CN115813367A - Multi-modal brain network computing method, device, equipment and medium related to structure function - Google Patents
Multi-modal brain network computing method, device, equipment and medium related to structure function Download PDFInfo
- Publication number
- CN115813367A CN115813367A CN202211508683.2A CN202211508683A CN115813367A CN 115813367 A CN115813367 A CN 115813367A CN 202211508683 A CN202211508683 A CN 202211508683A CN 115813367 A CN115813367 A CN 115813367A
- Authority
- CN
- China
- Prior art keywords
- module
- modal
- brain
- discriminator
- connection matrix
- 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.)
- Pending
Links
Images
Abstract
The invention discloses a multi-modal brain network calculation method, a device, equipment and a medium with structure function association, which are applied to training a brain disease prediction model, wherein the model comprises an association perception double-channel generation module, a disease feature regression module, a topological structure discriminator and a time-space joint discriminator.
Description
Technical Field
The present application relates to the field of machine learning technologies, and in particular, to a method, an apparatus, a device, and a medium for multi-modal brain network computation with structure function association.
Background
At present, brain diseases become a ubiquitous health problem in the world today, and seriously endanger the life safety of patients. Therefore, there is an increasing interest in the detection and diagnosis of brain diseases, and the research direction of brain junction is one aspect of the research, which helps to diagnose neurodegenerative diseases and trace pathology by analyzing brain junction, and Alzheimer's Disease (AD) is taken as an example, and brain junction changes occur in Alzheimer's disease patients during the course of disease development. These variable features can be obtained by fMRI, DTI and other brain images, and the conventional method is that a professional physician sets specific parameters through a software template, and the parameters are manually registered and image correction is effectively connected. The traditional pathological feature analysis method highly depends on the experience of professional doctors, has high time cost and labor cost, has great influence on the output effect due to the parameter setting of a software template, and is not beneficial to personalized accurate diagnosis and treatment.
With the development of artificial intelligence technology, a number of brain-connected intelligent computing systems have emerged that do not rely on specialized physicians. Effectively connected intelligent computing systems can be divided into two broad categories: 1) Effective connection intelligent calculation based on single mode; 2) Efficient connected intelligent computing based on multiple modalities. However, the monomodal signals mainly reflect the activity characteristics of brain regions, and have the main defect that the structural characteristics of nerve fibers between the brain regions are lacked, so that the causal action relationship with directionality between the brain regions cannot be guided by using the overall brain topological structure information, and therefore, the learning capability and precision of the model are limited. On the other hand, the existing intelligent computing system based on the multi-modal neuroimaging data only uses an affine splicing or weighted summation mode to fuse the multi-modal data, and the methods have the problems that the heterogeneity-heterogeneity of different modal data is neglected, so that the complementary information among different modalities is difficult to deeply mine, the performance of the model is limited, and the finally obtained model has poor practicability and low precision.
Disclosure of Invention
In view of this, the present application provides a method, an apparatus, a device and a medium for multi-modal brain network computation with structure function association, so as to solve the problems of poor practicability and low accuracy of the existing brain disease prediction model.
In order to solve the technical problem, the application adopts a technical scheme that: the method is applied to training a brain disease prediction model, wherein the brain disease prediction model comprises an association perception double-channel generation module, a disease characteristic regression module, a topological structure discriminator and a time-space joint discriminator; the method comprises the following steps: acquiring brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data; inputting brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data into a correlation perception double-channel generation module to perform interactive correlation perception fusion to obtain multi-modal brain region activity signal characteristics, a multi-modal effective connection matrix and a reconstruction structure connection matrix; inputting the multi-modal effective connection matrix into a disease feature regression module for prediction, inputting the reconstructed structure connection matrix into a topological structure discriminator for prediction, and inputting the multi-modal brain region activity signal features into a time-space joint discriminator for prediction; and reversely updating the associated perception dual-channel generation module, the disease feature regression module, the topological structure discriminator and the time-space joint discriminator according to the prediction result and the pre-constructed loss function.
As a further improvement of the application, the correlation perception dual-channel generation module comprises a brain region feature extraction module, a structure-to-function conversion module, a function-to-structure conversion module, a directional overall cause and effect inference module and a structure decoding module.
As a further improvement of the application, the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data are input to the associated perception dual-channel generation module to carry out interactive associated perception fusion, and the multi-mode brain region activity signal characteristics, the multi-mode effective connection matrix and the reconstruction structure connection matrix are obtained, and the method comprises the following steps: extracting first initial features and second initial features from the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data by using a brain region feature extraction module; inputting the first initial feature and the second initial feature into the structure-to-function conversion module, performing weighted fusion on the feature output by the structure-to-function conversion module and the first initial feature to obtain a new first initial feature, and repeatedly executing the step to finally obtain the multi-modal brain region activity signal feature; inputting the first initial feature and the second initial feature into the function-to-structure conversion module, performing weighted fusion on the feature output by the function-to-structure conversion module and the second initial feature to obtain a new second initial feature, and repeatedly executing the step to finally obtain the multi-modal structural feature; and inputting the multi-modal brain region activity signal characteristics to the directional integral cause and effect inference module to obtain a multi-modal effective connection matrix, and inputting the multi-modal structural characteristics to the structure decoding module to obtain a reconstructed structure connection matrix.
As a further improvement of the present application, inputting the multi-modal effective connection matrix into a disease feature regression module for prediction, inputting the reconstructed structure connection matrix into a topology structure discriminator for prediction, and inputting the multi-modal brain region activity signal features into a time-space joint discriminator for prediction, the method includes: inputting the multi-mode effective connection matrix into a disease characteristic regression module for prediction to obtain disease state prediction probability; inputting the reconstructed structure connection matrix and an empirical structure connection matrix output by the preprocessing software template into a topological structure discriminator for prediction to obtain the probability of whether the reconstructed structure connection matrix is output by the correlation perception dual-channel generation module or the preprocessing software template; inputting the multi-modal brain region activity signal characteristics and the empirical blood oxygen signal output by the preprocessing software template into a time-space joint discriminator for prediction to obtain the probability of whether the multi-modal brain region activity signal characteristics are output by the correlation perception dual-channel generation module or the preprocessing software template.
