Seizure prediction using scalp electroencephalogram Ivo Drury, a,b, * Brien Smith, b Dingzhou Li, c and Robert Savit a,c,d a Diagnostic Neurodynamics LLC, Ann Arbor, MI 48104, USA b Department of Neurology, Henry Ford Health System, Detroit, MI 48202, USA c Department of Physics, University of Michigan, Ann Arbor, MI 48109, USA d Biophysics Research Division, University of Michigan, Ann Arbor, MI 48109, USA Received 15 June 2003; accepted 15 June 2003 Abstract Using a measure of nonlinear dynamical changes we term marginal predictability, we report evidence of robust changes in this parameter on scalp EEG in a cohort of patients with medically refractory mesiobasal temporal lobe epilepsy (MBTLE). In the baseline (interictal) state there are distinct differences in this nonlinear measure between epileptic and neurologically normal subjects. At baseline, in patients with MBTLE there are differences in these measures between electrodes adjacent to the ictal onset zone and more remotely placed electrodes. The character of these differences evolves over a period of approximately 30 min before a seizure. We discuss and integrate our findings with two emerging concepts in epileptology, first, the concept of a preictal or transition phase rather than an abrupt movement from interictal to ictal activity, and second, the notion of an epileptic neural network with changes in areas of brain remote from what has traditionally been considered the ictal onset zone influencing “ictogenesis.” © 2003 Elsevier Inc. All rights reserved. Keywords: Mesial temporal lobe epilepsy; Nonlinear dynamics; Seizure prediction; Marginal predictability Introduction Despite a significant number of new antiepileptic com- pounds that have become available worldwide in the past 15 years, epilepsy refractory to best available medical therapy continues to affect nearly 1 in 4 patients with a seizure disorder. Current estimates are that some 500,000 persons have uncontrolled seizures in the United States today. Even when these patients’ seizures are relatively infrequent there is a major impact on their quality of life (Gilliam, 2002). There is also sizeable morbidity and mortality and direct and indirect economic burdens (Begley et al., 2000). The availability of reliable methods of seizure prediction could enhance the quality and safety of patients with epilepsy, facilitate implementation of short-term interventions to abort a seizure, and have the potential of reducing the economic burden of this disease. Efforts to predict when a seizure was going to occur began with Viglione and Walsh (1975), but have gathered momentum over the past decade with the ready availability of high-speed computer process- ing and the application of sophisticated mathematical tech- niques to biological processes. Part of the rationale for attempting to predict seizure onset relies on the fact that at least some patients with seizures experience a premonitory phase lasting for several minutes to hours that is distinct from an aura. There are, in fact, several strands of evidence from both human and experimental reports that support the notion of a transition phase of at least several minutes and potentially up to 1 h before the ictal event proper begins and against the concept of ictal activity happening as if a light switch had just been turned on. Initial attempts at anticipating seizures relied on standard linear statistical methods (Gotman et al., 1985; Osorio, 1998). Following the renaissance of nonlinear dynamics and the realization that many natural processes embodied non- linearities in their dynamics, researchers in a variety of fields began looking for evidence of nonlinearities and spe- cifically chaos in a wide range of data sets. The hope was * Corresponding author. Diagnostic Neurodynamics LLC, 728 Onon- daga Street, Ann Arbor, MI 48104, USA. Fax: +1-734-747-9239 E-mail address: idrury@comcast.net (I. Drury). R Available online at www.sciencedirect.com Experimental Neurology 184 (2003) S9 –S18 www.elsevier.com/locate/yexnr 0014-4886/$ – see front matter © 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0014-4886(03)00354-6