CN114757334A - Model construction method and device, storage medium and electronic equipment - Google Patents

Abstract

The disclosure discloses a model construction method and device, a storage medium and electronic equipment, which are used for constructing a spiking neural network brain model based on biological brain topological constraint, relate to the technical field of computational neurology and solve the problem that the spiking neural network brain model lacks biological rationality. The model construction method comprises the following steps: performing brain region division on functional nuclear magnetic resonance image data to be processed to obtain M brain region image data; generating M model nodes based on the M brain area image data; generating N model edges based on a correlation coefficient matrix among the M model nodes; based on a preset network topology threshold value, screening the N model edges to obtain S model edges meeting preset conditions; generating topological constraint of a brain-like model based on the biological brain function network based on the M model nodes and the S model edges; and constructing a brain-like model based on topological constraint. The present disclosure can improve the biological rationality of a brain-like model constructed based on a spiking neural network.

Description

Model construction method and device, storage medium and electronic device

technical field

The present disclosure belongs to the technical field of computational neuroscience, and in particular relates to a model construction method and device, a storage medium and an electronic device.

Background technique

In recent years, the subject of computational neuroscience has developed rapidly, and the application of artificial neural network models has become more and more extensive. As a new generation of artificial network model, spiking neural network has powerful processing ability for complex nonlinear spatiotemporal information, plays an important role in the field of computational neurology, and is the necessary theoretical and model basis for computational neuroscience. However, with the development of computational neuroscience toward intelligence, brain-inspired models have emerged, and brain-inspired models based on spiking neural networks lack the constraints of biological brain structure, and the problem of insufficient biological rationality has become increasingly prominent, thus limiting spiking neural networks. Developments in Computational Neuroscience.

SUMMARY OF THE INVENTION

In view of this, the present disclosure provides a model construction method and apparatus, storage medium and electronic device for constructing a spiking neural network brain-like model based on biological brain topology constraints, so as to solve the lack of biological question of rationality.

In a first aspect, an embodiment of the present disclosure provides a model building method for building a spiking neural network brain-like model based on biological brain topology constraints. The model building method includes: dividing the brain regions of the functional magnetic resonance image data to be processed to obtain M brain region image data; and generating M model nodes based on the M brain region image data, wherein the model node representation includes the corresponding model nodes is the brain region of the brain region image data; based on the correlation coefficient matrix between M model nodes, N model edges are generated, where the correlation coefficient matrix is used to represent the brain function network connection strength between the M model nodes; Set the network topology threshold, screen N model edges, and obtain S model edges that meet the preset conditions, where S is a positive integer less than or equal to N; based on M model nodes and S model edges, generate spiking neural networks The network brain-like model is based on the topological constraints of the biological brain function network; based on the topological constraints, a spiking neural network brain-like model is constructed.

With reference to the first aspect, in some implementations of the first aspect, building a spiking neural network brain-like model based on topological constraints, including: generating a spiking neural network brain-like model based on a preset second-order neuron model and M model nodes The network node of the model, wherein the preset second-order neuron model includes the Izhikevich neuron model; based on the network nodes and topology constraints of the spiking neural network brain-like model, the spiking neural network brain-like model is constructed.

In combination with the first aspect, in some implementations of the first aspect, building a spiking neural network brain-like model based on network nodes and topological constraints of the spiking neural network brain-like model, including: based on a preset synaptic plasticity model and S Model edge, generate the network edge of the spiking neural network brain-like model; based on the network edge of the spiking neural network brain-like model, network nodes and topology constraints of the spiking neural network brain-like model, construct the spiking neural network brain-like model.

In combination with the first aspect, in some implementations of the first aspect, the preset synaptic plasticity model includes a synaptic plasticity model co-regulated by excitability and inhibition, and based on the preset synaptic plasticity model and the S models , before generating the network edges of the brain-like model, the method further includes: based on neuroanatomical experimental data, determining the number ratio of excitatory neurons and inhibitory neurons included in the synaptic plasticity model; based on the number ratio, generating preset synapses touch plasticity model.

With reference to the first aspect, in some implementations of the first aspect, the model building method further includes: a preset network topology threshold is determined based on parameters capable of characterizing network topology characteristics. The parameters characterizing the network topology characteristics include at least one of network density, average node degree, small-world attribute and scale-free attribute.

In combination with the first aspect, in some implementations of the first aspect, M is 980, and the functional MRI data to be processed is divided into brain regions to obtain M brain region image data, including: using the Zalesky_980 template, the function to be processed The brain regions were divided according to the MRI impact data, and the image data of 980 brain regions were obtained.

With reference to the first aspect, in some implementations of the first aspect, before dividing the functional magnetic resonance image data to be processed into brain regions to obtain the image data of M brain regions, the method further includes: acquiring the subject's initial Functional magnetic resonance imaging data; preprocessing the initial functional magnetic resonance imaging data to obtain functional magnetic resonance imaging data to be processed. Among them, the preprocessing includes temporal layer correction processing and spatial normalization processing, and the preprocessing also includes head motion correction processing, smoothing processing and filtering processing.