As a further improvement of the application, the time-space combined discriminator comprises a time difference discrimination module and a space phase discrimination module, wherein the time difference discrimination module is used for restricting the associated perception double-channel generation module from the time continuity characteristic of the brain region activity time sequence signal, and the space phase discrimination module restricts the associated perception double-channel generation module from the space field distribution of the brain region activity signal.
As a further refinement of the present application, the loss functions include disease feature regression loss, topological countermeasure loss, topological perception loss, temporal-spatial joint countermeasure loss, and attribution metric constraint loss;
the disease feature regression loss is used to guide the disease feature regression module and the associated perceptual dual channel generation module to update parameters, which are expressed as:
wherein, the first and the second end of the pipe are connected with each other,
representing regression loss characteristic of the disease, A representing a multimodality effective connection matrix, y representing a disease state, p c (yA) represents the disease state prediction probability,
expressing the expectation of the disease state probability predicted by the model under the distribution of the real labels as a loss function for guiding the learning of the model;
the topological antagonistic loss is used for guiding the updating of parameters of a topological structure discriminator and an associated perception dual-channel generation module, and is expressed as follows:
wherein the content of the first and second substances,
representing a loss function that directs the learning of the topology discriminator,
representing a loss function learned by a topology discriminator directed generator,s represents a reconstructed structure connection matrix, S' represents an empirical structure connection matrix output by a preprocessing software template, D top A representation topology discriminator;
the topology perception loss is used for guiding the update of parameters of the association perception dual-channel generation module, and is expressed as follows:
wherein the content of the first and second substances,
representing topology perception loss, | · | | non-calculation 2 Expressing the Frobenius norm of the matrix, and expressing a preset hyper-parameter by lambda;
the time-space joint countermeasure loss is used for guiding the updating of parameters of a time difference judging module, a space phase judging module and an associated perception double-channel generating module, and is expressed as follows:
wherein, the first and the second end of the pipe are connected with each other,
represents a loss function that directs the learning of a temporal-spatial joint arbiter,
representing a loss function learned by a temporal-spatial joint arbiter directing the generator, D tmp Time difference indicating discrimination module, D spa A representation space phase discrimination module;
the attribution metric constraint loss is used to guide the update of the parameters of the correlation-aware two-channel generation module, which is expressed as:
wherein the content of the first and second substances,
representing attributed metric constraint loss, B represents a multi-modal brain region activity signal feature.
As a further improvement of the application, the topology structure discriminator comprises a multilayer nonlinear topology sensing network and a full connection layer, and the updating formula of the multilayer nonlinear topology sensing network is expressed as follows:
wherein S represents a connection matrix of a reconstruction structure, D represents a weighted dispersion matrix corresponding to the connection matrix of the reconstruction structure, and F ( l ) Representing topological features of layer l, F ( l +1) Represents the topological feature of layer l +1, W ( l ) Is a weight matrix learnable in layer l, b ( l ) Is a non-linear deviation that can be learned in layer l, and σ represents a sigmoid activation function.
In order to solve the above technical problem, the present application adopts another technical solution: a structurally-functionally-linked multimodal brain network computing device is provided, comprising: the acquisition module is used for acquiring brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data; the fusion module is used for inputting the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data into the correlation perception double-channel generation module of the brain disease prediction model to carry out interactive correlation perception fusion so as to obtain multi-modal brain region activity signal characteristics, a multi-modal effective connection matrix and a reconstruction structure connection matrix; the prediction module is used for inputting the multi-modal effective connection matrix into a disease characteristic regression module of the brain disease prediction model for prediction, inputting the reconstructed structure connection matrix into a topological structure discriminator of the brain disease prediction model for prediction, and inputting the multi-modal brain region activity signal characteristics into a time-space combined discriminator of the brain disease prediction model for prediction; and the updating module is used for reversely updating the association perception dual-channel generating module, the disease characteristic regression module, the topological structure discriminator and the time-space combined discriminator according to the prediction result and the pre-constructed loss function.
In order to solve the above technical problem, the present application adopts another technical solution that: there is provided a computer device comprising a processor, a memory coupled to the processor, the memory having stored therein program instructions which, when executed by the processor, cause the processor to perform the steps of a method of multi-modal brain network computing in structural functional association as in any one of the above.
In order to solve the above technical problem, the present application adopts another technical solution that: there is provided a storage medium storing program instructions capable of implementing the method of multi-modal brain network computation associated with structural functionality of any of the above.
The beneficial effect of this application is: according to the structure function associated multi-modal brain network computing method, cross fusion is carried out on brain function magnetic resonance data and magnetic resonance diffusion tensor imaging data through an associated perception double-channel generating module of a brain disease prediction model, a multi-modal brain region activity signal characteristic, a multi-modal effective connection matrix and a reconstructed structure connection matrix are obtained, non-linear multi-level fusion of multi-modal heterogeneous-heterogeneous data is achieved, then countercheck learning is carried out on a disease characteristic regression module through the multi-modal brain region activity signal characteristic, countercheck learning is carried out through the multi-modal effective connection matrix and a topological structure discriminator, countercheck learning is carried out through the reconstructed structure connection matrix and a time-space joint discriminator, a multi-element cooperation generation countercheck strategy is constructed, learning of the model is comprehensively guided from three aspects of time continuity, space field distribution and a topological structure of brain region activity time sequence signals, bidirectional constraint on the function state and the inner structure of multi-modal effective connection is achieved, and accuracy, robustness and generalization ability of the model are greatly improved.
