Method and apparatus for deriving effective connectivity information, method and apparatus for training digital twin brain model, and device

Abstract

Disclosed in the present application are a method and apparatus for deriving effective connectivity information, a method and apparatus for training a digital twin brain model, and a device. The method comprises: determining perturbation data of a first brain node among at least two brain nodes and brain neural data of the at least two brain nodes in a time series; by means of a digital twin brain model, and, on the basis of the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series, obtaining first predicted neural data of the at least two brain nodes at the next moment; and, on the basis of the first predicted neural data of the at least two brain nodes at the next moment and second predicted neural data of the at least two brain nodes at the next moment, determining brain effective connectivity information from the first brain node to a second brain node. On the basis of a training data set, a time series prediction network is trained to obtain the digital twin brain model.

Description

Effective methods for deriving information, training methods, devices, and equipment for digital twin brain models.

[0001] Cross-references to related applications

[0002] This application claims priority to Chinese Patent Application No. 202411569048.4, filed on November 4, 2024, entitled “Method for Derivation of Effective Connection Information, Method, Apparatus, and Device for Training Digital Twin Brain Model”, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This invention relates to the field of brain neuroscience technology, and in particular to a method for deriving effective brain connectivity information, a training method for digital twin brain models, and related devices and equipment. Background Technology

[0004] Effective connectivity (EC) characterizes causal interactions between brain regions and is fundamental to understanding brain information processing. EC can be obtained through experimental and data-driven methods.

[0005] However, common experimental methods are not applicable to stimulation and observation across the entire human brain; common data-driven methods for inferring effective connectivity are computationally complex, with model-based methods being particularly difficult to compute, while model-free methods can only distinguish the existence of directed connections.

[0006] It is evident that neither common experimental methods nor data-driven approaches can obtain accurate information about effective brain connectivity. Summary of the Invention

[0007] This application provides a method for deriving effective brain connectivity information, a method for training a digital twin brain model, an apparatus, and a device that can obtain accurate effective brain connectivity information.

[0008] The technical solution of this application embodiment is implemented as follows:

[0009] In a first aspect, embodiments of this application provide a method for deriving effective brain connectivity information, the method comprising:

[0010] Determine the perturbation data of the first brain node among at least two brain nodes and the brain neural data of the at least two brain nodes in the time series;

[0011] Using a digital twin brain model, based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series, the first predicted neural data of the at least two brain nodes at the next time step is obtained; wherein, the digital twin brain model is obtained by training based on a time series prediction network;

[0012] Based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step, the effective brain connection information from the first brain node to the second brain node is determined; wherein, the second brain node is the other brain node among the at least two brain nodes besides the first brain node; the second predicted neural data represents the predicted data obtained without adding the perturbation data.

[0013] In the embodiments of this application, based on a digital twin brain model obtained through training, the brain's response to perturbation stimuli is predicted and effective brain connectivity information is inferred by adding perturbation data to brain nodes. The digital twin brain model is obtained through training on a time-series prediction network; therefore, it can effectively and accurately predict the dynamic changes in brain neural data, thereby obtaining more accurate information on effective brain connectivity.

[0014] Secondly, embodiments of this application provide a method for training a digital twin brain model, the method comprising:

[0015] A time-series prediction network is trained based on a training dataset to obtain a digital twin brain model; wherein, the digital twin brain model is used to predict the neural data of the next time step based on the brain neural data of the brain nodes in the time series.

[0016] The training dataset includes neural training data for at least two brain nodes corresponding to a preset time length p+1; wherein, the neural training data corresponding to the preset time length p+1 includes neural training data from time mp to time m, and the training dataset also includes neural training data for the at least two brain nodes at time m+1; p is an integer greater than or equal to 0, and m is an integer greater than p.

[0017] The process of training a time-series prediction network based on a training dataset to obtain a digital twin brain model includes:

[0018] Using the time series prediction network, based on the neural training data of the at least two brain nodes from the mp-th time to the m-th time, the prediction data of the at least two brain nodes at the (m+1)-th time is obtained;

[0019] Based on the prediction data of the at least two brain nodes at time m+1 and the neural training data of the at least two brain nodes at time m+1, the time series prediction network is modified to obtain the digital twin brain model.

[0020] In the embodiments of this application, time-series prediction networks can be used to learn and predict dynamic changes in brain neural signals to obtain a digital twin brain model with time-series prediction capabilities. This digital twin brain model can effectively and accurately predict the dynamic changes in brain neural data.

[0021] Thirdly, embodiments of this application provide a device for deriving effective brain connectivity information, the device comprising:

[0022] A determining unit is used to determine the perturbation data of the first brain node among at least two brain nodes and the brain neural data of the at least two brain nodes in a time series;

[0023] The acquisition unit is used to obtain the first predicted neural data of the at least two brain nodes at the next moment based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series using a digital twin brain model; wherein, the digital twin brain model is obtained by training based on a time series prediction network;

[0024] The determining unit is further configured to determine effective brain connectivity information between the first brain node and the second brain node based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step; wherein, the second brain node is the other brain node among the at least two brain nodes besides the first brain node; the second predicted neural data represents the predicted data obtained without adding the perturbation data.

[0025] Fourthly, embodiments of this application provide a training device for a digital twin brain model, the training device for the digital twin brain model comprising:

[0026] The training unit is used to train the time series prediction network based on the training dataset to obtain a digital twin brain model; wherein, the digital twin brain model is used to predict the neural data of the next time step based on the brain neural data of the brain nodes in the time series.

[0027] The training dataset includes neural training data for at least two brain nodes corresponding to a preset time length p+1; wherein, the neural training data corresponding to the preset time length p+1 includes neural training data from time mp to time m, and the training dataset also includes neural training data for the at least two brain nodes at time m+1; p is an integer greater than or equal to 0, and m is an integer greater than p.

[0028] The process of training a time-series prediction network based on a training dataset to obtain a digital twin brain model includes:

[0029] Using the time series prediction network, based on the neural training data of the at least two brain nodes from the mp-th time to the m-th time, the prediction data of the at least two brain nodes at the (m+1)-th time is obtained;

[0030] Based on the prediction data of the at least two brain nodes at time m+1 and the neural training data of the at least two brain nodes at time m+1, the time series prediction network is modified to obtain the digital twin brain model.

[0031] Fifthly, embodiments of this application provide a computer device, the computer device comprising: a processor and a memory; wherein,

[0032] The memory is used to store computer programs that can run on the processor;

[0033] The processor is configured to execute the method described in the first or second aspect when running the computer program.

[0034] Sixthly, embodiments of this application provide a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the method described in the first or second aspect above.

[0035] In a seventh aspect, embodiments of this application provide a computer program product, including a computer program or instructions, which, when executed by a processor, implement the method described in the first or second aspect above.

[0036] Therefore, in the embodiments of this application, the digital twin brain model is obtained based on time series prediction network training. Thus, the digital twin brain model can effectively and accurately predict the dynamic changes of brain neural data, thereby obtaining more accurate information on effective brain connectivity. Attached Figure Description

[0037] Figure 1 is a schematic diagram of the implementation process of the method for deriving effective brain connectivity information proposed in this application embodiment;

[0038] Figure 2 is a schematic diagram of the brain node proposed in the embodiment of this application;

[0039] Figure 3 is a schematic diagram of the implementation process of the method for deriving effective brain connectivity information proposed in the embodiments of this application.

[0040] Figure 4 is a schematic diagram illustrating the derivation of effective brain connectivity information proposed in an embodiment of this application.

[0041] Figure 5 is a schematic diagram (2) illustrating the derivation of effective brain connectivity information proposed in an embodiment of this application;

[0042] Figure 6 is a schematic diagram illustrating the derivation of effective brain connectivity information proposed in an embodiment of this application.

[0043] Figure 7 is a schematic diagram illustrating the derivation of effective brain connectivity information proposed in an embodiment of this application.

[0044] Figure 8 is a schematic diagram illustrating the derivation of effective brain connectivity information proposed in an embodiment of this application.

[0045] Figure 9 is a schematic diagram of the implementation process of training the digital twin brain model proposed in the embodiments of this application;

[0046] Figure 10 is a second schematic diagram of the brain node proposed in the embodiment of this application;

[0047] Figure 11 is a schematic diagram of neural data analysis under real stimulation proposed in the embodiments of this application;

[0048] Figure 12 is a schematic diagram illustrating the implementation of the derivation of effective brain connectivity information based on a digital twin brain model obtained through training, as proposed in an embodiment of this application.

[0049] Figure 13 is a schematic diagram of the composition structure of the brain effective connectivity information derivation device proposed in the embodiment of this application;

[0050] Figure 14 is a schematic diagram of the composition structure of the training device for the digital twin brain model proposed in the embodiments of this application;

[0051] Figure 15 is a schematic diagram of the composition structure of the computer device proposed in the embodiments of this application. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the relevant application and not for limiting the application. Furthermore, it should be noted that, for ease of description, only the parts related to the relevant application are shown in the accompanying drawings.

[0053] Brain network models are commonly used to characterize the connections between different brain regions and the transitions in brain states, serving as an important method for understanding the working mechanisms of the brain at different scales and under different conditions. Exploring the causal relationships between different regions in brain networks is crucial for a deeper understanding of the brain's higher cognitive functions and for locating lesions in functional brain diseases.

[0054] Causal relationships in brain networks describe how one brain region directly influences other brain regions. Constructing causal brain networks requires determining the magnitude and direction of these interactions and distinguishing between excitatory and inhibitory effects; such causal relationships are also called effective connections.

[0055] However, the dynamic changes of brain networks are highly nonlinear, and constructing linear dynamic models with finite state variables is insufficient to accurately describe the complexity of whole-brain network dynamics. Furthermore, the direction and magnitude of information flow between different brain regions and the causality of their interactions remain unclear. Therefore, accurately establishing brain network dynamic models to describe the complex dynamic processes in the brain is a significant challenge.

[0056] Currently, effective connections can be obtained through experimental and data-driven methods.

[0057] Experimental methods involve perturbing brain signals in one region using electrical or magnetic stimulation and observing how this perturbation propagates to other brain regions to obtain effective connections. However, simultaneously applying stimulation and observation across the entire brain is technically impractical. Therefore, many studies attempt to infer effective connections from neural data, such as functional magnetic resonance imaging (fMRI) and electroencephalography (EEG) data.

[0058] Data-driven approaches include model-based approaches and model-free approaches.

[0059] Model-based methods typically parameterize effective connections within a generative model, fitting neural signals to estimate these connections based on model assumptions. For example, dynamic causal models use biophysical models where effective connections are treated as parameters in neurodynamics, converting generated neural signals into blood oxygenation level-dependent signals via hemodynamic functions to fit functional magnetic resonance imaging (fMRI) data. However, while dynamic causal models have been widely used to estimate effective connections between a few regions, their high computational complexity in parameter estimation makes them unsuitable for studying whole-brain effective connections that require considering signals from all brain regions. Another limitation of such model-based methods is their reliance on model assumptions, which can lead to significant inference bias if the model does not match actual brain dynamics.

