A Classification Method and Device for Autistic Toddlers Based on Brain Network Features

Abstract

The present invention discloses a method and device for classifying children with autism based on brain network features. The method comprises: obtaining resting-state functional magnetic resonance imaging (fMRI) data of children of different age groups to be tested, and performing data preprocessing to obtain training samples; selecting feature vectors based on the training samples, and training the feature vectors to obtain a support vector machine classifier; obtaining the resting-state fMRI data to be classified as test samples, and inputting the test samples into the support vector machine classifier for classification to obtain classification results. The method of the present invention combines functional connectivity between brain networks and graph theory indicators of brain networks during feature selection, which can better reveal the neural mechanisms of abnormal brain development in children with autism spectrum disorders and improve the accuracy, sensitivity, and specificity of the support vector machine classifier.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a method and device for classifying autistic children based on brain network characteristics. Background Art

[0002] Autism Spectrum Disorders (ASD) are a group of pervasive developmental disorders with a neural basis. Studies have shown that ASD can be present at birth or even earlier. The prevalence of ASD in children has increased significantly in recent years, prompting increased public interest in the causes, symptoms, diagnosis, and treatment of ASD. Therefore, identifying effective biomarkers and developing classification models to distinguish ASD from typically developing children (TD) can help better understand the neural mechanisms underlying abnormal brain development in children with ASD.

[0003] Resting-state functional magnetic resonance imaging (rs-fMRI) is widely used in the medical field to detect various brain diseases due to its non-invasive and high-resolution advantages. Therefore, using rs-fMRI to effectively model the functional brain networks of patients with autism can help detect biomarkers and better study the brain neural activity of patients with autism. However, existing technologies often use a single classification model based on functional connectivity between brain networks or graph-theoretic indicators of brain networks, resulting in low classification accuracy of support vector machine classifiers.

[0004] Therefore, the existing technology still needs to be improved and developed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method and device for classifying autistic children based on brain network characteristics in response to the above-mentioned defects of the existing technology, aiming to solve the problem that the existing technology selects a single classification model of functional connections between brain networks or graph theory indicators of brain networks when modeling, resulting in low classification accuracy of the support vector machine classifier.

[0006] The technical solutions adopted by the present invention to solve the technical problems are as follows:

[0007] In a first aspect, the present invention provides a method for classifying children with autism based on brain network features, wherein the method comprises:

[0008] Acquiring resting-state functional magnetic resonance imaging data from children of different age groups and performing data preprocessing to obtain training samples; wherein the preprocessing includes: format conversion, removal of data from the first three time points, inter-slice time correction, head motion correction, structural-functional image registration, spatial normalization, spatial smoothing, bandpass filtering, and removal of physiological noise;

[0009] Selecting feature vectors based on the training samples, and training the feature vectors to obtain a support vector machine classifier;

[0010] Resting state functional magnetic resonance imaging data to be classified are obtained as test samples, and the test samples are input into the support vector machine classifier for classification to obtain classification results.

[0011] In one implementation, selecting a feature vector based on the training sample and training the feature vector to obtain a support vector machine classifier includes:

[0012] Acquiring a resting-state brain network of each training sample, and obtaining a functional connectivity matrix of the resting-state brain network according to the resting-state brain network;

[0013] Obtaining brain network functional connections with significant differences based on the functional connection matrix of the resting-state brain network;

[0014] Obtaining a brain region corresponding to each of the resting-state brain networks, constructing a brain network map based on the brain regions, and obtaining a first graph theory index and a second graph theory index based on the brain network map;

[0015] The brain network functional connections with significant differences, the first graph theory index and the second graph theory index are used as the feature vectors, and the fitcsvm function is used to train the feature vectors to obtain the support vector machine classifier.

[0016] In one implementation, obtaining the resting-state brain network of each training sample and obtaining a functional connectivity matrix of the resting-state brain network according to the resting-state brain network includes:

[0017] Determine seven resting-state brain networks for the left and right cerebral hemispheres of each training sample; wherein the seven resting-state brain networks include: default network, executive control network, salience network, dorsal attention network, auditory network, motor network, and visual network;

[0018] Obtaining the time series mean of each resting-state brain network of the children with autism spectrum disorder in the training sample to obtain fourteen first time series means;

[0019] Obtaining the time series mean of each resting-state brain network of typically developing children in the training sample to obtain fourteen second time series means;

[0020] Calculate the Pearson correlation coefficient between any two first time series means to obtain the first correlation coefficient;

[0021] Calculate the Pearson correlation coefficient between any two second time series means to obtain the second correlation coefficient;

[0022] Obtaining a functional connectivity matrix of a first resting-state brain network according to the first correlation coefficient;

[0023] According to the second correlation coefficient, a functional connectivity matrix of the second resting-state brain network is obtained.

[0024] In one implementation, obtaining significantly different brain network functional connections based on the functional connectivity matrix of the resting-state brain network includes:

[0025] The functional connectivity matrix of the first resting-state brain network and the functional connectivity matrix of the second resting-state brain network are subjected to permutation test and multiple comparison correction to obtain the functional connectivity of the brain networks with significant differences.

