Development of Data Analysis Software for BCI Acquired Signals

Abstract

Seizures are the abnormal electrical activities in the brain which cause disruptive symptoms in the body. These seizures characterize Epilepsy which is a group of chronic neurological disorder. For the diagnosis of seizures, electrical activity of the brain is observed using electroencephalography(EEG). From person to person different types of seizures and presentations vary. So there is no generic model avail- able to distinguish seizure case from normal brain activity. Manually identifying seizure periods from EEG is tedious and time-consuming task. Therefore, automatic detection of seizures is very essential. In this thesis Brain-Computer Interface (BCI) system is studied and algorithm for seizure detection is proposed. Discrete Wavelet Transform (DWT) and Logistic Regression (LR) based automatic detection of epileptic seizures is presented. First, wavelet decomposition method for multi- channel EEG was applied with 5 levels decomposition, and Energy ratio feature were extracted from these sub-bands. Also important statistical features like min, max, standard deviation, mean were extracted from EEG raw data. These features data are given as input to logistic regression model for detection and prediction of epileptic seizures. An accuracy of 79.21% was achieved from this method.