Brain disease structure-function coupling analysis method based on dynamic pulse neural network

Abstract

The invention provides a brain disease structure-function coupling analysis method based on a dynamic spiking neural network, belongs to the technical field of intelligent auxiliary medical diagnosis, and effectively solves the technical problem that an existing structure-function coupling analysis method is insufficient in the aspects of capturing brain dynamic changes and nonlinear coupling characteristics and integrating multi-modal data. According to the technical scheme, firstly, a dynamic pulse neural network model is constructed; extracting a pulse structure connection matrix from the original fMRI time sequence; then extracting a pulse function connection matrix from the function brain network, and inputting the pulse function connection matrix into a pulse coupling pooling module to obtain pulse structure-function coupling; and finally, inputting the obtained coupling information into a classification layer to obtain a disease recognition result. The method has the advantages that a brand new analysis tool and view angle are provided for brain disease research, deep understanding of brain disease neurological mechanisms is promoted, and the method has wide application potential in clinical application.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent auxiliary medical diagnosis, and in particular to a brain disease structure-function coupling analysis method based on a dynamic pulse neural network. Background Art

[0002] Brain diseases affect millions of people worldwide and include a variety of diseases and disorders, ranging from stroke and Alzheimer's disease to multiple sclerosis, epilepsy, traumatic brain injury, etc., and are important medical conditions with profound impacts on human health. Recent studies on functional and structural networks of brain diseases have revealed specific disruptions in connectivity and network topology, thus leading to a shift in modern brain disease research from "focus" to "network". Analyzing brain networks can provide a new approach to explore the association between brain functional deficits and related underlying structural damage.

[0003] Structural connectivity (SC) is the physical inter-regional pathways in the brain, which is usually derived from magnetic resonance imaging (MRI) or diffusion tensor imaging (DTI), and functional connectivity (FC) is the temporal co-activation between regional brain activities, which is constructed from electroencephalogram (EEG) data and functional magnetic resonance imaging (fMRI) data. There is a tight coupling between SC and FC, which is reflected not only in the strength of inter-regional connections, but also in the pattern of connectivity that supports multiple cognitive functions. SC-FC coupling is an emerging comprehensive measure that links functional and structural networks and has been recognized as a more sensitive modality for detecting subtle changes in brain activity than any single modality index.

[0004] There are some significant deficiencies in the current structure-function coupling analysis, which restricts its in-depth application in the field of brain science. First, the existing methods are mostly based on linear or simple nonlinear models, which are difficult to fully capture the complex nonlinear coupling relationship between brain function and structure. As a complex dynamic system, the structural connections of the brain often affect functional activities through multiple levels, time delays and dynamic processes, and these characteristics are often simplified or ignored. Secondly, many analysis methods are based on static structural networks and functional connection matrices, ignoring the dynamic characteristics of brain function over time, and cannot effectively reveal the time-varying characteristics of coupling relationships under different tasks or states.

[0005] The paper Relationships between brain structure-function coupling in normalaging and cognition: A cross-ethnicity population-based study proposed a novel quantitative index, namely the brain structure connectome and functional connectome coupling (C-BSF) index. This index is constructed by evaluating the topological similarity between structural networks and functional networks, providing a new perspective and method for exploring the complex relationship between brain structure and function. However, the C-BSF index also has certain limitations. For example, it may be more suitable for capturing static structure-function coupling characteristics, but it is relatively insufficient in revealing dynamic changes in the brain and changes in real-time functional connectivity.

[0006] The paper Representing Functional Connectivity with Structural Detour: A New Perspective to Decipher Structure-Function Coupling Mechanism proposed a new connectomics representation, detour connection (DC), which characterizes direct functional connection (FC) by weighted connection strength on directed structural connection (SC) paths, and also designed a Transformer-style graph neural network framework to effectively quantify SC-FC coupling through a self-attention mechanism. Although this method can provide quantitative measures of structure-function coupling, the neurobiological mechanisms behind these measures may be difficult to interpret, especially in complex neural networks.

[0007] Therefore, by utilizing the neurobiological mechanisms of dynamic spiking neurons to simulate neuronal dynamics in the brain, we can more deeply study and analyze the complex mechanisms of brain structure-function coupling, thereby significantly improving the accuracy of brain disease diagnosis. Summary of the invention

[0008] In view of the shortcomings of existing structure-function coupling analysis methods in capturing dynamic changes in the brain, nonlinear coupling characteristics and integrating multimodal data, the purpose of the present invention is to provide a brain disease structure-function coupling analysis method based on a dynamic spiking neural network to improve the accuracy of brain disease identification.

[0009] The inventive concept of the present invention is: by combining the multimodal data of brain structural networks and functional networks, the bio-inspired characteristics of dynamic spiking neural networks are used to simulate the dynamic behavior of neurons to capture the complex relationship between brain structure-function coupling. Traditional structure-function coupling analysis methods often use static or linear models, which are difficult to fully characterize the nonlinear dynamic characteristics of brain networks. As a computational model that can simulate biological nervous systems, dynamic spiking neural networks have the unique advantage of capturing time-varying activity patterns and nonlinear coupling relationships between neurons.

[0010] In order to better achieve the above-mentioned invention object, the present invention is implemented by the following technical scheme: a brain disease structure-function coupling analysis method based on a dynamic spiking neural network, the specific steps are as follows:

[0011] A. A dynamic spiking neural network (DSNN) model is constructed using the Leaky Integrate-and-Fire spiking neuron with infinite impulse response (IIR) filter proposed by Fang et al. Let Axon(·) represent the operation of transmitting the pulse to the next neuron, let represents the behavior of neurons updating membrane potential and emitting pulses. Therefore, the DSNN model simulated by IIR filter can be divided into Axon(·) and operate;

[0012] B. Extract the original functional time series information from fMRI, Represents the time series matrix of fMRI, N corresponds to the number of node sets in the brain region, T corresponds to the length of the time series, and G fc As the feature input of DSNN, through Axon(·) and Operations were performed to obtain the spike functional connectivity (FC) matrix;

[0013] C. Extracting structural brain networks from diffusion tensor imaging DTI, let Represents the adjacency matrix of the structural brain network, N corresponds to the number of node sets in the brain region, and then the edge-to-edge convolution E2E and edge-to-node convolution E2N operations proposed by Kawahara are used to obtain the structural brain network information feature map. The feature map of the structural brain network after E2E convolution is recorded as The feature map after E2N convolution is Among them, C1 and C2 represent the number of filters of the two convolution operations. Finally, the structural brain network feature map G after E2E and E2N convolution operations is sc As the structural input of DSNN, through the linear network and Operation to obtain the impulse structure connection (SC) matrix;

[0014] D. Construct a pulse coupled pooling (SCP) module, input the pulse FC and SC matrices into the SCP module to obtain the pulse structure-function (SC-FC) coupling, and finally pass the linear network and Operation, we can get the pulses emitted by each output neuron over time It is a time-varying vector where m is the number of output categories;

[0015] E. Input to the classification layer, calculate the maximum pulse count of each output neuron to obtain the predicted category, and use the cross entropy loss between the predicted category and the true category as the loss function for training, to obtain an accurate brain disease recognition model based on pulse structure-function coupling.

[0016] As the brain disease structure-function coupling analysis method based on dynamic spiking neural network provided by the present invention, the specific steps of step B are as follows:

[0017] a. The original fMRI time series matrix G fc As the feature input of DSNN, the formula is as follows:

[0018]

[0019] Where Axon(·) represents the operation of transmitting the pulse to the next neuron, l is the layer index, and when l = 1, Y l-1 =G fc ;

[0020] b. Extract the pulse FC matrix using the dynamic pulse metamodel. The formula is as follows:

[0021]

[0022] in, It represents the behavior of neurons updating membrane potential and emitting pulses. The final output pulse FC matrix is

[0023] As the brain disease structure-function coupling analysis method based on dynamic spiking neural network provided by the present invention, the specific steps of step C are as follows:

[0024] a. The SC feature map G after E2E and E2N convolution operations sc As the structural input of DSNN, implicit pulse encoding is performed through internal pulse neurons. The formula is as follows:

[0025]

[0026] Where D is the feature dimension and T is the time step. It is worth noting that the choice of T is determined by the length of the fMRI time series.sc By G sc After T repetitions, we get through the linear network, It is through So it is a binarized impulse tensor;

[0027] b. Extract the pulse SC matrix using the linear network and dynamic pulse metamodel. The formula is as follows:

[0028]

[0029] Among them, X sc Represents the final output pulse SC matrix;

[0030] As the brain disease structure-function coupling analysis method based on dynamic spiking neural network provided by the present invention, the specific steps of step D are as follows:

[0031] a. The functional and structural inputs of SCP can be described as Y fc =(y0,...,y t ,...,y T ), and X sc =(x0,...,x t ,...,x T ), For ease of explanation, we will focus on the processing at time t (t∈[0,T]). Specifically, given the FC feature vector y at time t t and SC eigenvector x t , the simplest bilinear transformation is defined as:

[0032]

[0033] in, is the projection matrix, and (tr) represents the transpose operator. In order to obtain an o-dimensional output, we need to learn This results in a large number of parameters.

[0034] b. Decompose W into two low-rank matrices:

[0035]

[0036] Among them, k is the decomposition rank, is the element-wise multiplication of two eigenvectors, is an all-one vector, Therefore, to obtain an output feature vector

[0037] Two 3D tensors are required and Without loss of generality, a simple reshaping operation can be used to convert and Re-expressed as two-dimensional matrices and

[0038] c. Final fusion vector It is obtained by performing sum pooling of a one-dimensional non-overlapping window of size k on the projection vector using P′ and Q′:

[0039]

[0040] The above operation generates a feature vector that combines FC temporal information and SC spatial information. It only represents the coupling relationship between FC and SC at time t. By repeating operations a, b, and c T times, we can get a pulse matrix consisting of T eigenvectors: This is called pulse SC-FC coupling, which not only contains the temporal FC information, but also retains the spatial SC information, comprehensively characterizing the dynamic coupling relationship between FC and SC.

[0041] d. Finally, through the linear network and Operation, we can get the pulses emitted by each output neuron over time

[0042]

[0043] Where m is the number of output categories, is a vector that varies with time.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. Compared with most previous structure-function coupling analysis models, the brain disease structure-function coupling analysis method based on dynamic pulse neural network proposed in the present invention can more effectively capture the nonlinear and dynamic coupling characteristics between brain structure and function, overcome the limitations of traditional methods in static and linear modeling, and thus provide a more accurate and comprehensive analysis of brain coupling mechanisms.

[0046] 2. The brain disease structure-function coupling analysis method based on dynamic spiking neural network proposed in the present invention innovatively integrates multimodal data from fMRI and DTI, and combines the advantages of spiking neural network to effectively improve the accuracy and robustness of data integration. It can comprehensively reflect the structural and functional interactions of the brain at different time scales, and improve the accuracy of brain disease identification.

[0047] 3. The brain disease structure-function coupling analysis method based on dynamic pulse neural network proposed in the present invention provides a new technical means and theoretical framework, which can deeply reveal the structure-function coupling mechanism of the brain, has strong clinical transformation potential, and can play an important role in the early diagnosis, prediction and treatment of brain diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention but do not constitute a limitation of the present invention.

[0049] Figure 1 It is an overall framework diagram of the brain disease structure-function coupling analysis method based on dynamic spiking neural network of the present invention;

[0050] Figure 2 This is a model diagram of the brain disease structure-function coupling analysis method based on dynamic pulse neural network of the present invention. DETAILED DESCRIPTION

[0051] In order to deepen the understanding of the present invention, the present invention will be further described in detail below in conjunction with embodiments and drawings. The embodiments are only used to explain the present invention and do not constitute a limitation on the protection scope of the present invention.

[0052] Example 1

[0053] See also Figure 1 , Figure 2 This embodiment is specifically implemented through the following technical solutions:

[0054] like Figure 1 and Figure 2 As shown in the figure, the brain disease structure-function coupling analysis method based on dynamic spiking neural network has the following specific steps:

[0055] A. A dynamic spiking neural network (DSNN) model is constructed using the Leaky Integrate-and-Fire spiking neuron with infinite impulse response (IIR) filter proposed by Fang et al. Let Axon(·) represent the operation of transmitting the pulse to the next neuron, let represents the behavior of neurons updating membrane potential and emitting pulses. Therefore, the DSNN model simulated by IIR filter can be divided into Axon(·) and operate;

[0056] B. Extract the original functional time series information from fMRI, Represents the time series matrix of fMRI, N corresponds to the number of node sets in the brain region, T corresponds to the length of the time series, and G fcAs the feature input of DSNN, through Axon(·) and Operation to obtain the pulse functional connectivity (FC) matrix; the specific steps are as follows:

[0057] a. The original fMRI time series matrix G fc As the feature input of DSNN, the formula is as follows:

[0058]

[0059] Where Axon(·) represents the operation of transmitting the pulse to the next neuron, l is the layer index, and when l = 1, Y l-1 =G fc ;

[0060] b. Extract the pulse FC matrix using the dynamic pulse metamodel. The formula is as follows:

[0061]

[0062] in, It represents the behavior of neurons updating membrane potential and emitting pulses. The final output pulse FC matrix is

[0063] The number of samples in the brain disease dataset is 217, and the decision-making attributes are divided into two categories, namely, patients diagnosed with brain diseases and normal people. The fMRI time series matrix is The final generated pulse FC matrix is

[0064] C. Extracting structural brain networks from diffusion tensor imaging DTI, let Represents the adjacency matrix of the structural brain network, N corresponds to the number of node sets in the brain region, and then the edge-to-edge convolution E2E and edge-to-node convolution E2N operations proposed by Kawahara are used to obtain the structural brain network information feature map. The feature map of the structural brain network after E2E convolution is recorded as The feature map after E2N convolution is Among them, C1 and C2 represent the number of filters of the two convolution operations. Finally, the structural brain network feature map G after E2E and E2N convolution operations is sc As the structural input of DSNN, through the linear network and Operation to obtain the impulse structure connection (SC) matrix; the specific steps are as follows:

[0065] a. The SC feature map G after E2E and E2N convolution operations sc As the structural input of DSNN, implicit pulse encoding is performed through internal pulse neurons. The formula is as follows:

[0066]

[0067] Where D is the feature dimension and T is the time step. It is worth noting that the choice of T is determined by the length of the fMRI time series. sc By G sc After T repetitions, we get through the linear network, It is through So it is a binarized impulse tensor;

[0068] b. Extract the pulse SC matrix using the linear network and dynamic pulse metamodel. The formula is as follows:

[0069]

[0070] Among them, X sc Represents the final output pulse SC matrix;

[0071] The connection matrix of the structural brain network is The E2E convolutional layer consists of 32 1×90 and 32 90×1 filters, and the generated structural feature map is The E2N convolutional layer consists of 64 1×90×32 filters, and the generated structural feature map is

[0072] D. Construct a pulse coupled pooling (SCP) module, input the pulse FC and SC matrices into the SCP module to obtain the pulse structure-function (SC-FC) coupling, and finally pass the linear network and Operation, we can get the pulses emitted by each output neuron over time It is a time-varying vector where m is the number of output categories; the specific steps are as follows:

[0073] a. The functional and structural inputs of SCP can be described as Y fc =(y0,...,y t ,...,y T ), and X sc =(x0,...,x t ,...,x T ), For ease of explanation, we will focus on the processing at time t (t∈[0,T]). Specifically, given the FC feature vector y at time t t and SC eigenvector x t , the simplest bilinear transformation is defined as:

[0074]

[0075] in, is the projection matrix, and (tr) represents the transpose operator. In order to obtain an o-dimensional output, we need to learn This results in a large number of parameters.

[0076] b. Decompose W into two low-rank matrices:

[0077]

[0078] Among them, k is the decomposition rank, is the element-wise multiplication of two eigenvectors, is an all-one vector, Therefore, to obtain an output feature vector

[0079] Two 3D tensors are required and Without loss of generality, a simple reshaping operation can be used to convert and Re-expressed as two-dimensional matrices and

[0080] c. Final fusion vector It is obtained by performing sum pooling of a one-dimensional non-overlapping window of size k on the projection vector using P′ and Q′:

[0081]

[0082] The above operation generates a feature vector that combines FC temporal information and SC spatial information. It only represents the coupling relationship between FC and SC at time t. By repeating operations a, b, and c T times, we can get a pulse matrix consisting of T eigenvectors: This is called pulse SC-FC coupling, which not only contains the temporal FC information, but also retains the spatial SC information, comprehensively characterizing the dynamic coupling relationship between FC and SC.

[0083] d. Finally, through the linear network and Operation, we can get the pulses emitted by each output neuron over time

[0084]

[0085] Where m is the number of output categories, is a vector that varies with time.

[0086] E. Input to the classification layer, calculate the maximum pulse count of each output neuron to obtain the predicted category, and use the cross entropy loss between the predicted category and the true category as the loss function for training, to obtain an accurate brain disease recognition model based on pulse structure-function coupling.

[0087] Using the five-fold cross-validation method, the model was trained and tested independently for five rounds, with different training sets, test sets, and validation sets divided in each round. In these rounds, the highest accuracy recorded was 97.89%, and the comprehensive average accuracy was 93.64%. These data show that the model can significantly improve the accuracy of brain disease identification.

[0088] Example 2

[0089] This implementation example is trained on an NVIDIA RTX 3090 GPU graphics card. For comparison, this embodiment selects 6 comparison models, such as Statistics, MPCA, Self-calibrated, MMP-GCN, Brain-GRL, and Brain-DNN.

[0090] In order to highlight the advantages of the model in the operating efficiency of the present invention, the batch size will be continuously increased until the graphics card memory usage exceeds the limit, and the model with the best test results will be recorded. The initial batch size is set to 32, the epoch is 100, and the learning rate is 0.0001. The performance of the model is evaluated using the following indicators: accuracy (ACC), sensitivity (SEN) and specificity (SPE). The experimental results are shown in Table 1:

[0091] Table 1 Experimental results of the technology proposed by the present invention and other technologies

[0092]

[0093] It can be concluded from Table 1 that the DSNN model proposed in the present invention has achieved obvious advantages in the identification of brain diseases. It is not only superior to the existing methods in multiple evaluation indicators, but also provides a new perspective for the study of SC-FC coupling.

[0094] Example 3

[0095] This embodiment will introduce an applicable scenario of the present invention:

[0096] The coupling between brain structural connectivity (SC) and functional connectivity (FC) (SC-FC coupling) is the core mechanism of brain information processing and functional integration. In recent years, studies have found that abnormalities in SC-FC coupling are closely related to the occurrence and development of a variety of brain diseases, such as neuropsychiatric diseases such as epilepsy, Alzheimer's disease and schizophrenia. Therefore, a deep understanding of the mechanism of SC-FC coupling is crucial for the early diagnosis and prediction of brain diseases. However, existing SC-FC coupling analysis methods mostly focus on revealing the static association between SC and FC, ignoring the dynamic characteristics of neural activity, especially the pulse timing of information transmission between brain regions.

[0097] In order to solve this problem, the present invention proposes a brain disease structure-function coupling analysis method based on dynamic spiking neural network (DSNN). By quantifying the dynamic characteristics of SC-FC coupling, DSNN can help doctors evaluate the brain function status of different patients, and then provide support for personalized treatment. For example, in the treatment process of schizophrenia patients, DSNN can monitor the effects of drugs or therapies on brain networks, evaluate the treatment effect, and adjust the treatment plan according to the individual's brain network state to improve the effectiveness of treatment. DSNN can also be used in brain-computer interface systems and neurofeedback training. By analyzing the dynamic characteristics of SC-FC coupling between brain regions in real time, DSNN can provide more accurate real-time data for neurofeedback, thereby improving the individual's brain function performance in specific tasks, such as attention training, emotion regulation, etc.

