A model training method, device, equipment and medium for simulating brain activity

Abstract

The embodiment of the present invention discloses a model training method, device, equipment and medium for simulating brain activity, which relates to the field of data processing technology. The method includes: constructing a whole-brain structural network of a target object; creating an information interaction function and a state change function for the whole-brain structural network; wherein the information interaction function is used to characterize the interaction information between brain region nodes, and the state change function is used to characterize the functional state of brain region nodes under the action of the interaction information; based on the information interaction function and the state change function, the whole-brain structural network is model trained according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activity. This scheme adopts a graph neural network based on information interaction to construct a brain dynamics model for simulating brain activity. The model can simulate brain activity more accurately and generate interpretable interaction information between brain regions, which can simultaneously meet the requirements for model accuracy and interpretability, and enhance the applicability of the model.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a model training method, device, equipment and medium for simulating brain activity. Background Art

[0002] The human brain is a complex adaptive system that can be simulated as a network of anatomically interconnected brain regions. In computational neuroscience, the dynamic activities of the brain can be described using potential causal mechanisms and complex dynamics analysis. The current problem that needs to be solved is the need for a brain activity simulation system that can meet both accuracy and interpretability.

[0003] In the related art, a neural mass model is used to simulate the whole brain activity. When constructing the model, the whole brain is usually divided into several neuron groups, and each neuron group is represented by an ordinary differential equation.

[0004] However, the above model is obtained by greatly simplifying the neural cell dynamics equations. The simplification process may not be reasonable, and the results obtained are difficult to directly compare with the real signal. Therefore, the effectiveness of the model is questioned, the model is poorly interpretable, and the scope of application of the model is narrow. Summary of the invention

[0005] The present invention provides a model training method, device, equipment and medium for simulating brain activity. A graph neural network based on information interaction is used to construct a brain dynamics model that simulates brain activity. The model can simulate brain activity more accurately and generate interpretable interaction information between brain regions. It can simultaneously meet the requirements for model accuracy and interpretability, and enhance the applicability of the model.

[0006] According to one aspect of the present invention, a model training method for simulating brain activity is provided, the method comprising:

[0007] Constructing a whole-brain structural network of the target object; wherein the whole-brain structural network is a graph neural network including brain region nodes and edges between the brain region nodes;

[0008] Creating an information interaction function and a state change function for the whole brain structure network; wherein the information interaction function is used to characterize the interaction information between the brain region nodes, and the state change function is used to characterize the functional state of the brain region nodes under the action of the interaction information;

[0009] Based on the information interaction function and the state change function, the whole brain structure network is model trained according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activities.

[0010] According to another aspect of the present invention, there is provided a model training device for simulating brain activity, the device comprising:

[0011] A whole-brain structure network construction module, used to construct a whole-brain structure network of a target object; wherein the whole-brain structure network is a graph neural network including brain region nodes and edges between the brain region nodes;

[0012] A function creation module, used to create an information interaction function and a state change function for the whole brain structure network; wherein the information interaction function is used to characterize the interaction information between the brain region nodes, and the state change function is used to characterize the functional state of the brain region nodes under the action of the interaction information;

[0013] A model training module is used to perform model training on the whole brain structure network based on the information interaction function and the state change function according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activity.

[0014] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0015] at least one processor; and,

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the model training method for simulating brain activity described in any embodiment of the present invention.

[0018] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the model training method for simulating brain activity described in any embodiment of the present invention when executed.

[0019] The technical solution of the embodiment of the present invention first constructs a whole-brain structural network of the target object; wherein the whole-brain structural network is a graph neural network including brain region nodes and edges between brain region nodes; then creates information interaction functions and state change functions for the whole-brain structural network; wherein the information interaction function is used to characterize the interaction information between brain region nodes, and the state change function is used to characterize the functional state of brain region nodes under the action of interaction information; finally, based on the information interaction function and the state change function, the whole-brain structural network is model trained according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activity. This technical solution uses a graph neural network based on information interaction to construct a brain dynamics model for simulating brain activity. The model can simulate brain activity more accurately and generate interpretable interaction information between brain regions. It can simultaneously meet the requirements for model accuracy and interpretability, so as to simulate whole-brain activity, analyze its mechanism, and distinguish whole-brain activity in disease state from that in normal state, thereby enhancing the applicability of the model.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 is a flow chart of a model training method for simulating brain activity provided according to Embodiment 1 of the present invention;

[0023] Figure 2 is a schematic diagram of function calculation provided according to Embodiment 1 of the present invention;

[0024] Figure 3 is a schematic diagram of model operation provided according to the first embodiment of the present invention;

[0025] Figure 4 is a schematic diagram of a brain dynamics model provided according to Embodiment 1 of the present invention;

[0026] Figure 5 is a flow chart of a model training method for simulating brain activity provided according to Embodiment 2 of the present invention;

[0027] Figure 6is a schematic diagram of the structure of a model training device for simulating brain activity provided according to Embodiment 3 of the present invention;

[0028] Figure 7 It is a structural schematic diagram of an electronic device for implementing a model training method for simulating brain activities according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", "target", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] Embodiment 1

[0032] Figure 1 This is a flowchart of a model training method for simulating brain activity provided in Embodiment 1 of the present invention. This embodiment is applicable to the case where a brain dynamics model simulating brain activity is constructed based on a graph neural network with information interaction. The method can be executed by a model training device simulating brain activity. The model training device simulating brain activity can be implemented in the form of hardware and / or software. The model training device simulating brain activity can be configured in an electronic device with data processing capabilities. Figure 1 As shown, the method includes:

[0033] S110, constructing the whole-brain structural network of the target object.

