Automatic Detection of Epileptic Seizure Through HilbertHuang Transform

HilbertHuang Transform (HHT) consists of the Empirical Mode Decomposition (EMD) and Hilbert Transform (HT), which has certain advantages over traditional signal processing methods on nonlinear and nonstationary signal analysis due to its complete adaptability and flexibility of signal decomposition. In this paper, we begin to use the EMD to analyze the EEG signals, i.e., decompose each EEG signal into a number of Intrinsic Mode Functions (IMFs). Then, HT is implemented on these IMFs and thus the mean, variance as well as the core frequency interval of each IMF can be extracted to form the IMFs feature. The important IMFs of an EEG signal are selected and their features are combined together to form the feature vector of the EEG signal, being referred to as the HilbertHuang frequency ring. According to these feature vectors of the EGG signals, we utilize the support vector machine to learn and make the epileptic seizure and epileptic seizurefree classification and the grid search technique is used to optimize the parameters. It is demonstrated by the experiments on a typical epileptic seizure dataset using the fivefold crossvalidation that our proposed method can achieve the stateoftheart performance on epileptic seizure automatic detection.