CN110473635B

CN110473635B - Analysis method of relation model of teenager brain structure network and brain function network

Info

Publication number: CN110473635B
Application number: CN201910749392.4A
Authority: CN
Inventors: 邢建川; 丁志新; 张栋; 王翔; 卢胜; 孔渝峰; 王艺颖; 徐志敏
Original assignee: University of Electronic Science and Technology of China
Current assignee: University of Electronic Science and Technology of China
Priority date: 2019-08-14
Filing date: 2019-08-14
Publication date: 2023-02-28
Anticipated expiration: 2039-08-14
Also published as: CN110473635A

Abstract

The invention discloses an analysis method of a relation model of a brain structure network and a brain function network of teenagers; comprises the following steps; step one, establishing tested data information; step two, analyzing the autism brain structure network; thirdly, analyzing a brain function network of the autism; analyzing the cause-effect network of the brain; and step five, establishing the relationship between the structural network and the functional network. The analysis and research on the structural network can provide clues for mining the autism treatment target in medicine, and the analysis and research on the functional network, the resting state brain local activity and the cause-effect network can provide reliable basis for the diagnosis of doctors, and has important scientific and theoretical significance for the research on the relationship between the structural network and the functional network; is beneficial to the early diagnosis and treatment of the autism and has important social significance. The field of application of the invention can be found in; the health care product comprises (1) diagnosis and treatment of diseases, (2) smoking addiction, network addiction and network game addiction, and (3) cognition and other health fields.

Description

Analysis method of relation model of teenager brain structure network and brain function network

Technical Field

The invention relates to the field of brain network construction, in particular to an analysis method of a relation model of a brain structure network and a brain function network of teenagers.

Background

Autism is one of the representative diseases of pervasive developmental disorders that are common in the pediatric population. From the current statistical data, the prevalence rate of autism is about three-thousandths, wherein the invention rate of boys is much higher than that of girls, about 3.5 times.

The autism seriously interferes the brain intelligence development and the psychological normal development of the sick children, if the sick children are not treated in time, the sick children will have the symptoms of social communication disorder, communication disorder with people, interest stenosis and the like, and the sick children are also decocted in families. In addition, the children patients are also at risk of other concurrent mental diseases, for example, about 3/4 of the children patients will have symptoms of mental retardation, and even 1/4-1/3 of the children patients will have epilepsy. Even some children with the disease have poor intelligence on one hand and have supernormal performance on music, mechanical memory and the like on the other hand. Therefore, it is reasonable to suspect that the brain network of the autistic infant has abnormal development compared to the normal one.

Since the last half century, with the development of brain imaging techniques such as magnetic resonance, humans have never so intuitively observed the living brain, the most complex system of this known living body. If the imaging technologies can be used for more accurately diagnosing the autistic infant in the early stage, the imaging technologies are very significant and are a great thing which benefits all mankind.

Disclosure of Invention

Therefore, in order to solve the above-mentioned deficiencies, the present invention provides a method for analyzing a relation model of a brain structural network and a brain functional network of a teenager. The method is based on the magnetic resonance image data of the autism sick children and the normal control group, tries to analyze the network difference of the sick children and the normal children in the brain structure network and the brain function network, and provides clues for exploring the autism pathology examination scheme. Firstly, a structural network and a functional network are constructed, and theory in a complex network and a graph theory is introduced to research the difference between the two groups. Thereafter, signal activity and functional symmetry in the brain at rest were analyzed. Thereafter, a causal network was constructed for each test subject according to the granger causal test, and the differences in clustering coefficients among the groups were analyzed. And then, analyzing the consistency of edges in the structural network and the functional network corresponding to the two groups of tested objects so as to analyze the abnormity of the patient network. And finally, mining the relation between the structural network and the functional network by analyzing the properties of consistency, clustering coefficient, local efficiency and the like of edges in the tested structural network and the tested functional network of three groups of normal people in different age groups.

The invention is realized in this way, and constructs an analysis method of relation model of teenager brain structure network and brain function network, which is characterized in that: comprises the following steps;

step one, establishing tested data information;

step two, analyzing the autism brain structure network;

thirdly, analyzing a brain function network of the autism;

analyzing the cause-effect network of the brain;

and step five, establishing the relationship between the structural network and the functional network.

The invention relates to an analysis method of a relation model of a brain structure network and a brain function network of a teenager, which is characterized by comprising the following steps: the method comprises the following steps of analyzing a cerebral structural network of the autism, wherein the analyzing step comprises the following steps of;

(2.1) constructing a brain structure network: preprocessing by using DPARSF software to obtain an image of a gray matter part of a brain; then, the AAL90 template is used for dividing the brain image into 90 parts according to the brain area; each brain region in the template is defined as a node of the structural network, and an edge between any two brain regions is defined as a correlation coefficient of gray matter volume sequences of the two brain regions tested in the group (ASD/HC); wherein the gray matter volume on a certain brain region refers to the mean of gray matter volumes in all voxels (voxels) on the brain region; after obtaining the structural network with the weight, selecting a threshold value according to the principle that no isolated point exists in the network and the density of the graph is minimum to obtain a 0-1 binary network;

(2.2) replacement inspection treatment; the random permutation is used for calculation, and the specific steps are as follows:

1) H0 hypothesis is proposed: namely, the autistic patient group (30 cases) and the normal control group (79 cases) are from the same population, and the clustering coefficients of the respective structural networks are not significantly different (the significance level is 0.05);

2) Calculating the difference DC0 of the clustering coefficients of the initial two groups of tested structure networks;

3) Mixing the two groups of data to generate N (200 in the application) random arrangements, dividing the first 30 cases and the last 79 cases into two groups for the ith arrangement, respectively constructing a structural network, and calculating the difference value DCi of the clustering coefficients of the two structural networks;

4) The 200 differences DC1, DC2, are counted, and the number M of the differences DC200 that is greater than DC0 is counted, and the p value is calculated as: p = M/200;

5) To make an inference: if p is less than 0.05, the current sample is abnormal under the assumption that the two groups of tested samples originally belong to the same population, and the H0 assumption is rejected, namely the difference of the clustering coefficients of the two groups of tested samples has statistical significance; otherwise, the difference of the clustering coefficients of the two groups of tested samples is considered to have no statistical significance;

(2.3) brain structure network analysis; when analyzing the global property of the complex network, a series of network densities (18 in total, from 0.10 to 0.44, step size 0.02) are selected, and the complex network property difference between two groups of structural networks under each density threshold value is analyzed.

The invention relates to an analysis method of a relation model of a teenager brain structure network and a brain function network, which is characterized by comprising the following steps: in (2.3) brain structural network analysis,

in order to analyze the recovery capability of the brain network to acute and focal injuries, the situation when the network is damaged is simulated by a method of deleting nodes or edges, and the recovery capability of the network is measured by calculating the performance index of the damaged network;

on the rule of deleting a node, three rules are used: 1) In the initial state, calculating the average value of each node (brain area) in two groups of tested structure networks, sorting the nodes in a descending order, and then deleting the nodes in sequence according to the order until the deletion is finished; 2) The same as 1), carrying out descending order arrangement on the betweenness of each node, and then sequentially deleting the nodes according to the order; 3) Deleting nodes randomly until the nodes are deleted, carrying out 2000 experiments, and taking a mean value;

in the rule of deleting edges, the average value of edge betweenness of each edge in two groups of tested structure networks in an initial state is calculated, the average value is sorted in a descending order, and then the edges are deleted in sequence according to the order;

with respect to the performance metrics of the corrupted network, the present application attempts two metrics: 1) Global efficiency of the network after corruption; 2) The efficiency of the largest blob in the network after corruption; but for index 2), actually, in the later stage of node deletion, a plurality of lumps consisting of one edge and two nodes exist in the network, and the efficiency is known to be 1 according to a calculation formula of the network efficiency; if the nodes or edges are continuously deleted at this time, one sub-network always exists in the rest networks before all the nodes or edges are deleted, and the network efficiency after being damaged is even higher than that in the initial state;

in the analysis of the local properties of the structural network, the minimum density for ensuring the network connectivity is used as a threshold value for analysis; the results show that the nodularity is abnormally reduced in some brain areas, mainly including the left Olfactory cortex (olfactry _ L), the right Rectus muscle (Rectus _ R), and the right Lingual gyrus (Lingual _ R); the olfactory cortex of the human brain is closely related to the memory, and the abnormal reduction of the node degree in the area indicates that the average memory of the autism patient group in the experiment is lower than that of the normal human group; the tongue of brain mainly participates in the processing and logic analysis of visual memory, and the abnormal reduction of the part indicates that the logic analysis capability of the autistic patients is degraded.

