CN110889496B - Human brain effect connection identification method based on countermeasure generation network - Google Patents
Human brain effect connection identification method based on countermeasure generation network Download PDFInfo
- Publication number
- CN110889496B CN110889496B CN201911269814.4A CN201911269814A CN110889496B CN 110889496 B CN110889496 B CN 110889496B CN 201911269814 A CN201911269814 A CN 201911269814A CN 110889496 B CN110889496 B CN 110889496B
- Authority
- CN
- China
- Prior art keywords
- brain
- data
- effect connection
- fmri
- brain region
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
- 210000004556 brain Anatomy 0.000 title claims abstract description 144
- 230000000694 effects Effects 0.000 title claims abstract description 85
- 238000000034 method Methods 0.000 title claims abstract description 75
- 238000002599 functional magnetic resonance imaging Methods 0.000 claims abstract description 59
- 230000001364 causal effect Effects 0.000 claims abstract description 42
- 208000009119 Giant Axonal Neuropathy Diseases 0.000 claims abstract description 27
- 201000003382 giant axonal neuropathy 1 Diseases 0.000 claims abstract description 27
- 238000004088 simulation Methods 0.000 claims abstract description 15
- 230000008485 antagonism Effects 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 21
- 239000012636 effector Substances 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 8
- 238000013461 design Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims description 2
- 238000013507 mapping Methods 0.000 claims description 2
- 238000013459 approach Methods 0.000 abstract description 5
- 238000004422 calculation algorithm Methods 0.000 description 8
- 125000002015 acyclic group Chemical group 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 208000014644 Brain disease Diseases 0.000 description 2
- 230000003925 brain function Effects 0.000 description 2
- VZCCETWTMQHEPK-QNEBEIHSSA-N gamma-linolenic acid Chemical compound CCCCC\C=C/C\C=C/C\C=C/CCCCC(O)=O VZCCETWTMQHEPK-QNEBEIHSSA-N 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 208000024827 Alzheimer disease Diseases 0.000 description 1
- 206010012335 Dependence Diseases 0.000 description 1
- 208000012902 Nervous system disease Diseases 0.000 description 1
- 208000025966 Neurological disease Diseases 0.000 description 1
- 208000018737 Parkinson disease Diseases 0.000 description 1
- 208000028017 Psychotic disease Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000013399 early diagnosis Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000002595 magnetic resonance imaging Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000004770 neurodegeneration Effects 0.000 description 1
- 208000015122 neurodegenerative disease Diseases 0.000 description 1
- 238000002610 neuroimaging Methods 0.000 description 1
- 230000008506 pathogenesis Effects 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 230000000284 resting effect Effects 0.000 description 1
- 201000000980 schizophrenia Diseases 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses a human brain effect connection identification method based on an antagonism generation network, which is characterized in that data similar to real fMRI data is generated through GANs, so that priori knowledge in a data set is learned, and then human brain effect connection identification is performed through a model driving method (SEM model). When the effect connection learned by the SEM model approaches the true standard effect connection, the generated simulation data will also be consistent with the true data. Similarly, as the generated simulated data gradually approaches the true data distribution, the effect connection will also truly reflect the causal relationship between the brain regions. Compared with the traditional model driving method, the method provided by the invention can be used for more accurately and effectively identifying the human brain effect connection without priori knowledge, and has good use effect and wide application prospect.
Description
Technical Field
The invention relates to a human brain effect network identification method, in particular to an identification method based on an antagonism generation network (Generative Adversarial Networks, GANs).
Background
The identification of human brain effector junctions from neuroimaging data is an important study topic in human brain junction group studies and has become an effective means for evaluating normal brain function and its associated damage to neurodegenerative diseases. Therefore, the human brain effect connection is accurately identified, and the method has important significance for understanding the causal effect of brain regions in the human brain network, understanding the pathogenesis of brain diseases, and carrying out early diagnosis of brain diseases and research on pathology. In particular, a human brain effect connection network is a graph model consisting of nodes representing brain regions and directed edges, i.e., effect connections, that delineate the causal effects of neural activity exerted by one brain region on another.
