Autism spectrum disorder subtype division method and device, medium and program product

Abstract

The embodiment of the invention discloses an autism spectrum disorder subtype division method and device, a medium and a program product. The method comprises the following steps: constructing a connection brain map based on neuroimaging data and a functional brain region division template of an ASD individual; constructing a brain age regression model, predicting the social brain age of the ASD individual based on the connection brain map and the brain age regression model, and obtaining the brain age difference of the ASD individual in combination with the actual brain age of the ASD individual; obtaining an ADOS social score of the ASD individual, and carrying out clustering analysis on the ADOS social score and the brain age difference to obtain a clustering subtype; and carrying out behavioral verification and neural dimension verification on the clustering subtypes, and constructing a combined portrait among the subtypes, the behavior features and the neural features based on a verification result. According to the method, the social brain age can be predicted by constructing the connection brain map, the clustering subtypes are divided, the combined portrait is constructed, a doctor can be accurately assisted to detect ASD subtype neural development differences, and discovery and application of subtype specific biomarkers are assisted.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a method, apparatus, medium and program product for classifying autism spectrum disorder subtypes. Background Technology

[0002] Autism spectrum disorder (ASD) is a complex neurodevelopmental disorder characterized by social impairments, communication difficulties, and repetitive, stereotyped behaviors. In the DSM-5 diagnostic criteria, ASD is no longer subdivided into multiple subtypes but is treated as a single, unified diagnostic category. However, clinical observations and studies have revealed significant heterogeneity among ASD patients in terms of symptom presentation, cognitive abilities, behavioral characteristics, and responses to treatment. This heterogeneity suggests the possible existence of different subtypes, which may be associated with specific neurobiological mechanisms, genetic background, or environmental factors. Therefore, subtyping ASD is crucial for understanding disease mechanisms, guiding clinical diagnosis, and developing personalized intervention strategies.

[0003] However, existing classification methods mainly rely on behavioral observation and clinical assessment, which may be influenced by subjective judgment and are difficult to capture the neurobiological heterogeneity of ASD. In addition, most graph neural network (GNN) based methods focus on characterizing whole-brain structural features when processing neuroimaging data, but fail to predict and analyze functional brain age for specific ASD subtypes. This makes it difficult to fully assist doctors in detecting differences in neural developmental pathways between ASD subtypes, thus limiting the discovery and application of subtype-specific biomarkers. Summary of the Invention

[0004] In view of this, the present disclosure provides a method, device, medium and program product for classifying autism spectrum disorder subtypes, which can predict social brain age, classify cluster subtypes and construct joint profiles by constructing connectivity brain maps, accurately assist doctors in detecting neurodevelopmental differences in ASD subtypes and help discover and apply subtype-specific biomarkers.

[0005] In a first aspect, embodiments of this disclosure provide a method for classifying autism spectrum disorder subtypes, employing the following technical solution: Obtain neuroimaging data and functional brain region division templates for individuals with ASD. Construct a connectivity brain map based on the neuroimaging data and the functional brain region division templates. Herein, ASD refers to autism spectrum disorder. Nodes in the connectivity brain map represent brain regions, edges in the connectivity brain map represent functional connections between two brain regions, and the weight of the edge represents the strength of the functional connection. A brain age regression model is constructed, and based on the connectivity brain map and the brain age regression model, the social brain age of individuals with ASD is predicted. Based on the actual brain age and the social brain age of individuals with ASD, the brain age difference of individuals with ASD is obtained; Obtain the ADOS social score of individuals with ASD, and perform cluster analysis on the ADOS social score and the brain age difference to obtain cluster subtypes; wherein, ADOS refers to the Autism Diagnosis Observation Scale. Behavioral and neural dimension verifications are performed on the cluster subtypes. Based on the verification results, a joint profile of the subtypes, behavioral features, and neural features is constructed.

[0006] Optionally, constructing a connectivity brain map based on the neuroimaging data and the functional brain region segmentation template includes: The neuroimaging data is standardized and preprocessed to obtain standard images; According to the functional brain region division template, the standard image is divided into multiple brain regions; Extract the BOLD time-series signal of the brain region, and construct the connection matrix between brain regions based on the BOLD time-series signal; The connection matrix is ​​transformed into a graph-structured connection mind map.

[0007] Optionally, the brain age regression model includes an input layer, a multi-layer combination layer, a multi-scale brain region feature splicing layer, and a multi-layer perceptron regression layer; each of the multi-layer combination layers includes a graph convolutional layer, a social brain region enhancement mechanism layer, and a pooling layer. The input layer is used to provide the connectivity brain map for the multi-layer combined layer; Each combined layer is used to update the nodes in the connection mind map and output a new connection mind map; The graph convolutional layer is used to extract brain region features from the connected brain map; The social brain region enhancement mechanism layer is used to add a social brain region regularization term to the loss function based on the brain region features, form a new loss function and obtain the total loss, and optimize the weight parameters of the pooling layer based on the total loss. The pooling layer obtains the importance score of the node based on the weight parameter, updates the node in the connection brain map based on the importance score, and outputs a new connection brain map; The multi-scale brain region feature splicing layer is used to splice the connectivity brain maps output by each combined layer into a unified connectivity brain map. The multilayer perceptron regression layer is used to perform regression calculations on the unified connectivity brain map and output the social brain age of individuals with ASD.

[0008] Optionally, the formula for calculating the total loss is: In the formula, Indicates the total loss; Indicates the standard loss value; Represents the regularization parameter; This represents the total number of layers in the composite layer; Indicates the first The social brain region constraint regularization term of the combined layer.

[0009] Optionally, the step of performing cluster analysis on the ADOS social score and the brain age difference to obtain the cluster subtype of the ASD individual includes: Based on the ADOS social score and the brain age difference, a sample feature matrix is ​​constructed; Hierarchical clustering is performed on the sample feature matrix, and the optimal number of clusters is determined using a pruning method; Based on the optimal number of clusters, the corresponding cluster divisions are extracted from the hierarchical clustering results, and each ASD individual is assigned the number of the cluster to which it belongs as a cluster subtype label. Based on the cluster subtype label, the cluster subtype of all ASD individuals is obtained.

[0010] Optionally, the step of performing behavioral and neural dimension verification on the cluster subtypes, and constructing a joint profile of the subtypes, behavioral features, and neural features based on the verification results, includes: Multiple behavioral scoring indicators for individuals with ASD are extracted from the ADOS social score. Obtain the one-way difference values ​​of different behavioral scoring indicators among all cluster subtypes, and determine the target behavioral scoring indicator based on the one-way difference values; Obtain the significant difference value of the target behavior score index between each pair of cluster subtypes, and obtain the mean brain age difference value for each cluster subtype; Based on the mean brain age difference, the single-factor difference value, and the significant difference value, a joint profile between subtype and behavioral characteristics is constructed; To obtain indicators of differences in brain functional connectivity among individuals with ASD; Extract unique abnormal features of individuals with ASD from brain functional connectivity difference indicators; Obtain the modulus length with unique abnormal features, and use a non-parametric test method to perform an overall significant difference test on the modulus length to obtain the overall significant difference value; Based on the modulus of the unique abnormal features of individual ASD individuals and the overall significant difference value, a joint profile between subtype and neural features is constructed. The joint profile of the subtype and behavioral features, and the joint profile of the subtype and neural features are merged into a joint profile of the subtype, behavioral features, and neural features.

