Epileptic Spike Detection Based On Wavelet Packet Transform And Morphological Component Analysis

Epilepsy is a transient brain dysfunction syndrome caused by repeated seizures that induced by abnormal discharge of brain neurons.Due to the different locations of abnormal discharge of brain neurons,the harm to patients is also different,It can damage patients’ nerves and cognitive functions and may also cause consciousness and emotional disorders.Abnormal waveforms such as spikes,fast-spikes,slow-spikes and fast-spikes and slow-spike complexes,usually appear in EEG signals of patients during epileptic seizures.However,due to the obvious transient characteristics such as large amplitude,short time limit,and vertical rise and fall in potential,the detection of spike waves in epilepsy EEG signals is a tough work.This paper emploies a wavelet packet transform and a morphological component analysis to implement the detection of the epileptic spikes.It also analyzes and compares the detection performance of the two algorithms.(1)Epileptic EEG spike detection based on the wavelet packet transformSeizure detection requires that the EEG signal has the highest possible frequency or time resolution in a specific time period or frequency section,and also requires a high spike detection rate.Therefore,this paper combines the wavelet packet decomposition with the physical characteristics(i.e..,amplitude and frequency)of the spike wave to detect the epileptic spikes.First,a three-layer wavelet packet decomposition on the frequency of the epileptic brain wave(0-30Hz)is performed based on the wavelet packet transform.Secondly,a reconstruction of the third layer node frequency S(3,0)(0-10.85Hz)according to the frequency range of the brain wave S(3,1)(10.85Hz-21.7Hz)and S(3,2)(21.7Hz-32.55Hz)EEG signals.Finally,the amplitude of the spike is used as the detection threshold to extract the health state and seizure state of spikes during intervals and seizures.The experimental results show that when the selected data sampling frequency is 173.61 Hz and the signal length is 23.6s,the algorithm using wavelet packet transform and spike physical characteristics can not only extract multiple spike signals of different patients in different periods(positive phase spike,negative phase spike or bidirectional spike),but also can detect the epileptic spikes with a high detection accuracy(the detection accuracy is 87%).The false detection rate of spike wave is 12%,and the missed detection rate is 12.3%.(2)Epileptic EEG spike detection based on the morphological component analysisMorphological component analysis(MCA)is based on the assumption that each epilepstic EEG signal can be expressed as a number of linear combinations with different morphological components.And for each morphological component,it is necessary to find a sparse representation redundant dictionary.In this analysis,firstly,the collected EEG signals are taken as combination of spike signal and background signal.Secondly,the DIRAC,DCT and LDCT dictionary are selected in the sparse representation.Finally,the matching sparse algorithm is used to obtain the most sparsest representation and achieve better detection results.The experimental results show that when the DIRAC is used as the dictionary for extracting the spikes and the DCT is used as the dictionary for extracting the background signal,the false detection rate of spike wave is 11.18%,and the missed detection rate is 8.93%.And when the DIRAC was used as the dictionary for extracting the spikes and the LDCT was used as the dictionary for extracting the false detection rate of spike wave is 11.08%,and the missed detection rate is 6.01%.