Proceedings of 2015 RAECS UIET Panjab University Chandigarh 21-22 nd December 2015 978-1-4673-8253-3/15/$31.00 ©2015 IEEE Detection of Epileptic Seizure using Wavelet Transformation and Spike based Features Gurwinder Singh Department of CSE SLIET Longowal Email:kareergurwinder@hotmail.com Manpreet Kaur Department of Instrumentation and Control SLIET Longowal Email:aneja_mpk@yahoo.com Dalwinder Singh Department of CSE SLIET Longowal Email:dalwindercheema@outlook.com Abstract—Electroencephalogram (EEG) covers the detailed information regarding the neurological activity of human brain which is further used to analyze abnormal activities of which one of the abnormal activity is epileptic seizure which occurs due to sudden excitement of large number of neuron cells simultaneously. In this paper, spikes based parameters are used for epilepsy detection, as spikes are the main characteristics of seizure prone EEG signal. The signal is preprocessed by wavelet transformation and after that parameters are extracted from both normal and ictal (seizure activity) signal. Artificial Neural Network (ANN) is considered for classification and performance is measured on the basis of accuracy, sensitivity and specificity. A comparison of the proposed method with the other techniques shows the acceptable nature of this proposed method for seizure detection. Keywords—Electroencephalogram (EEG); Epileptic seizure; wavelet ; Artifical Neural Network (ANN) I. INTRODUCTION Epilepsy is one of the most common neurological disorders. Nearly 4% of world population experience seizure at some stage of their life out of which 1% establish epilepsy[1]. Epilepsy is an unprovoked abnormal activity of brain, in which neurons produces extra electrical charges which lead to disturbance in normal body functioning. These electrical activities of brain are recorded in the form of a graph known as Electroencephalogram (EEG), which is used to analyze the effected part of brain. EEG is a non-invasive method and is very efficient to understand dynamics of brain. Epilepsy patients are put under analysis for long term which results into a very large EEG signal, manual analysis of which is a time consuming process[2]. Hence the need of automatic method of epilepsy detection is sought by many technicians[3]. The use of an automatic method during initial phase may reduce the data presented to technicians for analysis of epilepsy. Research related to the automation of seizure detection from an EEG started in 1970s. In paper[4], authors developed an algorithm which relied on the frequency based characteristics to differentiate seizure prone EEG from a normal EEG. In a time domain method, authors have searched for periodic and rhythmic pattern in EEG signal which occurred during epileptic EEG signal [5]. Wavelet Transformation (WT) provides information about the signal in both frequency and time domain which is best suited for non- stationary signal as it captures spikes related, entropy based and correlation dimension features in signal. Since EEG is also a non-stationary signal so WT is the best suited. WT is used by many researchers for the EEG analysis in the context of epilepsy detection with amplitude, frequency, entropy and wavelet coefficients as discriminating parameters of two classes (epileptic and non-epileptic EEG) [2], [6]–[10]. Another widely used approach is Empirical Mode Decomposition (EMD), which provides instantaneous frequency data of a signal in form of Intrinsic Mode Functions (IMFs). EMD is deployed in seizure detection by calculating frequency and amplitude based parameters from epileptic and normal EEG signal [11]–[13]. In this work a new spike based parameters have been proposed for the detection of epileptic seizure from an EEG signal. The EEG signal is preprocessed by Discrete Wavelet Transformation (DWT) into sub-bands and purposed spikes based parameters are extracted from these sub-bands. The method is tested on the public available dataset and results show better classification accuracy as compared to work of other researchers who have used same dataset. The paper is organized as follows: Section II describes proposed method in which subsection A gives information about dataset, subsection B about wavelet transformation, subsection C describes proposed features and subsection D about Artificial Neural Network. In section III, results of the proposed method are discussed and last section conclude the whole study. II. PROPOSED METHOD A. Dataset A dataset has been obtained from Bonn University, Germany which contains five sets A, B, C, D and E [14]. Each set consists of 100 single channel EEG signals of 23.6 sec duration each, which has been recorded by 128 channel amplifier and digitized by 12 bit A/D convertor at 173.61 Hz sampling frequency. The 173.61 Hz sampling frequency means each second of signal provides 173.61 data points so the total number of data points in each signal are 4097. The acquisition system has 0.5 to 85 Hz bandwidth. The set A and B contain the EEG’s of five healthy patients with their eyes open and closed under awake condition, using standard placement of electrodes. Sets C, D and E contains EEG’s from five epileptic patients, out of which C and E are recorded from epileptogenic zone and D from opposite to an epileptogenic