Dynamic Addiction Neural Circuit Generation Method and System Based on Weakly Supervised Contrastive Learning

Abstract

The present invention relates to a method and system for generating dynamic addictive neural circuits based on weakly supervised contrastive learning, belonging to the field of artificial intelligence technology. The generation method includes: based on a convolutional neural network, reducing the voxels of multiple groups of fMRI to the brain region node attributes, and generating multiple groups of dynamic brain connection maps containing time series according to the brain region node attributes; extracting the spatio-temporal features of the brain connections in each of the dynamic brain connection maps; inputting the spatio-temporal features into an abnormal connection detection network, calculating the abnormal probability of the brain connections based on contrastive learning, and obtaining the brain connections with the maximum abnormal probability at each moment; generating dynamic addictive neural circuits according to neuroscience prior knowledge and the brain connections with the maximum abnormal probability. Directly inputting the fMRI images into the model eliminates redundant and complex preprocessing calculations. Through contrastive learning, the differences in brain connections of different samples are obtained, and combined with a small amount of neuroscience prior knowledge, the mechanism of addictive neural circuits is revealed, which is easy to train and has high accuracy.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method and system for generating dynamic brain addiction neural circuits based on weakly supervised contrast learning. Background Art

[0002] Addiction is a disease characterized by the seeking of addictive substances. Taking smoking addiction as an example, currently, the number of tobacco users in China has exceeded 300 million, and the number of people who die from tobacco-related diseases each year is as high as 1 million. From the perspective of inducing smoking behavior, nicotine addiction is the main cause of smoking and the main obstacle for smokers to quit smoking. Addiction is also regarded as a chronic relapsing functional brain disease.

[0003] Currently, the detection of nicotine addiction neural circuits is to analyze and calculate abnormal neural circuits from fMRI (functional magnetic resonance imaging). However, traditional statistics-based methods require complex preprocessing operations on image data, with a large amount of redundancy. Moreover, fMRI image data has the defect of small samples, resulting in difficult training and low training accuracy. Summary of the Invention

[0004] A method for generating dynamic addiction neural circuits based on weakly supervised contrast learning provided by the present invention adopts the following technical solutions:

[0005] A method for generating dynamic addiction neural circuits based on weakly supervised contrast learning includes:

[0006] Based on a convolutional neural network, reducing the voxels of multiple groups of fMRI images to brain region node attributes, and generating multiple groups of dynamic brain connection maps including time series according to the brain region node attributes; wherein, each group of fMRI includes multiple fMRI images collected at a fixed frequency within a preset time period;

[0007] Extracting the spatio-temporal features of the brain connections in each of the dynamic brain connection maps;

[0008] Inputting the spatio-temporal features into an abnormal connection detection network, calculating the abnormal probability of the brain connections based on contrast learning, and obtaining the brain connections with the maximum abnormal probability at each moment;

[0009] Generating dynamic addiction neural circuits according to neuroscience prior knowledge and the brain connections with the maximum abnormal probability at each moment.

[0010] By adopting the above technical solutions, a weakly supervised contrast learning model is constructed based on the group labels of each group of samples and a small amount of prior knowledge in neuroscience, combined with contrast learning. By directly inputting fMRI images into the weakly supervised contrast learning model, redundant and complex preprocessing calculations are eliminated. Moreover, on the premise of small samples, through contrast learning, the brain connection differences between different groups of samples are obtained, and combined with a small amount of prior knowledge in neuroscience, the addictive neural circuit mechanism is revealed, with low training difficulty and high training accuracy.

[0011] Optionally, before the step of extracting the spatio-temporal features of the brain connections in the dynamic brain connection map, the following is further included:

[0012] Taking the standard brain connection map as the true sample and the dynamic brain connection map as the false sample, generating adversarial learning to update the dynamic brain connection map.

[0013] By adopting the above technical solutions, through adversarial learning, a dynamic brain connection map that not only conforms to the spatial position information of the standard brain template but also has the original fMRI voxel space information is obtained, which is helpful for the training accuracy; moreover, the amount of data of the standard spatial brain template as prior knowledge is small and the learning difficulty is low.

