SC and FC mapping method and system based on bidirectional causal graph Transform

Abstract

The invention relates to the technical field of medical image processing and artificial intelligence, in particular to an SC and FC mapping method and system based on a bidirectional causal graph Transform. The method comprises the steps that a self-attention mechanism is combined with DAG constraint, causal relationship mapping between a brain structure connection SC and a function connection FC is captured, and a causal relationship graph between the SC and the FC is obtained; based on the causal relationship graph between the SC and the FC, the mapping performance of the SC and the FC is improved by introducing a non-causal disturbance strategy and an invariance constraint; and performing a bidirectional mapping experiment of predicting the FC from the SC and reversely deducing the SC from the FC, identifying a brain region and connection which play a key role in mapping, and obtaining causal level interpretation of a brain structure connection and function connection relationship. According to the method, more accurate prediction and explanation of the brain structure-function relationship can be realized, a prediction result with higher authenticity and higher interpretability is obtained, the potential mechanism mining capability of the brain structure-function relationship is enhanced, and the method has important significance in brain scientific research and brain disease diagnosis.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing and artificial intelligence technology, and in particular to a method and system for SC and FC mapping based on bidirectional causal graph Transformer. Background Technology

[0002] In neuroscience and neuroimaging research, structural connectivity (SC) obtained from diffusion tensor imaging (DTI) and functional connectivity (FC) extracted from resting-state functional magnetic resonance imaging (fMRI) are two core representation methods for characterizing the structure and function of brain networks.

[0003] Numerous studies have demonstrated that constructing mapping models from SC to FC or FC to SC allows for the quantitative study of the relationship between brain structure and function, thus becoming an important tool for revealing the mechanisms of human brain function. Specifically, structural connectivity provided by DTI reflects the anatomical connections between brain regions, while functional connectivity provided by resting-state fMRI reveals the patterns of coordinated activity between brain regions. The mapping between the two not only helps explain the constraints of structure on function but can also be used to infer underlying structural mechanisms.

[0004] Therefore, accurately realizing the bidirectional mapping between SC and FC is of great significance for improving the scientific rigor and accuracy of brain structure and function research. High-quality mapping models can: (1) reveal the potential mechanisms of brain structure-function relationships; (2) improve the reliability of brain disease diagnosis, individualized prediction, and clinical decision support; and (3) provide a new modeling paradigm for interdisciplinary research in neuroscience and artificial intelligence.

[0005] In recent years, graph neural networks (GNNs) have become a powerful tool for studying SC and FC mappings, thanks to their ability to capture the synergistic effects between SC and FC, thus significantly improving the accuracy of the mappings.

[0006] However, existing models of the SC and FC mapping relationship for graph neural networks still have the following common problems:

[0007] Too many unidirectional and linear assumptions: Many methods only model from SC to FC or FC to SC in a unidirectional way, or use linear / static assumptions, making it difficult to characterize the inherent nonlinearity, hierarchy and bidirectionality between structure and function;

[0008] Lack of causal interpretability: Traditional correlation or attention weights only reflect statistical associations and cannot distinguish between causality and covariation, making it difficult to answer "which connections / brain regions play a causal role in the mapping";

[0009] Insufficient generalization and robustness: Existing methods have weak adaptability to different datasets and sites, and are easily affected by non-causal features and noise, resulting in unstable prediction results.

[0010] In recent years, Generative Neural Networks (GNNs) have become an important tool for mapping SC and FC. Their advantage lies in their ability to effectively capture complex synergistic interactions between brain regions, thereby improving the accuracy and generalization performance of the mapping to a certain extent.

[0011] Existing GNN-based SC and FC mapping methods can be broadly classified into two categories:

[0012] FC to SC mapping strategy: This type of method uses models such as graph convolutional networks and multi-layer graph convolutional networks to infer structural connections from functional connections, aiming to reveal the coupling pattern between brain structure and function.

[0013] SC to FC mapping strategy: This type of method is based on graph adversarial networks, graph diffusion models, etc., and predicts functional connectivity from structural connectivity, attempting to learn the complex correspondence between brain region structural connectivity and functional activity patterns.

