Complex Network Analysis Of Biological Signals Based On Visibility Graph

Nonlinear time series is one of the most important research objects in digital signal processing and contains abundant information.With the development of time series complex network transformation algorithm complex network has become the main tool for nonlinear time series analysis.Visibility graph alogorithm,as one of them,has been widely used in various fields of real life.Based on visibility graph alogorithm,this thesis improves the mapping rules of complex network nodes and edges,and proposes new features of visibility network.All the above improved methods are applied to the analysis and research of biomedical signals,including electrocardiogram(ECG)signals,electroencephalogram(EEG)signals and eye staining images,to explore their potential characteristics and realize the detection of different types of signals.The main research work is summarized as follows:(1)An improved multiplex visibility graph algorithm is proposed to analyze ECG signal and realize automatic detection of myocardial infarction.The 12-lead electrocardiogram signals of human body are converted into multiplex visibility graph,each lead is taken as the node,and the inter-layer information between the two leads is taken as the edge weight,then the time series are mapped to a complex network.Since the fully connected network of different populations show the same topology,threshold is introduced to reconstruct the network,and weight degree and weighted clustering coefficients are extracted.The results show that the complex network of healthy subjects show more regular structure,higher complexity and connectivity,and could be distinguished from patients with myocardial infarction.The recognition accuracy of both parameters reach 93.3%.(2)A multivariate horizontal joint motif entropy algorithm is proposed for the recognition of multi-dimensional emotional EEG signals.Firstly,the key frequency band and key channel of emotion recognition are extracted by extracting the horizontal VG motif entropy feature.On this basis,the multiplex horizontal VG network is combined in pairs to extract multivariate horizontal joint motif entropy,and emotion EEG recognition is carried out under different signal segmentation windows.The results show that when the cutting window size is 10 s,accuracy of the classification of positive EEG/negative EEG,positive EEG/neutral EEG,negative EEG/neutral EEG is95.07%,97.73%,90.26%,and the accuracy of classification is 93.67%.(3)The image visibility graph is constructed and the node degree characteristics is redefined.The image of corneal ulcer staining is taken as the research object,and the image visibility graph is applied to the medical image.Firstly,the key regions extracted is constructed as images visibility graph network and the average node degree is calculated for eye disease image recognition.Then,the node-degree features of the images are extracted and mapped to the degree topology,which is used for image filtering,gray co-occurrence matrix is constructed and parameters are extracted,proving the role of the degree topology in image texture extraction,and contrast is used to achieve effective detection of eye staining images of different categories and different severity.In this thesis,we use visibility graph and its improved algorithm to convert ECG/EEG signals and medical image into complex network and analyze complex network,which is an important attempt of clinical medical detection.The visibility graph algorithm is not only simple and intuitive,but also can effectively inherit the inherent characteristics of the original time series,which provides a new idea for the follow-up research in the field of biology and medicine.