[0014] Optionally, the step of extracting the spatio-temporal features of the brain connections in the dynamic brain connection map specifically includes:

[0015] Using a graph convolutional neural network based on the topological space brain connection attention mechanism to extract the spatial features of the brain connections in the dynamic brain connection map;

[0016] Using a gated recurrent unit based on the temporal brain connection attention mechanism to extract the temporal features of the brain connections in the dynamic brain connection map.

[0017] By adopting the above technical solutions, the spatial features and temporal features of the dynamic brain connection map are respectively extracted, thereby obtaining the spatio-temporal features of the brain connections, which fully characterize the attributes of the dynamic addictive brain connections.

[0018] Optionally, the step of inputting the spatio-temporal features into an abnormal connection detection network and calculating the brain connection abnormality probability based on contrast learning specifically includes;

[0019] The brain connection abnormality probability is: F score = MLP(H t );

[0020]

[0021] h i is the hidden state feature vector of the i-th node, h jis the hidden state feature vector of the j-th node, σ(·) is the sigmoid function, a and b are the optimization parameters of the output layer, β and μ are a pair of hyperparameters, and ω represents the weight of edge e ij ;

[0022] Contrast loss function: L S = min(max{0, f(H S ) - ∑ a∈A f(H a ) + γ});

[0023] L Δ = min(max{0, f(H a ) - f(H a′ ) + γ}), a ∈ A, a' ∈ A, a ≠ a';

[0024] L TOTAL = L S + ∑ a∈A,a′∈A L Δ ;

[0025] Among them, max{0, ·} is the hinge loss, H S is the spatio-temporal feature of the saline injection group, H a is the spatio-temporal feature of the nicotine injection group, A = {1, 2,..., A'}, A' is the total number of groups in the nicotine injection group, and γ is a hyperparameter.

[0026] By adopting the above technical solution, based on the contrast learning strategy, the saline injection group and all nicotine injection groups are compared and learned, and pairwise comparison and learning are performed on multiple groups of nicotine injections, which can fully analyze the individual differences of the dynamic brain connection maps in different control groups and facilitate the discovery of abnormal brain connections related to addiction.

[0027] Optionally, the step of generating a dynamic addiction neural circuit according to neuroscience prior knowledge and the brain connection with the highest abnormal probability at each moment specifically includes:

[0028] According to neuroscience prior knowledge, correct and integrate the brain connection with the highest abnormal probability to obtain the addiction neural circuit at each moment, and generate a dynamic addiction neural circuit according to the time series.

[0029] By adopting the above technical solution, according to neuroscience prior knowledge, correct and normalize the brain connection with the highest abnormal probability at each moment, thus ensuring the accuracy of the generated addiction neural circuit.

[0030] In a second aspect, the present invention provides a dynamic addiction neural circuit generation system based on weakly supervised contrast learning, adopting the following technical solution:

[0031] A dynamic addictive neural circuit generation system based on weakly supervised contrast learning, comprising:

[0032] A dynamic brain connectivity map generation module, configured to reduce the voxels of multiple groups of fMRI images to

[0033] the brain region node attributes based on a convolutional neural network, and generate multiple groups of dynamic brain connectivity maps including time series according to the brain region node attributes; wherein, each group of fMRI includes multiple fMRI images collected at a fixed frequency within a preset time period;

[0034] A spatio-temporal feature extraction module, configured to extract the spatio-temporal features of the brain connections in each of the dynamic brain connectivity maps;

[0035] A most abnormal brain connection acquisition module, configured to input the spatio-temporal features into an abnormal connection detection network, calculate the abnormal probability of the brain connections based on contrast learning, and obtain the brain connections with the maximum abnormal probability at each moment;

[0036] A dynamic addictive neural circuit generation module, configured to generate a dynamic addictive neural circuit according to neuroscience prior knowledge and the brain connections with the maximum abnormal probability at each moment.

[0037] In a third aspect, the present invention provides an electronic device, adopting the following technical solution:

[0038] The electronic device includes a memory and a processor, and a computer program capable of being loaded and executed by the processor for the method described above is stored on the memory.