[0060] Model-free methods do not rely on explicit assumptions about underlying neural dynamics, using statistical methods to infer effective connections from observed neural signals. For example, Granger causality uses time-series analysis to determine the extent to which past neural activity in one region predicts future neural activity in another region. This method can be used to infer the strength and orientation of effective connections, but it cannot distinguish between excitatory and inhibitory connections. Furthermore, the effectiveness of Granger causality analysis is controversial due to conceptual and performance limitations.

[0061] Besides statistical methods, several deep learning-based causal inference models have been proposed in recent years. These neural network-based causal inference models typically represent brain networks using graph structures. Utilizing a data-driven approach, they treat effective connections in neural signals as implicit tasks, using explicit tasks of neural signal prediction or classification to drive the learning of connections between different brain regions in the implicit tasks. These data-driven causal inference neural network models have achieved excellent performance in several fields, such as trajectory prediction in physical systems, and have great application potential for inferring effective connections in the brain. However, most of these methods can only distinguish the existence of directed connections, but cannot characterize the weights and signs of connections. The resulting connections show little inter-individual variation, limiting their application in target selection for precise neuromodulation.

[0062] Therefore, common experimental methods, such as perturbing brain signals in one region through electrical or magnetic stimulation and observing how this perturbation propagates to other brain regions to obtain effective connectivity, are not suitable for stimulation and observation across the entire human brain. Furthermore, common data-driven methods for inferring effective connectivity suffer from high computational complexity in model-based approaches, while model-free methods can only distinguish the existence of directed connections.

[0063] In other words, neither common experimental methods nor data-driven approaches can obtain accurate information about the brain's effective connectivity.

[0064] To address the aforementioned issues, in the embodiments of this application, on the one hand, a time-series prediction network can be used to learn and predict the dynamic changes in brain neural signals to obtain a digital twin brain model with time-series prediction capabilities; on the other hand, based on the trained digital twin brain model, perturbation data is added to brain nodes to predict the brain's response to perturbation stimuli and infer effective brain connectivity information. Therefore, in the embodiments of this application, the digital twin brain model is obtained based on training with a time-series prediction network. Thus, the digital twin brain model can effectively and accurately predict the dynamic changes in brain neural data, thereby obtaining more accurate information on effective brain connectivity.

[0065] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0066] One embodiment of this application provides a method for deriving effective brain connectivity information. This method can be applied to a device for deriving effective brain connectivity information or a computer device; this application does not impose specific limitations. Below, using a device for deriving effective brain connectivity information as an example, the method for deriving effective brain connectivity information proposed in this application will be described exemplarily.

[0067] Furthermore, in the embodiments of the application, Figure 1 is a schematic diagram of the implementation process of the method for deriving effective brain connectivity information proposed in this application. As shown in Figure 1, the method for deriving effective brain connectivity information may include the following steps:

[0068] Step 101: Determine the perturbation data of the first brain node in at least two brain nodes and the brain neural data of at least two brain nodes in the time series.

[0069] In the embodiments of this application, the device for deducing effective brain connectivity information can first determine the perturbation data of the first brain node among at least two brain nodes, and can also determine the brain neural data of at least two brain nodes in the time series.

[0070] It is understood that the method for deriving effective brain connectivity information proposed in this application can determine the effective brain connectivity information between different brain nodes. A brain node can represent a brain region; that is, one brain node can be considered to correspond to one brain region. The brain can be divided into at least two brain regions, and correspondingly, one brain can correspond to at least two brain nodes.

[0071] For example, in some embodiments, Figure 2 is a schematic diagram of the brain nodes proposed in the embodiments of this application. As shown in Figure 2, assuming that the biological brain is divided into multiple regions, such as region A, region B, region C, and region D, it can be considered that the brain has at least brain node a, brain node b, brain node c, and brain node d.

[0072] It is understood that, in the embodiments of this application, during the derivation of effective brain connectivity information, the brain neural data corresponding to the time series of each of at least two brain nodes can be obtained first. That is, the device for derivation of effective brain connectivity information needs to first obtain brain neural data at multiple time points corresponding to each brain node.

[0073] In the embodiments of this application, the time series may include continuous time sampling intervals or discrete time sampling points; this application does not impose any specific limitations.

[0074] Furthermore, in embodiments of this application, the brain neural data of at least two brain nodes within a time series may include brain neural data of at least two brain nodes corresponding to a preset time length p+1. When predicting brain neural data at the next moment, the preset time length can be determined by specifying the time length of the brain neural data used in the prediction process. Here, p is an integer greater than or equal to 0.

[0075] It is understood that, in the embodiments of this application, the time length of the brain neural data used in the prediction process, i.e., the preset time length p+1, can be pre-set. For example, if p=2 is preset, then the value of the preset time length p+1 is 3.

[0076] It is understood that, in the embodiments of this application, the brain neural data of at least two brain nodes in the time series may include brain neural data of at least two brain nodes corresponding to multiple time points. The brain neural data of these multiple time points includes, but is not limited to, brain neural data of at least two brain nodes corresponding to a preset time length p+1, that is, the time series length corresponding to the time series is greater than or equal to the time length of the brain neural data used in the prediction process.

[0077] For example, in some embodiments, the time series corresponds to a time series length of 4800, while the time length of the brain neural data used in the prediction process is 3.

[0078] It is understood that in the embodiments of this application, for a given preset time length, the time length of the data used for predicting brain neural data is fixed. For example, for a preset time length p+1, brain neural data from the previous p+1 moments can be used to predict brain neural data from the next moment.

[0079] In other words, in the embodiments of this application, brain neural data from one or more time points can be used to predict brain neural data for the next time point. For example, if p=2 is preset, then brain neural data from three consecutive time points (time 1, time 2, and time 3) can be used to predict brain neural data for time 4, brain neural data from three consecutive time points (time 2, time 3, and time 4) can be used to predict brain neural data for time 5, and brain neural data from three consecutive time points (time 3, time 4, and time 5) can be used to predict brain neural data for time 6.

[0080] For example, in some embodiments, assuming a preset time length of p+1, the time length indicated by the preset time length includes p+1 times from time tp to time t. Accordingly, the brain neural data corresponding to the preset time length p+1 includes brain neural data from time tp to time t, where t is an integer greater than p.

[0081] In other words, in some embodiments, the time series may include p+1 consecutive time points from time tp to time t. Accordingly, the brain neural data of each brain node in the time series may include p+1 brain neural data points corresponding to that brain node at time point p+1.

[0082] Furthermore, in the embodiments of this application, various methods can be used to determine the brain neural data of any brain node in the time series, and this application does not impose specific limitations.

[0083] For example, in some embodiments, brain neural data can be acquired using existing bioinformatics databases, which include, but are not limited to, one or more public datasets such as human fMRI datasets, EEG datasets, and MEG datasets.

[0084] For example, in some embodiments, brain neural data can also be generated using computational models.

[0085] Furthermore, in embodiments of this application, the first brain node can be any one of at least two brain nodes. For example, the first brain node can be brain node a, or it can be brain node b.

[0086] In the embodiments of this application, during the derivation of effective brain connectivity information, it is also necessary to apply certain perturbation data to the first brain node, so as to use the changes in the activity of other brain nodes besides the first brain node after the perturbation data is added to characterize the effective connectivity of the first brain node to other brain nodes.

[0087] It is understood that, in the embodiments of this application, the perturbation data of the first brain node can be understood as a virtual perturbation corresponding to the first brain node. This application does not specifically limit the form in which the perturbation data of the first brain node is represented. For example, an N-dimensional virtual perturbation vector corresponding to N brain nodes can be pre-constructed, where N is an integer greater than 1. The perturbation data of the first brain node can be an element corresponding to this N-dimensional virtual perturbation vector, with the value being a non-zero value representing the intensity of the virtual perturbation added to the first brain node. Other elements have values ​​of 0, meaning that other brain nodes besides the first brain node do not experience any virtual perturbation.

[0088] Furthermore, in the embodiments of this application, when determining the perturbation data of the first brain node among at least two brain nodes, the perturbation data of the first brain node can be set based on a preset perturbation intensity.

[0089] For example, in some embodiments, for an N-dimensional virtual perturbation vector P corresponding to N brain nodes, it is assumed that each element p in the N-dimensional virtual perturbation vector P corresponds to a brain node. If the i-th brain node is the first brain node to which the virtual perturbation is applied, then when determining the perturbation data of the first brain node, the i-th element p corresponding to the first brain node in the N-dimensional virtual perturbation vector P can be selected. i Set to Δ, where Δ can be a preset perturbation strength; simultaneously, the i-th element p in the N-dimensional virtual perturbation vector P can be... i All other elements are set to 0. For example, in formula (1) below: p i =Δ,p k,k≠i =0 (1);

[0090] Step 102: Using a digital twin brain model, based on the perturbation data of the first brain node and the brain neural data of at least two brain nodes in the time series, obtain the first predicted neural data of at least two brain nodes at the next time step; wherein, the digital twin brain model is obtained by training a time series prediction network.

[0091] In the embodiments of this application, after determining the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series, the first predicted neural data of the at least two brain nodes at the next moment can be further determined by the digital twin brain model based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series. That is, in the case that there is a virtual perturbation in the first brain node, the brain neural data of other brain nodes can be predicted by the digital twin brain model.

[0092] Furthermore, in the embodiments of this application, the digital twin brain model can be obtained by training a time-series prediction network. Accordingly, the digital twin brain model can be used to determine the effective connectivity information of the brain based on the perturbation data of brain nodes and the brain neural data of brain nodes in the time series.

[0093] In other words, in the embodiments of this application, the digital twin brain model is obtained based on time series prediction network training. Therefore, the digital twin brain model can effectively predict the dynamic changes of brain neural signals (brain neural data), thereby obtaining more accurate information on effective brain connectivity.

[0094] Furthermore, in the embodiments of this application, the time series prediction network used to train the digital twin brain model can be any type and structure of artificial neural network for time series prediction, and this application does not impose any specific limitations.

[0095] For example, in some embodiments, the time series prediction network can use classic models such as multilayer perceptron, convolutional neural network, and recurrent neural network, or other network architectures suitable for time series prediction can be selected. This application does not make specific limitations.

[0096] For example, in some embodiments, for a time series of p+1 consecutive time periods from time tp to time t, when obtaining the first predicted neural data of at least two brain nodes at the next time period using a digital twin brain model based on the perturbation data of the first brain node and the brain neural data of at least two brain nodes in the time series, the perturbation-reset brain neural data of at least two brain nodes at time t can be determined first based on the perturbation data of the first brain node and the brain neural data of at least two brain nodes at time t; then, the first predicted neural data of at least two brain nodes at time t+1 can be obtained further using a digital twin brain model based on the perturbation-reset brain neural data of at least two brain nodes at time t and the brain neural data of at least two brain nodes from time tp to time t-1.

[0097] It is understood that, in the embodiments of this application, during the process of dynamically predicting brain neural data through a digital twin brain model, the perturbation data with added virtual perturbation corresponding to the current time can be generated based on the perturbation data of the first brain node corresponding to the current time (e.g., time t) and the brain neural data of all brain nodes corresponding to the current time.

[0098] Furthermore, in the embodiments of this application, based on the perturbation data of the first brain node and the brain neural data of at least two brain nodes at time t, the perturbation-reduced brain neural data at time t can be generated in various ways. For example, it includes, but is not limited to, performing mathematical operations on the perturbation data of the first brain node and the brain neural data of at least two brain nodes at time t to obtain the perturbation-reduced brain neural data at time t.