Common brain image data includes, but is not limited to, functional magnetic resonance imaging (functional magnetic resonance imaging, fMRI), brain waves (EEG), and the like. Among these, functional magnetic resonance imaging techniques are one that is non-invasive, atraumatic and very efficient brain function imaging techniques based on magnetic resonance imaging. Because the method has a plurality of characteristics of reliable theoretical basis, good space-time resolution and the like, the method provides favorable conditions for experimental study of cognitive neuroscience, has important clinical significance in pathological study, and has wide application prospect and important scientific value. In particular, by using fMRI data to identify human brain effector junctions, it is possible to help understand the operation of the complex human brain and to provide assistance in understanding psychoses and neurological disorders such as alzheimer's and parkinson's disease, schizophrenia, addiction, depression, and the like.
The existing brain effect connection recognition methods can be generally divided into two types, namely a recognition method based on model driving and a recognition method based on data driving. 1) Model-driven methods are a validated method that performs validity checking on brain effect connections based on an a priori connection model, and representative methods include methods of structural equation model (structural equation modeling, SEM), dynamic causal model (dynamic causal modeling, DCM), and the like. Research shows that such methods perform better when the task fMRI data and brain region scale are smaller, and are not suitable for analysis of resting fMRI data or lack of a priori model. 2) The data driving-based method does not need priori knowledge and assumptions, and can directly extract the causal relationship between brain regions from the data. The method comprises a Grangel causal model (Granger causality, GC), a Linear Non-Gaussian acyclic model (Linear Non-Gaussian Acyclic causal Models, liNGAM), a generalized Linear synchronous model (Generalized synchronization, GS), a Bayesian Network (BN) and the like. The data driven approach, while not relying on a priori knowledge, can directly identify effect connections from the data. However, the methods also have the defects of sensitivity to data noise, low recognition accuracy, high calculation complexity, certain limitation on data distribution and the like.
In recent years, the countermeasure generation networks (Generative Adversarial Networks, GANs) have exhibited very excellent performance in unsupervised learning tasks and have been successful in many application fields such as image generation, image synthesis, image conversion, data complementation, causal reasoning, and the like. Specifically, the GANs consists of two parts, a Generator (Generator) and a Discriminator (Discriminator), respectively. The generator is used for learning the real data distribution, and the discriminator is used for judging whether the data is generated or real. The GANs does not generate analog data highly similar to real world samples in the course of the countermeasure learning of the generator and the arbiter, and has obvious advantages compared with other traditional generation methods especially under the conditions of high data noise and few samples.
According to the invention, data similar to real fMRI data is generated through GANs, so that priori knowledge in a data set is learned, and then human brain effect connection identification is performed through a model driving method (SEM model). When the effect connection learned by the SEM model approaches the true standard effect connection, the generated simulation data will also be consistent with the true data. Similarly, as the generated simulated data gradually approaches the true data distribution, the effect connection will also truly reflect the causal relationship between the brain regions.
Compared with the traditional model driving method, the method provided by the invention can more accurately and effectively identify the human brain effect connection without priori knowledge, and has good use effect and wide application prospect.
Disclosure of Invention
Aiming at the defects of the current human brain effect recognition method, the invention provides a human brain effect connection recognition method (EC-GANs) based on an antagonism generation network. The method fully utilizes the advantages of the antagonism generation network, and automatically learns priori knowledge and model parameters required by a Structural Equation Model (SEM) in the process of generating data, thereby effectively identifying effect connection and overcoming the defect that the prior knowledge is required by the traditional model driving method. Experimental results show that the method provided by the invention can identify more reasonable and reliable human brain effect connection without human intervention.
The main idea for realizing the invention is as follows: the present invention is described with fMRI data as an example. Firstly, acquiring real fMRI time series data (non-image data); then, pre-training the EC-GANs of the method provided by the invention by using simulated fMRI data with a standard network, and determining a network structure and super parameters; then, the true fMRI data is taken as input, data similar to the true data is generated by using a Generator (Generator), and judgment is performed by using a Discriminator (Discriminator); training the distinguishing capability of the discriminator to enable the discriminator to distinguish between true fMRI data and generate fMRI data; the generating capacity of the training generator enables the training generator to generate simulated fMRI data which approximates to real fMRI data, so that the discriminator cannot distinguish correctly; finally, when the countermeasure training is finished, namely the generated fMRI data is very close to the real fMRI data, extracting causal parameters of a structural equation model in the current generator, and carrying out human brain effect connection identification by using the causal parameters; in the real experience standard network data, the method is compared with other methods, and the validity of the method is verified.