The invention relates to an analysis method of a relation model of a brain structure network and a brain function network of a teenager, which is characterized by comprising the following steps: step three autism brain function network analysis is implemented as follows;

(3.1) constructing a brain function network; the resting brain function image original data is a group of 4D images, and the 4D images with 146 time points are obtained after preprocessing by using DPARSF software; unlike brain structure networks, here each subject can build its own brain function network; firstly, extracting signal sequences of each brain region (using an AAL90 template, and total 90), and then calculating the absolute value of a Pearson correlation coefficient between the signal sequences to express the connection strength between the brain regions, thereby completing the construction of a weight network;

(3.2) brain function network analysis; selecting a density threshold interval with 0.04-0.42 and the step length of 0.02, wherein 20 density thresholds are selected in total, and investigating the network attribute difference of 0-1 functional networks under different density thresholds;

(3.3) brain signal analysis in resting state; mining differences between groups by calculating the low-frequency amplitude ratio of the brains and the local consistency of voxels in two groups of tested resting states, and searching for a brain region with abnormal signals;

firstly, extracting fALFF values of each tested brain area (AAL 90 template) by using DPARSF software, and performing Fisher-z transformation and smoothing (smoothening) operation; the Fisher-z transformation is used for making fALFF values obey normal distribution for convenient analysis, and is specifically shown in a formula (3-1);

wherein mu and sigma are respectively the mean value and standard deviation of fALFF values of all voxels; then carrying out double-sample T test on the fALFF values of the two groups of tested blocks, setting the significance level to be 0.05, setting the number of the elements in the blocks to be not less than 40, and using a displacement test method; still using random permutation to replace for 1000 times; meanwhile, the average head movement (means FD _ Jenkinson) obtained in the pretreatment is used as a covariate to regress (NuisanceCovariates Regression);

obtaining a double-sample T test result, and performing three-dimensional visualization by using BrainNet software; the processing steps for ReHo are essentially the same;

(3.4) analyzing the functional symmetry of the brain in the resting state; with respect to the measure of symmetry, the present application uses VMHC for calculation; the specific calculation method is still to calculate the pearson correlation coefficient between the signal sequences of the spatially symmetric voxels;

the method comprises the steps of preprocessing original image data by using DPARSF software, after extracting fALFF values, only retaining data on a frequency band of 0.01-0.1 Hz through filtering, then matching resting state images on a template which is symmetrical in left and right space to reduce the influence caused by geometric difference of left and right hemispheres of a tested brain, smoothing, extracting VMHC values, performing Fisher-z transformation and smoothing operation after obtaining the VMHC values, then performing double-sample T test, setting steps and parameters in the same process as the fALFF processing process, adding tested head dynamic parameters as covariates for regression, finally obtaining double-sample T test results, and performing three-dimensional visualization by using BrainNet software.

The invention relates to an analysis method of a relation model of a teenager brain structure network and a brain function network, which is characterized by comprising the following steps: step four, performing cause-effect network analysis of the brain in the following mode;

(4.1) Glanberg cause and effect relationship; the specific steps for performing the granger causal test are as follows:

1) The stationarity of the time series X and Y is checked, and the unit root check is generally carried out on the stationarity of the time series X and Y by using an augmented diy-fullerene test (ADF test);

2) The original hypothesis "H0: x is not the cause of the glange change in Y "the following two regression models were first estimated:

2.1 Unconstrained regression model, see formula (3-2);

2.2 There is a constrained regression model, see equation (3-3);

wherein alpha is ₀ Representing a constant term, p and q being the maximum number of lag periods, ε, of variables Y and X, respectively _t Is white noise;

2.3 Calculating residual square sum RSS of the two regression models _u And RSS _r Constructing F statistic, see equation (3-4);

where n is the sample size, RSS _u And RSS _r See formula (3-5);

2.4 Selecting a level of significance

If it is not

Then beta is ₁ 、β ₂ 、...、β _q Significantly different from 0, the first original hypothesis "H0: x is not the glargine cause of the change in Y ", i.e. X is the glargine cause of the change in Y, step 2.5) is continued; otherwise, the original hypothesis cannot be rejected, and the method is finished;

2.5 X and Y are exchanged, and the second original hypothesis "H0: y is not the Glanberg cause causing the change of X, when the test result is that the original hypothesis is accepted, the first original hypothesis is rejected in the step 2.3) to draw the final conclusion that X is the Glanberg cause of Y;

(4.2) constructing a factor network of the resting state brain; analyzing the causal relationship between any two brain area signal sequences of the brain in a resting state, and constructing a brain causal network according to the causal relationship; the method comprises the following specific steps:

(4.2.1) extracting a time sequence of each brain region in the resting state functional image;

(4.2.2) calculating a caudality value between time series X and Y of any two brain regions using Rest software;

(4.2.3) setting a threshold value threshold, if the reusability is greater than the threshold, considering that directional connection from X to Y exists, and juxtaposing the corresponding position in the adjacent matrix to be 1, otherwise, considering that directional connection in the direction does not exist;

(4.3) clustering coefficients of the directed network; the clustering coefficient of the directed network is calculated by the following method;

the adjacency matrix with directed network is A = (a) _ij ) _N×N I is more than or equal to 1, j is less than or equal to N, wherein a _ij =1 indicates the existence of an edge of nodes i to j or else a _ij =0, and further specifies a _ii I is not less than 0,1 and not more than N, namely, no self-loop exists in the network; the out-degree, in-degree and total degree of the node i in the network can be calculated by formulas (3-6), (3-7) and (3-8);

wherein, I _N A column vector of all 1's;

in addition, the total number of the existing bidirectional edges of the node i is calculated by the formula (3-9);

the clustering coefficient of a certain node i of the directed network, namely the ratio of the actual number of the directed triangles taking i as the vertex to the maximum possible number of the directed triangles taking i as the vertex, is calculated by a formula (3-10);

the invention relates to an analysis method of a relation model of a brain structure network and a brain function network of a teenager, which is characterized by comprising the following steps: the relationship between the structure network and the function network comprises the following contents;

wherein, the definition of network consistency is that for any two brain areas i and j in the network, the connection between the brain areas i and j exists in the structural network and the functional network at the same time or does not exist at the same time, and a specific calculation formula is shown in an expression (3-11);

wherein N is the number of brain regions, as is the structural network adjacency matrix, and Af is the functional network adjacency matrix.

The invention has the following advantages: the invention provides an analysis method of relation models of a teenager brain structure network and a brain function network; the method is based on the magnetic resonance image data of the autism sick children and the normal control group, tries to analyze the network difference of the sick children and the normal children in the brain structure network and the brain function network, and provides clues for exploring the autism pathology examination scheme. Firstly, a structural network and a functional network are constructed, and the theory in a complex network and a graph theory is introduced to research the difference between the two groups. Thereafter, signal activity in the brain in resting state and functional symmetry were analyzed. Then, a causal network was constructed for each test according to the granger causal test, and the differences in clustering coefficients among groups were analyzed. And then, analyzing the consistency of edges in the structural network and the functional network corresponding to the two groups of tested objects so as to analyze the abnormity of the patient network. And finally, mining the relation between the structural network and the functional network by analyzing the properties of consistency, clustering coefficient, local efficiency and the like of edges in the tested structural network and the tested functional network of three groups of normal people in different age groups.