[0099] For example, in some embodiments, when determining the perturbed brain neural data of at least two brain nodes at time t based on the perturbation data of the first brain node and the brain neural data of the first brain node at time t, it is possible to superimpose the perturbation data of the first brain node and the brain neural data of the first brain node at time t, and then combine it with the brain neural data of other brain nodes at time t to obtain the perturbed brain neural data of at least two brain nodes at time t.

[0100] It is understood that, in the embodiments of this application, the perturbed brain neural data of at least two brain nodes at time t and the brain neural data of at least two brain nodes from time tp to time t-1 can be used as the perturbed brain neural data of the digital twin brain model to predict the brain neural data at the next time step, and finally the first predicted neural data of at least two brain nodes at time t+1 can be output. Here, the perturbed brain neural data of at least two brain nodes at time t is generated based on the perturbed data of the first brain node; therefore, it can be considered that the digital twin brain model predicts brain neural data with the addition of virtual perturbation.

[0101] In other words, in the embodiments of this application, the first predicted neural data of at least two brain nodes at time t+1 is the brain neural data predicted after adding virtual perturbation.

[0102] For example, in some embodiments, first predicted neural data of at least two brain nodes at time t+1 are determined using a digital twin brain model. The process can be referred to the following formula (2):

[0103] The digital twin brain model can be represented as f(·), where P is the perturbation data, which can be an N-dimensional virtual perturbation vector, and x (t-p) ,x (t-p+1) ,…,x t It can represent brain neural data from time point (tp) to time point t.

[0104] For example, in some embodiments, first predicted neural data of at least two brain nodes at time t+1 are determined using a digital twin brain model. The process can also refer to the following formula (3):

[0105] Where θ represents all parameters in the digital twin brain model that can be optimized through training iterations.

[0106] Furthermore, in the embodiments of this application, Figure 3 is a schematic diagram of the second implementation flow of the method for deriving effective brain connectivity information proposed in the embodiments of this application. As shown in Figure 3, the method for deriving effective brain connectivity information may further include the following steps:

[0107] Step 104: Using a digital twin brain model, based on the brain neural data of at least two brain nodes in the time series, obtain the second predicted neural data of at least two brain nodes at the next time step.

[0108] In the embodiments of this application, after determining the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series, the second predicted neural data of the at least two brain nodes at the next moment can be further determined by the digital twin brain model based on the brain neural data of the at least two brain nodes in the time series. That is, in the absence of virtual perturbation in the first brain node, the brain neural data of other brain nodes can be predicted by the digital twin brain model.

[0109] For example, in some embodiments, for a time series of p+1 consecutive time periods from time tp to time t, when obtaining the first predicted neural data of at least two brain nodes at the next time period based on the brain neural data of at least two brain nodes in the time series using a digital twin brain model, the brain neural data of at least two brain nodes from time tp to time t can be used as input data of the digital twin brain model to predict the brain neural data at the next time period, and finally the second predicted neural data of at least two brain nodes at time t+1 can be output.

[0110] It is understood that, in the embodiments of this application, compared with the first predicted neural data, the second predicted neural data of at least two brain nodes at time t+1 can be understood as the predicted brain neural data without added virtual perturbation.

[0111] In other words, in the embodiments of this application, the first predictive neural data can characterize the predictive data obtained when perturbation data is added, while the second predictive neural data characterizes the predictive data obtained when no perturbation data is added.

[0112] For example, in some embodiments, second predicted neural data of at least two brain nodes at time t+1 are determined using a digital twin brain model. The process can be referred to the following formula (4):

[0113] The digital twin brain model can be represented as f(·), x (t-p) ,x (t-p+1) ,…,xt It can represent brain neural data from time point (tp) to time point t.

[0114] For example, in some embodiments, second predicted neural data of at least two brain nodes at time t+1 are determined using a digital twin brain model. The process can also refer to the following formula (5):

[0115] Where θ represents all parameters in the digital twin brain model that can be optimized through training iterations.

[0116] Step 103: Based on the first predicted neural data of at least two brain nodes at the next time step and the second predicted neural data of at least two brain nodes at the next time step, determine the effective brain connection information from the first brain node to the second brain node; wherein, the second brain node is the other brain node among the at least two brain nodes besides the first brain node; the second predicted neural data represents the predicted data obtained without adding perturbation data.

[0117] In the embodiments of this application, after predicting and obtaining the first and second predicted neural data at the next time step by using a digital twin brain model with and without virtual perturbation, the effective brain connection information from the first brain node to the second brain node can be further determined based on the first predicted neural data of at least two brain nodes at the next time step and the second predicted neural data of at least two brain nodes at the next time step.

[0118] It is understood that, in the embodiments of this application, the second brain node can be any of the at least two brain nodes other than the first brain node. For example, the first brain node is brain node a, and the second brain node can be one or more of brain nodes b, c, and d.

[0119] In the embodiments of this application, the effective brain connection information from the first brain node to the second brain node can be used to determine the corresponding changes produced by the second brain node after a virtual perturbation is applied to the first brain node.

[0120] It is understood that, in the embodiments of this application, the first predictive neural data represents the predictive data obtained when perturbation data is added, and the second predictive neural data represents the predictive data obtained when no perturbation data is added. Therefore, based on the first and second predictive neural data, the effective brain connectivity information between the first brain node and any other brain node can be determined.

[0121] Furthermore, in the embodiments of this application, the effective brain connection information from the first brain node to the second brain node determined based on the first predicted neural data and the second predicted neural data may include, but is not limited to, the effective connection type, the effective connection direction, and the effective connection strength.

[0122] It is understood that, in the embodiments of this application, the effective connection type can determine and distinguish the nature of the influence of effective connections between two brain nodes. The effective connection type can be, but is not limited to, excitatory and inhibitory types.

[0123] It is understood that, in the embodiments of this application, the effective connection direction can be determined as the transmission direction of the effective connection between two brain nodes.

[0124] It is understood that, in the embodiments of this application, the effective connection strength can be determined as the signal strength of the effective connection between two brain nodes.

[0125] Therefore, in the embodiments of this application, the effective brain connection information between two brain nodes obtained based on the digital twin brain model can not only clarify the causal interactions in the brain, but also provide effective connection data with directionality, intensity, and excitation / inhibition characteristics.

[0126] Furthermore, in the embodiments of this application, when determining the effective brain connectivity information from the first brain node to the second brain node based on the first predicted neural data and the second predicted neural data of at least two brain nodes at the next time step, the effective connectivity type can be determined to be excitatory if the first predicted neural data of the second brain node at the next time step is greater than the second predicted neural data of the second brain node at the next time step; and if the first predicted neural data of the second brain node at the next time step is less than the second predicted neural data of the second brain node at the next time step. That is, in the embodiments of this application, by comparing the first and second predicted neural data obtained with and without virtual perturbation, the corresponding effective connectivity type can be further determined based on the comparison result. Specifically, if the first predicted neural data is greater than the second predicted neural data, it can be considered that the neural activity of the second brain node has increased when perturbation data exists for the first brain node, and the corresponding effective connectivity type can be determined to be excitatory; if the first predicted neural data is less than the second predicted neural data, it can be considered that the neural activity of the second brain node has decreased when perturbation data exists for the first brain node, and the corresponding effective connectivity type can be determined to be inhibitory.

[0127] Therefore, in the embodiments of this application, the effective connection type can reflect the increase or decrease of neural activity corresponding to other brain nodes after adding virtual perturbation to the first brain node, and the resulting effects are respectively excitatory or inhibitory.

[0128] For example, in some embodiments, for a time series of p+1 consecutive time points from time tp to time t, when determining the effective brain connectivity information from the first brain node to the second brain node based on the first predicted neural data and the second predicted neural data of at least two brain nodes at the next time point, the effective connectivity type can be determined to be excitatory if the first predicted neural data of the second brain node at time t+1 is greater than the second predicted neural data of the second brain node at time t+1; and if the first predicted neural data of the second brain node at time t+1 is less than the second predicted neural data of the second brain node at time t+1, the effective connectivity type can be determined to be inhibitory. In other words, in the embodiments of this application, perturbation data is added to the first brain node at time t, and the first predicted neural data of the second brain node at time t+1 is predicted by combining the brain neural data of the first and second brain nodes from time tp to time t. At the same time, the second predicted neural data of the second brain node at time t+1 is directly predicted based on the brain neural data of the first and second brain nodes from time tp to time t. If the first predicted neural data of the second brain node at time t+1 is greater than the second predicted neural data, then it can be considered that the neural activity corresponding to the second brain node has increased, that is, the perturbation data has an excitatory effect on the second brain node, so the effective connection type can be determined to be excitatory. Conversely, if the first predicted neural data of the second brain node at time t+1 is smaller than the second predicted neural data, then it can be considered that the neural activity corresponding to the second brain node has decreased, that is, the perturbation data has an inhibitory effect on the second brain node, so the effective connection type can be determined to be inhibitory.

[0129] Furthermore, in the embodiments of this application, when determining the effective brain connection information from the first brain node to the second brain node, the effective connection direction can be directly determined as from the first brain node to the second brain node.

[0130] It is understood that, in the embodiments of this application, when monitoring the neural data of other brain nodes under the condition of applying virtual perturbation to the first brain node, it can be considered that the effective connection direction corresponding to the effective brain connection information between the first brain node and the second brain node is from the first brain node to the second brain node.

[0131] Furthermore, in the embodiments of this application, when determining the effective brain connection information from the first brain node to the second brain node based on the first predicted neural data of at least two brain nodes at the next time step and the second predicted neural data of at least two brain nodes at the next time step, the strength information of the first predicted neural data of the second brain node at time step t+1 can also be determined, and the strength information of the second predicted neural data of the second brain node at time step t+1 can also be determined; then, based on the strength information of the first predicted neural data and the strength information of the second predicted neural data, the effective connection strength is determined.

[0132] Understandably, in the embodiments of this application, perturbation data is added to the first brain node at time t, and combined with the brain neural data of the first and second brain nodes from time tp to time t, the first predicted neural data of the second brain node at time t+1 is predicted; simultaneously, the second predicted neural data of the second brain node at time t+1 is directly predicted based on the brain neural data of the first and second brain nodes from time tp to time t; the strength information of the first and second predicted neural data is determined respectively, and then the effective connection strength is further determined based on the strength information of these two predicted neural data. For example, a difference operation can be performed on the strength information of the first and second predicted neural data, and the effective connection strength can be determined based on the difference result.

[0133] Furthermore, in the embodiments of this application, after determining the strength information of the first predicted neural data and the strength information of the second predicted neural data respectively, and comparing the two, if the comparison result is that the strength information of the first predicted neural data and the strength information of the second predicted neural data are equal, it can be considered that no matter whether perturbation data is added to the first brain node, the predicted neural data corresponding to the second brain node will not change, that is, the perturbation data added to the first brain node will not affect the second brain node. At this time, it can be considered that there is no effective brain connection information between the first brain node and the second brain node.

