WO2023108418A1 - Procédé de construction d'un atlas cérébral et de détection d'un circuit neuronal et produit associé - Google Patents

Procédé de construction d'un atlas cérébral et de détection d'un circuit neuronal et produit associé Download PDF

Info

Publication number
WO2023108418A1
WO2023108418A1 PCT/CN2021/137913 CN2021137913W WO2023108418A1 WO 2023108418 A1 WO2023108418 A1 WO 2023108418A1 CN 2021137913 W CN2021137913 W CN 2021137913W WO 2023108418 A1 WO2023108418 A1 WO 2023108418A1
Authority
WO
WIPO (PCT)
Prior art keywords
neural
feature
neural network
mentioned
brain atlas
Prior art date
Application number
PCT/CN2021/137913
Other languages
English (en)
Chinese (zh)
Inventor
王书强
潘俊任
申妍燕
Original Assignee
深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Priority to PCT/CN2021/137913 priority Critical patent/WO2023108418A1/fr
Publication of WO2023108418A1 publication Critical patent/WO2023108418A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

  • This application relates to artificial intelligence, in particular to a method and related products for brain map construction and neural circuit detection.
  • artificial intelligence usually uses medical images to analyze and mine patients' medical data to improve the accuracy of medical diagnosis.
  • the functional or structural information of the brain is usually obtained based on traditional medical images, combined with the experience of doctors to judge the abnormalities in the brain.
  • the embodiment of the present application discloses a method of brain atlas construction and neural circuit detection and related products, which can detect brain neural circuits or brain connections, and are suitable for studying abnormalities caused by brain neural circuits or brain connections. disease.
  • the embodiment of the present application provides a method for brain atlas construction and neural circuit detection, including:
  • the first target data is obtained according to the first image and the second target data is obtained according to the second image
  • the first target data represents the time series features of each region in the first image
  • the second target data represents the The connection strength of neurons between the regions in the second image
  • the first image is the resting state functional magnetic resonance imaging (resting-state functional magnetic resonance imaging, rs-fMRI) of the brain to be processed
  • the second The image is magnetic resonance diffusion tensor imaging (diffusion tensor imaging, DTI); the region in the first image and the region in the second image are determined according to a template;
  • the first brain atlas is used to characterize the relationship between the first target data and the second target data;
  • the first image The region of the region and/or the region of the second image corresponds to the graph node of the first brain atlas;
  • the label of the graph node of the first brain atlas is the anatomical label serial number of the template corresponding to the region;
  • the node feature of the graph node is the time series feature of the corresponding area included in the first target data;
  • first neural network Inputting the first brain atlas into a first neural network, and outputting first features, the first features being used to represent high-order topological features of the first brain atlas;
  • the second neural network is used to decouple regions in the first brain atlas
  • the first neural circuit is used to Represents the connection relationship between the regions of the first brain atlas.
  • the rs-fMRI and DTI data are used as multimodal data, combined with the prior knowledge of anatomical brain region segmentation templates, and encoded into the first brain atlas that can perform high-order topological feature analysis, which is conducive to simultaneous Preserve the brain region functional information and physical neuron connection information of the three-dimensional brain internal organization, better mine the complementary information between different modalities, make the detection more accurate, and input the first brain atlas into the first network layer Perform feature extraction, obtain relevant first features, input the first features into the second network layer, use the second network layer to analyze and decouple the first brain atlas, and obtain the target neural circuit, which can be obtained by Image processing achieves the purpose of detecting target neural circuits or brain connections.
  • the inputting the first brain map into the first neural network, and outputting the first feature includes:
  • the ⁇ represents a nonlinear activation function
  • the l represents the number of layers of the first neural network
  • the W represents a weight matrix to be learned
  • the b represents a deviation to be learned
  • the X G represents the first A target data
  • the A G represents the second target data
  • the D G represents the third target data.
  • the inputting the first feature into the second neural network and outputting the first neural loop includes:
  • the parameters of the first feature and the second neural network satisfy the following relationship:
  • the d represents the vector to be learned
  • the b represents the deviation to be learned
  • the Dec(v i ) represents the first value
  • the ⁇ represents a nonlinear activation function
  • the W represents the weight matrix to be learned
  • the preset conditions include:
  • the v i represents the region
  • the ⁇ represents a preset hyperparameter
  • the d represents a vector to be learned.
  • the method before acquiring the first image and the second image, the method further includes:
  • first sample data representing the time series of each region in the first sample image
  • second sample data representing the time series of each region in the first sample image
  • the two sample data represent the connection strength of neurons between regions in the second sample image
  • the first sample image is a rs-fMRI sample image
  • the second sample image is a DTI sample image
  • the second features are high-order topological features
  • the second feature is input into a second neural network, and a second neural circuit is output, and the second neural network is also used to decouple the regions in the second brain atlas;
  • the first neural network and the second neural network are trained according to the second neural loop.
  • the multimodal sample data rs-fMRI and DTI data are processed and encoded into a second brain atlas that is more suitable for high-order topological feature analysis, and the second brain atlas is decoupled to obtain
  • the second neural loop, training the first neural network and the second neural network according to the second neural loop can improve detection accuracy.
  • the training the first neural network and the second neural network according to the second neural loop includes:
  • the first neural network, the second neural network, the third neural network, and the fourth neural network are trained according to the first loss.
  • the first loss guiding the decoupling module to detect the most relevant neural loop is calculated according to the second neural loop, and the decoupling module is trained according to the first loss to realize automatic detection of the neural loop the goal of.
  • the inputting the second neural loop into a third neural network, and outputting relevant features affecting the second neural loop include:
  • the nodes of the first hypergraph are regions in the second brain atlas, and the hyperedges of the first hypergraph are the second neural circuit , the first hypergraph includes a third feature and a first matrix, the third feature is a node feature of the first hypergraph, and the first matrix is a matrix associated between the node and the hyperedge;
  • the third feature and the first matrix are input into a third neural network to obtain the fourth feature, and the fourth feature is a feature that a neural circuit affects a disease.
  • the embodiment of the present application by embedding the second neural loop into a hypergraph to represent the complex connection relationship between the neural loops, the embodiment of the present application can analyze the neural loop from the overall feature distribution, Improve the accuracy of neural circuit detection.
  • the inputting the third feature and the first matrix into a third neural network, and obtaining the fourth feature includes:
  • the parameters of the third feature, the first matrix and the third neural network satisfy the following relationship:
  • the l represents the number of layers of the third neural network, so said said and said Represents the weight matrix and bias to be learned, the H represents the first matrix, the representation hypergraph region features, the Represents node features, and the ⁇ represents a nonlinear activation function.
  • the training the first neural network and the second neural network according to the second neural loop includes:
  • the generative network and the discriminative network are trained according to the second loss and the first neural loop and the second neural loop are trained according to the third loss.
  • the first neural network and the second neural network are trained by using the generative confrontation network to overcome the problem of the small amount of data samples, and update the training parameters by calculating the second loss and the third loss, It is ensured that the embodiment of the present application has high robustness and stability, so as to achieve the purpose of detecting neural circuits.
  • the inputting the random vector and the second neural circuit into the generation network, and obtaining the third brain map includes:
  • the sixth feature being a sparse topological feature of the second neural circuit
  • the third brain atlas is generated through the sparse brain network representation of the brain neural circuit, the authenticity of the third brain atlas is improved, and the third brain atlas is used for loop decoupling, so that the samples with a small sample size
  • the repeated use of multi-modal neuroimaging data solves the problem of overfitting of high-dimensional data in small samples of medical images, so that the present invention has the advantages of less model parameters, high robustness, and strong generalization ability, and integrates the neural circuit
  • the combination of the sparse brain network and the reconstructed brain network of the random vector in the latent space makes the target distribution of the detected neural circuit reflect the advantages of "maximizing the inter-class interval and minimizing the intra-class dispersion".
  • the obtaining the third loss according to the third neural circuit and the second neural circuit includes:
  • the third loss is designed to quantitatively describe the high-order topological differences between brain neural circuits, effectively solving the problem of overfitting in the training process, enhancing the generalization ability of the model in the training process, and making The embodiments of the present application achieve the purpose of detecting neural circuits.
  • the obtaining the third loss according to the second hypergraph and the third hypergraph includes:
  • L cap Sim spatial (H,H')+Sim spectral (H,H');
  • the N p represents the For the second neural circuit
