Research On Feature Fusion And Brain Network Of EEG And FNIRS For Depressive Disorder Patients

Since the beginning of the new century,the economy and society have developed rapidly,and life pressure has continued to increase.Mental diseases such as depression have increasingly plagued people,and the accuracy of the diagnosis of depression is relatively low.This paper combines the synchronized three-channel EEG signals and the functional near-infrared spectroscopy signals of the prefrontal lobe obtained from the audio stimulation experiment to assist in identifying depression,and studies the feature fusion of EEG and fNIRS signals and the brain network for depression.The main research is as follows:(1)Due to the high time resolution of the EEG signal and the low time resolution of the near-infrared signal,an experimental paradigm based on audio stimulation was designed.The three-channel EEG and fNIRS acquisition equipment were used to collect the synchronized EEG and near-infrared signals under the audio stimulation.(2)Preprocess the EEG signal and the near-infrared signal separately,and then extract linear and non-linear features to classify patients with depression and healthy people after preprocessing.Since EEG data has only three channels,filtering,segmentation,and baseline correction are needed to complete preprocessing.Perform band-pass filtering,motion artifact removal,segmentation,baseline correction on the near-infrared signal,and then I complete the preprocessing.After the preprocessing is completed,I extract linear features such as variance and peak value,non-linear features such as power spectrum entropy,Kolmogorov entropy,and C0 complexity of the EEG signal;I also extract linear features such as variance and peak value,non-linear features such as Pearson correlation coefficient.Then I use SVM,KNN and BP neural network as classifiers to classify.(3)First,directly fuse the features of EEG and fNIRS.The PCA and LDA algorithms are used to fuse features,and the classification effects of the two feature fusion methods are discussed,and compared with the previous feature classification effect.The results show that when LDA fusion features are used and the KNN classifier is used for discrimination,a classification accuracy of 87.5% can be achieved.(4)Because the EEG signal has only three channels,and the near-infrared signal has 22 leads,the near-infrared signal is used to construct the brain network connection diagram,and the clustering coefficient,average shortest path length and the global efficiency are calculated.This method is to use the features of the brain network to classify.The results show that the classification accuracy rate is the highest when using the K-nearest neighbor classifier for the brain network features,which is 84.38%.The experimental results show that for the prefrontal lobe,we choose an appropriate algorithm to perform feature fusion of the forehead three-channel EEG signal and the near-infrared spectroscopy signal,which can get a higher recognition rate of depression than the single-modal signal.This result reflects the advantages of the feature fusion.