[0039] In a fourth aspect, the present invention provides a computer-readable storage medium, adopting the following technical solution:

[0040] The computer-readable storage medium stores a computer program capable of being loaded and executed by the processor for the method described above.

[0041] In summary, the present invention includes at least one of the following beneficial technical effects:

[0042] 1. According to the group labels of each group of samples and a small amount of neuroscience prior knowledge, combined with contrast learning, a weakly supervised contrast learning model is constructed. By directly inputting fMRI images into the weakly supervised contrast learning model, redundant and complex preprocessing calculations are eliminated.

[0043] 2. On the premise of a small sample size, through contrast learning, the brain connection differences between different groups of samples are obtained, and combined with a small amount of neuroscience prior knowledge, the addictive neural circuit mechanism is revealed, which is easy to train and has high training accuracy.

[0044] 3. Adopt a contrastive learning strategy to conduct contrastive learning between the saline injection group and all nicotine injection groups, and conduct pairwise contrastive learning among multiple nicotine injection groups, which can fully analyze the individual differences in the dynamic brain connectivity maps of different control groups and facilitate the discovery of abnormal brain connectivity related to addiction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flowchart of a method for generating dynamic addiction neural circuits based on weakly supervised contrastive learning.

[0046] Figure 2 is a flowchart of a method for extracting spatio-temporal features of brain connectivity.

[0047] Figure 3 is a block diagram of the structure of a system for generating dynamic addiction neural circuits based on weakly supervised contrastive learning.

[0048] Figure 4 is a block diagram of the structure of a spatio-temporal feature extraction module.

[0049] Figure 5 is a schematic diagram of an electronic device. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the following further describes the present invention in detail with reference to Figure 1 .4 and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] An embodiment of the present invention discloses a method for generating dynamic addiction neural circuits based on weakly supervised contrastive learning. Referring to Figure 1 , the method for generating dynamic addiction neural circuits based on weakly supervised contrastive learning includes:

[0052] S11. Based on a convolutional neural network, reduce the voxels of multiple groups of fMRI images to the attributes of brain region nodes, and generate multiple groups of dynamic brain connectivity maps containing time series according to the attributes of the brain region nodes.

[0053] Use the fMRI images of the rat animal model in the nicotine addiction experiment to generate dynamic addiction neural circuits based on weakly supervised contrastive learning. Multiple groups of fMRI images include the fMRI images of rats in the saline injection group and the nicotine injection group. Among them, the number of groups of rats in the saline group is 1, and the number of groups of rats in the nicotine injection group can be set according to specific actual situations, preferably 1.4. Each group of fMRI images includes multiple fMRI images collected at a fixed frequency within a preset time period. Among them, the preset time period is the duration of rats injecting saline or nicotine, which can be set to two weeks, three weeks, four weeks or other durations, and is not limited herein.

[0054] Specifically, rats in the saline injection group and rats in the nicotine injection group were used for the experiment. The rats in the nicotine group

[0055] can be divided into multiple groups according to different concentrations, or only one group can be set. Each group of rats was continuously injected for a preset period of time to simulate the situation of nicotine intake and addiction. During the preset period, fMRI images of each group of rats were collected at a fixed frequency. The fMRI images of each group of rats reflected the changes in brain nerves over time after injecting saline or nicotine at the corresponding concentration.

[0056] Using a convolutional neural network to reduce the dimension of the voxels of each group's fMRI can, while maintaining the time resolution unchanged, segment the time segments of fMRI with a certain meaningful time resolution, and each segment is used as a time point to generate a dynamic brain connectivity map, fully extracting the temporal dynamic information of fMRI.

[0057] The dynamic brain connectivity map G is as follows: Specifically, in the dynamic brain connectivity map G, with the fMRI time segments as the time sequence, each time sequence as a time point, brain regions as nodes, and brain connections as edges. Brain connection refers to the functional connection between brain regions. T is the total number of time points, and G t is a snapshot of the dynamic brain connectivity map at time t, V t and E t respectively represent the node set and edge set therein. The edge e i,j ∈E t , indicating that at time t, there is a brain functional connection between the i-th and j-th nodes, and its weight is ω. represents the set of node attributes, N represents the total number of nodes, is the attribute vector of the i-th node at time t. The fMRI images of the saline injection group I s and the nicotine injection group I a were respectively input into the above dynamic brain connectivity map G to obtain the dynamic brain connectivity maps G s 、G a , where a ∈ A, A = {1, 2,..., A'}, and A' is the total number of groups in the nicotine injection group.