The invention adopts a technical scheme that the human brain effect connection identification method (EC-GANs) based on an antagonism generation network comprises the following steps:
step 1, acquiring real fMRI time sequence data; to compare the method with existing brain effect connection network identification methods, a set of real fMRI datasets with empirically calibrated networks is used. Because the data in the fMRI data set has the characteristic of high dimension, the whole brain time sequence data is directly used, and the data volume is very large. Therefore, when fMRI data is used, it is necessary to reduce the dimension of the image. And obtaining a representative voxel time sequence on each ROI by using a method for averaging the voxel time sequences of the region of interest, so as to achieve the purpose of reducing the dimension of the data.
Step 2 design is based on brain effect connection identification method (EC-GANs model) of the antagonism generation network, including generator design, arbiter design and loss function design.
Step 3, model pre-training; the generator and the discriminator are both composed of fully connected neural networks, so before use, the number of layers of the neural networks, the number of neurons of each layer of network, an activation function, a learning rate, a sparsification coefficient and the maximum number of parent nodes of each brain region are required to be determined. Training generates simulation data using public simulation fMRI data with a standard network or using a public data generation toolbox. When the EC-GANs model obtains an optimal solution in simulation data, namely the effect connection is highly consistent with the standard effect connection result, the current model parameters are reserved and used as the initial parameters of the EC-GANs model.
Step 4, training an EC-GANs model; the EC-GANs model training comprises two processes, namely, training a discriminator to effectively discriminate real data and generated simulation data, and generating simulation data which is highly similar to the real fMRI data by a training generator. The generator consists of n effect connection generators, wherein n is the number of brain regions. The generator takes as input the n-1 brain region voxel time sequences (excluding the current brain region), the causal parameters and gaussian noise, and generates the voxel time sequences of the corresponding brain regions. When the generated voxel time sequence is consistent with the real voxel time sequence of the brain region, the causal parameter reflects the causal relation between the brain region and other brain regions. The causal parameters are therefore used for the identification of human brain effect connections.
Step 5, human brain effect connection identification; human brain effect connection identification is carried out through causal parameters of an EC-GANs model, and firstly, a causal parameter matrix A epsilon R of all brain areas is obtained n×n If element A in the matrix ij Given a threshold value m, it is assumed that there is one effector connection from brain region j to brain region i, whereas there is no effector connection.
Step 6, result comparison analysis; the algorithm provided by the method is compared with other classical algorithms on a real data set, whether the effect connection identified by the method is consistent with the real data or not is observed, and the advantages and disadvantages of the new method are explored.
Compared with the prior art, the invention has the following obvious advantages and beneficial effects;
(1) The invention inherits the advantages of the traditional model driving method without priori knowledge, and is an effective unsupervised human brain effect connection identification method.
(2) The invention provides a novel method for combining a traditional model driving method with an countermeasure generation network, and compared with other methods, the model can obtain a more real and reliable human brain effect connection network.
(3) Compared with most of traditional effect connection identification methods, the method has no ring constraint and unidirectional constraint, and can identify the ring structure and bidirectional effect connection.
Drawings
Figure 1 is a flow chart of a method according to the invention.
Fig. 2 algorithm architecture framework.
The specific implementation mechanism of the effect connection generator of the five brain regions of fig. 3.
Fig. 4 results of a true fMRI data network construction.
Detailed Description
The following describes embodiments and detailed steps of the invention in terms of two parts, a simulated fMRI dataset and a real fMRI dataset:
and (3) obtaining fMRI voxel time series data. The data set uses real task fMRI data, the data is acquired through a 3T scanner, the repetition Time (TR) is 2s, the number of time points to be tested is 160, and the number of tests is 9. The region of interest was selected as 8 regular brain regions plus one external input, 9 total. The voxel time series of each region of interest is represented by the mean of all voxel time series within the brain region. The regions of interest used in the present invention are shown in table 1.
In particular, since there is no standard result for real fMRI data, an effective method is to introduce input variables and then observe the changes caused by the input variables, so as to detect the recognition effect of different algorithms on brain effect connection in the task state.