[0011] Optionally, the formula for calculating the brain functional connectivity difference index is as follows: In the formula, An index representing an individual with ASD; Indicates the first Indicators of differences in brain functional connectivity among individuals with ASD; Indicates the first The matrix formed by the functional connectivity strength between all adjacent brain regions in the initial connectivity brain map of an ASD individual; This represents the mean strength of functional connectivity between all adjacent brain regions in the TD population; The standard deviation represents the strength of functional connectivity between all adjacent brain regions in the TD population.

[0012] Secondly, this disclosure also provides a subtype classification system for autism spectrum disorder, employing the following technical solution: The brain map construction module is used to acquire neuroimaging data and functional brain region division templates of individuals with ASD, and to construct a connectivity brain map based on the neuroimaging data and the functional brain region division templates; wherein, ASD refers to autism spectrum disorder, nodes in the connectivity brain map represent brain regions, edges in the connectivity brain map represent functional connections between two brain regions, and the weight of the edge represents the strength of the functional connection. The brain age prediction module is used to construct a brain age regression model and predict the social brain age of individuals with ASD based on the connected brain map and the brain age regression model. The difference acquisition module is used to acquire the brain age difference of an ASD individual based on the individual's actual brain age and social brain age. The clustering analysis module is used to obtain the ADOS social score of individuals with ASD, and to perform clustering analysis on the ADOS social score and the brain age difference to obtain cluster subtypes; wherein, ADOS refers to the Autism Diagnosis Observation Scale. The profile building module is used to perform behavioral and neural dimension verification on the cluster subtypes, and based on the verification results, to build a joint profile of the subtypes, behavioral features and neural features.

[0013] Thirdly, this disclosure also provides a computer device, which adopts the following technical solution: The computer device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform any of the above-described autism spectrum disorder subtype classification methods.

[0014] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing computer instructions for causing a computer to execute any of the autism spectrum disorder subtype classification methods described above.

[0015] Fifthly, embodiments of this disclosure also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of any of the methods described above.

[0016] The autism spectrum disorder subtype classification method provided in this disclosure constructs a connectivity brain map by acquiring neuroimaging data and functional brain region segmentation templates of ASD individuals. This allows for an intuitive graphical representation of different brain regions and their functional connections. Nodes represent brain regions, edges represent functional connections, and edge weights indicate connection strength. This helps researchers more accurately understand the structure and characteristics of the brain functional network in ASD individuals, providing a detailed and comprehensive data foundation for subsequent analysis. Brain age is an important indicator reflecting the maturity of brain development, while social brain age focuses on the development of brain functions related to social interaction. By constructing a brain age regression model and combining it with a connectivity brain map to predict social brain age, a new quantitative indicator is provided for assessing the social functional development of ASD individuals, contributing to a deeper understanding of the brain development characteristics of ASD individuals in social aspects. Brain age difference is the difference between the actual brain age and social brain age of an ASD individual. It can intuitively reflect the degree of mismatch between an individual's brain development related to social function and overall brain development. By performing cluster analysis on ADOS social scores and brain age differences, ASD individuals can be classified into different cluster subtypes. Because autism is a highly heterogeneous disease, different subtypes of patients may exhibit different symptoms, causes, and treatment responses. Therefore, subtyping helps physicians more accurately understand the heterogeneity of autism, providing more targeted directions for subsequent research and treatment. Conducting behavioral and neurological validation of clustered subtypes, and constructing a joint profile of subtypes, behavioral characteristics, and neurological characteristics based on the validation results, integrates multi-dimensional information from behavioral and neuroimaging perspectives. This contributes to a comprehensive and in-depth understanding of the characteristics of individuals with different ASD subtypes, assists physicians in revealing the neuropathological mechanisms of autism, and provides a more comprehensive reference for clinical diagnosis and treatment. The joint profile can assist physicians in developing personalized treatment plans and rehabilitation programs for each subtype of patient, enabling precise intervention based on the patient's behavioral and neurological characteristics, improving the effectiveness and targeting of treatment, and promoting the development of autism treatment towards personalized medicine.

[0017] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the autism spectrum disorder subtype classification method provided in this embodiment of the disclosure; Figure 2 A schematic flowchart illustrating the connection mind map construction method provided in this embodiment of the disclosure; Figure 3 A flowchart illustrating the social brain age acquisition method provided in this embodiment of the disclosure; Figure 4 This is a schematic diagram of the brain age regression model provided in the embodiments of this disclosure; Figure 5 A flowchart illustrating the clustering subtype acquisition method provided in this embodiment of the disclosure; Figure 6 A flowchart illustrating the joint profile construction method provided in this embodiment of the disclosure; Figure 7 A schematic diagram of the autism spectrum disorder subtype classification system provided in this embodiment of the disclosure; Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. Detailed Implementation

[0020] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0021] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0022] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0023] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0024] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0025] Reference Figure 1 This disclosure provides a method for classifying autism spectrum disorder subtypes, including the following steps: S1: Obtain neuroimaging data and functional brain region division templates for individuals with ASD, and construct a connectivity brain map based on the neuroimaging data and functional brain region division templates; S2: Construct a brain age regression model to predict the social brain age of individuals with ASD based on connectivity brain maps and the brain age regression model; S3: Obtain the brain age difference of ASD individuals based on their actual brain age and social brain age; S4: Obtain ADOS social scores for individuals with ASD, perform cluster analysis on ADOS social scores and brain age differences, and obtain cluster subtypes; S5: Perform behavioral and neural dimension verification on the cluster subtypes, and construct a joint profile of subtypes, behavioral features and neural features based on the verification results.

[0026] ASD refers to Autism Spectrum Disorder. Nodes in the brain map represent brain regions, and edges in the brain map represent functional connections between two brain regions. The weight of the edge represents the strength of the functional connection. ADOS refers to the Autism Diagnosis Observation Scale.

[0027] The autism spectrum disorder subtype classification method disclosed herein constructs connectivity brain maps by acquiring neuroimaging data and functional brain region segmentation templates of ASD individuals. This allows for the intuitive graphical representation of different brain regions and their functional connections. Nodes represent brain regions, edges represent functional connections, and edge weights indicate connection strength. This helps researchers gain a more precise understanding of the structure and characteristics of the brain functional networks of ASD individuals, providing a detailed and comprehensive data foundation for subsequent analysis.

[0028] Brain age is an important indicator reflecting the maturity of brain development, while social brain age focuses on the development of brain functions related to social interaction. By constructing a brain age regression model and combining it with connectivity mapping to predict social brain age, a new quantitative indicator is provided for assessing the social functioning development of individuals with ASD, which helps to gain a deeper understanding of the brain development characteristics of ASD individuals in social aspects.

