Local Transformed Features for Epileptic Seizure Detection in EEG Signal

Abstract

Epilepsy is a well known neurological disorder characterized by the presence of recurrent seizures. Electroencephalograms (EEGs) record electrical activity in the brain and are used to detect epilepsy. Traditional EEG analysis methods for epileptic seizure detection are time-consuming, which has led to the recent proposal of several automated seizure detection frameworks. Feature extraction and classification are two important steps in this procedure. Feature extraction focuses on finding the informative features that could be used in the classification step for correct decision making; therefore, proposing some effective feature extraction techniques for seizure detection is of great significance. This paper introduces two novel feature extraction techniques: local centroid pattern (LCP) and one-dimensional local ternary pattern (1D-LTP) for seizure detection in EEG signal. Both the techniques are computationally simple and easy to implement. In both the techniques, the histograms are formed in the first step using the transformation code and then these histogram-based feature vectors are fed into a classifier in the second step. The performance of the proposed techniques was evaluated through 10-fold cross-validation tested on the benchmark dataset. Different machine learning classifiers were used for the classification. The experimental results show that LCP and 1D-LTP achieved the highest accuracy of 100% for the classification between normal and seizure EEG signals with the artificial neural network classifier. Nine different experimental cases have been tested. The results achieved for different experimental cases were higher than the results of some existing techniques in the literature. The experimental results indicate that LCP and 1D-LTP could be effective feature extraction techniques for seizure detection.