7. 7. High frequency tonic firing was also seen. Some such activity could be observed to undergo spontaneous changes from silence to high frequency tonic firing and then to bursts. 8. 8. Within a burst, the timing of spikes may be very repeatable (stereotyped bursts) in some cases. In a few cases, the structured bursts reported in chronic monkey foci have been observed, where there seems to be a characteristic pause in the firing after the first one or two spikes and then a resumption of high frequency firing in a manner identical to the stereotyped bursts. These structured timing patterns have been considered a clue towards the identification of primarily dysfunctional epileptic neurons (in contrast to normal cells recruited into bursting firing patterns by an abnormally large synaptic input). 9. 9. We would conclude that there is a good correspondence between the chronic alumina "epileptic" foci in animals and the human disease, insofar as the inter-ictal firing patterns of neurons near the focus is concerned.
The statistical properties of pre-ictal EEG spike activity in medial temporal lobe sites were analyzed in 6 patients with medically refractory complex partial seizures. A total of 24 1 h pre-ictal periods (2-6 periods per patient) were evaluated by quantifying the rate of occurrence of individual spatial patterns of spike activity derived from a subset (n = 6) of the recording channels. The channels chosen for analysis always included those medial temporal lobe sites which were documented to be most likely to initiate seizures, as well as their respective contralateral homologues. Each 1 h pre-ictal period was divided into 360 10 sec bins which were then visually classified into 1 of 64 possible spatial patterns of spike activity. These patterns, in turn, were grouped into 1 of 5 general spatial patterns and evaluated for trends across 3 20 min pre-ictal segments. Pooling these data across patients yielded the following results: (1) Focal patterns of spike activity tended to decline significantly in rate of occurrence several minutes prior to seizures, while the rate of independent contralateral patterns did not change. (2) The rate of occurrence of patterns of bilateral loosely coupled spike activity (involving focal and contralateral sites) increased significantly across the 20 min pre-ictal segments and was clearly augmented during the 20 min prior to seizures. These findings indicate that the degree of bilateral independence in medial temporal lobe spike activity tends to decrease several minutes prior to the localized onset of temporal lobe seizures; such changes may reflect the mechanisms responsible for the inter-ictal-ictal transition.