[0029] Brain age difference is the discrepancy between an individual's actual brain age and social brain age in autism dysplasia (ASD). It directly reflects the degree of mismatch between an individual's brain development related to social functioning and overall brain development. Cluster analysis of ADOS social scores and brain age difference allows for the classification of ASD individuals into different subtypes. Because autism is a highly heterogeneous disorder, different subtypes may present with different symptoms, causes, and treatment responses. Therefore, subtyping helps physicians more accurately understand the heterogeneity of autism, providing more targeted directions for subsequent research and treatment.

[0030] By performing behavioral and neurological validation on cluster subtypes and constructing a joint profile of subtypes, behavioral characteristics, and neurological features based on the validation results, multi-dimensional information from behavioral and neuroimaging perspectives can be integrated. This helps to comprehensively and deeply understand the characteristics of individuals with different ASD subtypes, assists doctors in revealing the neuropathological mechanisms of autism, and provides a more comprehensive reference for clinical diagnosis and treatment. The joint profile can assist doctors in developing personalized treatment plans and rehabilitation programs for each subtype of patient, enabling precise intervention based on the patient's behavioral and neurological characteristics, improving the effectiveness and targeting of treatment, and promoting the development of autism treatment towards personalized medicine.

[0031] In S1, refer to Figure 2 The flowchart illustrating the connectivity brain map construction method shows that "constructing a connectivity brain map based on neuroimaging data and functional brain region segmentation templates" includes the following steps: S11: Standardize and preprocess the neuroimaging data to obtain standard images; S12: Divide the standard image into multiple brain regions according to the functional brain region division template; S13: Extract the BOLD time-series signal of the brain region, and construct the connection matrix between brain regions based on the BOLD time-series signal; S14: Transform the connection matrix into a connection mind map of graph structure.

[0032] In S11, ASD refers to individuals with autism spectrum disorder. Neuroimaging data refers to functional magnetic resonance imaging (rs-fMRI) data of ASD individuals in a resting state. The scanner continuously acquires images of the brain over a period of time, taking one image at regular intervals to form a time-series collection of scan images.

[0033] The purpose of standardizing preprocessing neuroimaging data is to improve data quality, ensuring spatiotemporal consistency and signal stability, thereby laying the foundation for accurate and reliable subsequent analysis. In neuroimaging research, raw data may be affected by various factors. Without preprocessing, these interfering factors may mask the true neural activity information of the brain, leading to biased research results. Standardized preprocessing can effectively eliminate or reduce these interferences, making the data more consistent with research needs, enhancing the comparability of data from different subjects, and improving the stability of data from the same subject at different time points.

[0034] The standardized preprocessing includes unstable scan image removal, slice timing, head motion correction, spatial registration, signal smoothing, and linear drift correction.

[0035] At the start of a magnetic resonance imaging (MRI) scan, the system may not yet be in a stable state, and the early acquired scan images often contain a lot of noise and instability. This noise may originate from factors such as parameter adjustments during equipment startup and magnetic field instability, which can interfere with subsequent data analysis. Therefore, unstable scan image removal refers to removing the scan images from the first N time points to eliminate interference factors that may affect subsequent analysis and improve data quality. The specific value of N is usually determined based on the scanning equipment and experimental settings, and is generally 3-5 images.

[0036] Because of the time differences in acquiring different slices during MRI, data from different slices can deviate temporally. A slice refers to an image obtained at a specific point in time, where the scanner cuts the brain's three-dimensional structure into different layers. Each layer of the image is a slice, and numerous slices combined together present the complete three-dimensional structural information of the brain. In other words, the scan image acquired at each time point is actually composed of multiple slice images from different layers. The purpose of time-slice correction is to align the data from all slices temporally, ensuring that the data from each slice accurately reflects the brain's activity at the same point in time during subsequent analysis. Typically, a reference slice is selected, and the data from other slices are time-corrected to match the acquisition time with the reference slice, ensuring consistency in the temporal dimension of the data from different slices.

[0037] Individuals with ASD may experience slight head movements during scanning. These head movements can cause artifacts and spatial displacement in the scanned images, affecting image quality and the accuracy of subsequent analysis. Head motion correction involves estimating and correcting motion in the acquired scanned images, correcting the displacement and rotation caused by head movements, ensuring that all scanned images are spatially consistent, and reducing the impact of head movements on the data.

[0038] The brains of different individuals with ASD exhibit variations in anatomical structure and spatial location. Furthermore, scans of the same individual with ASD may show spatial variations at different time points. Spatial registration aims to unify scans of all individuals with ASD, or scans of the same individual at different time points, into a common spatial coordinate system. By selecting a standard anatomical template (such as the MNI template) as a reference, the scans are matched and transformed with the reference template, making different scans spatially comparable and facilitating subsequent group analysis and comparison.

[0039] Scanned images may contain random noise and high-frequency signals, which can affect the accurate extraction and analysis of brain activity signals. Signal smoothing involves processing the scanned images using smoothing filters (such as Gaussian filters) to reduce noise and high-frequency signals, making the signal more continuous and smooth. Smoothing can improve the signal-to-noise ratio of the scanned images, enhance signal stability, and facilitate subsequent detection and analysis of brain activity signals.

[0040] During prolonged scanning, the scanned image signal may exhibit linear drift, meaning the signal changes linearly over time. This linear drift can be caused by factors such as the thermal stability of the equipment or physiological factors, and can affect the accurate analysis of the signal. Linear drift correction involves fitting a linear model to remove the linear trend in the scanned image signal, making the signal more stable and more accurately reflecting the true activity of the brain.

[0041] After the above standardized preprocessing steps, standard scan images that meet the requirements can be obtained. These scan images are consistent in time and space, and the signals are more stable, providing a reliable data foundation for subsequent brain region division, functional connectivity analysis and other studies.

[0042] In S12, recognized functional brain region segmentation templates are used, such as the CC200 template (Craddock 200 functional region segmentation template). These templates are developed based on extensive neuroscience research and anatomical knowledge, dividing the brain into multiple regions with specific functions. The standardized preprocessed images are then segmented according to the selected functional brain region segmentation templates. Each segmented region is called a region of interest (ROI), also known as a brain region. Each ROI can be viewed as a functional unit, representing a relatively independent functional module in the brain. Through this segmentation, we can decompose complex brain images into multiple meaningful small regions, laying the foundation for subsequent analysis.

[0043] In S13, the blood oxygen level dependent (BOLD) signal of all voxels in each brain region is extracted, and the BOLD signal of each voxel changes over time to form a time series. The BOLD signals of all voxels in the brain region are averaged to obtain a representative BOLD time series signal of that brain region throughout the entire scanning process. This signal can represent the overall functional activity of that brain region during the scanning period.

[0044] Based on the BOLD time-series signals of brain regions, the Pearson correlation coefficient is used to calculate the functional connectivity strength between brain regions. The Pearson correlation coefficient measures the degree of linear correlation between two variables, specifically the correlation between the BOLD time-series signals of two brain regions. The correlation is calculated as follows: for each pair of different brain regions, their BOLD time-series signals are used, and the Pearson correlation coefficient is calculated using the formula to obtain a value representing the functional connectivity strength between the two brain regions.