[0026] In one implementation, obtaining the brain region corresponding to each of the resting-state brain networks, constructing a brain network map based on the brain regions, and obtaining a first graph theory indicator and a second graph theory indicator based on the brain network map include:

[0027] Obtaining the brain region corresponding to each of the resting-state brain networks, and extracting the mean of the time series of each brain region as the brain network node corresponding to each brain region; wherein the mean of the time series of the brain region is the arithmetic mean of the time series of all voxels in the brain region;

[0028] Calculate the time series mean x of any two brain regions i and j i , x j , Pearson correlation coefficient r between i, j∈{1,...,M} i，j ;

[0029] According to the Pearson correlation coefficient r i，j , and obtain the functional connectivity matrix R of the brain region = (r i，j ) M×M×N , where M is the number of brain regions and N is the number of training samples;

[0030] The functional connectivity matrix R of the brain region is binarized to obtain a binarized brain network matrix W = (b i，j ) M×M×N ;

[0031] Obtaining a brain network graph according to the brain network nodes and the binarized brain network matrix;

[0032] The first graph theory indicator and the second graph theory indicator are obtained according to the brain network graph.

[0033] In one implementation, the functional connectivity matrix R of the brain region is binarized to obtain a binarized brain network matrix W=(b i，j ) M×M×N ,include:

[0034] If the Pearson correlation coefficient r in the functional connectivity matrix of the brain region i，j If the connectivity coefficient b is greater than or equal to the preset first threshold γ, i，j = 1, there is a connection between brain regions i and j, that is, there is an edge between the corresponding brain network nodes;

[0035] If the Pearson correlation coefficient r in the functional connectivity matrix of the brain region i，j is less than a preset first threshold γ, then the connectivity coefficient b i，j = 0, there is no connection between brain regions i and j, that is, there is no edge between the corresponding brain network nodes;

[0036] The binary brain network matrix W is obtained according to the connectivity coefficient. i，j ) M×M×N .

[0037] In one implementation, obtaining the first graph theory indicator and the second graph theory indicator according to the brain network graph includes:

[0038] The first graph theory index obtained according to the brain network graph is:

[0039]

[0040] Where V is the set of all brain network nodes, is the clustering coefficient of brain network node i, K i is the number of connections between the brain region i and other brain regions; e i is the subnetwork graph G consisting of brain network node i and adjacent brain network nodes i The actual number of edges in , G is the brain network graph;

[0041] The second graph theory index obtained according to the brain network graph is:

[0042]

[0043] Where E(G i ) is the brain network subgraph G composed of the brain network nodes adjacent to the brain network node i i local efficiency.

[0044] In a second aspect, an embodiment of the present invention further provides a device for classifying children with autism based on brain network features, wherein the device comprises:

[0045] Age collection module, used to collect the age of the child to be tested;

[0046] a training sample acquisition module, configured to collect resting-state functional magnetic resonance imaging data according to the age, and perform data preprocessing to obtain training samples;

[0047] A support vector machine classifier acquisition module is used to select feature vectors based on the training samples and train the feature vectors to obtain a support vector machine classifier;

[0048] The classification module is used to obtain resting state functional magnetic resonance imaging data to be classified as a test sample, and input the test sample into the support vector machine classifier for classification to obtain a classification result.

[0049] In a third aspect, an embodiment of the present invention further provides an intelligent terminal, wherein the intelligent terminal includes a memory, a processor, and an autism child classification program based on brain network features stored in the memory and runnable on the processor. When the processor executes the autism child classification program based on brain network features, the steps of the autism child classification method based on brain network features as described in any one of the above schemes are implemented.

[0050] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a program for classifying autistic children based on brain network features. When the program for classifying autistic children based on brain network features is executed by a processor, the steps of the method for classifying autistic children based on brain network features as described in any one of the above schemes are implemented.

[0051] Beneficial effects: Compared with the existing technology, the present invention provides a method for classifying autistic children based on brain network characteristics. First, resting-state functional magnetic resonance imaging data of children to be tested in different age groups are obtained, and data preprocessing is performed to minimize the impact of head movement, equipment noise and physiological noise on the subjects during the data acquisition process, so as to obtain training samples. Then, feature vectors are selected based on the training samples. When selecting features, multiple feature vectors are obtained by combining the functional connections between brain networks and graph theory indicators of brain networks, thereby improving the accuracy, sensitivity and specificity of the support vector machine classifier. Finally, the resting-state functional magnetic resonance imaging data to be classified are obtained as test samples, and the test samples are input into the support vector machine classifier for classification, so as to distinguish children with autism spectrum disorders from children with typical development. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 This is a flowchart of a specific implementation of the method for classifying children with autism based on brain network features provided by an embodiment of the present invention.

[0054] Figure 2 This is a diagram showing the classification effect of the support vector machine for the 18-24 month group provided by an embodiment of the present invention.

[0055] Figure 3 This is a diagram showing the classification effect of the support vector machine for the 25-48 month group provided by an embodiment of the present invention.

[0056] Figure 4 This is a principle block diagram of a device for classifying children with autism based on brain network features provided by an embodiment of the present invention.

