Feature Extraction Of Eeg Signals Based On Pattern Recognition Method And Application In Diagnosis Of Stroke

Stroke,especially ischemic stroke whose percentage is 60?80%,is one of major causes for the death of human.Time window of treatment,just four to five hours,is too short so that timely diagnosis and treatment are very important.For the moment,diagnosis methods are almost based on imaging diagnosis including Computed Tomography(CT),nuclear magnetic resonance(MRI),transcranial Doppler(TCD)and angiography.However,it needs to take a long time before getting the result of the imaging diagnosis,and it's invasive to inject contrast agent for a much clearer imaging.Electroencephalography(EEG)is quite different.The advantages of EEG signal including high accuracy,easy to operate of acquisition process and noninvasive for patients make it always an important research direction in the diagnosis of stroke.The data of this article come from Beijing Tiantan Hospital of Capital Medical University including 8 left cerebral infarction patients,4 right cerebral infarction patients and 4 healthy persons as a contrast.The research contents of this article include following 3 parts:(1)The methods of Short Time Fourier Transform(STFT)and frequency band energy is used to get three indicators which are delta frequency band energy percentage(DFBEP),alpha frequency band energy percentage(AFBEP)and ratio of the latter to the former(AD).The results are in the following.It is generally symmetrical for the three indicators between left and right cerebral hemisphere for the healthy persons.For the stroke patients,DFBEP in the ipsilateral hemisphere is always higher than the contralateral side,while AFBEP and AD are the opposite.The above find can help determine the lesion of the stroke.An EEG signal of a left cerebral infarction patient that lasts for 15 hours is further analyzed.Compared to the medical record of this patient,the energy of each frequency content changes with the progress of treatment and recovery of patient accordingly,which could be used to monitor the patient condition during the treatment.(2)The hierarchical clustering method is used to cluster 8 groups of EEG signals of each person from time domain,frequency domain and time-frequency domain.The results show that the cluster analysis of healthy people is consistent,and the stroke patients are messy,but there is no clustering result consistent with healthy people,so it can be used to determine whether a person is a healthy person.(3)The combination of STFT,principal component analysis(PCA)and box plot is used.Using EEG data of healthy people as training data and EEG data of left and right stroke patients as the test data,a method for classifying patients with left and right cerebral ischemic stroke was obtained.The influence of two factors,with/without signal preprocessing and window length,on the classification results is discussed.The result is as follows.With/without signal preprocessing has little effect on this classification method.When the window length reaches 60s in left stroke patients,and reaches 90s in right stroke patients,the results tend to be stable.