Research On Feature Extraction Algorithms Of Task-related Eeg For Attention Mechanism Of Depression

Human brain is a vast complex system composed of billions of neurons,which has varieties of cognitive functions relying on the neurons’ discharge synchronously.With the rapid development of the neural electrophysiological techniques,non-invasive EEG technology(EEG)can help us tracking brain activity with its high temporal resolution.EEG technology has made a great progress in detecting how our brain works.Especially,with the Event-Related potential(ERPs)technology,we can not only record dynamically the information processing procedures of the brain which evoked by external stimulus or internal mental activity in brain,but also get the activated patterns of the brain and the working states related to the higher cognitive functions.This has also made it possible for us to study the different states and levels of human mental activity.Analyzing event-related potentials induced by different cognitive tasks and extracting the characteristics of ERPs has always been the focus of the researches in brain science and cognitive neuroscience,and it also has been an important research direction in the field of biomedical signal processing.Traditional psychological research methods only focus on the time-locked features of ERPs,and simply analyze the geometric characteristics of ERP waveform.By doing so,it will lose the abundant physiological and psychological information contained in ERPs.In recent years,the research techniques in information science and signal processing are more concerned with the unlock-time characteristics of the evoked ERPs.By using computer technology and modern signal processing methods,based on the precise models of computer processing and the quantitative analysis of multidimensional characteristics of event related potentials,we can get a deep mining of the brain information,which will help us finding more characteristics of the brain signal to characterize the neural mechanisms of the brain.Therefore,how to analyze EEG signal reasonably and establish the accurate digital computational analyzing methods effectively is still an extremely important problem which has not been completely solved.Event-related potential is an unstable nonlinear signal involving rich information,which has been studied only in a single dimension of time or frequency,and it cannot be described its characteristics on the space scale.So,in this article,a systematic study on extraction algorithms of task-related EEG signal processing in time,frequency and space domains was carried to extract the multidimensional features.The task-related ERP components which represent the attention cognitive functions of the brain have been found.Especially,through the emotion processing cognitive tasks completed by patients with depression,we have paid attention to the underlying mechanisms of attention.After the further researches on the activated regions of brain structures and brain functions,we use these brain network features as an auxiliary tool in diagnosing patients with mood disorders and mental illness.The main contributions and innovations are as follows:First,we developed an analytical method based on multivariable domain including time,frequency and space,through which multivariate tensor decomposition was used to extract different dimensional characteristics of task-related EEG.Based on the theory of tensor model,the method of multidimensional tensor decomposition was analyzed systematically.And a multivariable Tucker model was used to construct a multidimensional tensor decomposition algorithm.This method can extract the multidomain features of ERP signals,and construct the multidimensional feature extraction algorithm based on the regularized multivariable estimation model,which can adaptively decompose the different phase locking components.Compared with the traditional ERP component extraction methods used in psychology,this algorithm gave an effective improvement to them.It has changed the way that ERP components analyzed by experts’ experience.This analytical method can not only make the personal factors less in extracting ERP component effectively,but also make the analysis of ERP components more objective.Second,the source estimating methods based on MNE model were systematically analyzed.According to the above feature extraction algorithm of the multivariable tensor decomposition based on time,frequency and spatial domains,the minimum norm estimation algorithm based on the different emotional valences used as weights were constructed by taking the characteristic components of the different obtained ERP components.MNE estimation model was used to trace the activated brain regions related to ERP components.The evoked ERPs were weighted with the different threshold values according to the different emotional levels which were usedas parameters of calculating of the minimum norm values.The MNE model was improved because of adding the computational factors of emotion.This tracing method is more accurate than the traditional sLORETA with which calculates the dipole current density.It can represent the activated brain regions and reflect the brain structure characteristics which are relative to cognitive tasks.Besides,it can also imply the neural mechanisms of brain activity which are relative to cognition.In this paper,we investigated the specificity of the activated functional brain areas.Both the brain functional areas which were changed and the ERP components which can reflect the abnormality in patients with major depressive disorder were found during the emotional attention task.It had provided an auxiliary basis for quantitative analysis of the diagnosis of depression.Third,according to the ROI scalp nodes traceable to the source,the brain network algorithm based on the phase coupling characteristics of the ERP components was constructed.In this algorithm,we used the phase coupling characteristics of the ERP components to construct the brain network.It can describe the coupling characteristics of the different brain functional areas while performing cognitive tasks.By analyzing the value of measurements of clustering coefficient,characteristic path length,global effect and modularity,it is found that the brain networks constructed based on PLV characteristics of ERPs can reflect the brain network activities related to tasks better.The characteristics of specific brain function network of depressed patients were analyzed.The differences of brain functional structure between depressed patients and healthy population were investigated.At last,based on the above three parts of my work,the analysis of ERP signals were situated gradually in a thorough process from outer brain to inner.We have made the scalp EEG analysis method longitudinally extend to the inner cortex,from the task-related ERP features to the task activated brain regions,dynamically analyzed the coordinated responses in different brain functional areas.We studied the neural mechanisms of attentional cognitive tasks of brain activity.By constructing algorithms to analyze and extract event-related ERP features according to the cognitive experimental paradigm,both the cognitive neural mechanisms of task-related attention and the modulated mechanisms of different emotions on cognitive attention in patients with severe depression were discussed.The results showed that the influence of negative sad emotion on attention control of patients with major depressive disorder appeared in the early and late stages of cognitiveattention,and the cognitive components founded could be applied to the clinical diagnosis and intervention treatment of major depression.To sum up,by constructing the algorithms of extracting the characteristics of task-related ERPs,we effectively extracted the event-related potential components of emotion-induced attention and analyzed their spatial and temporal characteristics.By using the obtained components to construct the source locating modeling,the functional brain areas activated by tasks were located.Then,the task-related brain networks were constructed.Combined with the cognitive experimental paradigm,we focused on the neural mechanisms of cognitive attention in patients with major depressive disorder to explore both the cognitive psychological characteristics and the dynamic neural mechanisms of the brain information processing.These results will benefit for the further understanding of brain neural mechanisms of depression,mood disorders,psychological disorders,and other mental diseases.Furthermore,it will have important meaning to study the pathogenesis and the characteristics of clinicopathology of neuropsychiatric disorders.