[0045] Create a symmetric matrix with the number of rows and columns equal to the number of brain regions. Each element of the matrix corresponds to the functional connectivity strength value between a pair of brain regions. Fill the corresponding position in the matrix with the calculated functional connectivity strength values ​​between each pair of brain regions. Since the correlation is symmetric (i.e., the correlation between brain region A and brain region B is the same as the correlation between brain region B and brain region A), the matrix is ​​symmetric. The final matrix is ​​the connectivity matrix, which will serve as the basis for subsequent graph construction.

[0046] In S14, the connectivity matrix is ​​graphically represented, mapping each ROI to a node in the graph structure, with each node representing a brain region. The correlation value calculated using the Pearson correlation coefficient between pairs of brain regions is used as the weight of the edge. An edge represents a functional connection between two brain regions, and the edge weight represents the strength of that functional connection. This constructs a weighted undirected graph, which is the connectivity brain map. It can intuitively display the functional connectivity relationships between various brain regions, providing a clear visual model for further neuroscience research and analysis.

[0047] In S2, the brain age regression model includes an input layer, a multi-layer combination layer, a multi-scale brain region feature splicing layer, and a multi-layer perceptron regression layer; each combination layer in the multi-layer combination layer includes a graph convolutional layer, a social brain region enhancement mechanism layer, and a pooling layer. (See reference...) Figure 3 The flowchart illustrating the method for obtaining social brain age, "Predicting the social brain age of individuals with ASD based on connectivity brain maps and brain age regression models," includes the following steps: S21: The input layer is used to provide a connectivity mind map for multi-layer composite layers; S22: Each combined layer is used to update the nodes in the connection mind map and output a new connection mind map; S23: The multi-scale brain region feature stitching layer is used to stitch together the connectivity brain maps output by each combined layer into a unified connectivity brain map; S24: The multilayer perceptron regression layer is used to perform regression calculations on the unified connectivity brain map and output the social brain age of individuals with ASD.

[0048] In S21-S22, refer to Figure 4The diagram illustrates the structure of the brain age regression model, which is a GNN model. The input layer provides the raw input data, i.e., the connectivity brain map, for the entire model. The connectivity brain map first passes through multiple combination layers. Each combination layer updates the connectivity brain map. Specifically, the graph convolutional layer extracts brain region features from the connectivity brain map; the social brain region enhancement mechanism layer adds a social brain region regularization term to the loss function based on brain region features, forming a new loss function and obtaining the total loss; the pooling layer optimizes the weight parameters based on the total loss; the pooling layer obtains the importance score of nodes based on the weight parameters, updates the nodes in the connectivity brain map based on the importance score, and outputs a new connectivity brain map. Specifically, the Region-Aware Graph Convolution (Ra-GConv) layer, based on the Region of Interest (ROI), extracts local structural features based on the feature relationships between nodes and their neighbors in the connectivity brain map. This differs from traditional GNN methods where convolutional layers use a shared convolutional kernel for every node in the entire graph. Ra-GConv assigns a specific convolutional kernel to each node. This convolutional kernel is based on several functional community convolutional kernels. Community affiliation weight of nodes This is obtained through linear weighting. A functional community is an artificial division of the brain into different regions based on a connectivity brain map. A functional community contains multiple different brain regions that functionally cooperate to perform specific cognitive functions. For example, the social cognitive functional community includes multiple brain regions such as the medial prefrontal cortex (mPFC) and the temporoparietal junction (TPJ), which play a collaborative role in processing social information and understanding others' intentions. The expression for the convolution kernel (long vector form) of a node is: In the formula, Indicates the index of the composite layer; Indicates the index of the node; Indicates the first In the first composite layer The learnable weight matrix of each node; This indicates that the weight matrix is... Straighten it into a long vector; Indicates the first The number of manually defined functional communities in each composite layer; Indicates the first In the first composite layer, the first... The node belongs to the node. The membership degree of a functional community (referred to as community affiliation weight) is essentially a scalar, calculated by a multilayer perceptron (MLP) from the location encoding of the brain region; Indicates the first In the first composite layer The basis vectors corresponding to each functional community are parameters that are automatically learned by the entire brain age regression model through end-to-end training. Indicates the first The bias terms in each combined layer are used to adjust the output and increase the flexibility of the brain age regression model.

[0049] In summary, this convolutional kernel is a unique, learnable parameter matrix, ensuring that the parameter matrix of each node is different, resulting in different captured topological and functional information. This allows for better adaptation to the different structural and functional characteristics of each functional community. The expression for updating the embedding vector of each node through the Ra-GConv convolution operation is as follows: In the formula, Indicates the index of adjacent nodes; Indicates the first Within the first composite layer The updated embedding vector for each node; Denotes a nonlinear activation function, defined as follows: , ; Indicates the first layer of the previous combination layer Convolutional kernels for each node; Indicates the first The first combination layer Embedding vectors of each node; Indicates the first Each node The set of all adjacent nodes in a combined layer; Indicates the first In the first composite layer The node and the first The weight of the edges between nodes, also known as the functional connection strength.

[0050] The output form of brain region features extracted by the convolutional layer is an embedding feature matrix, which is composed of embedding vectors from all nodes. The matrix formed The embedded feature matrix is ​​input into the pooling layer, also known as the R-Pool Region of Interest Pooling Layer. This layer selects important brain regions using learnable projection vectors, enhancing the model's interpretability. Specifically, the pooling layer calculates the projection scores of nodes, calculates standardized scores based on these scores, determines the nodes to be retained based on the standardized scores, and updates the node features and adjacency matrix based on the retained nodes. This updates the nodes in the connectivity brain map and outputs a new connectivity brain map. The projection scores of all nodes in each component layer form a set vector, expressed as: In the formula, Indicates the first A vector of the set of projected scores of all nodes in a combined layer; Indicates the first The embedding feature matrix output by the convolutional layer in the combined layer; Indicates the first The learnable projection vectors of the combined layers, express of Norm, which is the magnitude of a vector.

[0051] The projected scores are standardized to obtain the standardized scores of the nodes. The standardized scores of all nodes in each component layer form a set vector, the expression of which is: In the formula, Indicates the first A vector of standardized scores for all nodes in a composite layer; Indicates the first The mean projected score of all nodes in the combined layer; Indicates the first The standard deviation of the projected scores of all nodes in a combined layer.

[0052] from Extract the standardized score of each node, and determine which nodes to retain based on the standardized scores. The expression for retaining nodes is: In the formula, Indicates the index of the reserved node; Indicates from Select the top with the highest standardized scores Name nodes are retained.

[0053] The connection mind map is updated based on retained nodes. The expression for the new connection mind map is: In the formula, Indicates the first The embedding matrix of all nodes in the combined layer, where each row corresponds to the embedding vector of a node. Indicates to Normalize, Compressed to the (0,1) interval, i.e. , ; Indicates by index Select rows; This represents element-wise multiplication, i.e., multiplication of the characteristic matrix. Each row (i.e., the embedding vector of each node) is multiplied by its corresponding sigmoid score: ,in, Indicates the first The total number of nodes in each composite layer, then according to... By index Select the row to get the first... The final output feature matrix of each combined layer. Indicates the first The functional connectivity matrix output by the combined layer is the new connectivity mind map; Indicates from the adjacency matrix Select to keep only nodes By connecting the elements and removing other rows and columns, the update operation of the functional connection matrix is ​​completed.