The method comprises the steps of firstly utilizing the structural gray matter volume and the functional image time sequence of each brain area to respectively construct a structural network and a functional network, then sequentially analyzing the attribute difference of the complex networks of the two networks, then analyzing several signals and functional symmetry in the brain in a resting state, then utilizing the grand causal effect to construct an effective network and analyze the difference of clustering coefficients of the effective network among groups, and finally analyzing the relationship between the two networks by analyzing the properties of consistency, clustering coefficients, local efficiency and the like of the edges in the structural network and the functional network. The analysis and research on the structural network can provide clues for excavating autism treatment targets in medicine, the analysis and research on the functional network, resting state brain local activity and the cause-effect network can provide reliable basis for diagnosis of doctors, and the analysis and research on the relationship between the structural network and the functional network has important scientific and theoretical significance. The results are better applied to practice, the early diagnosis and early treatment of the autism are facilitated, and the social significance is very important.

Drawings

FIG. 1 is a schematic diagram of two sets of structural networks to be tested (a) the structural network with weights in the ASD group, (b) the structural network with weights in the HC group, (c) the structural network after the binarization in the ASD group, and (d) the structural network after the binarization in the HC group);

FIG. 2 is a three-dimensional view of two sets of 0-1 structured networks tested (set (a) ASD, (b) HC);

FIG. 3 shows two sets of structural network clustering coefficient comparisons ((a) original comparison, (b) replacement test results);

FIG. 4 is a schematic diagram showing comparison of the lengths of the characteristic paths of two sets of tested structural networks ((a) original comparison, (b) replacement test results);

FIG. 5 is a schematic diagram showing the comparison of the small-world attributes of two sets of tested structural networks (a) original comparison, (b) replacement test results);

FIG. 6 shows the comparison of global efficiency of two sets of tested structural networks ((a) original comparison, (b) replacement test result);

FIG. 7 is a schematic diagram showing comparison of local efficiency of two sets of tested structural networks ((a) original comparison, (b) replacement test result);

FIG. 8 is a schematic diagram showing comparison between average node numbers of two tested structural networks ((a) original comparison, (b) replacement test result);

FIG. 9 is a schematic diagram showing comparison of average edge indexes of two groups of tested structure networks ((a) original comparison, (b) replacement test result);

FIG. 10 is a schematic diagram showing the modular comparison of two sets of tested structural networks ((a) original comparison, (b) replacement test result);

FIG. 11 is a comparison diagram of the overall efficiency reduction of two tested structural networks under attack ((a) sequentially deleting nodes in descending order of degree of initial nodes in the network, (b) sequentially deleting nodes in descending order of betweenness of initial nodes in the network, (c) randomly deleting nodes, taking the average value of 2000 times, (d) deleting edges in descending order of betweenness of initial edges in the network);

FIG. 12 is a graph illustrating normalized node contrast for a structured network;

FIG. 13 is a diagram illustrating the comparison of betweenness between standardized nodes in a structured network;

FIG. 14 is a diagram illustrating a comparison of normalized local clustering coefficients of a structural network;

FIG. 15 is a schematic diagram showing a comparison of clustering coefficients of functional networks;

FIG. 16 is a comparison of characteristic path lengths of functional networks;

FIG. 17 is a schematic diagram of a comparison of the properties of a functional network worlds;

FIG. 18 is a comparison graph of global efficiency of a functional network;

FIG. 19 is a comparison of local efficiency of a functional network;

FIG. 20 is a functional network synchronization comparison schematic;

FIG. 21 is a schematic diagram of a functional network hierarchy comparison;

FIG. 22 is a schematic diagram showing comparison between two groups of signals showing the local activity of the brain ((a) fALFF value, (b) ReHo value);

FIG. 23 is a graph showing the difference between two sets of VMHC values tested;

FIG. 24 is a schematic diagram showing comparison of tested VMHC values of three groups of normal persons;

FIG. 25 is a diagram illustrating a comparison of clustering coefficients between an effect network and a functional network;

FIG. 26 is a schematic diagram showing comparison of structural network and functional network consistency ((a) ASD vs HC0, (b) HC0 vs HC1 vs HC 2);

FIG. 27 is a graph showing the comparison of the attributes of three tested structural and functional networks ((a) clustering coefficients and (b) local efficiency).

Detailed Description

The present invention will be described in detail below with reference to fig. 1 to 27, and the technical solutions in the embodiments of the present invention will be clearly and completely described, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.

The invention provides an analysis method of a relation model of a brain structure network and a brain function network of a teenager by improvement, which comprises the following steps;

information of tested data: the data set used in the experiment of the application is collected by the nerve development and imaging research center of Kennedy-Crigger institute of Hopkins university, the scanner is a Philips 3T scanner, and each tested object only collects data of one Session, including structural images and resting state functional images. Of the 211 patients tested, 56 patients with autism and 155 patients with normal control were tested. The children participating in the program all meet the following conditions:

1) The age is between 8 years old and 0 months and 12 years old and 11 months and 30 days;

2) Parents or guardians provide informed consent and children provide consent;

3) The Wechsler child mental Scale WISC-IV or WISC-V full intelligence quotient exceeds 80. If there is an index difference of 12 points or more, the speech understanding index (WISC-IV or WISC-V), the perceptual reasoning index (WISC-IV) or the visual-spatial index or the fluid reasoning index (WISC-V) must exceed 80 and the lower of the indices must be greater than 65.

Furthermore, all ASD patients in this dataset met the ADOS-G or ADI-R based ASD criteria and clinical confirmation by children neurologists with a rich experience in autism diagnosis (SHM); all HC group children in this data set had no ASD first degree relatives and no history of psychiatric disease.

The data set does not provide other personal information to be tested, and the application defaults to such information as not being statistically significantly different. In addition, the data has a plurality of different time point parameters, the number of the tested objects counted by the time point is shown in the table 3-1, and the main parameters of the nuclear magnetic resonance are shown in the table 3-2.

TABLE 3-1 statistical data tested at different scanning time points

Timepoints	ASD	HC
			124	1	0
128	12	34
			156	43	120
162	0	1
			Is totaled	56	155

TABLE 3-2 NMR Scan parameters

(II) analyzing the brain structure network of the autism: the analysis work of the present application is mainly realized by means of GAT software.

(2.1) brain structure network construction:

the brain structure image raw data is a group of 3D data, and the application uses DPARSF software for preprocessing to obtain images of gray matter parts of the brain. The brain image is then segmented into 90 parts by brain region using the AAL90 template which is currently very common. Each brain region in the template is defined as a node of the structural network, and the edge between any two brain regions is defined as the correlation coefficient of the gray matter volume sequences of the two brain regions tested in the group (ASD/HC), and the pearson correlation coefficient is used in this application. Where the gray matter volume on a certain brain region refers to the mean of the gray matter volumes in all voxels (voxels) on that brain region. After the weighted structural network is obtained (as shown in fig. 1 (a) and 1 (b)), the threshold is selected according to the principle that no isolated point exists in the network and the graph density is minimized to obtain a 0-1 binary network (as shown in fig. 1 (c) and 1 (d)), and the threshold satisfying the condition in the application is 0.37428. Two groups of structural networks tested are shown in fig. 2 (a) and 2 (b) using the three-dimensional plots displayed by the BrainNet software.

(2.2)Permutation Test：

Permutation Test, the Permutation Test, was proposed by Fisher in the 30's last century. Its advantages and disadvantages are obvious. On one hand, the displacement test has low requirement on the overall distribution of data, and is particularly suitable for the research of small sample data with unknown overall distribution, and the brain network research just accords with the point, so the application is more. On the other hand, since it is necessary to participate in the calculation using the full arrangement or a large number of random arrangements of the sample data, the calculation amount is particularly large. If the permutation is performed using the full permutation of the data, the data size increases from 5 to 10, but the amount of calculation increases by about 3 ten thousand times. This is also one reason why its early application was less. Studies have shown that the results obtained by displacement tests are similar to classical parametric tests (T-test, F-test, etc.) when the sample content is large. When the sample content is small, the self-help sampling is superior to the parameter test, and the test efficiency is higher than that of the rank sum test.

