The Study Of EEG Microstate Analysis And Identification In Generalized Anxiety Disorder

Generalized Anxiety Disorder(GAD)is a common chronic psychiatric disorder that causes worry or anxiety as well as various physical discomforts.The etiology of GAD is still unclear and its comorbidity with other mental disorders is high,making it difficult to correctly identify in reality.Currently,the conventional diagnosis of GAD relies on a "Symptom Description" approach,which request psychiatrists to use psychological assessment scales,self-report questionnaires,and patient interview records to get comprehensive judgement.However,these traditional diagnostic methods have subjective bias,shallow diagnostic coverage,poor generalization ability,and a lack of quantitative standards.These problems urgently need to be resolved.Functional Magnetic Resonance Imaging(f MRI)studies based on the Blood Oxygenation Level Dependent(BOLD)principle have shown that GAD can lead to abnormal activation of specific brain regions in patients,thereby affecting the normal processing of anxiety signals in the brain.However,due to the lack of sufficient temporal resolution,this method cannot effectively explain potential dynamic changes in the brain related to GAD at the sub-second scale,and corresponding temporal features have not yet been proposed.To address the above mentioned issues,this paper explores the temporal dynamics of the brain in GAD patients during anxiety-induced experiments using microstate analysis and source localization techniques based on a high temporal resolution multichannel scalp electroencephalogram(EEG).Meanwhile,this study also constructs an effective neural network classification model to achieve an efficient auxiliary diagnosis of generalized anxiety disorder by combining the time domain features with the spatial domain features of EEG signals,and provides a new way to identify the disease quantitatively.The main research contents of this paper are as follows:(1)The purpose of this study is to investigate the differences in brain dynamics and activation states during the execution of a situational recall paradigm in 15 GAD patients and 14 healthy controls(HC).First,a standard microstate analysis process was performed on the task-state EEG data,and five microstate averaging templates were generated using a modified K-means algorithm,and two atypical microstate topologies(MS-4 and MS-5)were identified.In addition,by exploring the correlation between EEG microstate category templates at different frequencies,it was found that microstate components in Alpha and Beta bands have higher contribution to global brain activity.Then,the Duration,Occurrence,Time coverage parameters and transition probability(TP)features of microstate sequences were extracted for statistical analysis between groups.The results showed that compared with the HC group,the Duration and Coverage of MS-5 microstates in the GAD group were significantly reduced,and there was a trend of positive correlation with the scores of the GAD-7 scale.Finally,the source localization results revealed that MS-5 and MS-4 class microstates had overlapping regions involving the prefrontal,parietal,and temporal lobes and the cingulate gyrus related to brain emotion cognition and memory,and showed significant intergroup activation differences.Taken together,the findings of this paper reveal that abnormalities exist in the higher sensory cortex of the brains of GAD patients,which affect the processing of anxiety signals,and that MS-4 and MS-5 class microstates may be important markers of abnormal baseline brain activity in patients with GAD.(2)In this paper,an improved capsule network is proposed for capturing spatial features of EEG signals and combining them with temporal parameters of microstates for more reliable automatic GAD recognition.To better facilitate the spatial feature extraction of EEG signals,sensor-level and source-level data transformation ways of EEG input data are performed.Among them,the sensor-level transformation aims to convert multi-channel EEG sequences into planar frame sequences based on the position coordinates of scalp electrodes,while the source-level transformation maps EEG signals into source signal frames from different brain regions by source reconstruction to get more accurately represent spatial information.Subsequently,the EEG data with two different structures are fed into a neural network for spatial feature extraction.In this paper,an attention mechanism is innovatively introduced in classical capsule network model,which is dedicated to learning a set of weights to enhance the contribution of certain features to GAD recognition or to weaken the interference effect of redundant signals.Finally,the spatial features extracted by the model are combined with microstate temporal parameters to obtain a classification accuracy of up to 92.70% in the GAD dataset.Therefore,the spatio-temporal feature fusion framework proposed in this paper provides a new scheme for GAD-aided diagnosis study.