(step 2) brain effect connection identification method (EC-GANs model) based on the antagonism generation network. A model overall frame diagram, as shown in fig. 2. Inputs are fMRI real data, outputs are generated fMRI data and effect connections (causal parameters). The EC-GANs model is produced byThe device comprises a forming device and a judging device. Specifically, the generator comprises a number of effect connection generators (equal in number to the number of brain regions), each consisting of a Structural Equation Model (SEM). Assume that there are n brain regions X i (i=1.,), n), each brain region was then expressed as using SEM models:
X i ＝f i (PA(X i ),ε i ),for i＝1,...,n, (1)
wherein, PA (X) i ) Representing brain region X i Is set of parent nodes f i (. Cndot.) is a mapping function, i.e. each brain region X i And the causal relationship of its parent and parent nodes. Epsilon i Is each brain region X i Random gaussian noise of (c). If f is used ij (. Cndot.) represents brain region X j For brain region X i And then equation (1) can be written as:
wherein f ij (·)，j∈PA(X i ) Representing brain region X i Causal relationship with its parent node, if there is a brain effector junction slave brain region X j To brain region X i I.e. X j ->X i F is then ij (. Cndot.) is not equal to 0. Thus, given data set D, brain effect connections for all n brain regions are identified, reduced to computing causal relationships for all brain regions and their parent nodes:
θ D ＝(f 12 ,f 13 ,...,f 1n ,f 21 ,...,f 2n ,...,f n1 ,...,f n(n-1) ) (3)
where D is fMRI voxel time series data, which can be expressed as:
given the causal relationship (effect link) inference method described above, introducing the method described above into an effect link generator creates for each effect linkA generator for generating (estimating) brain region X i The process of voxel time series can be calculated as:
wherein A is ij For brain region X i Causal parameters with other brain regions, A ij By 0 is meant that there is no effector connection between brain region i and brain region j, in particular, when i=j we set a ij =0, i.e. the brain region and its own effector junctions are not considered. Fig. 3 illustrates a specific mechanism for the effect connection generator to generate a temporal sequence of brain region voxels, taking five brain regions as an example. Finally, the effect connection generator is set to add a generation network with a multi-layer fully connected neural network and a than activation function to the estimation model of equation 5. The arbiter is a two-layer feed-forward fully connected neural network and uses the Sigmod function as the activation function. The generator takes the real fMRI data and noise as input to generate analog data similar to the real fMRI data, and then the discriminator distinguishes the two data to give probability values of the data belonging to the real and generated two conditions.
Since the human brain effect connection is sparse in general, we add an L1 regularization term to the loss function to make the effect connection parameter matrix A ij Resulting in a lean fluffing. The sparseness penalty function is:
wherein t is the fMRI time series data length, and lambda is the sparse super-parameter. Finally, the overall model loss function is:
wherein L is p For the sparseness penalty, D (X) represents the probability (function) that fMRI data X is from the true fMRI dataset rather than generating the fMRI dataset, G i (. Cndot.) is brain region X i Is a generator i generates a brain region X i Voxel time series data of (c)),
to not include brain region X i Other brain region voxel time series data of the brain region. />
To expect, P D (X) true fMRI data distribution, < + >>
The spatial (space for generating fMRI data) distribution is assumed. The effect of the loss function is to enable the effect connection generator to generate a lean solution and avoid the over-fitting problem.
(step 3) model pre-training; the model consists of a generator and a discriminator, wherein the generator comprises n effect connection generators (n is the number of brain regions), each generator consists of a fully connected neural network comprising m neurons, each neuron uses a Tanh activation function, and x hidden layers. The arbiter consists of a fully connected neural network containing m neurons, each using a Sigmod activation function, y hidden layers, and after the fully connected layers, batch normalization (batch normalization) operations. In order to select model hyper-parameters, namely the number m of neurons and the number x, y of hidden layers, the learning rate alpha, the sparsification coefficient lambda and the maximum number MaxP of parent nodes. Training and using public simulation fMRI data with a standard network, and when an EC-GANs model obtains an optimal solution in simulation data, namely the effect connection is highly consistent with a standard effect connection result, reserving current model parameters as model initial parameters. The network model parameters that are ultimately used for real fMRI data are: n=10, m=100, x=1, y=2, α=0.1, λ=5, maxp=5.
(step 4) identifying brain effect connections using the EC-GANs model; the model identification effect connection comprises two processes, on one hand, a training discriminator can effectively discriminate real data and generated analog numbersAccording to another aspect, the training generator may generate simulation data that is highly similar to real fMRI data. Wherein the generator takes as input n-1 brain region voxel time sequences (excluding the current brain region), causal parameters and gaussian noise, and generates voxel time sequences of the corresponding brain regions. When the generated voxel time sequence is consistent with the voxel time sequence which is true of the brain region, the causal parameter A epsilon R n×n I.e. it is true that the relationship between this brain region and other brain regions is reflected. The causal parameter can then be used for the identification of human brain effect connections.