[0034] The whole brain structure network is a graph neural network including brain region nodes and edges between brain region nodes, which can be used to describe the structural connections between brain regions. The target object can refer to an object (person) for which brain functional activity prediction and analysis is required.

[0035] In this embodiment, it is first necessary to construct a whole-brain structural network of the target object. Optionally, constructing the whole-brain structural network of the target object includes the following steps A1-A2:

[0036] A1. Obtain a brain image of the target object based on magnetic resonance diffusion weighted imaging, and divide the brain image into brain regions based on a preset template atlas to obtain brain region nodes.

[0037] Among them, magnetic resonance diffusion weighted imaging (DWI) can be understood as an imaging method that provides information on brain physiological status based on the movement of water molecules. The preset template atlas can be used as a basis for dividing brain images into brain regions, and can be set according to actual needs. For example, the preset template atlas can use the ICBM152 template atlas and the HCP MMP 1.0 atlas defined by Glasser et al.

[0038] Specifically, the target object is first subjected to diffusion-weighted magnetic resonance imaging to obtain a brain image, including a single DWI image and a T1-weighted DWI image. Then, the single DWI image and the T1-weighted DWI image are registered using an affine transformation, and the structural image of the target object is registered to the ICBM152 template, thereby obtaining an inverse transformation from the standard space to the DWI space. The ICBM152 template atlas is then mapped to the sample space based on this inverse transformation, and the HCP MMP1.0 atlas is used to divide the brain regions of the single DWI image, and each divided brain region is used as a node.

[0039] A2. Construct edges between brain region nodes according to the connection status between brain region nodes to obtain the whole brain structure network of the target object.

[0040] The weight corresponding to each edge is determined based on the brain region connection weight between brain region nodes, and the brain region connection weight is obtained by fiber bundle tracing between brain region nodes. Exemplarily, the connection state between brain region nodes can be used to characterize whether a certain brain region node is connected to other brain region nodes, how many brain region nodes are connected to a certain brain region node, which brain region nodes they are, and what the brain region connection weight is.

[0041] Specifically, the fiber bundles are tracked by calculating the eigenvalues ​​and eigenvectors of water molecules in the brain in three directions. A connection matrix can be constructed based on the fiber bundles between two brain region nodes, and the connection matrix stores the brain region connection weights between the brain region nodes. The brain region connection weights between the brain region nodes are used as the weights of the edges between the brain region nodes, and the whole brain structure network of the target object can be constructed based on the edges between the brain region nodes.

[0042] Through such a setting, based on the constructed whole-brain structural network, this scheme can intuitively and clearly describe the structural connections between the various brain regions of the target object through the brain region nodes in the whole-brain structural network and the edges between the brain region nodes.

[0043] S120, creating information interaction functions and state change functions for the whole brain structure network.

[0044] Among them, the information interaction function is used to characterize the interaction information between brain region nodes, and the state change function is used to characterize the functional state of brain region nodes under the action of interaction information.

[0045] In this embodiment, it is assumed that the state change of a brain region in the brain depends on three parameters: the state of the brain region itself, the state of the brain region connected to the brain region, and the connection state between brain regions, and the interactions between brain regions all conform to the same pattern, that is, the brain regions are homogeneous. Under this assumption, the goal of model training is to accurately predict the changes in fMRI signals over the next period of time by combining the connection parameters between brain regions with the known functional magnetic resonance imaging (fMRI) signals at one or more time points, so as to capture the potential dynamic processes in the brain whose mechanisms have not yet been clearly defined.

[0046] In this scheme, an algebraic equation model is constructed to predict the next or a period of fMRI signal by inputting fMRI signal. Specifically, the model can be regarded as an implicit function , denoted as , the overall operation process can be expressed as .in, fMRI signals representing input and output, Represents the whole brain structural network obtained based on DWI. It is assumed that the information interaction between different brain regions follows the same set of equations The information interaction process includes two parts: information transmission and reception. Disassembled into information interaction functions and the state change function The two functions of can simplify this model to the brain region level.

[0047] Among them, the information interaction function It is transmitted between brain nodes and can be expressed as .in, Represents brain region nodes Nodes in the brain The interactive information that needs to be transmitted between Represents brain region nodes Nodes in the brain The connection characteristics between them (i.e., brain region connection weights).

[0048] Assume that the information transfer between brain regions is additive, that is, for brain region nodes For example, the effects of different brain region nodes on it are directly additive, so the functional state change of the brain region can be simplified by the state change function, which can be specifically expressed as .in, Represents brain region nodes functional status, Indicates that the brain area node and each brain region node that is connected to it The interactive information between them is summed.

[0049] In order to reduce the complexity of the model as much as possible and increase the interpretability of the model, a multi-layer perceptron can be used to construct an information interaction function in this embodiment. and the state change function , where the multilayer perceptron includes a linear layer and an activation function.

[0050] In this embodiment, optionally, creating an information interaction function for the whole brain structure network includes the following steps B1-B3:

[0051] B1. Using the first multi-layer perceptron to construct a first interaction function between a first brain region node and a second brain region node among multiple brain region nodes of a whole-brain structure network.