  • the N p ' represents the third neural circuit.
  • the third target data is obtained according to the second target data.
  • the embodiment of the present application provides a device for brain atlas construction and neural circuit detection, including:
  • an acquisition unit configured to acquire first target data according to the first image and second target data according to the second image, the first target data representing time-series features of each region in the first image, and the second
  • the target data represent the connection strength of neurons between regions in the second image
  • the first image is the resting state functional magnetic resonance imaging rs-fMRI of the brain to be processed
  • the second image is the rs-fMRI to be processed Magnetic resonance diffusion tensor imaging DTI
  • the region in the first image and the region in the second image are determined according to a template
  • a determining unit configured to determine a first brain atlas according to the first target data and the second target data, the first brain atlas being used to characterize the relationship between the first target data and the second target data;
  • the region of the first image and/or the region of the second image corresponds to the graph node of the first brain atlas;
  • the label of the graph node of the first brain atlas is the template corresponding to the region Anatomical marker serial number;
  • the node feature of the graph node is the time series feature of the corresponding region included in the first target data;
  • a first output unit configured to input the first brain atlas into a first neural network, and output first features, where the first features are used to represent high-order topological features of the first brain atlas;
  • a second output unit configured to input the first feature into a second neural network and output a first neural circuit, the second neural network is used to decouple regions in the first brain atlas, the The first neural circuit is used to represent the connection relationship between regions of the first brain atlas.
  • the first output unit is specifically configured to obtain the first feature according to the first brain atlas and parameters of the first neural network
  • the ⁇ represents a nonlinear activation function
  • the l represents the number of layers of the first neural network
  • the W represents a weight matrix to be learned
  • the b represents a deviation to be learned
  • the X G represents the first A target data
  • the A G represents the second target data
  • the D G represents the third target data
  • the third target data is the dispersion matrix of the first brain atlas
  • the dispersion matrix represents the Topological divergence of each graph node in the first Brain Atlas.
  • the second output unit is specifically configured to obtain a first value according to the first feature and parameters of the second neural network
  • the second output unit is specifically configured to decouple regions in the first brain atlas to obtain the first neural circuit when the first value satisfies a preset condition
  • the parameters of the first feature and the second neural network satisfy the following relationship:
  • the d represents the vector to be learned
  • the b represents the deviation to be learned
  • the Dec(v i ) represents the first value
  • the ⁇ represents a nonlinear activation function
  • the W represents the weight matrix to be learned
  • the preset conditions include:
  • the v i represents the region
  • the ⁇ represents a preset hyperparameter
  • the d represents a vector to be learned.
  • the acquiring unit is further configured to acquire the first sample data according to the first sample image and the second sample data according to the second sample image, the first sample This data represents the time series of each region in the first sample image, the second sample data represents the connection strength of neurons between regions in the second sample image, and the first sample image is rs-fMRI sample image, the second sample image is a DTI sample image;
  • the determining unit is further configured to determine a second brain atlas according to the first sample data and the second sample data;
  • the first output unit is further configured to input the second brain atlas into the first neural network, and output second features, the second features being high-order topological features;
  • the second output unit is also used to input the second feature into a second neural network and output a second neural circuit, and the second neural network is also used to decouple regions in the second brain atlas;
  • a training unit configured to train the first neural network and the second neural network according to the second neural loop.
  • the training unit is specifically configured to input the second neural loop into a third neural network, and output relevant features affecting the second neural loop;
  • the training unit is specifically configured to input the relevant features into the fourth neural network, and output a first probability, where the first probability represents the degree of influence on the second neural loop;
  • the training unit is specifically configured to obtain a first loss according to the target label information and the first probability
  • the training unit is specifically configured to train the first neural network, the second neural network, the third neural network, and the fourth neural network according to the first loss.
  • the training unit is specifically configured to construct a first hypergraph according to the second neural circuit, and the nodes of the first hypergraph are the second brain atlas
  • the hyperedge of the first hypergraph is the second neural loop
  • the first hypergraph includes the third feature and the first matrix
  • the third feature is the first hypergraph Node features
  • the first matrix is a matrix associated between the node and the hyperedge;
  • the training unit is specifically configured to input the third feature and the first matrix into a third neural network to obtain the fourth feature, and the fourth feature is a feature that a neural circuit affects a disease.
  • the training unit is specifically configured to obtain a fourth feature according to the third feature, the first matrix, and parameters of the third neural network.
  • the training unit is specifically configured to input a random vector and the second neural circuit into a generation network to obtain a third brain atlas;
  • the training unit is specifically configured to input the third brain atlas and the second brain atlas into a discriminant network to obtain a second loss;
  • the training unit is specifically configured to input the third brain atlas into the second neural network and output a third neural circuit
  • the training unit is specifically configured to obtain a third loss according to the third neural circuit and the second neural circuit;
  • the training unit is specifically configured to train the generation network and the discrimination network according to the second loss, and train the first neural circuit and the second neural circuit according to the third loss.
  • the training unit is specifically configured to input the random vector into a fifth neural network to obtain a fifth feature, where the fifth feature is a topological feature of the random vector ;
  • the training unit is specifically configured to embed the second neural circuit into a sparse brain network to obtain a sixth feature, where the sixth feature is a sparse topological feature of the second neural circuit;
  • the training unit is specifically configured to input the fifth feature and the sixth feature into a sixth neural network to obtain the third brain atlas.
  • the training unit is specifically configured to construct a second hypergraph according to the second neural circuit, and construct a third hypergraph according to the third neural circuit;
  • the training unit is specifically configured to obtain the third loss according to the second hypergraph and the third hypergraph.
  • the training unit is specifically configured to obtain the third loss according to parameters of the second hypergraph and parameters of the third hypergraph.
  • an embodiment of the present application provides an electronic device, including: a memory for storing instructions; a processor for executing the above-mentioned instructions stored in the above-mentioned memory, and when the above-mentioned instructions are executed, it is as in the first aspect or the first aspect A method in any one of the possible implementations is implemented.
  • the embodiment of the present application provides a computer-readable storage medium, the above-mentioned computer storage medium stores a computer program, and when the above-mentioned computer program is executed, it is as in the first aspect or any possible implementation manner of the first aspect The method in is implemented.
  • the embodiment of the present application provides a computer program, which, when the above computer program is run on a computer, causes the computer to execute the method in the first aspect or any possible implementation manner of the first aspect.
  • FIG. 1 is a schematic diagram of a method for disease diagnosis provided in an embodiment of the present application
  • Fig. 2 is a flow chart of a method for brain atlas construction and neural circuit detection provided in the embodiment of the present application;
  • FIG. 3 is a flowchart of a data processing method provided in an embodiment of the present application.
  • Fig. 4 is a flow chart of a training method provided by the embodiment of the present application.
  • Fig. 5 is a flow chart of a training method provided by the embodiment of the present application.
  • FIG. 6 is an application scene diagram of a method for brain atlas construction and neural circuit detection provided by the embodiment of the present application.
  • Fig. 7a is a schematic structural diagram of a brain atlas construction and neural circuit detection device provided in the embodiment of the present application.
  • Fig. 7b is a schematic structural diagram of a brain atlas construction and neural circuit detection device provided in the embodiment of the present application.
  • FIG. 8 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • an embodiment means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the present application.
  • the occurrences of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is understood explicitly and implicitly by those skilled in the art that the embodiments described herein can be combined with other embodiments.
  • At least one (item) means one or more
  • “multiple” means two or more
  • “at least two (items)” means two or three and three
  • “and/or” is used to describe the association relationship of associated objects, which means that there can be three kinds of relationships, for example, "A and/or B” can mean: only A exists, only B exists, and A and B exist at the same time A case where A and B can be singular or plural.
  • the character “/” generally indicates that the contextual objects are an "or” relationship.
  • “At least one of the following” or similar expressions refer to any combination of these items. For example, at least one item (piece) of a, b or c can mean: a, b, c, "a and b", “a and c", “b and c", or "a and b and c ".
