Construction method and system of common space-time brain calculation model

Abstract

The invention provides a construction method and system of a common space-time brain calculation model, and belongs to the technical field of computational neuroscience and artificial intelligence. The problem that an existing model is not enough in brain neurodynamics activity simulation precision is solved. Comprising the following steps: continuously scanning a brain to obtain functional magnetic resonance imaging data of the brain; preprocessing the obtained functional magnetic resonance imaging data to obtain BOLD signals of different brain regions at different moments; constructing a brain calculation model: dividing the BOLD signal into a plurality of transmission intensity matrixes, and extracting a space-time weight matrix containing brain internal common information from the transmission intensity matrixes; according to the space-time weight matrix, simulating a brain activity signal, and calculating a corresponding phase space similarity loss function; determining parameters of an optimal brain calculation model through a phase-space similarity loss function so as to construct a personalized common space-time brain calculation model; the method is applied to brain calculation model construction.

Description

Technical Field

[0001] The present invention provides a method and system for constructing a synchronous spatiotemporal brain computing model, and belongs to the technical fields of computational neuroscience and artificial intelligence. Background Art

[0002] With the advent of the era of artificial intelligence, understanding human behavior and cognitive processes can build smarter and more adaptive intelligent systems. Human cognition mainly comes from the release of electrical and chemical signals between neurons in the brain system, which is also called brain activity. With the development of neuroimaging technology, different recording techniques and computational analysis methods have provided some insights into the study of internal brain activities. Brain computational models based on neurodynamic theory are able to simulate the synchronization patterns between biologically real neuronal populations, thereby directly generating spontaneous or stimulus-induced activities, thereby reproducing the changes in dynamic activities in brain networks and further explaining their mechanisms. Therefore, using this technology to build a virtual brain and simulate and analyze internal activity states creates opportunities for better descriptions of brain dynamics and their relationship to brain function. However, existing brain computational models are limited by observed brain imaging data and are mainly built based on symmetrical and positive-only anatomical structures, which are parameterized by connectivity estimated by white matter integrity from diffusion imaging, which may contain potential erroneous inferences; the loss function in the later stage of the model also only considers the similarity between functional networks, while ignoring the unique dynamic characteristics within the brain. This series of reasons leads to the current simulation results being unsatisfactory and it is impossible to obtain a reliable and interpretable simulated brain model. Summary of the invention

[0003] In order to solve the problem that existing models have insufficient simulation accuracy for brain neural dynamic activities, the present invention proposes a method and system for constructing a synchronous spatiotemporal brain computational model to accurately simulate the brain's neural activities and functions.

[0004] The technical solution adopted by the present invention is: a method for constructing a synchronous spatiotemporal brain computing model, comprising the following steps:

[0005] Step 1: Obtain functional magnetic resonance imaging data of the brain by continuously scanning the brain;

[0006] Step 2: Preprocess the acquired functional magnetic resonance imaging data to obtain the BOLD signals of different brain regions at different times;

[0007] Step 3: Build a computational model of the brain, including:

[0008] Step 3.1: Divide the BOLD signal into multiple transmission intensity matrices, and extract the spatiotemporal weight matrix containing the intrinsic common information of the brain from these transmission intensity matrices;

[0009] Step 3.2: According to the spatiotemporal weight matrix, the brain activity signal is simulated by calculating the relationship between the spontaneous local activities of different brain regions and their output to other brain regions. The process of simulating brain neural activity is as follows:

[0010] Based on the theory of neural dynamics, the spatiotemporal weight matrix constructed above is used as the input of the brain computational model, and the evolution process between each brain region is described by the following differential equation:

[0011]

[0012] Where: η t is the noise picked up by the brain area, describes the spatiotemporal reconstruction of information captured from brain activity, is the relationship between the spontaneous local activity of a brain area and its output to other brain areas;

[0013] The simulated neural activity is converted into a simulated BOLD signal through the hemodynamic response Balloon-Windkessel model;

[0014] Step 4: Collect data based on experience and different The brain activity simulated by the brain computational model under the parameters is used to calculate the corresponding phase space similarity loss function;

[0015] Step 5: Determine the optimal brain computing model parameters through the phase space similarity loss function to build a personalized co-temporal and spatial brain computing model.

