Autism biomarker identification method based on functional magnetic resonance imaging

Abstract

The application relates to an autism biomarker identification method based on functional magnetic resonance imaging, which fully utilizes the topological structure and functional connection characteristics of the brain, performs autism fMRI identification through a graph propagation network, and then utilizes an attention guiding module to locate significant brain areas and significant functional connections, identifies autism biomarkers, provides interpretability for a model, and provides reference significance for analysis of an autism lesion site; in addition, two-stage statistical tests are used to verify the reliability of the biomarkers.

Description

Technical Field

[0001] This application relates to the field of biometric identification technology, specifically to a method for identifying autism biomarkers based on functional magnetic resonance imaging. Background Technology

[0002] In recent years, an increasing number of researchers have used functional magnetic resonance imaging (fMRI) to diagnose autism. Other medical methods, such as CT scans, ultrasound, and X-rays, can cause radiation damage to the brain. EEG has a high signal-to-noise ratio but low spatial resolution. In contrast, fMRI's low-radiation and non-invasive nature has led to its rapid rise in importance in the diagnosis of autism and other brain functions. With the rapid development of computer-aided diagnosis and artificial intelligence, the use of machine learning and deep learning methods to analyze medical data offers significant possibilities for autism diagnosis. Graph structures, which satisfy the topological characteristics of the brain, can be used to calculate non-Euclidean spatial distances and are more suitable for analyzing brain data. Compared to traditional convolutional neural networks, which treat brain features as one-dimensional vectors, graph propagation networks treat brain regions as nodes and functional connections between brain regions as edges, providing endless possibilities for exploring brain networks. Currently, diagnosis of mental illness based on graph propagation networks has become a hot topic in the field of artificial intelligence and has achieved significant success. However, the shortcomings are that most existing graph propagation networks used for disease diagnosis are based on brain node-driven approaches and do not focus on utilizing the characteristics of functional connectivity. Furthermore, after identifying autism patients, autism biomarkers are roughly derived through node aggregation or backpropagation without further interpretation and verification, which fails to provide interpretability for the model. Summary of the Invention

[0003] To overcome at least one deficiency in the prior art, this application provides a method for identifying autism biomarkers based on functional magnetic resonance imaging.

[0004] Firstly, a method for identifying autism biomarkers based on functional magnetic resonance imaging is provided, including:

[0005] Acquire functional magnetic resonance imaging (fMRI) images of the brains of multiple patients to be identified;

[0006] Based on the brain atlas, the brain regions in the functional magnetic resonance imaging (fMRI) images are divided, and the correlation between any two brain regions in the fMRI images is determined to obtain the functional connectivity matrix.

[0007] A graph propagation network is constructed, which includes an edge-driven update layer. Each brain region is used as a graph node of the graph propagation network, and the functional connectivity matrix is ​​used as the edge of the graph propagation network. The node features of each graph node and the edge features of each edge are updated based on the edge-driven update layer of the graph propagation network, so as to obtain the updated node features of each graph node and the updated edge features of each edge.

[0008] Based on the updated node features, the patient to be identified is determined to be either a person with autism or a normal person.

[0009] Based on the updated node and edge features corresponding to all autism patients, biomarkers were identified, including significant functional connectivity and significant brain regions.

[0010] In one embodiment, the correlation between any two brain regions in a functional magnetic resonance imaging (fMRI) image is determined to obtain a functional connectivity matrix, including:

[0011] Align functional magnetic resonance imaging (fMRI) images with brain atlases to delineate various brain regions in the fMRI images;

[0012] Extract signals from various brain regions in functional magnetic resonance imaging;

[0013] The correlation between signals from any two brain regions in functional magnetic resonance imaging (fMRI) images is calculated to form a functional connectivity matrix.