[0134] For example, in some embodiments, assuming brain node i is the first brain node and brain node j is the second brain node, then, based on the above formulas (3) and (5), the effective brain connection information EC between the two brain nodes is determined by the digital twin brain model. ij The process can be described by the following formula (6): EC ij =f(x) (t-p) ,x (t-p+1) ,…,x t +P,θ) j -f(x (t-p) ,x (t-p+1) ,…,xt ,θ) j (6);

[0135] For example, in some embodiments, assuming brain node i is the first brain node and brain node j is the second brain node, then, based on the above formulas (3) and (5), the effective brain connection information EC between the two brain nodes is determined by the digital twin brain model. ij The process can also be described by the following formula (7): EC ij =E t [f(x (t-p) ,x (t-p+1) ,…,x t +P,θ) j -f(x (t-p) ,x (t-p+1) ,…,x t ,θ) j (7);

[0136] Among them, E t (·) indicates the mean value over time t.

[0137] Furthermore, in the embodiments of this application, each of the at least two brain nodes can be traversed, thereby determining the effective brain connection information between any two of the at least two brain nodes, and thus obtaining the whole-brain effective connection of the at least two brain nodes.

[0138] In other words, in the embodiments of this application, a virtual perturbation can be applied to each brain node, that is, perturbation data can be added as input information. Then, the brain neural data of other brain nodes can be predicted by the digital twin brain model. Combined with the brain neural data of other brain nodes predicted without applying virtual perturbation, the effective brain connection information between any two brain nodes can be further determined.

[0139] It is understood that, in the embodiments of this application, after determining the effective brain connectivity information between every two brain nodes, the determination of the effective connectivity of the whole brain can be further completed.

[0140] For example, in some embodiments, Figure 4 is a schematic diagram illustrating the derivation of effective brain connectivity information proposed in this application. As shown in Figure 4, assuming brain node a is the first brain node, after applying a virtual perturbation to brain node a, the brain neural data of brain node a at time t is superimposed with the corresponding perturbation data. Simultaneously, the brain neural data of brain nodes b, c, and d at time t are combined to obtain the perturbed brain neural data of these four brain nodes at time t, which serves as the input to the digital twin brain model. Furthermore, the brain neural data of these four brain nodes from time tp to time t-1 can also be used as the input to the digital twin brain model. Through the digital twin brain model, the brain neural data of brain nodes a, b, c, and d at the next time step can be predicted. That is, the first predicted neural data obtained can include the first predicted neural data of brain nodes a, b, c, and d at time t+1. Simultaneously, the neural data of these four brain nodes from time tp to time t can be used as input to the digital twin brain model. The digital twin brain model then predicts and obtains second predictive neural data. This second predictive neural data can include the second predictive neural data of brain nodes a, b, c, and d at time t+1. By comparing the first and second predictive neural data of brain nodes b, c, and d at time t+1, it can be determined that the neural data of brain nodes b and c remain unchanged, while the neural data of brain node d shows an increase. Therefore, the effective brain connection information from brain node a to brain node d can be determined to include: the effective connection direction is from brain node a to brain node d, the effective connection type is excitatory, and the effective connection strength is the change in the intensity of the neural data of brain node d.

[0141] For example, in some embodiments, Figure 5 is a schematic diagram of the derivation of effective brain connectivity information proposed in the embodiments of this application. As shown in Figure 5, assuming brain node b is the first brain node, after applying a virtual perturbation to brain node b, the brain neural data of brain node b at time t is superimposed with the corresponding perturbation data. At the same time, the brain neural data of brain nodes a, c, and d at time t are combined to obtain the perturbed brain neural data of these four brain nodes at time t, which is used as the input of the digital twin brain model. Simultaneously, the brain neural data of these four brain nodes from time tp to time t-1 can also be used as the input of the digital twin brain model. Through the digital twin brain model, the brain neural data of brain nodes a, b, c, and d at the next time step can be predicted. That is, the first predicted neural data obtained can include the first predicted neural data of brain nodes a, b, c, and d at time t+1. Simultaneously, the neural data of these four brain nodes from time tp to time t can be used as input to the digital twin brain model. The digital twin brain model then predicts and obtains second predictive neural data. This second predictive neural data can include the second predictive neural data of brain nodes a, b, c, and d at time t+1. Comparing the first and second predictive neural data of brain nodes a, c, and d at time t+1, it can be determined that the neural data of brain node d remains unchanged, while the neural data of brain node a weakens, and the neural data of brain node c strengthens. Therefore, the effective connection information from brain node b to brain node a can be determined as follows: the effective connection direction is from brain node b to brain node a, the effective connection type is inhibitory, and the effective connection strength is the intensity change value of the neural data of brain node a. Similarly, the effective connection information from brain node b to brain node c can be determined as follows: the effective connection direction is from brain node b to brain node c, the effective connection type is excitatory, and the effective connection strength is the intensity change value of the neural data of brain node c.

[0142] For example, in some embodiments, Figure 6 is a schematic diagram three illustrating the derivation of effective brain connectivity information proposed in the embodiments of this application. As shown in Figure 6, assuming brain node c is the first brain node, after applying a virtual perturbation to brain node c, the brain neural data of brain node c at time t is superimposed with the corresponding perturbation data. Simultaneously, the brain neural data of brain nodes b, a, and d at time t are combined to obtain the perturbed brain neural data of these four brain nodes at time t, which serves as the input to the digital twin brain model. Furthermore, the brain neural data of these four brain nodes from time tp to time t-1 can also be used as the input to the digital twin brain model. Through the digital twin brain model, the brain neural data of brain nodes a, b, c, and d at the next time step can be predicted. That is, the first predicted neural data obtained can include the first predicted neural data of brain nodes a, b, c, and d at time t+1. Simultaneously, the neural data of these four brain nodes from time tp to time t can be used as input to the digital twin brain model. The digital twin brain model then predicts and obtains second predictive neural data. This second predictive neural data can include the second predictive neural data of brain nodes a, b, c, and d at time t+1. By comparing the first and second predictive neural data of brain nodes b, a, and d at time t+1, it can be determined that the neural data of brain nodes b and d remain unchanged, while the neural data of brain node a weakens. Therefore, the effective connection information from brain node c to brain node a can be determined as follows: the effective connection direction is from brain node c to brain node a, the effective connection type is inhibitory, and the effective connection strength is the change in the intensity of the neural data of brain node a.

[0143] For example, in some embodiments, Figure 7 is a schematic diagram of the derivation of effective brain connectivity information proposed in the embodiments of this application. As shown in Figure 7, assuming brain node d is the first brain node, after applying a virtual perturbation to brain node d, the brain neural data of brain node d at time t is superimposed with the corresponding perturbation data. At the same time, the brain neural data of brain nodes b, a, and c at time t are combined to obtain the perturbed brain neural data of these four brain nodes at time t, which is used as the input of the digital twin brain model. Simultaneously, the brain neural data of these four brain nodes from time tp to time t-1 can also be used as the input of the digital twin brain model. Through the digital twin brain model, the brain neural data of brain nodes a, b, c, and d at the next time step can be predicted. That is, the first predicted neural data obtained can include the first predicted neural data of brain nodes a, b, c, and d at time t+1. Simultaneously, the neural data of these four brain nodes from time tp to time t can be used as input to the digital twin brain model. The digital twin brain model then predicts and obtains second predictive neural data. This second predictive neural data can include the second predictive neural data of brain nodes a, b, c, and d at time t+1. By comparing the first and second predictive neural data of brain nodes b, a, and c at time t+1, it can be determined that the neural data of brain nodes b and a remain unchanged, while the neural data of brain node c weakens. Therefore, the effective connection information from brain node d to brain node c includes: the effective connection direction is from brain node d to brain node c, the effective connection type is inhibitory, and the effective connection strength is the change in the intensity of the neural data of brain node c.

[0144] For example, in some embodiments, Figure 8 is a schematic diagram of the derivation of brain effective connection information proposed in the embodiments of this application. As shown in Figure 8, after traversing brain node a, brain node b, brain node c and brain node d according to the above method, the brain effective connection information between every two brain nodes under the condition of applying virtual perturbation to different brain nodes can be determined, thereby completing the determination of the whole brain effective connection. The whole brain effective connection may include the effective connection type, effective connection direction and effective connection strength between different brain nodes.

[0145] Furthermore, in the embodiments of this application, since the length of the time series corresponding to the time series is greater than or equal to the length of the brain neural data used in the prediction process, it is possible to select to predict the effective brain connectivity information at multiple time points through a digital twin brain model, and to calculate the average value based on the multiple effective brain connectivity information obtained from the prediction to determine the effective connectivity in an average sense. The effective connectivity in an average sense can more accurately and comprehensively reflect the activity between brain nodes.

[0146] Furthermore, in the embodiments of this application, a digital twin brain model can be used to first obtain multiple first predicted neural data for at least two brain nodes corresponding to multiple times, based on multiple perturbation data of the first brain node corresponding to multiple times and brain neural data of at least two brain nodes in the time series. Simultaneously, a digital twin brain model can be used to obtain multiple second predicted neural data for at least two brain nodes corresponding to multiple times, based on brain neural data of at least two brain nodes in the time series. Then, based on the multiple first predicted neural data and the multiple second predicted neural data, multiple effective brain connection information from the first brain node to the second brain node corresponding to multiple times can be determined. Finally, based on the multiple effective brain connection information, the target effective connection information from the first brain node to the second brain node can be determined.

[0147] It is understood that, in the embodiments of this application, the first predicted neural data is brain neural data predicted after adding virtual perturbation, and the second predicted neural data is brain neural data predicted without adding virtual perturbation.

[0148] It is understood that, in the embodiments of this application, the mean value is calculated based on the effective connection information of multiple brain nodes from the first brain node to the second brain node corresponding to multiple time points. The final mean value is the target effective connection information from the first brain node to the second brain node. This target effective connection information can characterize the effective connection from the first brain node to the second brain node in an average sense.

[0149] In other words, in the embodiments of this application, for a sufficiently large time series length, brain neural data can be predicted for multiple optional parameters t based on the brain neural data within the time series. This involves predicting and obtaining multiple predicted neural data sets, including multiple first predicted neural data sets obtained with added perturbation data at multiple time points and multiple second predicted neural data sets obtained without added perturbation data. Finally, the effective connections in an average sense are determined using the multiple first predicted neural data sets and multiple second predicted neural data sets corresponding to multiple time points. For a single time point t, the effective connections obtained at that given t can be considered to depend on the current state. For multiple time points t, the effective connections are calculated and averaged across all optional t values ​​to obtain the effective connections in an average sense (target effective connection information).

[0150] For example, in some embodiments, it is assumed that the digital twin brain is a 3-step prediction of a 1-step process (using brain neural data from three consecutive time points to predict brain neural data from the next time point), i.e., p=2, with a preset time length of 3. The brain neural data in the time series includes brain neural data from time 1 to time 5, i.e., the time series length is 5. Then, based on the brain neural data at times 1, 2, and 3; times 2, 3, and 4; and times 3, 4, and 5, the effective brain connectivity information at times 4, 5, and 6 can be obtained respectively. The results obtained depend on the brain state represented by different time points, and the results obtained at different time points are not exactly the same. Finally, the effective brain connectivity information at times 4, 5, and 6 can be further averaged, and the average result can be used as the effective connectivity in an average sense, i.e., the target effective connectivity information.