[0058] In this step, the fMRI images were directly input into the weakly supervised adversarial learning model to obtain the dynamic brain connectivity map, eliminating redundant and complex preprocessing calculations.

[0059] S12: Using the standard brain connectivity map as the true sample and the dynamic brain connectivity map as the false sample, generate adversarial learning to update the dynamic brain connectivity map.

[0060] The standard brain connectivity map is a brain functional connectivity map preprocessed from a standard spatial brain template containing prior information of the brain atlas. The standard brain connectivity map is used as the true sample G True , and the dynamic brain connectivity map obtained in the previous step is used as the fake sample G Fake . Generative adversarial learning is performed to update the dynamic brain connectivity map.

[0061] The formula for generative adversarial learning is:

[0062]

[0063] Among them, D is the discriminator, representing the standard brain connectivity map, and G is the generator, representing the abnormal brain connectivity map. In this application, G specifically refers to the dynamic brain connectivity map. Generative adversarial learning is performed on the dynamic brain connectivity maps of three groups of categories respectively to update the dynamic brain connectivity map, that is, input G

[0064] and G s and G a into the above formula for updating respectively. The updated dynamic brain connectivity map is a dynamic brain connectivity map approximately G with respect to the spatial information of the brain atlas, representing the abnormal brain connectivity patterns of nicotine addiction under different categories.

[0065] It should be noted that step S12 is optional. If step 12 is executed, through adversarial learning, a dynamic brain connectivity map that not only conforms to the spatial position information of the standard brain template but also has the original fMRI voxel space information is obtained, improving the training accuracy; moreover, by preprocessing the standard spatial brain template containing prior information of the brain atlas to obtain the labeled brain functional connectivity map, the data volume of the standard spatial brain template is small and the learning difficulty is low.

[0066] S13. Extract the spatio-temporal features of the brain connections in each of the dynamic brain connectivity maps.

[0067] Input each group of dynamic brain connectivity maps into a neural network for learning to extract the spatial and temporal features of the brain connections in the dynamic brain connectivity maps.

[0068] Specifically, referring to Figure 2 , step S13 includes the following sub-steps:

[0069] S1311. Use a graph convolutional neural network based on the topological spatial brain connection attention mechanism to extract the spatial features of the brain connections in the dynamic brain connectivity map.

[0070] The graph convolutional neural network based on the topological spatial brain connection attention mechanism is composed of a multi-head anomalous connection focus block (MACFB), which has a powerful graph learning ability and can learn the geometric spatial features between nodes in the connection network.

[0071] Specifically, the total number of abnormal connection attention blocks is C, and the value of C is preferably 2, 3, or 4.

[0072] Hidden state feature vector set:

[0073] It represents the multi-head abnormal connection attention block of the l-th layer at time t, where t ∈ {1, 2, …, T}, Y is the total number of time points, i represents the node, i ∈ {1, 2, …, N}, and N is the total number of nodes. It represents the multi-head abnormal connection attention block of the i-th node at the l-th layer at time t.

[0074] Query-node attribute vector:

[0075] Among them, i represents the node, and c represents the abnormal connection attention block. is the learnable weight matrix for calculating the query-node attribute vector. is the offset for calculating the query-node attribute vector.

[0076] Key-node attribute vector:

[0077] Among them, j represents the node. is the learnable weight matrix for calculating the key-node attribute vector. is the offset for calculating the key-node attribute vector.

[0078] Additional brain connection feature vector: e c,ij = W c,e e ij + b c,e

[0079] Among them, e ij is the edge in the dynamic brain connection graph, representing the brain connection between the i-th and j-th nodes at time t, and its weight is ω. W c,e is the learnable weight matrix for calculating the additional brain connection feature vector, and b c,e is the offset for calculating the additional brain connection feature vector.