[0014] In one embodiment, the edge-driven update layer based on the graph propagation network updates the node features of each graph node and the edge features of each edge to obtain the updated node features of each graph node and the updated edge features of each edge, including:

[0015] The edge-driven update layer comprises multiple intermediate layers; each intermediate layer updates the node features of each graph node and the edge features of each edge, including:

[0016] For the kth intermediate layer, the graph propagation coefficient matrix of the kth intermediate layer is applied to the updated functional connectivity matrix obtained from the (k-1)th intermediate layer to perform a dot product operation, and the dot product result is the updated functional connectivity matrix of the kth intermediate layer. The updated functional connectivity matrix includes the updated edge features of each edge.

[0017] The dot product result is multiplied by a tensor with the updated node feature vector obtained from the (k-1)th intermediate layer to obtain the updated node feature vector of the kth intermediate layer. The updated node feature vector includes the updated node features of each graph node.

[0018] In one embodiment, determining whether the patient to be identified is a person with autism or a normal person based on the updated node features includes:

[0019] Graph propagation networks also include linear layers and sigmoid layers;

[0020] The updated node features from the last intermediate layer are input into a linear layer for aggregation to obtain brain activation values.

[0021] Brain activation values ​​are input into the Sigmoid layer to determine whether the patient to be identified is autistic or normal.

[0022] In one embodiment, biomarkers are determined based on the updated node features and updated edge features corresponding to all autism patients, including:

[0023] For each intermediate layer, the contribution of each node feature to the classification task is calculated based on the node feature graph formed by the updated node features of each graph node, and the contribution of each edge feature to the classification task is calculated based on the edge feature graph formed by the updated edge features of each edge.

[0024] Based on the contribution of each node feature corresponding to all intermediate layers to the classification task, the brain region activation map of each graph node is obtained; based on the contribution of each edge feature corresponding to all intermediate layers to the classification task, the functional connectivity activation map of each edge is obtained.

[0025] Calculate the mean of the brain region activation maps corresponding to all autistic patients to obtain the average brain region activation map; the average brain region activation map includes the average brain region activation value corresponding to each graph node.

[0026] Find the mean of the functional connectivity activation graphs for all autistic patients, and the average functional connectivity activation graph; the average functional connectivity activation graph includes the average functional connectivity activation value for each edge.

[0027] At least one significant brain region is determined based on the average brain region activation value corresponding to each graph node, and at least one significant functional connection is determined based on the average functional connection activation value corresponding to each edge. At least one significant brain region and at least one significant functional connection constitute a biomarker.

[0028] In one embodiment, acquiring multiple functional magnetic resonance imaging (fMRI) images of the brains of patients to be identified includes:

[0029] The raw functional magnetic resonance imaging (fMRI) images of the brain of the patient to be identified were preprocessed to obtain the functional fMRI images of the brain of the patient to be identified. The preprocessing included removing unstable time points, temporal layer correction, head movement correction, registration, and spatial smoothing.

[0030] In one embodiment, the method further includes:

[0031] The biomarkers were validated using a two-sample t-test. If the p-value was <0.05 / N, the biomarker was considered valid. Biomarker If so, the biomarker is reliable, N Biomarker The number of biomarkers.

[0032] Secondly, a device for identifying autism biomarkers based on functional magnetic resonance imaging is provided, comprising:

[0033] Functional magnetic resonance imaging acquisition module, used to acquire multiple functional magnetic resonance images of the brains of patients to be identified;

[0034] The functional connectivity matrix acquisition module is used to divide the brain regions in functional magnetic resonance imaging (fMRI) images according to the brain atlas, determine the correlation between any two brain regions in the fMRI images, and obtain the functional connectivity matrix.

[0035] The graph propagation network construction module is used to construct a graph propagation network. The graph propagation network includes an edge-driven update layer, which uses each brain region as a graph node and the functional connectivity matrix as an edge. Based on the edge-driven update layer of the graph propagation network, the node features of each graph node and the edge features of each edge are updated to obtain the updated node features of each graph node and the updated edge features of each edge.