[0057] Figure 5 This is a block diagram of the internal structure of the smart terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0059] In recent years, the prevalence of autism spectrum disorders (ASD) in young children has increased significantly, prompting widespread public attention to the causes, symptoms, diagnosis, and treatment of children with ASD. Therefore, identifying effective biomarkers and constructing classification models to distinguish ASD from typically developing children (TD) can help better understand the neural mechanisms underlying abnormal brain development in children with ASD. Resting-state functional magnetic resonance imaging (rs-fMRI), due to its non-invasive and high-resolution properties, is widely used in the medical field to detect various brain disorders. Therefore, using rs-fMRI to effectively model the functional brain networks of patients with autism would facilitate the detection of biomarkers and facilitate the study of neural activity in the brains of patients with autism. However, existing models typically use a single classification model based on functional connectivity between brain networks or graph-theoretic metrics of brain networks, resulting in low classification accuracy for support vector machine classifiers.

[0060] Therefore, to address the above-mentioned issues, this embodiment provides a method for classifying children with autism based on brain network features. This method can combine functional connectivity between brain networks and graph-theoretic indicators of brain networks during feature selection, thereby better revealing the neural mechanisms underlying abnormal brain development in children with autism spectrum disorders and improving the accuracy, sensitivity, and specificity of the support vector machine classifier. Specifically, this embodiment first obtains resting-state functional magnetic resonance imaging (fMRI) data of children of different age groups to be tested, performs data preprocessing, and obtains training samples. Feature vectors are then selected based on the training samples, and the feature vectors are trained to obtain a support vector machine classifier. Finally, the resting-state fMRI data to be classified is obtained as a test sample, and the test sample is input into the support vector machine classifier for classification. Thus, the classification results of the resting-state fMRI data to be classified are obtained.

[0061] Exemplary Methods

[0062] This embodiment provides a method for classifying autistic children based on brain network features. Figure 1 As shown, the method includes the following steps:

[0063] Step S10: Acquire resting-state functional magnetic resonance imaging data of children of different age groups to be tested, and perform data preprocessing to obtain training samples; wherein the preprocessing includes: format conversion, removal of data from the first three time points, inter-layer time correction, head motion correction, structural functional image registration, spatial normalization, spatial smoothing, bandpass filtering, and removal of physiological noise.

[0064] Resting-state functional MRI (rsfMRI) involves performing functional MRI scans on subjects without any conscious thought or explicit movement, perception, or cognitive processing. During the imaging process, sedatives can be injected into the child to induce and maintain anesthesia while MRI data are collected. This experimental simplicity makes rsfMRI data easy to collect and share, suitable for interspecies comparison, and ideal for biomarker development.

[0065] Specifically, this example uses 24 months of age as the cutoff and divides ASD and TD children into 18-24 month and 25-48 month groups for resting-state functional magnetic resonance imaging (r-fMRI) data acquisition. The acquired resting-state functional magnetic resonance (r-fMRI) data are preprocessed using AFNI (Analysis of Functional Neuro Images). This preprocessing includes format conversion (converting the r-fMRI data to AFNI format), removal of the first three time points, inter-slice time correction, motion correction, structural-functional image registration, spatial normalization, spatial smoothing, bandpass (0.01-0.1 Hz) filtering, and removal of physiological noise (average signal of cerebrospinal fluid and white matter). These steps minimize the effects of head movement, equipment noise, and physiological noise during data acquisition. The removal of the first three time points accounts for the uneven magnetic field during startup of the MRI machine. Inter-slice time correction is performed because MRI data are acquired by scanning alternate slices, which results in time differences between scanned slices. Therefore, interpolation is required for inter-slice correction. Furthermore, since subjects experience head movement during MRI experiments, which significantly impacts data quality, head movement correction based on a reference time point is necessary to eliminate the influence of sampling points with significant head movement. Finally, the individual functional imaging data were projected into standard space, resulting in training samples for the 18-24 month group and the 25-48 month group.

[0066] Step S20: Select feature vectors based on the training samples, and train the feature vectors to obtain a support vector machine classifier.

[0067] Support Vector Machines (SVMs) are a type of generalized linear classifier that performs binary classification using supervised learning. Their decision boundary is a maximum-margin hyperplane calculated from the learning examples. SVMs use a hinge loss function to calculate empirical risk and incorporate regularization terms into the solution to optimize structural risk. They are sparsity- and robust-based classifiers. SVMs can also perform nonlinear classification using kernel methods and are a common kernel learning method.

[0068] Specifically, the support vector machine classifier in this embodiment is a binary classifier for distinguishing ASD / TD infants based on individual feature vectors. The feature vectors of the training samples consisting of resting-state functional magnetic resonance imaging data of the 18-24 month group and the 25-48 month group are input respectively, and the feature vectors of the 18-24 month group and the 25-48 month group are trained respectively to obtain the support vector machine classifier for the 18-24 month group and the support vector machine classifier for the 25-48 month group.