The basic idea is as follows: under the premise that H0 is established, a test statistic is constructed according to the research purpose, the sample data is used, the sample is sequentially replaced, the statistical test quantity is recalculated, the empirical distribution is constructed, and then the p-value is obtained on the basis of the empirical distribution and is estimated. In practice, because the number of permutation-combinations is too large, a random permutation is used to simulate its approximate distribution, and then the probability p of the observed and more extreme samples appearing in the distribution is found, and a statistical inference is made by comparing with the significance level (typically 0.05). When p >0.05, indicating that the H0 hypothesis is true, it is trivial to observe the appearance of the sample, i.e. not reject the H0 hypothesis; otherwise, when the H0 hypothesis is satisfied, the occurrence of the observation sample is a small probability event, i.e., the H0 hypothesis can be considered not to be satisfied.

In combination with the research content of the present application, because the total array calculation amount is too large, the random array is used for calculation in the present experiment, and the following concrete steps are described by taking the inspection of the clustering coefficient as an example:

(1) The H0 hypothesis is presented: namely, the autistic patient group (30 cases) and the normal control group (79 cases) are from the same population, and the clustering coefficients of the respective structural networks are not significantly different (the significance level is 0.05);

(2) Calculating the difference DC0 of the clustering coefficients of the initial two groups of tested structure networks;

(3) The two sets of data were mixed to generate N (200 in this application) random permutations. For the ith arrangement, dividing the first 30 cases and the last 79 cases into two groups, respectively constructing a structural network, and calculating the difference value DCi of the clustering coefficients of the two structural networks;

(4) Counting the 200 difference values DC1, DC 2., the number M of DC200 that is greater than DC0, and calculating the p value as: p = M/200;

(5) Making an inference: if p is less than 0.05, the current sample is abnormal under the assumption that the two groups of tested samples originally belong to the same population, and the H0 assumption is rejected, namely the difference of the clustering coefficients of the two groups of tested samples has statistical significance; otherwise, the difference between the two tested clustering coefficients is considered to have no statistical significance.

(2.3) brain structure network analysis;

in order to reduce the influence of the subjective threshold selection method on the experimental results, a series of network densities (18 in total, from 0.10 to 0.44, step size 0.02) are selected when the global properties of the complex network are analyzed, and the complex network attribute difference between two groups of structural networks under each density threshold is analyzed.

Through calculation, we found that there was no significant difference in the clustering coefficients between the ASD group and the HC group (as shown in fig. 3), but the clustering coefficients of the normal human structural network were generally higher, indicating that the edges in the normal human structural network were relatively intensively distributed around some brain regions. The ASD group and the HC group have significant difference in characteristic path length (as shown in fig. 4), especially in the density range of 0.28-0.38, the characteristic path length of the normal human network is significantly smaller than that of the patient (p <0.05, corrected), that is, the information transmission speed of the brain network of the patient is significantly reduced. As can also be seen in FIG. 5, the small world attributes of the patient structural network have degraded as defined by the small world attributes of chapter II. Wherein the small-world attributes of the normal human structural network are all higher than those of the patient group over the density interval of 0.16-0.44, while the small-world attributes of the normal human are statistically significantly higher than those of the patient (p <0.05, corrected) over the density interval of 0.28-0.34. In combination with the changes of the reduction of the clustering coefficient and the increase of the characteristic path length of the patient group, the brain structure network of the autistic patient can be seen to not evolve to a random network (the characteristic path length and the clustering coefficient are both reduced relative to a small-world network) or a regular network (the characteristic path length and the clustering coefficient are both increased relative to the small-world network).

As can be seen from fig. 6, the global efficiency of the normal human group network is higher than that of the patient group at all network densities. This phenomenon coincides with a difference in feature path length, taking into account the relationship between global efficiency and feature path length. While in local efficiency (as shown in fig. 7), the difference between the two groups is relatively smaller (p <0.05, corrected). The combination of the two phenomena indicates that a small part of central nodes or masses in the brain network of the autistic patient are degenerated or abnormally connected, but the network can still normally work on the whole, and the efficiency is close to that of a normal person. This is consistent with the situation in real life where autistic patients can still maintain normal life.

The difference between the two sets of data is similar between the average node betweenness (as shown in FIG. 8) and the average edge betweenness (as shown in FIG. 9). The difference is shown in two aspects, on one hand, on the density interval of 0.28-0.38, the average node medium number and the average side medium number of the network of normal people are obviously smaller than those of patients; on the other hand, when the network density is low (density < 0.2), the average node number and average edge number of the normal human network are higher than those of the patient, and when the density is high (density > = 0.2), the opposite is true.

The results of the analysis on the modularity properties of the structural network (as shown in fig. 10) show that at all graph densities the differences between the two groups tested were not significant (p <0.05, corrected), and the structural network modularity performance of the normal control group was slightly better than that of the autistic patient group.

In addition, the method analyzes the performance of the structure network under attack. In order to analyze the recovery capability of the brain network to acute and focal injuries, the method simulates the situation when the network is damaged by deleting nodes or edges, and measures the recovery capability of the network by calculating the performance index of the damaged network.

In the rule of deleting the node, the application uses three rules: 1) In the initial state, calculating the average value of each node (brain area) in two groups of tested structure networks, sorting the nodes in a descending order, and then deleting the nodes in sequence according to the order until the deletion is finished (as shown in fig. 11 (a)); 2) Same as 1), arranging the betweenness of each node in a descending order, and then deleting the nodes in sequence according to the order (shown in fig. 11 (b); 3) Nodes are randomly deleted until deletion is completed (as shown in fig. 11 (c)), and the experiment is performed 2000 times and averaged.

In the rule of deleting edges, the present application first calculates the average value of edge betweenness of each edge in two sets of tested structure networks in the initial state, and sorts the edge betweenness in a descending order, and then deletes edges in sequence according to the order (as shown in fig. 11 (d)).

With respect to the performance metrics of the corrupted network, the present application attempts two metrics: 1) Global efficiency of the network after corruption; 2) The efficiency of the largest blob in the network after corruption. But for index 2), in fact, at the later stage of node deletion, a plurality of lumps consisting of one edge and two nodes exist in the network, and the efficiency is known to be 1 according to a calculation formula of the network efficiency. If the nodes or edges are deleted continuously at this time, one such sub-network always exists in the remaining networks before all the nodes or edges are deleted, and thus the network efficiency after being destroyed is even higher than that in the initial state, which is obviously unreasonable. Therefore, the present application only discusses the global efficiency of the network after being destroyed, and the vertical axis of all the partial graphs in fig. 11 is the ratio of the global efficiency of the network after being destroyed to the initial global efficiency.

As can be seen from fig. 11: when deleting nodes according to the node degree sequence (as shown in fig. 11 (a)), the recovery capability of the normal control group is slightly better, but the difference is not large; when the nodes are deleted according to the node betweenness order (as shown in fig. 11 (b)), the two groups of recovery capabilities are alternated, but the overall difference is small; when the nodes are deleted randomly (as shown in fig. 11 (c)), the two groups are tested to have almost no difference; when edges are deleted in edge-betweenness order (as shown in fig. 11 (d)), the recovery of the normal human network is better than that of the autistic patient group until 65% of the edges in the network are deleted. When the edge with the edge betweenness ordering of 27th is deleted, the network performance of the autism patient group is greatly reduced, the edge corresponds to the connection between the right top edge corner loop and the right edge top loop in the AAL90 template, the connection strength between the two brain areas of the two groups of tested patients is 0.7823 (ASD) and 0.8535 (HC), so that the connection between the two brain areas of the autism patient is easier to damage, and auditory aphasia is easy to occur when the corner loops and the like are damaged.

