| Depression is a global mental illness,and the number of patients suffering from it accounts for about 4.3% of the global population.Depression also has various negative effects on people's physical health and is the main cause of disability,and depression patients also have a high suicide rate.Therefore,depression has brought a great health and economic burden to the entire society.At present,the diagnosis of depression patients is basically carried out in the form of scale interviews,which is very subjective,which makes the diagnosis accuracy rate of depression low.In recent years,with the development of biological information and computer technology,EEG signals are increasingly used in the research of depression and other diseases.Previous studies have shown that compared with healthy controls,the functional network of patients with depression has changed.In this study,EEG data were collected from 27 mild depression patients and 27 healthy controls under the cuetarget paradigm.In this paper,two types of functional networks are calculated by using regions of interest and electrode as network nodes respectively.First,based on the source location method,84 regions of interest are defined as nodes to construct a functional network,and then 63 electrodes are used as nodes to construct a functional network.Analyze the difference between the functional networks corresponding to depression patients and healthy controls and classify them based on the features extracted from the functional networks.(1)First,use independent log-F ratio statistics to analyze the areas where there are differences in electrophysiological activities between the two groups.The results showed that in delta,theta,alpha and beta bands,the electrophysiological activity of depression in Broadman area 7 is stronger than that of healthy controls.Then,the functional network is constructed using four methods: lagged coherence(LC),lagged linear connectivity(LLC),lagged nonlinear connectivity(LNC)and lagged phase synchronization(LPS).Use the threshold method to binarize the obtained function matrix.The clustering coefficient and characteristic path length were calculated on the binarization matrix and the multivariate repeated measure analysis of variance was used for statistical analysis.It was found that when the threshold value was 0.014,the clustering coefficient and characteristic path length calculated by the LNC method and the LPS method were statistically different in the delta frequency band.Under the condition of the cue is color block and the cue is consistent with the target,compared with healthy controls,patients with depression have a higher clustering coefficient and a lower characteristic path length.Perform statistical analysis on the functional connections under the condition that the cue is color block and the cue is consistent with the target.It is found that the nodes with statistical differences in functional connections are mainly located in the left and right occipital lobe in the delta frequency band,mainly located in the left frontal lobe in the theta frequency band,and mainly located in the left parietal lobe in the alpha frequency band.(2)Four different functional networks are calculated,two of which are networks constructed with electrodes as nodes using coherence(COH)and phase locked value(PLV)methods,the other two are the functional networks calculated by LC and LPS with the region of interests as nodes,and then feature selection and small sample classification are performed on the four frequency bands.The feature selection strategy is CFS+BF,and the classifier uses traditional naive bayes,logistic regression,K nearest neighbor and random forest.The connectivity corresponding to the LPS method reached the highest classification accuracy rate of 98.15% under the condition of the cue is color block and the cue is inconsistent with the target.The clustering coefficient,characteristic path length and global efficiency of the network are calculated,and multifactor repeated measurement variance analysis is performed.Under the condition of the cue is color block and the cue is consistent with the target,the characteristic path length calculated by the LPS method in the delta frequency band and the COH method in the theta frequency band have differences between groups,and the characteristic path length of depression is less than that of the healthy control.Under the condition of the cue is color block and the cue is consistent with the target,the global efficiency calculated by the coherence method in theta frequency band has differences between groups,and the global efficiency of depression is greater than that of healthy controls.For the statistical analysis of local efficiency and betweenness centrality under delta and theta conditions,the phase locking value method has no difference nodes.At the nodes where the local efficiency of the other three methods are different,the local efficiency of the depression group is greater than that of the healthy control group,and the difference node is mainly located in the frontotemporal area.Under each condition,use statistically different local efficiency and betweenness centrality for classification.The highest accuracy rate of 79.63% was obtained on the delta frequency band,and the highest accuracy rate of 94.44% was obtained on the theta frequency band.Through the above classification research and statistical analysis of functional connectivity network,it was found that there were differences in network characteristics between the depressed patients and the healthy controls under the condition of the cue is color block and the cue is consistent with the target.The global efficiency and local efficiency of the depression group are higher than that of the healthy control,and the characteristic path length is shorter than that of the healthy control.The main difference between the two groups is the frontotemporal area.The functional connectivity estimation method based on source location has better performance than the connectivity estimation method that directly uses electrodes as nodes. |
PDF Full Text Request