Feature Extraction Research Of Epilepsy EEG Signal Based On Image Analysis

Epilepsy, as a brain-related disease, seriously affects people’s health. It often causes great trouble in the patients’daily life. The electroencephalogram (EEG) is the signal that can most directly reflect the brain activity among all bioelectrical signals. The characteristics of EEG include small amplitude, strong noise, low frequency, great randomness and nonlinearity. These features make the analysis and processing of EEG a great challenge. The analysis results of epilepsy EEG can be used to diagnose the disease. Feature extraction is the most important part during analyzing the epilepsy EEG. The effectiveness and reliability of the extracted feature information have a direct influence on the epilepsy diagnosis result.Feature extraction of the epilepsy EEG has always been the hot topic in the domain of biomedical signal processing all over the world. Currently, according to the related literature, the methods of EEG feature extraction research are focused on the traditional one-dimensional EEG processing and the nonlinear dynamics methods. But the traditional methods cannot easily detect the nonstationarity in EEG. Even though the nonlinear dynamic methods can extract the nonstationary features of EEG effectively, the algorithms are too complex and inefficient.In this paper, a new method was put forward to realize the classification of the normal period, the pre-ictal and ictal in the epilepsy EEG. It was a feature extraction method based on the image analysis of the epilepsy EEG. First, we reconstructed the original one-dimensional EEG signal into two dimensions based on the two-dimensional reconstruction theory. Then the two-dimensional signal were converted into image according to the correspondence between the matrix signal and the image. At last, the processing and analyzing were made to the image and the related matrix. In this paper, the image entropy, zeroPer and magDWT were used as the characteristics of classification. And the results of our experiment were analyzed. The results showed that the classification method with the image entropy as the characteristic value could recognize the normal signal easily. The accuracy rate could be as high as 92%. For the zeroPer as the characteristic value, YzeroPer30 could also distinguish noramal signal easily, and its accuracy rate is as high as 96%. For the magDWT characteristic value, YmagDWT30 could distinguish the ictal signal well, and the accuracy rate reaches as high as 96%. The results proved the method we used worked well classification of every state in the epilepsy EEG signal. Besides, our study could provide a new guiding to the study on the other biological signals, such as EMG, ECG and so on. And it has significant meaning for the diagnosis and treatment of the related diseases.