[0052] The second brain region node refers to a brain region node that has a connection relationship with the first brain region node in the whole brain structure network, and the input of the first multi-layer perceptron is the magnetic resonance functional imaging signal of the first brain region node and the second brain region node.

[0053] Figure 2 A schematic diagram of function calculation provided in the first embodiment of the present invention, wherein the Interaction operator and the Region operator represent information interaction functions respectively. and the state change function Calculation process. For example, Figure 2As shown in the Interaction operator, the first multilayer perceptron can include five linear layers (dark vertical bars) and four activation functions (light vertical bars). It should be noted that the use of multiple linear layers and activation functions can better describe the nonlinear relationship between input and output, and can better fit the nonlinear characteristics of brain information interaction to enhance the practicality of the model. Specifically, in the Interaction operator, the red sphere and the yellow sphere are used as the input of the first multilayer perceptron, indicating the fMRI signal values ​​of two different brain region nodes (the first brain region node and the second brain region node) that are connected. The white sphere is the output of the first multilayer perceptron, and the functional relationship between the output and the input is used as the first interaction function between the first brain region node and the second brain region node.

[0054] B2. Determine a second interaction function between the first brain region node and the second brain region node according to the first interaction function and the edge between the first brain region node and the second brain region node.

[0055] Exemplarily, the first interaction function can be multiplied by the weight of the edge between the first brain region node and the second brain region node to obtain the second interaction function between the first brain region node and the second brain region node, that is, the input of the second interaction function is the fMRI signal of the first brain region node and the second brain region node, and the output of the second interaction function is the interaction information between the first brain region node and the second brain region node. Figure 2 The Interaction operator in , multiplying the white ball (the output of the first multi-layer perceptron) and the connecting line between the red ball and the yellow ball (the weight of the edge between the two) can get the interaction information between the red ball and the yellow ball (the light gray ball).

[0056] B3. Use the second interaction function as the information interaction function between the first brain region node and the second brain region node.

[0057] In this embodiment, after determining the second interaction function, the second interaction function can be directly used as the information interaction function between the first brain region node and the second brain region node. Exemplarily, the information interaction function can be expressed as .in, represents the first interaction function (i.e., the first multilayer perceptron), and the right side of the equal sign represents the second interaction function.

[0058] In this embodiment, optionally, creating a state change function for the whole brain structure network includes the following steps C1-C3:

[0059] C1. Determine the single interaction information between the first brain region node and each second brain region node according to the information interaction function.

[0060] Specifically, Figure 2The Interaction operator in the figure only shows the information interaction process between a group of brain area nodes with connection relationships. Furthermore, for the red sphere, there may be multiple brain area nodes connected to it. According to the calculation process of the Interaction operator, the interaction information between the red sphere and each brain area node connected to the red sphere is determined respectively, thereby obtaining multiple single interaction information (i.e., light gray spheres).

[0061] C2. Add up each single interaction information corresponding to the first brain region node to obtain comprehensive interaction information between the first brain region node and all the second brain region nodes.

[0062] By adding up all the single interaction information related to the red ball, we can get the sum of all the interaction information that has information interaction with the red ball. The sum of all the interaction information is taken as the comprehensive interaction information (i.e. Figure 2 The dark grey sphere in the Regionoperator in the diagram).

[0063] C3. Use the second multi-layer perceptron to construct the state change function of the first brain region node in the whole brain structure network.

[0064] The input of the second multilayer perceptron is the functional magnetic resonance imaging signal of the first brain region node and the comprehensive interaction information corresponding to the first brain region node. Figure 2 As shown in the Region operator in, the second multi-layer perceptron may include five linear layers (dark vertical bars) and four activation functions (light vertical bars). It should be noted that for the first multi-layer perceptron and the second multi-layer perceptron, the number of linear layers and activation functions of the two may be the same or different, and may be set according to actual needs, which is not limited in this embodiment. Specifically, the red sphere and the comprehensive interaction information corresponding to the red sphere (dark gray sphere) are input into the state change function (i.e., the second multi-layer perceptron), and the output (red concave sphere) obtained is the functional state of the red sphere under the action of the comprehensive interaction information.

[0065] S130, based on the information interaction function and the state change function, a model of the whole brain structure network is trained according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activities.

[0066] In this embodiment, after the whole brain structure network, information interaction function and state change function are constructed, the magnetic resonance functional imaging signal of the target object can be obtained, and based on the information interaction function and the state change function, the whole brain structure network is model trained according to the magnetic resonance functional imaging signal of the target object, so as to obtain a brain dynamics model for simulating brain activity. Optionally, based on the information interaction function and the state change function, the whole brain structure network is model trained according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activity, including the following steps D1-D4:

[0067] D1. Grouping the functional magnetic resonance imaging signals of the target object in chronological order to obtain a training data set.

[0068] The training data set includes multiple input-output groups, and the input and output in each input-output group are magnetic resonance functional imaging signals with the same time length.

[0069] Exemplarily, based on 20 time points (one time point corresponds to 0.72s), the fMRI signal of the target object is divided in time order to obtain several input-output groups. In each input-output group, the data of the first 10 time points are used as input fMRI signals, and the data of the last 10 time points are used as output fMRI signals. In this case, the model input is the whole-brain fMRI signal value at 10 time points, and the model output is the predicted value of the whole-brain fMRI signal at the next 10 time points.

