Complex Network Analysis Of EEG And ECG Signals

Physiological signals reflect various functions of the human body,including emotion,fatigue and healthy state.The most typical and common used physiological signals include EEG,ECG.et al.Complex network has become a crucial theory for data analysis,which allows excavating important information underlying physiological signals.This dissertation develops visibility graph theory for analyzing the RR interval signals,the epileptic EEG signals and the fatigue driving EEG signals.ECG signals directly reflect the state of the heart.An accurate and timely identification of abnormal ECG signal is of great significance.We previously introduced a novel limited penetrable visibility graph,which presents a good anti-noise ability.This dissertation firstly develops a novel time-dependent limited penetrable visibility graph(TDLPVG)to infer complex networks from RR interval time series.The results suggest that our TDLPVG method allows characterizing the time-varying behaviors and classifying heart states of healthy,congestive heart failure and atrial fibrillation from RR interval time series.Combining random forest classifier and our TDLPVG method,a high classification accuracy of 93.5% has been obtained.Seizures can seriously impair the mental and physical health of patients.EEG is considered as an indispensable information source for diagnosing epilepsy in clinical applications.Therefore,the characterization of epileptic seizure from EEG signals becomes quite important.This dissertation proposes a novel multiscale limited penetrable horizontal visibility graph to classify EEG signals recorded from healthy subjects and epilepsy patients,and the classification accuracy is 100%.The classification of EEG signals in seizure free intervals and epileptic seizures represents a more challenging task than that of the classification of normal and epileptic seizures.This dissertation proposes a novel method of adaptive optimal kernel visibility graph for detecting epileptic seizures from seizure free intervals.The results show that our method can effectively distinguish these two kinds of EEG signals,and the classification accuracy is over 98%.Fatigue driving has attracted more and more attention in that it directly accounts for of the increasing traffic accidents.EEG signals has been deemed as an effective sources of information for fatigue detection.In this dissertation,the experiment of fatigue driving is elaborately designed to obtain the multi-channel EEG signals in normal and fatigue states.On this basis,we propose a multilayer limited penetrable horizontal visibility graphs to detect fatigue state from EEG signals.The results indicate that this method can effectively identify two brain states of normal and fatigue during human driving,which also provides a new way for complex network analysis of multivariate time series.