[0054] Pooling layers filter redundant information and highlight task-relevant functional brain regions by projecting node features onto learnable vectors and calculating standardized scores, retaining the nodes with the highest scores and their connections. This mechanism not only solves the computational challenges posed by high-dimensional fMRI data but also endows the model with inherent interpretability, as the retained nodes directly correspond to biomarker regions with predictive value. A social brain region enhancement mechanism layer (SAS, Social Attention Selector) is introduced during the pooling stage. This layer provides weight boosts based on whether the functional community contains preset brain regions such as mPFC, TPJ, Amygdala, and STS. Specifically, this function calculates the social brain region score of the current combined layer, constructs a social brain region regularization term based on the score, and adds this term to the model's loss function as a guiding reward mechanism for the pooling weights. The formula for calculating the social brain region score is as follows: In the formula, Indicates the first Social brain region scores of each combined layer; The index represents the preset brain region; U represents the preset brain region set. Indicates the first The first composite layer contains the first Scores for each preset brain region.

[0055] The presupposed brain regions include Cingulate, Precuneus, Angular, Temporal, Prefrontal, Orbital, Inferior Frontal, Superior Temporal, Middle Temporal, Amygdala, Postcentral, Parietal, TPJ (temporoparietal junction), STS (superior temporal sulcus), and mPFC (medial prefrontal cortex).

[0056] The expression for the social brain region constraint regularization term is: In the formula, Indicates the first The social brain region constraint regularization term of the combined layer.

[0057] Adding a social brain region regularization term to the loss function creates a new loss function. The total loss can be obtained based on this new loss function, and the formula for calculating the total loss is: In the formula, Indicates the total loss; The standard loss value refers to the loss value calculated by the loss function before introducing the social brain region regularization term. ,in This represents the mean squared error (MSE) loss. ; This represents the Mean Absolute Error (MAE). , This represents the total number of individuals with ASD. Represents the actual value. Indicates the predicted value; This represents the regularization parameter, which can be adjusted manually. This represents the total number of layers in the composite layer.

[0058] Following the social brain region enhancement mechanism layer, prior structural information representing the social brain region (including but not limited to: medial prefrontal cortex mPFC, temporoparietal junction TPJ, amygdala, and superior temporal sulcus STS) is input into the pooling scoring network in the form of a mask or weight matrix. This adds attention to the social brain region to the pooling layer, selectively filtering out relevant parts of the social brain region, which can improve the efficiency and interpretability of the brain age regression model. Then, the graph embedding features are output and the social brain age is predicted by regression. Furthermore, the parameters of the brain age regression model can be optimized through the total loss, including optimizing the weight parameters of the pooling layer.

[0059] In this method, after connecting the brain map input combination layer, convolution calculations are performed through multiple graph convolutional layers to extract brain region features at different levels. Each graph convolutional layer output is then fed into a pooling layer guided by social brain region priors, selecting key brain region nodes for downstream prediction. The pooling process is constrained by the social brain region enhancement module, and regularization terms guide the model to focus on regions related to social function. The node representation vectors after multiple Ra-GConv+R-pool combinations are concatenated into a unified representation, fusing structural information from different layers. The concatenated brain map features are then regressed using a multilayer perceptron, ultimately outputting a predicted value (Brain age), i.e., the social brain age of an ASD individual. The brain age regression model's network structure combines the powerful representational capabilities of graph neural networks in structural modeling with the interpretability advantages of prior knowledge of brain region functions. It achieves high-order feature extraction from the functional connectivity graph through a multilayer Ra-GConv structure, while effectively filtering key brain region nodes using the R-pooling mechanism, resulting in good stability and selectivity when processing high-dimensional brain map data. Furthermore, this brain age regression model explicitly introduces brain region structures related to social function as guiding information by embedding a social brain region enhancement mechanism layer. This makes the brain age regression model not only data-driven in structural learning but also consistent with brain science research at the neural mechanism level, improving the biological interpretability of the prediction results. In addition, the brain age regression model retains multi-scale brain region representations after multi-layer pooling and achieves accurate prediction of social brain age through feature splicing and MLP. Compared with traditional fully connected models, it has a faster convergence speed, higher training efficiency, and lower prediction error.

[0060] In S3, the difference between social brain age and actual brain age is the brain age difference of individuals with ASD. The brain age difference can serve as a sensitive indicator of the degree of deviation in brain development of individuals with ASD, providing quantitative support for the identification and precise intervention of heterogeneous subgroups.

[0061] In S4, refer to Figure 5 The flowchart illustrating the method for obtaining cluster subtypes, "Cluster analysis of ADOS social scores and brain age differences to obtain cluster subtypes," includes the following steps: S41: Construct a sample feature matrix based on ADOS social score and brain age difference; S42: Perform hierarchical clustering on the sample feature matrix and use pruning methods to determine the optimal number of clusters; S43: Based on the optimal number of clusters, extract the corresponding cluster partitions from the hierarchical clustering results, and assign the cluster number to each ASD individual as a cluster subtype label. Based on the cluster subtype label, obtain the cluster subtype of all ASD individuals.

[0062] In S41, ADOS (Autism Diagnostic Observation Schedule) is a standardized tool widely used to assess autism spectrum disorder (ASD). The ADOS social score includes social dimension scores, communication dimension scores, and clinical behavior scale indicators. The clinical behavior scale indicators refer to assessment indicators of other clinical behavioral performances related to ASD, such as repetitive and stereotyped behaviors, adaptability to environmental changes, and narrowing of interests. These indicators combined can more comprehensively reflect the symptom characteristics of individuals with ASD.

[0063] Since ADOS social scores and brain age differences may have different dimensions and ranges, data standardization can be performed to avoid certain features having an excessive impact on subsequent clustering results. For example, Z-score standardization can be used, which involves subtracting the mean of each feature value from its standard deviation. The standardized ADOS social scores and brain age differences are then combined into a two-dimensional sample feature matrix. Each row of this matrix represents an ASD individual, with the first column containing the brain age difference and the second column containing the ADOS social score.

[0064] In S42, Euclidean distance or cosine similarity is chosen to perform hierarchical clustering on the sample feature matrix. Initially, each sample in the feature matrix is ​​treated as a separate cluster. Based on the selected similarity index, similar clusters are continuously merged until all samples are merged into a large cluster, forming a hierarchical clustering structure. Pruning methods such as silhouette coefficient, Gap Statistic, or Calinski-Harabasz index are used to determine the optimal number of clusters K.

[0065] In S43, based on the determined optimal number of clusters, the corresponding levels are extracted from the tree structure formed by hierarchical clustering to obtain K clusters. Each cluster is assigned a unique number, and the cluster number to which each ASD individual belongs is used as its cluster subtype label. The cluster subtype labels corresponding to each ASD individual are collected and summarized to obtain the cluster subtypes of all ASD individuals.

[0066] In S5, refer to Figure 6The flowchart illustrating the joint profile construction method shows the following steps: "Performing behavioral and neural dimension verification on cluster subtypes, and constructing a joint profile based on the verification results, which includes the following steps:" S51: Extracting multiple behavioral scoring indicators for individuals with ASD from ADOS social scores; S52: Obtain the one-way difference values ​​of different behavioral scoring indicators among all cluster subtypes, and determine the target behavioral scoring indicator based on the one-way difference values; S53: Obtain the significant difference value of the target behavior score index between each two cluster subtypes, and obtain the average brain age difference value for each cluster subtype; S54: Based on the mean brain age difference, univariate difference value, and significant difference value, construct a joint profile between subtype and behavioral characteristics; S55: Obtain indicators of differences in brain functional connectivity among individuals with ASD; S56: Extract unique abnormal features of individuals with ASD from brain functional connectivity difference indicators; S57: Obtain the modulus length of unique abnormal features, and use a non-parametric test method to perform an overall significance difference test on the modulus length to obtain the overall significance difference value; S58: Based on the modulus of unique abnormal features of ASD individuals and the overall significant difference value, construct a joint profile between subtype and neural features; S59: Merge the joint profile between subtype and behavioral characteristics, and the joint profile between subtype and neural characteristics into a joint profile between subtype, behavioral characteristics and neural characteristics.

[0067] In S51 and S52, behavioral scoring indicators include social dimension scores, communication dimension scores, and clinical behavior scale indicators. Data for each behavioral scoring indicator are grouped according to cluster subtypes. Each group corresponds to a specific behavioral scoring indicator value within a given cluster subtype. The mean of the behavioral scoring indicator for each group is calculated. Based on the mean of the behavioral scoring indicator for each group, between-group mean squares and within-group mean squares are obtained. The between-group mean square reflects the degree of difference between different cluster subtypes, indicating how much the mean of each cluster subtype deviates from the overall mean. The within-group mean square reflects the dispersion of data within each cluster subtype, that is, the difference between the sample data within the same cluster subtype and the group mean. A one-way ANOVA method is used to divide the calculated between-group mean square by the within-group mean square to obtain the F-statistic. The F-statistic is an indicator used to measure the magnitude of between-group differences relative to within-group differences. A larger F-statistic value indicates that the between-group differences are more significant than the within-group differences, meaning that there are significant differences between different cluster subtypes on that behavioral scoring indicator. The F-statistic follows an F-distribution. Using the probability density function of the F-distribution, the corresponding P-value (Probability Value) can be calculated. The P-value represents the probability of obtaining the current sample data or more extreme data given the null hypothesis (i.e., the population mean of the behavioral rating indicator is equal among different cluster subtypes) is true. This P-value is the one-way difference value of different behavioral rating indicators among all cluster subtypes. By comparing the P-value with a preset significance level threshold (e.g., 0.05), it can be determined whether the difference in the mean of the behavioral rating indicator among different cluster subtypes is statistically significant. If the P-value is less than the significance level threshold, the null hypothesis is rejected, and it is considered that there is a behavioral difference among different cluster subtypes. The behavioral rating indicator corresponding to the P-value is defined as the target behavioral rating indicator because it reflects a significant difference among different cluster subtypes and is an indicator worthy of focus in subsequent analysis. If the P-value is not less than the significance level threshold, the difference in the behavioral rating indicator corresponding to the P-value between each pair of cluster subtypes is not further investigated.

[0068] In S53, the target behavior scoring index has been determined through one-way ANOVA in S52. When the results of one-way ANOVA show that there are significant differences in the target behavior scoring index between different cluster subtypes (P value is less than the preset significance level), the multiple comparison method can be used to test the difference in the target behavior scoring index between each pair of cluster subtypes to obtain the corresponding significance difference value. The magnitude of this value can clarify which cluster subtype pairs have significant differences. Here, the Tukey HSD (Tukey Honestly Significant Difference, Tukey Multiple Comparison) method is selected for multiple comparisons to effectively control the overall Type I error rate.

[0069] The average brain age difference of each cluster subtype, along with the obtained single-factor difference values ​​and significant difference values, are combined to form a joint profile between the subtype and the behavioral characteristics. The storage structure of this joint profile can be a table, bar chart, heat map, etc.

[0070] In S55, based on the connectivity map of each individual with ASD (referring to the initial connectivity map input into the brain age regression model), the brain functional connectivity difference index of the individual with ASD is obtained, and its calculation formula is as follows: In the formula, An index representing an individual with ASD; Indicates the first Indicators of differences in brain functional connectivity among individuals with ASD; Indicates the first The matrix formed by the functional connectivity strength between all adjacent brain regions in the initial connectivity brain map of an ASD individual; This represents the mean strength of functional connectivity between all adjacent brain regions in the TD population; This represents the standard deviation of the functional connectivity strength between all adjacent brain regions in the TD group. TD stands for Typically Developing, and the TD group refers to the typical developmental group. This group's development in various aspects such as physical, cognitive, linguistic, and socio-emotional aspects follows the general developmental patterns and timelines of most people, without obvious developmental disorders or neurodevelopmental diseases.

[0071] Based on this method, the standardized difference in average functional connectivity (FC) between individuals with ASD and the TD population can be calculated, defined as AFC. AFC precisely reveals abnormalities in brain connectivity at the individual level by standardizing and comparing the functional connectivity of each ASD individual with the TD population mean. It preserves spatial localization information, supports connectivity-level analysis, and quantifies the degree of deviation for each connection, thus more clearly describing the distribution and direction of brain functional abnormalities. Compared to directly using raw functional connectivity, the standardization process of AFC improves comparability across connectivity and across individuals, making it suitable for individualized analysis, subtype classification, brain network visualization, and symptom correlation studies.

[0072] In S56, COBE (Common Orthogonal Basis Extraction) is used to extract unique aberrant features from brain functional connectivity difference indices for individuals with ASD. COBE is a matrix factorization technique that aims to represent multiple samples as a shared part (common space) and an individual-specific part (residual). Applied to ASD AFC data, this means decomposing the brain functional connectivity difference indices for each ASD individual, such that AFC = shared aberrant features (common / shared) + individual-specific aberrant features (individual-specific).

[0073] Specifically, for Given 10 individuals with ASD, each with a vector length of d representing a differential index of brain functional connectivity, concatenate these indices into a matrix. The expression for this matrix is: In the formula, A splicing matrix representing differences in brain functional connectivity; Indicates the first Indicators of differences in brain functional connectivity among individuals with ASD; Represents the range of real numbers, and its matrix size is d rows. List.

[0074] COBE's goal is to The matrix is ​​decomposed into two parts: in, Represents a shared basis matrix; This represents the projection coefficients of each ASD individual onto the shared space, which is a matrix vector composed of the shared anomalous features of all ASD individuals. This represents the residual matrix, which is a matrix vector composed of the unique anomalous features of each ASD individual.

[0075] From the decomposed Each decomposed brain functional connectivity difference index can be extracted from the matrix. The expression for the decomposed brain functional connectivity difference index is: in, Indicates the first Shared anomalous characteristics of individuals with ASD; Indicates the first The shared AFC portion for each individual. Indicates the first The unique abnormal characteristics of each individual.