[0016] Furthermore, functional magnetic resonance imaging data is collected by an MRI machine.

[0017] Furthermore, the acquired functional magnetic resonance imaging data are preprocessed including: data deletion, image time layer correction, image motion correction, spatial smoothing, filtering, and image covariate regression processing.

[0018] Furthermore, the preprocessed functional magnetic resonance imaging data were partitioned using the brain atlas segmentation template to extract the BOLD signal of each brain region.

[0019] Furthermore, the process of constructing the spatiotemporal weight matrix is ​​as follows: the BOLD signal is divided into multiple time blocks by dividing the sliding time window method, and the Pearson correlation between brain regions in each time block is calculated respectively to construct the corresponding transmission intensity matrix; then the common information of these transmission intensity matrices is extracted through the fusion algorithm of augmented Lagrange multiplier and principal component tracking to construct a spatiotemporal weight matrix containing the intrinsic common information of the brain.

[0020] Furthermore, the process of constructing the phase space similarity loss function is as follows:

[0021] The brain activities simulated in step 3.2 are converted into the phase space corresponding to each brain region, and then the spatial similarity of the phase space between brain regions is measured using the following formula to construct a phase space similarity matrix:

[0022]

[0023] Where: PSA ij is the phase space similarity between the i-th and j-th brain regions, PS i is the phase space diagram of the ith brain region, PS j is the phase space diagram of the jth brain region, is the average value of the corresponding term, σ PS is the standard deviation of the corresponding item, c1=(k 1 L) 2 , c2=(k 2 L) 2 are constants used to maintain stability, where L is the dynamic range of the input image, k 1 =0.01, k 2 =0.03;

[0024] The phase space similarity matrices of the brain activity collected empirically and the brain activity simulated by the Balloon-Windkessel model are calculated respectively, and then the similarity between the two matrices is compared through Pearson correlation to construct the phase space similarity loss function.

[0025] Furthermore, the process of constructing the synchronous spatiotemporal brain computing model is as follows: Genetic algorithm is used to solve different The phase space similarity loss function value of the brain computational model under the parameters is used to find the parameters that are most similar to the empirical BOLD signal and the simulated BOLD signal, and to build a personalized spatiotemporal brain computational model.

[0026] Furthermore, the empirically acquired data in step 4 is functional magnetic resonance imaging data acquired by a magnetic resonance imaging machine.

[0027] A system for constructing a synchronous spatiotemporal brain computing model, comprising:

[0028] An image data acquisition module is used to acquire brain image data collected by a nuclear magnetic resonance machine;

[0029] The preprocessing module is used to preprocess the brain image data to obtain the BOLD signal of each brain region;

[0030] A spatiotemporal weight matrix building module is used to extract the common information in the brain;

[0031] A neural activity simulation module to describe the evolutionary process between each brain region;

[0032] A loss function building block for determining the optimal parameters of the brain computational model and evaluating the accuracy of the simulated BOLD signal;

[0033] The synchronous spatiotemporal brain computing model construction module is used to construct a synchronous spatiotemporal brain computing model based on the determined optimal parameters to simulate accurate and personalized brain activities.

[0034] The beneficial effects of the present invention over the prior art are as follows: compared with the existing brain computing model construction method based on neurodynamic theory, the method and system for constructing a synchronous and spatiotemporal brain computing model proposed by the present invention can simulate neural activities and neurophysiological characteristics that are closer to the real brain, accurately and comprehensively explain the functional characteristics and neural activity mechanisms of the brain, and transform the cross-sectional brain of anatomical biological research into a vivid dynamic brain, so as to study the brain working mechanism in a more systematic and dynamic manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described below in conjunction with the accompanying drawings:

[0036] Figure 1 It is a flow chart of the method for constructing the synchronous spatiotemporal brain computing model provided by the present invention.

[0037] Figure 2 It is a schematic diagram of the construction principle and mathematical representation of the synchronous spatiotemporal brain computing model provided by the present invention.

[0038] Figure 3 It is a structural schematic diagram of the system for constructing the synchronous spatiotemporal brain computing model provided by the present invention.