[0036] The patient identification module is used to determine whether the patient to be identified is a person with autism or a normal person based on the updated node features.

[0037] The biomarker identification module is used to identify biomarkers based on the updated node features and updated edge features corresponding to all autism patients. Biomarkers include significant functional connectivity and significant brain regions.

[0038] Thirdly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the aforementioned method for identifying autism biomarkers based on functional magnetic resonance imaging.

[0039] Fourthly, a computer program product is provided, including a computer program / instructions, which, when executed by a processor, implement the aforementioned method for identifying autism biomarkers based on functional magnetic resonance imaging.

[0040] Compared with the prior art, this application has the following beneficial effects: This application makes full use of the brain's topological structure and functional connectivity characteristics, identifies autism fMRI through graph propagation networks, and then uses an attention-guided module to locate significant brain regions and significant functional connections, thereby identifying autism biomarkers, thus providing interpretability for the identification model and providing reference for the analysis of autism lesion sites; and verifies the credibility of biomarkers through two-stage statistical tests. Attached Figure Description

[0041] This application can be better understood by referring to the description given below in conjunction with the accompanying drawings, which, together with the detailed description below, are incorporated in and form part of this specification. In the drawings:

[0042] Figure 1A flowchart of a method for identifying autism biomarkers based on functional magnetic resonance imaging according to an embodiment of this application is shown;

[0043] Figure 2 A schematic diagram illustrating the determination of a functional connection matrix according to an embodiment of this application is shown;

[0044] Figure 3 A schematic diagram of the structure of a graph propagation network according to an embodiment of this application is shown;

[0045] Figure 4 A schematic diagram of the process for determining biomarkers according to an embodiment of this application is shown;

[0046] Figure 5 A schematic diagram illustrating the validation of biomarkers according to an embodiment of this application is shown;

[0047] Figure 6 This is a visual representation of biomarkers according to embodiments of this application;

[0048] Figure 7 A structural block diagram of an autism biomarker identification device based on functional magnetic resonance imaging according to an embodiment of this application is shown. Detailed Implementation

[0049] Exemplary embodiments of the present application will be described below with reference to the accompanying drawings. For clarity and brevity, not all features of the actual embodiments are described in the specification. However, it should be understood that many embodiment-specific decisions can be made in the development of any such actual embodiment to achieve the developer’s specific objectives, and these decisions may vary as the embodiments differ.

[0050] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the device structure closely related to the solution according to this application is shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0051] It should be understood that this application is not limited to the described embodiments by virtue of the following description with reference to the accompanying drawings. In this document, embodiments may be combined with each other, features may be substituted or borrowed between different embodiments, and one or more features may be omitted in one embodiment, where feasible.

[0052] This application provides a method for identifying autism biomarkers based on functional magnetic resonance imaging. Figure 1 A flowchart illustrating a method for identifying autism biomarkers based on functional magnetic resonance imaging according to an embodiment of this application is shown. See also... Figure 1 The methods include:

[0053] Step S1: Acquire functional magnetic resonance imaging (fMRI) images of the brains of multiple patients to be identified;

[0054] Specifically, in this step, the raw functional magnetic resonance imaging (fMRI) images of the brain of the patient to be identified are acquired, and preprocessed using DPABI_V4.3_200401 and SPM12. DPABI and SPM are both Matlab toolboxes for processing fMRI images. The preprocessing includes removing unstable time points, temporal layer correction, head motion correction, registration, and spatial smoothing.

[0055] Here, removing unstable time points involves deleting the first four time points to eliminate the influence of unstable factors on the experiment; temporal correction involves performing temporal regression processing to eliminate interference signals and ensure that the acquisition time of all brain slices is theoretically consistent; head movement correction requires similar overlap through affine transformations to minimize the influence of head movements; registration involves matching the brains of all individuals to the same brain space to address the differences in brain morphology among different subjects and the inconsistency in spatial position during scanning; spatial smoothing involves smoothing all voxel acquisitions using a Gaussian filter to reduce spatial noise in the fMRI data and performing mean processing to increase the likelihood of its normal distribution.