[0014] Although the above methods have shown great potential in the mapping study of SC and FC, the following two prominent problems still exist:

[0015] (1) Lack of interpretability: Due to the inherent "black box" nature of GNN models, it is difficult to provide causal explanations during the mapping process, and it is impossible to clearly identify the brain regions and connections that play a key role in the relationship between SC and FC. This problem significantly limits its application value in neuroscience.

[0016] (2) Neglecting bidirectional interaction: Most existing methods focus on unidirectional mapping, that is, only on the unidirectional mapping from SC to FC or from FC to SC, while ignoring the bidirectional interaction between structure and function. Therefore, the lack of bidirectional modeling leads to bias in the model when reflecting the real interaction mechanism of the brain, reducing the reliability of the mapping results and the interpretability of neuroscience.

[0017] (3) Insufficient generalization and robustness: Existing methods are weak in adaptability to different datasets and sites, and are easily affected by non-causal features and noise, resulting in unstable prediction results.

[0018] Therefore, there is an urgent need for a bidirectional relationship modeling framework for SC and FC, which introduces causal constraints into depth graph modeling, so as to support bidirectional modeling and ensure interpretability and robustness. Summary of the Invention

[0019] To address the technical problems of insufficient interpretability, neglect of bidirectional interaction, and inadequate generalization and robustness in existing technologies, this invention provides a method and system for SC and FC mapping based on a bidirectional causal graph Transformer. The technical solution is as follows:

[0020] On the one hand, a method for mapping SC and FC based on bidirectional causal graph Transformer is provided, characterized in that the method includes:

[0021] S1. By utilizing the self-attention mechanism combined with DAG graph constraints, the causal relationship mapping between the brain's structural connectivity (SC) and functional connectivity (FC) is captured, and a causal relationship graph between SC and FC is obtained.

[0022] S2. Based on the causal relationship graph between SC and FC, the mapping performance of SC and FC is trained by introducing a non-causal perturbation strategy and invariance constraints.

[0023] S3. Obtain a trained SC and FC mapping model based on a bidirectional causal graph Transformer, conduct bidirectional mapping experiments to predict FC from SC and infer SC from FC, identify brain regions and connections that play a key role in the mapping, and obtain a causal explanation of the relationship between brain structure and function.

[0024] Optionally, S1 utilizes a self-attention mechanism combined with DAG constraints to capture the causal relationship mapping between SC and FC, obtaining a causal relationship graph between SC and FC, including:

[0025] Using the SC adjacency matrix and FC adjacency matrix Where n represents the number of brain regions and the corresponding node feature matrix. and As input, through two graph neural networks and They are mapped to the latent space respectively to obtain the features. and and will and The splicing operation is performed to obtain ;

[0026] exist The autonomous attention mechanism is calculated and then passed through an MLP layer to obtain the initial causal attention graph matrix. Where d is the dimension of the feature;

[0027] Through the By imposing DAG graph constraints, a causal effect matrix expressing the relationship between SC and FC is obtained. ;

[0028] The causal relationship between SC and FC is obtained through element-wise multiplication. .

[0029] Optionally, the input node feature matrix and It is obtained by linear transformation of the corresponding adjacency matrix.

[0030] Optionally, in The autonomous attention mechanism is calculated and then passed through an MLP layer to obtain the initial causal attention graph matrix. ,include:

[0031] Calculate the initial causal attention map matrix using the following formula. :

[0032]

[0033] in, and Both are learnable parameter matrices, and It is used to encode the causal relationship between SC and FC features.

[0034] Optionally, a causal relationship graph between SC and FC can be obtained by element-wise multiplication. ,include:

[0035] Based on the following formula, the causal relationship diagram between SC and FC can be obtained by element-wise multiplication. :

[0036] .

[0037] Optionally, the causal relationship graph between SC and FC in S2 is used to train the mapping performance of SC and FC by introducing a non-causal perturbation strategy and invariant constraints, including:

[0038] Obtain the causal relationship graph between SC and FC. ;

[0039] Divide the four causal subgraphs into two groups. and And perform bidirectional mapping between SC and FC;

[0040] Invariance constraints are used to ensure that the output obtained from the perturbation sample is consistent with the output of the causal graph sample;

[0041] Construct a loss function and optimize the bidirectional mapping performance of SC and FC through training.

