Early identification method of autism spectrum disorder based on artificial intelligence

Abstract

The present invention belongs to the field of autism identification, and specifically refers to an early identification method for autism spectrum disorder based on artificial intelligence, which includes collecting rs-fMRI data, partitioning and screening, optimizing neighbor distribution and autism identification. This scheme creatively proposes a functional connection network feature extraction method combined with brain region of interest screening, first partitioning, then screening, eliminating brain regions of interest with poor correlation with autism spectrum disorder, reducing computational complexity and computing cost while improving the training efficiency of the recognition model; a graph embedding recognition method combining graph neural network and K nearest neighbor is proposed, and the separation effect within and between the same type is enhanced through triple loss, and K nearest neighbor is used to identify whether the patient to be identified has the risk of autism, making full use of the prior expert knowledge data set to improve the recognition accuracy.

Description

Technical Field

[0001] The present invention relates to the field of autism identification, and specifically refers to an early identification method for autism spectrum disorders based on artificial intelligence. Background Art

[0002] Autism spectrum disorder is a neurodevelopmental disorder characterized by persistent repetitive behavior patterns and social interaction deficits. The AI-based early identification method for autism spectrum disorder refers to the use of deep learning technology to identify potential autistic patients, effectively reducing the economic burden of early screening for patients.

[0003] Among the existing approximate solutions, for example, CN118096776A, a method and device for identifying autism brain images based on heterogeneous graph isomorphic networks, addresses the technical problem that graph neural networks have black boxes and are difficult to explain classification results in a neuroscience-interpretable way. It uses heterogeneous graphs based on human brain structure and pre-selected brain regions of interest to achieve the technical effect of interpretable brain imaging for autism spectrum disorders using graph neural networks. However, there is a problem that brain regions of interest are not screened, and the directly preset regions of interest have a weak correlation with autism spectrum disorders.

[0004] Traditional classification algorithms can realize autism identification. CN109620259A is a system for automatically identifying autistic children based on eye movement technology and machine learning. This solution aims at the problem that traditional diagnostic methods rely heavily on the subjective judgment of doctors, the diagnostic process is long, the process is complicated and is limited by the developmental stage of the children. The K nearest neighbor classifier is used to classify predefined eye movement features. The eye movement features only include the mean, median, standard deviation, skewness, slope, interquartile range, minimum coordinate value, maximum coordinate value and eye movement sampling number, and the technical effect of directly diagnosing autism based on biological testing methods is achieved. However, there are technical problems such as large scale and large number of brain functional connection features, difficulty in predefining potential connection rules, and difficulty in directly transplanting the K nearest neighbor algorithm from feature analysis based on eye movement to feature analysis based on brain functional magnetic resonance imaging.

[0005] Therefore, the traditional autism identification model has technical problems that the brain regions of interest are not screened, and the directly preset regions of interest have a weak correlation with autism spectrum disorders; there are technical problems that the brain functional connection features are large in scale and large in number, making it difficult to predefine potential connection rules, and it is difficult to directly transplant the K-nearest neighbor algorithm from feature analysis based on eye movements to feature analysis based on brain functional magnetic resonance imaging. Summary of the invention

[0006] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an early identification method of autism spectrum disorder based on artificial intelligence;

[0007] In view of the technical problems that the traditional autism recognition model has unscreened brain regions of interest and the directly preset regions of interest have weak correlation with autism spectrum disorders, this solution creatively proposes a functional connection network feature extraction method combined with brain region of interest screening, first partitioning, then screening, and eliminating brain regions of interest (ROI, Region of Interest) with poor correlation with autism spectrum disorders, so as to achieve accurate extraction of functional connection networks of autistic patients, reduce computational complexity and computing cost, and improve the training efficiency of the recognition model; In view of the technical problems that the traditional autism recognition model has large scale and large number of brain functional connection features, it is difficult to predefine potential connection rules, and it is difficult to directly transplant the K nearest neighbor algorithm from feature analysis based on eye movement to feature analysis based on brain functional magnetic resonance imaging, this solution creatively proposes a graph embedding recognition method combining graph neural network and K nearest neighbor, enhances the separation effect within the same type and between types through triple loss, uses K nearest neighbor to identify whether the patient to be identified has the risk of autism, makes full use of prior expert knowledge data sets, and improves recognition accuracy;

[0008] The technical solution adopted by the present invention is as follows: The present invention provides an artificial intelligence-based early identification method for autism spectrum disorder, the method comprising the following steps:

[0009] Step S1: Acquire rs-fMRI data and construct a magnetic resonance imaging dataset;

[0010] The rs-fMRI data refers to brain activity data collected when the subject is not performing a specific task. The rs-fMRI data is a 4D object consisting of 3D brain images taken at n time points;

[0011] The magnetic resonance imaging data set includes rs-fMRI data of subjects with autism spectrum disorder and rs-fMRI data of healthy subjects, wherein the subjects with autism spectrum disorder are recorded as the first category and the healthy subjects are recorded as the second category;

[0012] Step S2: Partition screening, used to divide and screen brain regions of interest to obtain functional connectivity submaps;

[0013] Step S3: Optimize the neighbor distribution, extract graph embeddings from the functional connectivity subgraph, and use a graph neural network to perform distance optimization on all corresponding graph embeddings in the magnetic resonance imaging dataset to obtain K neighbor subjects;

[0014] Step S4: Loneliness identification. Among neighbor subjects, if the number of the first category is greater than or equal to the number of the second category, the patient to be identified has the risk of autism spectrum disorder. If the number of the first category is less than the number of the second category, the patient to be identified does not have the risk of autism spectrum disorder.

[0015] Furthermore, in step S2, the partition screening specifically includes the following steps:

[0016] Step S21: partitioning brain regions of interest, specifically, partitioning a group of brain regions of interest related to autism spectrum disorders;

[0017] Step S22: characterizing the functional connectivity between brain regions of interest, specifically, extracting rs-fMRI data of each brain region of interest from the magnetic resonance imaging data set, and performing covariance analysis to obtain a sample covariance matrix;

[0018] Step S23: generating an inverse covariance matrix, specifically, randomly initializing two Wishart distributions with different degree of freedom parameters, using a Bayesian hierarchical model, randomly sampling from one Wishart distribution to generate an inverse covariance matrix of the first category equal to the total number of subjects of the first category, and randomly sampling from another Wishart distribution to generate an inverse covariance matrix of the second category equal to the total number of subjects of the second category;

[0019] Step S24: Calculate the likelihood, specifically, by integrating the distribution of the inverse covariance matrix, and calculating the likelihood function of the inverse covariance matrix of the first type, the formula used is as follows:

[0020] ;

[0021] In the formula, represents the likelihood function of the inverse covariance matrix of the first kind, represents the inverse covariance matrix of the first kind, represents the rs-fMRI data of the i-th subject belonging to the first class, represents the proportionality coefficient, represents the number of subjects in the first category, represents the degrees of freedom parameter of the Wishart distribution of the first type, represents continuous product, represents the number of sampling time points for the i-th subject, represents the sample covariance matrix of the i-th subject in the first category;

[0022] Step S25: binary classification synchronization, specifically, repeating step S24, taking the same operation on the inverse covariance matrix of the second category, and obtaining the likelihood function of the inverse covariance matrix of the second category;

[0023] Step S26: Screening ROIs for adjusting the inverse covariance matrix and generating an indicator matrix, specifically, constructing an objective function, and finding the optimal values ​​of the first type of inverse covariance matrix, the second type of inverse covariance matrix and the indicator matrix by maximizing the objective function to obtain the indicator matrix, wherein the indicator matrix indicates whether any brain region of interest is selected, and the formula used is as follows:

[0024] ;

[0025] In the formula, When the objective function takes the maximum value, the solution is and and The best value of represents the inverse covariance matrix of the second kind, Indicator matrix, indicates the transpose of the matrix, represents the likelihood function of the inverse covariance matrix of the second kind, , , and They represent the adjustment parameters of the weights, and denote the L1 norm of the first and second categories respectively;

[0026] Step S27: Generate a functional connectivity subgraph, specifically, according to the indicator matrix, screen brain regions of interest related to autism spectrum disorder from the inverse covariance matrices of the first category and the second category to obtain a functional connectivity subgraph.

[0027] Furthermore, in step S3, the optimization of the nearest neighbor distribution specifically includes the following steps:

[0028] Step S31: connection mapping, used to capture the global features of the functional connection subgraph, specifically, using a graph neural network to map the functional connection subgraph into a node embedding vector in the Euclidean space, and merging the node embedding vector into a graph embedding;

[0029] Step S32: Adjust the graph embedding distance to increase the distance between graph embeddings corresponding to subjects of different categories and reduce the distance between graph embeddings corresponding to subjects of the same category. Specifically, randomly select a subject from the magnetic resonance imaging dataset, use the graph embedding corresponding to the subject as the anchor graph, use subjects with autism spectrum disorder as positive samples, that is, use the first category as positive samples, use healthy subjects as negative samples, that is, use the second category as negative samples, use triplet loss to train the graph neural network, and calculate the triplet loss of the entire magnetic resonance imaging dataset. The formula used is as follows:

[0030] ;

[0031] In the formula, represents the sum of all triplet losses, represents a triple consisting of an anchor map, a positive sample, and a negative sample. Indicates the maximum value operation. represents the mapping function, represents the anchor graph in the triple, represents the graph embedding corresponding to the positive sample, represents the graph embedding corresponding to the negative sample, represents the preset margin hyperparameter, represents the square of the Euclidean distance;

[0032] The constraints followed during training are as follows:

[0033] ;

[0034] Step S33: Collect neighbor subjects. Specifically, based on the rs-fMRI data of the patient to be identified, a K nearest neighbor algorithm is used to calculate the distance between the image embedding of the patient to be identified and the image embedding of each subject in the magnetic resonance imaging dataset, and the K neighbor subjects closest to the patient to be identified are collected.

[0035] The present invention provides an early identification method for autism spectrum disorder based on artificial intelligence. The beneficial effects achieved by the present invention using the above scheme are as follows:

[0036] (1) In view of the technical problem that the traditional autism recognition model has unscreened brain regions of interest and the directly preset regions of interest have weak correlation with autism spectrum disorders, this scheme creatively proposes a functional connection network feature extraction method combined with brain region of interest screening, which first partitions and then screens to eliminate brain regions of interest (ROI) with poor correlation with autism spectrum disorders, thereby achieving accurate extraction of the functional connection network of autistic patients, reducing computational complexity and computing cost while improving the training efficiency of the recognition model;

[0037] (2) In view of the technical problems that traditional autism recognition models have, such as large scale and large number of brain functional connection features, difficulty in predefining potential connection rules, and difficulty in directly transplanting the K nearest neighbor algorithm from feature analysis based on eye movement to feature analysis based on brain functional magnetic resonance imaging, this scheme creatively proposes a graph embedding recognition method that combines graph neural network and K nearest neighbor. The triplet loss is used to enhance the separation effect within the same type and between types. K nearest neighbor is used to identify whether the patient to be identified has the risk of autism, making full use of the prior expert knowledge data set to improve the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of the process of the early identification method of autism spectrum disorder based on artificial intelligence provided by the present invention.

[0039] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0041] Example 1, see Figure 1 The present invention provides an artificial intelligence-based early identification method for autism spectrum disorders, the method comprising the following steps:

[0042] Step S1: Acquire rs-fMRI data and construct a magnetic resonance imaging dataset;

[0043] The rs-fMRI data refers to brain activity data collected when the subject is not performing a specific task. The rs-fMRI data is a 4D object consisting of 3D brain images taken at n time points;

[0044] The magnetic resonance imaging data set includes rs-fMRI data of subjects with autism spectrum disorder and rs-fMRI data of healthy subjects, wherein the subjects with autism spectrum disorder are recorded as the first category and the healthy subjects are recorded as the second category;

[0045] Step S2: Partition screening, used to divide and screen brain regions of interest to obtain functional connectivity submaps;

[0046] Step S3: Optimize the neighbor distribution, extract graph embeddings from the functional connectivity subgraph, and use a graph neural network to perform distance optimization on all corresponding graph embeddings in the magnetic resonance imaging dataset to obtain K neighbor subjects;

[0047] Step S4: Loneliness identification. Among neighbor subjects, if the number of the first category is greater than or equal to the number of the second category, the patient to be identified has the risk of autism spectrum disorder. If the number of the first category is less than the number of the second category, the patient to be identified does not have the risk of autism spectrum disorder.

[0048] Example 2, see Figure 1This embodiment is based on the above embodiment. In step S2, the partition screening specifically includes the following steps:

[0049] Step S21: partitioning brain regions of interest, specifically, partitioning a group of brain regions of interest related to autism spectrum disorders;

[0050] Step S22: characterizing the functional connectivity between brain regions of interest, specifically, extracting rs-fMRI data of each brain region of interest from the magnetic resonance imaging data set, and performing covariance analysis to obtain a sample covariance matrix;

[0051] Step S23: generating an inverse covariance matrix, specifically, randomly initializing two Wishart distributions with different degree of freedom parameters, using a Bayesian hierarchical model, randomly sampling from one Wishart distribution to generate an inverse covariance matrix of the first category equal to the total number of subjects of the first category, and randomly sampling from another Wishart distribution to generate an inverse covariance matrix of the second category equal to the total number of subjects of the second category;

[0052] Step S24: Calculate the likelihood, specifically, by integrating the distribution of the inverse covariance matrix, and calculating the likelihood function of the inverse covariance matrix of the first type, the formula used is as follows:

[0053] ;

[0054] In the formula, represents the likelihood function of the inverse covariance matrix of the first kind, represents the inverse covariance matrix of the first kind, represents the rs-fMRI data of the i-th subject belonging to the first class, represents the proportionality coefficient, represents the number of subjects in the first category, represents the degrees of freedom parameter of the Wishart distribution of the first type, represents continuous product, represents the number of sampling time points for the i-th subject, represents the sample covariance matrix of the i-th subject in the first category;

[0055] Step S25: binary classification synchronization, specifically, repeating step S24, taking the same operation on the inverse covariance matrix of the second category, and obtaining the likelihood function of the inverse covariance matrix of the second category;

[0056] Step S26: Screening ROIs for adjusting the inverse covariance matrix and generating an indicator matrix, specifically, constructing an objective function, and finding the optimal values ​​of the first type of inverse covariance matrix, the second type of inverse covariance matrix and the indicator matrix by maximizing the objective function to obtain the indicator matrix, wherein the indicator matrix indicates whether any brain region of interest is selected, and the formula used is as follows:

[0057] ;

[0058] In the formula, When the objective function takes the maximum value, the solution is and and The best value of represents the inverse covariance matrix of the second kind, Indicator matrix, indicates the transpose of the matrix, represents the likelihood function of the inverse covariance matrix of the second kind, , , and They represent the adjustment parameters of the weights, and denote the L1 norm of the first and second categories respectively;

[0059] Step S27: Generate a functional connectivity subgraph, specifically, according to the indicator matrix, screen brain regions of interest related to autism spectrum disorder from the inverse covariance matrices of the first category and the second category to obtain a functional connectivity subgraph.

[0060] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S3, the optimization of the neighbor distribution specifically includes the following steps:

[0061] Step S31: connection mapping, used to capture the global features of the functional connection subgraph, specifically, using a graph neural network to map the functional connection subgraph into a node embedding vector in the Euclidean space, and merging the node embedding vector into a graph embedding;

[0062] Step S32: Adjust the graph embedding distance to increase the distance between graph embeddings corresponding to subjects of different categories and reduce the distance between graph embeddings corresponding to subjects of the same category. Specifically, randomly select a subject from the magnetic resonance imaging dataset, use the graph embedding corresponding to the subject as the anchor graph, use subjects with autism spectrum disorder as positive samples, that is, use the first category as positive samples, use healthy subjects as negative samples, that is, use the second category as negative samples, use triplet loss to train the graph neural network, and calculate the triplet loss of the entire magnetic resonance imaging dataset. The formula used is as follows:

[0063] ;

[0064] In the formula, represents the sum of all triplet losses, represents a triple consisting of an anchor map, a positive sample, and a negative sample. Indicates the maximum value operation. represents the mapping function, represents the anchor graph in the triple, represents the graph embedding corresponding to the positive sample, represents the graph embedding corresponding to the negative sample, represents the preset margin hyperparameter, represents the square of the Euclidean distance;

[0065] The constraints followed during training are as follows:

[0066] ;

[0067] Step S33: Collect neighbor subjects. Specifically, based on the rs-fMRI data of the patient to be identified, a K nearest neighbor algorithm is used to calculate the distance between the image embedding of the patient to be identified and the image embedding of each subject in the magnetic resonance imaging dataset, and the K neighbor subjects closest to the patient to be identified are collected.

[0068] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S21, the brain regions of interest are divided into anterior insula (left side), anterior insula (right side), anterior cingulate cortex (left side), anterior cingulate cortex (right side), middle cingulate cortex (left side), middle cingulate cortex (right side), posterior insula (left side), posterior insula (right side), posterior cingulate cortex (left side), posterior cingulate cortex (right side), thalamus (left side), thalamus (right side), primary somatosensory cortex (left side), primary somatosensory cortex (right side), dorsolateral Prefrontal cortex (left), dorsolateral prefrontal cortex (right), inferolateral parietal lobe (left), inferolateral parietal lobe (right), ventromedial prefrontal cortex (left), ventromedial prefrontal cortex (right), secondary somatosensory cortex (left), secondary somatosensory cortex (right), supplementary motor area (left), supplementary motor area (right), temporal pole (left), temporal pole (right), amygdala (left), amygdala (right), middle temporal gyrus (right), caudate nucleus (left), caudate nucleus (right), periaqueductal gray.

[0069] Example 5, see Figure 1This embodiment is based on the above embodiment. In step S27, the functional connection subgraph is obtained, specifically, on the basis of dividing a group of brain regions of interest in step S21, further screening brain regions of interest according to the indicator matrix, screening brain regions of interest related to autism spectrum disorder from the inverse covariance matrix, and obtaining the functional connection subgraph;

[0070] The indicator matrix is ​​a The diagonal matrix of is the number of regions of interest, and the indicator matrix diagonal consists of 0s and 1s, which are used to indicate whether the corresponding brain region of interest is included in the functional connectivity submap.

[0071] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0072] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0073] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An artificial intelligence-based method for early identification of autism spectrum disorders, characterized by: The method comprises the following steps: Step S1: Acquire rs-fMRI data and construct a magnetic resonance imaging dataset; The rs-fMRI data refers to brain activity data collected when the subject is not performing a specific task. The rs-fMRI data is a 4D object consisting of 3D brain images taken at n time points; The magnetic resonance imaging data set includes rs-fMRI data of subjects with autism spectrum disorder and rs-fMRI data of healthy subjects, wherein the subjects with autism spectrum disorder are recorded as the first category and the healthy subjects are recorded as the second category; Step S2: Partition screening, used to divide and screen brain regions of interest to obtain functional connectivity submaps; Step S3: Optimize the neighbor distribution, extract graph embeddings from the functional connectivity subgraph, and use a graph neural network to perform distance optimization on all corresponding graph embeddings in the magnetic resonance imaging dataset to obtain K neighbor subjects; Step S4: loneliness identification, among the neighbor subjects, if the number of the first category is greater than or equal to the number of the second category, the patient to be identified has a risk of suffering from autism spectrum disorder; if the number of the first category is less than the number of the second category, the patient to be identified does not have a risk of suffering from autism spectrum disorder; In step S3, the optimization of the nearest neighbor distribution specifically includes the following steps: Step S31: connection mapping, used to capture the global features of the functional connection subgraph, specifically, using a graph neural network to map the functional connection subgraph into a node embedding vector in the Euclidean space, and merging the node embedding vector into a graph embedding; Step S32: Adjust the graph embedding distance to increase the distance between graph embeddings corresponding to subjects of different categories and reduce the distance between graph embeddings corresponding to subjects of the same category. Specifically, randomly select a subject from the magnetic resonance imaging dataset, use the graph embedding corresponding to the subject as the anchor graph, use subjects with autism spectrum disorder as positive samples, that is, use the first category as positive samples, use healthy subjects as negative samples, that is, use the second category as negative samples, use triplet loss to train the graph neural network, and calculate the triplet loss of the entire magnetic resonance imaging dataset. The formula used is as follows: ； In the formula, represents the sum of all triplet losses, represents a triple consisting of an anchor map, a positive sample, and a negative sample. Indicates the maximum value operation. represents the mapping function, represents the anchor graph in the triple, represents the graph embedding corresponding to the positive sample, represents the graph embedding corresponding to the negative sample, represents the preset margin hyperparameter, represents the square of the Euclidean distance; The constraints followed during training are as follows: ； Step S33: Collect neighbor subjects. Specifically, based on the rs-fMRI data of the patient to be identified, a K nearest neighbor algorithm is used to calculate the distance between the image embedding of the patient to be identified and the image embedding of each subject in the magnetic resonance imaging dataset, and the K neighbor subjects closest to the patient to be identified are collected.

