Computer-Aided Diagnosis of Epilepsy Based on the Time-Frequency Texture Descriptors of EEG Signals Using Wavelet Packet Decomposition and Artificial Neural Network

摘要

An adaptive time-frequency (t-f) representation of electroencephalographic (EEG) signals with high time and frequency resolutions using wavelet packet decomposition are introduced in this paper for automated diagnosis of epilepsy. The novel texture pattern techniques namely local neighbor descriptive pattern (LNDP) and symmetric weighted LNDP (SWLNDP) are proposed to obtain distinct features from the t-f images. Proposed texture pattern techniques are insensitive to local and global variations as the consecutive neighboring pixels are compared. SWLNDP is a modified version of LNDP which improves the computational efficiency of the system by reducing the feature vector length. The histogram based features are extracted from the texture pattern of t-f images and fed into artificial neural network (ANN) for classification of signals. The obtained results show that ANN attained an accuracy of 100% using proposed techniques for classifying epileptic and normal signal. Further the performance of the proposed system was analyzed for fifteen different cases using University of Bonn EEG dataset.