[0042] Optionally, the causal relationship graph between SC and FC is obtained in S2. Divide the four causal subgraphs into two groups. and And perform bidirectional mapping between SC and FC, including:

[0043] Causal relationship diagram between SC and FC Divide the four causal subgraphs into two groups. and These are used for bidirectional mapping of SC and FC respectively:

[0044]

[0045]

[0046] Among them, the graph neural network that maps FC and SC and Each input consists of four graphs, two of which are causal graphs, and the other two are generated by perturbing non-causal features to produce two features. and ;

[0047] Perform a bidirectional mapping between SC and FC according to the following formula:

[0048] ,

[0049] ,

[0050] ,

[0051] .

[0052] Optionally, a loss function is constructed, and the bidirectional mapping performance of SC and FC is optimized through training, including:

[0053] Construct a loss function and optimize the bidirectional mapping performance of SC and FC through training:

[0054]

[0055] in, Let represent the total optimization loss function for the bidirectional mapping between SC and FC.

[0056] On the other hand, a bidirectional causal graph Transformer-based SC and FC mapping system is provided. This system implements the aforementioned bidirectional causal graph Transformer-based SC and FC mapping method. The system includes:

[0057] The Bidirectional Causal Graph Transformer module is used to capture the causal relationship mapping between structural connections (SC) and functional connections (FC) by using a self-attention mechanism combined with DAG graph constraints, and obtain the causal relationship graph between SC and FC.

[0058] The causal invariant bidirectional mapping module is used to train the mapping performance of SC and FC based on the causal relationship graph between SC and FC by introducing non-causal perturbation strategies and invariance constraints.

[0059] The model output and application module is used to obtain a trained SC and FC mapping model based on a bidirectional causal graph Transformer, conduct bidirectional mapping experiments to predict FC from SC and infer SC from FC, identify brain regions and connections that play a key role in the mapping, and obtain a causal explanation of the relationship between brain structure and function.

[0060] On the other hand, a bidirectional causal graph Transformer-based SC and FC mapping device is provided, the bidirectional causal graph Transformer-based SC and FC mapping device comprising: a processor; and a memory storing computer-readable instructions, which, when executed by the processor, implement any of the methods described above for bidirectional causal graph Transformer-based SC and FC mapping.

[0061] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored therein, the at least one instruction being loaded and executed by a processor to implement any of the above-described methods of SC and FC mapping based on bidirectional causal graph Transformer.

[0062] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0063] SC and FC bidirectional mapping framework: Construct a bidirectional mapping model between brain structural connectivity (SC) and functional connectivity (FC) to realize mutual mapping between SC and FC; this framework is the foundation for realizing the core function of this invention, and without this feature, bidirectional mapping between SC and FC cannot be completed.

[0064] The Bidirectional Causal Graph Transformer module generates interpretable causal subgraphs on a DAG-constrained graph neural network and Transformer architecture. It can characterize causal effects in bidirectional mappings of SC and FC and provide biological interpretability. This feature is key to achieving causal explanation and visualization. Without this module, the causal interpretability of the model cannot be demonstrated.

[0065] Causal Invariance Bidirectional Mapping Module: Through non-causal perturbations and invariance constraints, it ensures the stability and robustness of the bidirectional mapping between SC and FC; it can effectively alleviate the heterogeneity of multimodal data, noise interference, and cross-individual / cross-center data bias; this module is the core technology to improve the model's generalization ability and task-related decoupling effect. Without this module, the model is susceptible to noise and has poor stability. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a flowchart illustrating the SC and FC mapping method based on bidirectional causal graph Transformer provided in an embodiment of the present invention.

[0068] Figure 2 A block diagram of an SC and FC mapping system based on a bidirectional causal graph Transformer provided in an embodiment of the present invention;

[0069] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0070] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0071] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0072] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0073] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0074] This invention provides a method for SC and FC mapping based on a bidirectional causal graph Transformer. This method can be implemented by a device for SC and FC mapping based on a bidirectional causal graph Transformer, which can be a terminal or a server. Figure 1 The flowchart shown is a method for mapping SC and FC based on a bidirectional causal graph Transformer, as follows: Figure 1 As shown, this invention proposes a method for mapping SC and FC based on a bidirectional causal graph Transformer. The processing flow of this method may include the following steps:

[0075] S1. By utilizing the self-attention mechanism combined with DAG graph constraints, the causal relationship mapping between the brain's structural connectivity (SC) and functional connectivity (FC) is captured, and a causal relationship graph between SC and FC is obtained.