(step 5) human brain effect connection identification; human brain effect connection identification is performed by model causal parameters, first, a causal parameter matrix (A E R) of all brain regions is obtained n×n ) If element A in the matrix ij Given a threshold value m, it is assumed that there is one effector connection from brain region j to brain region i, whereas there is no effector connection. In order to avoid the influence caused by improper selection of the threshold value and manual operation, the value of the threshold value m is determined by the maximum parent node number MaxP.
(step 6) result comparison analysis; the algorithm provided by the invention is compared with other classical algorithms on a real data set, whether the effect connection identified by the method is consistent with the real data is observed, and the advantages and disadvantages of the new method are explored. The comparison algorithm comprises:
(1) linear Non-Gaussian acyclic model (Linear Non-Gaussian Acyclic causal Models, liNGAM)
(2) GES (Greedy Equivalence Search), a greedy equivalence class search-based bayesian network method);
as can be seen from fig. 4, the relationship between the Input variable and the brain region, i.e. the two effect connections of Input- > LOCC and Input- > LACC, can be correctly identified by the method provided by the present invention, but the two effect connections cannot be identified by the other two comparison algorithms. The result shows that the method provided by the invention can accurately identify the direction of brain connection and reasonably find the causal relationship between brain areas. Furthermore, the method of the present invention allows to identify bi-directional effect connections, such as between LACCS and RACC, LIFG and RIFG. The result on the true task state fMRI data shows that the method can effectively identify the causal relationship between brain areas, accurately identify human brain effect connection, and has good use effect and wide application prospect.
Claims (4)
1. A human brain effect connection identification method based on an antagonism generation network is characterized by comprising the following steps of: comprises the steps of,
step 1, acquiring real fMRI time sequence data; using a set of real fMRI datasets with empirically derived standard networks; obtaining a representative voxel time sequence on each ROI by using a method of averaging the voxel time sequences of the region of interest, thereby reducing the dimension of the data;
step 2, designing an EC-GANs (brain effect connection recognition model) based on an antagonism generation network, wherein the EC-GANs comprise a generator design, a discriminator design and a loss function design;
step 3, model pre-training; the generator and the discriminator are composed of fully connected neural networks, the number of layers of the neural networks is determined, and the number of neurons, the activation function, the learning rate, the sparsification coefficient and the maximum parent node number of each brain region of each layer of the network are determined; training generates simulation data using public simulation fMRI data with a standard network or using a public data generation toolbox; when the EC-GANs model obtains an optimal solution in simulation data, reserving current model parameters as initial parameters of the EC-GANs model;
step 4, training an EC-GANs model; the EC-GANs model training comprises two processes, namely, on one hand, a training discriminator can effectively discriminate real data and generated simulation data, and on the other hand, a training generator generates the simulation data; the generator consists of n effect connection generators, wherein n is the number of brain regions; the generator takes n-1 brain region voxel time sequences, causal parameters and Gaussian noise as inputs to generate voxel time sequences of corresponding brain regions; when the generated voxel time sequence is consistent with the real voxel time sequence of the brain region, the causal parameter reflects the causal relation between the brain region and other brain regions; the causal parameters are used for identifying human brain effect connection;
step 5, human brain effect connection identification; human by causal parameters of the EC-GANs modelThe identification of brain effect connection, firstly, the causal parameter matrix A epsilon R of all brain areas is obtained n×n If element A in the matrix ij Given a threshold, it is assumed that there is one effector link from brain region j to brain region i, whereas there is no effector link;
step 6, result comparison analysis;
fMRI voxel time series data acquisition; the data set uses real task state fMRI data, the data is obtained through a 3T scanner, the repetition time is 2s, the number of each tested time point is 160, and the number of tested points is 9; the selection of the interested areas is that 8 conventional brain areas are added with one external input, and 9 areas are added;
