Research On DIE-based Brain Network State Observation Matrix Dimension Reduction Method

Nuclear magnetic resonance imaging technology provides a favorable means for studying the characteristics of the human brain,and because of its high temporal and spatial resolutionand the resting state magnetic resonance imaging with blood oxygen level dependent signals played an important role in deep research for human brain functional network dynamic characteristics,in the mean time the brain network reconstruction technique has become one of the most powerful tools to study the features of human brain.In view of the complexity of the human brain network,when extracting the properties of human brain network,a kind of human brain network state observation matrix is constructed with rather high dimension and it is difficult to identify its main features,so the research of dimensionality reduction and clustering method for human brain network state observation matrix is very necessary.Considering the above-mentioned situation,focusing on the dimensionality reduction and clustering of high-dimensional human brain network state observation matrix,some research works has been discussed in this thesis at the basis of deep learning theory.Firstly a dimensionality reduction algorithm based on deep autoencoder is presented and a five-layer restricted boltzmann machine framwork is constructed and implemented.By this method the brain network features in high dimensional space can be mapped into low-dimensional space,and which provide a new solution for the dimensionality reduction of high-dimensional brain network state observation matrix.Secondly in order to verify the reliability and validity of this deepencoder dimension reduction algorithm,the self-organization mapping method is used to cluster the state observation vectors in low dimension space after dimensionality reduction.Finally the experiments results show the feasibility of this method by different samples and this work provides the necessary basis for further study of the dynamic characteristics of the human brain network.