[0076] In one feasible implementation, the present invention proposes a bidirectional causal graph Transformer that incorporates DAG constraints to enhance the causal interpretability of the bidirectional mapping between SC and FC.

[0077] In one feasible implementation, S1, using a self-attention mechanism combined with DAG constraints, the causal relationship mapping between SC and FC is captured to obtain a causal relationship graph between SC and FC, including:

[0078] Using the SC adjacency matrix and FC adjacency matrix (where n represents the number of brain regions), and the corresponding node feature matrix. and As input, through two graph neural networks and They are mapped to the latent space respectively to obtain the features. and Each graph neural network consists of two residual graph convolutional networks (GCNs). Each residual GCN block consists of a GCN layer, a dropout layer, a ReLU layer, and a residual connection.

[0079] Optionally, the input node feature matrix and It is obtained by linear transformation of the corresponding adjacency matrix.

[0080] Through Multilayer Perceptron (MLP) and Transformation to latent space features and and will and The splicing operation is performed to obtain ;

[0081] exist The autonomous attention mechanism is calculated and then passed through an MLP layer to obtain the initial causal attention graph matrix. Where d is the dimension of the feature;

[0082] Through the By imposing DAG graph constraints, a causal effect matrix expressing the relationship between SC and FC is obtained. ;

[0083] The causal relationship between SC and FC is obtained through element-wise multiplication. .

[0084] In one feasible implementation, the input node feature matrix and Obtained by applying a linear transformation to the adjacency matrix, including:

[0085] The input node feature matrix is ​​obtained by applying a linear transformation to the adjacency matrix according to the following formula. and :

[0086] .

[0087] In one feasible implementation, in The autonomous attention mechanism is calculated and then passed through an MLP layer to obtain the initial causal attention graph matrix. ,include:

[0088] Calculate the initial causal attention map matrix using the following formula. :

[0089]

[0090] in, and Both are learnable parameter matrices, and It is used to encode the causal relationship between SC and FC features.

[0091] In one feasible implementation, the causal relationship graph between SC and FC is obtained through element-wise multiplication. ,include:

[0092] The causal relationship between SC and FC can be obtained by element-wise multiplication using the following formula. :

[0093] .

[0094] The above describes the initial causal graph. The DAG constraint operations are as follows:

[0095]

[0096] Where I is the identity matrix, This represents the trace operation. The above constraints can be transformed into the following optimization function:

[0097]

[0098]

[0099] in, For Lagrange multipliers, This is a penalty parameter. It is updated iteratively. , and Obtain the internal causal relationship between SC and FC to explain the mapping relationship between SC and FC.

[0100] S2. Based on the causal relationship graph between SC and FC, the mapping performance of SC and FC is trained by introducing a non-causal perturbation strategy and invariance constraints.

[0101] In one feasible implementation, the present invention improves the mapping performance between SC and FC by introducing a non-causal perturbation strategy and invariance constraints. Based on the principle of causal invariance, a robust causal model should maintain stable generation results even when perturbed by non-causal features.

[0102] In one feasible implementation, in S2, based on the causal relationship graph between SC and FC, the mapping performance of SC and FC is trained by introducing a non-causal perturbation strategy and invariance constraints, including:

[0103] Obtain the causal relationship graph between SC and FC. ;

[0104] Divide the four causal subgraphs into two groups. and And perform bidirectional mapping between SC and FC;

[0105] Invariance constraints are used to ensure that the output obtained from the "perturbed sample" is consistent with the output of the "causal graph sample".

[0106] Construct a loss function and optimize the bidirectional mapping performance of SC and FC through training.

[0107] In one feasible implementation, a causal relationship graph between SC and FC is obtained. Divide the four causal subgraphs into two groups. and And perform bidirectional mapping between SC and FC, including:

[0108] Causal relationship diagram between SC and FC Divide the four causal subgraphs into two groups. and These are used for bidirectional mapping of SC and FC respectively:

[0109]

[0110]

[0111] Among them, the graph neural network that maps FC and SC and Each input consists of four graphs, two of which are causal graphs, and the other two are generated by perturbing non-causal features to produce two features. and The calculation formula is as follows:

[0112]

[0113]

[0114] in, Represented as a matrix of all 1s, and for The corresponding binarized matrix, It is a random noise matrix.

