Research And Applied To The Key Technology Of Constructing Brain Protein Network Model Based On PET Image

As the most complex system in nature,human brain includes different regions.These regions carry different brain functions.In daily life,thinking and physical activities are completed through cooperation of different brain regions.It is an important branch in human brain that explored the specific cooperation process of brain regions and how the different diseases affect the brain.In recent years,complex networks and graph theory analysis have been introduced into brain research.It is popular to deeply study the brain from the network perspective.Combined with machine learning methods,it has broad application prospects in disease diagnosis and prediction.For different diseases or different modal data,the construction and analysis methods of brain network are different.Previous studies mainly focused on pathological analysis on group network,but lack of the research in individual network.There are many kinds of brain images,how to explore the internal relationship between different brain images is also an urgent problem.Positron emission computed tomography(PET)is an advanced clinical imaging technology in nuclear medicine field.Large numbers of PET data have been used in the research of brain diseases,especially in Alzheimer's disease(AD)and mild cognitive impairment(MCI).In recent years,Tau protein probe(AV-1451)and A? protein probe(AV-45)have been developed and widely used.These two proteins are closely related to the diseases of AD and MCI.AD is a neurodegenerative disease with irreversibility characteristics.MCI is seen as the precursor of AD,in addition MCI can be recovered.Compared with the irreversibility of AD,the research on MCI patients will be more important.Based on the protein PET image,this paper analyzes the properties of brain protein network in MCI patients,and provides basis for computer-aided diagnosis.There are three main contributions and innovations:First,we used parallel independent component analysis to study the relationship between Tau protein and A? protein.The study covered two groups of people: MCI and normal people.First,we use the two sample t-test to find the most significant brain function subnetwork components in each protein between normal people and patients with MCI.The 8 different components of each protein data were paired for correlation analysis.Finally,we get a group of the mostrelevant difference components between the two proteins.The results can well reveal the spatial distribution of Tau protein and A? protein in MCI.Secondly,Tau protein is thought to be highly related to the process of MCI,so we focused on analyzing the brain connection network of Tau protein.Previous studies focused on how the MCI effect Tau protein and this paper focused on the relationship between the factors(Apolipoprotein e4(APOE 4)genotype and abnormal physiological indexes of cerebrospinal fluid(CSF))that affect the MCI and Tau protein network.By building Pearson's correlation network,calculating small world propertites,brain network propertites(clustering coefficient(Cp),shortest path length(Lp),node centrality(HUB),modularity,etc.)and the robustness of the whole brain network,we found that these pathogenic factors did impact on Tau protein brain connection network,and the damaged brain areas are related to human emotion,memory and motor function.Among them,APOE 4 gene is more closely related to disease than other factors.Thirdly,we studied the contribution of brain protein network to MCI assisted diagnosis.Firstly,regions of interest(ROIs)extracted by parallel independent component analysis were used as the feature for classification analysis.The results showed that these ROI regions could achieve to82.14% for distinguishing MCI population.Then,we constructed high-order sparse individual brain network based on multiple protein features which named multi features high order protein network(MHPN).MHPN could show the abnormal protein network organization in MCI patients.Using F-score feature selection method,linear SVM classifier and 10 fold cross validation,the network properties were classified as features.The results showed that the classification accuracy of MHPN network properties for MCI was reached 95.24%.This study confirmed that brain protein network is of great value in MCI research and auxiliary diagnosis.