Fractal Based Feature Extraction Method for Epileptic Seizure Detection in Long-Term EEG Recording

Abstract

One of the most common brain disorders is epilepsy. A person who has epilepsy is not able to have normal days like the others. It's characterized by more than two unprovoked seizures. However, the faster detection and treatment of epileptic seizures, the quicker reduction of the disease abnormal level. Neurologists are still diagnosing, detecting, and testing a seizure manually by observing the Electroencephalogram (EEG) signals. This takes a very long time because of the irregularity of EEG signals. Hence, a Computer-Aided Diagnosis (CAD) is developed by many scientists to help neurologists in detecting seizures automatically. In this research, a CAD system was developed at CHB-MIT dataset. The EEG signals were processed at several stages through this system, namely pre-processing, decomposition, feature extraction, and classification. In pre-processing, the EEG signals were uniformed by selecting the most appropriate channels and filtered using Butterworth Bandpass Filter (BPF) to remove noise. The process continued to the decomposition and feature extraction stage using Empirical Mode Decomposition (EMD) and fractal dimension-based methods, i.e. Higuchi, Katz, and Sevcik, respectively. Then, the features were classified by Support Vector Machine (SVM). The proposed method achieved the highest accuracy at 94.72% on the Chb07 record. Meanwhile, the average accuracy was 81.2% for all records. The proposed study is expected to be applied for the detection of seizure onset in a real-time system.