Research On Algorithms Of Multi-scale EEG Analysis And The Applications

In recent years,the incidence of depression has increased in all age groups.EEG biomarkers can help improve the efficiency and accuracy of depression diagnosis.Emotional stimuli could change the state of the brain,and depressed patients with long-term negative mood have abnormal brain structure and network.Nonlinear and nonstationary electroencephalogram(EEG)signals are the external summations of the neuronal electrical activities,reflecting the state of brain.Traditional nonlinear dynamic analysis has been widely used in EEG feature extraction and has achieved good results in emotion recognition and depression diagnosis,but there are still the following problems:(1)scale rigidity,(2)dimension confinement,and(3)high frequency feature extraction is difficult.Reconstructing the dynamic system is the core of solving the above three problems,multi-scale reconstruction technology can break the scale rigidity of traditional methods in the time domain and frequency domain,and spatial domain reconstruction can break the dimension confinement.Therefore,this paper aimed to propose new nonlinear EEG indicators to reflect the influence of emotional stimuli on EEG complexity and reveal the abnormal EEG complexity and balance of depressed patients.This paper focused on the difficulties of nonlinear analysis on EEG,according to the multi-scale reconstruction theory,the research on multi-scale EEG analysis was carried out from the following two perspectives:(1)EEG oscillations of some brain regions are active during emotion induction,therefore the multi-scale feature analysis algorithms were proposed in the time domain and frequency domain to evaluate the complexity and long-range correlation of EEG during the emotional induction process;(2)depressed patients with long-term negative mood have abnormal brain structure,while the traditional single-channel EEG analysis methods ignore the structural characteristics of the brain and the coupling and synchronization of brain regions,therefore a new multichannel multi-scale EEG analysis algorithm and a balance indicator were proposed by using the state-based method for multichannel space reconstruction.The new indicators could measure the time-spatial complexity and balance of the resting EEG,and quantitatively evaluate the severity of depressed patients.The major work of this thesis is described as follows:(1)Aiming at the scale rigidity and difficulty in high-frequency EEG feature extraction in EEG nonlinear analysis,the multiscale information analysis method and multiorder detrended fluctuation analysis were proposed by combining multi-scale reconstruction and adaptive algorithms in the time domain and frequency domain to measure the complexity and long-range correlations of EEG oscillations.The results demonstrated that(a)high-arousal negative emotional stimuli increased the complexity and long-range correlation of EEG oscillations;(b)the accumulation effect of negative emotions;(c)EEG features of negative emotional stimuli were associated with individual emotion regulation strategies.(2)Aiming at the dimension confinement problem of common nonlinear analysis methods used in EEG feature extraction,the multichannel multiscale state entropy algorithm based on multi-scale reconstruction was proposed to measure the time-spatial complexity of multichannel EEG,by compressing the space information of multivariate sequence.Based on the common characteristics of time-varying complexity in healthy complex systems,the complexity attenuation rate was proposed,to evaluate the complexity balance.(3)Aiming at the abnormal brain structure and synchronization of depressed patients,the proposed multichannel multiscale state entropy and complexity attenuation rate were used to measure time-spatial complexity and balance indicators of resting EEG in depressed patients.The time-spatial complexity and balance indicators were used as quantitative EEG biomarkers for depression screening and prediction of antidepressant drug treatment effect on multiple depression datasets.(4)Aiming at the current difficulty of large sample depression screening,a new wearable frontal lobe EEG monitor was developed,and the signal quality of the device was verified in all aspects.A high-accuracy screening model for mild to moderate depression was trained on the time-spatial complexity and balance features of frontal lobe resting EEG.The results demonstrated the application value of wearable frontal lobe EEG device and multi-scale EEG feature analysis in clinical research of depression.