Topology Analysis And Application Of Uncertain Brain Network Based On Frequent Subgraph Mining

The brain is a sophisticated and precise organ.Although human beings have accumulated some knowledge about the composition and function of the brain,in view of the complexity of the operation between one functional area and another related area of the brain,human exploration of the brain continues.Studies have shown that the emergence of brain diseases and some areas of the brain cannot work properly.In recent years,the advent of the f MRI technology ushered provides technical support for guarding against and diagnosing of brain diseases.The introduction of graph theory made brain network become a hot spot.Researchers combine brain network with computer technology widely used in the field of artificial intelligence,such as machine learning,deep learning and other computer technologies to mine frequent subgraphs,extract and select features,and classify brain network distinguishing depression from normal people,so as to obtain biomarkers related to diseases.Traditional brain network research focused on the certain graph,neglecting the uncertainty information between brain networks,and the problem of sparsity selection should be considered when constructing the certain graph.At present,there is no gold standard for the selection of sparsity.Therefore,some researchers apply the uncertainty map to the modeling of brain,and construct the uncertain brain network,which can reflect the uncertainty information of brain interval well.By comparing the existing research of uncertain brain network and certain brain network,we cannot explain which model is effective for brain modeling from the angle of classification accuracy.Moreover,there is no gold standard for the selection of brain network models.In addition,in the traditional uncertain brain network research,the statistical indexes such as mean,variance and extreme deviation were used to extract features.However,the statistical indicators of mean has the problems of poor classification accuracy and bad generalization performance;variance reflects the average difference between subgraph modes,and the result takes square for sample value,which cannot directly reflect individual differences between subgraphs;the range can be seen from the calculation formula that it reflects the biggest difference among subgraphs,which is greatly affected by the minimum value of maximum value,and cannot directly balance the difference between quantum graphs.Therefore,in view of the existing problems in the current research,the main works are as follows:(1)The method of constructing uncertain brain network based on independent component analysis is proposedConsidering the uncertainty of brain network,this paper applies the uncertain graph to construct the brain network.Moreover,the nodes corresponding to the uncertain graph are divided by group independent component analysis.This method using blind source signal separation technique,adds group elements to analyze the data,and the results obtained are statistical significance.Comparing with the traditional method,this method has no prior template,which avoids the influence of different templates on the experimental results,and can better reflect the spatial relationship between different components.(2)A comparison method between uncertain brain network and certain brain network based on discriminant subgraphs is proposed.Considering the uncertainty of brain network,the traditional method uses certain graph to model brain,which ignores the uncertainty of brain.But by comparing the classification results of uncertain brain network and certain brain network in the existing experiments,it is difficult to elaborate the strengths and weakness of the two models in terms of final performance.At present,there is no corresponding research that shows which model is better to use for brain network modeling,which has no effective reference for the use of the two models in the future.Therefore,based on the discriminant subgraphs,this paper introduces a two-way comparison method of sparsity and weighting of the model of the uncertain graph.And then,the classification performance of the uncertain brain network model and the certain brain network model is compared.It is appropriate and effective to determine which model is selected to model the brain from the angle of classification,and the probability of brain disease was predicted.(3)A feature extraction method based on relative extreme difference is proposedBy analyzing the existing statistical indicators' problems,this paper introduces a new statistical index to extract the feature of relative range.It is the combination of mean and range.It can effectively overcome the influence of the maximum minimum value of the statistical index in the original method,considering the maximum difference between subgraph modes and the group difference between subgraphs The large and the difference of subgraphs cannot be directly measured.Then,the feature selection is carried out by the corresponding discriminant score function,and the feature matrix is constructed to classify.The above results testify that the classification of uncertain brain network is better than that of the certain brain network,while the sparsity is in the range of(0.25-0.40),and the classification performance of the certain brain network is better than that of the uncertain brain network.This conclusion supplies a reference for the model selection of brain network in terms of disease diagnosis.In addition,the relative range proposed in this paper is a new feature extraction method,and its classification performance is significantly higher than the mean,variance and range of the existing statistical indicators.Moreover,the index has good classification performance and strong generalization under different feature selection methods.This paper focuses on the comparison of uncertain brain network model and certain brain network model,and proposes a new feature extraction method to find the most suitable model for brain network and the most discriminative features,in order to supply a distinct method for the prevention and diagnosis of brain diseases.