[0076] In S57, for each ASD individual, the magnitude of the AFC vector in its individual-specific subspace is taken, which is the unique anomalous feature, also known as the MAFCIS (Magnitude of Altered Functional Connectivity in the Individual-Specific Subspace) value. The formula for calculating the magnitude is: In the formula, Indicates the first The modulus of the unique abnormal characteristics of each individual.

[0077] Finally, the MAFCIS (Magnitude of Altered Functional Connectivity in the Individual-Specific Subspace) value is obtained for each individual, representing the intensity of its anomalous features in its individual-specific AFC space. The larger the MAFCIS, the more severe the deviation of the functional connectivity of the ASD individual from the TD.

[0078] Finally, the Kruskal-Wallis H test (essentially a non-parametric test) was used to test the overall significance of the MAFCIS distributions for all cluster subtypes, yielding significant difference values. The Kruskal-Wallis H test method first merges the MAFCIS observations for all cluster subtypes and ranks them, then calculates the rank sum for each group. The H statistic is calculated by combining the between-group rank sum with the within-group sample size. The H statistic measures the degree of variation in between-group rank differences relative to within-group rank variations; a larger value indicates a more significant difference in rank distribution among different cluster subtypes on this behavioral scoring indicator. The H statistic approximately follows a chi-square distribution (χ² distribution) with one fewer group than the number of degrees of freedom, and the corresponding P-value can be calculated using the probability density function of this distribution. The P-value represents the probability of observing a rank difference in the current sample or a more extreme rank difference if the null hypothesis (i.e., the distributions of different cluster subtypes on this behavioral scoring indicator are the same) is true; this P-value is the overall significant difference value.

[0079] The MAFCIS distribution of each cluster subtype and the obtained overall significance difference value are combined to form a joint profile of the subtype across the neural dimension. The storage structure of this joint profile can be a table, bar chart, violin plot, etc.

[0080] In S59, the joint profile between subtype and behavioral characteristics, and the joint profile between subtype and neural characteristics are fused to form a joint profile between subtype, behavioral characteristics and neural characteristics. This profile displays relevant data at three levels: individual ASD, cluster subtype and global. It can provide doctors with accurate brain development indicators, help doctors detect differences in neural development pathways between ASD subtypes, and facilitate the discovery and application of subtype-specific biomarkers by medical experts.

[0081] This study evaluated the validity of the proposed method on the Autism Brain Imaging Data Exchange (ABIDE) dataset. The ABIDE project has established a database containing a large number of resting-state functional and structural images of individuals with autism spectrum disorder (ASD) and healthy participants (TD). The project has conducted two rounds of data collection, compiling thousands of images from multiple experimental centers, and selected data from 505 patients and 530 healthy participants for validation.

[0082] To evaluate the effectiveness of the Social Brain Region Enhancement Module (SAS) in the brain age prediction task and the impact of optimized parameters on model performance, the prediction performance of the original GNN model and GNN models with different regularization parameters using the SAS module were compared. As shown in Table 1, the original GNN model had a MAE of 3.80 years and an RMSE of 5.02 years on the healthy participant test set. The GNN with SAS, with a regularization parameter of 0.1, performed best, achieving an MAE of 3.15 years (a 17% reduction) and an RMSE of 4.17 years (also a 17% reduction) on the healthy participant test set. The GNN with SAS, with a regularization parameter of 0.2, achieved an MAE of 3.42 years and an RMSE of 4.44 years on the healthy participant test set. For the ASD patient test set, all three models achieved higher MAEs and RMSEs than those on the healthy participant test set. RMSE (Mean Squared Error) is a commonly used evaluation metric in regression tasks, used to measure the difference between predicted and true values. It is the square root of MSE. Both MAE and RMSE are commonly used evaluation metrics in model regression tasks; the smaller the value, the closer the model's predicted value is to the true value, and the better the model's performance. Table 1 is as follows: Table 1 In Table 1, BrainGNN represents the original GNN model, and 0.1SAS and 0.2SAS represent social brain region regularization terms with different weights added when calculating the total loss. , The table shows that when The model performs best when the time is right.

[0083] This method proposes a subtype classification approach for autism spectrum disorder based on social brain age prediction, possessing strong theoretical innovation and practical application value. Compared with traditional whole-brain modeling methods, this method explicitly introduces prior structural information from social cognition-related brain regions (such as mPFC, TPJ, and Amygdala), guiding the model to focus on key brain regions with social functions during the graph neural network pooling stage, thereby achieving a deep integration of neuroscience knowledge and data-driven algorithms. This design not only improves the model's performance in the social brain age prediction task but also enhances its interpretability and controllability.

[0084] Empirical analysis on the ABIDE dataset validated the effectiveness of the Social Brain Region Enhancement (SAS) mechanism. In brain age prediction for healthy control groups, the model incorporating the SAS module and setting appropriate regularization parameters reduced the MAE and RMSE by approximately 17% compared to the original GNN model, demonstrating superior fitting ability and generalization performance. Furthermore, this method successfully identified differences in brain age and behavioral dimensions among individuals with ASD. By combining unsupervised clustering, behavioral scoring tests, and neural connectivity validation, a multimodal joint ASD subtype classification mechanism was constructed, significantly improving the accuracy and biological interpretability of ASD heterogeneity modeling. Overall, this method possesses advantages such as high modeling accuracy, transparent model structure, strong interpretability of neural mechanisms, and high rationality of subtype classification, showing promising prospects and clinical value in brain functional development modeling, assisting intelligent diagnosis of neurological diseases, and personalized intervention design.

[0085] Reference Figure 7 This disclosure provides a subtype classification system for autism spectrum disorder, including: The brain map construction module 101 is used to acquire neuroimaging data and functional brain region division templates of individuals with ASD, and to construct a connectivity brain map based on the neuroimaging data and functional brain region division templates; where ASD refers to autism spectrum disorder, nodes in the connectivity brain map represent brain regions, edges in the connectivity brain map represent functional connections between two brain regions, and the weight of the edge represents the strength of the functional connection. The brain age prediction module 102 is used to construct a brain age regression model, which predicts the social brain age of individuals with ASD based on the connectivity brain map and the brain age regression model. The difference acquisition module 103 is used to acquire the brain age difference of an individual with ASD based on the individual's actual brain age and social brain age. The cluster analysis module 104 is used to obtain the ADOS social score of individuals with ASD, perform cluster analysis on the ADOS social score and brain age difference, and obtain cluster subtypes; where ADOS refers to the Autism Diagnosis Observation Scale. The profile building module 105 is used to perform behavioral and neural dimension verification on cluster subtypes. Based on the verification results, a joint profile is built between subtypes, behavioral features and neural features.

[0086] The various variations and specific examples of the autism spectrum disorder subtype classification method provided above are also applicable to the autism spectrum disorder subtype classification system provided in this disclosure. Through the foregoing detailed description of the autism spectrum disorder subtype classification method, those skilled in the art can clearly understand the implementation method of the autism spectrum disorder subtype classification system. For the sake of brevity, it will not be described in detail here.

[0087] A computer device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0088] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory, causing the computer device to perform all or part of the steps of the autism spectrum disorder subtype classification method of the foregoing embodiments of this disclosure.