  • Artificial intelligence in the medical field usually uses medical image analysis and mining of patient medical data.
  • Figure 1 As shown in Figure 1, in the study of human brain diseases, the functional or structural information of the brain is usually obtained based on traditional medical images, and combined with the doctor's experience, the abnormalities in the brain are judged , but traditional medical imaging cannot provide brain connection information, and is not suitable for diseases caused by abnormal brain circuits or brain connections.
  • the embodiment of the present application proposes a method for brain atlas construction and neural circuit detection, the above method includes:
  • the above-mentioned first target data represents the time series features of each region in the above-mentioned first image
  • the above-mentioned second target data represents the connection strength of neurons between the regions in the above-mentioned second image
  • the above-mentioned first image is Brain resting state functional magnetic resonance imaging rs-fMRI of the brain
  • the second image is the magnetic resonance diffusion tensor imaging DTI to be processed
  • the area in the first image and the area in the second image are determined according to the template
  • the above The first image and the aforementioned second image correspond to the same target.
  • the above template may be an automatic anatomical labeling (Anatomical Automatic Labeling, AAL) template, which is not limited in this solution.
  • the aforementioned acquisition of the first target data based on the first image can build a brain activity time series extraction framework 4D dynamic feature perception module through the deep learning framework, and then learn rs-fMRI through the above brain activity time series extraction framework 4D dynamic feature perception module
  • the time series features of each voxel in the data combined with the AAL brain region segmentation template, comprehensively calculate the time series average of all voxels contained in each brain region, as the time series features of each brain region, due to the above rs-fMRI
  • the time series signal measured by the data is the time series signal of the blood oxygen level of each voxel in the brain.
  • the above voxel corresponds to the three-dimensional coordinates
  • the time series signal corresponding to the above voxel corresponds to the time coordinate. Therefore, the above time series characteristics can reflect Temporal and spatial information in the brain.
  • the above-mentioned acquisition of the second target data based on the second image can build a fiber bundle target detection module through the deep learning framework, automatically extract and analyze the main white matter fiber bundles in the brain based on the panda database, calculate the connection strength of brain neurons, and combine the AAL brain region Segment the template, count the overall connection strength of neurons contained between different brain regions (that is, the above-mentioned second target data), characterize the topological information between each brain region, and store the data in the form of a matrix.
  • the graph nodes of the above-mentioned first brain atlas are the regions of the above-mentioned first image and/or the regions of the above-mentioned second images;
  • the labels of the graph nodes of the above-mentioned first brain atlas are the serial numbers of the anatomical markers of the above-mentioned templates corresponding to the above-mentioned regions;
  • the node feature of the above-mentioned graph node is the time series feature of the corresponding region in the above-mentioned first target data;
  • the dispersion matrix of the above-mentioned first brain atlas is the third target data, and the above-mentioned third target data represents each graph node in the above-mentioned first brain atlas topological divergence.
  • the above-mentioned third target data can be obtained by calculating the second target data, and the above-mentioned second target data and the third target data satisfy the following relationship:
  • the above-mentioned D G represents the above-mentioned third object data
  • the above-mentioned A G represents the above-mentioned second object data.
  • the above-mentioned first feature is used to represent the high-order topological features of the above-mentioned first brain atlas
  • the above-mentioned first neural network includes two or more layers of neural networks
  • the above-mentioned first neural network is used to extract the information of the above-mentioned first brain atlas.
  • Topological features inputting the above-mentioned first brain atlas into the first neural network to obtain the above-mentioned first features.
  • the above-mentioned inputting the above-mentioned first brain atlas into the first neural network, and outputting the first features include: obtaining the above-mentioned first feature, the first feature above.
  • the above-mentioned first brain atlas and the parameters of the above-mentioned first neural network satisfy the following relationship:
  • the above-mentioned ⁇ represents a nonlinear activation function
  • the above-mentioned l represents the number of layers of the above-mentioned first neural network
  • the above-mentioned W represents the weight matrix to be learned
  • the above-mentioned b represents the deviation to be learned
  • the above-mentioned X G represents the above-mentioned first target data
  • the above-mentioned A G represents the above-mentioned second target data
  • the above-mentioned D G represents the above-mentioned third target data
  • the above-mentioned weight matrix to be learned and the above-mentioned deviation to be learned can be obtained through training.
  • the above-mentioned second neural network is used to decouple the regions in the above-mentioned first brain atlas
  • the above-mentioned first neural loop is used to represent the connection relationship between the regions of the first brain atlas
  • the above-mentioned first neural loop Include at least one neural circuit.
  • Input the above-mentioned first feature into the second neural network decouple the brain regions of the above-mentioned first atlas, and obtain the first neural circuit with the first feature.
  • the above-mentioned first neural circuit can be the first brain At least one neural circuit that causes disease in the atlas and other neural circuits with the first feature, which are not limited in this scheme.
  • the above-mentioned inputting the above-mentioned first feature into the second neural network, and outputting the first neural loop includes:
  • the first value is obtained; when the above-mentioned first value satisfies the preset condition, the regions in the above-mentioned first brain atlas are decoupled to obtain the above-mentioned first neural network. loop.
  • the parameters of the above-mentioned first feature and the above-mentioned second neural network satisfy the following relationship:
  • the above Indicates the above-mentioned first characteristic
  • the above-mentioned area v i the above-mentioned d represents the vector to be learned
  • the above-mentioned b represents the deviation to be learned
  • the above-mentioned Dec(v i ) represents the first value
  • the above-mentioned vector to be learned and the deviation to be learned can be obtained through training
  • the above ⁇ represents a nonlinear activation function
  • the above W represents a weight matrix to be learned
  • the above preconditions include:
  • the above v i represents the above region
  • the above ⁇ represents a pre-set hyperparameter
  • the above d represents a vector to be learned, which can be obtained through training.
  • the complementary information between different modalities is deeply excavated, the high-order topological features between neural circuits are preserved, and redundant interference information is effectively eliminated.
  • Use the first network to extract high-order features of the brain atlas, and then use the second neural network to analyze and decouple the brain neural circuits in the brain atlas to obtain neural circuits to achieve in-depth analysis of brain atlas physics-frequency domain space. The purpose of first-order topological features, and finally realize the accurate detection of neural circuits.
  • FIG. 2 describes the process of processing multi-modal images and identifying target neural circuits, which can be applied to electronic devices.
  • the above-mentioned electronic devices can be computers, which is not limited in this solution.
  • the above-mentioned first neural network and the above-mentioned second neural network may be neural networks received from other devices, or may be trained by the brain map construction and neural loop detection device itself, therefore, the brain map construction and neural network Taking the neural network trained by the loop detection device as an example, the embodiment of the present application provides a training method.
  • Fig. 3 is a flow chart of a training method provided by the embodiment of the present application. As shown in Fig. 3, the above method includes:
  • the above-mentioned first sample data represents the time series of each region in the above-mentioned first sample image
  • the above-mentioned second sample data represents the connection strength of neurons between regions in the above-mentioned second sample image
  • the above-mentioned first This image is an rs-fMRI sample image
  • the above-mentioned second sample image is a DTI sample image
  • the area in the above-mentioned first sample image and the area in the above-mentioned second sample image are determined according to the template
  • the above-mentioned first sample image and the above-mentioned second sample image The two sample images correspond to the same target.
  • the above-mentioned first sample image and the above-mentioned second sample image can be obtained through rs-fMRI and DTI in the Alzheimer's disease neuroimaging initiative (ADNI) database or other databases, the above-mentioned first
  • the category of this image and the above second sample image may include normal elderly control group, early mild cognitive impairment, late mild cognitive impairment and Alzheimer's disease, and the age coverage range is between 57 and 93 years old, But this program does not limit it.
  • step 301 For a specific description of step 301, reference may be made to the relevant method shown in step 201 above, and details are not repeated here.
  • the graph nodes of the above-mentioned second brain atlas are the regions of the above-mentioned first sample image and/or the regions of the above-mentioned second sample image; the labels of the graph nodes of the above-mentioned second brain atlas are the Anatomical marker serial number; the node feature of the above-mentioned map node is the time series feature of the corresponding region in the above-mentioned second sample data; the dispersion matrix of the above-mentioned second brain atlas is the third sample data, and the above-mentioned third sample data represents the above-mentioned second brain atlas The topological divergence of each graph node.
  • the above-mentioned third target data may be obtained by calculating the second target data.