In the analysis of the local properties of the structural network, the minimum density for ensuring the network connectivity is used as a threshold value for the analysis. The results show that the nodularity is abnormally reduced in some brain regions (as shown in fig. 12), mainly including the left Olfactory cortex (olfactary _ L), the right Rectus muscle (Rectus _ R), and the right Lingual gyrus (angual _ R). The olfactory cortex of the human brain is closely related to memory, and the abnormal reduction of the nodosity in the area indicates that the average memory of the autistic patient group in the experiment is lower than that of the normal human group. The tongue of the brain mainly participates in the processing and logic analysis of visual memory, and the abnormal reduction of the part indicates that the logic analysis capability of the autistic patient is degraded.

Medians were abnormally elevated or lowered in some brain regions (as shown in fig. 13), abnormally elevated brain regions were mainly left supraorbital Frontal gyrus (front _ Mid _ Orb _ L), right Medial and lateral cingulate gyrus (Cingulum _ Mid _ R), left Thalamus (thaalamus _ L), and abnormally lowered brain regions were mainly right Medial suprafrontal gyrus (front _ Sup _ media _ R), right peritalar cortex (Calcarine _ R). And the clustering coefficient of the right edge top loop (supraglobal _ R) is abnormally decreased (as shown in fig. 14). The orbital-frontal cortex is a main neural mechanism generated by human emotion and is a main neural area for generating emotions of regret, pleasure, embarrassment, anger, sadness and the like. It is normally able to control the occurrence of emotions in different social situations and to modify our behaviour in accordance with our emotional response. The medial and lateral cingulated gyroids are important components of emotional circuits and participate in self-evaluation. The betweenness of the autistic disorder is abnormal, which indicates that the emotional control ability of the autistic disorder patient is abnormal. The thalamus is the transfer station of various sensory transmission of human beings, and the abnormal increase of its betweenness indicates that part of the sensory organs of the patients become abnormally sensitive or dull. In addition, the right lateral zygoma cortex is primarily closely related to vision, and abnormalities in this area indicate abnormalities in visual ability in autistic patients. These above inferences regarding the association of abnormal brain regions with the actual patient performance are limited to the data set used in this experiment, and further examination is generally needed.

The application also researches a network center (Networkhub) of the structure network, and if the degree/centrality of a certain node meets the deg _i /betw _i And > mu +2 sigma, is defined as the center of the network, where mu, sigma are the mean and standard deviation of the degree/centrality of all nodes, respectively. The results show that the autism group and the normal control group have a common network center, namely right prefrontal gyrus. In addition, the autism group had network centers such as right triangle inferior rostrum, right dorsum lateral rostrum, left precordial rostrum, left lateral margin superior rostrum, etc., and the normal group had network centers such as left triangle inferior rostrum, anterior cuneium, etc.

(iii) autism brain function network analysis: the analysis work of the present application is mainly realized by means of Gretna software.

(3.1) constructing a brain function network: the resting brain function image raw data is a group of 4D images, and 4D images with 146 time points are obtained after preprocessing by using DPARSF software. Unlike the brain structure network, here each subject can build its own brain function network. First we extract the signal sequences of each brain region (using AAL90 template, total 90), then we calculate the absolute value of the pearson correlation coefficient between the signal sequences to represent the connection strength between the brain regions as described in the second chapter, and this completes the construction of the weight network.

(3.2) brain function network analysis:

in the method, a density threshold interval with the step length of 0.04-0.42 and the step length of 0.02 are selected, and the network attribute difference of 0-1 functional networks under different density thresholds is inspected by 20 density thresholds in total.

Through calculation, the functional network clustering coefficient of the autism group is found to be smaller than that of the normal control group under all densities (as shown in fig. 15), but the difference between the two groups becomes smaller as the density of the graph increases. While there is little difference between the two groups in the characteristic path length (as shown in figure 16). It can also be seen from the definition of the small world attributes that the small world attributes of the patient functional network are weaker than the normal group, as can also be seen in fig. 17. The small world attribute of the brain function network of the autistic patient is slightly degraded, and the trend of the autistic patient towards the development of a random network can be seen according to the fact that the clustering coefficient is reduced and the characteristic path length is almost unchanged.

In terms of global efficiency (as shown in fig. 18), the efficiency of the functional network of the autism group is better than that of normal people when the density is high (0.16-0.42), while in terms of local efficiency (as shown in fig. 19), the same situation exists, which indicates that although the functional network of the autism patient is abnormal, the normal operation can still be maintained, and even higher efficiency is achieved.

It is known from the theory related to complex networks that the synchronization of a regular network is poor, and in the process of evolving to a random network, the synchronization capability of the network is enhanced along with the increase of the reconnection probability (Rewiring probability p). In a small-world network, the smaller the average distance between nodes (i.e., the characteristic path length), the better the synchronization capability of the network. In addition, for the same p, the larger the scale, the stronger the synchronization capability of the small world network. By comparing the synchronicity of the two groups of tested brain function networks (as shown in fig. 20), it can be seen that the synchronicity of the network of the normal human group is slightly better than that of the autistic patient group. However, the previous analysis of the present application has mentioned that there is no significant difference in the characteristic path lengths of the two groups of subjects, and the two groups of subjects are identical in scale (the number of nodes and the number of edges are the same) under the same network density, so that it can be found that the functional network of the autism patient group in the present application is closer to the regular network than that of the normal person, and the brain of the patient has a problem of developmental delay.

On one hand, the complex problem to be processed can be decomposed into a plurality of simpler parts with small coupling to be handed to the next layer for processing respectively, so that the processing efficiency is improved; on the other hand, the independence between layers is strong, the structure is clear, and the management and the maintenance are easy. In a complex network, layering is also a very important property, and the network efficiency and stability of the network with clear layers are good. By comparing the degrees of stratification of the functional networks of the two groups of tested brains (as shown in fig. 21), it can be seen that the normal group is slightly better than the group of autistic patients. In addition, the layering and modularization are closely related, and the modularization degree of a normal human structural network is slightly higher than that of an autism patient through previous analysis, which is reasonable.

(3.3) resting state brain signal analysis:

studies have shown that the human brain consumes a significant portion of its total energy consumption when resting, and that the task state usually does not increase energy consumption by more than 5% compared to the resting state. In addition, the energy consumed by spontaneous neural activity within the brain accounts for more than 60% of the total brain energy consumption. Therefore, the resting state occupies an important position in the research of the human brain function.

This section will find the brain area with abnormal signal by calculating the Low Frequency Amplitude ratio (fALFF) and the local consistency (ReHo) of the voxels of the brain in the two groups of the tested resting states to find the difference between the groups. These two signals are briefly described below.

fALFF was proposed in 2008 by improvements in a defect based on ALFF (Low frequency amplitude). It is known that ALFF represents the total activity intensity of all low-frequency signals of the brain in a resting state, but contains a lot of noise with no physiological significance including the ventricular position, and fALFF only considers the ratio of the total amplitude value in a certain frequency band to the total amplitude of the whole frequency band. For example, if the sum of the amplitudes is SA1 over the entire band and the sum of the amplitudes is SA2 in the 0.01 to 0.08Hz band, the ratio of SA2 to SA1 is fALFF. fALFF eliminates the interference of noise in other frequency bands, and greatly improves the specificity and sensitivity of signal detection.

ReHo, in 2004, was proposed by Zang et al, and was mainly used to analyze the consistency of spontaneous activity of each voxel and surrounding voxels when the brain was at rest. In the experiments embodied in the present application, for each brain function image tested, we obtained a 4D data containing 146 time points after preprocessing. We can then extract the active sequence for each voxel, length 146, and we can then compute the consistency of a certain voxel with surrounding voxels. With respect to the definition of surrounding voxels, there are generally 3 options: 6,18, 26. Take 27 cubes with a side length of 1 stacked into a large cube with a side length of 3 as an example: assuming that the voxel to be studied is in the mid-center, the surrounding 6 voxels are small cubes in the center of the 6 faces of the cube, the surrounding 18 voxels are small cubes in the center of the 6 faces of the cube and in the center of the 12 edges, and the surrounding 26 voxels are all small cubes except for it. In the experiments of the present application 26 surrounding voxels were used. Regarding the measure of consistency, reHo is calculated by using a kender correlation coefficient between signal sequences, and the larger the value of ReHo indicates that the signal activity similarity of the voxel and surrounding voxels is higher. The Kendell correlation coefficient is not described in detail here.

The application first extracts the fALFF values of each tested brain region (AAL 90 template) by using DPARSF software, and performs Fisher-z transformation and smoothing (smooth) operations. The Fisher-z transformation is to make the fALFF value follow a normal distribution for convenient analysis, and a specific formula is shown in an equation (3-1).

Wherein, mu and sigma are respectively the mean value and standard deviation of fALFF values of all voxels. Then, two groups of tested fALFF values are subjected to double-sample T test, the significance level is set to be 0.05, the number of the elements in the clumps is not less than 40, and a replacement test method is used. Because the calculation amount of the full permutation is huge, the random permutation is still used for carrying out the permutation in the part, and the times are 1000 times. Meanwhile, the average head movements (means fd _ Jenkinson) obtained in the preprocessing are regressed as covariates (NuisanceCovariates Regression).

The results of the two-sample T-test were obtained and visualized three-dimensionally using the BrainNet software, and the results are shown in fig. 22 (a), and the specific T-values and p-values (uncorrected) are shown in tables 3 to 3. As can be seen from the table, the abnormally elevated fALFF values of the brain areas include the areas of the brain such as the region of the central region, the supplementary motor area, the lateral lobules of the center, and the superior limbic gyrus; the abnormally lowered brain areas include the areas of the forehead, the lower back of the triangle, the medial side and the lateral cingulate gyrus.

The processing steps for ReHo are basically the same, and the three-dimensional visualization result is shown in fig. 22 (b), and the specific t value and p value (uncorrected) are shown in tables 3-4. As can be seen from the table, the abnormally elevated ReHo values in the brain areas include those in the central, apical, precordial, and parahippocampal regions; the abnormally lowered brain regions include the brain regions such as the prefrontal gyrus and the cerebral islands.

Tables 3-3 two sets of tested fALFF Dual sample T test results (uncorrected for p-value)

Brain region (ASD)>HC)	t value	p value	Brain region (ASD)<HC)	t value	p value
						Left central posterior	3.2584	0.00150	Left forehead and middle forehead	-2.1297	0.03549
Left side top up	3.4122	0.00091	Left triangle forehead downward return	-2.0966	0.03839
						Left top lower edge angle	2.8162	0.00579	Left temporal mediastinum	-2.7802	0.00642
Central anterior gyria of left side	2.2248	0.02819	Right edge upward return	-3.1061	0.00243
						The central withdrawal of the right side	3.0933	0.00253	The right triangle part of the forehead returns	-3.0536	0.00285
Right temporal mediastinum	2.2593	0.02589	Right forehead centre	-2.9797	0.00357
						Anterior wedge left	2.0089	0.04707	The medial left side and the lateral cingulum	-3.3175	0.00124
Left supplemental motion zone	2.3163	0.02245	The medial and lateral cingulum of the right side	-2.2632	0.02564
						Left central lateral leaflet	2.1268	0.03574
Right supplemental exercise area	4.0441	0.00010
						Lateral leaflet of right side center	3.2736	0.00143
Anterior wedge leaf on the right	2.3019	0.02328

Tables 3-4 two sets of test ReHo values Dual sample T test results (uncorrected for p-value)

The apical-inferior margin is the visual-linguistic center, and the fALFF signal enhancement indicates that the visual-linguistic ability of the autistic patient is stronger than that of a normal person. A decline in the fALFF signal from the antenatal return suggests a deterioration in the patient's short-term memory and decision making ability. The hippocampus is an important structure for the function of the hippocampus, and the ReHo signal enhancement of the hippocampus shows that the memory of an autism patient is more prominent than that of a common person.

The central paraleaflet and the supplementary motor area are closely related to the motor function of the human body, and the fALFF signal and the ReHo signal are enhanced, which indicates that the autistic patient has abnormal enhancement or degeneration in the motor function. Enhancement of both fALFF and ReHo signals in the central and apical loops indicates that autistic patients are more sensory. The presence of a decrease in both the fALFF and ReHo signals in the prefrontal gyrus suggests a deterioration in the patient's short-term memory and decision making abilities.

(3.4) analysis of brain functional symmetry in resting State

The human brain is spatially divided into left and right hemispheres, which are substantially symmetrical in geometry, but not symmetrical in actual function, i.e., there is a significant division of work in processing information. Studies have shown that the left hemisphere of the brain is mainly responsible for processing information related to language and analyzing specific things, while the right hemisphere is mainly responsible for processing visual signals and auditory signals, etc., and for integrating various signals. If the functional symmetry of some brain regions in the brain of a patient is improved, it can be considered that the region is abnormal. Regarding the measure of symmetry, VMHC (volume-distorted homeotropic connectivity) is used herein ^[50] To calculate. The specific calculation method is still to calculate the pearson correlation coefficient between the signal sequences of spatially symmetric voxels.

The present application also uses DPARSF software to pre-process raw image data. After the extraction of fALFF value is finished, only the data on the frequency band of 0.01-0.1 Hz is reserved through filtering. Then, the rest state image is matched on a template which is symmetrical in left and right space, so that the influence caused by geometric difference of the left hemisphere and the right hemisphere of the tested brain is reduced, and smoothness is performed. The VMHC value may then be extracted. After obtaining the VMHC value, fisher-z transformation and smoothing are performed. Then, a double-sample T test is performed, and the steps and parameter setting are the same as those of the fALFF processing process. The test head motion parameters were also added here to regress as covariates. The results of the double-sample T-test were finally obtained and visualized in three dimensions using the BrainNet software, as shown in fig. 23, with specific T-values, p-values (uncorrected) in tables 3-5. As can be seen from the table, the brain areas with abnormally increased VMHC values include the brain areas such as the juxtacral lobule and the supplementary motor area; the abnormally lowered brain areas include the frontal middle gyrus, temporal middle gyrus, and cortex around the talus. The central paraleaflet and the supplementary exercise area are closely related to the human body exercise function, and the abnormal increase of the VMHC value indicates that the related function is abnormal. An abnormality in VMHC values of the perilachrymal cortex indicates an abnormality (prominence or deterioration) in the processing of visual information by the patient group.

The present application also compares the symmetry of brain function in several groups of normal persons of different ages, as shown in fig. 24. Wherein HC0 is the normal control group in the application, and the age is 8-13 years; HC1 is partial data in the Beijing _ Zang data set in the fcon _1000 project, all normal persons, 20 tested persons in total are screened in the application, and the age is 18-23 years; HC2 is the Baltimore data set in fcon-1000 project, all normal persons, and 16 total subjects were screened in this application, with the age of 27-40 years. HC1 and HC2 both use the same process steps as HC 0. As can be seen from the figure, the functional symmetry of most brain areas is reduced after the age of 27, which shows that the division of work in the left and right hemispheres of the brain is more clear and the whole information processing capability is stronger.

Tables 3-5 two sets of test VMHC values Dual sample T test results (uncorrected p value)

(IV) brain cause-effect network analysis

(4.1) Glanberg cause and effect relationship: the grangian causal test (grangerkasaity test), proposed in 1969 by grangian (Granger), is a statistical method used to analyze whether causal relationships exist between variables, such as testing whether there is a causal relationship between time series X and time series Y.

On the one hand, if the variable X does not contribute to predicting the variable Y, then the reason that X is not Y is explained; on the other hand, if X is the cause of Y, then these conditions should be met: (1) X is helpful for predicting Y, namely, compared with the method for solving Y by using the past value of Y back and forth, the regression interpretation capability can be obviously improved by adding the past value of X as an independent variable for regression; (2) Y should not help predict X. Therefore, two original hypotheses "H0" need to be proposed for causal relationship testing: x is not the glange cause of Y change "and" H0: y is not the cause of glangel for X changes.

The specific steps for performing the granger causal test are as follows:

(1) The stationarity of the time series X, Y is checked, typically using the augmented diy-fullerene test (ADF test) to perform a unit root test on the stationarity of the X, Y time series, respectively.