[0039] Figure 4 It is a comparison chart of simulation signals and empirical data obtained by 30 healthy subjects using the method and system provided by the present invention on multiple evaluation indicators. DETAILED DESCRIPTION

[0040] like Figures 1 to 4 As shown, the present invention provides a method for constructing a synchronous spatiotemporal brain computing model, comprising the following steps:

[0041] Step 1: Obtain functional magnetic resonance imaging (fMRI) data of the brain by continuously scanning the brain;

[0042] Step 2: Preprocess the acquired fMRI data to obtain the BOLD signals of different brain regions at different times. Specifically, the process of preprocessing fMRI data is as follows:

[0043] The first 10 time points of the collected data were removed to ensure the image quality of the subjects;

[0044] Perform time slice correction on the images acquired by the MRI machine to reduce the systematic errors in subsequent data analysis caused by differences in acquisition time;

[0045] Perform head motion correction on the images collected by the MRI machine so that each image can be aligned with the standard template;

[0046] Gaussian kernel function is used to spatially smooth the image to suppress the influence of noise and improve the signal-to-noise ratio of the image;

[0047] Use ICA-AROMA strategy to reduce image noise, effectively remove motion artifacts in functional magnetic resonance imaging data, and improve sensitivity to brain signals;

[0048] The images collected by the MRI machine were bandpass filtered (0.01 Hz-0.09 Hz) to eliminate noise interference;

[0049] Perform covariate regression on the images collected by the MRI machine to remove interference noise from white matter, cerebrospinal fluid, and whole brain mean, and reduce interference from non-neuronal signals;

[0050] The brain map segmentation template is used to partition the images collected by the MRI machine and extract the BOLD signal of each brain region.

[0051] Step 3: Build a computational model of the brain, including:

[0052] Step 3.1: Divide the BOLD signal into multiple transmission intensity matrices by dividing the sliding time window, and extract the spatiotemporal weight matrix containing the intrinsic common information of the brain from these matrices. Specifically, the process of constructing the spatiotemporal weight matrix is ​​as follows:

[0053] The BOLD signal is divided into multiple time blocks by dividing the sliding time window method, and the Pearson correlation between brain regions in each time block is calculated to construct the corresponding transmission intensity matrix; then the common information of these matrices is extracted by the fusion algorithm of augmented Lagrange multiplier and principal component tracking to construct a spatiotemporal weight matrix containing the common information of the brain. This process can be expressed by the following formula:

[0054]

[0055] Where: {X i (t), t=1, ...T, i=1, ..., M} is the collected brain activity, N is the number of time blocks, L n =[l1,...,l N] is a rank The matrix S n =[s1, ..., s N ] is a sparse matrix with non-zero entries, W L is a spatiotemporal weight matrix containing the intrinsic shared information of the brain.

[0056] For the decomposition process, we consider solving the following convex optimization problem:

[0057]

[0058] Where: is the matrix L n The nuclear norm of , that is, the sum of all its singular values, ‖S n ‖1=Σ ij |s nij |It is S n The l1 norm of , and λ1>0 is a regularization parameter that controls L n and S n The trade-off between the two matrices, ‖Bvec(L n )‖1 is regularized by adjusting the parameter λ2>0, which encourages individual-induced group homogeneity by penalizing differences in connectivity profiles. B is a The matrix defined is D, which is the first-order difference matrix It is an M 2 ×M 2 The identity matrix of .

[0059] Step 3.2: According to the spatiotemporal weight matrix, the brain activity signal is simulated by calculating the relationship between the spontaneous local activities of different brain regions (i.e., neuronal groups) and their output to other brain regions. Specifically, the process of simulating brain neural activity is as follows:

[0060] Based on the theory of neural dynamics, the spatiotemporal weight matrix constructed in step 3.2 above is used as the input of the co-temporal brain computational model, and a set of differential equations is used to describe the evolution process between each brain region:

[0061]

[0062] Where: η t is the noise picked up by the brain area, describes the spatiotemporal reconstruction of information captured from brain activity, It is the relationship between the spontaneous local activity of a brain area (i.e. a population of neurons) and its output to other brain areas.