  • step 302 For a specific description of step 302, reference may be made to the related method shown in step 202 above, which will not be repeated here.
  • the above-mentioned second feature is a high-order topological feature of the second brain atlas.
  • step 303 For a specific description of step 303, reference may be made to the related method shown in step 203 above, which will not be repeated here.
  • the above-mentioned second neural network is also used for decoupling the regions in the second brain atlas, and the above-mentioned second neural loop includes at least one neural loop.
  • step 304 For a specific description of step 304, reference may be made to the relevant method shown in step 204 above, which will not be repeated here.
  • the training of the above-mentioned first neural network and the above-mentioned second neural network according to the above-mentioned second neural loop can be performed by comparing the label information of the above-mentioned second neural loop with the above-mentioned first sample image and/or the second sample image , to obtain a loss function, the above-mentioned loss function is used to represent the difference between the above-mentioned second neural loop and the label information in the first sample image and/or the second sample image, through the above-mentioned loss function and backpropagation algorithm, update the above-mentioned first A neural network and parameters in the above-mentioned second neural network.
  • the embodiment of the present application transforms the detection problem of the neural loop into a high-order topological feature analysis problem in the physical-frequency domain space of the brain atlas, avoiding the non-convex optimization brought about by searching the feature space of the target brain region to realize the detection of the neural loop and singularity framing problems.
  • the embodiment of the present application proposes a training method, as shown in FIG. 4, the method includes:
  • the third neural network may adopt a deep learning network, for example, the feature data of each second image may be acquired through a ResNet network. Further, the ResNet network may include i convolutional layers, and the second image may be sequentially processed through convolution of the i convolutional layers to obtain feature data of the second image.
  • the above-mentioned second neural loop is input into the third neural network, and the relevant features that affect the above-mentioned second neural loop are output include:
  • the nodes of the above-mentioned first hypergraph are regions in the above-mentioned second brain atlas
  • the hyperedges of the above-mentioned first hypergraph are the above-mentioned second neural loops
  • the above-mentioned first hypergraph includes the third feature and the first matrix
  • the above-mentioned third feature is the node feature of the above-mentioned first hypergraph
  • the above-mentioned first matrix is a matrix related to the above-mentioned node and the above-mentioned hyperedge
  • the above-mentioned fourth feature is a feature of the influence of the neural circuit on the disease.
  • node features are defined as:
  • the above X H represents the node characteristics, and the above Represents the features of the hyperedges that contain each node.
  • the above-mentioned first hypergraph corresponds to a matrix of dimension
  • the above-mentioned V represents the set of nodes in the above-mentioned first hypergraph (that is, the set of all regions in the above-mentioned second brain atlas)
  • the above-mentioned ⁇ represents the above-mentioned
  • the set formed by the hyperedges of the first hypergraph (i.e. the set formed by all the second neural circuits in the above-mentioned second brain atlas) hypergraph H corresponds to an association matrix H whose dimension is
  • the first probability is obtained according to the above-mentioned third feature, the above-mentioned first matrix, and the parameters of the above-mentioned third neural network;
  • the above-mentioned relevant features and the parameters of the above-mentioned third neural network satisfy the following relation:
  • the above l represents the number of layers of the third neural network, so and above shows the weight matrix to be learned, the above and above Represents the deviation to be learned, the above ⁇ represents the nonlinear activation function, the above H represents the above first matrix, and the above represents Hypergraph region features, above Represents the node features, the above initial features
  • the above-mentioned weight matrix to be learned and the above-mentioned deviation to be learned can be obtained after training.
  • the third feature, the first matrix, and the parameters of the third neural network relevant features can be obtained.
  • the above-mentioned first probability represents the degree of influence on the above-mentioned second neural loop
  • the above-mentioned fourth neural network can use a fully connected network
  • the above-mentioned fully connected network can be realized by convolution operation, and one feature space is linearly transformed into another feature space. Space, integrate the above related features, and output the first probability.
  • the above-mentioned target label information is label information of the first sample image and/or the second sample image, and the first loss is obtained according to the above-mentioned target label information and the first probability.
  • the above-mentioned target label information includes the above-mentioned first
  • the real disease label of a sample image and/or the second sample image, or other effects caused by the second neural circuit in the first sample image and/or the second sample image, are not limited in this solution.
  • the parameters of the above-mentioned first neural network, the parameters of the above-mentioned second neural network, the parameters of the above-mentioned third neural network, and the parameters of the above-mentioned fourth neural network are updated.
  • the parameters of a neural network include a weight matrix to be learned and a deviation to be learned
  • the parameters of the second neural network include a vector to be learned and a deviation to be learned
  • the hypergraph embedding transformation is performed on the abnormal neural loop, and the neural loop can be analyzed from the level of overall feature distribution.
  • a neural loop hyperedge neuron algorithm is designed to achieve the purpose of in-depth analysis of high-order topological features in the physical-frequency domain space of the brain atlas, and finally realize the accurate detection of neural loops.
  • the embodiment of the present application proposes a training method, as shown in FIG. 5 , the method includes:
  • the above random vectors are randomly sampled vectors in the latent space
  • the above third brain atlas is a reconstructed brain atlas output by the generation network
  • the above latent space includes at least two random vectors.
  • the aforementioned inputting the random vector and the aforementioned second neural loop into the generating network, and obtaining the third brain map includes: inputting the aforementioned random vector into the fifth neural network, and obtaining the fifth feature;
  • the above-mentioned second neural circuit is embedded into the sparse brain network to obtain the sixth feature;
  • the above-mentioned fifth feature and the above-mentioned sixth feature are input into the sixth neural network to obtain the above-mentioned third brain atlas.
  • the above-mentioned fifth feature is the topological feature of the above-mentioned random vector
  • the above-mentioned sixth feature is the sparse topological feature of the above-mentioned second neural loop
  • the above-mentioned fifth neural network includes two or more deconvolution layers
  • the above-mentioned first The six neural network is used to fuse the above-mentioned fifth feature with the sixth feature
  • the above-mentioned sparse brain network is a brain function network with sparsity.
  • the above-mentioned random variable is input into the fifth neural network to obtain the fifth feature
  • the above-mentioned fifth feature is a feature matrix
  • the sixth feature is obtained after embedding the above second neural circuit into the sparse brain network, the above second feature is a feature matrix, and the above second feature can be defined as:
  • the above represents the above-mentioned second feature
  • the above-mentioned v i and the above-mentioned v j represent the regions of the above-mentioned second brain atlas
  • the above-mentioned N p represents the second neural circuit.
  • the above-mentioned discriminant network includes two or more layers of perceptrons and a fully connected network, and the characteristics of the above-mentioned second brain atlas and the above-mentioned third brain atlas are obtained by the perceptrons on the above-mentioned second brain atlas and the above-mentioned third brain atlas.
  • the brain atlas is obtained from the probability and label information generated by the generator network to obtain the second loss.
  • the label information includes the brain atlas that is originally coded or the brain atlas output by the generator network.
  • the above Indicates the discriminative network confrontation loss, that is, the discriminative network's ability to distinguish the above reconstructed brain map
  • the above G(z,N 1 ,N 2 ,...,N k ) represents the above third brain atlas
  • the above A represents the above second brain map.
  • the above-mentioned second neural network is also used for decoupling the third brain atlas, and the above-mentioned third neural loop represents the connection relationship between regions in the above-mentioned third brain atlas.
  • step 503 For a specific description of step 503, reference may be made to the above-mentioned related method shown in FIG. 2 , which will not be repeated here.
  • the above third loss represents the difference between the above third neural loop and the second neural loop, and is used to constrain the first neural network and the second neural network.
  • obtaining the third loss according to the above-mentioned third neural circuit and the above-mentioned second neural circuit includes: constructing a second hypergraph according to the above-mentioned second neural circuit, and according to The above-mentioned third neural circuit constructs a third hypergraph; according to the above-mentioned second hypergraph and the above-mentioned third hypergraph, the above-mentioned third loss is obtained.
  • the nodes of the second hypergraph are regions in the second brain atlas, and the hyperedges of the second hypergraph are the second neural circuits.
  • the nodes of the third hypergraph are regions in the third brain atlas, and the hyperedges of the third hypergraph are the third neural circuits.
  • the above-mentioned third loss is obtained according to the parameters of the above-mentioned second hypergraph and the parameters of the above-mentioned third hypergraph; the parameters of the above-mentioned second hypergraph and the parameters of the above-mentioned third hypergraph The parameters satisfy the following relationship:
  • L cap Sim spatial (H,H')+Sim spectral (H,H');