[0080] Brain connection attention coefficient: Among them, <q, k> is the exponential dot product function, and N(i) is the value excluding i.

[0081] After calculating the brain connection attention coefficient, message passing in the graph is performed to output the hidden state feature vector of the next layer of nodes.

[0082]

[0083] is the learnable weight matrix for computing , and is the offset for computing .

[0084] The set of hidden state feature vectors of the nodes in the next layer is used as the input to the multi-head abnormal connection attention block in the next layer, or as the final output result of the graph convolutional neural network. If it is used as the final output result of the graph convolutional neural network, it represents the spatial features of the abnormal brain connection patterns in the extracted dynamic brain connection graph.

[0085] S1312. Use a gated recurrent unit based on the temporal brain connection attention mechanism to extract the temporal features of the dynamic brain connection graph.

[0086] The gated recurrent unit combined with the brain connection attention mechanism learns short-term and long-term hidden state features, so that the hidden state features of each temporal brain functional connection graph contain the context brain connection hidden state information of other time series, thereby learning the temporal features of the dynamic brain connection graph.

[0087] In the gated recurrent unit, the hidden state feature vector at time t is required to calculate the hidden state feature vector at the next time t + 1.

[0088] Set of long-term hidden state feature vectors: Current t+1 t+1 = MACFB(H t )

[0089] In the initial state, Ht represents the output result of the graph convolutional neural network, that is, the spatial features of the abnormal brain connection patterns in the extracted dynamic brain connection graph.

[0090] Calculate the short-term hidden state feature vector according to the brain connection attention window; within the brain functional connection attention window, the output of the node hidden state feature vector:

[0091] where the window size is w + 1, and i represents the node;

[0092] Temporal brain connection attention coefficient:

[0093] where r is the weight vector and Q is the weight matrix;

[0094]

[0095] Node short-term hidden state feature vector:

[0096] Set of hidden state feature vectors of brain connections at the next moment: H t+1= GRU(Current t+1 , Short t+1 );

[0097] Short t+1 represents the set of short - term hidden state feature vectors of all nodes, where \(i\in\{1,2,\cdots,N\}\) and \(N\) is the total number of nodes.

[0098] Furthermore, the specific calculation process of \(H t+1 = GRU(Current t+1 , Short t+1 ) is as follows:

[0099] Brain connection state update gate: \(P t+1 =\sigma(U P Current t+1 +W P Short t+1 +b P );

[0100] Among them, \(U P \) and \(W P \) are both learnable weight matrices for calculating \(P t+1 \), and \(b P \) is the offset for calculating \(P t+1 \).

[0101] Brain connection state reorganization gate: \(R t+1 =\sigma(U R Current t+1 +W R Short t+1 +b R )

[0102] Among them, \(U R \) and \(W R \) are both learnable weight matrices for calculating \(R t+1 \), and \(b R \) is the offset for calculating \(R t+1 \).

[0103]

[0104] Among them, \(U H \) and \(W H \) are learnable weight matrices for calculating .

[0105]

[0106] Since the input is spatial features when calculating temporal features, therefore, when using the gated recurrent unit based on the temporal brain connection attention mechanism to extract temporal features, the set of hidden state feature vectors \(H\) output at time \(t\)t , representing the spatio-temporal characteristics of the abnormal brain connection patterns in the extracted dynamic brain connection map.

[0107] S14. Input the spatio-temporal characteristics into the abnormal connection detection network, calculate the abnormal probability of the brain connection based on contrastive learning, and obtain the brain connection with the maximum abnormal probability at each moment.

[0108] In the AnomalyConnectivitiesDetectionNetwork, at time t, the output H of the set of hidden state feature vectors t is calculated through a fully connected layer to obtain the abnormal probability of the brain connection at time t as:

[0109] F score = MLP(H t );

[0110] The abnormal probability represents the abnormal probability of the edge e i,j ∈ E t at time t, that is, the abnormal probability of each edge in the dynamic brain connection maps G s and G a

[0111]

[0112] where h i is the hidden state feature vector of the i-th node, h j is the hidden state feature vector of the j-th node, σ(·) is the sigmoid function, a and b are the optimization parameters of the output layer, β and μ are a pair of hyperparameters, and ω represents the weight of the edge e ij ; Input the spatio-temporal characteristics of each group of dynamic brain connection maps into the above function f(·) to obtain f(H S ) and f(H a ).

[0113] Adopt the strategy of contrastive learning, and define the contrastive loss function for abnormal brain connection learning as:

[0114] Contrastive loss function: L S = min(max{0, f(H S - ∑ a∈A f(H a ) + γ)});

[0115] L Δ = min(max{0, f(H a ) - f(H a′ ) + γ}), a ∈ A, a' ∈ A, a ≠ a';

[0116] L​TOTAL = L s + ∑ a∈A,a′∈A LΔ

[0117] where max{0, ·} is the hinge loss, H S is the spatio-temporal feature of the saline injection group, H a is the spatio-temporal feature of the nicotine injection group, A = {1, 2, …, A'}, A' is the total number of groups in the nicotine injection group, and γ is a hyperparameter.

[0118] Adopting the above-mentioned contrastive learning strategy, comparing the saline injection group with all nicotine injection groups for contrastive learning, and pairwise comparing multiple groups of nicotine injection groups, can fully analyze the individual differences in the dynamic brain connection maps of different control groups, facilitating the discovery of abnormal brain connections related to addiction.

[0119] After calculating the abnormal probability of the brain connections in each dynamic brain connection map through contrastive learning, for each moment, obtain the brain connection with the maximum abnormal probability among multiple groups of dynamic brain connection maps.

[0120] As an implementation, before obtaining the brain connection with the maximum abnormal probability at each moment, input the brain connection abnormal probability into a preset classifier for iterative training until the classification accuracy converges to a preset accuracy.

[0121] Specifically, at time t, input the abnormal probability of each brain connection in each group of dynamic brain connection maps into a preset classifier for iterative training. The preset classifier can be selected according to actual needs, and conventional classifiers can all achieve the above functions, such as SVM, KNN, Naive Bayes, Decision Tree, Logistic Regression, Neural Network Algorithm, etc., which are not specifically limited here. The preset accuracy is set according to actual needs and is not limited here.

[0122] S15. Generate a dynamic addiction neural circuit based on prior knowledge of neuroscience and the brain connection with the maximum abnormal probability at each moment.

[0123] Combining prior knowledge of neuroscience, for each moment t, correct and integrate the brain functional connection with the maximum abnormal probability to obtain the addiction neural circuit at each moment, and then generate a dynamic addiction neural circuit according to the entire time series. Correcting the brain connection with the maximum abnormal probability at each moment means deleting incorrect brain connections and / or adjusting brain connections with large deviations according to prior knowledge of neuroscience; integrating the most abnormal brain functional connections means integrating the corrected brain functions into a complete addiction neural circuit for each moment.

[0124] In this embodiment, according to the group labels of each group of samples and a small amount of neuroscience prior knowledge, combined with contrastive learning, a weakly supervised contrastive learning model is constructed, and the training difficulty is low. Moreover, the fMRI images are directly input into the weakly supervised contrastive learning model, eliminating redundant and complex preprocessing calculations. Further, on the premise of small samples, through contrastive learning, the brain connection differences between different groups of samples are obtained, and combined with a small amount of neuroscience prior knowledge, the addictive neural circuit mechanism is revealed, which is easy to train and has high training accuracy.

[0125] An embodiment of the present invention also discloses a dynamic addictive neural circuit generation system based on weakly supervised contrastive learning. Refer to Figure 3 , the dynamic addictive neural circuit generation system based on weakly supervised contrastive learning includes:

[0126] A dynamic brain connection map generation module 10, configured to reduce the voxels of multiple groups of fMRI images to brain region node attributes based on a convolutional neural network, and generate multiple groups of dynamic brain connection maps including time series according to the brain region node attributes; wherein, each group of fMRI includes multiple fMRI images collected at a fixed frequency within a preset time period;

[0127] A dynamic brain connection update module 20, configured to generate an adversarial learning with the standard brain connection map as the true sample and the dynamic brain connection map as the false sample to update the dynamic brain connection map;

[0128] A spatio-temporal feature extraction module 30, configured to extract the spatio-temporal features of the brain connections in each of the dynamic brain connection maps;

[0129] A most abnormal brain connection acquisition module 40, configured to input the spatio-temporal features into an abnormal connection detection network, calculate the abnormal probability of the brain connection based on contrastive learning, and acquire the brain connection with the maximum abnormal probability at each moment;

[0130] A dynamic addictive neural circuit generation module 50, configured to generate a dynamic addictive neural circuit according to neuroscience prior knowledge and the brain connection with the maximum abnormal probability at each moment.