[0070] The input value is constructed using the fMRI sequence of each scan of the target subject. , output value , so that the number of time points of input value and output value are consistent. Among them, Represents the number of input-output groups. The whole cerebral cortex was divided according to the HCP MMP 1.0 atlas defined by Glasser et al., and the fMRI signal in each brain region was averaged to obtain the fMRI signal value of 360 brain regions. Therefore, the input value at each time point is a 10×360 matrix, and the output value is a 10×360 dimensional vector. The four scanning fMRI sequences of the target object were constructed in the above way to form the training data set of the target object.

[0071] After obtaining the training data set, the training data set is divided into training set, validation set and test set in a ratio of 8:1:1, that is, the first 90% of the data is used as the training set and validation set, and the last 10% of the data is used as the test set. In each epoch of training, the data of the training set and validation set are randomly shuffled and re-divided.

[0072] D2. Determine the comprehensive interaction information corresponding to each first brain region node in the whole brain structure network according to the information interaction function.

[0073] The comprehensive interaction information is the interaction information between the first brain region node and all the second brain region nodes. Specifically, for each first brain region node in the whole brain structure network, the interaction information between the first brain region node and each second brain region node with a connection relationship is first determined according to the output of the information interaction function, and then the individual interaction information is summarized (simple sum) to obtain the comprehensive interaction information corresponding to each first brain region node.

[0074] Optionally, step D2 specifically includes the following steps D21-D23:

[0075] D21. Determine the second brain region nodes to which each first brain region node is connected based on the whole brain structural network.

[0076] Specifically, in the whole-brain structural network, two brain region nodes with a connecting line (i.e., edge) represent brain region nodes with a connection relationship, thereby determining one or more second brain region nodes connected to each first brain region node based on the whole-brain structural network.

[0077] D22. Input the input functional magnetic resonance imaging signal in the training data set into the information interaction function to obtain a single interaction information between the first brain region node and each second brain region node.

[0078] Figure 3 A schematic diagram of a model operation provided in the first embodiment of the present invention. Figure 3 , taking a three-node graph neural network as an example. For the three brain region nodes of red, yellow and blue, the fMRI signals of each pair of brain region nodes with connections (edges) are input into the information interaction function In the example, we get the single interactive message msg corresponding to the connection (i.e. Figure 3 gray sphere in the image).

[0079] D23. Add up each single interaction information corresponding to the first brain region node to obtain comprehensive interaction information between the first brain region node and all the second brain region nodes.

[0080] Among them, for each first brain region node in the whole brain structure network, there will be one or more single interaction information corresponding to it, which depends on the number of second brain region nodes connected to the first brain region node. Simply summing up the single interaction information corresponding to the first brain region node can obtain the comprehensive interaction information between the first brain region node and all the second brain region nodes. For example, see Figure 3For each first brain region node in the whole brain structure network, the sum of the msgs on all edges directly connected to it is calculated as the comprehensive interaction information of the first brain region node (i.e. Figure 3 dark grey sphere in the image).

[0081] D3. Input the input functional magnetic resonance imaging signal and the comprehensive interaction information corresponding to the first brain region node into the state change function to obtain the functional state prediction value of the first brain region node under the action of the comprehensive interaction information.

[0082] For example, see Figure 3 For each first brain region node in the whole brain structure network, the comprehensive interaction information of the first brain region node and the fMRI signal of the first brain region node itself are input into the state change function In the above example, we can obtain the predicted functional state value of the first brain region node under the comprehensive interaction information (i.e. Figure 3 The concave ball in the middle).

[0083] D4. Determine the loss value based on the output magnetic resonance functional imaging signal and the functional state prediction value corresponding to the node of the first brain region, optimize the model parameters of the whole-brain structural network based on the loss value, and use the trained whole-brain structural network as a brain dynamics model for simulating brain activity when the conditions for ending model training are met.

[0084] Among them, the end condition of model training can be set to reach the total number of model training times or the loss value is less than the preset threshold. Specifically, when performing model training, the Data() function in torch_geometric can be used to load data, and the optimizer uses Adam. In addition, the loss function uses the sum of absolute errors, and L1 regularization is used, and the regularization weight is 1e-4. The output fMRI signal and the predicted functional state corresponding to the node of the first brain region are input into the loss function to obtain the loss value. The model parameters of the whole brain structure network are optimized based on the loss value, and it is judged whether the model training end condition is met at the end of each training. If it is met, the model training process can be stopped, and the whole brain structure network obtained from the last training is used as a brain dynamics model for simulating brain activity.

[0085] Figure 4 The brain dynamics model obtained through model training is presented. The model is based on graph neural network and adopts Estimate the information generated by the interaction between brain regions using Estimate the predicted value of the brain region state. Input the fMRI signal of the target object in a certain period of time into the brain dynamics model, and predict the fMRI signal of the next period of time based on the model output. Input the predicted fMRI signal into the brain dynamics model again, and continue to obtain the predicted value of the fMRI signal of the next period of time. Repeat the above process to obtain a complete fMRI prediction sequence.

[0086] The technical solution of the embodiment of the present invention first constructs a whole-brain structural network of the target object; wherein the whole-brain structural network is a graph neural network including brain region nodes and edges between brain region nodes; then creates information interaction functions and state change functions for the whole-brain structural network; wherein the information interaction function is used to characterize the interaction information between brain region nodes, and the state change function is used to characterize the functional state of brain region nodes under the action of interaction information; finally, based on the information interaction function and the state change function, the whole-brain structural network is model trained according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activity. This technical solution uses a graph neural network based on information interaction to construct a brain dynamics model for simulating brain activity. The model can simulate brain activity more accurately and generate interpretable interaction information between brain regions. It can simultaneously meet the requirements for model accuracy and interpretability, so as to simulate whole-brain activity, analyze its mechanism, and distinguish whole-brain activity in disease state from that in normal state, thereby enhancing the applicability of the model.

[0087] In this embodiment, optionally, after creating an information interaction function and a state change function for the whole-brain structure network, and before performing model training on the whole-brain structure network according to the magnetic resonance functional imaging signal of the target object based on the information interaction function and the state change function, it also includes: acquiring the magnetic resonance functional imaging signal of the reference object; based on the information interaction function and the state change function, performing hyperparameter optimization on the whole-brain structure network according to the magnetic resonance functional imaging signal of the reference object; wherein the hyperparameters include the output dimension of the information interaction function, the learning rate, the weight attenuation and the total number of training times.

[0088] The reference object can refer to a person who is different from the target object, that is, the reference object and the target object are not the same person. Specifically, before model training, a single individual (i.e., the reference object) is randomly selected for model hyperparameter optimization, and the individual does not participate in subsequent training and analysis. In the process of hyperparameter debugging, the global parameter is set to 0.6 according to the experience of previous studies.

[0089] Exemplarily, grid search is used to optimize model hyperparameters, and the search parameters include the dimension of the interactive information between the two brain regions, i.e. The output dimension of msg_dim∈[20,30,50,200], The number of layers L_M∈[2,3,4,5,6], The number of layers L_N∈[2,3,4,5,6]; learning rate Lr∈[1×10^(-3),5×10^(-4),1×10^(-4),5×10^(-5)], weight decay Wd∈[1×10^(-6),1×10^(-7),1×10^(-8)], and the total number of training epochs∈[50,100,300,400,1000,2000].

[0090] The functional magnetic resonance imaging signals of the reference object are also divided into training set, validation set and test set. The training set and validation set are used to train the models under different hyperparameter combinations, and multiple candidate models can be obtained. Among them, the specific model training process can refer to the model training process of the target object mentioned above. The mean absolute error of the test set is then used to screen each candidate model, and finally a set of optimal hyperparameter combinations is determined as follows: msg_dim=30, L_M=5, L_N=5, Lr=1×10^(-4), Wd=1×10^(-8). Considering the training effect and time, epochs is set to 100.

[0091] Through such a setting, this scheme optimizes the model's hyperparameters through model pre-training based on the fMRI signals of the reference object before formal model training, thereby determining a set of optimal hyperparameter combinations, and then performs formal model training based on the fMRI signals of the target object based on the optimal hyperparameter combinations, thereby improving the model performance and enabling the trained model to simulate brain activity more accurately.

[0092] Embodiment 2

[0093] Figure 5 A flowchart of a model training method for simulating brain activity provided in the second embodiment of the present invention, this embodiment is optimized based on the above embodiment. The specific optimization is: based on the information interaction function and the state change function, the whole brain structure network is model trained according to the magnetic resonance functional imaging signal of the target object, including: data preprocessing of the magnetic resonance functional imaging signal of the target object; wherein the data preprocessing includes data correction and registration, data smoothing and filtering, and interference data removal; based on the information interaction function and the state change function, the whole brain structure network is model trained according to the magnetic resonance functional imaging signal of the target object after data preprocessing.

[0094] like Figure 5 As shown, the method of this embodiment specifically includes the following steps:

[0095] S210, constructing a whole-brain structural network of the target object.

[0096] Among them, the whole brain structure network is a graph neural network including brain region nodes and edges between brain region nodes.

[0097] S220, creating information interaction functions and state change functions for the whole brain structure network.

[0098] Among them, the information interaction function is used to characterize the interaction information between brain region nodes, and the state change function is used to characterize the functional state of brain region nodes under the action of interaction information.

[0099] S230, performing data preprocessing on the functional magnetic resonance imaging signal of the target object.

[0100] Among them, data preprocessing includes data correction and registration, data smoothing and filtering, and interference data removal.

[0101] Specifically, the nonlinear deformation of the gradient is first corrected, and then rigidly aligned to a single-band reference image. The deformed spin echo EPI (Echo Planar Imaging) image is then rigidly aligned with the reference image. The nonlinear transformation generated in this alignment process is used to correct the fMRI signal, and finally resampled using single spline interpolation. On this basis, fine-tuning is performed to ensure that the fMRI data and structural data are aligned.

[0102] Next, the surface data is processed. First, the voxels in the cortex are mapped to the surface, and then aligned to the 32k Conte69 grid. The time series on the surface is smoothed using a check sequence with a half-width of 2 mm. The signal is then processed using a high-pass filter with a half-width greater than 2000 s. The ICA-FIX method is then used to remove the noise components in the fMRI signal, and the data is regressed using the regressor composed of the global signal. Finally, a bandpass filter of 0.009-0.08 Hz is used to remove signals in other frequency bands to remove most of the noise in the fMRI signal.