  • the above is the eigenvalue of the Laplacian matrix ⁇ H of the second hypergraph above, and the above is the eigenvalue of the Laplacian matrix ⁇ H' of the above-mentioned third hypergraph
  • the above-mentioned H represents the above-mentioned second hypergraph
  • the above-mentioned H' represents the above-mentioned third hypergraph
  • the above-mentioned N p represents the above-mentioned second neural loop
  • the above-mentioned N p ′ represents the above-mentioned third neural circuit
  • the above-mentioned DV and D ⁇ are the degree matrix of the brain region of the above-mentioned second hypergraph correlation matrix H and the above-mentioned second neural circuit respectively.
  • the above-mentioned parameters include the above-mentioned bias to be learned and the above-mentioned weight matrix to be learned.
  • the discriminant network in the above-mentioned second loss is against The loss is used to update the parameters in the discriminative network, and the generation network adversarial loss in the above second loss is used to update the parameters in the generation network. Update the parameters of the above-mentioned first neural network and the parameters of the above-mentioned second neural network according to the above-mentioned third loss and backpropagation algorithm.
  • the parameters of the network include the vectors to be learned and the biases to be learned.
  • the multi-modal neuroimaging data with a small sample size is used repeatedly and efficiently by using the confrontation generation network, and the third loss is designed to quantitatively describe the high-order topological differences between brain neural circuits.
  • the decoupling mechanism reduces unnecessary parameters, effectively solving the problems of weak robustness, poor generalization ability and overfitting in the learning of small samples of medical images, which significantly enhances the generalization ability of the model and achieves the purpose of neural loop detection , to improve the accuracy of detection.
  • the embodiment of the present application takes Alzheimer's disease as an example, and proposes a method for detecting abnormal neural circuits in Alzheimer's disease, as shown in Figure 6, the above method can be applied to a brain atlas construction and Neural circuit detection device, said device includes a decoupling module, an analysis module, a multi-scale brain atlas generation module (i.e. the generation network in the training method shown in the above-mentioned figure 5) and a discrimination module (i.e. the training method shown in the above-mentioned figure 5 discriminative network in ).
  • a decoupling module i.e. the generation network in the training method shown in the above-mentioned figure 5
  • a discrimination module i.e. the training method shown in the above-mentioned figure 5 discriminative network in
  • the input of the above-mentioned decoupling module is the brain atlas encoding of rs-fMRI and DTI (ie, the second brain atlas in the training method shown in Figure 5 above), and the output is several neural circuits that affect Alzheimer's disease (ie, The second neural loop in the training method shown in Fig. 5 above).
  • the decoupling module includes an Alzheimer's disease topological feature extraction layer (that is, the first neural network in the training method shown in Figure 5 above) and an abnormal neural circuit decoupling layer (that is, the first neural network in the training method shown in Figure 5 above). second neural network).
  • the input of the above analysis module is the abnormal neural circuit related to Alzheimer's disease detected by the decoupling module, and the output is the probability distribution of the neural circuit affecting the development of Alzheimer's disease.
  • the parsing module includes a neural circuit hyperedge neuron layer (that is, the third neural circuit in the training method shown in Figure 5 above) and a fully connected layer (that is, the fourth neural circuit in the training method shown in Figure 5 above). ).
  • the input of the above-mentioned multi-scale brain map generator is the abnormal neural circuit related to Alzheimer's disease detected by the above-mentioned decoupling module and the latent space composed of random vectors, and the output is the reconstructed brain map (that is, the above-mentioned Figure 5
  • the above-mentioned third brain atlas in the training method shown includes a deconvolution layer (that is, the fifth neural circuit in the training method shown in Figure 5 above) and a multi-scale brain atlas feature fusion layer (that is, the fifth neural circuit in the training method shown in Figure 5 above). six neural circuits).
  • the input of the above-mentioned discriminant module is a brain map
  • the above-mentioned brain map includes the brain map of the initial encoding (that is, the above-mentioned second brain map in the training method shown in Figure 5 above) and the reconstructed brain map output by the multi-scale brain map generation module.
  • Atlas the output is the discriminant probability of whether the brain atlas is generated by the multi-scale brain atlas generation module.
  • the discriminative module consists of a multilayer perceptron and a fully connected layer.
  • the brain map G obtained according to rs-fMRI and DTI processing is input into the decoupling module to obtain all abnormal neural circuits affecting Alzheimer's disease N 1 , N 2 , . . . , N k .
  • the abnormal neural circuit reference may be made to the related method shown in step 204 in FIG. 2 above, which will not be repeated here.
  • the above-mentioned decoupling module includes the above-mentioned first neural network and the above-mentioned second neural network, and the above-mentioned abnormal neural loop is input into the analysis module to obtain the probability distribution of the neural loop affecting the development of Alzheimer's disease (that is, as shown in the above-mentioned Figure 5
  • the above-mentioned analysis module includes the above-mentioned third neural network and the above-mentioned fourth neural network, and the above-mentioned y represents the development of Alzheimer's disease Stages of different states (i.e. normal control NC, early mild cognitive impairment, late mild cognitive impairment, and Alzheimer's disease).
  • the above-mentioned methods for obtaining the analysis loss include:
  • the above N 1 , N 2 ,...,N k and the above N 1 ′, N 2 ′,...,N k ′ are used to represent the abnormal neural loop obtained through the decoupling module, and the above loop analysis loss is backpropagated, Guide the parameter update of the decoupling module and the parsing module.
  • the atlas generation module includes the fifth neural network and the sixth neural network.
  • the reconstructed brain atlas and the initial brain atlas are input into the discrimination module, and the features of the topological matrix of the brain atlas are extracted by a multi-layer perceptron, and the extracted features are sent to the Fully connected layer, the output of the fully connected layer is the discriminative probability of whether the brain atlas is generated by the generator.
  • the adversarial loss is obtained according to the above discriminative probability, which includes the discriminant module adversarial loss and generative modules against loss (That is, the above-mentioned second loss in the training method shown in Figure 5 above.
  • the above-mentioned confrontation loss is backpropagated to guide the update of the multi-scale brain atlas generation module and the discrimination module parameters.
  • the above-mentioned reconstructed brain atlas generated by the above multi-scale brain atlas generation module is re-inputted into the decoupling module to obtain abnormal brain neural circuits N 1 ′, N 2 ′, ..., N k ′ after cyclic decoupling. Constrain the decoupling module by discrimination (that is, the above-mentioned third loss in the training method shown in Figure 5 above), and guide the parameter update of the decoupling module.
  • the calculation of the above-mentioned sparse capacity loss can refer to the step 504 in the above-mentioned Figure 5 related methods, which will not be repeated here.
  • the rs-fMRI and DTI of the patient to be assisted in diagnosis and treatment are processed to obtain the time series signal of each voxel and the connection strength of neurons, which are encoded into a brain atlas combined with a template, and the above brain atlas Input the decoupling module that has been trained, and the several neural circuits output by the decoupling module are the abnormal brain neural circuits affecting Alzheimer's disease of the patient.
  • a method for decoupling deep features through the combination of physical space and spectral space is applied to the detection of abnormal neural circuits in Alzheimer's disease. It can efficiently process neuroimaging data of different modalities while ensuring the accuracy of abnormal neural circuit detection.
  • brain rs-fMRI images are encoded as brain atlas data by the time series of voxel features of the three-dimensional region of interest and the structural connection matrix of DTI images, which is beneficial to simultaneously retain the brain function of the three-dimensional brain internal organization Information and physical neuron connection information.
  • the embodiment of the present application provides a device for brain atlas construction and neural circuit detection, as shown in Figure 7a, the device includes:
  • An acquisition unit 701 configured to acquire first target data according to the first image and second target data according to the second image, the first target data represents the time-series features of each region in the first image, and the second target data Represents the connection strength of neurons between the regions in the above second image, the above first image is the resting state functional magnetic resonance imaging rs-fMRI of the brain to be processed, and the above second image is the magnetic resonance diffusion tensor to be processed Imaging DTI; the region in the above-mentioned first image and the region in the above-mentioned second image are determined according to the template;
  • the determining unit 702 is configured to determine a first brain atlas according to the first target data and the second target data, the first brain atlas is used to characterize the relationship between the first target data and the second target data; the first image The region of the above-mentioned second image and/or the region of the above-mentioned second image corresponds to the graph node of the above-mentioned first brain atlas; the label of the graph node of the above-mentioned first brain atlas is the anatomical label serial number of the above-mentioned template corresponding to the above-mentioned region; the node characteristics of the above-mentioned graph node is the time series feature of the corresponding region included in the first target data;
  • the first output unit 703 is configured to input the above-mentioned first brain atlas into the first neural network, and output the first feature, and the above-mentioned first feature is used to represent the high-order topological features of the above-mentioned first brain atlas;
  • the second output unit 704 is configured to input the above-mentioned first feature into the second neural network and output the first neural loop, the above-mentioned second neural network is used to decouple the regions in the above-mentioned first brain atlas, and the above-mentioned first neural network The loop is used to represent the connection relationship between the regions of the first brain atlas.