Drawings
Fig. 1 is a schematic structural diagram of a brain disease prediction model according to an embodiment of the present invention;
FIG. 2 is a flow chart of the method for computing a multi-modal brain network with structural and functional associations according to the embodiment of the invention;
FIG. 3 is a schematic structural diagram of an associative perception dual channel generation module according to an embodiment of the present invention;
FIG. 4 is a functional block diagram of a structural and functional association multimodal brain network computing device according to an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a computer apparatus according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a storage medium according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The terms "first", "second" and "third" in this application are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or as implying a number of indicated technical features. Thus, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of the feature. In the description of the present application, "plurality" means at least two, e.g., two, three, etc., unless explicitly specified otherwise. In the embodiment of the present application, all the directional indicators (such as upper, lower, left, right, front, and rear … …) are used only to explain the relative positional relationship between the components, the movement, and the like in a specific posture (as shown in the drawing), and if the specific posture is changed, the directional indicator is changed accordingly. Furthermore, the terms "include" and "have," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those steps or elements listed, but may alternatively include other steps or elements not listed, or inherent to such process, method, article, or apparatus.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the 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 explicitly and implicitly understood by one skilled in the art that the embodiments described herein may be combined with other embodiments.
Fig. 1 is a schematic structural diagram of a brain disease prediction model according to an embodiment of the present invention. As shown in FIG. 1, the brain disease prediction model comprises an associative perception dual-channel generation module, a disease feature regression module, a topological structure discriminator and a time-space joint discriminator. The correlation perception dual-channel generation module is used for performing multi-level nonlinear mutual conversion on brain region image features extracted from brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data, and fusing brain region functional information and physical neuron connection information of internal brain tissues. The disease characteristic regression module is used for predicting the probability of the patient suffering from the tested brain disease according to the multi-mode effective connection matrix output by the correlation perception double-channel generation module. The topological structure discriminator and the time-space combined discriminator are used for comprehensively guiding the learning of the model from three aspects of time continuity, space field distribution and topological structure of the brain region activity time sequence signals, and realizing the bidirectional constraint on the functional state and the intrinsic structure of the multi-mode effective connection.
Fig. 2 is a flow chart of a method for computing a multi-modal brain network associated with structural functions according to an embodiment of the present invention. It should be noted that the method of the present invention is not limited to the flow sequence shown in fig. 2 if the results are substantially the same. As shown in fig. 2, the method for computing a multi-modal brain network with structural and functional associations includes the steps of:
step S101: and acquiring brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data.
Specifically, the present embodiment trains a brain disease prediction model with pre-acquired brain functional Magnetic Resonance Imaging (fMRI) and Magnetic Resonance Diffusion Tensor Imaging (DTI) as sample data.
Step S102: and inputting the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data into an association perception dual-channel generation module for interactive association perception fusion to obtain multi-modal brain region activity signal characteristics, a multi-modal effective connection matrix and a reconstruction structure connection matrix.
Specifically, referring to fig. 3, the correlation perception dual-channel generation module includes a brain region feature extraction module, a structure-to-function conversion module, a function-to-structure conversion module, a directional overall cause and effect inference module, and a structure decoding module. The associated perception dual-channel generation module can extract complementary information of different scales and different levels in heterogeneous data by adopting an alternating multilayer interpenetration structure, and carries out deep complementary information fusion in a nonlinear repeated interaction mode to achieve the effect of efficiently fusing heterogeneous characteristics
Further, step S102 specifically includes:
1. and extracting the first initial characteristic and the second initial characteristic from the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data by using a brain region characteristic extraction module.
Specifically, a brain region feature extraction module is used for extracting brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data respectively, and first initial features extracted from the brain functional magnetic resonance data are recorded as
Recording a second initial feature extracted from the magnetic resonance diffusion tensor imaging data as
Wherein n represents the number of brain regions, and p and q represent each brain region with respect to the first initial feature vectorAnd a dimension for the second initial feature vector.
2. Inputting the first initial characteristic and the second initial characteristic into the structure-to-function conversion module, performing weighted fusion on the characteristic output by the structure-to-function conversion module and the first initial characteristic to obtain a new first initial characteristic, and repeatedly executing the step to finally obtain the multi-modal brain region activity signal characteristic.
3. Inputting the first initial characteristic and the second initial characteristic into the function-to-structure conversion module, performing weighted fusion on the characteristic output by the function-to-structure conversion module and the second initial characteristic to obtain a new second initial characteristic, and repeatedly executing the step to finally obtain the multi-modal structural characteristic.
In this embodiment, the first initial feature and the second initial feature are mutually transformed by the function-to-structure transforming module and the structure-to-function transforming module, and the two transforming modules are implemented based on the correlation sensing Transformer. Wherein, the output formula of the function-to-structure conversion module is as follows:
F2S(X,Y)＝Attention(q(X),k(Y),v(X||η·Y))；
wherein eta represents a hyper-parameter, the default value is 0.1, the symbol | represents the feature association aggregation, q (-), k (-), v (-) are transformation functions formed by a neural network, and X is respectively mapped to the dimension of
Mapping Y to a dimension of
Mapping X | | | η · Y to a dimension of
The feature space of (1).
Similarly, the output formula of the structure-to-function conversion module is as follows:
S2F(X,Y)＝Attention(q(Y),k(X),v(Y||η·X))。
wherein mapping X to a dimension of
Mapping Y to a dimension of
Mapping Y | | η · X to a dimension of
Of the feature space.
In this embodiment, the formula for Attention mechanism Attention is as follows:
wherein, the first and the second end of the pipe are connected with each other,
specifically, when a first initial feature and a second initial feature are input into the structure-to-function conversion module, and the feature output by the structure-to-function conversion module and the first initial feature are subjected to weighting fusion to obtain a new first initial feature, and the weighting coefficient is set to 0.1, similarly, a new second initial feature is obtained by the function-to-structure conversion module, so that the first interactive correlation perception fusion is completed, then the process is performed again on the new first initial feature and the new second initial feature, and after a plurality of times of correlation perception fusion, a multi-modal brain region activity signal feature T = (T =) is finally obtained 1 ,t 2 ,…,t n ) And multimodal structural features D = (D) 1 ,d 2 ,…,d n )。
4. And inputting the multi-modal brain region activity signal characteristics to the directional integral cause and effect inference module to obtain a multi-modal effective connection matrix, and inputting the multi-modal structural characteristics to the structure decoding module to obtain a reconstructed structure connection matrix.
The associated perception dual-channel generation module of the embodiment utilizes an alternative multi-layer interpenetration structure to realize the overall nonlinear fusion of different levels of features, and compared with the traditional cooperative fusion or tower-type fusion, the alternative multi-layer interpenetration structure provided by the embodiment can extract complementary information of different scales and different levels in heterogeneous data, and performs deep complementary information fusion in a nonlinear repeated interaction mode to achieve the effect of efficiently fusing the heterogeneous features.