[0103] The whole cerebral cortex was divided according to the HCP MMP 1.0 atlas defined by Glasser et al., and the average value of the fMRI signal in each brain region was calculated to obtain the fMRI signal value of 360 brain regions. In order to remove the influence of head movement on the data, the time points (interference data) with head movement greater than 0.25 were deleted in this scheme, and the data were divided into small segments with unequal intervals. Each small segment was divided into a group of 20 time points with overlap, such as 1-20 time points were divided into a group, and 2-21 time points were divided into a group, so that the training data set could be expanded. The first ten time points of each group were input data, and the last ten time points were the true values ​​of the output data. If the small segment had less than 20 time points, the data of the small segment was abandoned. Individuals with input-output groups greater than 2000 were retained to ensure sufficient training data.

[0104] S240, based on the information interaction function and the state change function, a model training is performed on the whole brain structure network according to the magnetic resonance functional imaging signal of the target object after data preprocessing to obtain a brain dynamics model for simulating brain activities.

[0105] Among them, the specific implementation method of S210-S220 and S240 can refer to the relevant description in Example 1, and will not be repeated here.

[0106] It should be noted that compared with the existing technology, the goodness of fit of the brain dynamics model trained in this scheme to the whole brain signal reached 0.89, which is much higher than other models (0.35); and the correlation coefficient between the functional connection obtained by the whole brain functional signal generated by the brain dynamics model in this scheme and the real functional connection is 0.975, which is higher than other models (0.7). Therefore, the brain dynamics model of this scheme can simulate the functional signals of the whole brain more accurately.

[0107] In addition, the brain dynamics model of this scheme can derive the interaction information between different brain regions of the whole brain that cannot be represented by other models. The fluctuation value of the interaction information (such as standard deviation) is used to perform a two-sample t-test between Alzheimer's patients and the normal control group, and the p-value is less than 0.05, that is, the fluctuation value of the interaction information can effectively distinguish Alzheimer's patients from the normal control group, which shows that this model can help understand and diagnose the dynamic mechanism of brain diseases.

[0108] The technical solution of the embodiment of the present invention, after constructing the whole-brain structural network of the target object and creating information interaction functions and state change functions for the whole-brain structural network, performs data preprocessing (including data correction and alignment, data smoothing and filtering, and interference data removal) on the magnetic resonance functional imaging signal of the target object, and constructs a training data set based on the preprocessed data, and then performs model training on the whole-brain structural network based on the information interaction function and the state change function according to the training data set to obtain a brain dynamics model for simulating brain activity, thereby improving the accuracy of the model training data and helping to further enhance the performance of the model so as to more accurately simulate brain activity through the model.

[0109] Embodiment 3

[0110] Figure 6 This is a schematic diagram of the structure of a model training device for simulating brain activity provided in Embodiment 3 of the present invention. The device can execute the model training method for simulating brain activity provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. Figure 6 As shown, the device comprises:

[0111] A whole-brain structure network construction module 310 is used to construct a whole-brain structure network of a target object; wherein the whole-brain structure network is a graph neural network including brain region nodes and edges between the brain region nodes;

[0112] A function creation module 320 is used to create an information interaction function and a state change function for the whole brain structure network; wherein the information interaction function is used to characterize the interaction information between the brain region nodes, and the state change function is used to characterize the functional state of the brain region nodes under the action of the interaction information;

[0113] The model training module 330 is used to perform model training on the whole brain structure network based on the information interaction function and the state change function according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activity.

[0114] Optionally, the whole-brain structure network construction module 310 is specifically used for:

[0115] Acquiring a brain image of a target object based on magnetic resonance diffusion weighted imaging, and dividing the brain image into brain regions based on a preset template atlas to obtain brain region nodes;

[0116] The edges between the brain region nodes are constructed according to the connection status between the brain region nodes to obtain the whole-brain structural network of the target object; wherein the weight corresponding to each edge is determined based on the brain region connection weight between the brain region nodes, and the brain region connection weight is obtained by fiber bundle tracing between the brain region nodes.

[0117] Optionally, the function creation module 320 is used to:

[0118] A first multilayer perceptron is used to construct a first interaction function between a first brain region node and a second brain region node among the plurality of brain region nodes of the whole-brain structure network; wherein the second brain region node refers to a brain region node in the whole-brain structure network that has a connection relationship with the first brain region node, and the input of the first multilayer perceptron is a magnetic resonance functional imaging signal of the first brain region node and the second brain region node;

[0119] determining a second interaction function between the first brain region node and the second brain region node according to the first interaction function and an edge between the first brain region node and the second brain region node;

[0120] The second interaction function is used as the information interaction function between the first brain region node and the second brain region node.

[0121] Optionally, the function creation module 320 is further used to:

[0122] Determining single interaction information between the first brain region node and each second brain region node according to the information interaction function;

[0123] Adding up each single interaction information corresponding to the first brain region node to obtain comprehensive interaction information between the first brain region node and all second brain region nodes;

[0124] A second multi-layer perceptron is used to construct a state change function of the first brain region node in the whole-brain structure network; wherein the input of the second multi-layer perceptron is the magnetic resonance functional imaging signal of the first brain region node and the comprehensive interaction information corresponding to the first brain region node.