[0063] The simulated neural activity is converted into a simulated BOLD signal through the hemodynamic response Balloon-Windkessel model. In short, the synaptic spike activity S in each brain region i Causes vasodilation signal Z i These signals are influenced by autoregulatory feedback, blood flow f i With signal Z i proportional response, with blood volume v i and deoxyhemoglobin content q i The specific equation is as follows:

[0064]

[0065] Where: S i represents the neural activity signal, ρ = 0.34 is the resting oxygen extraction fraction, k = 0.65 (s -1 ) is the signal attenuation rate, Z i is the vasodilation signal, γ=0.41(s -1 ) is the blood flow-dependent elimination rate, f i is the blood inflow, v i is the blood volume, τ=0.98 (s) represents the hemooxygen dynamics time constant, α=0.32 is the Grubb constant, V0=0.02 is the resting blood volume ratio, and a set of parameter values ​​related to the 3T magnetic field strength are k1=2.38, k2=2 and k3=0.48.

[0066] Step 4: Collect data (fMRI data) based on experience and different The brain activity simulated by the brain computational model under the parameters is used to calculate the corresponding phase space similarity loss function. Specifically, the process of constructing the phase space similarity loss function is as follows:

[0067] The obtained brain activities are converted into the phase space corresponding to each brain region, and then the spatial similarity of the phase space between brain regions is measured using the following formula to construct the phase space similarity matrix:

[0068]

[0069] Where: PSA ij is the phase space similarity between the i-th and j-th brain regions, PS i is the phase space diagram of the ith brain region, PS j is the phase space diagram of the jth brain region, is the average value of the corresponding term, σ PS is the standard deviation of the corresponding item, c1=(k 1 L) 2 , c2=(k 2 L)2 are constants used to maintain stability, where L is the dynamic range of the input image, k 1 =0.01, k 2 =0.03;

[0070] The phase space similarity matrices of the brain activity collected empirically and the brain activity simulated by the brain computational model are calculated respectively, and then the similarity between the two matrices is compared through Pearson correlation to construct a phase space similarity loss function.

[0071] Step 5: Determine the optimal brain computational model parameters through the phase space similarity loss function to build a personalized synchronous spatiotemporal brain computational model. Specifically, the process of building a synchronous spatiotemporal brain computational model is as follows:

[0072] The genetic algorithm is used to solve the phase space similarity loss function value of the brain computing model under different parameters, and the parameters that are most similar to the empirical signal and the simulated signal are found to build a personalized co-temporal and spatial brain computing model.

[0073] Corresponds to Figure 1 The present invention also provides a method for constructing a synchronous spatiotemporal brain computing model, such as Figure 3 As shown, the system includes the following structures:

[0074] An image data acquisition module 201 is used to acquire brain image data collected by a nuclear magnetic resonance machine;

[0075] A preprocessing module 202 is used to preprocess the brain image data to obtain the BOLD signal of each brain region;

[0076] A spatiotemporal weight matrix construction module 203 is used to extract the common information inherent in the brain;

[0077] A neural activity simulation module 204, used to describe the evolution process between each brain region;

[0078] A loss function construction module 205 for determining optimal parameters of the brain computation model and evaluating the accuracy of the simulated signal;

[0079] The synchronous spatiotemporal brain computational model building module 206 is used to simulate accurate and personalized brain activities according to the determined optimal parameters.

[0080] Figure 4 It is a comparison chart of the evaluation indicators (edge ​​FC, node FC, DFC) of the simulated signals and the empirical data obtained by 30 healthy subjects using the method and system provided by the present invention.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a synchronous spatiotemporal brain computing model, characterized in that: The following steps are involved: Step 1: Obtain functional magnetic resonance imaging data of the brain by continuously scanning the brain; Step 2: Preprocess the acquired functional magnetic resonance imaging data to obtain the BOLD signals of different brain regions at different times; Step 3: Build a computational model of the brain, including: Step 3.1: Divide the BOLD signal into multiple transmission intensity matrices, and extract the spatiotemporal weight matrix containing the intrinsic common information of the brain from these transmission intensity matrices; Step 3.2: According to the spatiotemporal weight matrix, the brain activity signal is simulated by calculating the relationship between the spontaneous local activities of different brain regions and their output to other brain regions. The process of simulating brain neural activity is as follows: Based on the theory of neural dynamics, the spatiotemporal weight matrix constructed above is used as the input of the brain computational model, and the evolution process between each brain region is described by the following differential equation: Where: η t is the noise received by the brain region, ζ(W) describes the spatiotemporal reconstruction information captured from brain activity, is the relationship between the spontaneous local activity of a brain area and its output to other brain areas; The simulated neural activity is converted into a simulated BOLD signal through the hemodynamic response Balloon-Windkessel model; Step 4: Collect data based on experience and different The brain activity simulated by the brain computational model under the parameters is used to calculate the corresponding phase space similarity loss function; Step 5: Determine the optimal brain computing model parameters through the phase space similarity loss function to build a personalized co-temporal and spatial brain computing model.