2. The method for early identification of autism spectrum disorders based on artificial intelligence according to claim 1, characterized in that: In step S2, the partition screening specifically includes the following steps: Step S21: partitioning brain regions of interest, specifically, partitioning a group of brain regions of interest related to autism spectrum disorders; Step S22: characterizing the functional connectivity between brain regions of interest, specifically, extracting rs-fMRI data of each brain region of interest from the magnetic resonance imaging data set, and performing covariance analysis to obtain a sample covariance matrix; Step S23: generating an inverse covariance matrix, specifically, randomly initializing two Wishart distributions with different degree of freedom parameters, using a Bayesian hierarchical model, randomly sampling from one Wishart distribution to generate an inverse covariance matrix of the first category equal to the total number of subjects of the first category, and randomly sampling from another Wishart distribution to generate an inverse covariance matrix of the second category equal to the total number of subjects of the second category; Step S24: Calculate the likelihood, specifically, by integrating the distribution of the inverse covariance matrix, and calculating the likelihood function of the inverse covariance matrix of the first type, the formula used is as follows: ； In the formula, represents the likelihood function of the inverse covariance matrix of the first kind, represents the inverse covariance matrix of the first kind, represents the rs-fMRI data of the i-th subject belonging to the first class, represents the proportionality coefficient, represents the number of subjects in the first category, represents the degrees of freedom parameter of the Wishart distribution of the first type, represents the continuous product, represents the number of sampling time points for the i-th subject, represents the sample covariance matrix of the i-th subject in the first category; Step S25: binary classification synchronization, specifically, repeating step S24, taking the same operation on the inverse covariance matrix of the second category, and obtaining the likelihood function of the inverse covariance matrix of the second category; Step S26: Screening ROIs for adjusting the inverse covariance matrix and generating an indicator matrix, specifically, constructing an objective function, and finding the optimal values ​​of the first type of inverse covariance matrix, the second type of inverse covariance matrix and the indicator matrix by maximizing the objective function to obtain the indicator matrix, wherein the indicator matrix indicates whether any brain region of interest is selected, and the formula used is as follows: ； In the formula, When the objective function takes the maximum value, the solution is and and The best value of represents the inverse covariance matrix of the second kind, Indicator matrix, indicates the transpose of the matrix, represents the likelihood function of the inverse covariance matrix of the second kind, , , and They represent the adjustment parameters of the weights, and denote the L1 norm of the first and second categories respectively; Step S27: Generate a functional connectivity subgraph, specifically, according to the indicator matrix, screen brain regions of interest related to autism spectrum disorder from the inverse covariance matrices of the first category and the second category to obtain a functional connectivity subgraph.

Images