(2) The original hypothesis "H0: x is not the cause of glangel for Y changes. The following two regression models were first estimated:

(2.1) unconstrained regression model, see formula (3-2).

(2.2) there is a constrained regression model, see equation (3-3).

Wherein alpha is ₀ Representing a constant term, p and q being the maximum number of lag periods, ε, of variables Y and X, respectively _t Is white noise.

(2.3) calculating residual sum of squares RSS of the two regression models _u And RSS _r F statistics are constructed, see equation (3-4).

Where n is the sample size, RSS _u And RSS _r See formula (3-5).

(2.4) selecting a level of significance

If it is not

Then beta is ₁ 、β ₂ 、...、β _q Significantly different from 0, the first original hypothesis "H0: x is not the Glanberg cause of the change in Y, i.e., X is the Glanberg cause of the change in Y, continue with step (2.5); otherwise, the original hypothesis cannot be rejected, and the method is ended.

(2.5) exchange X with Y, and examine the second original hypothesis "H0: y is not the Glandoy cause causing the change of X, and when the test result is that the original hypothesis is accepted, the first original hypothesis is rejected in combination (2.3) to draw the final conclusion that X is the Glandoy cause of Y.

(4.2) use of the grande causal test:

glange causal tests are widely used in the fields of economy, ecology, industry, biomedicine, and the like. The first application of the method in the economic field is that Sarpagriya Ray [51] explores the relationship among a plurality of macroscopic economic variables in stock markets between 1990 and 2010 in India in 2012, and finds that gold and oil prices have important negative effects on the stock markets, and domestic production total values, foreign exchange reserves, trade balance and the like have positive effects on the stock markets. In the ecological field, the cause of global warming was studied in 2012 by Antonello Pasini using glange causal test, and more evidence was obtained to show the causal decoupling between the last solar radiation and temperature. In the industrial field, yemane WoldeRufael [52] in 2014 applied the Glanberg causal test to investigate whether a causal relationship exists between power consumption and economic growth of 15 transformation economies 35 years ago in 1975. The research provides important basis for each transformation economy to establish power supply policy. Since the birth of the concept of grand cause and effect relationship in the biomedical field, the application thereof in the fields of computational neuroscience, signal processing and the like, particularly in the field of functional brain networks is increasing. In 2004, brovelli et al studied the monkey that pressed the lever in an awake state using the frangie causal relationship method and identified a causal effect from the primary somatosensory cortex to the motor cortex at a β frequency band of 15-30 Hz.

(4.3) constructing a factor network of the resting brain:

the method is combined with the principle of the grand causal effect, the causal relationship between any two brain area signal sequences of the brain in the resting state is analyzed, and a brain causal network is constructed according to the causal relationship. The method comprises the following specific steps:

(1) Extracting time sequences of all brain regions in the resting state functional image;

(2) Calculating calusality values between time series X and Y for any two brain regions using Rest software [54 ];

(3) Setting a threshold value threshold, if cautuality > threshold, considering that there is directional connection from X to Y, and juxtaposing the corresponding position in the adjacent matrix as 1, otherwise, considering that there is no directional connection in the direction;

(4.4) clustering coefficient of the directed network: the clustering coefficient can reflect the clustering degree of the network, and is different from a calculation method in an undirected 0-1 network, and the clustering coefficient of the directed network is used in the method. The calculation method is briefly described here.

The adjacency matrix with directed network is A = (a) _ij ) _N×N I is more than or equal to 1, j is less than or equal to N, wherein a _ij =1 indicates the existence of an edge of nodes i to j or else a _ij =0, and further specifies a _ii I ≦ N =0,1 ≦ i ≦ N, i.e., no self-loops exist in the network. The out-degree, in-degree and total degree of the node i in the network can be calculated by the formulas (3-6), (3-7) and (3-8).

Wherein, I _N A column vector of all 1's.

In addition, the total number of bidirectional edges present at node i is calculated by equation (3-9).

The clustering coefficient of a certain node i of the directed network, namely the ratio of the actual number of the directed triangles with i as the vertex to the maximum possible number of the directed triangles with i as the vertex, is calculated by the formula (3-10).

The method mainly analyzes and compares the clustering coefficients of two groups of cause-effect networks which are tested to be constructed under a group of different thresholds (0.04-0.46, and the step length is 0.02), as shown in fig. 25. It can be seen that there was no significant difference in the clustering coefficients of the two tested causal networks at almost all thresholds. The effective network is closer to the actual brain connection of human beings, which also indicates that the clustering coefficient of the brain network of the autism patient has no obvious abnormality to a certain extent. In addition, the clustering coefficients of the cause network and the function network are compared in the graph, and the clustering coefficient of the function network is higher under the same density, but the difference between the two is smaller and smaller along with the increase of the density.

(V) relationship between structural network and functional network

(5.1) consistency of connection; differences in the consistency of the average functional networks of the autistic patient group and the normal control group with the structural networks of the respective groups were examined. Wherein, the definition of network consistency is that for any two brain areas i and j in the network, the connection between them exists in the structural network and the functional network or does not exist at the same time, and the specific calculation formula is shown in formula (3-11).

Wherein N is the number of brain regions, as is the structural network adjacency matrix, and Af is the functional network adjacency matrix. In this subsection, a set of density thresholds (0.04-0.46, step size 0.02) was selected and the edge-to-edge consistency between the structural network and the average functional network corresponding to the patient group and the normal group at each density was compared, as shown in fig. 26 (a). It can be seen that the structural network and the functional network of the normal person are slightly better in consistency than the patient, but the difference is small, which also indicates that no obvious abnormality occurs in the brain connection of the patient as a whole. In addition, this subsection also compares the consistency of the structural network and functional network edges of three tested groups of normal people of different ages, as shown in fig. 26 (b). As a result, the consistency of the two networks of normal people of different ages is not obviously different, which shows that the change of the brain structural network and the function network has certain synchronism in the human development or aging process.

(5.2) differences between network attributes

Here, the clustering coefficients and local efficiencies of the structural network and the functional network of the tested normal persons of three different age groups were compared, as shown in fig. 27. As a result, it was found that with age, both the clustering coefficient (as shown in fig. 27 (a)) and the local efficiency (as shown in fig. 27 (b)) of the structural network gradually decreased, indicating a tendency of gradual degradation of the physiological connections between the brain regions inside the brain. The clustering coefficient and the local efficiency of the functional network are changed differently, and it can be seen from the graph that the clustering coefficient and the local efficiency both tend to increase and then decrease from the childhood stage to the juvenile stage and then to the middle-aged stage of normal people, and the middle-aged stage is lower than the juvenile stage. On one hand, the topology change of the structural network cannot be completely synchronized to the topology of the functional network, and certain independence exists between the two; on the other hand, the performance of the normal brain function network reaches a higher extreme point in the lifetime in adolescence.

The method comprises the steps of using the published autism magnetic resonance data, firstly, respectively constructing a structure network and a function network by using a structure image gray matter volume and a function image time sequence of each brain area, then, sequentially analyzing the attribute difference of the complex networks of the two networks, then, analyzing several signals and function symmetry in the brain in a resting state, then, constructing a cause-effect network by using a Glange cause-effect, analyzing the difference of cluster coefficients of the cause-effect network among groups, and finally, analyzing the relation between the two networks by analyzing the properties of consistency, cluster coefficients, local efficiency and the like existing at the edges of the structure network and the function network. The analysis and research on the structural network can provide clues for excavating autism treatment targets in medicine, the analysis and research on the functional network, resting state brain local activity and the cause-effect network can provide reliable basis for diagnosis of doctors, and the analysis and research on the relationship between the structural network and the functional network has important scientific and theoretical significance. The results are better applied to practice, the early diagnosis and early treatment of the autism are facilitated, and the social significance is very important.