[0115] Perform a bidirectional mapping between SC and FC according to the following formula:

[0116] ;

[0117] ;

[0118] ;

[0119] .

[0120] The above and It also consists of two residual GCNs.

[0121] In one feasible implementation, the invariance constraint aims to ensure that the output obtained from the "perturbed sample" is consistent with the output of the "causal graph sample," and its mathematical form is:

[0122]

[0123] in, Indicates the first The mapping SC (or FC) matrix of subjects, Show the first Causal invariant mapping SC (or FC) matrix for each subject. It is the matrix after the target is mapped.

[0124] In one feasible implementation, a loss function is constructed, and the bidirectional mapping performance of SC and FC is optimized through training, including:

[0125] For the bidirectional mapping between SC and FC, the following loss function is used for optimization:

[0126] and The squared error and absolute error functions are used to maintain consistency in the overall connection strength between the true SC (FC) and the mapped FC (SC); and This represents the global and local correlation functions used to maintain the correlation between the true SC (FC) and mapped FC (SC) connectivity strengths in each brain region. and Let represent the covariance and standard deviation, respectively. In summary, the total optimal loss function for the bidirectional mapping of SC and FC is:

[0127] in, Let represent the total optimization loss function for the bidirectional mapping between SC and FC.

[0128] S3. Obtain a trained SC and FC mapping model based on a bidirectional causal graph Transformer, conduct bidirectional mapping experiments to predict FC from SC and infer SC from FC, identify brain regions and connections that play a key role in the mapping, and obtain a causal explanation of the relationship between brain structure and function.

[0129] In one feasible implementation, after training, this invention can achieve a bidirectional mapping from SC to FC and from FC to SC. Secondly, through the analysis of causal subgraphs, this invention can automatically identify brain regions and connections that play a key role in the mapping, thereby achieving a causal explanation of the relationship between brain structure and function. Finally, this method can be applied to cognitive neuroscience research and clinical fields, such as: analyzing the mechanisms of structure-function relationships, early detection and intervention of brain diseases, and personalized brain network modeling.

[0130] In one feasible implementation, the following experiment was conducted to demonstrate the accuracy of the bidirectional SC and FC method provided by the present invention.

[0131] Experimental setup: The dataset was randomly divided into training, validation, and test sets in an 8:1:1 ratio. The experimental development environment used was PyTorch 1.9.1 on an NVIDIA GeForce RTX 4060 Ti graphics processor. The model training ran for 100 epochs using the Adam optimizer, with an initial learning rate of 0.003 and a batch size of 4. The experimental environment and specific settings are shown in Table 1.

[0132] Table 1 Summary of Experimental Parameter Settings

[0133]

[0134] In addition, the present invention uses mean square error and Pearson correlation as evaluation indicators in the experiment to evaluate the effect of the mapping method provided in this embodiment. See Table 2, which shows that the mapping method provided in this embodiment has a good effect.

[0135] Table 2. Bidirectional mapping results in this embodiment.

[0136]

[0137] This invention provides a brain network mapping method based on a bidirectional causal graph Transformer. By encoding the SC (Structured Brain) obtained from DTI (Digital Transformation Theory) and FC (Functional Brain) obtained from resting-state fMRI using a graph neural network, and extracting their node feature representations, a dual-channel causal decoupling module is used to automatically identify the causal relationship between structure and function under causal DAG constraints. Bidirectional feature interaction fusion is achieved through multi-head attention and gating mechanisms to construct causal subgraph relationships. Subsequently, causal invariance constraints are introduced to suppress non-causal perturbations, thereby optimizing the mapping results at both the global and local levels. Through the above design, this invention can achieve bidirectional prediction of structural connections to functional connections and functional connections to structural connections. The resulting mapping results are more accurate and stable in terms of node features and network topology information. Simultaneously, this invention can automatically identify brain regions and causal connections that play a key role in the mapping process, thereby improving the interpretability of the model. Furthermore, this invention maintains strong generalization and robustness under cross-dataset and cross-site conditions, avoiding interference from non-causal features and noise on the prediction. Therefore, this invention enables more accurate prediction and interpretation of the brain structure-function relationship, yielding more accurate and interpretable prediction results, enhancing the ability to uncover the potential mechanisms of the brain structure-function relationship, and is of great significance to brain science research and brain disease diagnosis.

