Research On Signal Processing And Pain Recognition Based On EEG

Electroencephalogram(EEG)is an overall reflection of the electrophysiological activity of brain cells on the surface of the cerebral cortex or scalp,and contains a large amount of physiological and pathological information.Recently,studies of the pain-related EEG have been one of the hot and hard issues in the fields of cognitive sciences and clinical care.Objective assessment of pain can help patients with infants,postoperative patients and other patients who cannot express pain information to make objective judgments on pain,and is of great value for clinical pain treatment.In this paper,the techniques of signal preprocessing,feature extraction,feature selection and classification identification of multi-channel clinical pain EEG signals are studied to achieve the objective recognition of three pain levels(no pain vs.pain vs.heavy pain).The collection scheme of clinical EEG signals in pain is firstly introduced.Because the EEG signal is weak and susceptible to various noises(such as ocular electricity,ECG,etc.),the independent component analysis method is used for preprocessing to remove the noise while retaining useful information.In order to extract the features related to pain,a feature combination method based on discrete wavelet transform is proposed to extract four combinations of the wavelet energy ratios,coefficient statistics,sample entropy,and phase-locking values.ANOVA is used for rough feature selection,and the features with significant differences at different pain levels are selected.Then the support vector machine is used for pattern recognition.Furthermore,the influence of each sub-band signal on the recognition ability of pain is analyzed,which indicates that the changes of brain activity caused by different pain intensities are reflected in the EEG signals of each bands.In order to further improve the classification accuracy of the pain recognition,the combination of Random Forest(RF)and Sequential Backward Selection(SBS)is used to select features and eliminate a large number of redundant features.Then,the optimal feature subsets are used for training and prediction.Comparing the classification results of ANOVA-SVM and RF-SBS algorithm,it is concluded that RF-SBS algorithm is more suitable for pain recognition,and the classification accuracies are 80.95%for 3 pain intensity levels(no pain vs.pain vs.heavy pain),100%for 2 pain intensity levels(no pain vs.pain),and 92.85%for 2 pain intensity levels(pain vs.no pain or heavy pain),respectively.In addition,the statistical analysis of the feature categories contained in the optimal feature sets shows that the sample entropy and phase-locked value can reflect the change of pain.