Chaos based nonlinear analysis of epileptic seizure R.Sahu 1 , T.Parija 1 ,B.Mohapatra 1 , B.Rout 1 ,S.Sahu 2 , R.Panda 2 , P. Pal 3 , T. Gandhi 4 Abstract– Feature extraction and classification of electro-physiological signals is an important issue in development of disease diagnostic expert system (DDES). In this paper we propose a method based on chaos methodology for EEG signal classification. The nonlinear dynamics of original EEGs are quantified in the form of entropy, largest Lyapunov exponent (LLE), correlation dimension (CD), capacity dimension (CAD) and were considered for discrimination of various categories of EEG signals. After calculating the above mentioned parameters for signals, we found that without going for rigorous time-frequency domain analysis, only chaos based parameters is also suitable to classify various EEG signals. Keywords: electro-physiological signal, DDES, chaos theory, Lyapunov exponent, correlation dimension, capacity dimension I. INTRODUCTION Epilepsy is a chronic disorder characterized by recurrent seizures, which may vary from a brief lapse of attention or muscle jerks, to severe and prolonged convulsions [1], [15], [16]. The seizures are caused by sudden, usually brief, excessive electrical discharges in a group of brain cells (neurons) (Cited in WHO, http://www.who.int/topics/epilepsy/en/). About 1million of the entire population in the world are characterized by intermittent abnormal firing of neurons in the brain. Eighty percent of the epileptic seizure activity can be controlled or can be treated effectively, if properly detected and diagnosed (Cited in WHO, http://www.who.int/topics/epilepsy/en/). The brain activity in the ictal state (during a seizure) differs significantly from the activity in the normal state with respect to frequency and pattern of neuronal firing due to presence of higher amount myogenic artifacts [15]. The detection of seizure is very challenging because of its nature and origin. Electroencephalograms (EEGs) contain very useful information relating to different physiological states of brain and thus are very effective tool to understand the complex dynamics of the brain. Since most cases EEG are non-invasive, it can be recorded over a long time span to monitor the incidental disorder like epileptic seizure for presurgical evaluations to determine the epileptogenic foci of the brain. These EEG recordings are visually inspected by trained neurophysiologist for detecting epileptic seizures [2] or other abnormalities present. This information is then used for proper clinical diagnosis and then accordingly various therapies, medications or surgical treatments are administered to the subjects. Moreover, due to the human error, leads to the improper diagnose of the disease causing fatal to human life. Thus a lot of effort has been devoted by biomedical engineers, researchers to develop an automated expert system for seizure and epilepsy detection which might help not only the physician to speed up the process with greater accuracy but also reduce the amount of data needs to be stored (Oeak, 2008a; Subasi, 2007; Tazel & OZbay, 2009). Various techniques have been proposed in the literature (Oeak, 2008b; Iasemidis, Shiau, & Chaovalitwongsa, 2003; Khan & Gotman, 2003) for the detection of seizures and epilepsy and various different physiological conditions of brain by analyzing the EEG in various domains. Iasemidis and Sackellares were first to state about non-linear dynamics. Bullmore et al. differentiate between normal and epileptic seizure signals. Hively et al. proposed a chi- square statistics and phase space visitation frequency to quantify chaos in EEGs and to detect the transition from nonseizure to epileptic activity. According to the knowledge of author there are many paper based on chaos which is used for EEG classification but there are some shortcomings as they are giving false indications sometimes. An effective classification must be proposed for the distinctions which consider some points like decrease sensitivities to noise and muscle artifacts and increase sensitivities towards the disorder which is required must for our classification. The other drawbacks are the numbers of data on which the research is performed; as by taking less numbers of the data sets the chances of getting error is more. That’s why for getting effective results the detection must be performed on large numbers of data sets which enhances the accuracy of the classification. This paper focuses on chaos methodology for analysis of various groups of EEG signals. The methodology is applied on five subgroups and the simple non-linear chaos parameters like Lyapunov 1 Biomedical Engg., MIET, BBSR, Orissa 2 Biomedical Engg., TAT, BBSR, Orissa 3 Biomedical Engg., NIT, Raipur 4 Biomedical Engg., AIIMS, New Delhi Corresponding Author: gandhitk@gmail.com Third International Conference on Emerging Trends in Engineering and Technology 978-0-7695-4246-1/10 $26.00 © 2010 IEEE DOI 10.1109/ICETET.2010.111 594 Third International Conference on Emerging Trends in Engineering and Technology 978-0-7695-4246-1/10 $26.00 © 2010 IEEE DOI 10.1109/ICETET.2010.111 594 Third International Conference on Emerging Trends in Engineering and Technology 978-0-7695-4246-1/10 $26.00 © 2010 IEEE DOI 10.1109/ICETET.2010.111 594 Third International Conference on Emerging Trends in Engineering and Technology 978-0-7695-4246-1/10 $26.00 © 2010 IEEE DOI 10.1109/ICETET.2010.111 594