[0125] Optionally, the model training module 330 includes:

[0126] A training data set determination unit is used to group the functional magnetic resonance imaging signals of the target object in time sequence to obtain a training data set; wherein the training data set includes a plurality of input-output groups, and the input and output in each input-output group are functional magnetic resonance imaging signals with the same time length;

[0127] A comprehensive interaction information determination unit, configured to determine the comprehensive interaction information corresponding to each first brain region node in the whole brain structure network according to the information interaction function; wherein the comprehensive interaction information is the interaction information between the first brain region node and all the second brain region nodes;

[0128] a functional state prediction value determination unit, configured to input the input functional magnetic resonance imaging signal corresponding to the first brain region node and the comprehensive interaction information into the state change function, and obtain the functional state prediction value of the first brain region node under the action of the comprehensive interaction information;

[0129] A brain dynamics model determination unit is used to determine a loss value based on the output magnetic resonance functional imaging signal corresponding to the first brain region node and the functional state prediction value, optimize the model parameters of the whole-brain structure network based on the loss value, and use the trained whole-brain structure network as a brain dynamics model for simulating brain activity when the model training end conditions are met.

[0130] Optionally, the comprehensive interaction information determining unit is specifically configured to:

[0131] Determining a second brain region node connected to each first brain region node according to the whole brain structural network;

[0132] Inputting the input functional magnetic resonance imaging signal in the training data set into the information interaction function to obtain single interaction information between the first brain region node and each of the second brain region nodes;

[0133] Each single interaction information corresponding to the first brain region node is summed up to obtain comprehensive interaction information between the first brain region node and all the second brain region nodes.

[0134] Optionally, the device further comprises:

[0135] A reference object signal acquisition module, used for acquiring a functional magnetic resonance imaging signal of a reference object after creating an information interaction function and a state change function for the whole-brain structure network and before performing model training on the whole-brain structure network according to the functional magnetic resonance imaging signal of the target object based on the information interaction function and the state change function;

[0136] A hyperparameter optimization module, configured to optimize the hyperparameters of the whole brain structure network based on the information interaction function and the state change function and according to the functional magnetic resonance imaging signal of the reference object;

[0137] Among them, the hyperparameters include the output dimension, learning rate, weight decay and total training times of the information interaction function.

[0138] Optionally, the model training module 330 further includes:

[0139] A data preprocessing unit, used for performing data preprocessing on the functional magnetic resonance imaging signal of the target object; wherein the data preprocessing includes data correction and registration, data smoothing and filtering, and interference data removal;

[0140] A model training unit is used to perform model training on the whole brain structure network based on the information interaction function and the state change function according to the magnetic resonance functional imaging signal of the target object after data preprocessing.

[0141] A model training device for simulating brain activity provided by an embodiment of the present invention can execute a model training method for simulating brain activity provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0142] Embodiment 4

[0143] Figure 7 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0144] like Figure 7 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0145] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0146] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs the various methods and processes described above, such as a model training method for simulating brain activity.

[0147] In some embodiments, the model training method for simulating brain activity may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the model training method for simulating brain activity described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the model training method for simulating brain activity in any other appropriate manner (e.g., by means of firmware).

[0148] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0149] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0150] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, device, or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0151] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0152] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0153] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0154] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0155] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A model training method for simulating brain activity, characterized in that: The method comprises: Constructing a whole-brain structural network of the target object; wherein the whole-brain structural network is a graph neural network including brain region nodes and edges between the brain region nodes; Creating an information interaction function and a state change function for the whole brain structure network; wherein the information interaction function is used to characterize the interaction information between the brain region nodes, and the state change function is used to characterize the functional state of the brain region nodes under the action of the interaction information; Based on the information interaction function and the state change function, model training is performed on the whole brain structure network according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activity; Wherein, creating an information interaction function and a state change function for the whole brain structure network includes: A first multilayer perceptron is used to construct a first interaction function between a first brain region node and a second brain region node among the plurality of brain region nodes of the whole-brain structure network; wherein the second brain region node refers to a brain region node in the whole-brain structure network that has a connection relationship with the first brain region node, and the input of the first multilayer perceptron is a magnetic resonance functional imaging signal of the first brain region node and the second brain region node; determining a second interaction function between the first brain region node and the second brain region node according to the first interaction function and an edge between the first brain region node and the second brain region node; Using the second interaction function as the information interaction function between the first brain region node and the second brain region node; Determining single interaction information between the first brain region node and each second brain region node according to the information interaction function; Adding up each single interaction information corresponding to the first brain region node to obtain comprehensive interaction information between the first brain region node and all second brain region nodes; A second multilayer perceptron is used to construct a state change function of a first brain region node in the whole brain structure network; wherein the input of the second multilayer perceptron is a magnetic resonance functional imaging signal of the first brain region node and comprehensive interaction information corresponding to the first brain region node; Based on the information interaction function and the state change function, the whole brain structure network is trained according to the magnetic resonance functional imaging signal of the target object to obtain a brain dynamics model for simulating brain activity, including: The functional magnetic resonance imaging signals of the target object are grouped in time sequence to obtain a training data set; wherein the training data set includes a plurality of input-output groups, and the input and output in each input-output group are functional magnetic resonance imaging signals with the same time length; Determining the comprehensive interaction information corresponding to each first brain region node in the whole brain structure network according to the information interaction function; wherein the comprehensive interaction information is the interaction information between the first brain region node and all the second brain region nodes; Inputting the input functional magnetic resonance imaging signal corresponding to the first brain region node and the comprehensive interaction information into the state change function to obtain a functional state prediction value of the first brain region node under the action of the comprehensive interaction information; A loss value is determined based on the output magnetic resonance functional imaging signal corresponding to the first brain region node and the functional state prediction value, and model parameters of the whole-brain structure network are optimized based on the loss value. When the model training end conditions are met, the trained whole-brain structure network is used as a brain dynamics model for simulating brain activity.

