EEG Signal Analysis for Effective Classification of Brain States

EEG (Electroencephalogram) is a non-stationary signal that has been well established to be used for studying various states of the brain, in general, and several disorders, in particular. This work presents efficient signal processing and classification of the EEG signal. The digital filters used during decomposition of the input EEG signal have transfer functions which are simple and easily realizable on digital signal processors (DSP) and embedded systems. The features selected in this study; energy, entropy and variance; are among the most efficient and informative to analyze the EEG signal strength and distribution for detecting brain disorders such as seizure. Training and testing of the extracted features are performed using linear kernel (Support Vector Machine) SVM and thresholding in DSP algorithms and hardware, respectively. The experimental results for the digital signal processing algorithms show a high classification accuracy of 95% in the occurrence of seizure in epileptic patients. The techniques in this work are also under investigation for classifying other brain states/disorders such as sleep stages, sleep apnea and multiple sclerosis