Step S103: inputting the multi-mode effective connection matrix into a disease characteristic regression module for prediction, inputting the reconstructed structure connection matrix into a topological structure discriminator for prediction, and inputting the multi-mode brain region activity signal characteristics into a time-space joint discriminator for prediction.
Specifically, after obtaining a multi-modal effective connection matrix, a reconstructed structure connection matrix and multi-modal brain region activity signal characteristics, the multi-modal effective connection matrix, the reconstructed structure connection matrix and the multi-modal brain region activity signal characteristics are respectively input into a disease characteristic regression module, a topological structure discriminator and a time-space joint discriminator for counterstudy, wherein the disease characteristic regression module is used for predicting the prediction probability of the brain disease to be tested, and the topological structure discriminator and the time-space joint discriminator are used for comprehensively guiding the study of the model from three aspects of time continuity, space field distribution and topological structure of brain region activity time sequence signals.
Further, step S103 specifically includes:
1. and inputting the multi-mode effective connection matrix into a disease characteristic regression module for prediction to obtain the disease state prediction probability.
Specifically, the disease characteristic regression module takes the multi-modal effective connection matrix as an input and outputs the prediction probability of the tested disease state. The disease characteristic regression module consists of a characteristic perceptron, an information aggregation layer, a high-order characteristic extraction layer, an integral characteristic analysis layer and a state probability prediction network, and finally the probability of the state of the tested disease is obtained. And the correlation perception dual-channel generation module is guided to learn by comparing with the known tested real disease state label.
2. Inputting the reconstructed structure connection matrix and the empirical structure connection matrix output by the preprocessing software template into a topological structure discriminator for prediction to obtain the probability of whether the reconstructed structure connection matrix is output by the correlation perception dual-channel generation module or the preprocessing software template.
Specifically, the topological structure discriminator takes a reconstructed structure connection matrix and an empirical structure connection matrix output by the preprocessing software template as input, and outputs the probability whether the reconstructed structure connection matrix is output by the correlation perception dual-channel generation module or the preprocessing software template. The topological structure discriminator comprises a multilayer nonlinear topological perception network and a full connection layer, and the updating formula of the multilayer nonlinear topological perception network is as follows:
wherein S represents a connection matrix of a reconstruction structure, D represents a weighted dispersion matrix corresponding to the connection matrix of the reconstruction structure, and F ( l ) Indicates the topological characteristics of the l-th layer, F ( l +1) Represents the topological feature of layer l +1, W ( l ) Is a weight matrix learnable in layer I, b ( l ) Is a learnable non-linear bias in layer l, σ denotes the sigmoid activation function, sigmoid is a library function in the deep learning framework.
Specifically, the multilayer nonlinear topology aware network directly quantifies and calculates topological features of each order from a structural connection matrix by using a graph topology iteration technology and sensing a coherence relationship in a structural network, and compared with a traditional method which uses a feature extraction mode of a multilayer perceptron, the multilayer nonlinear topology aware network provided by the embodiment focuses on the learning of the topological features, eliminates the interference of other irrelevant features, and can more comprehensively, systematically and integrally depict structural connection from the topological features.
3. Inputting the multi-modal brain region activity signal characteristics and the empirical blood oxygen signal output by the preprocessing software template into a time-space joint discriminator for prediction to obtain the probability of whether the multi-modal brain region activity signal characteristics are output by the correlation perception dual-channel generation module or the preprocessing software template.
The time-space combined discriminator comprises a time difference discrimination module and a space phase discrimination module. The time difference judging module consists of a time sequence second order difference layer, an oscillation fitting layer, a nonlinear fusion layer and a continuity analysis network; the space discriminator consists of a phase sensing layer, a field intensity detection layer, a field action path calculation layer, a nonlinear fusion layer and a field distribution prediction layer. The time difference judging module is used for restraining the correlation perception double-channel generating module from the time continuity characteristics of the brain region activity time sequence signals, and the space phase judging module is used for restraining the correlation perception double-channel generating module from the space field distribution of the brain region activity signals, so that the bidirectional restraint of the functional state and the built-in structure of the multi-mode effective connection matrix is realized.
Specifically, the time difference judging module and the spatial phase judging module take the multi-modal brain region activity signal characteristics and the empirical blood oxygen signal output by the preprocessing software template as input, and output the probability that the multi-modal brain region activity signal characteristics are output by the correlation perception dual-channel generating module or output by the preprocessing software template.
Step S104: and reversely updating the associated perception dual-channel generation module, the disease feature regression module, the topological structure discriminator and the time-space joint discriminator according to the prediction result and the pre-constructed loss function.
Wherein the loss functions include disease feature regression loss, topological countermeasure loss, topological perception loss, spatio-temporal union countermeasure loss, and attribution measure constraint loss.
The disease characteristic regression loss is constructed based on Kullback-Leibler divergence and is used for guiding the parameter updating of a disease characteristic regression module and an associated perception dual-channel generation module, and the parameter updating is represented as follows:
wherein, the first and the second end of the pipe are connected with each other,
indicating regression loss characteristic of the disease, A indicating a multimodality effective connection matrix, and y indicating a disease state, for example, alzheimer's diseaseThe disease state includes healthy control group, mild cognitive impairment, late stage cognitive impairment, alzheimer's disease, p c (yA) represents the disease state prediction probability,
expressing the expectation of the disease state probability predicted by the model under the distribution of the real labels as a loss function for guiding the learning of the model; .
The topological countermeasure loss is used for guiding the updating of parameters of a topological structure discriminator and an associated perception dual-channel generation module, and is expressed as follows:
wherein the content of the first and second substances,
representing a loss function that directs the learning of the topology discriminator,
representing a loss function which is learned by a generator guided by a topological structure discriminator, wherein the loss function and the loss function jointly form topological countermeasure loss and aim to learn the distribution of structure connection, S represents a reconstructed structure connection matrix, S' represents an empirical structure connection matrix output by a preprocessing software template, and D top A topology discriminator is represented.
In order to better capture the high-order topological structure difference between the reconstructed structure connection and the empirical structure connection, in this embodiment, a topology sensing loss is designed to guide the update of parameters of the associated sensing dual-channel generation module, and the topology sensing loss is expressed as:
wherein the content of the first and second substances,
representing topology perception loss, | · | | non-conducting phosphor 2 The Frobenius norm of the matrix is represented, and λ represents a preset hyper-parameter. Compared with the traditional confrontation learning method which directly compares the brain region characteristics, the topological perception loss describes the low-order-high-order comprehensive topological difference between the reconstructed structure brain network and the empirical structure brain network obtained by the preprocessing software template, after model training is completed, no special physician marks are needed, and the model can accurately learn the topological structure characteristics in the multi-modal data, so that the accuracy of multi-modal effective connection calculation is improved.
The time-space joint countermeasure loss is used for guiding the updating of parameters of a time difference judging module, a space phase judging module and an associated perception double-channel generating module, and is expressed as follows:
wherein the content of the first and second substances,
represents a loss function that directs the learning of a temporal-spatial joint arbiter,
representing a loss function learned by a temporal-spatial joint discriminator directed generator, aiming to let the multi-modal brain region activity signal generated by the model learn the time-frequency distribution, D, of the blood oxygen level dependency extracted from the fMRI data tmp Time difference indicating discrimination module, D spa And a spatial phase discrimination module is shown. The time-space joint proposed by the present embodiment will resist the lossThe time continuity characteristics and the spatial field distribution of the multi-modal brain region activity sequence are integrated into an organic whole, the continuity characteristics of the time sequence are described by introducing a two-step difference layering and an oscillation fitting layer, the spatial field distribution of the brain region activity is described by a phase sensing layer, a field intensity detection layer and a field action path calculation layer, and the two layers jointly constrain counterstudy, so that the problem of low accuracy of the traditional generation counterstudy strategy in time sequence generation and extraction is solved.
It should be understood that the operative connections depict causal relationships between the brain region activity signals, satisfying the structural equations
Is oriented global constraint of, where e i Representing noise, based on the structural equation, the embodiment designs an attribute metric constraint loss to constrain a multi-modal effective connection matrix learned by a model and multi-modal brain region activity signal features, and is used for guiding updating of parameters of a correlation perception dual-channel generation module, and the attribute metric constraint loss is represented as:
wherein the content of the first and second substances,
representing attributed metric constraint loss, B represents a multi-modal brain region activity signal feature. In the associated perception dual-channel generation module of this embodiment, a directional overall cause and effect inference module and an attribution measure constraint loss are designed based on a brain region activity signal structural equation theory in effective connection, and compared with a traditional effective connection calculation model which directly calculates effective connection, the associated perception dual-channel generation module provided by this embodiment simultaneously outputs a multi-modal effective connection matrix and multi-modal brain region activity signal characteristics, and simultaneously utilizes the attribution measure constraint loss to constrain the internal connection between the two, so that the output accuracy of the brain disease prediction model obtained by training of this embodiment is ensuredHigher than the traditional prediction model and stronger interpretability.
According to the multi-modal brain network computing method based on the structural function association, the multi-modal brain region activity signal characteristics, the multi-modal effective connection matrix and the reconstruction structure connection matrix are obtained by performing cross fusion on the brain function magnetic resonance data and the magnetic resonance diffusion tensor imaging data through the association perception double-channel generation module of the brain disease prediction model, the non-linear multi-level fusion of multi-modal heterogeneous-heterogeneous data is achieved, then the multi-modal brain region activity signal characteristics are used for countercheck learning on the disease characteristic regression module, the multi-modal effective connection matrix and the topological structure discriminator are used for countercheck learning, a multi-element cooperation generation countercheck strategy is constructed, learning of the model is comprehensively guided from three aspects of time continuity, space field distribution and topological structure of brain region activity time sequence signals, bidirectional constraint on the functional state and the inner structure of the multi-modal effective connection is achieved, and the precision, robustness and generalization capability of the model are greatly improved.
Furthermore, after the multi-modal brain network associated with the structural function is calculated, the brain disease prediction can be performed by using the brain disease prediction model, and the method for predicting the brain disease by using the brain disease prediction model comprises the following steps:
1. functional magnetic resonance data and magnetic resonance diffusion tensor imaging data of the brain of the patient are acquired.
2. And inputting the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data into the association perception dual-channel generation module for interactive association perception fusion to obtain a multi-mode effective connection matrix.
3. And inputting the multi-mode effective connection matrix into the disease characteristic regression module for prediction to obtain the prediction probability of the brain disease of the patient.
Fig. 4 is a functional module diagram of a multi-modal brain network computing device associated with structural functions according to an embodiment of the present invention. As shown in fig. 4, the structural and functional association multimodal brain network computing apparatus 20 includes an obtaining module 21, a fusion module 22, a prediction module 23, and an updating module 24.
An obtaining module 21, configured to obtain brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data;
the fusion module 22 is used for inputting the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data into the correlation perception dual-channel generation module of the brain disease prediction model to perform interactive correlation perception fusion so as to obtain multi-modal brain region signal characteristics, a multi-modal effective connection matrix and a reconstruction structure connection matrix;
the prediction module 23 is used for inputting the multi-modal effective connection matrix into a disease feature regression module of the brain disease prediction model for prediction, inputting the reconstructed structure connection matrix into a topological structure discriminator of the brain disease prediction model for prediction, and inputting the multi-modal brain region signal features into a time-space joint discriminator of the brain disease prediction model for prediction;
and the updating module 24 is used for reversely updating the association perception dual-channel generating module, the disease characteristic regression module, the topological structure discriminator and the time-space combined discriminator according to the prediction result and the pre-constructed loss function.
Optionally, the association perception dual-channel generation module includes a brain region feature extraction module, a structure-to-function conversion module, a function-to-structure conversion module, a directional overall cause and effect inference module, and a structure decoding module.