[0138] Figure 2 This is a block diagram of a bidirectional causal graph Transformer-based SC and FC mapping system 300, illustrated according to an exemplary embodiment. The system 300 is used for a bidirectional causal graph Transformer-based SC and FC mapping method. (Refer to...) Figure 2 The system includes a bidirectional causal graph Transformer module 310, a causal invariant bidirectional mapping module 320, and a model output and application module 330. Among them:

[0139] The bidirectional causal graph Transformer module 310 is used to capture the causal relationship mapping between the brain's structural connectivity (SC) and functional connectivity (FC) by utilizing the self-attention mechanism combined with DAG graph constraints, and obtain the causal relationship graph between SC and FC.

[0140] The causal invariant bidirectional mapping module 320 is used to train the mapping performance of SC and FC based on the causal relationship graph between SC and FC by introducing non-causal perturbation strategies and invariance constraints.

[0141] The model output and application module 330 is used to obtain the trained SC and FC mapping model based on bidirectional causal graph Transformer, conduct bidirectional mapping experiments to predict FC from SC and infer SC from FC, identify brain regions and connections that play a key role in the mapping, and obtain causal explanations of the connections between brain structure and function.

[0142] Optionally, the bidirectional causal graph Transformer module 310 is used for:

[0143] Using the SC adjacency matrix and FC adjacency matrix Where n represents the number of brain regions and the corresponding node feature matrix. and As input, through two graph neural networks and They are mapped to the latent space respectively to obtain the features. and and will and The splicing operation is performed to obtain ;

[0144] exist The autonomous attention mechanism is calculated and then passed through an MLP layer to obtain the initial causal attention graph matrix. Where d is the dimension of the feature;

[0145] Through the By imposing DAG graph constraints, a causal effect matrix expressing the relationship between SC and FC is obtained. ;

[0146] The causal relationship between SC and FC is obtained through element-wise multiplication. .

[0147] Optionally, the input node feature matrix and It is obtained by linear transformation of the corresponding adjacency matrix.

[0148] Optionally, the bidirectional causal graph Transformer module 310 is used for:

[0149] Calculate the initial causal attention map matrix using the following formula. :

[0150]

[0151] in, and Both are learnable parameter matrices, and It is used to encode the causal relationship between SC and FC features.

[0152] Optionally, the bidirectional causal graph Transformer module 310 is used for:

[0153] Based on the following formula, the causal relationship diagram between SC and FC can be obtained by element-wise multiplication. :

[0154] .

[0155] Optionally, the causal invariant bidirectional mapping module 320 is used for:

[0156] Obtain the causal relationship graph between SC and FC. ;

[0157] Divide the four causal subgraphs into two groups. and And perform bidirectional mapping between SC and FC;

[0158] Invariance constraints are used to ensure that the output obtained from the perturbation sample is consistent with the output of the causal graph sample;

[0159] Construct a loss function and optimize the bidirectional mapping performance of SC and FC through training.

[0160] Optionally, the causal invariant bidirectional mapping module 320 is used for:

[0161] Causal relationship diagram between SC and FC Divide the four causal subgraphs into two groups. and These are used for bidirectional mapping of SC and FC respectively:

[0162]

[0163]

[0164] Among them, the graph neural network that maps FC and SC and Each input consists of four graphs, two of which are causal graphs, and the other two are generated by perturbing non-causal features to produce two features. and ;

[0165] Perform a bidirectional mapping between SC and FC according to the following formula:

[0166] ,

[0167] ,

[0168] ,

[0169] .

[0170] Optionally, a loss function is constructed, and the bidirectional mapping performance of SC and FC is optimized through training, including:

[0171] Construct a loss function and optimize the bidirectional mapping performance of SC and FC through training:

[0172] in, Let represent the total optimization loss function for the bidirectional mapping between SC and FC.