the EC-GANs model is input into fMRI real data, and output into generated fMRI data and effect connection; each effect connection generator consists of a structural equation model SEM; there are n brain regions X i I=1,..n, each brain region was expressed as using SEM model:
X i ＝f i (PA(X i ),ε i ),for i＝1,...,n, (1)
wherein, PA (X) i ) Representing brain region X i Is set of parent nodes f i (. Cndot.) is a mapping function representing each brain region X i And the causal relationship of the parent node and the parent node; epsilon i Is each brain region X i Random gaussian noise of (a); by f ij (. Cndot.) represents brain region X j For brain region X i And then equation (1) is written as:
wherein f ij (·)，j∈PA(X i ) Representing brain region X i Causal relationship with its parent node, if there is a brain effector junction slave brain region X j To brain region X i F is then ij (. Cndot.) not equal to 0; given data set D, the brain effect connection of all n brain regions is identified, which is simplified to calculate the causal relationship between all brain regions and their parent nodes:
θ D ＝(f 12 ,f 13 ,...,f 1n ,f 21 ,...,f 2n ,...,f n1 ,...,f n(n-1) ) (3)
where D is fMRI voxel time series data, expressed as:
D＝{X 1 ,X 2 ,...,X n }
＝(X 1 ,X 2 ,...,X n ) T (4)
given a causality inference method, introduced into effect connection generators, brain region X is generated for each effect connection generator i The process of voxel time series is calculated as:
wherein A is ij For brain region X i Causal parameters with other brain regions, A ij =0 indicates that there is no effector connection between brain region i and brain region j, when i=j, then a ij ＝0。
2. A human brain effect connection identification method based on an antagonism generation network according to claim 1, wherein: an L1 regular term is added on the loss function, so that the effect is connected with the parameter matrix A ij Generating thin fluffing; the sparseness penalty function is:
wherein t is the length of fMRI time series data, and lambda is a sparse super-parameter; finally, the overall model loss function is:
wherein L is p For the sparsity penalty, D (X) represents the probability that fMRI data X is from the true fMRI dataset rather than generating the fMRI dataset, G i (. Cndot.) is brain region X i Is a microtransaction represented by a multi-layer perceptron,
to not include brain region X i Other brain region voxel time series data of (a); />
For hope of->
For a true fMRI data distribution,/a.about.>
To assume a spatial distribution.
3. A human brain effect connection identification method based on an antagonism generation network according to claim 1, wherein: each generator consists of a fully connected neural network comprising m neurons, each neuron using a Tanh activation function, x hidden layers; the arbiter consists of a fully connected neural network containing m neurons, each neuron using a Sigmod activation function, y hidden layers, and a batch normalization operation after the fully connected layers.
4. A human brain effect connection identification method based on an antagonism generation network according to claim 1, wherein: the value of the threshold in the step 5 is determined by the maximum number of the parent nodes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911269814.4A CN110889496B (en) | 2019-12-11 | 2019-12-11 | Human brain effect connection identification method based on countermeasure generation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911269814.4A CN110889496B (en) | 2019-12-11 | 2019-12-11 | Human brain effect connection identification method based on countermeasure generation network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110889496A CN110889496A (en) | 2020-03-17 |
| CN110889496B true CN110889496B (en) | 2023-06-06 |
Family
ID=69751532
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911269814.4A Active CN110889496B (en) | 2019-12-11 | 2019-12-11 | Human brain effect connection identification method based on countermeasure generation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110889496B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113095366B (en) * | 2021-03-15 | 2024-03-15 | 北京工业大学 | Brain intelligent analysis method based on task state neural image data fusion and uncertain reasoning |
| CN113222147B (en) * | 2021-05-11 | 2024-02-13 | 北华航天工业学院 | Construction method of conditional double-countermeasure learning reasoning model |
| CN113947157B (en) * | 2021-10-26 | 2024-04-30 | 北京工业大学 | Dynamic brain effect connection network generation method based on hierarchical clustering and structural equation model |
| CN114219069B (en) * | 2021-11-16 | 2024-04-26 | 北京工业大学 | Brain effect connection network learning method based on automatic variation self-encoder |
| CN114219068B (en) * | 2021-11-16 | 2024-04-02 | 北京工业大学 | Brain effect connection network learning method based on space-time diagram convolution model |