[0056] Step S2: Divide the brain regions in the functional magnetic resonance imaging (fMRI) images according to the brain atlas, determine the correlation between any two brain regions in the fMRI images, and obtain the functional connectivity matrix.

[0057] Step S3: Construct a graph propagation network. The graph propagation network includes an edge-driven update layer. Each brain region is used as a graph node of the graph propagation network, and the functional connectivity matrix is ​​used as the edge of the graph propagation network. The node features of each graph node and the edge features of each edge are updated based on the edge-driven update layer of the graph propagation network to obtain the updated node features of each graph node and the updated edge features of each edge.

[0058] Step S4: Based on the updated node features, determine whether the patient to be identified is a person with autism or a normal person.

[0059] Step S5: Based on the updated node features and updated edge features corresponding to all autism patients, biomarkers are identified, including significant functional connectivity and significant brain regions.

[0060] In this embodiment, the topological structure and functional connectivity characteristics of the brain are fully utilized to identify autism fMRI through graph propagation networks. Then, attention-guided modules are used to locate significant brain regions and significant functional connectivity, thereby identifying autism biomarkers. This provides interpretability for the model and offers reference for the analysis of autism lesion sites.

[0061] In one embodiment, Figure 2 A schematic diagram illustrating the determination of the functional connection matrix according to an embodiment of this application is shown. See also: Figure 2 In step S2, the correlation between any two brain regions in the functional magnetic resonance imaging (fMRI) image is determined to obtain the functional connectivity matrix, which includes:

[0062] Step S21: Align the functional magnetic resonance imaging (fMRI) images with the brain atlas and delineate the various brain regions in the fMRI images; here, the brain atlas can be the AAL atlas, Harvard Oxford atlas, CC200 atlas, CC400 atlas, etc.

[0063] Step S22: Extract signals from each brain region in the functional magnetic resonance imaging; here, the signal of each brain region is extracted, specifically by extracting the average time series of each brain region.

[0064] Step S23: Calculate the correlation between signals from any two brain regions in the functional magnetic resonance imaging (fMRI) image to form the functional connectivity matrix FC. Here, the Pearson correlation coefficient can be used to calculate the correlation coefficient between brain regions, thus forming the functional connectivity matrix FC.

[0065] In one embodiment, Figure 3 A schematic diagram of the structure of a graph propagation network according to an embodiment of this application is shown. See also: Figure 3 In step S3, the node features of each graph node and the edge features of each edge are updated based on the graph propagation network to obtain the updated node features of each graph node and the updated edge features of each edge, including:

[0066] The edge-driven update layer consists of multiple intermediate layers, each used for one update. Taking the update process of the kth intermediate layer as an example, the update process is as follows:

[0067] First, the graph propagation coefficient matrix of the kth intermediate layer is... The updated functional connectivity matrix obtained by applying the action on the (k-1)th intermediate layer Perform a dot product operation on the above to obtain the dot product result. That is, the updated functional connection matrix of the kth intermediate layer, which includes the updated edge features of each edge;

[0068] Then multiply the result The updated node feature vector obtained from the (k-1)th intermediate layer Perform tensor multiplication to obtain the updated node feature vector of the kth intermediate layer. The updated node feature vector includes the updated node features of each graph node.

[0069] The calculation formula is as follows:

[0070]

[0071] In this embodiment, there can be m intermediate layers. The initial value of the node feature vector of each graph node is 1. The correlation coefficient in the functional connectivity matrix is ​​used as the connection strength value of the edge and as the input of the graph propagation network. The edge-driven update layer includes multiple intermediate layers. The edge features extracted by each intermediate layer constitute an n×n two-dimensional edge feature vector. The node features extracted by each graph node constitute an n×1 two-dimensional node feature vector, where n represents the number of brain regions. One brain region is one graph node. The edge feature vector represents the updated result of the functional connectivity features between two brain regions. The node feature vector represents the activation status of brain regions.