[0151] In summary, the method for deriving effective brain connectivity information proposed in steps 101 to 104 above can be used to stimulate a digital twin brain model with different virtual stimuli. By virtually perturbing the digital twin brain model, the brain's response to stimulation can be predicted and the effective connectivity of the whole brain can be inferred. This method can consider a large number of brain regions at the same time, is suitable for inferring effective connectivity of the whole brain, and can simultaneously characterize the symbol (type of effective connectivity), weight (strength of effective connectivity), and direction (direction of effective connectivity) of the connectivity.

[0152] This application provides a method for deriving effective brain connectivity information. The device for deriving effective brain connectivity information determines the perturbation data of a first brain node and the brain neural data of at least two brain nodes in a time series. Using a digital twin brain model, based on the perturbation data of the first brain node and the brain neural data of at least two brain nodes in the time series, it obtains first predicted neural data of at least two brain nodes at the next time step. The digital twin brain model is obtained by training a time series prediction network. Based on the first predicted neural data of at least two brain nodes at the next time step and the second predicted neural data of at least two brain nodes at the next time step, it determines the effective brain connectivity information from the first brain node to the second brain node. The second brain node is any brain node other than the first brain node among the at least two brain nodes. The second predicted neural data represents the predicted data obtained without adding perturbation data. In other words, in the embodiments of this application, based on the digital twin brain model obtained through training, the brain's response to perturbation stimulation is predicted and the effective connection information of the brain is inferred by adding perturbation data to the brain nodes. The digital twin brain model can effectively and accurately predict the dynamic changes of brain neural data, thereby obtaining more accurate information on the effective connection of the brain.

[0153] Another embodiment of this application provides a training method for a digital twin brain model. This training method can be applied to a training device for a digital twin brain model or a computer device; this application does not impose specific limitations. Below, using a training device for a digital twin brain model as an example, the training method for a digital twin brain model proposed in this application will be described exemplarily.

[0154] Furthermore, in the embodiments of the application, Figure 9 is a schematic diagram of the implementation process of training the digital twin brain model proposed in the embodiments of this application. As shown in Figure 9, the training method of the digital twin brain model may include the following steps:

[0155] Step 901: Train the time series prediction network based on the training dataset to obtain a digital twin brain model; wherein, the digital twin brain model is used to predict the neural data of the next moment based on the brain neural data of the brain nodes in the time series.

[0156] In embodiments of this application, the training device for the digital twin brain model can use neural training data including at least two brain nodes within a time series to train a time series prediction network, thereby obtaining a digital twin brain model. The digital twin brain model obtained based on the time series prediction network training can predict the dynamic changes in neural data; that is, it can predict the neural data at the next moment based on the brain neural data of the brain nodes within the time series. Furthermore, based on the obtained predicted neural data, it can determine the effective brain connectivity information between brain nodes, thereby obtaining more accurate effective brain connectivity information.

[0157] It is understood that, in the embodiments of this application, a brain node can represent a brain region, that is, one brain node can be considered to correspond to one brain region. The brain can be divided into at least two brain regions, and correspondingly, one brain can correspond to at least two brain nodes.

[0158] For example, in some embodiments, Figure 10 is a schematic diagram of the brain node proposed in the embodiments of this application. As shown in Figure 10, assuming that the biological brain is divided into multiple regions, such as region A, region B and region C, it can be considered that the brain has at least brain node a, brain node b and brain node c.

[0159] Furthermore, in the embodiments of this application, the neural training data of at least two brain nodes in the time series may include the neural training data of each brain node corresponding to the time series. That is, the training device of the digital twin brain model needs to first acquire the neural training data of multiple moments corresponding to each brain node.

[0160] In the embodiments of this application, the time series may include continuous time sampling intervals or discrete time sampling points; this application does not impose any specific limitations.

[0161] Furthermore, in embodiments of this application, the training dataset may include brain neural data corresponding to at least two brain nodes with a preset time length p+1; wherein, when predicting brain neural data at the next moment, the preset time length may be determined by specifying the time length of the brain neural data used in the prediction process. Here, p is an integer greater than or equal to 0.

[0162] It is understood that, in the embodiments of this application, the time length of the brain neural data used in the prediction process, i.e., the preset time length p+1, can be pre-set. For example, if p=2 is preset, then the value of the preset time length p+1 is 3.

[0163] It is understood that, in the embodiments of this application, the training dataset may include brain neural data at least two brain nodes corresponding to multiple time points. These multiple time points of brain neural data include, but are not limited to, brain neural data at least two brain nodes corresponding to a preset time length p+1. That is, the time series length corresponding to the training dataset is greater than or equal to the time length of the brain neural data used in the prediction process.

[0164] For example, in some embodiments, the time series length corresponding to the training dataset is 5000, while the time series length of the brain neural data used in the prediction process is 3.

[0165] It is understood that in the embodiments of this application, for a given preset time length, the time length of the data used for predicting brain neural data is fixed. For example, for a preset time length p+1, brain neural data from the previous p+1 moments can be used to predict brain neural data from the next moment.

[0166] In other words, in the embodiments of this application, brain neural data from one or more time points can be used to predict brain neural data for the next time point. For example, if p=2 is preset, then brain neural data from three consecutive time points (time 1, time 2, and time 3) can be used to predict brain neural data for time 4, brain neural data from three consecutive time points (time 2, time 3, and time 4) can be used to predict brain neural data for time 5, and brain neural data from three consecutive time points (time 3, time 4, and time 5) can be used to predict brain neural data for time 6.

[0167] For example, in some embodiments, assuming a preset time length of p+1, the time length indicated by the preset time length includes p+1 times from time mp to time m. Accordingly, the neural training data corresponding to the preset time length p+1 includes neural training data from time mp to time m, where m is an integer greater than p.

[0168] Furthermore, in embodiments of this application, the training dataset also includes neural training data of at least two brain nodes at time m+1.

[0169] Furthermore, in the embodiments of this application, various methods can be used to determine the neural training data of any brain node in the time series, and this application does not impose specific limitations.

[0170] For example, in some embodiments, neural training data can be acquired using existing bioinformatics databases, which include, but are not limited to, one or more public datasets such as human fMRI datasets, EEG datasets, and MEG datasets.

[0171] For example, in some embodiments, neural training data can also be generated using computational models.

[0172] Furthermore, in the embodiments of this application, the time series prediction network used to train the digital twin brain model can be any type and structure of artificial neural network for time series prediction, and this application does not impose any specific limitations.

[0173] For example, in some embodiments, the time series prediction network can use classic models such as multilayer perceptron, convolutional neural network, and recurrent neural network, or other network architectures suitable for time series prediction can be selected. This application does not make specific limitations.

[0174] Furthermore, in the embodiments of this application, when training the time series prediction network based on the training dataset to obtain a digital twin brain model, the time series prediction network can be used to obtain prediction data for at least two brain nodes at time m+1 based on the neural training data of at least two brain nodes from time mp to time m; based on the prediction data of at least two brain nodes at time m+1 and the neural training data of at least two brain nodes at time m+1, the time series prediction network is corrected to obtain the digital twin brain model.

[0175] For example, in some embodiments, when training a time series prediction network based on a training dataset, the predicted data for at least two brain nodes at time m+1 is determined using a digital twin brain model. The process can be referred to the following formula (8):

[0176] The time series prediction network can be represented as f(·), x (m-p) ,x (m-p+1) ,…,x m It can represent neural training data from time point (mp) to time point m.

[0177] For example, in some embodiments, when training a time series prediction network based on a training dataset, the predicted data for at least two brain nodes at time m+1 is determined using a digital twin brain model. The process can be referred to the following formula (9):

[0178] Where θ represents all parameters in the time series prediction network that can be optimized through training iterations.

[0179] Furthermore, in the embodiments of this application, when modifying the time series prediction network based on the prediction data of at least two brain nodes at time m+1 and the neural training data of at least two brain nodes at time m+1 to obtain a digital twin brain model, the model parameters are determined based on the prediction data of at least two brain nodes at time m+1, the neural training data of at least two brain nodes at time m+1, and the preset time length corresponding to the training dataset; the digital twin brain model is determined based on the model parameters.

[0180] It is understood that, in the embodiments of this application, after predicting neural data based on the input neural training data through a time series prediction network and obtaining the predicted data for the next time step, the time series prediction network can be further trained by minimizing the loss function in combination with the neural training data for the next time step, thereby obtaining the corresponding model parameters. Finally, the corresponding digital twin brain model can be obtained using the model parameters.

[0181] In the embodiments of this application, the loss function can be defined as an objective function of any type. For example, the loss function can be defined by minimizing the mean squared error of time series prediction. This application does not impose any specific limitations.

[0182] For example, in some embodiments, it is assumed that the time series prediction network is trained by minimizing the mean squared error of the time series prediction, i.e., the loss function is defined as follows: (10)

[0183] Where T represents the number of predicted data points obtained for at least two brain nodes corresponding to multiple time points based on the training dataset. By minimizing the loss function L(θ) using optimization methods such as stochastic gradient descent and adaptive learning rate algorithms, network parameters suitable for describing the dynamics of large-scale brain networks can be obtained, i.e., model parameters. Based on these model parameters, an artificial neural network that can serve as a digital twin brain model can be obtained, i.e., a digital twin brain model can be obtained.

[0184] It is understood that in the embodiments of this application, the digital twin brain model is obtained based on time series prediction network training. Therefore, the digital twin brain model can effectively predict the dynamic changes of brain neural signals (brain neural data), thereby obtaining more accurate information on effective brain connectivity.

[0185] Furthermore, in the embodiments of this application, since the time series length corresponding to the training dataset is greater than or equal to the time series length of the brain neural data used in the prediction process, it is possible to select to predict data at multiple time points through a time series prediction network, and calculate the difference between the multiple predicted data obtained from the prediction and the multiple neural training data at the corresponding time points. The model parameters are determined by comprehensively considering the multiple errors obtained from the calculation, thereby obtaining a digital twin brain model with better prediction performance.

[0186] Furthermore, in the embodiments of this application, when training a time series prediction network based on a training dataset to obtain a digital twin brain model, the time series prediction network can be used to obtain multiple prediction data corresponding to multiple time points for at least two brain nodes based on the training dataset; based on the multiple prediction data and the neural training data corresponding to multiple time points for at least two brain nodes, multiple data errors corresponding to multiple time points are determined; model parameters are determined based on the multiple data errors; and the digital twin brain model is determined based on the model parameters.

[0187] In other words, in the embodiments of this application, for a sufficiently large time series length, multiple optional parameters t can be selected for neural data prediction based on the neural training data in the training dataset. That is, multiple predicted data are obtained by prediction, and finally, the difference between the multiple predicted data and the multiple neural training data corresponding to multiple time points is calculated to determine multiple data errors (multiple errors) corresponding to multiple time points. These data errors are comprehensively considered to determine the model parameters.

[0188] For example, in some embodiments, it is assumed that the digital twin brain is a 3-step prediction of a 1-step process (using neural training data from three consecutive time points to predict brain neural data from the next time point), i.e., p=2, with a preset time length of 3. The neural training data in the training dataset includes neural training data from five time points, from time 1 to time 5, i.e., a time series length of 5. Then, the predicted data for time points 4 and 5 can be obtained from the neural training data at times 1, 2, and 3, and from the neural training data at times 2, 3, and 4, respectively. The errors between the predicted data at time 4 and the neural training data, and between the predicted data at time 5 and the neural training data, are then calculated to obtain the data errors for time points 4 and 5. Finally, the model parameters are determined based on these two data errors.