Optionally, the fusion module 22 executes an operation of inputting the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data to the association perception dual-channel generation module for interactive association perception fusion to obtain the multi-modal brain region signal features, the multi-modal effective connection matrix, and the connection matrix of the reconstruction structure, and specifically includes: extracting first initial features and second initial features from the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data by using a brain region feature extraction module; inputting the first initial feature and the second initial feature into the structure-to-function conversion module, performing weighted fusion on the feature output by the structure-to-function conversion module and the first initial feature to obtain a new first initial feature, and repeatedly executing the step to finally obtain the multi-modal brain region signal feature; inputting the first initial feature and the second initial feature into the function-to-structure conversion module, performing weighted fusion on the feature output by the function-to-structure conversion module and the second initial feature to obtain a new second initial feature, and repeatedly executing the step to finally obtain the multi-modal structural feature; and inputting the multi-modal brain region signal characteristics to the directional overall cause and effect inference module to obtain a multi-modal effective connection matrix, and inputting the multi-modal structural characteristics to the structure decoding module to obtain a reconstructed structure connection matrix.
Optionally, the predicting module 23 performs operations of inputting the multi-modal effective connection matrix into the disease feature regression module for prediction, inputting the reconstructed structure connection matrix into the topology structure discriminator for prediction, and inputting the multi-modal brain region signal features into the time-space joint discriminator for prediction, which specifically includes: inputting the multi-mode effective connection matrix into a disease characteristic regression module for prediction to obtain disease state prediction probability; inputting the reconstructed structure connection matrix and an empirical structure connection matrix output by the preprocessing software template into a topological structure discriminator for prediction to obtain the probability of whether the reconstructed structure connection matrix is output by a correlation perception dual-channel generation module or the preprocessing software template; inputting the multi-modal brain area signal characteristics and the empirical blood oxygen signals output by the preprocessing software template into a time-space joint discriminator for prediction to obtain the probability of whether the multi-modal brain area signal characteristics are output by the correlation perception dual-channel generation module or the preprocessing software template.
Optionally, the time-space joint discriminator includes a time difference discrimination module and a space phase discrimination module, the time difference discrimination module is configured to constrain the associated perception dual-channel generation module from a time continuity feature of the brain region activity time series signal, and the space phase discrimination module constrains the associated perception dual-channel generation module from a space field distribution of the brain region activity signal.
Optionally, the loss function comprises disease feature regression loss, topological countermeasure loss, topological perceptive loss, spatio-temporal joint countermeasure loss, and attribution metric constraint loss;
the disease feature regression loss is used to guide the disease feature regression module and the associated perceptual dual channel generation module to update parameters, which are expressed as:
wherein, the first and the second end of the pipe are connected with each other,
representing regression loss characteristic of the disease, A representing a multimodality effective connection matrix, y representing a disease state, such as illustrated by Alzheimer's disease, including healthy controls, mild cognitive impairment, late cognitive impairment, alzheimer's disease, p c (|) represents the prediction probability of the disease state,
expressing the expectation of the disease state probability predicted by the model under the distribution of the real labels as a loss function for guiding the learning of the model;
the topological countermeasure loss is used for guiding the updating of parameters of a topological structure discriminator and an associated perception dual-channel generation module, and is expressed as follows:
wherein, the first and the second end of the pipe are connected with each other,
representing a loss function that directs the learning of the topology discriminator,
representing a loss function learned by a topology discriminator directed generator, which together form a topologyResisting loss, aiming at learning the distribution of structural connection, S represents a reconstructed structural connection matrix, S' represents an empirical structural connection matrix output by a preprocessing software template, D top A representation topology discriminator;
the topology perception loss is used for guiding the update of parameters of the association perception dual-channel generation module, and is expressed as follows:
wherein the content of the first and second substances,
representing topology perception loss, | · | | non-conducting phosphor 2 Expressing the Frobenius norm of the matrix, and expressing a preset hyper-parameter by lambda;
the time-space joint countermeasure loss is used for guiding the updating of parameters of a time difference judging module, a space phase judging module and an associated perception double-channel generating module, and is expressed as follows:
wherein the content of the first and second substances,
represents a loss function that directs the learning of a temporal-spatial joint arbiter,
represents a loss function which is learned by a time-space joint discriminator guide generator, aims to lead the multi-modal brain region activity signal generated by a model to learn the time-frequency distribution of the blood oxygen level dependency extracted from fMRI data, and D tmp Time difference indicating discrimination module, D spa Representing a spatial phase discrimination modelA block;
the attribution metric constraint loss is used to guide the update of the parameters of the correlation-aware two-channel generation module, which is expressed as:
wherein the content of the first and second substances,
representing attributed metric constraint loss, B represents a multi-modal brain region signal feature.
Optionally, the topology structure discriminator includes a multilayer nonlinear topology sensing network and a full connection layer, and the update formula of the multilayer nonlinear topology sensing network is represented as:
wherein S represents a connection matrix of a reconstruction structure, D represents a weighted dispersion matrix corresponding to the connection matrix of the reconstruction structure, and F () Representing topological features of layer l, F (+1) Represents the topological feature of layer l +1, W () Is a weight matrix learnable in layer l, b () Is a learnable non-linear bias in layer l, σ denotes the sigmoid activation function, sigmoid is a library function in the deep learning framework.
For other details of the technical solution implemented by each module in the multi-modal brain network computing apparatus in the foregoing embodiment, reference may be made to the description of the multi-modal brain network computing method in the foregoing embodiment, and details are not repeated here.
It should be noted that, in the present specification, the embodiments are all described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments may be referred to each other. For the device-like embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
Referring to fig. 5, fig. 5 is a schematic structural diagram of a computer device according to an embodiment of the present invention. As shown in fig. 5, the computer device 30 includes a processor 31 and a memory 32 coupled to the processor 31, wherein the memory 32 stores program instructions, and the program instructions, when executed by the processor 31, cause the processor 31 to perform the steps of the multi-modal brain network computing method according to any of the embodiments.
The processor 31 may also be referred to as a CPU (Central Processing Unit). The processor 31 may be an integrated circuit chip having signal processing capabilities. The processor 31 may also be a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
Referring to fig. 6, fig. 6 is a schematic structural diagram of a storage medium according to an embodiment of the invention. The storage medium of the embodiment of the present invention stores program instructions 41 capable of implementing the above multi-modal brain network computing method, where the program instructions 41 may be stored in the storage medium in the form of a software product, and include several instructions to enable a computer device (which may be a personal computer, a server, or a network device) or a processor (processor) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a portable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, or a computer device such as a computer, a server, a mobile phone, or a tablet.
In the several embodiments provided in the present application, it should be understood that the disclosed computer apparatus, device and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, a division of a unit is only a logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit may be implemented in the form of hardware, or may also be implemented in the form of a software functional unit. The above embodiments are merely examples and are not intended to limit the scope of the present disclosure, and all modifications, equivalents, and flow charts using the contents of the specification and drawings of the present disclosure or those directly or indirectly applied to other related technical fields are intended to be included in the scope of the present disclosure.
Claims (10)
1. A multi-modal brain network computing method of structure function association is characterized in that the method is applied to training a brain disease prediction model, and the brain disease prediction model comprises an association perception double-channel generation module, a disease feature regression module, a topological structure discriminator and a time-space joint discriminator; the method comprises the following steps:
acquiring brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data;
inputting the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data into the association perception dual-channel generation module for interactive association perception fusion to obtain multi-modal brain region activity signal characteristics, a multi-modal effective connection matrix and a reconstruction structure connection matrix;
inputting the multi-modal effective connection matrix into the disease feature regression module for prediction, inputting the reconstructed structure connection matrix into the topological structure discriminator for prediction, and inputting the multi-modal brain region activity signal features into the time-space joint discriminator for prediction;
and reversely updating the associated perception dual-channel generation module, the disease feature regression module, the topological structure discriminator and the time-space combined discriminator according to a predicted result and a pre-constructed loss function.
2. The method according to claim 1, wherein the correlation-aware two-channel generation module comprises a brain region feature extraction module, a structure-to-function conversion module, a function-to-structure conversion module, a directional global cause and effect inference module, and a structure decoding module.
3. The method for computing the multi-modal brain network in the structural functional relationship according to claim 2, wherein the inputting the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data into a dual-channel correlation perception generating module for interactive correlation perception fusion to obtain the multi-modal brain region activity signal features, the multi-modal effective connection matrix and the reconstruction structure connection matrix comprises:
extracting first initial features and second initial features from the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data respectively by using the brain region feature extraction module;
inputting the first initial feature and the second initial feature into the structure-to-function conversion module, performing weighted fusion on the feature output by the structure-to-function conversion module and the first initial feature to obtain a new first initial feature, and repeatedly executing the step to finally obtain the multi-modal brain region activity signal feature;
inputting the first initial feature and the second initial feature into the function-to-structure conversion module, performing weighted fusion on the features output by the function-to-structure conversion module and the second initial feature to obtain a new second initial feature, and repeatedly executing the step to finally obtain a multi-modal structure feature;
and inputting the multi-modal brain region activity signal characteristics to the directional overall cause and effect inference module to obtain the multi-modal effective connection matrix, and inputting the multi-modal structural characteristics to the structure decoding module to obtain the reconstructed structure connection matrix.
4. The method according to claim 1, wherein the inputting the multi-modal effective connection matrix into the disease feature regression module for prediction, the inputting the reconstructed structure connection matrix into the topology structure discriminator for prediction, and the inputting the multi-modal brain region activity signal features into the temporal-spatial joint discriminator for prediction comprises:
inputting the multi-mode effective connection matrix into the disease characteristic regression module for prediction to obtain disease state prediction probability;
inputting the reconstructed structure connection matrix and an empirical structure connection matrix output by a preprocessing software template into the topological structure discriminator for prediction to obtain the probability of whether the reconstructed structure connection matrix is output by the correlation perception dual-channel generation module or the preprocessing software template;
inputting the multi-modal brain region activity signal characteristics and the empirical blood oxygen signal output by the preprocessing software template into the time-space joint discriminator for prediction to obtain the probability that the multi-modal brain region activity signal characteristics are output by the association perception dual-channel generation module or the preprocessing software template.
5. The method according to claim 4, wherein the temporal-spatial joint discriminator comprises a temporal difference discrimination module and a spatial phase discrimination module, the temporal difference discrimination module is used for constraining the correlation perception two-channel generation module from the temporal continuity feature of the brain region activity time series signal, and the spatial phase discrimination module is used for constraining the correlation perception two-channel generation module from the spatial field distribution of the brain region activity signal.
6. The method of structural-functional associative multimodal brain network computation according to claim 5, wherein the loss functions include disease feature regression loss, topological countermeasure loss, topological perception loss, spatio-temporal joint countermeasure loss, and attribution metric constraint loss;
the disease feature regression loss is used to guide parameter updates for the disease feature regression module and the associative perception dual channel generation module, which are represented as:
wherein the content of the first and second substances,
representing regression loss characteristic of the disease, A representing a multimodality effective connection matrix, y representing a disease state, p c (|) represents the prediction probability of the disease state,
expressing the expectation of the disease state probability predicted by the model under the distribution of the real labels as a loss function for guiding the learning of the model;
the topological countermeasure loss is used for guiding the updating of parameters of the topological structure discriminator and the associated perception dual-channel generation module, and is expressed as follows:
wherein the content of the first and second substances,
show guidance rubbingThe loss function learned by the flapping structure discriminator,
representing a loss function learned by a topology discriminator directed generator, S representing a reconstructed structure connection matrix, S ′ Empirical structure connection matrix, D, representing outputs of preprocessed software templates top A representation topology discriminator;
the topology perception loss is used for guiding the parameter updating of the associated perception dual-channel generation module, and is represented as follows:
wherein the content of the first and second substances,
representing topology perception loss, | · | | non-conducting phosphor 2 Expressing the Frobenius norm of the matrix, and expressing a preset hyper-parameter by lambda;
the time-space joint countermeasure loss is used for guiding the update of parameters of the time difference discrimination module, the space phase discrimination module and the associated perception dual-channel generation module, and is represented as:
wherein the content of the first and second substances,
represents a loss function that directs the learning of a temporal-spatial joint arbiter,
representing a loss function learned by a temporal-spatial joint arbiter directing the generator, D tmp Means for discriminating time difference representation, D spa A representation space phase discrimination module;
the attribution metric constraint penalty is used to guide the associated perceptual dual channel generation module parameter update, which is expressed as:
wherein the content of the first and second substances,
representing attributed metric constraint loss, B represents a multi-modal brain region activity signal feature.