[0072] In one embodiment, step S4, determining whether the patient to be identified is an autistic patient or a normal person based on the updated node features, includes:

[0073] Graph propagation networks also include linear layers and sigmoid layers;

[0074] The updated node features from the last intermediate layer are input into a linear layer for aggregation to obtain brain activation values.

[0075] Brain activation values ​​are input into the Sigmoid layer to determine whether the patient to be identified is autistic or normal.

[0076] In one embodiment, Figure 4 A schematic flowchart illustrating the determination of biomarkers according to an embodiment of this application is shown. See also: Figure 4 In step S5, biomarkers are determined based on the node and edge features corresponding to all autism patients, including:

[0077] Step S51: For each intermediate layer k, where k is the label of the intermediate layer, k = 1, 2, ..., m, the updated node features of each graph node constitute a two-dimensional node feature map, and the updated edge features of each edge constitute a two-dimensional edge feature map. The node feature maps and edge feature maps of the m intermediate layers are concatenated to obtain a three-dimensional edge feature map and a three-dimensional node feature map. Then, the contribution of each node feature to the classification task is calculated. This represents the contribution of each edge feature of the k-th intermediate layer to the classification task. The contribution of each node feature in the k-th intermediate layer to the classification task is represented by the following formula:

[0078]

[0079]

[0080] Among them, yc For the output of the linear layer, F s (k,i,j) represents the edge feature at coordinate (i,j) in the edge feature map corresponding to the kth intermediate layer, F p (k,i,1) represents the node feature of the i-th graph node in the node feature graph corresponding to the k-th intermediate layer.

[0081] Step S52: Based on the contribution of each node feature in all intermediate layers to the classification task. Obtain the brain region activation map H for each graph node. s Brain region activation map H s This includes the activation values ​​of brain regions corresponding to each graph node; and the contribution of each edge feature corresponding to all intermediate layers to the classification task. Obtain the functional connection activation graph H of each edge. p Functional connection activation diagram H p It includes the activation values ​​of the functional connections corresponding to each edge.

[0082]

[0083]

[0084] Where ReLU is the ReLU activation function.

[0085] Step S53: Calculate the brain region activation map H corresponding to all autistic patients. s The mean value is used to obtain the average brain region activation map. Average brain region activation map It includes the average brain region activation value corresponding to each graph node.

[0086] Step S54: Calculate the functional connectivity activation graph H corresponding to all autistic patients. p The mean, average functional connection activation graph Average Functional Connection Activation Chart It includes the average functional connection activation value corresponding to each edge.

[0087] Step S55: Determine at least one significant brain region based on the average brain region activation value corresponding to each graph node, and determine at least one significant functional connection based on the average functional connection activation value corresponding to each edge. The at least one significant brain region and the at least one significant functional connection constitute a biomarker.

[0088] In this step, 505 autistic patients were included. Brain ROI maps were compared to determine which specific brain region in the map corresponded to each individual brain region. By setting thresholds, these activation maps were suppressed. All brain regions corresponding to the graph nodes that met the threshold requirements were selected as salient brain regions from the average activation values ​​of each graph node. Similarly, all functional connections corresponding to the edges that met the threshold requirements were selected as salient functional connections from the average functional connectivity activation values ​​of each edge.

[0089] In this embodiment, based on the Grad_CAM attention guidance module, the feature maps of each intermediate layer can be spliced ​​together to form an edge feature map and a node feature map. Then, the contribution of the node features of each graph node and the contribution of the edge features of each edge can be calculated.

[0090] In one embodiment, Figure 5 A schematic diagram illustrating the validation of biomarkers according to an embodiment of this application is shown. See also... Figure 5 The methods also include:

[0091] The biomarkers were validated using a two-sample t-test. The p-value was calculated; if the p-value < 0.05 / N, the biomarker was considered valid. Biomarker If so, the biomarker is reliable, N Biomarker The number of biomarkers.

[0092] In this embodiment, P-values ​​are calculated for significant brain regions and significant functional connectivity in the biomarkers. Specifically, this includes selecting each significant brain region of 20 autism patients and the corresponding brain regions of normal individuals to construct significant brain region matrices for autism patients and significant brain region matrices for normal individuals, respectively. The statistical two-sample t-test method is used for verification, and the P1 value is calculated. Here, a P1 value is calculated for each significant brain region.

[0093] Each significant functional connection of 20 autistic patients and the corresponding functional connection of normal individuals were selected to construct significant functional connection matrices for autistic patients and normal individuals, respectively. The statistical two-sample t-test method was used to verify the matrices and calculate the p-value. Here, a p-value was calculated for each significant functional connection.

[0094] If all P1 values ​​are less than 0.05 / N Biomarker And all P2 values ​​< 0.05 / N Biomarker Here, N Biomarker The number of significant brain regions and significant functional connections indicates the reliability of the biomarker. Figure 6 This is a visual representation of biomarkers according to an embodiment of this application.

[0095] To further demonstrate the accuracy of the autism patient identification results of this application, Table 1 shows the comparison results of autism patient identification using the method of this application and existing classification methods. According to Table 1, the identification accuracy of the method of this application can reach 70.7%, which is better than other methods. The better classification effect provides a premise for the localization of biomarkers.

[0096] Table 1 Comparative experimental results of different methods

[0097]

[0098] Employing the same inventive concept as the autism biomarker identification method based on functional magnetic resonance imaging (fMRI), this embodiment also provides a corresponding autism biomarker identification device based on fMRI. Figure 7 A structural block diagram of an autism biomarker identification device based on functional magnetic resonance imaging according to an embodiment of this application is shown, including:

[0099] Functional magnetic resonance imaging acquisition module 71 is used to acquire multiple functional magnetic resonance images of the brains of patients to be identified.

[0100] The functional connectivity matrix acquisition module 72 is used to divide the brain regions in the functional magnetic resonance imaging (fMRI) images according to the brain atlas, determine the correlation between any two brain regions in the fMRI images, and obtain the functional connectivity matrix.

[0101] Graph propagation network construction module 73 is used to construct a graph propagation network. The graph propagation network includes an edge-driven update layer, which uses each brain region as a graph node of the graph propagation network and a functional connectivity matrix as an edge of the graph propagation network. The edge-driven update layer of the graph propagation network updates the node features of each graph node and the edge features of each edge to obtain the updated node features of each graph node and the updated edge features of each edge.

[0102] The patient identification module 74 is used to determine whether the patient to be identified is an autistic patient or a normal person based on the updated node features;

[0103] The biomarker identification module 75 is used to identify biomarkers based on the updated node features and updated edge features corresponding to all autism patients. The biomarkers include significant functional connectivity and significant brain regions.

[0104] In other embodiments, the specific implementation functions of each module of the autism biomarker identification device based on functional magnetic resonance imaging are consistent with the specific implementation steps of the autism biomarker identification method based on functional magnetic resonance imaging in the foregoing embodiments, and will not be described in detail again.

[0105] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the above-described method for identifying autism biomarkers based on functional magnetic resonance imaging.

[0106] This application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the above-described method for identifying autism biomarkers based on functional magnetic resonance imaging.