[0173] This invention provides a brain network mapping method based on a bidirectional causal graph Transformer. By encoding the SC (Structured Brain) obtained from DTI (Digital Transformation Theory) and FC (Functional Brain) obtained from resting-state fMRI using a graph neural network, and extracting their node feature representations, a dual-channel causal decoupling module is used to automatically identify the causal relationship between structure and function under causal DAG constraints. Bidirectional feature interaction fusion is achieved through multi-head attention and gating mechanisms to construct causal subgraph relationships. Subsequently, causal invariance constraints are introduced to suppress non-causal perturbations, thereby optimizing the mapping results at both the global and local levels. Through the above design, this invention can achieve bidirectional prediction of structural connections to functional connections and functional connections to structural connections. The resulting mapping results are more accurate and stable in terms of node features and network topology information. Simultaneously, this invention can automatically identify brain regions and causal connections that play a key role in the mapping process, thereby improving the interpretability of the model. Furthermore, this invention maintains strong generalization and robustness under cross-dataset and cross-site conditions, avoiding interference from non-causal features and noise on the prediction. Therefore, this invention enables more accurate prediction and interpretation of the brain structure-function relationship, yielding more accurate and interpretable prediction results, enhancing the ability to uncover the potential mechanisms of the brain structure-function relationship, and is of great significance to brain science research and brain disease diagnosis.

[0174] Figure 3 This is a schematic diagram of the structure of an SC and FC mapping device based on a bidirectional causal graph Transformer provided in an embodiment of the present invention, as shown below. Figure 3 As shown, a SC to FC mapping device based on a bidirectional causal graph Transformer may include the above-mentioned... Figure 2The diagram illustrates a SC to FC mapping system based on a bidirectional causal graph Transformer. Optionally, a SC to FC mapping device 410 based on a bidirectional causal graph Transformer may include a first processor 2001.

[0175] Optionally, an SC to FC mapping device 410 based on a bidirectional causal graph Transformer may further include a memory 2002 and a transceiver 2003.

[0176] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0177] The following is combined Figure 3 The various components of a bidirectional causal graph Transformer-based SC to FC mapping device 410 are described in detail below:

[0178] The first processor 2001 is the control center of the SC to FC mapping device 410 based on a bidirectional causal graph Transformer. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0179] Optionally, the first processor 2001 can perform various functions of a bidirectional causal graph Transformer-based SC to FC mapping device 410 by running or executing software programs stored in memory 2002 and calling data stored in memory 2002.

[0180] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 3 CPU0 and CPU1 are shown in the diagram.

[0181] In a specific implementation, as one example, an SC to FC mapping device 410 based on a bidirectional causal graph Transformer may also include multiple processors, for example... Figure 3 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0182] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0183] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently, and may be connected via an interface circuit of a bidirectional causal graph Transformer-based SC to FC mapping device 410. Figure 3 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0184] The transceiver 2003 is used to communicate with network devices or with mobile terminals.

[0185] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 3 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0186] Optionally, the transceiver 2003 can be integrated with the first processor 2001 or exist independently, and can be interfaced through an interface circuit of an SC to FC mapping device 410 based on a bidirectional causal graph Transformer. Figure 3 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0187] It should be noted that, Figure 3 The structure of the SC and FC mapping device 410 based on the bidirectional causal graph Transformer shown in the figure does not constitute a limitation on the router. The actual knowledge structure identification device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0188] Furthermore, the technical effects of the SC and FC mapping device 410 based on a bidirectional causal graph Transformer can be referred to the technical effects of the SC and FC mapping method based on a bidirectional causal graph Transformer described in the above method embodiments, and will not be repeated here.

[0189] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0190] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0191] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable sensors. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0192] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0193] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0194] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for various specific applications, but such implementations should not be considered beyond the scope of this invention.

[0195] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0196] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0197] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0198] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for mapping SC and FC based on bidirectional causal graph Transformer, characterized in that, The method includes: S1. By utilizing the self-attention mechanism combined with DAG graph constraints, the causal relationship mapping between the brain's structural connectivity (SC) and functional connectivity (FC) is captured, and a causal relationship graph between SC and FC is obtained. S2. Based on the causal relationship graph between SC and FC, the mapping performance of SC and FC is trained by introducing a non-causal perturbation strategy and invariance constraints. S3. Obtain a trained SC and FC mapping model based on a bidirectional causal graph Transformer, conduct bidirectional mapping experiments to predict FC from SC and infer SC from FC, identify brain regions and connections that play a key role in the mapping, and obtain a causal explanation of the structural and functional connectivity of the brain.

