Seizure prediction

Post-publication activity

Florian Mormann, Department of Epileptology, University of Bonn, Germany

Figure 1: Seizure Prediction from EEG time series. Continuous multi-channel EEG recordings are analyzed by means of a moving-window analysis. The data covered by the orange window is transformed into a single value in the time profile of a multivariate characterizing measure. When this time profile crosses a certain pre-defined threshold, an alarm is issued.

Seizure prediction refers to predicting/forecasting/anticipating the occurrence of epileptic seizures, typically by monitoring the brain’s electrical activity by means of electroencephalographic (EEG) recordings. To date no study has provided sound evidence that prospective seizure prediction is possible above chance level. The field of seizure prediction is concerned with anticipating the occurrence of a forthcoming seizure in a patient who has epilepsy, not with assessing the prognostic or diagnostic value of pathological EEG findings.

Motivation

The sudden and seemingly unpredictable nature of seizures is one of the most compromising aspects of the disease epilepsy. Most epilepsy patients only spend a marginal part of their time actually having a seizure and show no clinical signs of their disease during the time between seizures, the so-called inter-ictal interval. But the constant fear of the next seizure and the feeling of helplessness associated with it often have a strong impact on the everyday life of a patient (Fisher et al. 2000). A method capable of reliably predicting the occurrence of seizures, e.g. from continuous EEG recordings, could significantly improve the quality of life for these patients and open new therapeutic possibilities. Apart from simple warning devices, fully automated closed-loop seizure prevention systems are conceivable. Treatment concepts could move from preventive strategies (e.g. long-term medication with anti-epileptic drugs) towards an EEG-triggered on-demand therapy (e.g. by excretion of fast-acting anticonvulsant substances or by electrical or other stimulation in an attempt to reset brain dynamics to a state that will no longer develop into a seizure (Theodore and Fisher 2004, Osorio et al. 2005, Morrell 2006, Stacey and Litt 2008)).

In principle there are two types of scenarios of how a seizure could occur (Lopes da Silva et al. 2003). It could either be caused by a sudden and abrupt transition in which case it would not be preceded by detectable dynamical changes in the EEG. Such a scenario would be conceivable for the initiation of seizures in primary generalized epilepsy. Alternatively, this transition could be a gradual change (or a cascade of changes) in dynamics which could in theory be detected. This type of transition could be more common in focal epilepsies.

The crucial question thus is whether a pre-seizure state exists that can be distinguished from the inter-ictal state. Such a state would reflect an increased probability of seizure occurrence and is sometimes also referred to as a pro-ictal state. Clinical findings in support of the existence of a pre-seizure state include an increase in cerebral blood flow, oxygen availability, and blood oxygen level dependent (BOLD) signal as well as changes in heart rate prior to seizure occurrence (see Mormann et al. 2007 and references therein); however, it remains unclear if this represents changes related to ongoing ictal activity (i.e. during a seizure) or pre-ictal (i.e. pre-seizure) events in a strict sense. A minority of patients have reported to experience premonitory symptoms (Rajna et al. 1997, Schulze-Bonhage et al. 2006, Haut et al. 2007), and reports have been published of seizure alert dogs that are presumed to be capable of anticipating the occurrence of seizures of their owners (Kirton et al. 2004). However, most of these findings were based on self-reported data (Hoppe et al. 2007) and until now have not been confirmed by a prospective and objective evaluation (Ortiz and Liporace 2005, Doherty and Haltiner 2007).

Algorithms for Seizure Prediction

Most EEG-based prediction techniques use a moving-window analysis in which some (linear or non-linear) characterizing measure (see e.g. Kantz and Schreiber 2004) is calculated from a window of EEG data with a pre-defined length, then the subsequent window of EEG is analyzed, and so forth. The duration of these analysis windows typically ranges between 10 and 40 s. Depending on whether the measure is used to characterize a single EEG channel or relations between two or more channels, it is referred to as a univariate, bivariate or multivariate measure, respectively. The moving-window analysis thus renders time profiles of a characterizing measure for different channels or channel combinations (Figure 1).