2. The method for constructing a synchronous spatiotemporal brain computing model according to claim 1, characterized in that: Functional magnetic resonance imaging data is collected by an MRI machine.

3. The method for constructing a synchronous spatiotemporal brain computing model according to claim 1, characterized in that: The preprocessing of the acquired functional magnetic resonance imaging data includes: data deletion, image time layer correction, image motion correction, spatial smoothing, filtering, and image covariate regression processing.

4. The method for constructing a synchronous spatiotemporal brain computing model according to claim 3, characterized in that: The preprocessed functional magnetic resonance imaging data were partitioned using the brain atlas segmentation template to extract the BOLD signal of each brain region.

5. The method for constructing a synchronous spatiotemporal brain computing model according to claim 4, characterized in that: The process of constructing the spatiotemporal weight matrix is ​​as follows: the BOLD signal is divided into multiple time blocks by dividing the sliding time window method, and the Pearson correlation between brain regions in each time block is calculated to construct the corresponding transmission intensity matrix; then the common information of these transmission intensity matrices is extracted through the fusion algorithm of augmented Lagrange multiplier and principal component tracking to construct a spatiotemporal weight matrix containing the intrinsic common information of the brain.

6. The method for constructing a synchronous spatiotemporal brain computing model according to claim 5, characterized in that: The process of constructing the phase space similarity loss function is as follows: The brain activities simulated in step 3.2 are converted into the phase space corresponding to each brain region, and then the spatial similarity of the phase space between brain regions is measured using the following formula to construct a phase space similarity matrix: Where: PSA ij is the phase space similarity between the i-th and j-th brain regions, PS i is the phase space diagram of the ith brain region, PS j is the phase space diagram of the jth brain region, is the average value of the corresponding term, σ PS is the standard deviation of the corresponding item, c1=(k 1 L) 2 , c2=(k 2 L) 2 are constants used to maintain stability, where L is the dynamic range of the input image, k 1 =0.01, k 2 =0.03; The phase space similarity matrices of the brain activity collected empirically and the brain activity simulated by the Balloon-Windkessel model are calculated respectively, and then the similarity between the two matrices is compared through Pearson correlation to construct the phase space similarity loss function.

7. The method for constructing a synchronous spatiotemporal brain computing model according to claim 6, characterized in that: The process of constructing the synchronous spatiotemporal brain computing model is as follows: Genetic algorithm is used to solve different The phase space similarity loss function value of the brain computational model under the parameters is used to find the parameters that are most similar to the empirical BOLD signal and the simulated BOLD signal, and to build a personalized spatiotemporal brain computational model.

8. The method for constructing a synchronous spatiotemporal brain computing model according to claim 7, characterized in that: The empirically acquired data in step 4 is functional magnetic resonance imaging data acquired by a magnetic resonance imaging machine.

9. A system for constructing a synchronous spatiotemporal brain computing model, characterized in that: include: An image data acquisition module, used to acquire brain image data collected by a nuclear magnetic resonance machine; The preprocessing module is used to preprocess the brain image data to obtain the BOLD signal of each brain region; A spatiotemporal weight matrix building module is used to extract the common information in the brain; A neural activity simulation module to describe the evolutionary process between each brain region; The loss function construction module is used to determine the optimal parameters of the brain computational model and evaluate the accuracy of the simulated BOLD signal; the synchronous and spatiotemporal brain computational model construction module is used to construct a synchronous and spatiotemporal brain computational model based on the determined optimal parameters to simulate accurate and personalized brain activities.

Images