Automatic Detection Of Epileptic Characteristics In EEG Signals Based On Sparse Representation And The Design Of An Application System

EEG (Electroencephalography, EEG) examination is by now the most convenient and mature manner for clinic to take the brain electrical signal by noninvasive access. It also provides an efficient approach to make clinical diagnosis of brain diseases and research on neurophysiology and brain science. In fact, the generation and collection of nerve evoked electrical signals, and also automatic detection and classification of epilepsy features have played a very important role in clinical testing, EEG ward, control and treatment of epilepsy and other brain diseases.The main focus in this thesis is the automatic detection of epileptic spikes in EEG signals based on sparse representation. The innovation results achieved in the thesis are summarized as follows.Firstly, a multi-component mathematical model is established for the producing process of EEG signals, taking the neuroelectricity activity dipole model and header model as the physiological foundation and the epileptic characteristics in abnormal EEG signals as the prior foundation. Then, a structure adaptive sparse decomposition model (SSDM) is proposed for EEG signals, based on the hypothesis that EEG signals can be of sparse representation. In fact, the two models have been the foundation of our research in this thesis.The second constribution in this thesis is the designed sparse dictionary for EEG signals. To represent EEG signals as sparsely as possible, the atoms in the designed dictionary should match the inherent structures in the EEG signals as closely as possible. Since EEG signals are locally observed much like Gaussian probability density, Gaussian wavelet, and Mexico-hat wavelet, the generation functions of the dictionary are intuitively chosen as the Gaussian and its first-order and second-order derivative functions, producing a sparse dictionary capable of matching kinds of characteristic epilepsy waves in abnormal EEG signals. Experiment results demonstrate that the designed sparse dictionary behaves more efficiently than the usual Gabor dictionary in sparse representation of the EEG signals. Moreover, compressive sampling of EEG signals is also discussed based on sparse representation.The core of this thesis is the newly proposed algorithm of automatic detection of epileptic spikes in EEG signals through our quantitative dectection criteria which have been established based on the detection criteria in clinic diagnosis and our designed sparse dictionary. In the first stage of the algorithm, an adaptive autoregressive prediction filter is used as a pre-detector to detect all the possible epileptiform transients. This pre-detection is not only able to reduce the complexity of the algorithm but also improve the overall detection performance of the procedure. In the second stage, the time-frequency parametrization of EEG signals is provided using our designed sparse dictionary and the matching pursuit method, capable of describing quantificationally the epilepsy characteristic waves in the abnormal EEG signals. Through comparing with the pre-established detection criteria, the time-frequency parametrization can be used to automatically dectect the epileptic spikes in the abnormal EEG signals. Numerous experiment results show that the proposed algorithm is not only capable of detecting the periodic spike sequences in EEG signals, but also effectively elliminating the influences of background rhythm and artifacts. Compared with previous detection methods based on the Gabor dictionary, our algorithm behaves more efficiently and accurately.Motivated by the need of clinical testing of epilepsy and other neurological diseases, a portable application system is designed for sampling and processing the EEG signals, which can easily implement the relevant processing algorithms based on sparse analysis. The design of hardware system has been based on technologies of Zigbee short-range wireless communi-cations and SD card interface for mobile computing devices such as PDA. The design of the application system includes a couple of aspects:(a) the design of SD card for short-range wireless telemetry and corresponding test terminals; (b) the programs for device drivers and data acquisition and processing running on the Windows CE operating system. The overall design process of hardware system is simple, and that of software system is perfect. Practical applications demonstrate that our designed system has met the most research on evoked potentials, in particular to those micro-intrusive types.