2. The method according to claim 1, characterized in that The whole-brain structural network of the target object is constructed, including: Acquiring a brain image of a target object based on magnetic resonance diffusion weighted imaging, and dividing the brain image into brain regions based on a preset template atlas to obtain brain region nodes; The edges between the brain region nodes are constructed according to the connection status between the brain region nodes to obtain the whole-brain structural network of the target object; wherein the weight corresponding to each edge is determined based on the brain region connection weight between the brain region nodes, and the brain region connection weight is obtained by fiber bundle tracing between the brain region nodes.

3. The method according to claim 1, characterized in that Determining the comprehensive interaction information corresponding to each first brain region node in the whole brain structure network according to the information interaction function includes: Determining a second brain region node connected to each first brain region node according to the whole brain structural network; Inputting the input functional magnetic resonance imaging signal in the training data set into the information interaction function to obtain single interaction information between the first brain region node and each of the second brain region nodes; Each single interaction information corresponding to the first brain region node is summed up to obtain comprehensive interaction information between the first brain region node and all the second brain region nodes.

4. The method according to claim 1, characterized in that: After creating an information interaction function and a state change function for the whole-brain structure network, and before performing model training on the whole-brain structure network based on the information interaction function and the state change function and according to the magnetic resonance functional imaging signal of the target object, the method further includes: acquiring a functional magnetic resonance imaging signal of a reference object; Based on the information interaction function and the state change function, performing hyperparameter optimization on the whole brain structure network according to the functional magnetic resonance imaging signal of the reference object; Among them, the hyperparameters include the output dimension, learning rate, weight decay and total training times of the information interaction function.

5. The method according to claim 1, characterized in that Based on the information interaction function and the state change function, model training is performed on the whole brain structure network according to the magnetic resonance functional imaging signal of the target object, including: Performing data preprocessing on the functional magnetic resonance imaging signal of the target object; wherein the data preprocessing includes data correction and registration, data smoothing and filtering, and interference data removal; Based on the information interaction function and the state change function, model training is performed on the whole brain structure network according to the magnetic resonance functional imaging signal of the target object after data preprocessing.

6. A model training device for simulating brain activity, characterized in that: The device comprises: A whole-brain structure network construction module, used to construct a whole-brain structure network of a target object; wherein the whole-brain structure network is a graph neural network including brain region nodes and edges between the brain region nodes; A function creation module, used to create an information interaction function and a state change function for the whole brain structure network; wherein the information interaction function is used to characterize the interaction information between the brain region nodes, and the state change function is used to characterize the functional state of the brain region nodes under the action of the interaction information; A model training module, used for performing model training on the whole brain structure network based on the information interaction function and the state change function according to the magnetic resonance functional imaging signal of the target object, so as to obtain a brain dynamics model for simulating brain activity; Wherein, the function creation module is used to: A first multilayer perceptron is used to construct a first interaction function between a first brain region node and a second brain region node among the plurality of brain region nodes of the whole-brain structure network; wherein the second brain region node refers to a brain region node in the whole-brain structure network that has a connection relationship with the first brain region node, and the input of the first multilayer perceptron is a magnetic resonance functional imaging signal of the first brain region node and the second brain region node; determining a second interaction function between the first brain region node and the second brain region node according to the first interaction function and an edge between the first brain region node and the second brain region node; Using the second interaction function as the information interaction function between the first brain region node and the second brain region node; Determining single interaction information between the first brain region node and each second brain region node according to the information interaction function; Adding up each single interaction information corresponding to the first brain region node to obtain comprehensive interaction information between the first brain region node and all second brain region nodes; A second multilayer perceptron is used to construct a state change function of a first brain region node in the whole brain structure network; wherein the input of the second multilayer perceptron is a magnetic resonance functional imaging signal of the first brain region node and comprehensive interaction information corresponding to the first brain region node; The model training module includes: A training data set determination unit is used to group the functional magnetic resonance imaging signals of the target object in time sequence to obtain a training data set; wherein the training data set includes a plurality of input-output groups, and the input and output in each input-output group are functional magnetic resonance imaging signals with the same time length; A comprehensive interaction information determination unit, configured to determine the comprehensive interaction information corresponding to each first brain region node in the whole brain structure network according to the information interaction function; wherein the comprehensive interaction information is the interaction information between the first brain region node and all the second brain region nodes; a functional state prediction value determination unit, configured to input the input functional magnetic resonance imaging signal corresponding to the first brain region node and the comprehensive interaction information into the state change function, and obtain the functional state prediction value of the first brain region node under the action of the comprehensive interaction information; A brain dynamics model determination unit is used to determine a loss value based on the output magnetic resonance functional imaging signal corresponding to the first brain region node and the functional state prediction value, optimize the model parameters of the whole-brain structure network based on the loss value, and use the trained whole-brain structure network as a brain dynamics model for simulating brain activity when the model training end conditions are met.

7. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the model training method for simulating brain activity described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the model training method for simulating brain activity according to any one of claims 1 to 5 when executed.

Images