Comparison of empirical mode decomposition and coarse-grained procedure for detecting pre-ictal and ictal condition in electroencephalography signal

This study evaluates the use of multiscale signal analysis to detect and predict seizures by finding the ictal and pre-ictal condition in electroencephalography (EEG) recordings. There are three processing stages in this study. The first is to decompose EEG signals by using empirical mode decomposition (EMD) and a coarse-grained (CG) procedure to obtain signal information in multiple scales. The second is extracting the features by calculating the fractal dimension of the decomposed signals. Eventually, k-NN, Random Forest, and support vector machine (SVM) classifiers are used to classify ictal and pre-ictal conditions. We evaluate the system using a public dataset from Bonn University. The combination of EMD with five IMFs, FD, and SVM is used for seizure detection (normal vs. ictal) and the three-class problem (normal vs. pre-ictal vs. ictal). The accuracy for seizure detection is 100%. For the three-class problem, we achieved a highest accuracy of 99.7%, and sensitivity and specificity of 99.7% and 99.9%, respectively. The combination of CG, FD, and SVM is proposed to predict a seizure (normal vs. pre-ictal) and achieves a maximum classification accuracy from 99.3% to 100%. These results indicate that the use of EMD with five IMFs is suitable for detecting seizures, while CG is suitable for predicting seizures in EEG signals.