  • the acquisition unit 701 processes the first image and the second image of the multimodal neuroimaging, obtains the first brain atlas according to the determination unit, and achieves the purpose of efficiently utilizing the complementary information of different modal image data, and then Through the first output unit 703 and the second output unit 704, the first neural circuit is obtained to achieve the purpose of detecting the neural circuit, and the detected neural circuit can be used to assist pathological analysis and traceability of diseases.
  • the device provided in the embodiment of the present application may further include:
  • the above-mentioned first output unit 703 specifically uses parameters according to the above-mentioned first brain atlas and the above-mentioned first neural network to obtain the above-mentioned first feature;
  • the above-mentioned ⁇ represents a nonlinear activation function
  • the above-mentioned l represents the number of layers of the above-mentioned first neural network
  • the above-mentioned W represents the weight matrix to be learned
  • the above-mentioned b represents the deviation to be learned
  • the above-mentioned X G represents the above-mentioned first target data
  • the above-mentioned A G represents the second target data
  • D G represents the third target data
  • the third target data is the dispersion matrix of the first brain atlas
  • the dispersion matrix represents the topological divergence of each graph node in the first brain atlas.
  • the above-mentioned second output unit 704 is specifically configured to obtain the first value according to the above-mentioned first feature and the above-mentioned parameters of the second neural network;
  • the second output unit 704 is specifically configured to decouple the regions in the first brain atlas to obtain the first neural circuit when the first value satisfies a preset condition;
  • the above Indicates the above-mentioned first characteristic
  • the above d represents the vector to be learned
  • the above b represents the deviation to be learned
  • the above Dec(v i ) represents the first value
  • the above ⁇ represents the nonlinear activation function
  • the above W represents the weight to be learned matrix
  • the above preconditions include:
  • the above v i represents the above region
  • the above ⁇ represents a preset hyperparameter
  • the above d represents a vector to be learned.
  • the acquisition unit 701 is further configured to acquire first sample data according to the first sample image and second sample data according to the second sample image, the first sample data representing the first Time series of each region in a sample image, the second sample data represents the connection strength of neurons between regions in the second sample image, the first sample image is an rs-fMRI sample image, the second The sample image is a DTI sample image;
  • the determination unit 702 is further configured to determine a second brain atlas according to the first sample data and the second sample data;
  • the above-mentioned first output unit 703 is further configured to input the above-mentioned second brain map into the above-mentioned first neural network, and output the second feature, and the above-mentioned second feature is a high-order topological feature;
  • the above-mentioned second output unit 704 is also used for inputting the above-mentioned second feature into the second neural network and outputting the second neural loop, and the above-mentioned second neural network is also used for decoupling the regions in the second brain atlas;
  • the training unit 705 is configured to train the above-mentioned first neural network and the above-mentioned second neural network according to the above-mentioned second neural loop.
  • the above-mentioned training unit 705 is specifically configured to input the above-mentioned second neural loop into a third neural network, and output relevant features that affect the above-mentioned second neural loop;
  • the above-mentioned training unit 705 is specifically configured to input the above-mentioned relevant features into the fourth neural network, and output the first probability, and the above-mentioned first probability indicates the degree of influence on the above-mentioned second neural loop;
  • the above training unit is specifically used to obtain the first loss according to the target label information and the above first probability
  • the training unit 705 is specifically configured to train the first neural network, the second neural network, the third neural network, and the fourth neural network according to the first loss.
  • the above-mentioned training unit 705 is specifically configured to construct a first hypergraph according to the above-mentioned second neural circuit, the nodes of the above-mentioned first hypergraph are regions in the above-mentioned second brain atlas, and the above-mentioned first
  • the hyperedge of the hypergraph is the above-mentioned second neural loop, the above-mentioned first hypergraph includes a third feature and a first matrix, the above-mentioned third feature is the node feature of the above-mentioned first hypergraph, and the above-mentioned first matrix is the above-mentioned node and the above-mentioned matrix associated with hyperedges;
  • the above-mentioned training unit 705 is specifically configured to input the above-mentioned third feature and the above-mentioned first matrix into the third neural network to obtain the above-mentioned fourth feature, and the above-mentioned fourth feature is a feature of the influence of the neural circuit on the disease.
  • the above-mentioned training unit 705 is specifically configured to obtain the fourth feature according to the above-mentioned third feature, the above-mentioned first matrix, and the above-mentioned parameters of the third neural network.
  • the above-mentioned training unit 705 is specifically configured to input the random vector and the above-mentioned second neural circuit into the generation network to obtain a third brain atlas;
  • the above-mentioned training unit 705 is specifically used to input the above-mentioned third brain atlas and the above-mentioned second brain atlas into the discriminant network to obtain the second loss;
  • the above-mentioned training unit 705 is specifically configured to input the above-mentioned third brain atlas into the above-mentioned second neural network, and output the third neural loop;
  • the aforementioned training unit 705 is specifically configured to obtain a third loss according to the aforementioned third neural loop and the aforementioned second neural loop;
  • the training unit 705 is specifically configured to train the generation network and the discrimination network according to the second loss, and train the first neural loop and the second neural loop according to the third loss.
  • the above-mentioned training unit 705 is specifically configured to input the above-mentioned random vector into a fifth neural network to obtain a fifth feature, where the above-mentioned fifth feature is a topological feature of the above-mentioned random vector;
  • the above-mentioned training unit 705 is specifically configured to embed the above-mentioned second neural loop into a sparse brain network to obtain a sixth feature, where the above-mentioned sixth feature is a sparse topological feature of the above-mentioned second neural loop;
  • the above-mentioned training unit 705 is specifically configured to input the above-mentioned fifth feature and the above-mentioned sixth feature into the sixth neural network to obtain the above-mentioned third brain atlas.
  • the above-mentioned training unit 705 is specifically configured to construct a second hypergraph according to the above-mentioned second neural loop, and construct a third hypergraph according to the above-mentioned third neural loop;
  • the above-mentioned training unit 705 is specifically configured to obtain the above-mentioned third loss according to the above-mentioned second hypergraph and the above-mentioned third hypergraph.
  • the training unit 705 is specifically configured to obtain the third loss based on parameters of the second hypergraph and parameters of the third hypergraph.
  • FIG. 8 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • the electronic device 80 includes a processor 801, a memory 802, and a communication interface 803; the processor 801, the memory 802, and the communication interface 803 are connected to each other through a bus.
  • the electronic device in FIG. 8 is used to execute the methods for brain map construction and neural circuit detection in the foregoing embodiments.
  • Memory 802 includes, but is not limited to, random access memory (random access memory, RAM), read-only memory (read-only memory, ROM), erasable programmable read-only memory (erasable programmable read only memory, EPROM), or Portable read-only memory (compact disc read-only memory, CDROM), the memory 802 is used for relevant instructions and data.
  • the communication interface 803 is used to receive and send data.
  • the processor 801 may be one or more CPUs. In the case where the processor 801 is one CPU, the CPU may be a single-core CPU or a multi-core CPU.
  • the processor 801 may implement the functions or steps performed by the first determining unit shown in FIG. 7a.
  • a computer-readable storage medium stores a computer program, and when the computer program is executed, the methods for brain map construction and neural circuit detection provided in the foregoing embodiments are realized .
  • the embodiment of the present application provides a computer program product containing instructions, which, when run on a computer, enables the computer to execute the methods for brain atlas construction and neural circuit detection provided in the foregoing embodiments.
  • the disclosed systems, devices and methods may be implemented in other ways.
  • the division of this unit is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored, or not implement.
  • the mutual coupling, or direct coupling, or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.
  • the method and device provided in this application can be applied to the same classification problem with other industry backgrounds, such as other medical image classification tasks. It only needs to be replaced with corresponding medical images as training samples in the training stage, and the system can be used for learning and detection.
  • the computer may be fully or partially implemented by software, hardware, firmware or any combination thereof.
  • software When implemented using software, it may be implemented in whole or in part in the form of a computer program product.
  • the computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the processes or functions according to the embodiments of the present application will be generated in whole or in part.
  • the computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable device.
  • the computer instructions may be stored in or transmitted over a computer-readable storage medium.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