[0069] In one implementation, step S20 in this embodiment includes the following steps:

[0070] Step S21, obtaining a resting-state brain network of each training sample, and obtaining a functional connectivity matrix of the resting-state brain network according to the resting-state brain network;

[0071] Step S22, obtaining brain network functional connections with significant differences based on the functional connectivity matrix of the resting-state brain network;

[0072] Step S23: obtaining the brain region corresponding to each of the resting-state brain networks, constructing a brain network map based on the brain regions, and obtaining a first graph theory index and a second graph theory index based on the brain network map;

[0073] Step S24: using the brain network functional connectivity with significant differences, the first graph theory index, and the second graph theory index as the feature vector, and using the fitcsvm function to train the feature vector to obtain the support vector machine classifier.

[0074] The resting-state brain network is composed of anatomically separated but functionally connected brain regions. There are multiple resting-state brain networks that can show high levels of sustained functional connectivity during rest. In addition to forming multiple resting-state networks, the human brain forms an integrated complex network that connects all brain regions and subnetworks into a complex system. Using graph theory analysis methods, the overall topological structure of the brain network can be examined, which can reveal local and global information about functional brain networks. In this embodiment, the use of graph theory indicators as feature vectors of training samples can more clearly reflect the neural mechanisms of abnormal brain development in ASD children.

[0075] Specifically, this embodiment obtains the functional connectivity matrix of the resting-state brain network through the acquired resting-state brain network, and obtains the functional connectivity of the brain network with significant differences by comparing the functional connectivity matrices. The brain functional connectivity with significant differences is used to reflect the degree of difference between the functional connectivity matrices of ASD children and TD children. The characteristic vector of the training sample is obtained by combining the first graph theory index and the second graph theory index obtained in the brain network graph theory. Finally, the characteristic vector is trained using the function fitcsvm provided by Matlab to obtain the classifier of the support vector machine for the 18-24 month and 25-48 month groups. fitcsvm supports the use of kernel functions to map predicted data, and supports sequential minimum optimization to minimize the objective function through quadratic programming, iterative single data algorithm, or L1 soft boundary minimization. fitcsvm supports three types of kernel functions: linear kernel "linear", polynomial kernel "polynomial", and Gaussian kernel "rbf" or "gaussian". After testing, the classification effect is best when the linear kernel "linear" is used, and then the leave one out cross-validation method is used to optimize the parameters of the training data. As Figure 2 、 Figure 3 As shown, the ROC curves of two support vector machines for the 18-24 month group and the 25-48 month group were drawn to evaluate the classification effect of the support vector machine.

[0076] In one implementation, step S21 in this embodiment includes the following steps:

[0077] Step S211, determining seven resting-state brain networks for the left and right cerebral hemispheres of each training sample respectively; wherein the seven resting-state brain networks include: default network, executive control network, salience network, dorsal attention network, auditory network, motor network, and visual network;

[0078] Step S212: Obtain the time series mean of each resting-state brain network of the children with autism spectrum disorder in the training sample to obtain fourteen first time series means;

[0079] Step S213, obtaining the time series mean of each resting-state brain network of the typically developing children in the training sample, and obtaining fourteen second time series means;

[0080] Step S214: Calculate the Pearson correlation coefficient between any two first time series means to obtain a first correlation coefficient;

[0081] Step S215: Calculate the Pearson correlation coefficient between any two second time series means to obtain a second correlation coefficient;

[0082] Step S216: obtaining a functional connectivity matrix of a first resting-state brain network according to the first correlation coefficient;

[0083] Step S217: Obtain a functional connectivity matrix of a second resting-state brain network based on the second correlation coefficient.

[0084] Specifically, seven resting-state brain networks were identified for each hemisphere, seven regions of interest were generated to represent the seven resting-state brain networks, and the mean of the time series (time series across all voxels) of these seven regions of interest was extracted. These seven brain networks are: the default mode network (DMN), the executive control network (ECN), the salience network (SN), the dorsal attention network (DAN), the auditory network (AN), the motor network (SMT), and the visual network (VN). Based on the ASD and TD groups, the Pearson correlation calculation was used to confirm the correlation coefficient between any two brain networks, thereby obtaining the functional connectivity matrix of the first resting-state brain network of the ASD children and the functional connectivity matrix of the second resting-state brain network of the TD children. The Pearson correlation coefficient (also known as the Pearson product-moment correlation coefficient (PPMCC or PCCs)) is used to measure the degree of correlation between two variables. The correlation coefficients of the ASD and TD groups were summarized separately to obtain the functional connectivity matrix of the resting-state brain networks of the ASD and TD groups.

[0085] In one implementation, step S22 in this embodiment includes the following steps:

[0086] Step S221: Perform a permutation test and multiple comparison correction on the functional connectivity matrix of the first resting-state brain network and the functional connectivity matrix of the second resting-state brain network to obtain the functional connectivity of the brain networks with significant differences.

[0087] Specifically, after using Pearson correlation to calculate the correlation coefficient between any two brain networks and obtaining the functional connectivity matrix of the resting-state brain network of each subject, permutation tests were performed on the functional connectivity matrices of ASD children and TD children in the 18-24 month group and the functional connectivity matrices of ASD children and TD children in the 25-48 month group, and multiple comparison corrections were performed to find the brain network functional connections with significant differences between ASD children and TD children in the two age groups.