[0131] Among them, the dynamic brain connection update module 20 is optional.

[0132] Refer to Figure 3 , the spatio-temporal feature extraction module 30 specifically includes:

[0133] A spatial feature extraction module 3010, configured to extract the spatial features of the brain connections in the dynamic brain connection map by using a graph convolutional neural network based on a topological space brain connection attention mechanism.

[0134] A time feature extraction module 3020, configured to extract the time features of the dynamic brain connection map by using a gated recurrent unit based on a temporal brain connection attention mechanism.

[0135] The described system embodiment can be used to execute the above method embodiment, and the principle and technical effect are similar, which will not be elaborated here.

[0136] Based on the same inventive concept, an embodiment of the present disclosure also provides an electronic device 500. Referring to Figure 5 As shown, the electronic device 500 includes a processor 501, a memory 502, and a bus. Among them, the memory 502 is used to store computer programs, including an internal memory 5021 and an external memory 5022; the internal memory 5021 is used to temporarily store the operation data in the processor 501 and the data exchanged with the external memory 5022 such as a hard disk, and the processor 501 exchanges data with the external memory 5022 through the internal memory 5021.

[0137] In the embodiment of the present application, the memory 502 is specifically used to store the computer program for executing the technical solution of the present application and is controlled by the processor 501 to execute. That is, when the electronic device 500 runs, the processor 501 communicates with the memory 502 through the bus, so that the processor 501 executes the computer program stored in the memory 502, and further executes the method described in any of the foregoing embodiments.

[0138] Among them, the memory 502 can be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read - only memory (PROM), an erasable programmable read - only memory (EPROM), etc.

[0139] The processor 501 may be an integrated circuit chip with signal processing capabilities. The above - mentioned processor may be a general - purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application - specific integrated circuit (ASIC), a field - programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general - purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0140] It can be understood that the structure illustrated in the embodiments of the present application does not constitute a specific limitation on the electronic device 500. In other embodiments of the present application, the electronic device 500 may include more or fewer components than those illustrated, or combine certain components, or split certain components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0141] This embodiment also provides a computer-readable storage medium, such as a floppy disk, optical disc, hard disk, flash memory, USB flash drive, SD (Secure Digital Memory Card) card, MMC (Multimedia Card) card, etc. In this readable storage medium, a computer program for implementing the above various steps is stored. The computer program can be executed by one or more processors to implement the method in the above embodiments.

[0142] In the embodiments provided in the present application, it should be understood that the disclosed device and method can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some communication interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical, mechanical, or other form.

[0143] In addition, in each embodiment of the present application, the various functional modules can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0144] The above are all the preferred embodiments of the present invention. The protection scope of the present invention is not limited by this. Any feature disclosed in this specification (including the abstract and drawings), unless specifically described, can be replaced by other equivalent or similar-purpose alternative features. That is, unless specifically described, each feature is only an example in a series of equivalent or similar features.

Claims

1. A method for generating dynamic addictive neural circuits based on weakly supervised contrast learning, characterized in that, it includes: Based on a convolutional neural network, reducing the voxels of multiple groups of fMRI images to brain region node attributes, and generating multiple groups of dynamic brain connection maps containing time series according to the brain region node attributes; wherein, each group of fMRI images includes multiple fMRI images collected at a fixed frequency within a preset time period; Extracting the spatio-temporal features of the brain connections in each of the dynamic brain connection maps; Inputting the spatio-temporal features into an abnormal connection detection network, calculating the abnormal probability of the brain connections based on contrast learning, and obtaining the brain connections with the maximum abnormal probability at each moment; Generating dynamic addictive neural circuits according to neuroscience prior knowledge and the brain connections with the maximum abnormal probability at each moment.