[0089] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0090] like Figure 8 This is a schematic diagram of a computer device provided for an embodiment of the present disclosure. It illustrates a structural schematic diagram suitable for implementing the computer device in the embodiments of the present disclosure. Figure 8 The computer device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0091] like Figure 8 As shown, a computer device may include a processor (such as a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) or programs loaded from storage devices into random access memory (RAM). The RAM also stores various programs and data required for the operation of the computer device. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0092] Typically, the following devices can be connected to the I / O interface: input devices, such as sensors or visual information acquisition devices; output devices, such as displays; storage devices, such as magnetic tapes or hard drives; and communication devices. Communication devices allow the computer device to communicate wirelessly or wiredly with other devices (such as edge computing devices) to exchange data. Although Figure 8 A computer apparatus with various devices is shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or included alternatively.

[0093] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the autism spectrum disorder subtype classification method of embodiments of this disclosure are performed.

[0094] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0095] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the autism spectrum disorder subtype classification method of the foregoing embodiments of the present disclosure are performed.

[0096] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0097] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0098] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0099] In this disclosure, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The block diagrams of devices, apparatuses, devices, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc., are open-ended terms meaning "including but not limited to," and are used interchangeably with them. The terms "or" and "and" as used herein refer to the terms "and / or," and are used interchangeably with them unless the context clearly indicates otherwise. The term "such as" as used herein refers to the phrase "such as but not limited to," and is used interchangeably with it.

[0100] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0101] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0102] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0103] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0104] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method for classifying subtypes of autism spectrum disorder, characterized in that, include: Obtain neuroimaging data and functional brain region division templates for individuals with ASD. Construct a connectivity brain map based on the neuroimaging data and the functional brain region division templates. Herein, ASD refers to autism spectrum disorder. Nodes in the connectivity brain map represent brain regions, edges in the connectivity brain map represent functional connections between two brain regions, and the weight of the edge represents the strength of the functional connection. A brain age regression model is constructed, and based on the connectivity brain map and the brain age regression model, the social brain age of individuals with ASD is predicted. Based on the actual brain age and the social brain age of individuals with ASD, the brain age difference of individuals with ASD is obtained; Obtain the ADOS social score of individuals with ASD, and perform cluster analysis on the ADOS social score and the brain age difference to obtain cluster subtypes; wherein, ADOS refers to the Autism Diagnosis Observation Scale. Behavioral and neural dimension verifications are performed on the cluster subtypes. Based on the verification results, a joint profile of the subtypes, behavioral features, and neural features is constructed.

2. The method for classifying autism spectrum disorder subtypes according to claim 1, characterized in that, The construction of a connectivity brain map based on the neuroimaging data and the functional brain region segmentation template includes: The neuroimaging data is standardized and preprocessed to obtain standard images; According to the functional brain region division template, the standard image is divided into multiple brain regions; Extract the BOLD time-series signal of the brain region, and construct the connection matrix between brain regions based on the BOLD time-series signal; The connection matrix is ​​transformed into a graph-structured connection mind map.

3. The method for classifying autism spectrum disorder subtypes according to claim 1, characterized in that, The brain age regression model includes an input layer, a multi-layer combination layer, a multi-scale brain region feature splicing layer, and a multi-layer perceptron regression layer; each of the multi-layer combination layers includes a graph convolutional layer, a social brain region enhancement mechanism layer, and a pooling layer. The input layer is used to provide the connectivity brain map for the multi-layer combined layer; Each combined layer is used to update the nodes in the connection mind map and output a new connection mind map; The graph convolutional layer is used to extract brain region features from the connected brain map; The social brain region enhancement mechanism layer is used to add a social brain region regularization term to the loss function based on the brain region features, form a new loss function and obtain the total loss, and optimize the weight parameters of the pooling layer based on the total loss. The pooling layer obtains the importance score of the node based on the weight parameter, updates the node in the connection brain map based on the importance score, and outputs a new connection brain map; The multi-scale brain region feature splicing layer is used to splice the connectivity brain maps output by each combined layer into a unified connectivity brain map. The multilayer perceptron regression layer is used to perform regression calculations on the unified connectivity brain map and output the social brain age of individuals with ASD.

4. The method for classifying autism spectrum disorder subtypes according to claim 3, characterized in that, The formula for calculating the total loss is as follows: In the formula, Indicates the total loss; Indicates the standard loss value; Represents the regularization parameter; This represents the total number of layers in the composite layer; Indicates the first The social brain region constraint regularization term of the combined layer.

5. The method for classifying autism spectrum disorder subtypes according to claim 1, characterized in that, The cluster analysis of the ADOS social score and the brain age difference to obtain the cluster subtypes of ASD individuals includes: Based on the ADOS social score and the brain age difference, a sample feature matrix is ​​constructed; Hierarchical clustering is performed on the sample feature matrix, and the optimal number of clusters is determined using a pruning method; Based on the optimal number of clusters, the corresponding cluster divisions are extracted from the hierarchical clustering results, and each ASD individual is assigned the number of the cluster to which it belongs as a cluster subtype label. Based on the cluster subtype label, the cluster subtype of all ASD individuals is obtained.

6. The method for classifying autism spectrum disorder subtypes according to claim 1, characterized in that, The process involves performing behavioral and neural dimension verification on the cluster subtypes, and based on the verification results, constructing a joint profile of the subtypes, behavioral features, and neural features, including: Multiple behavioral scoring indicators for individuals with ASD are extracted from the ADOS social score. Obtain the one-way difference values ​​of different behavioral scoring indicators among all cluster subtypes, and determine the target behavioral scoring indicator based on the one-way difference values; Obtain the significant difference value of the target behavior score index between each pair of cluster subtypes, and obtain the mean brain age difference value for each cluster subtype; Based on the mean brain age difference, the single-factor difference value, and the significant difference value, a joint profile between subtype and behavioral characteristics is constructed; To obtain indicators of differences in brain functional connectivity among individuals with ASD; Extract unique abnormal features of individuals with ASD from brain functional connectivity difference indicators; Obtain the modulus length with unique abnormal features, and use a non-parametric test method to perform an overall significant difference test on the modulus length to obtain the overall significant difference value; Based on the modulus of the unique abnormal features of individual ASD individuals and the overall significant difference value, a joint profile between subtype and neural features is constructed. The joint profile of the subtype and behavioral features, and the joint profile of the subtype and neural features are merged into a joint profile of the subtype, behavioral features, and neural features.

7. The method for classifying autism spectrum disorder subtypes according to claim 6, characterized in that, The formula for calculating the brain functional connectivity difference index is as follows: In the formula, An index representing an individual with ASD; Indicates the first Indicators of differences in brain functional connectivity among individuals with ASD; Indicates the first The matrix formed by the functional connectivity strength between all adjacent brain regions in the initial connectivity brain map of an ASD individual; This represents the mean strength of functional connectivity between all adjacent brain regions in the TD population; The standard deviation represents the strength of functional connectivity between all adjacent brain regions in the TD population.

8. A computer device, characterized in that, The computer device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the autism spectrum disorder subtype classification method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the autism spectrum disorder subtype classification method according to any one of claims 1-7.

10. A computer program product comprising computer instructions, characterized in that, When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1-7.

Images