EEG analysis with simulated neuronal cell models helps to detect pre-seizure changes K. Schindler * , R. Wiest, M. Kollar, F. Donati Department of Neurology, University Hospital of Bern, Inselspital, 3010 Bern, Switzerland Accepted 23 January 2002 Abstract Objectives: To test if a method for real-time detection of epileptic seizures based on electroencephalographic (EEG) analysis with simulated neuronal cell models can be modified to identify pre-seizure changes. Methods: Our EEG analysis method consists of two simulated leaky integrate and fire units (LIFU) connected to a signal preprocessing stage that marks parts of the EEG signals with slopes larger than a preset threshold Hth with unit pulses. The LIFUs change their spiking frequency depending on the rate and the synchrony of the impinging pulse trains. Here, we use our method in a high-sensitivity mode by setting Hth to low values, which causes the LIFUs to continuously spike during the interictal state. We test if the LIFUs spiking rates change before seizure onset. Results: We used 9 long-term EEGs (16 ^ 7 h) of 7 patients with drug resistant epilepsy. Fifteen seizures were analyzed and all were preceded by an increase of the time-averaged spiking rates SR av of the LIFUs. We defined a function F Sz , which quantifies the changes of SR av . F Sz increased and stayed above an individually set and fixed threshold 83 ^ 91 min (range: 4–330 min) before EEG seizure onset. Only two false alarms occurred. Conclusions: We conclude that EEG analysis with simulated neuronal cell models may be used to detect pre-seizure changes with high sensitivity and specificity. q 2002 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Pre-seizure changes; Electroencephalography; Neuronal cell models; Epilepsy; Foramen ovale 1. Introduction Developing reliable methods for detecting and predicting the onset of epileptic seizures is of major clinical and theo- retical importance. Clinically, the detection of the onset of epileptic seizures is, for example, necessary to trigger the injection of a radiotracer to localize precisely an epilepto- genic focus with ictal single-photon-emission-computed- tomography (SPECT) (So et al., 2000). Following seizure onset detection, therapeutic interventions that have rapid effects could be initiated like intravenous or local delivery of antiepileptic drugs (AED) (Eder et al., 1997; Stein et al., 2000) or electrical stimulation of peripheral and central parts of the nervous system to interfere with and possibly stop seizure activity (Velasco et al., 1987; Penry and Dean, 1990; Nicolelis, 2001). The anticipation of epileptic seizures minutes to hours before clinical onset would be important for treatments with slower effects, for example the oral intake of AED. Even in cases of epilepsy resistant to all therapies, the knowledge of when a seizure will occur would reduce the psychological stress of the patient and his social environment (Murray, 1993) and be useful to prevent accidents (Cockerell et al., 1994). From the point of view of basic neuroscience the development of robust seizure prediction methods that sense subtle pre-seizure changes of electrical activity could shed new light on the physiolo- gical mechanisms of the interictal to ictal transition by revealing the time courses of the underlying processes (Calvin, 1972; Sherwin, 1978; Depaulis et al., 1994; Le Van Quyen et al., 1999). Diverse methods have been applied to detect and predict epileptic seizures and promising results have been reported (Lehnertz and Elger, 1998; Martinerie et al., 1998; Osorio et al., 1998; Le Van Quyen et al., 2001; Litt et al., 2001). However, the different approaches also provoked a lively debate about which of the mathematical concepts and assumptions used are actually fulfilled by biological data sets (Webber and Zbilut, 1994; Theiler and Rapp, 1996; Thomasson et al., 2001). We have recently developed a novel computer-based method to detect the onset of epilep- Clinical Neurophysiology 113 (2002) 604–614 1388-2457/02/$ - see front matter q 2002 Elsevier Science Ireland Ltd. All rights reserved. PII: S1388-2457(02)00032-9 www.elsevier.com/locate/clinph CLINPH 2001644 * Corresponding author. Tel.: 141-31-711-22-58; fax: 141-31-632-96- 79. E-mail address: kschindler@access.ch (K. Schindler).