With respect to practical application, a prediction algorithm should ideally be prospective, i.e., its output at a given time should be a function of the information available at this time. Prediction algorithms usually employ certain thresholds. If the time profile of a characterizing measure crosses the threshold, the algorithm issues a warning in the form of an alarm (Figure 1). This alarm can be either true or false, depending on whether it is actually followed by a seizure or not. For this distinction, it is necessary to define a prediction horizon or warning time, i.e., the period after an alarm within which a seizure is expected to occur. If an alarm is followed by a seizure within the prediction horizon, it is classified as a true alarm (true positive), otherwise it is regarded as a false alarm (false positive) (Figure 2). Prediction horizons reported in the literature range from several minutes to a few hours.

Figure 2: Whenever the time profile of some characterizing measure rises above a pre-defined threshold, the prediction algorithm issues an alarm which is classified as true or false, depending on whether or not it is followed by a seizure within a specified warning time, termed prediction horizon. The area shaded in gray represents the time during which the patient is under false warning, i.e. awaiting a non-occurring seizure. The area shaded in green represents the time under correct warning. In principle, alarms are issued continuously for as long as the measure profile remains above the threshold. For better legibility not all of these continuous alarms are shown. For the last seizure, alarms begin too early with respect to the prediction horizon so they are classified as false warnings until the seizure falls within the prediction horizon. The time period shown on the x-axis could typically represent 1 day.

If a seizure is not preceded by an alarm within the prediction horizon, this will be counted as a missed seizure (false negative). The sensitivity of a prediction algorithm is usually quantified as the number of seizures with at least one alarm within the preceding prediction horizon divided by the total number of seizures. To quantify the specificity of a prediction algorithm, most studies have reported specificity rates measured in false predictions per hour. A reported false prediction rate cannot be judged independently of the prediction horizon since in a prospective prediction algorithm a false alarm will leave the patient erroneously awaiting a seizure for the entire duration of the warning time. It is only after this time that the patient can know that an alarm was a false warning. If another alarm occurs during the prediction horizon, the warning time should be restarted at this point. A better way to assess the specificity of a prediction algorithm is to report the portion of time during which a patient is not in the state of falsely awaiting a seizure. In general, any algorithm can be tuned (e.g. by varying the alarm threshold) to yield a higher sensitivity at the cost of a lower specificity and vice versa.

Another important issue in the evaluation of a prediction algorithm is the use of a posteriori information. For a prospective prediction algorithm, this type of information is not available. Two typical cases of using a posteriori information are found in the literature: (i) in-sample optimization of parameters of the algorithm and (ii) retrospective selection of one or more channels with particularly good performance. Such an optimization is likely to result in an over-estimated performance that will not be reproducible when applying the algorithm to other, out-of-sample testing data that was not used in the optimization process. While it may be useful and necessary to adapt an algorithm for an individual patient by adjusting parameters and selecting optimal channels, this optimization must be performed on a part of the data (training set) that is subsequently excluded from the data set used to assess the algorithm’s performance (test set) .

Once the out-of-sample performance in terms of sensitivity and specificity (e.g. as one minus fraction of time under false warning) of a prospective prediction algorithm has been assessed, it remains to be tested whether it is indeed superior to naïve prediction schemes such as random or periodic predictors. Such a statistical validation can be based on bootstrapping methods or Monte Carlo simulations (Andrzejak et al. 2003, Kreuz et al. 2004, Mormann et al. 2005, Andrzejak et al. 2009) or on comparison with performance estimates derived analytically for naïve prediction schemes (Winterhalder et al. 2003, Schelter et al. 2006).

A simple way to assess the performance of a prospective prediction algorithm is to compare the percentage of seizures that occurred while an alarm was active with the total percentage of time during which an alarm was active. For instance, if a ‘warning light’ is on 60% of the time, we would expect 60% of the seizures to be predicted just by chance. Only if the percentage of predicted seizures (i.e. the sensitivity) significantly exceeds the percentage of time under warning can a given algorithm be assumed to bear predictive power above chance level (cf. Snyder et al. 2008). The statistical significance of this performance, i.e. the probability of predicting S out of N seizures by chance for a given portion of time under warning of q is simply given by the binomial probability \[ P(N,S,q) = \sum_{K\ge S} \left( {\begin{array}{*{20}c} N \\ K \\ \end{array}} \right) \cdot q^K \cdot (1-q)^{N-K} \]

A Brief History of Seizure Prediction

This section briefly describes some of the major studies in the field. For a more comprehensive review see Mormann et al. 2007.