7. The method according to claim 1, wherein the topology discriminator comprises a plurality of layers of nonlinear topology-aware networks and a full connection layer, and the update formula of the plurality of layers of nonlinear topology-aware networks is expressed as:
wherein S represents a connection matrix of a reconstruction structure, D represents a weighted dispersion matrix corresponding to the connection matrix of the reconstruction structure, and F () Representing topological features of layer l, F (+1) Represents the topological feature of layer l +1, W () Is a weight matrix learnable in layer l, b () Is a non-linear deviation that can be learned in layer l, and σ represents a sigmoid activation function.
8. A structurally-functionally-linked multimodal brain network computing device, comprising:
the acquisition module is used for acquiring brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data;
the fusion module is used for inputting the brain functional magnetic resonance data and the magnetic resonance diffusion tensor imaging data into an association perception double-channel generation module of a brain disease prediction model to carry out interactive association perception fusion so as to obtain multi-modal brain region activity signal characteristics, a multi-modal effective connection matrix and a reconstruction structure connection matrix;
the prediction module is used for inputting the multi-modal effective connection matrix into a disease characteristic regression module of a brain disease prediction model for prediction, inputting the reconstructed structure connection matrix into a topological structure discriminator of the brain disease prediction model for prediction, and inputting the multi-modal brain region activity signal characteristics into a time-space joint discriminator of the brain disease prediction model for prediction;
and the updating module is used for reversely updating the associated perception dual-channel generating module, the disease feature regression module, the topological structure discriminator and the time-space combined discriminator according to a predicted result and a pre-constructed loss function.
9. A computer device, characterized in that the computer device comprises a processor, a memory coupled to the processor, in which memory program instructions are stored which, when executed by the processor, cause the processor to carry out the steps of the method of structural-functional associative multimodal brain network computation according to any one of claims 1 to 7.
10. A storage medium characterized by storing program instructions capable of implementing the structural-functional-associative multi-modal brain network computing method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211508683.2A CN115813367A (en) | 2022-11-29 | 2022-11-29 | Multi-modal brain network computing method, device, equipment and medium related to structure function |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211508683.2A CN115813367A (en) | 2022-11-29 | 2022-11-29 | Multi-modal brain network computing method, device, equipment and medium related to structure function |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115813367A true CN115813367A (en) | 2023-03-21 |
Family
ID=85532497
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211508683.2A Pending CN115813367A (en) | 2022-11-29 | 2022-11-29 | Multi-modal brain network computing method, device, equipment and medium related to structure function |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115813367A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116491958A (en) * | 2023-06-28 | 2023-07-28 | 南昌大学第一附属医院 | Target determination device, electronic device, and storage medium |
-
2022
- 2022-11-29 CN CN202211508683.2A patent/CN115813367A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116491958A (en) * | 2023-06-28 | 2023-07-28 | 南昌大学第一附属医院 | Target determination device, electronic device, and storage medium |
| CN116491958B (en) * | 2023-06-28 | 2023-09-19 | 南昌大学第一附属医院 | Target determination device, electronic device, and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jayatilake et al. | Involvement of machine learning tools in healthcare decision making | |
| Jiang et al. | Hi-GCN: A hierarchical graph convolution network for graph embedding learning of brain network and brain disorders prediction | |
| Yin et al. | Locally robust EEG feature selection for individual-independent emotion recognition | |
| WO2023077603A1 (en) | Prediction system, method and apparatus for abnormal brain connectivity, and readable storage medium | |
| Xu et al. | A novel ensemble of random forest for assisting diagnosis of Parkinson's disease on small handwritten dynamics dataset | |
| CN112735570B (en) | Image-driven brain atlas construction method, device, equipment and storage medium | |
| Pal et al. | Deep learning techniques for prediction and diagnosis of diabetes mellitus | |
| Liang et al. | A convolutional neural network combined with prototype learning framework for brain functional network classification of autism spectrum disorder | |
| CN115775626A (en) | Multi-modal brain network computing method, device, equipment and storage medium | |
| CN114415842B (en) | Brain-computer interface decoding method and device based on locus equivalent enhancement | |
| CN115813367A (en) | Multi-modal brain network computing method, device, equipment and medium related to structure function | |
| Deng et al. | SparNet: a convolutional neural network for EEG space-frequency feature learning and depression discrimination | |
| Devarajan et al. | Decision making in health care diagnosis: Evidence from Parkinson's disease via hybrid machine learning | |
| Elayaraja et al. | An efficient approach for detection and classification of cancer regions in cervical images using optimization based CNN classification approach | |
| Kouka et al. | EEG channel selection-based binary particle swarm optimization with recurrent convolutional autoencoder for emotion recognition | |
| Zuo et al. | Brain structure-function fusing representation learning using adversarial decomposed-VAE for analyzing MCI | |
| Chen et al. | DCTNet: Hybrid deep neural network-based EEG signal for detecting depression | |
| Alotaibi et al. | Stroke in-patients' transfer to the ICU using ensemble based model | |
| Gong et al. | Smart urban planning: Intelligent cognitive analysis of healthcare data in cloud-based IoT | |
| Akyol et al. | An ensemble approach for classification of tympanic membrane conditions using soft voting classifier | |
| WO2022188793A1 (en) | Electrophysiological signal classification processing method and apparatus, computer device and storage medium | |
| Liang et al. | Multi-Scale Attention-Based Deep Neural Network for Brain Disease Diagnosis. | |
| KR102639807B1 (en) | Multimodal Multitask Deep Learning Model for Alzheimer's Disease Progression Detection based on Time Series Data | |
| CN114359657A (en) | Method for constructing brain atlas and detecting nerve loop and related product | |
| Lei et al. | Autism spectrum disorder diagnosis based on deep unrolling-based spatial constraint representation |
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 |