Les modes de réalisation de la présente demande concernent un procédé de construction d'un atlas cérébral et de détection d'un circuit neuronal, ainsi qu'un produit associé. Le procédé comprend les étapes consistant à : acquérir des premières données cibles en fonction d'une première image et des secondes données cibles en fonction d'une seconde image ; déterminer un premier atlas cérébral en fonction des premières et secondes données cibles, le premier atlas cérébral étant utilisé pour représenter la relation entre les premières et secondes données cibles ; entrer le premier atlas cérébral dans un premier réseau neuronal de façon à délivrer en sortie une première caractéristique, la première caractéristique étant utilisée pour représenter une caractéristique topologique d'ordre supérieur du premier atlas cérébral ; et entrer la première caractéristique dans un second réseau neuronal de façon à délivrer en sortie un premier circuit neuronal. Le second réseau neuronal est utilisé pour découpler des zones dans le premier atlas cérébral. Le premier circuit neuronal est utilisé pour représenter une relation de connexion entre les zones du premier atlas cérébral. Ainsi l'objectif de détection d'un circuit neuronal peut-il être atteint.
PCT/CN2021/137913 2021-12-14 2021-12-14 Procédé de construction d'un atlas cérébral et de détection d'un circuit neuronal et produit associé WO2023108418A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2021/137913 WO2023108418A1 (fr) 2021-12-14 2021-12-14 Procédé de construction d'un atlas cérébral et de détection d'un circuit neuronal et produit associé