In a second aspect, an embodiment of the present disclosure provides a model building device for building a spiking neural network brain-like model based on biological brain topology constraints, the model building device includes: a brain region division module, used for functional nuclear magnetic resonance to be processed The resonance image data is divided into brain regions to obtain M brain region image data; the first generation module is used to generate M model nodes based on the M brain region image data, wherein the model node representation includes the brain region images corresponding to the model nodes The brain area of the data; the second generation module is used to generate N model edges based on the correlation coefficient matrix between the M model nodes, wherein the correlation coefficient matrix is used to represent the brain function network connection strength between the M model nodes ; the screening module is used to screen the N model edges based on the preset network topology threshold to obtain S model edges that meet the preset conditions, where S is a positive integer less than or equal to N; the third generation module, using Based on the M model nodes and S model edges, the topological constraints based on the biological brain function network of the spiking neural network brain-like model are generated; the building module is based on the topology constraints, and the spiking neural network brain-like model is constructed.

In a third aspect, an embodiment of the present disclosure provides an electronic device, the electronic device includes a processor; and a memory for storing instructions executable by the processor, wherein the processor is configured to execute the method mentioned in the first aspect.

In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, where the storage medium stores a computer program, and the computer program is used to execute the method mentioned in the first aspect.

The model construction method provided by the embodiment of the present disclosure is used to construct a spiking neural network brain-like model based on biological brain topology constraints. Since the topological constraints of the model are determined based on functional magnetic resonance image data, the embodiments of the present disclosure can fully Using the topological characteristics of the biological brain function network, and then using the biological brain function network to construct a spiking neural network for topological constraints, the purpose of improving the biological rationality of the brain-like model based on the spiking neural network is achieved. In combination with the embodiments of the present disclosure, the biological rationality of the brain-like model based on the spiking neural network can be effectively improved, thereby making the spiking neural network more widely used in computational neuroscience.

Description of drawings

The above and other objects, features and advantages of the present disclosure will become more apparent from the more detailed description of the embodiments of the present disclosure in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present disclosure, and constitute a part of the specification, and are used to explain the present disclosure together with the embodiments of the present disclosure, and do not limit the present disclosure.

FIG. 1 is a schematic diagram of an application scenario provided by an embodiment of the present disclosure.

FIG. 2 is a schematic flowchart of a model construction method according to an embodiment of the present disclosure.

FIG. 3 is a schematic flowchart of building a brain-like model of a spiking neural network based on topology constraints according to an embodiment of the present disclosure.

FIG. 4 shows a schematic flowchart of building a spiking neural network brain-like model based on network nodes and topology constraints of the spiking neural network brain-like model according to an embodiment of the present disclosure.

FIG. 5 shows a schematic flowchart of building a spiking neural network brain-like model based on network nodes and topology constraints of the spiking neural network brain-like model provided by another embodiment of the present disclosure.

FIG. 6 is a schematic flowchart of determining a preset network topology threshold according to an embodiment of the present disclosure.

FIG. 7 is a schematic flowchart of a model construction method provided by another embodiment of the present disclosure.

FIG. 8 is a schematic structural diagram of a model building apparatus according to an embodiment of the present disclosure.

FIG. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure.

Detailed ways

The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, but not all of the embodiments.

Computational neuroscience integrates cognitive neuroscience, electrical engineering, information science and computer science and other fields. Its purpose is to explain a series of phenomena related to the biological brain nervous system from a multidisciplinary perspective. The development of systemic disease research plays an important role. With the development of the information industry, more and more attention has been paid to computational neuroscience, which has further promoted the development of information science and brain science. Therefore, the application of artificial neural network models is becoming more and more extensive, and the spiking neural network, as a new generation artificial network model, is different from the traditional artificial neural network model. The subject area plays an important role and is the necessary theoretical and model basis for computational neuroscience. However, with the development of computational neuroscience in the direction of intelligence, brain-inspired models have emerged, and brain-inspired models based on spiking neural networks lack biological brain structure constraints, and the problem of insufficient biological rationality has become increasingly prominent. Developments in Computational Neuroscience.

It can be seen that how to improve the biological rationality of brain-like models based on spiking neural networks is an urgent problem to be solved. In order to solve the above problem, the embodiments of the present disclosure provide a model construction method for constructing a spiking neural network brain-like model, so as to solve the problem that the existing spiking neural network-based brain-like model lacks biological rationality.

The following briefly introduces application scenarios of the embodiments of the present disclosure with reference to FIG. 1 .

FIG. 1 shows a schematic diagram of an application scenario provided by an embodiment of the present disclosure. As shown in Figure 1, this scene is a scene for building a brain-like model of a spiking neural network. Specifically, the scenario for constructing the spiking neural network brain-like model includes a server 110 , a user terminal 120 communicatively connected to the server 110 , and a storage device 130 for functional magnetic resonance image data. model building method. Exemplarily, the server 110 is configured to perform: dividing the functional magnetic resonance image data to be processed into brain regions to obtain M brain region image data; and generating M model nodes based on the M brain region image data, wherein the model nodes represent The brain region containing the image data of the brain regions corresponding to the model nodes; based on the correlation coefficient matrix between the M model nodes, N model edges are generated, wherein the correlation coefficient matrix is used to represent the brain function network connection between the M model nodes Strength; based on the preset network topology threshold, filter N model edges to obtain S model edges that meet the preset conditions, where S is a positive integer less than or equal to N; based on M model nodes and S model edges , generating the topological constraints of the spiking neural network brain-like model based on the biological brain function network; based on the topological constraints of the brain-like model, building the spiking neural network brain-like model.