After some early work on the predictability of seizures dating back to the 1970s (Viglione and Walsh 1975), attempts to extract seizure precursors from surface EEG recordings of absence seizures were carried out by different groups using linear approaches such as pattern detection and spectral analysis (see Mormann et al. 2007).

Following the advent of the theory of nonlinear dynamics in the 1980s, time series analysts became aware of seizure prediction as a potential field of application. These and later studies predominantly analyzed EEG signals from patients undergoing video-EEG monitoring, with chronic electrodes implanted directly inside or on the surface of the brain to localize the seizure focus for possible surgical resection. During the 1990s several quantitative EEG studies reported pre-ictal phenomena using characterizing measures such as the largest Lyapunov exponent (Iasemidis et al. 1990), the correlation density (Martinerie et al. 1998) or a dynamical similarity index (Le Van Quyen et al. 1999, 2001). The common feature of these studies was that their focus of interest was entirely limited to the pre-ictal period and that they did not include an evaluation of control recordings from the seizure-free interval, so the specificity of the applied techniques was not assessed. Another group of proof-of-principle studies addressed the issue of specificity by comparing pre-ictal changes in dynamics to inter-ictal control recordings, although the reported findings remained on an anecdotal level (Mormann et al. 2000, Navarro et al. 2002, Chávez et al. 2003).

In the first controlled studies comprising defined groups of patients with pre-ictal and inter-ictal control recordings, measures like the correlation dimension (Lehnertz and Elger 1998), dynamical entrainment (Iasemidis et al. 2001), accumulated signal energy (Litt et al. 2001), simulated neuronal cell models (Schindler et al. 2002), or phase synchronization (Mormann et al. 2003) were shown to be capable to distinguish short segments of inter-ictal data from pre-ictal data.

These studies were followed by a number of studies (mostly carried out on more extensive data bases) that found a substantially poorer predictive performance than expected from earlier reports for measures like the correlation dimension (Aschenbrenner-Scheibe et al. 2003), the similarity index (Winterhalder et al. 2003), and accumulated energy (Maiwald et al. 2004). Furthermore a controversy evolved regarding both the reproducibility of earlier studies (De Clercq et al. 2003) and the general suitability of nonlinear measures used to characterize EEG time series (McSharry et al. 2003, Lai et al. 2003).

Around the turn of the millennium, when mass storage capacity became more widely available, epilepsy centers were able to store the complete data acquired during pre-surgical monitoring without the necessity of selecting sample recordings. In 2005, several groups published a series of studies that were carried out on a set of five continuous multi-day recordings provided by different epilepsy centers for the First International Collaborative Workshop on Seizure Prediction (Lehnertz and Litt 2005) held in 2002. The aim of this workshop was to have different groups test and compare their methods on a joint data set. Results from the different groups for the most part showed a poor performance of univariate measures (D'Alessandro et al. 2005, Esteller et al. 2005, Harrison et al. 2005, Jouny et al. 2005, Mormann et al. 2005). A better performance was reported for bi- and multi-variate measures (Mormann et al. 2005, Le Van Quyen et al. 2005, Iasemidis et al. 2005). The observed pre-ictal changes were found to be locally restricted to specific channels rather than occurring as a global phenomenon.

The first attempts to test seizure prediction algorithms in a prospective study design (D'Alessandro et al. 2005, Iasemidis et al. 2005, 2003) yielded sensitivities and specificity rates that most epileptologists would consider unacceptable for clinical implementation. Whether the performance of the algorithms was at all better than random was not investigated. Although two recent studies (Chaovalitwongse et al. 2005, Sackellares et al. 2006) attempted such a validation against a random predictor, they failed to carry out a proper statistical comparison.

Prior to the Third International Workshop on Seizure Prediction in 2007, the workshop organizers initiated a public competition, in which participants could download the first parts of three continuous long-term recordings from three different patients. After training their algorithms on this data and optimizing them individually for each patient, participants could then submit their algorithms to have them tested on the remaining parts of the data. The benchmark for winning the competition was merely to outperform chance level (i.e. to predict a percentage of seizures that was higher than the percentage of time under false warning). None of the submitted algorithms passed this test, so the competition continues to be open to the public.

Understanding the Mechanisms of Seizure Generation

The exact mechanisms by which seizures occur remain largely unknown except for certain distinct types of epilepsies such as reflex epilepsies (Parra et al. 2005). It is conceivable that there may be different mechanisms underlying the initiation of seizures in different brain structures, and thus different seizure-predicting algorithms may be necessary. This may also be true of the various pathologies underlying focal epilepsies. The predictive changes in the EEG before a seizure and the best methods for detecting them could thus vary from patient to patient.

A number of recent studies have attempted to increase our understanding of the dynamics of seizure generation in humans. In both temporal lobe and neocortical epilepsies, high-frequency oscillations were found to play a role in the initiation of epileptiform activity and seizures (Bragin et al. 1999, Worrell et al. 2004, Schiff et al. 2000, Bragin et al. 2002, Jirsch et al. 2006, Worrell et al. 2008). In another study on patients with temporal lobe epilepsy, the intracranial EEG signal was found to respond more strongly to electrical stimulation as the brain approached a seizure (Kalitzin et al. 2005). Remarkably, this study used active external perturbation of ongoing EEG activity rather than passive extraction of features from spontaneous EEG signals to detect impending seizures. In addition to empirical studies on seizure dynamics in humans (Franaszczuk et al. 1994, Jouny et al. 2007, Bartolomei et al. 2004, Schiff et al. 2005, Schindler et al. 2007) studies on animal models of epilepsy (McCormick and Contreras 2001, Avoli et al. 2002; Jefferys 2003; Beck and Yaari 2008) as well as computational models (Wendling et al. 2003, 2005, Suffczynski et al. 2005, Feldt et al 2007, Lytton 2008) have helped to gain insight into the dynamical processes potentially involved in seizure generation.

A better understanding of the mechanisms of seizure generation that takes into consideration the complex spatiotemporal interactions between different brain regions for different types of epilepsy may eventually stimulate the design of improved prediction methods and algorithms.

Summary

The more rigorous methodological design in recent seizure prediction studies has shown that many of the techniques previously considered suitable for seizure prediction turned out to perform no better than a random predictor. The next milestone in this field is to prove that seizure prediction algorithms can indeed be designed to run prospectively on unselected, out-of-sample data with a performance that is significantly higher than chance level.

References

Andrzejak RG, Mormann F, Kreuz T, Rieke C, Kraskov A, Elger CE, Lehnertz K. Testing the null hypothesis of the nonexistence of a preseizure state. Phys Rev E Stat Nonlin Soft Matter Phys 2003; 67: 010901.

Andrzejak RG, Chicharro D, Elger CE, Mormann F. Seizure Prediction: Any better than chance?. Clin Neurophysiol 2009; 120: 1465-78.

Aschenbrenner-Scheibe R, Maiwald T, Winterhalder M, Voss HU, Timmer J, Schulze-Bonhage A. How well can epileptic seizures be predicted? An evaluation of a nonlinear method. Brain 2003; 126: 2616-26.

Avoli M, D'Antuono M, Louvel J, Köhling R, Biagini G, Pumain R, D'Arcangelo G, Tancredi V. Network and pharmacological mechanisms leading to epileptiform synchronization in the limbic system in vitro. Prog Neurobiol 2002; 68: 167-207.

Bartolomei F, Wendling F, Régis J, Gavaret M, Guye M, Chauvel P. Pre-ictal synchronicity in limbic networks of mesial temporal lobe epilepsy. Epilepsy Res 2004; 61: 89-104.

Beck H, Yaari Y. Plasticity of intrinsic neuronal properties in CNS disorders. Nat Rev Neurosci 2008; 9: 357-69.

Bragin A, Engel J, Wilson CL, Fried I, Buzsáki G. High-frequency oscillations in human brain. Hippocampus 1999; 9: 137-42.

Bragin A, Mody I, Wilson CL, Engel J. Local generation of fast ripples in epileptic brain. J Neurosci 2002; 22: 2012-21.

Chaovalitwongse W, Iasemidis LD, Pardalos PM, Carney PR, Shiau D, Sackellares JC. Performance of a seizure warning algorithm based on the dynamics of intracranial EEG. Epilepsy Res 2005; 64: 93-113.

Chávez M, Le Van Quyen M, Navarro V, Baulac M, Martinerie J. Spatio-temporal dynamics prior to neocortical seizures: amplitude versus phase couplings. IEEE Trans Biomed Eng 2003; 50: 571-83.

D'Alessandro M, Vachtsevanos G, Esteller R, Echauz J, Cranstoun S, Worrell G, et al. A multi-feature and multi-channel univariate selection process for seizure prediction. Clin Neurophysiol 2005; 116: 506-16.

De Clercq W, Lemmerling P, Van Huffel S, Van Paesschen W. Anticipation of epileptic seizures from standard EEG recordings. Lancet 2003; 361: 970; author reply 970-1.

Doherty MJ, Haltiner AM. Wag the dog: skepticism on seizure alert canines. Neurology 2007; 68: 309.

Esteller R, Echauz J, D'Alessandro M, Worrell G, Cranstoun S, Vachtsevanos G, et al. Continuous energy variation during the seizure cycle: towards an on-line accumulated energy. Clin Neurophysiol 2005; 116: 517-26.

Feldt S, Osterhage H, Mormann F, Lehnertz K, Zochowski M. Internetwork and intranetwork communications during bursting dynamics: applications to seizure prediction. Phys Rev E Stat Nonlin Soft Matter Phys 2007; 76: 021920.

Fisher RS, Vickrey BG, Gibson P, Hermann B, Penovich P, Scherer A. The impact of epilepsy from the patient's perspective I. Descriptions and subjective perceptions. Epilepsy Res 2000; 41: 39-51.

Franaszczuk PJ, Bergey GK, Kamiński MJ. Analysis of mesial temporal seizure onset and propagation using the directed transfer function method. Electroencephalogr Clin Neurophysiol 1994; 91: 413-27.

Harrison MAF, Frei MG, Osorio I. Accumulated energy revisited. Clin Neurophysiol 2005; 116: 527-31.

Haut SR, Hall CB, LeValley AJ, Lipton RB. Can patients with epilepsy predict their seizures? Neurology 2007; 68: 262-6.

Hoppe C, Poepel A, Elger CE. Epilepsy: accuracy of patients seizure counts. Arch Neurol 2007; 64: 1595-9.

Iasemidis L, Pardalos P, Sackellares J, Shiau D. Quadratic Binary Programming and Dynamical System Approach to Determine the Predictability of Epileptic Seizures. Journal of Combinatorial Optimization 2001; 5: 9-26.

Iasemidis LD, Shiau D, Chaovalitwongse W, Sackellares JC, Pardalos PM, Principe JC, et al. Adaptive epileptic seizure prediction system. IEEE Trans Biomed Eng 2003; 50: 616-27.

Iasemidis LD, Shiau D, Pardalos PM, Chaovalitwongse W, Narayanan K, Prasad A, et al. Long-term prospective on-line real-time seizure prediction. Clin Neurophysiol 2005; 116: 532-44.

Iasemidis LD, Sackellares JC, Zaveri HP, Williams WJ. Phase space topography and the Lyapunov exponent of electrocorticograms in partial seizures. Brain Topogr 1990; 2: 187-201.

Jefferys JG. Models and mechanisms of experimental epilepsies. Epilepsia 2003; 44: 44-50.

Jirsch JD, Urrestarazu E, LeVan P, Olivier A, Dubeau F, Gotman J. High-frequency oscillations during human focal seizures. Brain 2006; 129: 1593-608.

Jouny CC, Adamolekun B, Franaszczuk PJ, Bergey GK. Intrinsic ictal dynamics at the seizure focus: effects of secondary generalization revealed by complexity measures. Epilepsia 2007; 48: 297-304.

Jouny CC, Franaszczuk PJ, Bergey GK. Signal complexity and synchrony of epileptic seizures: is there an identifiable preictal period? Clin Neurophysiol 2005; 116: 552-8.

Kalitzin S, Velis D, Suffczynski P, Parra J, da Silva FL. Electrical brain-stimulation paradigm for estimating the seizure onset site and the time to ictal transition in temporal lobe epilepsy. Clin Neurophysiol 2005; 116: 718-28.

Kantz H, Schreiber T. Nonlinear Time Series Analysis. Second Edition. Cambridge, UK: Cambridge University Press; 2004.

Kirton A, Wirrell E, Zhang J, Hamiwka L. Seizure-alerting and -response behaviors in dogs living with epileptic children. Neurology 2004; 62: 2303-5.

Kreuz T, Andrzejak RG, Mormann F, Kraskov A, Stögbauer H, Elger CE, et al. Measure profile surrogates: a method to validate the performance of epileptic seizure prediction algorithms. Phys Rev E Stat Nonlin Soft Matter Phys 2004; 69: 061915.

Lai Y, Harrison MAF, Frei MG, Osorio I. Inability of Lyapunov exponents to predict epileptic seizures. Phys Rev Lett 2003; 91: 068102.

Lehnertz K, Elger CE. Can Epileptic Seizures be Predicted? Evidence from Nonlinear Time Series Analysis of Brain Electrical Activity. Phys. Rev. Lett. 1998; 80: 5019.

Lehnertz K, Litt B. The First International Collaborative Workshop on Seizure Prediction: summary and data description. Clin Neurophysiol 2005; 116: 493-505.

Le Van Quyen M, Martinerie J, Baulac M, Varela F. Anticipating epileptic seizures in real time by a non-linear analysis of similarity between EEG recordings. Neuroreport 1999; 10: 2149-55.

Le Van Quyen M, Martinerie J, Navarro V, Boon P, D'Havé M, Adam C, et al. Anticipation of epileptic seizures from standard EEG recordings. Lancet 2001; 357: 183-8.

Le Van Quyen M, Soss J, Navarro V, Robertson R, Chavez M, Baulac M, et al. Preictal state identification by synchronization changes in long-term intracranial EEG recordings. Clin Neurophysiol 2005; 116: 559-68.

Litt B, Esteller R, Echauz J, D'Alessandro M, Shor R, Henry T, et al. Epileptic seizures may begin hours in advance of clinical onset: a report of five patients. Neuron 2001; 30: 51-64.

Lopes da Silva F, Blanes W, Kalitzin SN, Parra J, Suffczynski P, Velis DN. Epilepsies as dynamical diseases of brain systems: basic models of the transition between normal and epileptic activity. Epilepsia 2003; 44: 72-83.

Lytton WW. Computer modelling of epilepsy. Nat Rev Neurosci 2008; 9: 626-37.

Maiwald T, Winterhalder M, Aschenbrenner-Scheibe R, Voss HU, Schulze-Bonhage A, Timmer J. Nonlinear Phenomena : Comparison of three nonlinear seizure prediction methods by means of the seizure prediction characteristic. Physica D 2004; 194: 357-368.

Martinerie J, Adam C, Le Van Quyen M, Baulac M, Clemenceau S, Renault B, et al. Epileptic seizures can be anticipated by non-linear analysis. Nat Med 1998; 4: 1173-6.

McCormick DA, Contreras D. On the cellular and network bases of epileptic seizures. Annu Rev Physiol 2001; 63: 815-46.

McSharry PE, Smith LA, Tarassenko L. Prediction of epileptic seizures: are nonlinear methods relevant? Nat Med 2003; 9: 241-2; author reply 242.

Mormann F, Andrzejak RG, Elger CE, Lehnertz K. Seizure prediction: the long and winding road. Brain 2007; 130: 314-33.

Mormann F, Kreuz T, Andrzejak RG, David P, Lehnertz K, Elger CE. Epileptic seizures are preceded by a decrease in synchronization. Epilepsy Res 2003; 53: 173-85.

Mormann F, Kreuz T, Rieke C, Andrzejak RG, Kraskov A, David P, et al. On the predictability of epileptic seizures. Clin Neurophysiol 2005; 116: 569-87.

Mormann F, Lehnertz K, David P, E. Elger C. Mean phase coherence as a measure for phase synchronization and its application to the EEG of epilepsy patients. Physica D 2000; 144: 358-369.

Morrell M. Brain stimulation for epilepsy: can scheduled or responsive neurostimulation stop seizures? Curr. Opin. Neurol 2006; 19: 164-8.

Navarro V, Martinerie J, Le Van Quyen M, Clemenceau S, Adam C, Baulac M, et al. Seizure anticipation in human neocortical partial epilepsy. Brain 2002; 125: 640-55.

Ortiz R, Liporace J. "Seizure-alert dogs": observations from an inpatient video/EEG unit. Epilepsy Behav 2005; 6: 620-2.

Osorio I, Frei MG, Sunderam S, Giftakis J, Bhavaraju NC, Schaffner SF, et al. Automated seizure abatement in humans using electrical stimulation. Ann Neurol 2005; 57: 258–68.

Parra J, Kalitzin SN, Lopes da Silva FH. Photosensitivity and visually induced seizures. Curr Opin Neurol 2005; 18: 155-9.

Rajna P, Clemens B, Csibri E, Dobos E, Geregely A, Gottschal M, et al. Hungarian multicentre epidemiologic study of the warning and initial symptoms (prodrome, aura) of epileptic seizures. Seizure 1997; 6: 361-8.

Sackellares JC, Shiau D, Principe JC, Yang MCK, Dance LK, Suharitdamrong W, et al. Predictability analysis for an automated seizure prediction algorithm. J Clin Neurophysiol 2006; 23: 509-20.

Schelter B, Winterhalder M, Maiwald T, Brandt A, Schad A, Schulze-Bonhage A, et al. Testing statistical significance of multivariate time series analysis techniques for epileptic seizure prediction. Chaos 2006; 16: 013108.

Schiff SJ, Colella D, Jacyna GM, Hughes E, Creekmore JW, Marshall A, et al. Brain chirps: spectrographic signatures of epileptic seizures. Clin Neurophysiol 2000; 111: 953-8.

Schiff SJ, Sauer T, Kumar R, Weinstein SL. Neuronal spatiotemporal pattern discrimination: the dynamical evolution of seizures. Neuroimage 2005; 28: 1043-55.

Schindler K, Wiest R, Kollar M, Donati F. EEG analysis with simulated neuronal cell models helps to detect pre-seizure changes. Clin Neurophysiol 2002; 113: 604-14.

Schulze-Bonhage A, Kurth C, Carius A, Steinhoff BJ, Mayer T. Seizure anticipation by patients with focal and generalized epilepsy: A multicentre assessment of premonitory symptoms. Epilepsy Res, 2006; 70: 83-88

Snyder DE, Echauz J, Grimes DC, Litt B. The statistics of a practical seizure warning system. J Neural Eng 2008; 5: 392-401.

Stacey WC, Litt B. Technology insight: neuroengineering and epilepsy-designing devices for seizure control. Nat Clin Pract Neurol 2008; 4: 190-201.

Suffczynski P, Lopes da Silva F, Parra J, Velis D, Kalitzin S. Epileptic transitions: model predictions and experimental validation. J Clin Neurophysiol 2005; 22: 288-99.

Theodore WH, Fisher RS. Brain stimulation for epilepsy. Lancet Neurology 2004; 3: 111-8.

Viglione SS, Walsh GO. Proceedings: Epileptic seizure prediction. Electroencephalogr Clin Neurophysiol 1975; 39: 435-6.

Wendling F, Bartolomei F, Bellanger JJ, Bourien J, Chauvel P. Epileptic fast intracerebral EEG activity: evidence for spatial decorrelation at seizure onset. Brain 2003; 126: 1449-59.

Wendling F, Hernandez A, Bellanger J, Chauvel P, Bartolomei F. Interictal to ictal transition in human temporal lobe epilepsy: insights from a computational model of intracerebral EEG. J Clin Neurophysiol 2005; 22: 343-56.

Winterhalder M, Maiwald T, Voss HU, Aschenbrenner-Scheibe R, Timmer J, Schulze-Bonhage A. The seizure prediction characteristic: a general framework to assess and compare seizure prediction methods. Epilepsy Behav 2003; 4: 318-25.

Worrell GA, Gardner AB, Stead SM, Hu S, Goerss S, Cascino GJ, et al. High-frequency oscillations in human temporal lobe: simultaneous microwire and clinical macroelectrode recordings. Brain 2008; 131: 928-37.

Worrell GA, Parish L, Cranstoun SD, Jonas R, Baltuch G, Litt B. High-frequency oscillations and seizure generation in neocortical epilepsy. Brain 2004; 127: 1496-506.

Internal references