Brain Network Construction Analysis And Classification Based On Minimum Spanning Tree And Group Lasso

Brain network has been acknowledged as one of the most sophisticated mystery network systems which we have explored now in our realistic world.There are billions of neural cells in our brain,it records and processes tens of millions of messages every day,reflects different types of connection while recognition,feeling or action occur.Recently,complex brain network analysis attracts more and more attention especially in brain disease.And it must become a hot topic in medical revolution and an essential way in exploring brain mystery.The complex brain network not only can help to understand the mechanism of neural spirit disease,but also it can further provide potential imageology annotation and new methodology for clinical brain disease diagnosis.With the development and integration of cross-curricular interests,brain network analysis will play an important role in brain neuro spirit disease research.In this paper,two approaches of resting state brain network construction and analysis are proposed based on complex network theory in complicated brain network research.Both two new classification models are used in autism patient classification research.We apply functional brain network analyzing inter-group network metrics and explore the differences among autism patients and iconography symbols for depressive disorder diagnosing which can assist clinical diagnosing.The main contributions of this paper can be summarized as follows,(1)Brain network analysis based on minimum spanning tree.There are huge various clinical characterizations of autism in different ages,but these differences are difficult to be detected based on the imaging indicators.In order to solve this problem,this paper employs the minimum spanning tree analysis method based on the static state functional brain network.Node attributes,such as degree,betweenness centrality and eccentricity,are used to analyze the differences between different age groups(children-adolescent,adolescent-adult).The goal is discovering the variation rule of brain network on autism patients among different age groups.(2)Hypernetwork construction and analysis based on Group Lasso.Recently,some novel network construction methods are proposed,such as hypernetwork,according to the deficiency of traditional brain network construction.However,the ability of the typical hypernetwork construction method is limited due to the influence of brain inter-group.The constructed super-edge exists a certain degree of randomness,it makes the network lacking the ability to explain the grouping effect information and ultimately reduces the classification accuracy.Thus,we propose a Group Lasso based hypernetwork construction approach and put it in automatic diagnosing for autism disease patients.The proposed hypernetwork replaces the selection of a single variable of the traditional method with group variable selection,i.e.,variable selection is performed on the basis of a pre-defined variable group.Compared with the traditional hypernetwork construction method,the proposed method based on group lasso can effectively remove the effect of group effects and improve the classification accuracy.(3)The verification of SVM based classification models.While the brain network based on minimum spanning tree and hypernetwork based on Group Lasso are both constructed,the classification features are extracted according to the statistical significance differences.And then a framework with higher accuracy is constructed which fuses multiple parameters optimizing classification based SVM classification algorithm.The experimental results on autism patients of different ages show that there are significant differences in two groups(child-adolescent,adolescent-adult)comparative analysis,and the classification accuracy rates are 80.38% and 81.88%,respectively.This method provides new views and ideas for imaging analysis and auxiliary diagnosis of patients with autism at different ages.The experimental results of autism and normal group comparisons on hypernetwork based on group lasso also show that our proposed method obtains better classification accuracy and the average accuracy can rate to 87.84%.The results reflect the effectiveness of our proposed methods.And our proposed approaches give great help to autism disease diagnosing.