The field of application of the invention can be embodied; the health care product comprises (1) diagnosis and treatment of diseases, (2) smoking addiction, network addiction and network game addiction, and (3) health fields such as cognition and the like.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An analysis method of relation models of teenager brain structure networks and brain function networks is characterized in that: comprises the following steps;

step one, establishing tested data information;

step two, analyzing the autism brain structure network, including the following contents;

(2.1) brain structure network construction: preprocessing by using DPARSF software to obtain an image of a gray matter part of a brain; then, the AAL90 template is used for dividing the brain image into 90 parts according to the brain area; each brain region in the template is defined as a node of the structural network, and an edge between any two brain regions is defined as a correlation coefficient of gray matter volume sequences on the two brain regions tested in the group; wherein the gray matter volume on a certain brain region refers to the mean value of gray matter volumes in all voxels on the brain region; after obtaining the structural network with the weight, selecting a threshold value according to the principle that no isolated point exists in the network and the density of the graph is minimum to obtain a 0-1 binary network;

1) H0 hypothesis is proposed: namely, the autistic patient group and the normal control group are from the same population, and the clustering coefficients of the respective structural networks have no significant difference;

3) Mixing the two groups of data to generate N random arrangements, dividing the first 30 cases and the last 79 cases into two groups for the ith arrangement, respectively constructing a structural network, and calculating the difference value DCi of the clustering coefficients of the two structural networks;

4) Counting the 200 difference values DC1, DC2, the number M of DC200 that is greater than DC0, and calculating the p value as: p = M/200;

(2.3) brain structure network analysis; selecting a series of network densities when analyzing the global property of the complex network, and analyzing the attribute difference of the complex network between two groups of structural networks under each density threshold;

in (2.3) brain structural network analysis,

in order to analyze the recovery capability of the brain network to acute and focal injuries, the situation of the network when the network is damaged is simulated by a method of deleting nodes or edges, and the recovery capability of the network is measured by calculating the performance index of the damaged network;

on the rule of deleting a node, three rules are used: 1) In the initial state, calculating the average value of each node in two groups of tested structure networks, sequencing the nodes in a descending order, and then sequentially deleting the nodes according to the sequence until the deletion is finished; 2) The same 1), the betweenness of each node is arranged in a descending order, and then the nodes are deleted in sequence according to the order; 3) Deleting nodes randomly until the nodes are deleted, carrying out 2000 experiments, and taking a mean value;

with respect to the performance metrics of the corrupted network, the present application attempts two metrics: 1) Global efficiency of the network after corruption; 2) Efficiency of the largest blob in the network after corruption; but for index 2), actually, in the later stage of node deletion, a plurality of lumps consisting of one edge and two nodes exist in the network, and the efficiency is known to be 1 according to a calculation formula of the network efficiency; if the nodes or edges are continuously deleted at this time, one sub-network always exists in the rest networks before all the nodes or edges are deleted, and the network efficiency after damage is even higher than that in the initial state;

in the analysis of the local properties of the structural network, the minimum density for ensuring the network communication is used as a threshold value for analysis; the results show that the node degree is abnormally reduced in some brain areas, including left olfactory cortex, right rectus muscle and right lingual gyrus; the olfactory cortex of the human brain is closely related to the memory, and the abnormal reduction of the regional node degree indicates that the average memory of the autistic patient group in the experiment is lower than that of a normal human group; the tongue of brain participates in the processing and logic analysis of visual memory, and the abnormal reduction of the part indicates that the logic analysis capability of the autistic patient is degraded;

thirdly, analyzing the functional network of the brain of the autism, and implementing the analysis according to the following mode;

(3.1) constructing a brain function network; the resting brain function image original data is a group of 4D images, and the 4D images with 146 time points are obtained after preprocessing by using DPARSF software; unlike brain structural networks, here each subject is able to build up its own brain functional network; firstly, extracting a signal sequence of each brain region, and then calculating the absolute value of a Pearson correlation coefficient between the signal sequences to represent the connection strength between the brain regions, thereby completing the construction of a weight network;

(3.2) brain function network analysis; selecting a density threshold interval with the step length of 0.02 and the density threshold interval of 0.04-0.42, wherein 20 density thresholds are selected, and inspecting the network attribute difference of the 0-1 functional network under different density thresholds;

(3.3) resting state brain signal analysis; mining differences between groups by calculating the low-frequency amplitude ratio of the brains and the local consistency of voxels in two groups of tested resting states, and searching for a brain region with abnormal signals;

firstly, extracting fALFF values of each tested brain area by using DPARSF software, and performing Fisher-z transformation and smoothing operation; the Fisher-z transformation is used for making fALFF values obey normal distribution for convenient analysis, and is specifically shown in a formula (3-1);

wherein mu and sigma are respectively the mean value and standard deviation of fALFF values of all voxels; then carrying out double-sample T test on the fALFF values of the two groups of tested blocks, setting the significance level to be 0.05, setting the number of the elements in the blocks to be not less than 40, and using a displacement test method; still using random permutation to carry out replacement, wherein the times are 1000; meanwhile, the average head motion obtained in the preprocessing is regressed as a covariate;

obtaining a double-sample T test result, and performing three-dimensional visualization by using BrainNet software; the processing steps for ReHo are the same;

preprocessing original image data by using DPARSF software, after extracting fALFF value, only retaining data on 0.01-0.1 Hz frequency band through filtering, then matching resting state image on a template which is symmetrical in left and right space to reduce the influence caused by geometric difference of left and right hemispheres of tested brain, smoothing, then extracting VMHC value, after obtaining VMHC value, performing Fisher-z transformation and smoothing operation, then performing double-sample T Brabender test, setting steps and parameters in the same processing process as the fALFF, adding tested head dynamic parameters as covariates to regress, finally obtaining double-sample T test result, and performing three-dimensional visualization by using InNet software;

analyzing the cause-effect network of the brain; the method is implemented as follows;

(4.1) granger's causal relationship; the specific steps for performing the glangel causal test are as follows:

1) The stationarity of the time sequences X and Y is checked, and the unit root check is carried out on the stationarity of the time sequences X and Y by using an augmented diy-fullerene test;

2.1 Unconstrained regression model, see formula (3-2);

2.2 There is a constrained regression model, see equation (3-3);

2.3 Calculating residual square sum RSS of the two regression models _u And RSS _r Constructing F statistics, see equation (3-4);

where n is the sample size, RSS _u And RSS _r See formula (3-5);

2.4 Selecting a level of significance

If it is used

Then beta is ₁ 、β ₂ 、...、β _q Significantly different from 0, the first original hypothesis "H0: x is not the glancing cause for the change in Y, i.e. X is the glancing cause for the change in Y, step 2.5) is continued; otherwise, the original hypothesis cannot be rejected, and the method is ended;

2.5 Exchange X and Y), check the second original hypothesis "H0: y is not the Glanberg cause causing the change of X, when the test result is that the original hypothesis is accepted, the first original hypothesis is rejected in the step 2.3) to draw the final conclusion that X is the Glanberg cause of Y;

(4.2) constructing a factor network of the resting brain; analyzing the causal relationship between any two brain area signal sequences of the brain in a resting state, and constructing a brain causal network according to the causal relationship; the method comprises the following specific steps:

(4.2.2) calculating the cautuality value between time series X and Y of any two brain regions using Rest software;

(4.3) clustering coefficients of the directed network; the clustering coefficient of the directed network is calculated by the following method; the adjacency matrix with directed network is A = (a) _ij ) _N×N I is more than or equal to 1, j is less than or equal to N, wherein a _ij =1 indicates that there are edges of nodes i to j otherwise a _ij =0, and further specifies a _ii I is not less than 0,1 and not more than N, namely, no self-loop exists in the network; the out degree, the in degree and the total degree of the node i in the network can be calculated by formulas (3-6), (3-7) and (3-8);

wherein, I _N A column vector of all 1's;

establishing a relationship between the structural network and the functional network; comprises the following contents;

wherein, the definition of network consistency is that for any two brain areas i and j in the network, the connection between them exists or does not exist in the structural network and the functional network at the same time, the specific calculation formula is shown in formula (3-11);