Determination of Epileptic Seizure Onset From EEG Data Using Spectral Analysis and Discrete Finite Automata Rory A. Lewis 1,2 , Brian Parks 2 , Andrew M. White 1 1. Departments of Pediatrics & Neurology, University of Colorado Denver, Anschutz Medical Campus 2. Department of Computer Science, University of Colorado at Colorado Springs rory.lewis@ucdenver.edu, bparks@uccs.edu, andrew.white@ucdenver.edu Abstract—This paper is a continuation of the goal to connect power spectra and Deterministic Finite Automata (DFA) in a manner to enhance the detection of spikes and seizures in epileptiform activity from Electroencephalograms (EEG). The goal is to develop robust classification rules for identifying epileptiform activity in the human brain. This paper presents advancement using the author’s proprietary developed spectral analysis to link power spectra of rat EEGs experiencing epilepsy seizures with the authors DFA algorithm and their MATLAB spectral analysis. We present a system that links 1) power spectra of seizures, in sleep, spike and seizure states with 2) Deterministic Finite Automata (DFA). Combining power spectra with DFA to correctly predict and identify epileptiform activity (spikes) and epileptic seizures opens the door to creating classifiers for seizures. It is a common goal for those skilled in the art of epilepsy prediction to create classifiers to make rules and discretize events leading to an epileptic seizure. Herein we present a means to link time and frequency domain components from MATLAB and proprietary software to clinical epileptiform activity. Index Terms—epilepsy, seizure prediction, deterministic finite automata, spectral analysis I. I NTRODUCTION Epilepsy is a neurological disorder that makes people sus- ceptible to recurrent unprovoked seizures due to electrical disturbances in the brain. Unfortunately, 30% of patients that suffer from epilepsy are not well controlled on medication. Only a small fraction of these can be helped by seizure surgery [1]. Therefore, it would be life changing to a large number of individuals if a system could be developed that would predict a seizure hours, minutes, or even seconds before its clinical onset. The challenge in this problem is that the dimensionality is huge; in the human brain there are approximately 100 billion neurons, each with about 1000 connections (synapses)[2]. Even in the rat brain it is estimated that there are approxi- mately 200 million neurons [3], [4]. The connections are wired such that the problem is highly nonlinear. In a certain class of seizures it would be helpful if they could be detected even a few seconds prior to the start of a seizure. The dimensionality of the problem can be significantly reduced, with only a minimal loss of information by recording electrical potentials at multiple points on the surface of the skull or, using depth electrodes, in the hippocampus (EEG). EEGs are accepted as one of the best means of evaluating neurocognitive functions [5]. EEG spike/seizure detection and prediction is made more complicated by the following: (1) For a single individual, no two seizures or even their EEG correlates are exactly alike, (2) seizures from different individuals vary significantly, (3) there is no single metric that consistently changes during all seizures, (4) correlation among channels can change signif- icantly from one seizure to the next, and (5) even experts disagree as to what constitutes a seizure [6]. Occasionally, the reduction in dimensionality does result in an indeterminate mapping from EEG record to animal state (i.e. it is not surjective or onto). For the reasons listed above, rigid seizure detection rules do not produce good results [7], [8]. Interictal spikes are brief (20 - 70 ms) sharply contoured waveforms that stand out when compared to background EEG rhythms and may be indicative of an underlying epileptic process. Because they are considered an indicator of the presence of epileptic seizures, and may actually precede a seizure (sentinel spike), the detection of these interictal, transient spikes which may be confused with artifact or noise is indeed a crucial element in the prediction of epileptic conditions. II. RECORDING EPILEPTOGENESIS Until 1992 most EEG analysis was based on analysis of brain slices [9] or anesthetized animals [10]. Kainic acid, a chemoconvulsant extracted from seaweed, was introduced to induce seizures in animals. This provided a major break- through particularly with the advent of monitoring the animals on video, but the equally significant subclinical seizures were impossible to detect with video monitoring alone. The field was further advanced through the development of a tethered recording system [11] in which multi-channel cortical and sub-cortical recordings could be obtained. The quality of recordings was further improved by incorporating a small pre- amplifier close to the skull, allowing for a significant increase in the signal to noise (S/N) ratio. As shown in Figure 1, electrodes were placed stereotaxically in the hippocampus and secured in the skull [12], [13] Additional electrodes were placed directly on the dura. Dental cement was applied to hold the electrode pins together in a plastic cap that was later connected to the pre-amplifier. The pre-amplified signal was sent to an amplifier and from there to a computer for storage. Our facility has the capability of continuously monitoring up to 64 tethered or untethered rats. Untethered rats underwent