[0107] The above descriptions are merely various embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying autism biomarkers based on functional magnetic resonance imaging, characterized in that, include: Acquire functional magnetic resonance imaging (fMRI) images of the brains of multiple patients to be identified; Based on the brain atlas, the brain regions in the functional magnetic resonance imaging are divided, and the correlation between any two brain regions in the functional magnetic resonance imaging is determined to obtain the functional connectivity matrix. A graph propagation network is constructed, which includes an edge-driven update layer. Each brain region is used as a graph node of the graph propagation network, and the functional connectivity matrix is ​​used as an edge of the graph propagation network. The node features of each graph node and the edge features of each edge are updated based on the edge-driven update layer of the graph propagation network to obtain the updated node features of each graph node and the updated edge features of each edge. Based on the updated node features, the patient to be identified is determined to be either a person with autism or a normal person. Based on the updated node features and updated edge features corresponding to all autism patients, biomarkers are identified, including significant functional connectivity and significant brain regions; Determining the correlation between any two brain regions in the functional magnetic resonance imaging (fMRI) images yields a functional connectivity matrix, including: Align the functional magnetic resonance imaging (fMRI) images with the brain atlas to delineate the various brain regions in the fMRI images; Extract signals from each brain region in the functional magnetic resonance imaging; Calculate the correlation between signals from any two brain regions in the functional magnetic resonance imaging to construct the functional connectivity matrix; The edge-driven update layer of the graph propagation network updates the node features of each graph node and the edge features of each edge, resulting in the updated node features of each graph node and the updated edge features of each edge, including: The edge-driven update layer includes multiple intermediate layers; each intermediate layer is used to update the node features of each graph node and the edge features of each edge, including: For the kth intermediate layer, the graph propagation coefficient matrix of the kth intermediate layer is applied to the updated functional connectivity matrix obtained from the (k-1)th intermediate layer to perform a dot product operation, and the dot product result is the updated functional connectivity matrix of the kth intermediate layer. The updated functional connectivity matrix includes the updated edge features of each edge. The dot product result is multiplied by a tensor with the updated node feature vector obtained from the (k-1)th intermediate layer to obtain the updated node feature vector of the kth intermediate layer. The updated node feature vector includes the updated node features of each graph node. Based on the updated node features and updated edge features corresponding to all autism patients, biomarkers are identified, including: For each intermediate layer, the contribution of each node feature to the classification task is calculated based on the node feature graph formed by the updated node features of each graph node, and the contribution of each edge feature to the classification task is calculated based on the edge feature graph formed by the updated edge features of each edge. Based on the contribution of each node feature corresponding to all intermediate layers to the classification task, the brain region activation map of each graph node is obtained; based on the contribution of each edge feature corresponding to all intermediate layers to the classification task, the functional connectivity activation map of each edge is obtained. Calculate the mean of the brain region activation maps corresponding to all autistic patients to obtain the average brain region activation map; the average brain region activation map includes the average brain region activation value corresponding to each graph node. Find the mean of the functional connectivity activation graphs for all autistic patients, and the average functional connectivity activation graph; the average functional connectivity activation graph includes the average functional connectivity activation value for each edge. At least one significant brain region is determined based on the average brain region activation value corresponding to each graph node, and at least one significant functional connection is determined based on the average functional connection activation value corresponding to each edge. The at least one significant brain region and the at least one significant functional connection constitute the biomarker.

2. The method as described in claim 1, characterized in that, in, Based on the updated node features, determining whether the patient to be identified is autistic or normal includes: The graph propagation network also includes linear layers and sigmoid layers; The updated node features of the last intermediate layer are input into the linear layer for aggregation to obtain brain activation values; The brain activation values ​​are input into the Sigmoid layer to determine whether the patient to be identified is an autistic patient or a normal person.

3. The method as described in claim 1, characterized in that, in, The acquisition of multiple functional magnetic resonance imaging (fMRI) images of the brains of patients to be identified includes: The original functional magnetic resonance imaging (fMRI) images of the brain of the patient to be identified are preprocessed to obtain the functional fMRI images of the brain of the patient to be identified; the preprocessing includes removing unstable time points, temporal layer correction, head movement correction, registration, and spatial smoothing.

4. The method as described in claim 1, characterized in that, The method further includes: The biomarkers were validated using a two-sample t-test. If the p-value < 0.05 / N, the biomarker was considered valid. Biomarker If the biomarker is reliable, then N Biomarker The number of the biomarkers.