[0088] In one implementation, step S23 in this embodiment includes the following steps:

[0089] Step S231, obtaining the brain region corresponding to each of the resting-state brain networks, and extracting the mean of the time series of each brain region as the brain network node corresponding to each brain region; wherein the mean of the time series of the brain region is the arithmetic mean of the time series of all voxels in the brain region;

[0090] Step S232: Calculate the time series mean x of any two brain regions i and j i , x j , Pearson correlation coefficient r between i, j∈{1,...,M} i，j ;

[0091] Step S233: According to the Pearson correlation coefficient r i，j , and obtain the functional connectivity matrix R of the brain region = (r i，j ) M×M×N , where M is the number of brain regions and N is the number of training samples;

[0092] Step S234: Binarize the functional connectivity matrix R of the brain region to obtain a binary brain network matrix W = (b i，j ) M×M×N ;

[0093] Step S235: obtaining a brain network map according to the brain network nodes and the binarized brain network matrix;

[0094] Step S236: Obtain the first graph theory index and the second graph theory index according to the brain network graph.

[0095] Resting-state brain networks are interconnected and connected across various brain regions and subregions, while the dynamic connections between these networks form a larger single network. Graph theory primarily studies the relationships between nodes and edges. Using graph theory, a functional brain network can be defined as a graph G = (V, E), where V is the set of nodes representing brain regions, and E is the functional connections (also called edges) between these regions, describing the interactions between nodes. In brain network graph analysis, various graph-theoretic metrics can be used to characterize resting-state brain network connectivity.

[0096] Specifically, in this embodiment, seven brain regions corresponding to the resting-state brain networks are selected, and the mean of the time series (the time series in all voxels) of each brain region is extracted as the brain network node of each brain region. i and j are used to represent the identifiers of the brain network nodes, and a brain network node set V consisting of M brain network nodes is obtained, where M is the number of brain regions. The correlation coefficient r between the average time series of two brain regions is calculated using Pearson correlation. i，j, as the edges of the brain network nodes, the functional connectivity matrix of the brain regions corresponding to the subject's resting-state brain network is obtained. The functional connectivity matrix is then binarized to obtain a binarized brain network matrix. In this embodiment, a brain network graph is obtained using the brain network nodes and the binarized brain network matrix, and then graph theory indicators of the brain network graph are obtained based on graph theory.

[0097] In one implementation, step S234 in this embodiment includes the following steps:

[0098] Step S2341: If the Pearson correlation coefficient r in the functional connectivity matrix of the brain region i，j If the connectivity coefficient b is greater than or equal to the preset first threshold γ, i，j = 1, there is a connection between brain regions i and j, that is, there is an edge between the corresponding brain network nodes;

[0099] Step S2342: If the Pearson correlation coefficient r in the functional connectivity matrix of the brain region i，j is less than a preset first threshold γ, then the connectivity coefficient b i，j = 0, there is no connection between brain regions i and j, that is, there is no edge between the corresponding brain network nodes;

[0100] Step S2343: Obtain the binary brain network matrix W based on the connectivity coefficient. i，j ) M×M×N .

[0101] Specifically, the GRETNA software was used to calculate graph theory indicators for the 18-24 month old ASD and TD children and the 25-48 month old ASD and TD children. First, the connectivity between two brain network nodes i and j was set: the first threshold γ was set by a certain sparsity cost, and the sparsity cost value was a percentage. For example, the correlation coefficient r i，j Sort by large to small, if Cost = 10%, then the correlation coefficient r of the first 10% is i，j is considered connected, the correlation coefficient r of the last 90% i，j is considered to be disconnected. After determining the first threshold γ, if the correlation coefficient r i，j If the connectivity coefficient b is greater than or equal to the preset first threshold γ, i，j = 1, there is a connection between brain regions i and j, that is, there is an edge between the corresponding brain network nodes; otherwise b i，j =0, thus obtaining the brain network R=(r i，j ) M×M×N The binary brain network W=(b i，j ) M×M×N, where N is the number of training samples. In this embodiment, the cost value range is set to [2%, 50%], with a preset step size of 1%, and only positive correlations are considered. To avoid selecting a single threshold, this embodiment applies the area under the curve (AUC) method, which is widely used in graph theory-based network research. For each graph theory metric, the AUC is calculated within the defined threshold range.

[0102] In one implementation, step S236 in this embodiment includes the following steps:

[0103] Step S2361: According to the brain network diagram, the first graph theory index is obtained as

[0104]

[0105] Where V is the set of all brain network nodes, is the clustering coefficient of brain network node i, K i is the number of connections between the brain region i and other brain regions; e i is the subnetwork graph G consisting of brain network node i and adjacent brain network nodes i The actual number of edges in , G is the brain network graph;

[0106] Step S2361: According to the brain network diagram, the second graph theory index is obtained as

[0107]

[0108] Where E(G i ) is the local efficiency of the brain network subgraph Gi composed of the brain network nodes adjacent to the brain network node i.

[0109] Specifically, the present invention selects and calculates two commonly used graph theory indicators: the network average clustering coefficient and the network average local efficiency, where the network average clustering coefficient is the first graph theory indicator and the network average local efficiency is the second graph theory indicator. The clustering coefficient is a coefficient that describes the degree of clustering between a node in the brain network and its connected nodes. The calculation formula of the network average clustering coefficient is: Local efficiency is the average of the local efficiencies of nodes in a node’s neighborhood. It is considered an indicator of the network’s fault tolerance when a node is removed. The calculation formula for the network’s average local efficiency is:

[0110] Step S300: Acquire resting-state functional magnetic resonance imaging data to be classified as a test sample, and input the test sample into the support vector machine classifier for classification to obtain a classification result.

[0111] Specifically, the resting-state functional MRI data to be classified were input, and the feature vectors of the resting-state functional MRI data to be classified were extracted in the same way as the feature vectors of the training samples; and the feature vectors were input into the support vector machine classifier to obtain the classification results. The classification accuracy of the 18-24 month group reached 88%, that is, 7 out of 8 samples were correctly identified by the classifier, and the classification accuracy of the 25-48 month group reached 83%, that is, 10 out of 12 samples were correctly identified by the classifier, see Table 1.

[0112] Table 1. Classification results of training samples

[0113]

[0114] Exemplary devices

[0115] like Figure 4 As shown in , this embodiment also provides a device for classifying children with autism based on brain network features, the device comprising:

[0116] The training sample acquisition module 10 is used to obtain resting-state functional magnetic resonance imaging data of children of different age groups to be tested, and perform data preprocessing to obtain training samples;

[0117] A support vector machine classifier acquisition module 20 is used to select feature vectors based on the training samples and train the feature vectors to obtain a support vector machine classifier;

[0118] The classification module 30 is configured to obtain resting-state functional magnetic resonance imaging data to be classified as a test sample, and input the test sample into the support vector machine classifier for classification to obtain a classification result.

[0119] In one implementation, the support vector machine classifier acquisition module 20 includes:

[0120] a functional connectivity matrix acquisition unit, configured to acquire a resting-state brain network of each training sample, and obtain a functional connectivity matrix of the resting-state brain network according to the resting-state brain network;

[0121] A brain network functional connectivity acquisition unit with significant differences, configured to obtain brain network functional connectivity with significant differences based on the functional connectivity matrix of the resting-state brain network;

[0122] A first graph theory indicator and a second graph theory indicator acquisition unit, configured to acquire a brain region corresponding to each of the resting-state brain networks, construct a brain network map based on the brain regions, and obtain the first graph theory indicator and the second graph theory indicator based on the brain network map;

[0123] A support vector machine classifier acquisition unit is used to use the brain network functional connections with significant differences, the first graph theory index and the second graph theory index as the feature vector, and use the fitcsvm function to train the feature vector to obtain the support vector machine classifier.

[0124] In one implementation, the functional connectivity matrix acquisition unit includes:

[0125] A resting-state brain network acquisition subunit is used to determine seven resting-state brain networks for the left and right cerebral hemispheres of each training sample; wherein the seven resting-state brain networks include: default network, executive control network, salience network, dorsal attention network, auditory network, motor network, and visual network;

[0126] A first time series mean acquisition subunit is used to acquire the time series mean of each resting-state brain network of the children with autism spectrum disorder in the training sample, and obtain fourteen first time series means;

[0127] A second time series mean acquisition subunit is used to acquire the time series mean of each resting-state brain network of typically developing children in the training sample, to obtain fourteen second time series means;

[0128] A first correlation coefficient obtaining subunit is used to calculate the Pearson correlation coefficient between any two first time series means to obtain a first correlation coefficient;

[0129] A second correlation coefficient obtaining subunit is used to calculate the Pearson correlation coefficient between any two second time series means to obtain a second correlation coefficient;

[0130] a functional connectivity matrix acquisition subunit of the first resting-state brain network, configured to obtain the functional connectivity matrix of the first resting-state brain network according to the first correlation coefficient;

[0131] a functional connectivity matrix acquisition subunit of the second resting-state brain network, configured to obtain the functional connectivity matrix of the second resting-state brain network according to the second correlation coefficient;

[0132] In one implementation, the unit for acquiring brain network functional connections with significant differences includes:

[0133] a subunit for obtaining functional connectivity of brain networks with significant differences, configured to perform a permutation test and multiple comparison correction on the functional connectivity matrix of the first resting-state brain network and the functional connectivity matrix of the second resting-state brain network to obtain the functional connectivity of the brain networks with significant differences;

[0134] In one implementation, the unit for acquiring brain network functional connections with significant differences includes:

[0135] a brain network node acquisition subunit, configured to acquire the brain region corresponding to each of the resting-state brain networks, and extract the mean of the time series of each brain region as the brain network node corresponding to each brain region; wherein the mean of the time series of the brain region is the arithmetic mean of the time series of all voxels in the brain region;

[0136] Pearson correlation coefficient acquisition subunit, used to calculate the time series mean x of any two brain regions i and j i , x j , Pearson correlation coefficient r between i, j∈{1,...,M} i，j ;

[0137] The functional connectivity matrix is obtained for subunits according to the Pearson correlation coefficient r i，j , and obtain the functional connectivity matrix R of the brain region = (r i，j ) M×M×N , where M is the number of brain regions and N is the number of training samples;

[0138] The binary brain network matrix acquisition subunit is used to perform binary processing on the functional connection matrix R of the brain region to obtain a binary brain network matrix W=(b i，j ) M×M×N ;

[0139] a brain network map acquisition subunit, configured to obtain a brain network map according to the brain network nodes and the binarized brain network matrix;

[0140] The first graph theory indicator and the second graph theory indicator obtaining subunit are configured to obtain the first graph theory indicator and the second graph theory indicator according to the brain network graph.

[0141] Based on the above embodiment, the present invention also provides an intelligent terminal, whose principle block diagram can be shown as follows: Figure 5 As shown. The smart terminal includes a processor, a memory, a network interface, a display screen, and a temperature sensor connected via a system bus. The processor of the smart terminal is used to provide computing and control capabilities. The memory of the smart terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the smart terminal is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for classifying autistic children based on brain network characteristics is implemented. The display screen of the smart terminal can be a liquid crystal display screen or an electronic ink display screen, and the temperature sensor of the smart terminal is pre-set inside the smart terminal to detect the operating temperature of the internal device.

[0142] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, operating database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0143] In summary, the present invention discloses a method and device for classifying autistic children based on brain network features. The method comprises: obtaining resting-state functional magnetic resonance imaging data of children to be tested in different age groups, and performing data preprocessing to obtain training samples; selecting feature vectors based on the training samples, and training the feature vectors to obtain a support vector machine classifier; obtaining the resting-state functional magnetic resonance imaging data to be classified as test samples, and inputting the test samples into the support vector machine classifier for classification to obtain classification results. The method of the present invention combines the functional connectivity between brain networks and the graph theory indicators of brain networks during feature selection, which can better reveal the neural mechanism of abnormal brain development in children with autism spectrum disorders and improve the accuracy, sensitivity and specificity of the support vector machine classifier.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A device for classifying children with autism based on brain network characteristics, characterized in that: The device comprises: The training sample acquisition module is used to obtain resting-state functional magnetic resonance imaging data of children of different age groups to be tested, and perform data preprocessing to obtain training samples; A support vector machine classifier acquisition module is used to select feature vectors based on the training samples and train the feature vectors to obtain a support vector machine classifier; The support vector machine classifier acquisition module includes: a functional connectivity matrix acquisition unit, configured to acquire a resting-state brain network of each training sample, and obtain a functional connectivity matrix of the resting-state brain network according to the resting-state brain network; A brain network functional connectivity acquisition unit with significant differences is used to obtain brain network functional connectivity with significant differences based on the functional connectivity matrix of the resting-state brain network; A first graph theory indicator and a second graph theory indicator acquisition unit, configured to acquire a brain region corresponding to each of the resting-state brain networks, construct a brain network map based on the brain regions, and obtain the first graph theory indicator and the second graph theory indicator based on the brain network map; a support vector machine classifier acquisition unit, configured to use the significantly different brain network functional connectivity, the first graph theory index, and the second graph theory index as the feature vector, and train the feature vector using a fitcsvm function to obtain the support vector machine classifier; The functional connectivity matrix acquisition unit includes: a resting-state brain network acquisition subunit, configured to determine seven resting-state brain networks for the left and right cerebral hemispheres of each training sample; a first time series mean acquisition subunit, configured to acquire the time series mean of each resting-state brain network of the children with autism spectrum disorder in the training sample, to obtain fourteen first time series means; A second time series mean acquisition subunit is used to acquire the time series mean of each resting-state brain network of typically developing children in the training sample, to obtain fourteen second time series means; A first correlation coefficient obtaining subunit is used to calculate the Pearson correlation coefficient between any two first time series means to obtain a first correlation coefficient; A second correlation coefficient obtaining subunit is used to calculate the Pearson correlation coefficient between any two second time series means to obtain a second correlation coefficient; a functional connectivity matrix acquisition subunit of the first resting-state brain network, configured to obtain the functional connectivity matrix of the first resting-state brain network according to the first correlation coefficient; a functional connectivity matrix acquisition subunit of the second resting-state brain network, configured to obtain the functional connectivity matrix of the second resting-state brain network according to the second correlation coefficient; The brain network functional connection acquisition unit with significant differences includes: a subunit for obtaining functional connectivity of brain networks with significant differences, configured to perform a permutation test and multiple comparison correction on the functional connectivity matrix of the first resting-state brain network and the functional connectivity matrix of the second resting-state brain network to obtain the functional connectivity of the brain networks with significant differences; The classification module is used to obtain resting state functional magnetic resonance imaging data to be classified as a test sample, and input the test sample into the support vector machine classifier for classification to obtain a classification result.

2. An intelligent terminal, characterized in that: The intelligent terminal includes a memory, a processor, and an autism child classification program based on brain network features stored in the memory and executable on the processor. When the processor executes the autism child classification program based on brain network features, the steps of the autism child classification method based on brain network features are implemented. The steps of the autism child classification method based on brain network features include: Acquiring resting-state functional magnetic resonance imaging data from children of different age groups and performing data preprocessing to obtain training samples; wherein the preprocessing includes: format conversion, removal of data from the first three time points, inter-slice time correction, head motion correction, structural-functional image registration, spatial normalization, spatial smoothing, bandpass filtering, and removal of physiological noise; Selecting feature vectors based on the training samples, and training the feature vectors to obtain a support vector machine classifier; Acquiring resting-state functional magnetic resonance imaging data to be classified as a test sample, and inputting the test sample into the support vector machine classifier for classification to obtain a classification result; The selecting feature vectors based on the training samples and training the feature vectors to obtain a support vector machine classifier includes: Acquiring a resting-state brain network of each training sample, and obtaining a functional connectivity matrix of the resting-state brain network according to the resting-state brain network; Obtaining brain network functional connections with significant differences based on the functional connection matrix of the resting-state brain network; Obtaining a brain region corresponding to each of the resting-state brain networks, constructing a brain network map based on the brain regions, and obtaining a first graph theory index and a second graph theory index based on the brain network map; Using the significantly different brain network functional connectivity, the first graph theory index, and the second graph theory index as the feature vector, and using the fitcsvm function to train the feature vector to obtain the support vector machine classifier; The step of obtaining a resting-state brain network of each training sample and obtaining a functional connectivity matrix of the resting-state brain network according to the resting-state brain network comprises: Determining seven resting-state brain networks for the left and right cerebral hemispheres of each training sample; obtaining the time series mean of each resting-state brain network of the children with autism spectrum disorder in the training sample to obtain fourteen first time series means; Obtaining the time series mean of each resting-state brain network of typically developing children in the training sample to obtain fourteen second time series means; Calculate the Pearson correlation coefficient between any two first time series means to obtain the first correlation coefficient; Calculate the Pearson correlation coefficient between any two second time series means to obtain the second correlation coefficient; Obtaining a functional connectivity matrix of a first resting-state brain network according to the first correlation coefficient; According to the second correlation coefficient, a functional connectivity matrix of the second resting-state brain network is obtained; Obtaining significantly different brain network functional connections based on the functional connectivity matrix of the resting-state brain network includes: The functional connectivity matrix of the first resting-state brain network and the functional connectivity matrix of the second resting-state brain network are subjected to permutation test and multiple comparison correction to obtain the functional connectivity of the brain networks with significant differences.

3. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program for classifying children with autism based on brain network features. When the program is executed by the processor, the steps of the method for classifying children with autism based on brain network features are implemented. The steps of the method for classifying children with autism based on brain network features include: Acquiring resting-state functional magnetic resonance imaging data from children of different age groups and performing data preprocessing to obtain training samples; wherein the preprocessing includes: format conversion, removal of data from the first three time points, inter-slice time correction, head motion correction, structural-functional image registration, spatial normalization, spatial smoothing, bandpass filtering, and removal of physiological noise; Selecting feature vectors based on the training samples, and training the feature vectors to obtain a support vector machine classifier; Acquiring resting-state functional magnetic resonance imaging data to be classified as a test sample, and inputting the test sample into the support vector machine classifier for classification to obtain a classification result; The selecting feature vectors based on the training samples and training the feature vectors to obtain a support vector machine classifier includes: Acquiring a resting-state brain network of each training sample, and obtaining a functional connectivity matrix of the resting-state brain network according to the resting-state brain network; Obtaining brain network functional connections with significant differences based on the functional connection matrix of the resting-state brain network; Obtaining a brain region corresponding to each of the resting-state brain networks, constructing a brain network map based on the brain regions, and obtaining a first graph theory index and a second graph theory index based on the brain network map; Using the significantly different brain network functional connectivity, the first graph theory index, and the second graph theory index as the feature vector, and using the fitcsvm function to train the feature vector to obtain the support vector machine classifier; The step of obtaining a resting-state brain network of each training sample and obtaining a functional connectivity matrix of the resting-state brain network according to the resting-state brain network comprises: Determine seven resting-state brain networks for the left and right cerebral hemispheres of each training sample; Obtaining the time series mean of each resting-state brain network of the children with autism spectrum disorder in the training sample to obtain fourteen first time series means; Obtaining the time series mean of each resting-state brain network of typically developing children in the training sample to obtain fourteen second time series means; Calculate the Pearson correlation coefficient between any two first time series means to obtain the first correlation coefficient; Calculate the Pearson correlation coefficient between any two second time series means to obtain the second correlation coefficient; Obtaining a functional connectivity matrix of a first resting-state brain network according to the first correlation coefficient; According to the second correlation coefficient, a functional connectivity matrix of the second resting-state brain network is obtained; Obtaining significantly different brain network functional connections based on the functional connectivity matrix of the resting-state brain network includes: The functional connectivity matrix of the first resting-state brain network and the functional connectivity matrix of the second resting-state brain network are subjected to permutation test and multiple comparison correction to obtain the functional connectivity of the brain networks with significant differences.

Images