Exemplarily, in the actual application process, the user uses the user terminal 120 to send an instruction to the server 110 to construct an spiking neural network brain-like model for A. After the server 110 receives the instruction, it calls out the functional magnetic resonance image data of A in the storage device 130, generates a spiking neural network brain-like model for A based on the functional magnetic resonance image data, and then sends it to the user. The terminal 120 outputs the model so that the user terminal 120 can apply the spiking neural network brain-like model.

Exemplarily, the user terminal 120 mentioned above includes, but is not limited to, computer terminals such as desktop computers and notebook computers. The data stored in the above-mentioned storage device 130 includes, but is not limited to, the data in the Neuroimaging Tools & Resources Collaborative (NITRC) public neuroimaging database, and image data such as functional magnetic resonance imaging data input by the user.

The model building method of the present disclosure will be briefly introduced below with reference to FIGS. 2 to 7 .

FIG. 2 is a schematic flowchart of a model construction method provided by an embodiment of the present disclosure. As shown in FIG. 2 , the model construction method provided by the embodiment of the present disclosure includes the following steps.

Step S210, dividing the functional magnetic resonance image data to be processed into brain regions to obtain M brain region image data. Among them, M is a positive integer.

Exemplarily, the selected functional magnetic resonance imaging (Functional Magnetic Resonance Imaging, FMRI) data is selected from the image data of a healthy adult male in the NITRC public neuroimaging database, and then the obtained image data is divided into brain regions. , to obtain image data of M brain regions.

Step S220, based on the image data of the M brain regions, generate M model nodes.

Exemplarily, the model node contains the corresponding brain region image data, that is, each brain region is regarded as a node of the brain function network, and each node represents the corresponding brain region containing functional magnetic resonance imaging (FMRI) data after division. For example, according to the image data of M brain regions, M brain regions are used as nodes of the brain function network, and M network nodes are generated.

Step S230, based on the correlation coefficient matrix between the M model nodes, generate N model edges.

Exemplarily, the correlation coefficient matrix between model nodes is used to represent the brain functional network connection strength between M model nodes, that is, the functional connection strength between nodes is determined by the Pearson correlation coefficient between the average time series of different nodes.

For example, by calculating with the Pearson correlation coefficient correlation formula, a symmetric correlation coefficient matrix of M*M is obtained, wherein the correlation coefficient that is not 0 is 2*N, that is, the edges of N models are obtained. The mathematical expression of the Pearson correlation coefficient is specifically as follows.

和

分别为节点i和节点j的平均时间序列；r_ij为节点i和节点j间的相关系数；T为时间点数。Among them, x _i (t) and x _j (t) are the average time series of node i and node j at time t, respectively;

and

are the average time series of node i and node j, respectively; r _ij is the correlation coefficient between node i and node j; T is the number of time points.

Step S240 , based on a preset network topology threshold, filter N model edges to obtain S model edges that meet preset conditions, where S is a positive integer less than or equal to N.

Exemplarily, the preset network topology threshold is selected as a threshold that conforms to the biological brain network, and the connection status of network nodes is obtained according to the preset network topology threshold. According to the connection of network nodes, N model edges are screened to obtain S model edges that meet the preset network topology threshold.

Step S250 , based on the M model nodes and the S model edges, the topological constraints based on the biological brain function network of the spiking neural network brain-like model are generated.

Exemplarily, according to the M model nodes and the S model edges, a binary matrix can be obtained, and the binary matrix is the topological constraint based on the biological brain function network of the spiking neural network brain-like model.

Step S260, building a brain-like model of a spiking neural network based on topological constraints.

Exemplarily, based on the above topological constraints, a brain-like model based on a spiking neural network is constructed.

In the actual application process, the functional magnetic resonance imaging data to be processed is first divided into brain regions to obtain M brain region image data, and M model nodes are generated based on the M brain region image data, and then M model nodes are generated based on the relationship between the M model nodes. Then, based on the preset network topology threshold, the N model edges are screened to obtain S model edges that meet the preset conditions, and then based on M model nodes and S model edges, Generate the spiking neural network brain-like model based on the topological constraints of the biological brain functional network, and finally build the spiking neural network brain-like model based on the topological constraints.

Since the network topology used in constructing the brain-like model is obtained from functional MRI data, that is, based on the biological brain structure as the network topology constraint, the embodiments of the present disclosure can generate a spike-based neural network based on the biological brain structure as the network topology constraint. The brain-inspired model of SPI enables the spiking neural network to reflect the real brain network connection, making the spiking neural network more biologically rational, and solves the problem that the current brain-like model based on the spiking neural network lacks biological rationality.

FIG. 3 is a schematic flowchart of building a brain-like model of a spiking neural network based on topology constraints according to an embodiment of the present disclosure. On the basis of the embodiment shown in FIG. 2 , the embodiment shown in FIG. 3 is extended, and the following focuses on describing the differences between the embodiment shown in FIG. 3 and the embodiment shown in FIG. 2 , and the similarities will not be repeated.

As shown in FIG. 3 , in the embodiment of the present disclosure, the steps of constructing a brain-like model of a spiking neural network based on topological constraints include the following steps.

Step S310 , based on the preset second-order neuron model and the M model nodes, generate network nodes of the spiking neural network brain-like model.

Exemplarily, the preset second-order neuron model selects the Izhikevich neuron model. The Izhikevich neuron model can well reflect the discharge characteristics of biological neurons and has low time complexity, which is suitable for the construction of large-scale networks. When building the spiking neural network brain-like model, the Izhikevich neuron model is used as the node of the spiking neural network brain-like model. In this embodiment, the selected Izhikevich neuron model includes two firing modes, including regular firing mode and low-threshold firing mode. The regular firing mode simulates excitatory neuron firing, and the low-threshold firing mode simulates inhibitory neuron firing. For the dimensionless parameters in the mathematical model of the Izhikevich neuron model, different values are selected to represent the two firing modes. It can be understood that the preset second-order neuron model includes, but is not limited to, the Izhikevich neuron model.

Exemplarily, the mathematical model of the Izhikevich neuron model is shown below.

if

Among them, VI is the neuron membrane voltage; u is the membrane voltage recovery variable; _I is the sum of the external input current and the current conducted through multiple synapses. a is the time scale of the recovery variable u; b is the sensitivity of the recovery variable u to subdomain fluctuations in membrane voltage; c is the reset value of the membrane voltage induced by the fast high-threshold potassium conductance; d is the slow high-threshold potassium and sodium conductances Caused to restore the reset value of the variable. a, b, c and d are all dimensionless parameters, and various firing patterns of neurons can be simulated by adjusting their values. To sum up, in this embodiment, the excitatory neuron firing is simulated in a regular firing mode, and the parameters are selected as follows: a=0.02, b=0.2, c=-65, d=8; the inhibitory neuron firing is simulated in a low-threshold firing mode , the selected parameters are as follows: a=0.02, b=0.25, c=-65, d=2.

Step S320 , build a spiking neural network brain-like model based on the network nodes and topology constraints of the spiking neural network brain-like model.

Exemplarily, in the brain-like model constructed under the above topological constraints, the Izhikevich neuron model is set as the brain-like model network node to constitute the spiking neural network brain-like model.

Since the Izhikevich neuron model can well reflect the firing characteristics of biological neurons, the selected Izhikevich neuron model can express the two firing modes of biological neurons, and can well express the firing of excitatory neurons and the firing of inhibitory neurons. , which further increases the biological rationality of the constructed brain-like model from the perspective of network nodes. It can be seen that the embodiments of the present disclosure can further increase the biological rationality of the brain-like model based on the spiking neural network in the direction of the nodes of the brain-like model based on topological constraints.

FIG. 4 shows a schematic flowchart of building a brain-like model based on network nodes and topology constraints of the spiking neural network brain-like model provided by an embodiment of the present disclosure. On the basis of the embodiment shown in FIG. 3 , the embodiment shown in FIG. 4 is extended. The following focuses on describing the differences between the embodiment shown in FIG. 4 and the embodiment shown in FIG. 3 , and the similarities will not be repeated.

As shown in FIG. 4 , in the embodiment of the present disclosure, based on the network nodes and topology constraints of the spiking neural network brain-like model, the steps of constructing the spiking neural network brain-like model include the following steps.

Step S410 , based on the preset synaptic plasticity model and the S model edges, generate network edges of the spiking neural network brain-like model.

Exemplarily, the preset synaptic plasticity model is a prominent plasticity model in which excitatory and inhibitory synapses are jointly regulated, and both excitatory synapses and inhibitory synapses can regulate the spiking neural network through changes in synaptic conductance. According to the regulation law, preset the influence parameters of excitatory synapse and inhibitory synapse in both excitation and inhibition states, including reversal potential, excitatory synapse weight, inhibitory synapse weight, excitatory synapse conductance Decay Constant, Decay Constant of Inhibitory Synapse Conductance, Maximum Correction of Excitatory Synaptic Conductance, Minimum Correction of Excitatory Synaptic Conductance, Maximum Correction of Inhibitory Synaptic Conductance and Minimum Correction of Inhibitory Synaptic Conductance value. According to the parameter setting, a synaptic plasticity model is obtained, and the obtained synaptic plasticity model is combined with S model edges to generate edges of the brain-like model.

Illustratively, the synaptic output current has an approximately linear relationship with the input voltage, which is mathematically described as follows.

I _syn = g _syn (t)(EV _j (t))

Among them, I _syn is the synaptic current; g _syn is the synaptic conductance; V _j (t) is the membrane potential of the postsynaptic neuron; E is the reversal potential. In this embodiment, the reversal potential E _ex of excitatory synapses is selected to be 0 mV, and the reversal potential E _in of inhibitory synapses is selected to be -70 mV. Both excitatory synapses and inhibitory synapses achieve regulation of spiking neural networks through changes in synaptic conductance. When postsynaptic neuron j does not receive the action potential of presynaptic neuron i, excitatory synapses and The synaptic conductance of inhibitory synapses decays exponentially, as shown below.

Among them, g _ex represents the excitatory synaptic weight, gin represents the inhibitory synaptic weight; τ _ex and τ _in represent the decay constants of the excitatory synaptic conductance and inhibitory synaptic conductance _, respectively.

When postsynaptic neuron j receives an action potential from presynaptic neuron i, the excitatory synaptic conductance and inhibitory synaptic conductance change as shown below.

和

为由动作电位引起的兴奋性电导增量和抑制性电导增量，分别由兴奋性修正函数w_ij和抑制性修正函数m_ij进行调节。兴奋性修正函数w_ij和抑制性修正函数m_ij的数学描述如下述所示。in,

and

are the excitatory and inhibitory conductance increments caused by action potentials, which are adjusted by the excitatory correction function w _ij and the inhibitory correction function m _ij , respectively. The mathematical description of the excitatory correction function w _ij and the inhibitory correction function m _ij is as follows.

和

的最大值和最小值分别0.015和0；τ₊＝τ_-＝20ms，A₊＝0.1，A_-＝0.105，B₊＝0.02，B_-＝0.003Among them, A ₊ and A _- are the maximum correction value and minimum correction value of excitatory synaptic conductance, respectively; B ₊ and B _- are the maximum correction value and minimum correction value of inhibitory synaptic conductance, respectively. Δt is the firing interval between presynaptic neurons and postsynaptic neurons. τ ₊ and τ- are the time interval ranges of neuron firing during synaptic strengthening and synaptic weakening _, respectively. To sum up, in this embodiment, the selection of parameters is:

and

The maximum and minimum values are 0.015 and 0, respectively; τ ₊ =τ _- =20ms, A ₊ =0.1, A _- =0.105, B ₊ =0.02, B _- =0.003

Step S420 , based on the network edges of the spiking neural network brain-like model, network nodes and topology constraints of the spiking neural network brain-like model, construct the spiking neural network brain-like model.

Exemplarily, in the above-mentioned construction of the brain-like model based on topological constraints and network nodes, the above-mentioned synaptic plasticity model is set as an edge of the brain-like model network to form a spiking neural network brain-like model.

According to the research results of biological synapses, the brain-like model co-regulated by excitatory synapses and inhibitory synapses is more biologically complete. Biological plausibility of spiking neural network brain-like models. It can be seen that, based on the network nodes and topology constraints of the spiking neural network brain-like model, the embodiments of the present disclosure can further increase the biological rationality of the spiking neural network-based brain-like model in the direction of the edges of the spiking neural network brain-like model. sex.

The specific execution process of constructing the spiking neural network brain-like model based on the network nodes and topology constraints of the spiking neural network brain-like model is further described with reference to FIG. 5 .

FIG. 5 shows a schematic flowchart of constructing a brain-like model of a spiking neural network based on nodes and topology constraints of a spiking neural network brain-like model network provided by another embodiment of the present disclosure. The embodiment shown in FIG. 5 is extended on the basis of the embodiment shown in FIG. 4 . The following focuses on the differences between the embodiment shown in FIG. 5 and the embodiment shown in FIG. 4 , and the similarities are not repeated.

As shown in FIG. 5 , in another embodiment of the present disclosure, the preset synaptic plasticity model includes a synaptic plasticity model co-regulated by excitability and inhibition, and the network nodes and topological constraints based on the spiking neural network brain-like model , before the step of constructing the spiking neural network brain-like model, the following steps are also included.

Step S510, based on the experimental data of neuroanatomy, determine the number ratio of excitatory neurons and inhibitory neurons included in the synaptic plasticity model.

Exemplarily, based on the experimental data of neuroanatomy, the ratio of excitatory neurons to inhibitory neurons can be selected to be 4:1 to determine the model ratio of excitatory neurons and inhibitory neurons included in the synaptic plasticity model. .

Step S520, based on the quantity ratio, generate a preset synaptic plasticity model.

Exemplarily, according to the above ratio of excitatory neurons to inhibitory neurons, the models of excitatory neurons and inhibitory neurons in the synaptic plasticity model are randomly distributed at a ratio of 4:1 to generate a preset synaptic plasticity model. .

Since the ratio of the number of excitatory neurons and inhibitory neurons in the synaptic plasticity model is based on the experimental data of neuroanatomy, the synaptic plasticity model is more biologically reasonable. It can be seen from this that the embodiments of the present disclosure can build a spiking neural network brain-like model based on the network nodes and network topology constraints of the spiking neural network brain-like model, and adopt a more biologically rational synaptic plasticity model. The biological rationality of the brain-like model constructed by the network provides a premise.

FIG. 6 is a schematic flowchart of determining a preset network topology threshold according to an embodiment of the present disclosure. On the basis of the embodiment shown in FIG. 2 , the embodiment shown in FIG. 6 is extended. The following focuses on describing the differences between the embodiment shown in FIG. 6 and the embodiment shown in FIG. 2 , and the similarities will not be repeated.

As shown in FIG. 6 , in this embodiment of the present disclosure, the step of determining a preset network topology threshold includes the following steps.

In step S610, the threshold is adjusted within a certain range, and the parameters representing the network topology characteristics generated under different thresholds are obtained.

Exemplarily, the range of the adjustment threshold is selected to be between 0.1 and 0.6, and the step size is set to 0.1. According to different thresholds, different parameters expressing network topology characteristics can be generated. In this embodiment, the parameters representing the network topology characteristics are selected as network density, average node degree, small-world attribute and power-law index, and different network densities, average node-degree, small-world attribute and power-law index are obtained according to different thresholds.

Exemplarily, the network density is obtained according to a definition formula, and the definition formula of the network density is as follows.

Among them, N represents the number of nodes in the network, and L represents the actual number of connections in the network.

Exemplarily, the average node degree is obtained according to a definition formula, and the definition formula of the average node degree is as follows.

where D _i is the degree of node i. The mathematical expression of D _i is as follows.

Among them, h _ij indicates whether there is a connection between node i and node j, if there is a connection, it is 1, and if there is no connection, it is 0.

Exemplarily, the small-world attribute is determined by σ. When σ>1, it means that the network has the small-world attribute, and its mathematical expression is shown below.

Among them, C _real and C _random represent the clustering coefficients of the constructed network and random network, respectively; L _real and L _random represent the shortest path length of the constructed network and random network, respectively.

Step S620: Determine a preset network topology threshold according to the obtained different network topology characteristic parameters.

Exemplarily, according to the characteristics of biological brain topology, the network density is generally in the range of 5%-40%, the network has small-world properties and scale-free properties, the power-law index is around 2 and the clustering coefficient is high, according to The above-mentioned characteristics of the biological brain topology are comprehensively considered in this embodiment, and the selected threshold is 0.2.

Since the selection of the network topology threshold is based on the characteristics of the biological brain network topology, the biological rationality of the spiking neural network brain-like model is increased.

In the embodiment of the present disclosure, the step of dividing the functional MRI data to be processed into brain regions to obtain M brain region image data includes using the Zalesky_980 template to divide the functional MRI data to be processed into brain regions, and obtaining 980 image data of brain regions.

The specific execution process of building the model is further described with reference to FIG. 7 .

FIG. 7 is a schematic flowchart of a model construction method provided by another embodiment of the present disclosure. On the basis of the embodiment shown in FIG. 2 , the embodiment shown in FIG. 7 is extended. The following focuses on describing the differences between the embodiment shown in FIG. 7 and the embodiment shown in FIG. 2 , and the similarities will not be repeated.

As shown in FIG. 7 , in another embodiment of the present disclosure, the following steps are further included before dividing the functional magnetic resonance image data to be processed into brain regions to obtain image data of M brain regions.

Step S710, acquiring initial functional magnetic resonance image data of the subject.

Step S720, preprocessing the initial functional magnetic resonance image data to obtain functional magnetic resonance image data to be processed.

Exemplarily, the initial functional magnetic resonance imaging data is preprocessed, including temporal slice correction processing and spatial normalization processing. According to the actual application, the initial functional magnetic resonance imaging data has a time offset between slices, which requires time slice correction processing. Spatial normalization transforms the original fMRI data into standard MNI space to remove differences in brain shape and size. According to comprehensive consideration, the preprocessing of the initial functional MRI data in this embodiment not only includes temporal layer correction processing and spatial normalization processing, but also performs head motion correction processing, smoothing processing, and filtering processing. During the acquisition of functional MRI image data, subjects will inevitably have head movement, so head movement correction processing is performed to eliminate the influence of head movement on image positioning; in order to reduce the influence of random noise, improve To determine the signal-to-noise ratio of the data, the functional MRI data needs to be smoothed; in order to reduce low-frequency drift and high-frequency physiological noise, the functional MRI data is filtered by selecting a band-pass filter.

The method embodiments of the present disclosure are described in detail above with reference to FIGS. 2 to 7 , and the apparatus embodiments of the present disclosure are described in detail below with reference to FIGS. 8 and 9 . In addition, it should be understood that the descriptions of the method embodiments correspond to the descriptions of the apparatus embodiments, and therefore, for the parts not described in detail, reference may be made to the foregoing method embodiments.

FIG. 8 is a schematic structural diagram of a model building apparatus according to an embodiment of the present disclosure. As shown in FIG. 8 , the model construction apparatus provided by the embodiment of the present disclosure includes a brain region division module 810 , a first generation module 820 , a second generation module 830 , a screening module 840 , a third generation module 850 , and a construction module 860 .