5. A device for identifying autism biomarkers based on functional magnetic resonance imaging, characterized in that, include: Functional magnetic resonance imaging acquisition module, used to acquire multiple functional magnetic resonance images of the brains of patients to be identified; The functional connectivity matrix acquisition module is used to divide the brain regions in the functional magnetic resonance imaging (fMRI) images according to the brain atlas, determine the correlation between any two brain regions in the functional fMRI images, and obtain the functional connectivity matrix. The graph propagation network construction module is used to construct a graph propagation network. The graph propagation network includes an edge-driven update layer. Each brain region is used as a graph node of the graph propagation network, and the functional connection matrix is ​​used as an edge of the graph propagation network. The node features of each graph node and the edge features of each edge are updated based on the edge-driven update layer of the graph propagation network to obtain the updated node features of each graph node and the updated edge features of each edge. The patient identification module is used to determine whether the patient to be identified is an autistic patient or a normal person based on the updated node features. A biomarker determination module is used to determine biomarkers based on updated node features and updated edge features corresponding to all autism patients. The biomarkers include significant functional connectivity and significant brain regions. Determining the correlation between any two brain regions in the functional magnetic resonance imaging (fMRI) images yields a functional connectivity matrix, including: Align the functional magnetic resonance imaging (fMRI) images with the brain atlas to delineate the various brain regions in the fMRI images; Extract signals from each brain region in the functional magnetic resonance imaging; Calculate the correlation between signals from any two brain regions in the functional magnetic resonance imaging to construct the functional connectivity matrix; The edge-driven update layer of the graph propagation network updates the node features of each graph node and the edge features of each edge, resulting in the updated node features of each graph node and the updated edge features of each edge, including: The edge-driven update layer includes multiple intermediate layers; each intermediate layer is used to update the node features of each graph node and the edge features of each edge, including: For the kth intermediate layer, the graph propagation coefficient matrix of the kth intermediate layer is applied to the updated functional connectivity matrix obtained from the (k-1)th intermediate layer to perform a dot product operation, and the dot product result is the updated functional connectivity matrix of the kth intermediate layer. The updated functional connectivity matrix includes the updated edge features of each edge. The dot product result is multiplied by a tensor with the updated node feature vector obtained from the (k-1)th intermediate layer to obtain the updated node feature vector of the kth intermediate layer. The updated node feature vector includes the updated node features of each graph node. Based on the updated node features and updated edge features corresponding to all autism patients, biomarkers are identified, including: For each intermediate layer, the contribution of each node feature to the classification task is calculated based on the node feature graph formed by the updated node features of each graph node, and the contribution of each edge feature to the classification task is calculated based on the edge feature graph formed by the updated edge features of each edge. Based on the contribution of each node feature corresponding to all intermediate layers to the classification task, the brain region activation map of each graph node is obtained; based on the contribution of each edge feature corresponding to all intermediate layers to the classification task, the functional connectivity activation map of each edge is obtained. Calculate the mean of the brain region activation maps corresponding to all autistic patients to obtain the average brain region activation map; the average brain region activation map includes the average brain region activation value corresponding to each graph node. Find the mean of the functional connectivity activation graphs for all autistic patients, and the average functional connectivity activation graph; the average functional connectivity activation graph includes the average functional connectivity activation value for each edge. At least one significant brain region is determined based on the average brain region activation value corresponding to each graph node, and at least one significant functional connection is determined based on the average functional connection activation value corresponding to each edge. The at least one significant brain region and the at least one significant functional connection constitute the biomarker.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the autism biomarker identification method based on functional magnetic resonance imaging as described in any one of claims 1-4.

7. A computer program product, characterized in that, Includes a computer program / instruction, which, when executed by a processor, implements the autism biomarker identification method based on functional magnetic resonance imaging as described in any one of claims 1-4.

Images