[0098] In short, the practical application range of the DSNN model is very wide, covering a wide range of fields from clinical brain disease diagnosis to personalized treatment, neuromodulation, brain imaging analysis, etc. By more accurately capturing and analyzing the brain's pulse neural activity, it provides innovative ideas and technical support for solving many challenges in current brain science research and neurological disease treatment.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for analyzing brain disease structure-function coupling based on dynamic spiking neural network, characterized in that: The following steps are involved: S1. A dynamic spiking neural network DSNN model is constructed using a Leaky Integrate-and-Fire spiking neuron with an infinite impulse response IIR filter. Let Axon(·) represent the operation of transmitting a pulse to the next neuron, and let It represents the behavior of neurons updating membrane potential and emitting pulses. Therefore, the DSNN model simulated by IIR filter is divided into Axon(·) and operate; S2, extract the original functional time series information from the functional magnetic resonance imaging fMRI, let Represents the time series matrix of fMRI, N corresponds to the number of node sets in the brain region, T corresponds to the length of the time series, and G fc As the feature input of DSNN, through Axon(·) and Operations were performed to obtain the spike functional connectivity (FC) matrix; S3, extracting structural brain networks from diffusion tensor imaging DTI, let Represents the adjacency matrix of the structural brain network, N corresponds to the number of node sets in the brain region, and then the edge-to-edge convolution E2E and edge-to-node convolution E2N operations proposed by Kawahara are used to obtain the structural brain network information feature map. The feature map of the structural brain network after E2E convolution is recorded as The feature map after E2N convolution is Among them, C1 and C2 represent the number of filters of the two convolution operations. Finally, the structural brain network feature map G after E2E and E2N convolution operations is sc As the structural input of DSNN, through the linear network and Operation to obtain the impulse structure connection (SC) matrix; S4, construct the pulse coupling pooling SCP module, input the pulse FC and SC matrices into the SCP module to obtain the pulse structure-function SC-FC coupling, and finally pass the linear network and Operation, get the pulses emitted by each output neuron over time It is a time-varying vector where m is the number of output categories; S5. Input to the classification layer, calculate the maximum pulse count of each output neuron to obtain the predicted category, and use the cross entropy loss between the predicted category and the true category as the loss function for training, to obtain an accurate brain disease recognition model based on pulse structure-function coupling.

2. The method for analyzing brain disease structure-function coupling based on dynamic spiking neural network according to claim 1, characterized in that: The specific steps of obtaining the pulse FC matrix in step S2 are as follows: S21, the original fMRI time series matrix G fc As the feature input of DSNN, the formula is as follows: Where Axon(·) represents the operation of transmitting the pulse to the next neuron, l is the layer index, and when l = 1, Y l-1 =G fc ; S22. Extract the pulse FC matrix using the dynamic pulse metamodel. The formula is as follows: in, It represents the behavior of neurons updating membrane potential and emitting pulses. The final output pulse FC matrix is 3. The method for analyzing brain disease structure-function coupling based on dynamic spiking neural network according to claim 1, characterized in that: The specific steps of obtaining the pulse SC matrix in step S3 are as follows: S31, the SC feature map G after E2E and E2N convolution operations sc As the structural input of DSNN, implicit pulse encoding is performed through internal pulse neurons. The formula is as follows: Where D is the feature dimension, T is the time step, and the choice of T is determined by the length of the fMRI time series. sc By G sc After T repetitions, we get through the linear network, It is through The obtained value is a binarized impulse tensor; S32. Extract the pulse SC matrix using the linear network and dynamic pulse metamodel. The formula is as follows: Among them, X sc Represents the final output pulse SC matrix.

4. The method for analyzing brain disease structure-function coupling based on dynamic spiking neural network according to claim 1, characterized in that: The specific steps of obtaining pulse SC-FC coupling in step S4 are as follows: The functional and structural inputs of S41 and SCP are described as and The focus will be on describing the processing at time t (t∈[0,T]). Given the FC feature vector y at time t t and SC eigenvector x t , the bilinear transformation is defined as: in, is the projection matrix, (tr) represents the transposition operator, and in order to obtain an o-dimensional output, we need to learn This results in a large number of parameters; S42. Decompose W into two low-rank matrices: Among them, k is the decomposition rank, is the element-wise multiplication of two eigenvectors, is an all-one vector, Therefore, to obtain an output feature vector Two 3D tensors are required and Use the reshape operation to and Re-expressed as two-dimensional matrices and S43, final fusion vector It is obtained by performing sum pooling of a one-dimensional non-overlapping window of size k on the projection vector using P′ and Q′: The above operation generates a feature vector that combines FC temporal information and SC spatial information. It only represents the coupling relationship between FC and SC at time t. After repeating S41, S42 and S43 operations T times, a pulse matrix consisting of T eigenvectors is obtained. This is called pulsed SC-FC coupling, which not only contains the FC information in time, but also retains the SC information in space, and fully characterizes the dynamic coupling relationship between FC and SC; S44, finally through the linear network and Operation, get the pulses emitted by each output neuron over time Where m is the number of output categories, is a vector that varies with time.

Images