Specifically, the brain region dividing module 810 is used for dividing the brain regions of the functional magnetic resonance image data to be processed to obtain M brain region image data. The first generating module 820 is configured to generate M model nodes based on the M brain region image data, wherein the model nodes represent brain regions including the brain region image data corresponding to the model nodes. The second generation module 830 is configured to generate N model edges based on the correlation coefficient matrix between the M model nodes, wherein the correlation coefficient matrix is used to represent the connection strength of the brain function network between the M model nodes. The screening module 840 is configured to screen the N model edges based on the preset network topology threshold to obtain S model edges that meet the preset conditions, where S is a positive integer less than or equal to N. The third generation module 850 is configured to generate the topological constraints based on the biological brain function network of the spiking neural network brain-like model based on the M model nodes and the S model edges. The building block 860 is used to build a brain-like model of the spiking neural network based on topological constraints.

In some embodiments, the building module 860 is further configured to, based on the preset second-order neuron model and the M model nodes, generate network nodes of the spiking neural network brain-like model, wherein the preset second-order neuron model includes the Izhikevich neural network Metamodel; based on the network nodes and topological constraints of the spiking neural network brain-like model, a spiking neural network brain-like model is constructed.

In some embodiments, the building block 860 is further configured to, based on the synaptic plasticity model and the S model edges, generate network edges of the spiking neural network brain-like model; network edges based on the spiking neural network brain-like model, spiking neural network class The network nodes and topological constraints of the brain model, and the brain-like model of the spiking neural network is constructed.

In some embodiments, the building block 860 is also used to determine a preset synaptic plasticity model. Determining the preset synaptic plasticity model includes: determining the number ratio of excitatory neurons and inhibitory neurons included in the synaptic plasticity model based on neuroanatomical experimental data; and generating a preset synaptic plasticity model based on the number ratio.

In some embodiments, the screening module 840 is further configured to preset the determination of the network topology threshold. Specifically, the network threshold is determined according to a parameter capable of characterizing the network topology characteristic, wherein the parameter characterizing the network topology characteristic includes at least one of network density, average node degree, small-world attribute and scale-free attribute.

In some embodiments, the first generating module 820 is further configured to generate M pieces of image data with a preset M of 980. The preset M is 980, and M pieces of image data are generated, including: dividing the brain regions with the functional magnetic resonance image data to be processed, and obtaining 980 brain regions image data.

In some embodiments, the first generating module 820 is further configured to generate functional magnetic resonance image data to be processed. Specifically, the initial functional magnetic resonance imaging data of the subject is obtained; the initial functional magnetic resonance imaging data is preprocessed to obtain the functional magnetic resonance imaging data to be processed, wherein the preprocessing includes temporal layer correction processing and spatial standardization Processing, preprocessing also includes head movement correction processing, smoothing processing and filtering processing.

FIG. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. The electronic device 900 shown in FIG. 9 (the electronic device 900 may specifically be a computer device) includes a memory 901 , a processor 902 , a communication interface 903 and a bus 904 . The memory 901 , the processor 902 , and the communication interface 903 are connected to each other through the bus 904 for communication.

The memory 901 may be a read only memory (Read Only Memory, ROM), a static storage device, a dynamic storage device, or a random access memory (Random Access Memory, RAM). The memory 901 may store a program, and when the program stored in the memory 901 is executed by the processor 902, the processor 902 and the communication interface 903 are used to execute various steps of the model building method of the embodiment of the present disclosure.

The processor 902 may use a general-purpose central processing unit (Central Processing Unit, CPU), a microprocessor, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a graphics processing unit (Graphics Processing Unit, GPU), or one or more The integrated circuit is used to execute the relevant program, so as to realize the functions required to be performed by each unit in the model building apparatus of the embodiment of the present disclosure.

The processor 902 may also be an integrated circuit chip with signal processing capability. In the implementation process, each step of the model building method of the present disclosure may be completed by hardware integrated logic circuits in the processor 902 or instructions in the form of software. The above-mentioned processor 902 may also be a general-purpose processor, a digital signal processor (Digital Signal Processing, DSP), an application specific integrated circuit (ASIC), a Field Programmable Gate Array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components. The disclosed methods, steps and logical block diagrams in the embodiments of the present disclosure can be implemented or executed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the methods disclosed in conjunction with the embodiments of the present disclosure may be directly embodied as executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module may be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other storage media mature in the art. The storage medium is located in the memory 901, and the processor 902 reads the information in the memory 901 and, in combination with its hardware, completes the functions required to be performed by the units included in the model building apparatus of the embodiment of the present disclosure, or executes the model building of the method embodiment of the present disclosure method.

The communication interface 903 implements communication between the electronic device 900 and other devices or a communication network using a transceiving device such as, but not limited to, a transceiver. For example, the functional magnetic resonance imaging data signal can be acquired and processed through the communication interface 903 .

Bus 904 may include a pathway for communicating information between various components of electronic device 900 (eg, memory 901, processor 902, communication interface 903).

It should be noted that although the electronic device 900 shown in FIG. 9 only shows a memory, a processor, and a communication interface, in the specific implementation process, those skilled in the art should understand that the electronic device 900 also includes necessary components for normal operation. other devices. Meanwhile, according to specific needs, those skilled in the art should understand that the electronic device 900 may further include hardware devices that implement other additional functions. In addition, those skilled in the art should understand that the electronic device 900 may also only include the necessary devices for implementing the embodiments of the present disclosure, and does not necessarily include all the devices shown in FIG. 9 .

Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each particular application, but such implementations should not be considered beyond the scope of this disclosure.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described systems, devices and units may refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

In the several embodiments provided by the present disclosure, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the device embodiments described above are only illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated. to another system, or some features can be ignored, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as independent products, may be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present disclosure can be embodied in the form of software products in essence, or the parts that make contributions to the prior art or the parts of the technical solutions. The computer software products are stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure. The aforementioned storage medium includes: a U disk, a removable hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk and other media that can store program codes.

The above are only specific embodiments of the present disclosure, but the protection scope of the present disclosure is not limited to this. should be included within the scope of protection of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims (10)

1. a model building method is characterized in that, for building the spiking neural network brain-like model based on biological brain topology constraints, the method comprises: The functional magnetic resonance imaging data to be processed is divided into brain regions, and the image data of M brain regions are obtained; generating M model nodes based on the M brain region image data, wherein the model nodes represent brain regions including the brain region image data corresponding to the model nodes; generating N model edges based on the correlation coefficient matrix between the M model nodes, wherein the correlation coefficient matrix is used to represent the connection strength of the brain function network between the M model nodes; Based on a preset network topology threshold, the N model edges are screened to obtain S model edges that meet the preset conditions, where S is a positive integer less than or equal to N; generating, based on the M model nodes and the S model edges, a topology constraint based on a biological brain function network of the spiking neural network brain-like model; Based on the topological constraints, the spiking neural network brain-like model is constructed. 2. The model construction method according to claim 1, wherein the building the spiking neural network brain-like model based on the topology constraint comprises: generating a network node of the spiking neural network brain-like model based on a preset second-order neuron model and the M model nodes, wherein the preset second-order neuron model includes an Izhikevich neuron model; Based on the network nodes of the spiking neural network brain-like model and the topological constraints, the spiking neural network brain-like model is constructed. 3. The method for constructing a model according to claim 2, wherein, building the spiking neural network brain-like model based on the network nodes of the spiking neural network brain-like model and the topology constraints, comprising: generating network edges of the spiking neural network brain-like model based on the preset synaptic plasticity model and the S model edges; Based on the network edges of the spiking neural network brain-like model, the network nodes of the spiking neural network brain-like model, and the topology constraints, the spiking neural network brain-like model is constructed. 4. The model construction method according to claim 3, wherein the preset synaptic plasticity model comprises a synaptic plasticity model co-regulated with excitability and inhibition, and the preset synaptic plasticity model based on the preset synaptic plasticity model and the S model edges, before generating the network edge of the brain-like model, further comprising: Determine the number ratio of excitatory neurons and inhibitory neurons included in the synaptic plasticity model based on neuroanatomical experimental data; Based on the quantity scale, the preset synaptic plasticity model is generated. 5. The model construction method according to any one of claims 1 to 4, wherein the preset network topology threshold is determined based on parameters capable of characterizing network topology characteristics, wherein the parameters characterizing network topology characteristics include At least one of network density, average node degree, small-world property, and scale-free property. 6. The model construction method according to any one of claims 1 to 4, wherein M is 980, and the functional magnetic resonance image data to be processed is divided into brain regions to obtain M brain region images data, including: Using the Zalesky_980 template, the functional magnetic resonance imaging data to be processed is divided into brain regions to obtain 980 brain regions image data. 7. The model construction method according to any one of claims 1 to 4, wherein, before the brain region division is performed on the functional nuclear magnetic resonance image data to be processed to obtain M brain region image data, Also includes: Obtain the subject's initial functional magnetic resonance imaging data; Preprocessing the initial functional magnetic resonance imaging data to obtain the functional magnetic resonance imaging data to be processed, wherein the preprocessing includes temporal layer correction processing and spatial normalization processing, and the preprocessing further includes head movement Correction, smoothing, and filtering. 8. A model building device, characterized in that, for building a spiking neural network brain-like model based on biological brain topology constraints, the device comprising: The brain area division module is used to divide the brain area of the functional magnetic resonance image data to be processed, and obtain the image data of M brain areas; a first generation module, configured to generate M model nodes based on the M brain region image data, wherein the model nodes represent brain regions that include the brain region image data corresponding to the model nodes; The second generation module is configured to generate N model edges based on the correlation coefficient matrix between the M model nodes, wherein the correlation coefficient matrix is used to represent the brain function network connection between the M model nodes strength; a screening module, configured to screen the N model edges based on a preset network topology threshold to obtain S model edges that meet the preset conditions, where S is a positive integer less than or equal to N; a third generation module, configured to generate the topological constraints based on the biological brain function network of the spiking neural network brain-like model based on the M model nodes and the S model edges; A building block for building the spiking neural network brain-like model based on the topological constraints. 9. An electronic device, characterized in that, comprising: processor; memory for storing said processor-executable instructions, Wherein, the processor is configured to execute the model building method according to any one of claims 1 to 7. 10 . A computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute the model construction method according to any one of claims 1 to 7 . 11 .

Images