| Epilepsy is a chronic brain dysfunction syndrome which is caused by a number of factors. It is characterized by a large super-synchronous neuronal discharge, paroxysmal, emergency and temporary brain dysfunction. It is reported that there are approximately0.5%～2%of the people is affected by epilepsy in the world. Because epilepsy is a chronic, recurrent and long-term disease, it not only makes patients afflicted on the body, but also leads to mental and psychosocial barriers to patients. The pathogenesis of epilepsy is complex and it’s difficult to identify the cause for most cases and it is difficult to cure. Since EEG contains a wealth of physiological and pathological information, it is an effective method to diagnose brain disease by using EEG detection. This method is often used in clinical diagnosis. EEG can show the person’s brain signals in the form of an image and provide true and reliable information to the doctorto help health care professionals understand the patient’s condition timely and accurately. But the analysis of epileptic EEG is usually rely on computer graphics through health care professionals’naked eyes and combined with years of clinical experience to make judgments. Due to the high complexity and longer duration of EEG signals, the identification of EEG analysis is a relatively heavy work, and may get different results because of the experience of different doctors. Therefore, the design of EEG automatic system for classification and identification is particularly important.Automatic seizure detection system reduces the costs of long-term monitoring required by reducing the amount of data required to observe. It has become an important part of the EEG detection. The earliest widely used technique is proposed and improved by Gotman. In recent years, epileptic EEG classification is becoming a hot area; people continue to put forward some new methods to analyze EEG, such as Gabor transform, wavelet transform, etc. Meanwhile, a right features which is raised from EEG for seizure detection is essential. So far, various features have been proposed, such as, Volatility index, correlation dimension, Hurst exponent, the largest Lyapunov exponent, etc. All these features can be used instead of EEG signal for classification. In this paper, we use a new feature called Earth Mover’s Distance (EMD). Earth Mover’s Distance was first proposed by Rubner et al, used as a distribution, marking and histogram metrics. EMD is related to solve linear programming problems. Linear programming problem needs to calculate the minimum overall cost of a cost function with certain restrictions. In the calculation method of EMD, one distribution is regarded as mountain of Earth’s surface and the other distribution is regarded as low-lying parts of the Earth’s surface. The purpose of EMD is to find the minimum cost that filled in the low-lying part.It’s a hard work to design a new seizure detection system with high sensitivity and low false detection rate. Sensitivity and false detection rate has always been a contradiction. So, to find a suitable trade-off between sensitivity and false detection rate is a popular area in recent years. Previous work has proposed various techniques. In this paper, we provide a user-adjustable threshold to provide an acceptable trade-off.The main contents are as follows:Introduce the formation mechanism and characteristics of EEG. Research status at home and abroad on the automatic detection of epileptic EEG and analyze and compare the advantages and disadvantages of each method. Combining existing methods and characteristics of epileptic EEG proposed automatic classification based on the feature called EMD. Classify EEG signal and analysis the result automatically by using Boosting algorithm classification. Using a large number of EEG data to validate the algorithm. Finally, we make further instructions for prospects of the automatic classification of EEG, EEG feature extraction and other issues. |
PDF Full Text Request