[0189] Furthermore, in the embodiments of this application, in order to evaluate the effect of the digital twin brain model on the characterization of large-scale brain network dynamics, the performance of the trained digital twin brain model can be further measured and evaluated.

[0190] Furthermore, in the embodiments of this application, when measuring and evaluating the performance of the trained digital twin brain model, the determination coefficient corresponding to the digital twin brain model can be determined based on the training dataset and the prediction dataset corresponding to the training dataset; through the digital twin brain model, based on brain neural data and random noise, the first functional connectivity matrix predicted by the digital twin brain model is determined, and the matrix correlation coefficient between the first functional connectivity matrix and the second functional connectivity matrix corresponding to the training dataset is determined; based on the determination coefficient and the matrix correlation coefficient, the model performance parameters of the digital twin brain model are determined.

[0191] It is understood that, in the embodiments of this application, the training dataset may include neural training data of at least two brain nodes corresponding to T time points, and the prediction dataset may include prediction data of at least two brain nodes corresponding to T time points, that is, the prediction dataset may include prediction data of T time points corresponding to the neural training data of T time points.

[0192] Furthermore, when determining the determination coefficients of the digital twin brain model based on the training dataset and the prediction dataset corresponding to the training dataset, the mean of the data for at least two brain nodes corresponding to T time points is determined based on the neural training data for at least two brain nodes corresponding to T time points; the determination coefficients of the digital twin brain model are determined based on the mean of the data, the neural training data corresponding to T time points, and the prediction data corresponding to T time points.

[0193] Exemplarily, in some embodiments, the determination coefficient R introduced for evaluating the digital twin brain model obtained through training is... 2 The following formula (11) can be used for calculation:

[0194] Where N represents the number of nodes in a large-scale brain network (the number of brain nodes), x i,t This represents the value of node i at time point t in brain neural data. This represents the value of node i at time point t predicted by the digital twin brain model. Let represent the mean of node i at all time points in the brain neural data, and T represent the length of the brain neural data used to train the time series prediction network.

[0195] It is understood that, in the embodiments of this application, the determination coefficient R obtained through calculation is... 2 This allows for the evaluation of the performance of digital twin brain models. The better the digital twin brain model predicts brain neural data, the better it can characterize the state transition relationships of the brain network.

[0196] Furthermore, in the embodiments of this application, a first functional connectivity matrix predicted by the digital twin brain model can be determined based on brain neural data and random noise. Simultaneously, a second functional connectivity matrix can be predicted based on the training dataset, i.e., the second functional connectivity matrix corresponding to the training dataset is determined. Then, the matrix correlation coefficient can be further determined based on the first and second functional connectivity matrices. The matrix correlation coefficient can be used to determine the correlation between the first and second functional connectivity matrices.

[0197] It is understood that, in the embodiments of this application, the functional connectivity matrix is ​​the correlation matrix between the activities of different brain nodes. Specifically, the first functional connectivity matrix can be understood as the correlation matrix between the activities of different brain nodes predicted by the trained digital twin brain model; while the second functional connectivity matrix can be understood as the correlation matrix between the activities of different brain nodes calculated based on the training dataset.

[0198] Furthermore, in the embodiments of this application, when determining the first functional connectivity matrix based on brain neural data and random noise, a digital twin brain model can be used to generate data driven by random noise, thereby generating the corresponding first functional connectivity matrix.

[0199] For example, in some embodiments, the initial value can be all zeros (i.e., x). 0..p =0), through the formula x t+1 =f(x) (t-p)..t +δ (t-p)..t The simulation data is generated iteratively (δ, θ) until the length of the simulation data matches the length of the brain neural data used to train the model. t It is random noise with a mean of 0 that follows a normal distribution.

[0200] It is understood that, in the embodiments of this application, the functional connectivity matrix is ​​reconstructed using a digital twin brain model, and then the matrix correlation coefficient between the first functional connectivity matrix reconstructed by the digital twin brain model and the second functional connectivity matrix corresponding to the training dataset is further determined. This matrix correlation coefficient can determine the reconstruction effect of the functional connectivity matrix. The better the reconstruction effect of the functional connectivity matrix, the better the digital twin brain model can characterize the functional dependencies between brain nodes.

[0201] In other words, in the embodiments of this application, after training the digital twin brain model, in order to evaluate the digital twin brain model's ability to characterize large-scale brain network dynamics, the coefficient of determination and matrix correlation coefficient can be introduced to measure and evaluate the performance of the digital twin brain model. For example, the coefficient of determination predicted from brain neural data, the correlation coefficient between the functional connectivity matrix of the data generated by the digital twin brain model under random noise, and the functional connectivity matrix of the brain neural data (training dataset) used to train the model (matrix correlation coefficient) can be used to measure the effectiveness of the digital twin brain model.

[0202] Furthermore, in the embodiments of this application, after completing the digital twin brain model, the digital twin brain model can be used to further determine the effective brain connection information between different brain nodes by increasing virtual perturbation, in accordance with the brain effective connection information derivation method proposed in the above embodiments, thereby obtaining the whole-brain effective connection of at least two brain nodes.

[0203] In summary, the training method for the digital twin brain model proposed in step 901 above uses artificial neural networks (such as time series prediction networks) to learn and predict the dynamic changes of brain neural signals. The artificial neural network is used as the digital twin brain model, and the digital twin brain model is trained using reconstructed or predicted brain neural data. Furthermore, the prediction error and the functional connectivity of the generated data can be used as indicators to evaluate the digital twin brain model. This avoids dependence on specific model assumptions and can be flexibly applied to neural signals of different modalities without extensive adjustments.

[0204] This application provides a training method for a digital twin brain model. The training device for the digital twin brain model trains a time-series prediction network based on a training dataset to obtain the digital twin brain model. The digital twin brain model is used to predict neural data at the next time step based on the brain neural data of brain nodes within a time series. The training dataset includes neural training data of at least two brain nodes within a time series. In other words, in this application, a time-series prediction network can be used to learn and predict the dynamic changes of brain neural signals to obtain a digital twin brain model with time-series prediction capabilities. The digital twin brain model can effectively and accurately predict the dynamic changes of brain neural data.

[0205] Based on the above embodiments, another embodiment of this application proposes a method for deriving effective brain connectivity information and a method for training a digital twin brain model. On one hand, the training method for the digital twin brain model can use artificial neural networks to fit large-scale neurodynamics, construct a digital twin brain model, and characterize the functional dependencies between different brain regions. On the other hand, the method for deriving effective brain connectivity information, based on the trained digital twin brain model, applies virtual stimuli to the model and observes the responses of other regions to the stimuli to characterize the effective connections between brain nodes. This can predict the brain's response patterns under different states, thereby characterizing the symbols (effective connection types), weights (effective connection strengths), and directions (effective connection directions) of effective connections across the entire brain.

[0206] Figure 11 is a schematic diagram of neural data analysis under real stimulation proposed in an embodiment of this application. As shown in Figure 11, in general, effective connectivity can be evaluated through neural stimulation experiments. These experiments use real stimulation to perturb specific brain regions and monitor the neural responses in other regions, recording neural data synchronously to directly prove causal relationships. However, such a method limits the scalability of whole-brain analysis due to its invasiveness.

[0207] Furthermore, in the embodiments of this application, an artificial neural network can be used to construct a digital twin brain model, wherein the artificial neural network can be a time series prediction network.

[0208] For example, in some embodiments, the dynamics of large-scale brain networks are expressed using an artificial neural network f(·) as follows: (12)

[0209] The artificial neural network f(·) can use classic models such as multilayer perceptrons, convolutional neural networks, and recurrent neural networks, or other network architectures suitable for time series prediction can be selected. θ represents all parameters in the artificial neural network that can be optimized through training iterations, and x... (t-p)..t This represents brain neural data from time point (tp) to time point t. This represents the brain neural data at time point (t+1) predicted by the artificial neural network. The value of the hyperparameter p is determined by the nature of the brain neural data.

[0210] For example, in some embodiments, the artificial neural network is trained by minimizing the mean squared error of time series predictions, i.e., the loss function is defined according to formula (10). Here, T represents the length of the brain neural data used to train the artificial neural network. By minimizing the loss function L(θ) using optimization methods such as stochastic gradient descent and adaptive learning rate algorithms, parameters suitable for describing the dynamics of large-scale brain networks can be obtained, resulting in an artificial neural network that can serve as a digital twin brain model.

[0211] Furthermore, in the embodiments of this application, after training to obtain a digital twin brain model, the digital twin brain model can be evaluated using the functional connection between prediction error and generated data.

[0212] It is understood that, in the embodiments of this application, in order to evaluate the effect of the digital twin brain model on the characterization of large-scale brain network dynamics, the correlation coefficient between the determination coefficient of brain neural data prediction, the functional connectivity matrix of the data generated by the digital twin brain model under random noise driving, and the functional connectivity matrix of the brain neural data (training dataset) used to train the model can be used to measure the effect of the alternative model.

[0213] Exemplarily, in some embodiments, the coefficient of determination R for predicting brain neural data is... 2 It can be calculated using formula (12), where N represents the number of nodes in a large-scale brain network, and x i,t This represents the value of node i at time point t in brain neural data. This represents the value of node i at time point t predicted by the digital twin brain model. This represents the mean of node i across all time points in the brain neural data. The better the digital twin brain model predicts the brain neural data, the better the alternative model can characterize the state transition relationships of the brain network.

[0214] For example, in some embodiments, the functional connectivity matrix is ​​a correlation matrix between the activities of different nodes. When generating data using a digital twin brain model driven by random noise, it can be started with all-zero data (i.e., x). 0..p =0), through the formula x t+1 =f(x) (t-p)..t +δ (t-p)..t The simulation data is generated iteratively (δ, θ) until the length of the simulation data matches the length of the brain neural data used to train the model. t It is random noise with a mean of 0 that follows a normal distribution. The better the digital twin brain model reconstructs the functional connectivity matrix, the better the alternative model can characterize the functional dependencies between nodes.

[0215] Furthermore, in the embodiments of this application, after training to obtain a digital twin brain model with time series prediction capabilities, the effective connectivity of the whole brain can be inferred through virtual perturbation of the digital twin brain model.

[0216] Understandably, after obtaining and validating the effectiveness of the digital twin brain model, the effective connections of the brain network can be obtained by applying virtual perturbations to the model. When characterizing the effective connections from a given node to other nodes, the virtual perturbation can be represented as a slight rise in the signal of a given node at the last time point of the perturbated brain neural data. The effective connections of that node to other nodes can be characterized by observing the effect of applying the virtual perturbation on the output of the surrogate model.

[0217] For example, in some embodiments, the effective connection EC from node i to node j ij It can be calculated using formula (7), where E t (·) represents the mean value over time t, f(·) j The component j represents the output of the alternative model, and the virtual perturbation P is consistent with p. i =Δ,p k,k≠i =0 or other traditional experimental paradigms require an N-dimensional virtual perturbation vector, where Δ is the intensity of the virtual perturbation applied to node i.

[0218] Furthermore, in the embodiments of this application, the effective connections of the whole brain can be obtained by traversing all nodes i and j. The effective connections obtained through the derivation method of brain effective connection information proposed in this application can be brain effective connection information with symbols (effective connection type), weights (effective connection strength), and directions (effective connection direction), which can then be used to guide precise neuromodulation.

[0219] For example, in some embodiments, Figure 12 is a schematic diagram of the derivation of effective brain connectivity information based on a digital twin brain model obtained through training, as proposed in the embodiments of this application. As shown in Figure 12, during the training process of the digital twin brain model, the number of nodes in the model can correspond to the number of brain nodes. For example, there can be four model nodes corresponding to the four brain nodes a, b, c, and d. The training dataset used for model training can include the neural training data of the brain nodes in the time series, such as resting-state functional magnetic resonance imaging data from time t to time t+Δt. The neural training data x(t) at time t is used as input to predict the predicted data x(t+Δt) at time t+Δt. Combined with the neural training data at time t+Δt, the model can be iterated and corrected to finally obtain the digital twin brain (digital twin brain model). After obtaining the digital twin brain model through training, the predicted neural data obtained with and without perturbation data are further compared and analyzed by combining the digital twin brain model with virtual perturbation (perturbation data) applied to different brain nodes. This allows for the acquisition of effective brain connectivity information between any two brain nodes, ultimately resulting in accurate whole-brain effective connectivity, which includes effective connectivity type, effective connectivity direction, and effective connectivity strength.

[0220] Furthermore, the method for deriving effective brain connectivity information and the training method for the digital twin brain model proposed in this application were verified using three simulation models, demonstrating the feasibility of the method. Moreover, the inference results for effective connectivity proposed in this application are superior to traditional data-driven methods (Granger causality, dynamic causality models). The method for deriving effective brain connectivity information proposed in this application has been successfully applied to four brain atlases, obtaining whole-brain effective connectivity under different brain atlases based on resting-state functional magnetic resonance imaging data. Furthermore, the obtained connectivity shows significant differences between autistic patients and normal individuals.

[0221] In summary, in the embodiments of this application, firstly, artificial neural networks (such as time series prediction networks) are used to learn and predict the dynamic changes of brain neural signals. The artificial neural network serves as a digital twin brain model, and the model is trained using reconstructed or predicted brain neural data. Furthermore, prediction errors and functional connectivity of generated data can be used as indicators to evaluate the digital twin brain model. This avoids reliance on specific model assumptions and allows for flexible application to neural signals of different modalities without extensive adjustments. Secondly, the digital twin brain model can be stimulated with different virtual stimuli. By virtually perturbing the digital twin brain model, the brain's response to stimulation can be predicted, and effective connections throughout the brain can be inferred. This allows for the simultaneous consideration of a large number of brain regions, making it suitable for inferring effective connections throughout the brain and enabling the simultaneous characterization of the sign, weight, and direction of connections.

[0222] This application provides a method for deriving effective brain connectivity information and a method for training a digital twin brain model. The digital twin brain model is obtained by training a time series prediction network. Therefore, the digital twin brain model can effectively and accurately predict the dynamic changes of brain neural data, thereby obtaining more accurate effective brain connectivity information.

[0223] Based on the above embodiments, in another embodiment of this application, FIG13 is a schematic diagram of the composition structure of the brain effective connectivity information derivation device proposed in the embodiment of this application. As shown in FIG13, the brain effective connectivity information derivation device 130 proposed in the embodiment of this application may include a determining unit 1301 and an acquiring unit 1302.

[0224] The determining unit 1301 is used to determine the perturbation data of the first brain node among at least two brain nodes and the brain neural data of the at least two brain nodes in the time series;

[0225] The acquisition unit 1302 is used to obtain the first predicted neural data of the at least two brain nodes at the next moment based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series using a digital twin brain model; wherein, the digital twin brain model is obtained by training based on a time series prediction network.

[0226] The determining unit 1301 is further configured to determine effective brain connection information between the first brain node and the second brain node based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step; wherein, the second brain node is the other brain node among the at least two brain nodes besides the first brain node; the second predicted neural data represents the predicted data obtained without adding the perturbation data.

[0227] Furthermore, in the embodiments of this application, FIG14 is a schematic diagram of the composition structure of the training device for the digital twin brain model proposed in the embodiments of this application. As shown in FIG14, the training device 140 for the digital twin brain model proposed in the embodiments of this application may include a training unit 1401 and an evaluation unit 1402.

[0228] Training unit 1401 is used to train a time series prediction network based on a training dataset to obtain a digital twin brain model; wherein, the digital twin brain model is used to predict the neural data of the next time step based on the brain neural data of the brain nodes in the time series.

[0229] The training dataset includes neural training data for at least two brain nodes corresponding to a preset time length p+1; wherein, the neural training data corresponding to the preset time length p+1 includes neural training data from time mp to time m, and the training dataset also includes neural training data for the at least two brain nodes at time m+1; p is an integer greater than or equal to 0, and m is an integer greater than p.

[0230] The process of training a time-series prediction network based on a training dataset to obtain a digital twin brain model includes:

[0231] Using the time series prediction network, based on the neural training data of the at least two brain nodes from the mp-th time to the m-th time, the prediction data of the at least two brain nodes at the (m+1)-th time is obtained;

[0232] Based on the prediction data of the at least two brain nodes at time m+1 and the neural training data of the at least two brain nodes at time m+1, the time series prediction network is modified to obtain the digital twin brain model.

[0233] Further, in an embodiment of this application, the evaluation unit 1402 is configured to determine the determination coefficient corresponding to the digital twin brain model based on the training dataset and the prediction dataset corresponding to the training dataset; through the digital twin brain model, based on the brain neural data and random noise, determine the first functional connectivity matrix predicted by the digital twin brain model, and determine the matrix correlation coefficient between the first functional connectivity matrix and the second functional connectivity matrix corresponding to the training dataset; and determine the model performance parameters of the digital twin brain model based on the determination coefficient and the matrix correlation coefficient.

[0234] In the embodiments of this application, FIG15 is a schematic diagram of the composition structure of the computer device proposed in the embodiments of this application. As shown in FIG15, the computer device 150 proposed in the embodiments of this application includes a processor 1501 and a memory 1502 storing instructions executable by the processor 1501. Further, the computer device 1500 may also include a communication interface 1503 and a bus 1504 for connecting the processor 1501, the memory 1502 and the communication interface 1503.

[0235] In the embodiments of this application, the processor 1501 can be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that for different devices, the electronic device used to implement the above-mentioned processor function can also be other types, and the embodiments of this application do not specifically limit this. The computer device 1500 may also include a memory 1502, which can be connected to the processor 1501. The memory 1502 is used to store executable program code, which includes computer operation instructions. The memory 1502 may include high-speed RAM memory and may also include non-volatile memory, such as at least two disk drives.

[0236] In embodiments of this application, bus 1504 is used to connect communication interface 1503, processor 1501, and memory 1502, as well as the mutual communication between these devices.

[0237] In embodiments of this application, memory 1502 is used to store instructions and data.

[0238] Further, in embodiments of this application, the processor 1501 is configured to: determine perturbation data of a first brain node among at least two brain nodes and brain neural data of the at least two brain nodes in a time series; obtain first predicted neural data of the at least two brain nodes at the next time step based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series using a digital twin brain model; wherein the digital twin brain model is obtained by training based on a time series prediction network; determine effective brain connection information from the first brain node to the second brain node based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step; wherein the second brain node is another brain node among the at least two brain nodes besides the first brain node; the second predicted neural data represents the predicted data obtained without adding the perturbation data.

[0239] Furthermore, in the embodiments of this application, the processor 1501 is also used to: train a time series prediction network based on a training dataset to obtain a digital twin brain model; wherein, the digital twin brain model is used to predict the neural data of the next moment based on the brain neural data of the brain nodes in the time series.

[0240] The training dataset includes neural training data for at least two brain nodes corresponding to a preset time length p+1; wherein, the neural training data corresponding to the preset time length p+1 includes neural training data from time mp to time m, and the training dataset also includes neural training data for the at least two brain nodes at time m+1; p is an integer greater than or equal to 0, and m is an integer greater than p.

[0241] The process of training a time-series prediction network based on a training dataset to obtain a digital twin brain model includes:

[0242] Using the time series prediction network, based on the neural training data of the at least two brain nodes from the mp-th time to the m-th time, the prediction data of the at least two brain nodes at the (m+1)-th time is obtained;

[0243] Based on the prediction data of the at least two brain nodes at time m+1 and the neural training data of the at least two brain nodes at time m+1, the time series prediction network is modified to obtain the digital twin brain model.

[0244] In practical applications, the aforementioned memory 1502 can be volatile memory, such as random-access memory (RAM); or non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 1501.

[0245] Furthermore, in this embodiment, the functional modules 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. The integrated unit can be implemented in hardware or as a software functional module.

[0246] If the integrated unit is implemented as a software functional module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or 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, server, or network device, etc.) or processor to execute all or part of the steps of the method of this embodiment. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0247] This application provides a device for deriving effective brain connectivity information, a training device for a digital twin brain model, and a computer device. The digital twin brain model is obtained by training a time series prediction network. Therefore, the digital twin brain model can effectively and accurately predict the dynamic changes of brain neural data, thereby obtaining more accurate effective brain connectivity information.

[0248] Specifically, the program instructions corresponding to a method for deriving effective brain connectivity information in this embodiment can be stored on storage media such as optical discs, hard disks, and USB flash drives. When the program instructions corresponding to the method for deriving effective brain connectivity information in the storage media are read or executed by an electronic device, the following steps are included:

[0249] Determine the perturbation data of the first brain node among at least two brain nodes and the brain neural data of the at least two brain nodes in the time series;

[0250] Using a digital twin brain model, based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series, the first predicted neural data of the at least two brain nodes at the next time step is obtained; wherein, the digital twin brain model is obtained by training based on a time series prediction network;

[0251] Based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step, the effective brain connection information from the first brain node to the second brain node is determined; wherein, the second brain node is the other brain node among the at least two brain nodes besides the first brain node; the second predicted neural data represents the predicted data obtained without adding the perturbation data.

[0252] Specifically, the program instructions corresponding to the training method of a digital twin brain model in this embodiment can be stored on storage media such as optical discs, hard disks, and USB flash drives. When the program instructions corresponding to the training method of a digital twin brain model in the storage media are read or executed by an electronic device, the following steps are included:

[0253] A time-series prediction network is trained based on a training dataset to obtain a digital twin brain model; wherein, the digital twin brain model is used to predict the neural data of the next time step based on the brain neural data of the brain nodes in the time series.

[0254] The training dataset includes neural training data for at least two brain nodes corresponding to a preset time length p+1; wherein, the neural training data corresponding to the preset time length p+1 includes neural training data from time mp to time m, and the training dataset also includes neural training data for the at least two brain nodes at time m+1; p is an integer greater than or equal to 0, and m is an integer greater than p.

[0255] The process of training a time-series prediction network based on a training dataset to obtain a digital twin brain model includes:

[0256] Using the time series prediction network, based on the neural training data of the at least two brain nodes from the mp-th time to the m-th time, the prediction data of the at least two brain nodes at the (m+1)-th time is obtained;

[0257] Based on the prediction data of the at least two brain nodes at time m+1 and the neural training data of the at least two brain nodes at time m+1, the time series prediction network is modified to obtain the digital twin brain model.