| WO2023123380A1 (en) * | 2021-12-31 | 2023-07-06 | 深圳先进技术研究院 | Dynamic addiction neural circuit generation method and system based on weakly supervised contrastive learning |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110301431A1 (en) * | 2010-06-05 | 2011-12-08 | The Board Of Trustees Of The Leland Stanford Junior University | Methods of classifying cognitive states and traits and applications thereof |
| US11412975B2 (en) * | 2014-03-28 | 2022-08-16 | The General Hospital Corporation | System and method for measuring functional brain specialization |
| CN105022934B (en) * | 2015-06-29 | 2018-03-09 | 北京工业大学 | A kind of artificial immunity method for being used to build brain effective connectivity network from fMRI data |
| CN108510057A (en) * | 2017-02-27 | 2018-09-07 | 顾泽苍 | A kind of constructive method of the neural network model of ultra-deep confrontation study |
| CN108616470A (en) * | 2018-03-26 | 2018-10-02 | 天津大学 | Modulation Signals Recognition method based on convolutional neural networks |
| CN108564039A (en) * | 2018-04-16 | 2018-09-21 | 北京工业大学 | A kind of epileptic seizure prediction method generating confrontation network based on semi-supervised deep layer |
| CN110169770B (en) * | 2019-05-24 | 2021-10-29 | 西安电子科技大学 | Fine-grained visualization system and method for emotion electroencephalogram |
| CN110309861B (en) * | 2019-06-10 | 2021-05-25 | 浙江大学 | Multi-modal human activity recognition method based on generation of confrontation network |
| CN110503630B (en) * | 2019-07-19 | 2023-05-09 | 江苏师范大学 | Cerebral hemorrhage classifying, positioning and predicting method based on three-dimensional deep learning model |
| CN110503187B (en) * | 2019-07-26 | 2024-01-16 | 深圳万知达科技有限公司 | Implementation method for generating countermeasure network model for generating functional nuclear magnetic resonance imaging data |
-
2019
- 2019-12-11 CN CN201911269814.4A patent/CN110889496B/en active Active
Non-Patent Citations (2)
| Title |
|---|
| ｆＭＲＩ观察高温环境下人脑静息态神经活动变化及意义;钱绍文 等;中国医学影像技术;全文 * |
| Working memory capacity influences performance and brain networks: Evidence from effective connectivity analysis;Nayoung Kim 等;2018 6th International Conference on Brain-Computer Interface (BCI);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110889496A (en) | 2020-03-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110889496B (en) | Human brain effect connection identification method based on countermeasure generation network | |
| CN109376751B (en) | Human brain function network classification method based on convolutional neural network | |
| CN110188836B (en) | Brain function network classification method based on variational self-encoder | |
| Amra et al. | Students performance prediction using KNN and Naïve Bayesian | |
| CN113040715B (en) | Human brain function network classification method based on convolutional neural network | |
| Cheng et al. | Combining convolutional and recurrent neural networks for Alzheimer's disease diagnosis using PET images | |
| WO2023108712A1 (en) | Structural-functional brain network bidirectional mapping model construction method and brain network bidirectional mapping model | |
| Turkson et al. | Classification of Alzheimer’s disease using deep convolutional spiking neural network | |
| CN107133651B (en) | The functional magnetic resonance imaging data classification method of subgraph is differentiated based on super-network | |
| Solo et al. | Connectivity in fMRI: blind spots and breakthroughs | |
| Qiang et al. | Modeling task-based fMRI data via deep belief network with neural architecture search | |
| Yan et al. | Modeling spatio-temporal patterns of holistic functional brain networks via multi-head guided attention graph neural networks (Multi-Head GAGNNs) | |
| CN113947157B (en) | Dynamic brain effect connection network generation method based on hierarchical clustering and structural equation model | |
| Zhu et al. | Deep multi-modal discriminative and interpretability network for Alzheimer’s disease diagnosis | |
| CN113693563A (en) | Brain function network classification method based on hypergraph attention network | |
| CN113951830B (en) | Brain disease classification method based on 3D attention convolution and self-supervision learning | |
| Feng et al. | A review of methods for classification and recognition of ASD using fMRI data | |
| Qiang et al. | Learning brain representation using recurrent Wasserstein generative adversarial net | |
| Qiang et al. | A deep learning method for autism spectrum disorder identification based on interactions of hierarchical brain networks | |
| Qin et al. | Aiding Alzheimer's Disease Diagnosis Using Graph Convolutional Networks Based on rs-fMRI Data | |
| CN114357372A (en) | Aircraft fault diagnosis model generation method based on multi-sensor data driving | |
| Ouhmida et al. | Parkinson’s diagnosis hybrid system based on deep learning classification with imbalanced dataset | |
| Goutham et al. | Brain tumor classification using EfficientNet-B0 model | |
| DE102020215191A1 (en) | DETECTION OF ABNORMAL MOVEMENT SEQUENCES IN REPETITIVE HUMAN ACTIVITIES | |
| Zhao et al. | Spatial temporal graph convolution with graph structure self-learning for early MCI detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |