Anatomical connection prediction method based on functional connection and kinetic model

Abstract

The invention discloses a functional connection and dynamic model-based anatomical connection prediction method, which comprises the following steps of: firstly, constructing a neurodynamic model, taking a random matrix as initial input, combining functional connection data and the neurodynamic model, and generating a coupling matrix which is more in line with neurodynamic characteristics by iteratively optimizing the random matrix; and the coupling matrix is used for predicting anatomical connection and replacing anatomical connection data to construct a high-precision brain model. According to the method, the initial random matrix is iteratively optimized, so that the generated coupling matrix gradually approaches real anatomical connection, non-intrusive anatomical connection prediction is realized, dependence on intrusive structure connection data is remarkably reduced, and compared with an existing neurodynamics modeling method depending on dMRI data, the method has the advantages that the modeling efficiency is improved, and the modeling cost is reduced. The used DMF model achieves higher fitting degree in structure and function, has higher flexibility and adaptability, is beneficial to revealing deep connection between functional connection and anatomical connection, and provides more efficient technical support for brain network research.

Description

Technical Field

[0001] The present invention belongs to the interdisciplinary technical field of neuroscience and computer science, and specifically relates to an anatomical connection prediction method based on functional connection and dynamics models. Background Art

[0002] In neuroscience research, understanding the structure and function of brain networks has always been a core challenge, among which anatomical connectivity and functional connectivity are two key concepts. Anatomical connectivity represents the physical connections between neurons in the brain and forms the basis of brain networks. In existing technologies, anatomical connections are obtained through invasive means, such as using tracer injection technology in non-human primates (such as macaques). This type of technology can accurately track the connections between brain regions and provide ground truth data of brain connections. However, because these methods are highly invasive, they cannot be promoted for use in human studies. Therefore, how to effectively obtain anatomical connection data has always been a research problem.

[0003] To solve this problem, researchers have tried to indirectly infer anatomical connections through non-invasive methods such as diffusion tensor imaging (dMRI) and functional magnetic resonance imaging (fMRI). Diffusion MRI can infer the direction of nerve fibers by tracing the diffusion path of water molecules in brain tissue, thereby constructing a structural connection map of the brain. However, the resolution and accuracy of this method are limited by technology, especially when resolving complex crossing fibers. In addition, diffusion MRI requires a large amount of data and computing resources, making the research process cumbersome and time-consuming. In contrast, functional connectivity measures the interactive relationship between different brain regions by analyzing the coordinated activities between them. Resting-state functional magnetic resonance imaging (rs-fMRI) is a commonly used method that can capture the spontaneous functional activity of the brain at rest, thereby inferring the functional connectivity between brain regions. This method has the advantages of being non-invasive, easy to obtain and efficient, and has been widely used in the study of brain networks. Studies have shown that resting-state functional connectivity can effectively identify direct anatomical connections in the brain and reflect the characteristics of anatomical connections. However, since functional connectivity not only reflects direct anatomical connections but may also be affected by multi-synaptic or other complex network factors, inferring anatomical connectivity based on functional connectivity becomes more complicated and challenging.

[0004] Neural dynamic models aim to combine anatomy and dynamics to explain the formation of resting-state networks. In theory, anatomical connections can be inferred through functional connectivity and computational models. However, existing neural dynamic models often construct structural connectivity matrices based on dMRI, and then use these structural connections to simulate the dynamic behavior of brain networks. Due to the low imaging accuracy of dMRI and the limited structural connectivity information obtained, the model often fails to achieve the expected accuracy when analyzing complex networks and simulating real brain functional networks. To solve this problem, researchers have begun to explore methods to optimize anatomical connections using functional connectivity data. Recent studies have proposed a method for iteratively optimizing structural connections based on functional connectivity. This method uses functional connectivity data to iteratively optimize the initial dMRI structural connectivity matrix to make it more consistent with the characteristics of functional connectivity, thereby improving the accuracy and predictive ability of the neural dynamic model. Although this method improves the accuracy of the model to a certain extent, it still relies on the initial dMRI connectivity data, which limits the application of this method in the absence of anatomical data or when anatomical data is scarce. Summary of the invention

[0005] To solve the above technical problems, the present invention provides an anatomical connection prediction method based on functional connectivity and dynamic models, which solves the problem of lack of high-precision anatomical connection data. By constructing a connection matrix that conforms to dynamic coupling, anatomical connections can be effectively predicted, thereby achieving a high-precision brain model.

[0006] The technical solution adopted by the present invention is: a method for predicting anatomical connections based on functional connections and dynamic models, and the specific steps are as follows:

[0007] S1. Construct a neural dynamics model, including: input module, DMF module, and output module;

[0008] The neural dynamics model adopts a dynamic mean field DMF model, and the functional connection matrix obtained through simulation is used as output.

[0009] Among them, the input module receives structural connection data as the initial input of the model, which represents the connection strength between different brain regions; the DMF module adopts the DMF model to describe the electrical activities of different brain regions through neurodynamic equations, and converts electrical signals into blood oxygen level-dependent BOLD signals through transduction functions; the output module generates a simulated functional connection matrix, uses the brain area activity signals output by the DMF module to calculate the functional connection strength between each brain area, and finally generates a simulated functional connection matrix.

[0010] S2, the input module initializes a random matrix as the initial coupling matrix of the DMF model;

[0011] Initialize a random matrix as the initial coupling matrix C of the model(0) , which is the initial input matrix of the DMF model, and this matrix will serve as the starting point for optimization.

[0012] Each element in the input matrix represents the connection relationship between specific brain regions. Represents the initial structural connectivity between brain regions n and p.

[0013] S3, sending the initial input matrix obtained in step S2 to the DMF module, and converting the electrical signal output by the model into a blood oxygen level dependent BOLD signal;

[0014] The DMF module uses neurodynamic equations to describe the electrical activities of different brain regions. The DMF model at the whole brain level is represented by a set of coupled differential equations. The specific expression is as follows:

[0015]

[0016]

[0017] in, represents the input current of an excitatory E or inhibitory I neuron population in brain region n, represents the firing rate of a population of excitatory E or inhibitory I neurons in brain area n. represents the average excitatory E or inhibitory I synaptic gating variable in brain region n, and t represents time. And the overall effective external input I0 = 0.382nA, which is represented by the effective coupling constant W in the excitatory pool and inhibitory pool, respectively. E =1 and W I = 0.7 scaling. Local excitability recursion ω + =1.4, G represents the global coupling parameter, C np represents the structural connectivity matrix between brain regions n and p, with all excitatory synaptic coupling J NMDA =0.15nA. represents the current transduction function, which converts the incoming total input current into firing rate. The gains of the transduction function are g E =310nC -1 , g I =615nC -1 , and the other parameters are threshold current Noise factor d E =0.16, d I =0.087. And in the transduction function, it is necessary to ensure that the denominator is not zero, that is, The kinetic parameter γ = 0.641 / 1000, the factor 1000 means that everything is in milliseconds, and the effective time constants are τ E =τNMDA=100ms and τI =τGABA=10ms. n represents Gaussian noise, whose amplitude is σ=0.01nA. n represents the local feedback inhibition weight of the brain region, and the feedback inhibition control FIC algorithm is used to adjust the J of each brain region n , so that the firing rate of the excitation pool is kept at a low firing rate, that is

[0018] The generalized hemodynamic model is then used to convert the simulated mean field activity in the DMF model into a BOLD signal. When it rises, it triggers vasodilation signal x n Increase, blood flow proportional to the vasodilation signal f n Caused by deoxyhemoglobin content q n and blood volume n The specific expression is as follows:

[0019]

[0020] where k is the signal decay rate, e is the blood flow-dependent elimination rate, ρ is the resting oxygen uptake fraction, α is the venous resistance, and τ is the time constant.

[0021] BOLD signal is the content of deoxyhemoglobin q n and blood volume n The BOLD signal is a static nonlinear function of B, which includes the volume-weighted sum of the extravascular and intravascular signals. For each region n, the BOLD signal is expressed as B n The expression is as follows:

[0022]

[0023] Among them, the resting blood volume fraction V0=0.02, the dimensionless parameters k1=7ρ, k2=2, k3=2ρ-0.2.

[0024] S4, based on the BOLD signal obtained in step S3, outputting the initial simulated functional connectivity matrix through the output module;

[0025] The initial simulated functional connectivity matrix FC was generated by calculating the Pearson correlation coefficient between the simulated BOLD signals of different brain regions. sim , for the BOLD signal time series of brain regions i and j, the Pearson correlation coefficient r ij The calculation expression is as follows:

[0026]

[0027] Among them, T represents the overall time length, B i，t represents the BOLD signal value of brain area i at time t, Represents the average BOLD signal of brain area i.

[0028] The initial simulated functional connectivity matrix is ​​then compared with the real functional connectivity data FC emp For comparison, the Pearson correlation coefficient and mean square error MSE are calculated. The MSE calculation expression is as follows:

[0029]

[0030] Where N represents the total number of connections in the matrix.

[0031] Finally, the Pearson correlation coefficient and mean square error were used to evaluate the difference between the simulated functional connectivity and the real functional connectivity data as the optimization objective function.

[0032] S5, based on step S4, using the gradient descent method to update the coupling matrix;

[0033] The gradient descent method is used to adjust the coupling matrix using the difference between the simulated and true functional connection matrices, so that the simulated functional connection gradually approaches the true functional connection. The update expression of the coupling matrix is ​​as follows:

[0034]

[0035] Among them, C (m) represents the coupling matrix of the input of the mth iteration, ∈ represents the learning rate, which controls the update amplitude of each step, and m represents the mth iteration.

[0036] Then use the updated C (m+1) Re-run the DMF model to generate a new BOLD signal, and then calculate the Pearson correlation coefficient of the new BOLD signal to obtain a new simulated functional connectivity matrix The difference between the new simulated functional connectivity and the real functional connectivity data in the current iteration is calculated, and this difference is compared with the difference obtained in the previous iteration. If the difference no longer decreases and is close to the preset fitting accuracy, the optimization is considered to have converged; otherwise, the iteration continues until the preset fitting accuracy is reached.

[0037] S6, based on step S5, after the iterative optimization is completed, an optimized coupling matrix, i.e., an optimal coupling matrix, is generated to perform anatomical connection prediction;

[0038] After the iterative optimization is completed, the coupling matrix C that conforms to the neural dynamics characteristics is generated optimized , as the final input of the DMF model, is used to predict anatomical connections and replace anatomical connection data to build a higher-precision brain model.

[0039] Beneficial effects of the present invention: The method of the present invention first constructs a neurodynamic model, takes a random matrix as the initial input, combines functional connection data and a neurodynamic model, generates a coupling matrix that is more in line with neurodynamic characteristics by iteratively optimizing the random matrix, and uses the coupling matrix to predict anatomical connections, replacing anatomical connection data to build a high-precision brain model. The method of the present invention iteratively optimizes the initial random matrix so that the generated coupling matrix gradually approaches the real anatomical connection, realizes non-invasive anatomical connection prediction, and significantly reduces the dependence on invasive structural connection data. Compared with the existing neurodynamic modeling method that relies on dMRI data, the DMF model used achieves a higher degree of fit in structure and function, providing a new direction for the construction of a high-precision digital twin brain model, with greater flexibility and adaptability, helping to reveal the deep connection between functional connections and anatomical connections, and providing more efficient technical support for brain network research. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of an anatomical connection prediction method based on functional connection and dynamic model of the present invention.

[0041] Figure 2 Schematic diagram of the neural dynamics model in an embodiment of the present invention.

[0042] Figure 3 This is a schematic diagram of the result of the optimal coupling matrix predicting anatomical connections in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The method of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0044] like Figure 1 As shown, a flow chart of an anatomical connection prediction method based on functional connection and dynamic model of the present invention, the specific steps are as follows:

[0045] S1. Construct a neural dynamics model, including: input module, DMF module, and output module;

[0046] like Figure 2 As shown, the neural dynamics model adopts a dynamic mean field DMF (Dynamic Mean Field, DMF) model, and the functional connection matrix obtained by simulation is used as output.

[0047] Among them, the input module receives structural connection data as the initial input of the model, which represents the connection strength between different brain regions; the DMF module adopts the DMF model to describe the electrical activities of different brain regions through neurodynamic equations, and converts electrical signals into blood oxygen level dependent (BOLD) signals through transduction functions. By setting appropriate differential equations, the membrane potential changes of each brain region are simulated to reflect the activation state of neurons. The behavior of each brain region is affected by other brain regions. The connection strength (or coupling) is the key parameter in the model, which determines the degree of influence of one brain region on another brain region. Each brain region is described by a node equation, and the interaction between different brain regions is characterized by the input coupling matrix C; the output module generates a simulated functional connection matrix, and uses the brain region activity signals output by the DMF module to calculate the functional connection strength between each brain region, and finally generates a simulated functional connection matrix.

[0048] S2, the input module initializes a random matrix as the initial coupling matrix of the DMF model;

[0049] Initialize a random matrix as the initial coupling matrix C of the model (0) , which is the initial input matrix of the DMF model, and this matrix will serve as the starting point for optimization.

[0050] Each element in the input matrix represents the connection relationship between specific brain regions. Represents the initial structural connectivity between brain regions n and p.

[0051] S3, sending the initial input matrix obtained in step S2 to the DMF module, and converting the electrical signal output by the model into a blood oxygen level dependent BOLD signal;

[0052] In the DMF model, large-scale interconnected populations of excitatory and inhibitory neurons are simplified into two coupled populations (i.e., pools), whose activities are described by a set of simplified dynamic equations. I ) is mediated by GABA receptors, while the excitatory synaptic current (I E ) is mediated by NMDA receptors. Each brain region can be regarded as a coupled system composed of excitatory and inhibitory neuronal populations. The coupling between nodes n and p in different brain regions occurs only between their excitatory populations, and the long-range synaptic connections are mediated by AMPA receptors. In the DMF model, these coupled connections are scaled by structural connections.

[0053] The DMF module uses neurodynamic equations to describe the electrical activities of different brain regions. The DMF model at the whole brain level is represented by a set of coupled differential equations. The specific expression is as follows:

[0054]

[0055]

[0056] in, represents the input current of an excitatory E or inhibitory I neuron population in brain region n, represents the firing rate of a population of excitatory E or inhibitory I neurons in brain area n. represents the average excitatory E or inhibitory I synaptic gating variable in brain region n, and t represents time. And the overall effective external input I0 = 0.382nA, which is represented by the effective coupling constant W in the excitatory pool and inhibitory pool, respectively. E =1 and W I = 0.7 scaling. Local excitability recursion ω + =1.4, G represents the global coupling parameter, C np represents the structural connectivity matrix between brain regions n and p, with all excitatory synaptic coupling J NMDA =0.15nA. represents the current transduction function, which converts the incoming total input current into firing rate. The gains of the transduction function are g E =310nC -1 , g I =615nC -1 , and the other parameters are threshold current Noise factor d E =0.16, d I =0.087. And in the transduction function, it is necessary to ensure that the denominator is not zero, that is, The kinetic parameter γ = 0.641 / 1000 (the factor 1000 means that everything is in milliseconds), and the effective time constants are τ E =τNMDA=100ms and τ I =τGABA=10ms. n represents Gaussian noise with an amplitude of σ = 0.01 nA. In order to better predict the resting-state functional connectivity, each independent node is made to exhibit typical noisy spontaneous activity observed in electrophysiological experiments and have a low firing rate. The feedback inhibition control FIC algorithm is used to adjust the local feedback inhibition weight J of each brain region. n Keep the firing rate of the excitation pool around 3Hz.

[0057] The generalized hemodynamic model is then used to convert the simulated mean field activity in the DMF model into a BOLD signal. When it rises, it triggers vasodilation signal xn Increase, blood flow proportional to the vasodilation signal f n Caused by deoxyhemoglobin content q n and blood volume n The specific expression is as follows:

[0058]

[0059] where k is the signal decay rate, e is the blood flow-dependent elimination rate, ρ is the resting oxygen uptake fraction, α is the venous resistance, and τ is the time constant.

[0060] BOLD signal is the content of deoxyhemoglobin q n and blood volume n The BOLD signal is a static nonlinear function of B, which includes the volume-weighted sum of the extravascular and intravascular signals. For each region n, the BOLD signal is expressed as B n The expression is as follows:

[0061]

[0062] Among them, the resting blood volume fraction V0=0.02, the dimensionless parameters k1=7ρ, k2=2, k3=2ρ-0.2.

[0063] S4, based on the BOLD signal obtained in step S3, outputting the initial simulated functional connectivity matrix through the output module;

[0064] The initial simulated functional connectivity matrix FC was generated by calculating the Pearson correlation coefficient between the simulated BOLD signals of different brain regions. sim , for the BOLD signal time series of brain regions i and j, the Pearson correlation coefficient r ij The calculation expression is as follows:

[0065]

[0066] Among them, T represents the overall time length, B i,t represents the BOLD signal value of brain area i at time t, Represents the average BOLD signal of brain area i.

[0067] The initial simulated functional connectivity matrix is ​​then compared with the real functional connectivity data FC emp For comparison, the Pearson correlation coefficient and mean square error (MSE) of the two are calculated. The MSE calculation expression is as follows:

[0068]

[0069] Where N represents the total number of connections in the matrix.

[0070] Finally, the Pearson correlation coefficient and mean square error were used to evaluate the difference between the simulated functional connectivity and the real functional connectivity data as the optimization objective function.

[0071] S5, based on step S4, using the gradient descent method to update the coupling matrix;

[0072] The gradient descent method is used to adjust the coupling matrix using the difference between the simulated and true functional connection matrices, so that the simulated functional connection gradually approaches the true functional connection. The update expression of the coupling matrix is ​​as follows:

[0073]

[0074] Among them, C (m) represents the coupling matrix of the input of the mth iteration, ∈ represents the learning rate, which controls the update amplitude of each step, and m represents the mth iteration.

[0075] Then use the updated C (m+1) Re-run the DMF model to generate a new BOLD signal, and then calculate the Pearson correlation coefficient of the new BOLD signal to obtain a new simulated functional connectivity matrix The difference between the new simulated functional connectivity and the real functional connectivity data in the current iteration is calculated, and this difference is compared with the difference obtained in the previous iteration. In this embodiment, if the difference no longer decreases (fluctuation is less than 0.1) and is close to the preset fitting accuracy (Pearson correlation coefficient is greater than 0.9, mean square error is less than 0.1), the optimization is considered to have converged; otherwise, the iteration continues until the preset fitting accuracy is reached.

[0076] S6, based on step S5, after the iterative optimization is completed, an optimized coupling matrix, i.e., an optimal coupling matrix, is generated to perform anatomical connection prediction;

[0077] After the iterative optimization is completed, the coupling matrix C that conforms to the neural dynamics characteristics is generated optimized , as the final input of the DMF model, is used to predict anatomical connections and replace anatomical connection data to build a higher-precision brain model.

[0078] In this embodiment, step S7 is also included to evaluate the prediction accuracy, which is as follows:

[0079] Fitting accuracy assessment: Calculate the simulated functional connectivity matrix (FC) of the final model output after optimization sim ) and the true functional connectivity matrix (FC emp ) to evaluate the fitting accuracy, that is, using mean square error (MSE) and Pearson correlation coefficient as quantitative indicators.

[0080] Comparison with anatomical connectivity matrix: The optimized coupling matrix was compared with the true anatomical connectivity matrix to evaluate its predictive ability for anatomical structure.

[0081] Correlation Analysis: Calculating C optimized The Pearson correlation coefficient between the optimized coupling matrix and the true anatomical connection (AC) was used to determine to what extent the anatomical connection characteristics were preserved.

[0082] Binary classification indicators: The area under the receiver operating characteristic (ROC) curve (AUC), specificity, sensitivity and other indicators were used to evaluate the performance of the optimized coupling matrix in predicting anatomical connections. Figure 3 The results of predicting anatomical connections based on the optimal coupling matrix are shown below: Figure 3 (a) is the receiver operating characteristic curve (ROC), Figure 3 (b) is the specificity and sensitivity curve.

[0083] In summary, the method of the present invention aims to solve the problem of lack of high-precision anatomical connection data, and effectively predicts anatomical connections by constructing a connection matrix that conforms to dynamic coupling, thereby realizing a high-precision brain model. The neurodynamic model adopts a dynamic mean field DMF model, and the functional connection matrix obtained by simulation is used as the output. The optimization process of the model is based on the difference between the simulated functional connection and the real functional connection, and the input random matrix is ​​continuously adjusted until the output functional connection matrix and the real functional connection matrix reach the best fit. The coupling matrix finally generated can be used to predict anatomical connections more accurately. Compared with the existing neurodynamic modeling method that relies on dMRI data, the method of the present invention only requires functional connections, and the modeling is simple, so that the model can achieve a higher degree of fit in structure and function, and provides a new direction for the construction of a high-precision digital twin brain model. The method of the present invention has stronger flexibility and adaptability, which helps to reveal the deep connection between functional connection and anatomical connection, and provides more efficient technical support for brain network research.

[0084] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting anatomical connectivity based on functional connectivity and dynamics model, the specific steps are as follows: S1. Construct a neural dynamics model, including: Input module, DMF module, output module; The neural dynamics model adopts a dynamic mean field DMF model, and the functional connection matrix obtained by simulation is used as output; The input module receives structural connection data as the initial input of the model, which represents the connection strength between different brain regions. The DMF module adopts the DMF model to describe the electrical activities of different brain regions through neurodynamic equations, and converts the electrical signals into blood oxygen level-dependent BOLD signals through transduction functions. The output module generates a simulated functional connection matrix, and uses the brain region activity signals output by the DMF module to calculate the functional connection strength between each brain region, and finally generates a simulated functional connection matrix. S2, the input module initializes a random matrix as the initial coupling matrix of the DMF model; Initialize a random matrix as the initial coupling matrix C of the model (0) , which is the initial input matrix of the DMF model, and this matrix will be used as the starting point for optimization; Each element in the input matrix represents the connection relationship between specific brain regions. represents the initial structural connectivity between brain regions n and p; S3, sending the initial input matrix obtained in step S2 to the DMF module, and converting the electrical signal output by the model into a blood oxygen level dependent BOLD signal; The DMF module uses neurodynamic equations to describe the electrical activities of different brain regions. The DMF model at the whole brain level is represented by a set of coupled differential equations. The specific expression is as follows: in, represents the input current of an excitatory E or inhibitory I neuron population in brain region n, represents the firing rate of the excitatory E or inhibitory I neuron population in brain area n; represents the average excitatory E or inhibitory I synaptic gating variable in brain region n, t represents time; and the overall effective external input I0 = 0.382 nA, which is represented by the effective coupling constant W in the excitatory pool and the inhibitory pool, respectively. E =1 and W I = 0.7 scaling; local excitability recursion ω + =1.4, G represents the global coupling parameter, C np represents the structural connectivity matrix between brain regions n and p, with all excitatory synaptic coupling J NMDA =0.15nA; represents the current transduction function, which converts the incoming total input current into the firing rate; the gains of the transduction function are g E =310nC -1 , g I =615nC -1 , and the other parameters are threshold current Noise factor d E =0.16, d I =0.087; and in the transduction function, it is necessary to ensure that the denominator is not zero, that is, The kinetic parameter γ = 0.641 / 1000, the factor 1000 means that everything is in milliseconds, and the effective time constants are τ E =τNMDA=100ms and τ I =τGABA=10ms; v n represents Gaussian noise, whose amplitude is σ=0.01nA; J n represents the local feedback inhibition weight of the brain region, and the feedback inhibition control FIC algorithm is used to adjust the J of each brain region n , so that the firing rate of the excitation pool is kept at a low firing rate, that is The generalized hemodynamic model is then used to convert the simulated mean field activity in the DMF model into a BOLD signal; regulated by a self-regulatory feedback mechanism, when the firing rate of the neuron When it rises, it triggers vasodilation signal x n Increase, blood flow proportional to the vasodilation signal f n Caused by deoxyhemoglobin content q n and blood volume n The specific expression is as follows: Where k represents the signal attenuation rate, e represents the elimination rate dependent on blood flow, ρ represents the resting oxygen uptake fraction, α represents the venous resistance, and τ represents the time constant; BOLD signal is the content of deoxyhemoglobin q n and blood volume n A static nonlinear function of the volume weighted sum of the extravascular and intravascular signals; for each region n, the BOLD signal is expressed as B n The expression is as follows: Among them, resting blood volume fraction V0 = 0.02, dimensionless parameters k1 = 7ρ, k2 = 2, k3 = 2ρ-0.2; S4, based on the BOLD signal obtained in step S3, outputting the initial simulated functional connectivity matrix through the output module; The initial simulated functional connectivity matrix FC was generated by calculating the Pearson correlation coefficient between the simulated BOLD signals of different brain regions. sim , for the BOLD signal time series of brain regions i and j, the Pearson correlation coefficient r ij The calculation expression is as follows: Among them, T represents the overall time length, B i,t represents the BOLD signal value of brain area i at time t, represents the average value of BOLD signal in brain area i; The initial simulated functional connectivity matrix is ​​then compared with the real functional connectivity data FC emp For comparison, the Pearson correlation coefficient and mean square error MSE are calculated. The MSE calculation expression is as follows: Where N represents the total number of connections in the matrix; Finally, the Pearson correlation coefficient and mean square error were used to evaluate the difference between the simulated functional connectivity and the real functional connectivity data as the objective function for optimization; S5, based on step S4, using the gradient descent method to update the coupling matrix; The gradient descent method is used to adjust the coupling matrix using the difference between the simulated and true functional connection matrices, so that the simulated functional connection gradually approaches the true functional connection; the update expression of the coupling matrix is ​​as follows: Among them, C (m) represents the coupling matrix of the input of the mth iteration, ∈ represents the learning rate, which controls the update amplitude of each step, and m represents the mth iteration; Then use the updated C (m+1) Re-run the DMF model to generate a new BOLD signal, and then calculate the Pearson correlation coefficient of the new BOLD signal to obtain a new simulated functional connectivity matrix The difference between the new simulated functional connectivity and the real functional connectivity data in the current iteration is calculated, and this difference is compared with the difference obtained in the previous iteration. If the difference no longer decreases and is close to the preset fitting accuracy, the optimization is considered to have converged; otherwise, the iteration continues until the preset fitting accuracy is reached. S6, based on step S5, after the iterative optimization is completed, an optimized coupling matrix, i.e., an optimal coupling matrix, is generated to perform anatomical connection prediction; After the iterative optimization is completed, the coupling matrix C that conforms to the neural dynamics characteristics is generated optimized , as the final input of the DMF model, is used to predict anatomical connections and replace anatomical connection data to build a higher-precision brain model.

Images