[0258] This application also provides a computer program product.

[0259] In some embodiments, the computer program product may include a computer program or instructions.

[0260] In some embodiments, the computer program product can be applied to the computer device in the embodiments of this application, and the computer program instructions cause the computer to execute the corresponding processes implemented by the computer device in the various methods of the embodiments of this application. For the sake of brevity, they will not be described in detail here.

[0261] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0262] This application is described with reference to schematic and / or block diagrams of implementations of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the schematic and / or block diagrams, and combinations thereof, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the schematic and / or block diagrams.

[0263] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0264] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more blocks in a block diagram.

[0265] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Industrial applicability

[0266] This application discloses a method for deriving effective brain connectivity information, a training method for a digital twin brain model, an apparatus, and a device. The method includes: determining perturbation data of a first brain node and brain neural data of the at least two brain nodes within a time series; using a digital twin brain model, based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes within the time series, obtaining first predicted neural data of the at least two brain nodes at the next time step; and determining effective brain connectivity information from the first brain node to the second brain node based on the first predicted neural data and second predicted neural data of the at least two brain nodes at the next time step. A time series prediction network is trained using a training dataset to obtain the digital twin brain model. On one hand, based on the trained digital twin brain model, perturbation data is added to the brain nodes to predict the brain's response to perturbation stimuli and infer effective brain connectivity information. Since the digital twin brain model is obtained by training a time series prediction network, it can effectively and accurately predict the dynamic changes in brain neural data, thereby obtaining more accurate effective brain connectivity information. On the other hand, time-series prediction networks can be used to learn and predict dynamic changes in brain neural signals, resulting in digital twin brain models with time-series prediction capabilities. These digital twin brain models can effectively and accurately predict the dynamic changes in brain neural data.

Claims

A method for deriving information about the brain's effective connections, the method comprising: Determine the perturbation data of a first brain node out of at least two brain nodes and the brain neural data of the at least two brain nodes in a time series; wherein the perturbation data of the first brain node includes an element with a non-zero value in a virtual perturbation vector; Using a digital twin brain model, based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series, the first predicted neural data of the at least two brain nodes at the next time step is obtained; wherein, the digital twin brain model is obtained by training based on a time series prediction network; Based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step, the effective brain connection information from the first brain node to the second brain node is determined; wherein, the second brain node is the other brain node among the at least two brain nodes besides the first brain node; the second predicted neural data represents the predicted data obtained without adding the perturbation data. According to the method of claim 1, wherein, The method further includes: Using the digital twin brain model, based on the brain neural data of the at least two brain nodes in the time series, the second predicted neural data of the at least two brain nodes at the next time step is obtained. The method according to claim 1 or 2, wherein, The effective brain connectivity information includes the type of effective connectivity. The determination of the effective brain connectivity information from the first brain node to the second brain node, based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step, includes: If the first predicted neural data of the second brain node at the next time step is greater than the second predicted neural data of the second brain node at the next time step, the effective connection type is determined to be excitatory. If the first predicted neural data of the second brain node at the next time step is less than the second predicted neural data of the second brain node at the next time step, the effective connection type is determined to be inhibitory. The method according to any one of claims 1-3, wherein, The brain neural data of the at least two brain nodes in the time series includes brain neural data of the at least two brain nodes corresponding to a preset time length p+1; wherein, the brain neural data corresponding to the preset time length p+1 includes brain neural data from time tp to time t, where p is an integer greater than or equal to 0, and t is an integer greater than p. The method involves obtaining first predicted neural data for the at least two brain nodes at the next time step using a digital twin brain model, based on perturbation data of the first brain node and neural data of the at least two brain nodes within a time series. This includes: Based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes at time t, determine the perturbation-resolved brain neural data of the at least two brain nodes at time t. Using the digital twin brain model, based on the perturbed brain neural data of the at least two brain nodes at time t, and the brain neural data of the at least two brain nodes from time tp to time t-1, the first predicted neural data of the at least two brain nodes at time t+1 is obtained. The method according to claim 4, wherein, The determination of the perturbation-adjusted brain neural data of the at least two brain nodes at time t, based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes at time t, includes: The perturbation data of the first brain node and the brain neural data of the at least two brain nodes at time t are superimposed to obtain the perturbation-resolved brain neural data of the at least two brain nodes at time t. The method according to claim 4, wherein, The method of obtaining second predicted neural data for the at least two brain nodes at the next time step using a digital twin brain model, based on the brain neural data of the at least two brain nodes in the time series, includes: Using the digital twin brain model, based on the brain neural data of the at least two brain nodes from time tp to time t, the second predicted neural data of the at least two brain nodes at time t+1 is obtained. The method according to any one of claims 1-6, wherein, The effective brain connectivity information includes the effective connectivity direction, and determining the effective brain connectivity information from the first brain node to the second brain node includes: The effective connection direction is defined as from the first brain node to the second brain node. The method according to any one of claims 1-7, wherein, The effective brain connectivity information includes the type of effective connectivity. The determination of the effective brain connectivity information from the first brain node to the second brain node, based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step, includes: If the first predicted neural data of the second brain node at time t+1 is greater than the second predicted neural data of the second brain node at time t+1, the effective connection type is determined to be excitatory. If the first predicted neural data of the second brain node at time t+1 is less than the second predicted neural data of the second brain node at time t+1, the effective connection type is determined to be inhibitory. The method according to any one of claims 1-8, wherein, The effective brain connectivity information includes effective connectivity strength. The determination of the effective brain connectivity information from the first brain node to the second brain node, based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step, includes: Determine the strength information of the first predicted neural data of the second brain node at time t+1, and determine the strength information of the second predicted neural data of the second brain node at time t+1; The effective connection strength is determined based on the strength information of the first predicted neural data and the strength information of the second predicted neural data. The method according to any one of claims 1-9, wherein, The determination of perturbation data for the first brain node out of at least two brain nodes includes: The perturbation data of the first brain node is set based on a preset perturbation intensity; The method further includes: Traverse the at least two brain nodes and determine the effective brain connection information between any two brain nodes to obtain the whole-brain effective connection of the at least two brain nodes. The method according to any one of claims 1-10, wherein, The method further includes: Using the digital twin brain model, based on multiple perturbation data of the first brain node corresponding to multiple times and brain neural data of the at least two brain nodes in the time series, multiple first predicted neural data of the at least two brain nodes corresponding to the multiple times are obtained respectively. Using the digital twin brain model, based on the brain neural data of the at least two brain nodes in the time series, multiple second predicted neural data corresponding to the at least two brain nodes at the multiple times are obtained respectively; Based on the plurality of first predictive neural data and the plurality of second predictive neural data, determine the plurality of valid brain connections from the first brain node to the second brain node, corresponding to the plurality of time points; Based on the multiple valid brain connectivity information, the target valid connectivity information from the first brain node to the second brain node is determined. A method for training a digital twin brain model, the method comprising: A time-series prediction network is trained based on a training dataset to obtain a digital twin brain model; wherein, the digital twin brain model is used to predict the neural data of the next time step based on the brain neural data of the brain nodes in the time series; the digital twin brain model is used to perform the method for deriving effective brain connectivity information as described in any one of claims 1-11; The training dataset includes neural training data for at least two brain nodes corresponding to a preset time length p+1; wherein, the neural training data corresponding to the preset time length p+1 includes neural training data from time mp to time m, and the training dataset also includes neural training data for the at least two brain nodes at time m+1; p is an integer greater than or equal to 0, and m is an integer greater than p. The process of training a time-series prediction network based on a training dataset to obtain a digital twin brain model includes: Using the time series prediction network, based on the neural training data of the at least two brain nodes from the mp-th time to the m-th time, the prediction data of the at least two brain nodes at the (m+1)-th time is obtained; Based on the prediction data of the at least two brain nodes at time m+1 and the neural training data of the at least two brain nodes at time m+1, the time series prediction network is modified to obtain the digital twin brain model. The method according to claim 12, wherein, The process of refining the time-series prediction network based on the prediction data of the at least two brain nodes at time m+1 and the neural training data of the at least two brain nodes at time m+1 to obtain the digital twin brain model includes: The model parameters are determined based on the prediction data of the at least two brain nodes at the (m+1)th time, the neural training data of the at least two brain nodes at the (m+1)th time, and the preset time length corresponding to the training dataset. The digital twin brain model is determined based on the model parameters. The method according to claim 12 or 13, wherein, The process of training a time-series prediction network based on a training dataset to obtain a digital twin brain model includes: Using the time series prediction network, based on the training dataset, multiple predicted data corresponding to multiple time points for the at least two brain nodes are obtained; Based on the multiple prediction data and the neural training data of the at least two brain nodes corresponding to the multiple time points, multiple data errors corresponding to the multiple time points are determined; The model parameters are determined based on the aforementioned multiple data errors; The digital twin brain model is determined based on the model parameters. The method according to any one of claims 12-14, wherein, The method further includes: Based on the training dataset and the prediction dataset corresponding to the training dataset, the determination coefficient corresponding to the digital twin brain model is determined; Using the digital twin brain model, based on the brain neural data and random noise, the first functional connectivity matrix predicted by the digital twin brain model is determined, and the matrix correlation coefficient between the first functional connectivity matrix and the second functional connectivity matrix corresponding to the training dataset is determined. Based on the determination coefficient and the matrix correlation coefficient, the model performance parameters of the digital twin brain model are determined. The method according to any one of claims 12-15, wherein, The training dataset includes neural training data for at least two brain nodes corresponding to T time points, and the prediction dataset includes prediction data for the at least two brain nodes corresponding to T time points, where T is an integer greater than 0. Determining the determination coefficients corresponding to the digital twin brain model based on the training dataset and the prediction dataset corresponding to the training dataset includes: Based on the neural training data of the at least two brain nodes corresponding to T time points, determine the average data of the at least two brain nodes corresponding to T time points; Based on the mean of the data, the neural training data corresponding to T time points, and the prediction data corresponding to T time points, the determination coefficient corresponding to the digital twin brain model is determined. A device for deriving effective brain connectivity information, the device comprising: A determining unit is configured to determine perturbation data of a first brain node among at least two brain nodes and brain neural data of the at least two brain nodes in a time series; wherein the perturbation data of the first brain node includes an element with a non-zero value in a virtual perturbation vector; The acquisition unit is used to obtain the first predicted neural data of the at least two brain nodes at the next moment based on the perturbation data of the first brain node and the brain neural data of the at least two brain nodes in the time series using a digital twin brain model; wherein, the digital twin brain model is obtained by training based on a time series prediction network; The determining unit is configured to determine effective brain connectivity information between the first brain node and the second brain node based on the first predicted neural data of the at least two brain nodes at the next time step and the second predicted neural data of the at least two brain nodes at the next time step; wherein, the second brain node is the other brain node among the at least two brain nodes besides the first brain node; the second predicted neural data represents the predicted data obtained without adding the perturbation data. A computer device, the computer device comprising: Processor and memory; among which, The memory is used to store computer programs that can run on the processor; The processor is configured to, when running the computer program, perform the method as described in any one of claims 1-11 or 12-16.

Images