2. The method according to claim 1, characterized in that, S1 utilizes a self-attention mechanism combined with DAG constraints to capture the causal relationship mapping between SC and FC, obtaining a causal relationship graph between SC and FC, including: Using the SC adjacency matrix and FC adjacency matrix Where n represents the number of brain regions and the corresponding node feature matrix. and As input, through two graph neural networks and They are mapped to the latent space respectively to obtain the features. and and will and The splicing operation is performed to obtain ; exist The autonomous attention mechanism is calculated and then passed through an MLP layer to obtain the initial causal attention graph matrix. Where d is the dimension of the feature; Through the By imposing DAG graph constraints, a causal effect matrix expressing the relationship between SC and FC is obtained. ; The causal relationship between SC and FC is obtained through element-wise multiplication. .

3. The method according to claim 2, characterized in that, Input node feature matrix and It is obtained by linear transformation of the corresponding adjacency matrix.

4. The method according to claim 3, characterized in that, exist The autonomous attention mechanism is calculated and then passed through an MLP layer to obtain the initial causal attention graph matrix. ,include: Calculate the initial causal attention map matrix using the following formula. : ； in, and Both are learnable parameter matrices, and It is used to encode the causal relationship between SC and FC features.

5. The method according to claim 4, characterized in that, The causal relationship diagram between SC and FC is obtained by element-wise multiplication. ,include: Based on the following formula, the causal relationship diagram between SC and FC can be obtained by element-wise multiplication. : 。 6. The method according to claim 5, characterized in that, The causal relationship graph between SC and FC in S2 is used to train the mapping performance of SC and FC by introducing a non-causal perturbation strategy and invariant constraints, including: Obtain the causal relationship graph between SC and FC. ; Divide the four causal subgraphs into two groups. and And perform bidirectional mapping between SC and FC; Invariance constraints are used to ensure that the output obtained from the perturbation sample is consistent with the output of the causal graph sample; Construct a loss function and optimize the bidirectional mapping performance of SC and FC through training.

7. The method according to claim 6, characterized in that, Causal relationship diagram between obtaining SC and FC in S2 Divide the four causal subgraphs into two groups. and And perform bidirectional mapping between SC and FC, including: Causal relationship diagram between SC and FC Divide the four causal subgraphs into two groups. and These are used for bidirectional mapping of SC and FC respectively: ； ； Among them, the graph neural network that maps FC and SC and Each input consists of four graphs, two of which are causal graphs, and the other two are generated by perturbing non-causal features to produce two features. and ; Perform a bidirectional mapping between SC and FC according to the following formula: ， ， ， 。 8. The method according to claim 7, characterized in that, Construct a loss function and optimize the bidirectional mapping performance of SC and FC through training, including: Construct a loss function and optimize the bidirectional mapping performance of SC and FC through training: ； in, Let represent the total optimization loss function for the bidirectional mapping between SC and FC.

9. A system for mapping SC and FC based on a bidirectional causal graph Transformer, wherein the system is used to implement the method for mapping SC and FC based on a bidirectional causal graph Transformer as described in any one of claims 1-8, characterized in that, The system includes: The Transformer module for bidirectional causal graphs is used to capture the causal relationship mapping between the brain's structural connectivity (SC) and functional connectivity (FC) by utilizing the self-attention mechanism combined with DAG graph constraints, and to obtain the causal relationship graph between SC and FC. The causal invariant bidirectional mapping module is used to train the mapping performance of SC and FC based on the causal relationship graph between SC and FC by introducing non-causal perturbation strategies and invariance constraints. The model output and application module is used to obtain a trained SC and FC mapping model based on a bidirectional causal graph Transformer, conduct bidirectional mapping experiments to predict FC from SC and infer SC from FC, identify brain regions and connections that play a key role in the mapping, and obtain a causal explanation of the structural and functional connectivity of the brain.

10. A SC to FC mapping device based on a bidirectional causal graph Transformer, the SC to FC mapping device based on a bidirectional causal graph Transformer comprising: Processor; memory storing computer-readable instructions that, when executed by the processor, implement any one of the SC and FC mapping methods based on bidirectional causal graph Transformer as described in any one of claims 1-8.

Images