2. The method according to claim 1, characterized in that, Before the step of extracting the spatio-temporal features of the brain connections in each of the dynamic brain connection maps, it further includes: Taking the standard brain connection map as a true sample and the dynamic brain connection map as a false sample to generate adversarial learning and update the dynamic brain connection map.

3. The method according to claim 1 or 2, characterized in that, The step of extracting the spatio-temporal features of the brain connections in each of the dynamic brain connection maps specifically includes: Using a graph convolutional neural network based on a topological space brain connection attention mechanism to extract the spatial features of the brain connections in the dynamic brain connection map; Using a gated recurrent unit based on a temporal brain connection attention mechanism to extract the temporal features of the brain connections in the dynamic brain connection map.

4. The method according to claim 3, characterized in that, The step of inputting the spatio-temporal features into an abnormal connection detection network and calculating the abnormal probability of the brain connections based on contrast learning specifically includes; The abnormal probability of brain connection is: F score = MLP(H t ), where H t is the output of the set of hidden state feature vectors; where, h i is the hidden state feature vector of the i-th node, h j is the hidden state feature vector of the j-th node, σ(·) is the sigmoid function, a and b are the optimization parameters of the output layer, β and μ are a pair of hyperparameters, and ω represents the weight of the edge e ij ; Contrastive loss function: L S = min(max{0, f(H S ) - ∑ a∈A f(H a ) + γ}); L Δ = min(max{0, f(H a ) - f(H a′ ) + γ}), a ∈ A, a' ∈ A, a ≠ a'; L TOTAL = L S + ∑ a∈A,a′∈A L Δ ; Among them, max{0,·} is the hinge loss, and H S is the spatio-temporal feature of the saline injection group, and H a is the spatio-temporal feature of the nicotine injection group, A = {1, 2,..., A'}, A' is the total number of groups in the nicotine injection group, and γ is a hyperparameter.

5. The method according to claim 4, characterized in that, Before obtaining the brain connections with the maximum abnormal probability at each moment, it further includes: Inputting the abnormal probability of the brain connections into a preset classifier for iterative training until the classification accuracy converges to a preset accuracy.

6. The method according to claim 4, characterized in that, The step of generating dynamic addictive neural circuits according to neuroscience prior knowledge and the brain connections with the maximum abnormal probability at each moment specifically includes: Combining neuroscience prior knowledge, correcting and integrating the brain connections with the maximum abnormal probability to obtain the addictive neural circuits at each moment, and generating dynamic addictive neural circuits according to the time series.

7. A system for generating dynamic addictive neural circuits based on weakly supervised contrast learning, characterized in that, it includes: A dynamic brain connection map generation module, configured to reduce the voxels of multiple groups of fMRI images to brain regions node attributes based on a convolutional neural network, and generate multiple groups of dynamic brain connection maps containing time series according to the brain region node attributes; wherein, each group of fMRI images includes multiple fMRI images collected at a fixed frequency within a preset time period; A spatio-temporal feature extraction module, configured to extract the spatio-temporal features of the brain connections in each of the dynamic brain connection maps; The most abnormal brain connection acquisition module is used to input the spatio-temporal features into the abnormal connection detection network, calculate the abnormal probability of the brain connection based on contrast learning, and obtain the brain connection with the maximum abnormal probability at each moment; The dynamic addiction neural circuit generation module is used to generate a dynamic addiction neural circuit according to prior neuroscience knowledge and the brain connection with the maximum abnormal probability at each moment.

8. The system according to claim 7, wherein, it further includes a dynamic brain connection update module; The dynamic brain connection update module is used to generate adversarial learning with the standard brain connection map as the true sample and the dynamic brain connection map as the false sample, and update the dynamic brain connection map.

9. An electronic device, wherein: it includes a memory and a processor, and a computer program capable of being loaded and executed by the processor for the method according to any one of claims 1-6 is stored on the memory.

10. A computer-readable storage medium, wherein: a computer program capable of being loaded and executed by the processor for the method according to any one of claims 1-6 is stored.

Images