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2021/137913 WO2023108418A1 (fr) 2021-12-14 2021-12-14 Procédé de construction d'un atlas cérébral et de détection d'un circuit neuronal et produit associé

Publications (1)

Publication Number Publication Date
WO2023108418A1 true WO2023108418A1 (fr) 2023-06-22

Family

ID=86774937

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/137913 WO2023108418A1 (fr) 2021-12-14 2021-12-14 Procédé de construction d'un atlas cérébral et de détection d'un circuit neuronal et produit associé

Country Status (1)

Country Link
WO (1) WO2023108418A1 (fr)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117649917A (zh) * 2024-01-29 2024-03-05 北京大学 检验报告生成模型的训练方法及装置、检验报告生成方法
CN118079185A (zh) * 2024-01-03 2024-05-28 天津大学 基于脑机穿戴交互设备的智能睡眠调控系统
CN118470442A (zh) * 2024-07-10 2024-08-09 华东交通大学 一种基于多尺度超图及特征配准的小样本异常检测方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109528197A (zh) * 2018-11-20 2019-03-29 中国科学院上海生命科学研究院 基于脑功能图谱的猴-人跨物种迁移进行精神疾病的个体化预测方法和系统
CN110931123A (zh) * 2019-12-06 2020-03-27 深圳市德力凯医疗设备股份有限公司 一种矩阵型脑网络及其构建方法
US20200230413A1 (en) * 2019-01-22 2020-07-23 General Electric Company Brain connectivity atlas for personalized functional neurosurgery targeting and brain stimulation programming
CN112735570A (zh) * 2021-01-09 2021-04-30 深圳先进技术研究院 影像驱动的脑图谱构建方法、装置、设备及存储介质
CN112951386A (zh) * 2021-01-09 2021-06-11 深圳先进技术研究院 影像驱动的脑图谱构建方法、装置、设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109528197A (zh) * 2018-11-20 2019-03-29 中国科学院上海生命科学研究院 基于脑功能图谱的猴-人跨物种迁移进行精神疾病的个体化预测方法和系统
US20200230413A1 (en) * 2019-01-22 2020-07-23 General Electric Company Brain connectivity atlas for personalized functional neurosurgery targeting and brain stimulation programming
CN110931123A (zh) * 2019-12-06 2020-03-27 深圳市德力凯医疗设备股份有限公司 一种矩阵型脑网络及其构建方法
CN112735570A (zh) * 2021-01-09 2021-04-30 深圳先进技术研究院 影像驱动的脑图谱构建方法、装置、设备及存储介质
CN112951386A (zh) * 2021-01-09 2021-06-11 深圳先进技术研究院 影像驱动的脑图谱构建方法、装置、设备及存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
SUK HEUNG-IL; WEE CHONG-YAW; LEE SEONG-WHAN; SHEN DINGGANG: "State-space model with deep learning for functional dynamics estimation in resting-state fMRI", NEUROIMAGE, ELSEVIER, AMSTERDAM, NL, vol. 129, 14 January 2016 (2016-01-14), AMSTERDAM, NL , pages 292 - 307, XP029465395, ISSN: 1053-8119, DOI: 10.1016/j.neuroimage.2016.01.005 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118079185A (zh) * 2024-01-03 2024-05-28 天津大学 基于脑机穿戴交互设备的智能睡眠调控系统
CN117649917A (zh) * 2024-01-29 2024-03-05 北京大学 检验报告生成模型的训练方法及装置、检验报告生成方法
CN118470442A (zh) * 2024-07-10 2024-08-09 华东交通大学 一种基于多尺度超图及特征配准的小样本异常检测方法

Similar Documents

Publication Publication Date Title
Majid et al. Classification of stomach infections: A paradigm of convolutional neural network along with classical features fusion and selection
Suganyadevi et al. A review on deep learning in medical image analysis
CN110148142B (zh) 图像分割模型的训练方法、装置、设备和存储介质
US20230342918A1 (en) Image-driven brain atlas construction method, apparatus, device and storage medium
Mansour et al. Internet of things and synergic deep learning based biomedical tongue color image analysis for disease diagnosis and classification
CN112365980B (zh) 脑肿瘤多靶点辅助诊断与前瞻性治疗演化可视化方法及系统
WO2023108418A1 (fr) Procédé de construction d'un atlas cérébral et de détection d'un circuit neuronal et produit associé
CN111009321A (zh) 一种机器学习分类模型在青少年孤独症辅助诊断中的应用方法
WO2022127500A1 (fr) Procédé et appareil de segmentation d'image irm basés sur de multiples réseaux neuronaux, et dispositif
CN113096137B (zh) 一种oct视网膜图像领域适应分割方法及系统
CN111192660B (zh) 一种影像报告分析方法、设备及计算机存储介质
CN113724185B (zh) 用于图像分类的模型处理方法、装置及存储介质
WO2023108873A1 (fr) Réseau cérébral et procédé et appareil de calcul de liaison de dépendance cérébrale
CN113569891A (zh) 一种神经网络模型的训练数据处理、装置、电子设备及存储介质
CN111681247A (zh) 肺叶肺段分割模型训练方法和装置
Elayaraja et al. An efficient approach for detection and classification of cancer regions in cervical images using optimization based CNN classification approach
CN110503636B (zh) 参数调整方法、病灶预测方法、参数调整装置及电子设备
Soundrapandiyan et al. AI-based wavelet and stacked deep learning architecture for detecting coronavirus (COVID-19) from chest X-ray images
CN113935957A (zh) 医疗图像对比方法、装置、电子设备及存储介质
CN114359657B (zh) 脑图谱构建及神经环路检测的方法和相关产品
CN113822439A (zh) 任务预测方法、装置、设备及存储介质
CN111383217B (zh) 大脑成瘾性状评估的可视化方法、装置及介质
CN115937129A (zh) 基于多模态磁共振影像的左右半脑关系的处理方法及装置
Ayus et al. A novel hybrid ensemble based Alzheimer’s identification system using deep learning technique
Jaganathan et al. Modelling an effectual feature selection approach for predicting down syndrome using machine learning approaches

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21967558

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE