CN110934587A - Alzheimer disease auxiliary diagnosis method based on atlas neural network - Google Patents
Alzheimer disease auxiliary diagnosis method based on atlas neural network Download PDFInfo
- Publication number
- CN110934587A CN110934587A CN201911278900.1A CN201911278900A CN110934587A CN 110934587 A CN110934587 A CN 110934587A CN 201911278900 A CN201911278900 A CN 201911278900A CN 110934587 A CN110934587 A CN 110934587A
- Authority
- CN
- China
- Prior art keywords
- graph
- neural network
- brain
- node
- disease
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 50
- 238000003745 diagnosis Methods 0.000 title claims abstract description 47
- 238000000034 method Methods 0.000 title claims abstract description 46
- 208000024827 Alzheimer disease Diseases 0.000 title claims abstract description 32
- 210000004556 brain Anatomy 0.000 claims abstract description 46
- 230000003925 brain function Effects 0.000 claims abstract description 36
- 238000003062 neural network model Methods 0.000 claims abstract description 30
- 238000000547 structure data Methods 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 18
- 201000010099 disease Diseases 0.000 claims abstract description 17
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims abstract description 17
- 238000012795 verification Methods 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims abstract description 6
- 239000011159 matrix material Substances 0.000 claims description 27
- 238000012546 transfer Methods 0.000 claims description 23
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 230000001419 dependent effect Effects 0.000 claims description 3
- 238000005192 partition Methods 0.000 claims description 2
- 238000002599 functional magnetic resonance imaging Methods 0.000 claims 2
- 238000001514 detection method Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 7
- 238000011176 pooling Methods 0.000 description 4
- 208000014644 Brain disease Diseases 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 230000032683 aging Effects 0.000 description 2
- 238000004195 computer-aided diagnosis Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 239000003814 drug Substances 0.000 description 2
- 229940079593 drug Drugs 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 206010039966 Senile dementia Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 208000010877 cognitive disease Diseases 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000013399 early diagnosis Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 206010027175 memory impairment Diseases 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000000653 nervous system Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 208000015122 neurodegenerative disease Diseases 0.000 description 1
- 238000002610 neuroimaging Methods 0.000 description 1
- 230000000474 nursing effect Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/40—Detecting, measuring or recording for evaluating the nervous system
- A61B5/4076—Diagnosing or monitoring particular conditions of the nervous system
- A61B5/4088—Diagnosing of monitoring cognitive diseases, e.g. Alzheimer, prion diseases or dementia
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Neurology (AREA)
- General Health & Medical Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Developmental Disabilities (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Neurosurgery (AREA)
- Child & Adolescent Psychology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Hospice & Palliative Care (AREA)
- Signal Processing (AREA)
- High Energy & Nuclear Physics (AREA)
- Psychology (AREA)
- Radiology & Medical Imaging (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
An aided diagnosis method for Alzheimer's disease based on a atlas neural network comprises the following steps: processing the brain functional magnetic resonance image to obtain a time sequence of each brain region; calculating a Pearson correlation coefficient between any two time sequences in the time sequences of each brain region to obtain a brain function connection network which takes each time sequence as a node and the Pearson correlation coefficient as the weight of a connection edge between the two nodes; removing all edges with the weight less than a set threshold value, and simplifying a brain function connection network to obtain graph structure data; and designing a graph convolution neural network model, training the designed graph convolution neural network model by using graph structure data, finally obtaining a training result which is best represented on a verification set and is used as an auxiliary diagnosis model, and outputting a disease state corresponding to the whole graph structure. The method of the invention obtains better detection accuracy than the traditional method and obtains the advanced auxiliary diagnosis classification level on the disease.
Description
Technical Field
The invention relates to the technical field of computer-aided diagnosis, in particular to an aided diagnosis method for Alzheimer's disease based on a graph convolution neural network.
Background
Alzheimer's disease is a major cause of senile dementia, and is a progressive degenerative disease of the nervous system with hidden disease. After the disease progresses to a later stage, interpersonal communication, work and study and daily life of the old people are seriously influenced, and relatively heavy burden is brought to families and society. The incidence of this disease is about 5% in people over 65 years old and up to about 20% over 85 years old, and prevention and treatment of alzheimer's disease is a common concern of governments and medical circles around the world today where the aging problem of the population is growing. By 2016, about 1000 million patients with Alzheimer's disease exist in China, the number of the patients is the first global, and 30 new patients are added each year on average. Although no drug can completely cure the disease at present, medical research shows that the disease development can be greatly relieved if intervention can be performed in the early stage of the disease and medical means such as reasonable drugs are used for treatment, and then good nursing is performed. However, due to the occult onset of alzheimer's disease, in early stages of the disease, the patient may not have significant cognitive dysfunction or only exhibit mild memory impairment, which makes it difficult to distinguish the patient from normal aging. Meanwhile, the disease lacks specific indexes in diagnosis, so that the early diagnosis rate is low. Under the condition, the auxiliary diagnosis of the Alzheimer disease is carried out by analyzing the medical image by the computer, so that the method has wide application prospect and can promote the research of related brain diseases.
In recent years, convolutional neural networks have enjoyed great success in the fields of computer vision and the like. Compared with the traditional method, the method does not need to manually extract the features, directly generates a label through the network, directly combines the feature learning and classification together, and is an end-to-end learning mode. Compared with the traditional method, the neural network can learn more efficient characteristics and patterns. However, the learning objects of convolutional neural networks are limited to regular geometric data, such as two-dimensional, well-aligned images and one-dimensional, regular speech sequences. The human brain is a complex neural network aggregate, is difficult to manually analyze and find out regular features, and cannot be directly converted into images to learn the convolutional neural network. In this case, there are many challenges in performing computer-aided diagnosis of brain diseases, and most researchers manually perform feature selection and then perform modeling by using a conventional machine learning method in the aided diagnosis of alzheimer's disease. These methods have difficulty in further mining a large amount of information contained in brain neuroimaging data, and also have unsatisfactory model classification performance due to manual feature extraction.
Disclosure of Invention
The invention aims to provide an aided diagnosis method for Alzheimer's disease based on a atlas neural network. Compared with a traditional machine learning model, the accuracy rate obtained by the method is remarkably improved, and meanwhile, some important abnormal node information can be obtained and used as an indication of an abnormal region related to the brain diseases.
The technical scheme of the invention is as follows:
an aided diagnosis method for Alzheimer's disease based on a atlas neural network comprises the following steps: step 1) processing a brain functional magnetic resonance image to obtain a time sequence of each brain region; step 2) calculating a Pearson correlation coefficient between any two time sequences in the time sequences of each brain region to obtain a brain function connection network which takes each time sequence as a node and the Pearson correlation coefficient as the weight of a connection edge between the two nodes; step 3) removing all edges with the weight smaller than a set threshold value, and simplifying a brain function connection network to obtain graph structure data; and step 4), designing a graph convolution neural network model, training the designed graph convolution neural network model by using graph structure data, finally obtaining a training result which is best represented on a verification set and is used as an auxiliary diagnosis model, and outputting a disease state corresponding to the whole graph structure.
In the method for aided diagnosis of alzheimer's disease, in step 1), functional magnetic resonance image data of different brain regions is obtained for a period of time of a functional magnetic resonance image of a brain by using a brain region partition template, the data of each time point of a region is filtered and then an average value is taken as a value of the time point of the region, and values of a plurality of time points of the same region form a time sequence of the brain region, so that the time sequence of each brain region can be obtained.
In the above-mentioned aided diagnosis method for Alzheimer's disease, in step 4), for the graph convolution neural network model, set the graph G as (A, F), where A ∈ {0, 1}n×nIn the form of a contiguous matrix, the matrix,
the method is characterized in that a node feature matrix is adopted, each node has d-dimensional features, a classification structure of a graph convolution neural network model is based on a graph neural network, the whole graph convolution neural network comprises information transfer operation of the graph neural network of each layer and information transfer operation similar to convolution between adjacent layers, and the information transfer process of the graph neural network is shown as the following formula:
H(k)＝M(A，H(k-1)；θ(k)) (1)
wherein,
representing node embedding after information transfer through a k-times graph neural network, M being an information transfer function dependent on an adjacency matrix A, a trainable parameter theta(k)And last time messageNode embedding state after information transfer H(k-1)。
In the method for aided diagnosis of alzheimer's disease, a transfer function specifically used is represented by the following formula:
wherein
Is a trainable weight matrix, in each layer, using graph neural network operation iteration K times to obtain the final output node embedding structure, assuming the first layer as
K iterations are represented using Z ═ GNN (a, X), where a is the adjacency matrix and X is the initial input node feature matrix.
In the method for auxiliary diagnosis of Alzheimer's disease, a hierarchical structure is defined to extract hierarchical information in a graph structure, fine granularity is gradually reduced, high-abstraction low-dimensionality features of the whole graph are finally obtained to perform classification tasks, specifically, an output Z and an adjacent matrix A of the graph structure of a certain layer are obtained, and a convolution operation is performed to extract a more abstract graph of the next layer
And
the number of nodes is less than n, and the node distribution matrix of the extracted graph information obtained from the l-th layer is set as
S(l)Each row of (1) corresponds to one layer nlEach column of one of the nodes corresponds to one of the nodes in the layer l +1 and the node in the layer nl +1, and the process of information transfer between the calculation layers is shown as the following formula (3) and formula (4):
the overall graph neural network is divided into two parts, namely a graph neural network for calculating internal information propagation in each layer and a graph neural network for calculating information propagation between adjacent layers like convolution, wherein the calculation formula is shown as formula (5) and formula (6):
Z(l)＝GNNl，embed(A(l)X(l)) (5)
S(l)＝softmax(GNNl，pool(A(l)X(l))) (6)
the output dimensionality of each layer is the maximum node number defined in advance and is a hyper-parameter of the graph convolution neural network model.
The invention also provides an aided diagnosis of Alzheimer's disease based on the atlas neural network, which comprises a brain function connection network diagram structure creation module and an aided diagnosis network model module, wherein: the brain function connection network graph structure creating module is used for converting the brain function magnetic resonance image into a time sequence of brain regions, calculating a Pearson correlation coefficient between any two time sequences in the time sequence of each brain region to obtain a brain function connection network, and simplifying the brain function connection network to obtain graph structure data; and the auxiliary diagnosis network model module is used for designing a graph convolution neural network model, training the designed graph convolution neural network model by using graph structure data, finally obtaining a training result with the best performance on the verification set as an auxiliary diagnosis model, and outputting a disease state corresponding to the whole graph structure.
Compared with the prior art, the invention has the beneficial effects that:
the method of the invention uses the structure of the graph convolutional neural network to carry out classification diagnosis, directly inputs the structure into the brain function connection network, combines the characteristic learning and the classification together, leads the model to automatically mine the information in the network, and simultaneously carries out classification diagnosis according to the learned characteristic information. Compared with the traditional method, the method can save the step of feature selection, automatically learn the information in the brain function network and obtain more accurate classification results and abnormal nodes.
Drawings
The invention is further illustrated by way of example in the following with reference to the accompanying drawings:
fig. 1 is a flowchart of an aided diagnosis method for alzheimer's disease based on a atlas neural network of the present invention.
FIG. 2 is a diagram illustrating the transmission of graph structure data used in the present invention in a graph convolution neural network model as an auxiliary diagnostic model in the present invention.
Fig. 3 is a schematic diagram of a brain function connection network according to the method of the present invention.
Detailed Description
The invention provides an aided diagnosis method of Alzheimer's disease based on a graph convolution neural network, which is a full-automatic aided detection method of Alzheimer's disease based on brain functional magnetic resonance images. The method is characterized in that a model automatically learns a strategy of feature selection and feature fusion, is completely driven by data, and is trained by using a graph convolution neural network to obtain a model for classification diagnosis. Compared with the traditional method, the method utilizes the advantages of deep learning automatic learning characteristics and classification, simultaneously overcomes the problem that irregular data such as brain function connection network is applied to deep learning, and has obviously higher auxiliary diagnosis accuracy than the traditional method.
The principle of the invention is as follows: 1.) processing the brain functional magnetic resonance image to obtain a time sequence of each brain region; 2.) calculating a Pearson correlation coefficient between any two time sequences in each region time sequence to obtain a brain function connection network; 3.) removing all edges of which the correlation coefficients are smaller than a certain threshold value to obtain graph structure data of the input graph convolution neural network model; 4.) training the graph convolutional neural network model by using the graph structure data to obtain a network model finally used for classification as an auxiliary diagnosis model.
The invention provides an aided diagnosis method for Alzheimer's disease based on a graph convolution neural network, which comprises the following steps:
step 1: and processing the brain function magnetic resonance image to obtain a time sequence of each brain region. Specifically, for a period of time of brain functional magnetic resonance images, a brain region division template is used to obtain functional magnetic resonance image data of different brain regions, the data of each time point of a region is filtered and then the average value is taken as the value of the time point of the region, and the values of a plurality of time points of the same region form a time sequence of the brain region, so that the time sequence of each brain region can be obtained.
Step 2: and calculating a Pearson correlation coefficient between any two time sequences in the time sequences of the brain regions to obtain a brain function connection network which takes each time sequence as a node and the Pearson correlation coefficient as the weight of a connection edge between the two nodes. The correlation in the pearson correlation coefficients is non-directional, and thus a brain function connection network is obtained in which each time series is used as a node, and the pearson correlation coefficients between the time series of the regions are used as connection edge weights between the two nodes.
And step 3: and removing all edges with the weight smaller than the set threshold value, and simplifying the brain function connection network to obtain the graph structure data. Specifically, a threshold is set, all edges with the weight smaller than the threshold are removed, the brain function connection network is simplified, and simplified graph structure data are obtained to be used as the input of the graph convolution neural network model.
And 4, step 4: and designing a graph convolution neural network model, training the designed graph convolution neural network model by using graph structure data, finally obtaining a training result which is best represented on a verification set and is used as an auxiliary diagnosis model, and outputting a disease state corresponding to the whole graph structure.
An aided diagnosis system for Alzheimer's disease based on a atlas neural network comprises a brain function connection network graph structure creation module and an aided diagnosis network model module, wherein:
the brain function connection network graph structure creating module is used for converting the brain function magnetic resonance image into a time sequence of brain regions, calculating a Pearson correlation coefficient between any two time sequences in the time sequence of each brain region to obtain a brain function connection network, and simplifying the brain function connection network to obtain graph structure data, namely processing the brain function magnetic resonance image into a graph structure data form input by a graph convolution neural network designed by the invention; and
and the auxiliary diagnosis network model module is used for designing a graph convolution neural network model, training the designed graph convolution neural network model by using graph structure data, finally obtaining a training result which is best represented on the verification set and is used as an auxiliary diagnosis model, and outputting a disease state corresponding to the whole graph structure, namely the auxiliary diagnosis network model module is used for designing the graph convolution neural network model suitable for the graph structure data to perform a classification task.
Fig. 1 is a flowchart of an aided diagnosis method for alzheimer's disease based on a atlas neural network of the present invention. Dividing a time period of brain functional magnetic resonance image into brain regions, filtering data of each time point of each region, taking an average value as a value of the time point of the region, and forming a time sequence of the brain regions by values of a plurality of time points of the same region, so as to obtain the time sequence of each brain region; then, calculating a Pearson correlation coefficient between any two time sequences in the time sequences of the brain area, taking each time sequence as a node, and taking the Pearson correlation coefficient as the weight of a connecting edge between the two nodes to obtain a brain function connecting network; and finally, removing all edges with weights less than a certain threshold value, simplifying the brain function connection network, obtaining a simplified graph structure as the input of the graph convolution neural network model, training the graph convolution neural model by using graph structure data of a training set, obtaining the graph convolution neural network model which is best represented on a verification set, and outputting a disease state corresponding to the whole graph structure.
FIG. 2 is a diagram of graph structure data as an aid in the present inventionThe graph of the diagnostic model is convolved with the transmission schematic in the neural network model. Note that the number of layers and the size illustrated in the schematic diagram are merely exemplary and do not limit the number of layers and the size of a network in which the present invention is actually used. Let diagram G be (A, F), where A ∈ {0, 1}n×nIn the form of a contiguous matrix, the matrix,
is a node feature matrix, each node has d-dimensional features. The classification structure of the graph convolution neural network model is based on the graph neural network, and the whole graph convolution neural network comprises information transfer operation of the graph neural network of each layer and information transfer operation similar to convolution between adjacent layers. The information transmission process of the graph neural network is shown as the following formula:
H(k)＝M(A，H(k-1)；θ(k)) (1) wherein (1) in the above-mentioned formula,
representing node embedding after information transfer through a k-times graph neural network, M being an information transfer function dependent on an adjacency matrix A, a trainable parameter theta(k)And node embedding state H after last information transfer(k-1). Initial value H(0)F, i.e. the entered graph structure data. The transfer function used in particular is shown below:
wherein
Is a trainable weight matrix. In each layer, using graph neural network operation to iterate K times to obtain final output node embedded structure, for example, the first layer is
For simplicity, Z ═ GNN (a, X) is used to represent K iterations, where a is the adjacency matrix and X is the initial input node feature matrix.
The invention defines a hierarchical structure to extract hierarchical information in a graph structure, gradually reduces fine granularity, and finally obtains high-abstraction low-dimensionality characteristics of the whole graph to perform classification tasks. Assuming that a layer of graph structure output Z and an adjacent matrix A are obtained, the convolution operation extracts a next layer of more abstract graph
And
and the input of the operation of the neural network of the next layer graph is used, wherein the number m of the nodes is less than n. Setting the node distribution matrix of the extraction graph information obtained from the l-th layer as
S(l)Each row of (1) corresponds to one layer nlEach column of one of the nodes corresponds to one of the nodes in the layer l +1 and the node in the layer nl +1, and the process of information transfer between the calculation layers is shown as the following formula (3) and formula (4):
the overall graph neural network is divided into two parts, namely a graph neural network for calculating internal information propagation in each layer and a graph neural network for calculating information propagation between adjacent layers like convolution, wherein the calculation formula is shown as formula (5) and formula (6):
Z(l)＝GNNl，embed(A(l)X(l)) (5)
s(l)＝softmax(GNNl，pool(A(l)X(l))) (6)
the output dimensionality of each layer is the maximum node number defined in advance and is a hyper-parameter of the graph convolution neural network model.
Although the input of the two graph neural networks in the overall graph convolution neural network model is the same, the two graph neural networks independently play different roles, the formula (5) is used for generating a new embedded form of the layer graph structure, and the formula (6) is used for generating node probability distribution for transmitting the layer graph structure into the next layer graph structure. This process can be repeated several times, as shown in fig. 2, assuming that the input is graph structure data containing 17 nodes, and after the calculation of the above two graph neural networks, the nodes are divided into 5 clusters, and graph structure data containing 5 nodes, which is subjected to the operation of the layer 1 pooling network layer, is obtained; similarly, the graph structure data of 5 nodes operated by the pooling network layer 1 is divided into 2 clusters through the calculation of the pooling network layer 2, and the graph structure data of the next layer containing 2 nodes is obtained. After the last layer of computation, such as the computation of the 3 rd pooling network layer, all nodes are allocated to a cluster, that is, only the whole graph structure data of one node is available, and the embedded form of the node is input into the graph classification layer of the graph convolution neural network as a feature to obtain a final classification result.
Fig. 3 is a schematic diagram of a brain function connection network according to the present invention. Wherein L represents the left brain hemisphere and R represents the right brain hemisphere. Nodes represent various regions of the brain, and gray connecting lines represent correlations between the various regions. The node size is proportional to the number of connected edges, and the thickness degree of the edges is proportional to the weight of the edges, i.e., the absolute value of the correlation. And finally, the graph structure data of the graph convolution neural network model is input into a numerical expression form of the brain function network.
The above is a specific embodiment of the aided diagnosis method for alzheimer's disease based on the atlas neural network provided by the invention. Using the same feature and training test set on the published ADNI brain magnetic resonance data dataset, the accuracy of the method of the invention was 95.4% ± 0.5%, while the accuracy using the support vector machine was 83.2% ± 1.2%, and the accuracy using the discriminant analysis classifier was 87.6% ± 0.8%. The method provided by the invention obviously improves the accuracy of the assisted diagnosis of the Alzheimer's disease based on the functional magnetic resonance image, and reaches the current leading level.
Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that: any person skilled in the art can modify the technical solutions of the foregoing embodiments or easily conceive of changes, or equivalents of some of the technical features thereof, within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the embodiments of the present invention, and they should be construed as being included therein. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (6)
1. An aided diagnosis method for Alzheimer's disease based on a atlas neural network is characterized by comprising the following steps:
step 1) processing a brain functional magnetic resonance image to obtain a time sequence of each brain region;
step 2) calculating a Pearson correlation coefficient between any two time sequences in the time sequences of each brain region to obtain a brain function connection network which takes each time sequence as a node and the Pearson correlation coefficient as the weight of a connection edge between the two nodes;
step 3) removing all edges with the weight smaller than a set threshold value, and simplifying the brain function connection network to obtain graph structure data; and
and 4) designing a graph convolution neural network model, training the designed graph convolution neural network model by using the graph structure data, finally obtaining a training result which is best represented on a verification set and is used as an auxiliary diagnosis model, and outputting a disease state corresponding to the whole graph structure.
2. The method for aided diagnosis of Alzheimer's disease according to claim 1, wherein in step 1), for a period of time of functional magnetic resonance imaging of brain, functional magnetic resonance imaging data of different brain regions are obtained using a brain region partition template, the data of each time point of a region are filtered and then averaged to obtain the value of the time point of the region, and the values of a plurality of time points of the same region form a time series of brain regions, so that the time series of brain regions can be obtained.
3. The aided diagnosis method for Alzheimer's disease as claimed in claim 1, wherein, in step 4), for the graph convolution neural network model, set graph G as (A, F), where A e {0, 1}n×nIn the form of a contiguous matrix, the matrix,
the method is characterized in that the method is a node feature matrix, each node has d-dimensional features, the classification structure of the graph convolutional neural network model is based on a graph neural network, the whole graph convolutional neural network comprises information transfer operation of the graph neural network of each layer and information transfer operation similar to convolution between adjacent layers, and the information transfer process of the graph neural network is shown as the following formula:
H(k)＝M(A，H(k-1)；θ(k)) (1)
wherein,
representing node embedding after information transfer through a k-times graph neural network, M being an information transfer function dependent on an adjacency matrix A, a trainable parameter theta(k)And node embedding state H after last information transfer(k-1)。
4. The method for aided diagnosis of Alzheimer's disease according to claim 3, wherein the transfer function used is as follows:
wherein
Is a trainable weight matrix, in each layer, using graph neural network operation iteration K times to obtain the final output node embedding structure, assuming the first layer as
K iterations are represented using Z ═ GNN (a, X), where a is the adjacency matrix and X is the initial input node feature matrix.
5. The Alzheimer's disease auxiliary diagnosis method of claim 3, wherein a hierarchical structure is defined to extract hierarchical information in a graph structure, fine granularity is gradually reduced, high-abstraction low-dimensional features of the whole graph are finally obtained to perform classification tasks, specifically, assuming that a certain graph structure output Z and an adjacency matrix A are obtained, a convolution operation is performed to extract a next-layer more abstract graph
And
the number of nodes is less than n, and the node distribution matrix of the extracted graph information obtained from the l-th layer is set as
S(l)Each row of (1) corresponds to one layer nlOne node in each node, each column corresponds to l +1 layers nl+1One node in the nodes calculates the information transfer between the layers as shown in the following formulas (3) and (4):
the overall graph neural network is divided into two parts, namely a graph neural network for calculating internal information propagation in each layer and a graph neural network for calculating information propagation between adjacent layers like convolution, wherein the calculation formula is shown as formula (5) and formula (6):
Z(l)＝GNNl，embed(A(l)X(l)) (5)
S(l)＝softmax(GNNl，pool(A(l)X(l))) (6)
the output dimensionality of each layer is the maximum node number defined in advance and is a hyper-parameter of the graph convolution neural network model.
6. An aided diagnosis of Alzheimer's disease based on a atlas neural network comprises a brain function connection network graph structure creation module and an aided diagnosis network model module, wherein:
the brain function connection network graph structure creating module is used for converting the brain function magnetic resonance image into a time sequence of brain regions, calculating a Pearson correlation coefficient between any two time sequences in the time sequence of each brain region to obtain a brain function connection network, and simplifying the brain function connection network to obtain graph structure data; and
and the auxiliary diagnosis network model module is used for designing a graph convolution neural network model, training the designed graph convolution neural network model by using graph structure data, finally obtaining a training result with the best performance on the verification set as an auxiliary diagnosis model, and outputting a disease state corresponding to the whole graph structure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911278900.1A CN110934587A (en) | 2019-12-13 | 2019-12-13 | Alzheimer disease auxiliary diagnosis method based on atlas neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911278900.1A CN110934587A (en) | 2019-12-13 | 2019-12-13 | Alzheimer disease auxiliary diagnosis method based on atlas neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110934587A true CN110934587A (en) | 2020-03-31 |
Family
ID=69911019
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911278900.1A Pending CN110934587A (en) | 2019-12-13 | 2019-12-13 | Alzheimer disease auxiliary diagnosis method based on atlas neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110934587A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111588349A (en) * | 2020-05-28 | 2020-08-28 | 京东方科技集团股份有限公司 | Health analysis device and electronic equipment |
| CN111812507A (en) * | 2020-05-27 | 2020-10-23 | 浙江工业大学 | Motor fault diagnosis method based on graph convolution |
| CN111938592A (en) * | 2020-08-13 | 2020-11-17 | 天津工业大学 | Missing multi-mode representation learning algorithm for Alzheimer's disease diagnosis |
| CN112294338A (en) * | 2020-09-29 | 2021-02-02 | 山东师范大学 | Epilepsy detection system, equipment and medium based on graph attention network |
| CN112598021A (en) * | 2020-11-27 | 2021-04-02 | 西北工业大学 | Graph structure searching method based on automatic machine learning |
| CN113288131A (en) * | 2021-05-06 | 2021-08-24 | 广东工业大学 | Non-invasive blood glucose detection method, processor and device based on graph convolution network |
| CN113616209A (en) * | 2021-08-25 | 2021-11-09 | 西南石油大学 | Schizophrenia patient discrimination method based on space-time attention mechanism |
| CN113673610A (en) * | 2021-08-25 | 2021-11-19 | 上海鹏冠生物医药科技有限公司 | Image preprocessing method for tissue cell pathological image diagnosis system |
| CN114841280A (en) * | 2022-05-20 | 2022-08-02 | 北京安智因生物技术有限公司 | Prediction classification method, system, medium, equipment and terminal for complex diseases |
| CN118452841A (en) * | 2024-07-10 | 2024-08-09 | 北京智精灵科技有限公司 | Cognitive impairment evaluation method and system based on brain network anomaly detection |
| CN118507005A (en) * | 2024-07-18 | 2024-08-16 | 中国人民解放军联勤保障部队第九八〇医院 | Intelligent gastrointestinal surgery patient personalized nursing path management system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109255791A (en) * | 2018-07-19 | 2019-01-22 | 杭州电子科技大学 | A kind of shape collaboration dividing method based on figure convolutional neural networks |
| CN110111325A (en) * | 2019-05-14 | 2019-08-09 | 深圳大学 | Neuroimaging classification method, terminal and computer readable storage medium |
| CN110522448A (en) * | 2019-07-12 | 2019-12-03 | 东南大学 | A kind of brain network class method based on figure convolutional neural networks |
-
2019
- 2019-12-13 CN CN201911278900.1A patent/CN110934587A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109255791A (en) * | 2018-07-19 | 2019-01-22 | 杭州电子科技大学 | A kind of shape collaboration dividing method based on figure convolutional neural networks |
| CN110111325A (en) * | 2019-05-14 | 2019-08-09 | 深圳大学 | Neuroimaging classification method, terminal and computer readable storage medium |
| CN110522448A (en) * | 2019-07-12 | 2019-12-03 | 东南大学 | A kind of brain network class method based on figure convolutional neural networks |
Non-Patent Citations (1)
| Title |
|---|
| REX YING ET AL: "Hierarchical Graph Representation Learning with Differentiable Pooling", 《ARXIV:1806.08804V4[CS.LG]》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111812507A (en) * | 2020-05-27 | 2020-10-23 | 浙江工业大学 | Motor fault diagnosis method based on graph convolution |
| CN111812507B (en) * | 2020-05-27 | 2023-05-02 | 浙江工业大学 | Motor fault diagnosis method based on graph convolution |
| CN111588349A (en) * | 2020-05-28 | 2020-08-28 | 京东方科技集团股份有限公司 | Health analysis device and electronic equipment |
| CN111588349B (en) * | 2020-05-28 | 2023-12-01 | 京东方科技集团股份有限公司 | Health analysis device and electronic equipment |
| CN111938592A (en) * | 2020-08-13 | 2020-11-17 | 天津工业大学 | Missing multi-mode representation learning algorithm for Alzheimer's disease diagnosis |
| CN111938592B (en) * | 2020-08-13 | 2024-03-12 | 天津工业大学 | Missing multi-modal representation learning algorithm for Alzheimer disease diagnosis |
| CN112294338A (en) * | 2020-09-29 | 2021-02-02 | 山东师范大学 | Epilepsy detection system, equipment and medium based on graph attention network |
| CN112598021A (en) * | 2020-11-27 | 2021-04-02 | 西北工业大学 | Graph structure searching method based on automatic machine learning |
| CN113288131B (en) * | 2021-05-06 | 2022-07-12 | 广东工业大学 | Non-invasive blood glucose detection method, processor and device based on graph convolution network |
| CN113288131A (en) * | 2021-05-06 | 2021-08-24 | 广东工业大学 | Non-invasive blood glucose detection method, processor and device based on graph convolution network |
| CN113673610A (en) * | 2021-08-25 | 2021-11-19 | 上海鹏冠生物医药科技有限公司 | Image preprocessing method for tissue cell pathological image diagnosis system |
| CN113616209B (en) * | 2021-08-25 | 2023-08-04 | 西南石油大学 | Method for screening schizophrenic patients based on space-time attention mechanism |
| CN113616209A (en) * | 2021-08-25 | 2021-11-09 | 西南石油大学 | Schizophrenia patient discrimination method based on space-time attention mechanism |
| CN114841280A (en) * | 2022-05-20 | 2022-08-02 | 北京安智因生物技术有限公司 | Prediction classification method, system, medium, equipment and terminal for complex diseases |
| CN114841280B (en) * | 2022-05-20 | 2023-02-14 | 北京安智因生物技术有限公司 | Prediction classification method, system, medium, equipment and terminal for complex diseases |
| CN118452841A (en) * | 2024-07-10 | 2024-08-09 | 北京智精灵科技有限公司 | Cognitive impairment evaluation method and system based on brain network anomaly detection |
| CN118507005A (en) * | 2024-07-18 | 2024-08-16 | 中国人民解放军联勤保障部队第九八〇医院 | Intelligent gastrointestinal surgery patient personalized nursing path management system |
| CN118507005B (en) * | 2024-07-18 | 2024-10-01 | 中国人民解放军联勤保障部队第九八〇医院 | Intelligent gastrointestinal surgery patient personalized nursing path management system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110934587A (en) | Alzheimer disease auxiliary diagnosis method based on atlas neural network | |
| CN111476292B (en) | Small sample element learning training method for medical image classification processing artificial intelligence | |
| WO2023077603A1 (en) | Prediction system, method and apparatus for abnormal brain connectivity, and readable storage medium | |
| CN110084318B (en) | Image identification method combining convolutional neural network and gradient lifting tree | |
| CN110889852B (en) | Liver segmentation method based on residual error-attention deep neural network | |
| CN110889853B (en) | Tumor segmentation method based on residual error-attention deep neural network | |
| CN107273925B (en) | Lung parenchyma CT image processing device based on local receptive field and semi-supervised depth self-coding | |
| CN109528197B (en) | Individual prediction method and system for mental diseases based on brain function map | |
| CN112446891B (en) | Medical image segmentation method based on U-Net network brain glioma | |
| CN113693563B (en) | Brain function network classification method based on hypergraph attention network | |
| CN110534195B (en) | Alzheimer disease detection method based on data space transformation | |
| CN113344864A (en) | Ultrasonic thyroid nodule benign and malignant prediction method based on deep learning | |
| CN106780453A (en) | A kind of method realized based on depth trust network to brain tumor segmentation | |
| CN116309431B (en) | Visual interpretation method based on medical image | |
| CN114299006A (en) | Self-adaptive multi-channel graph convolution network for joint graph comparison learning | |
| CN112863664A (en) | Alzheimer disease classification method based on multi-modal hypergraph convolutional neural network | |
| Zuo et al. | Alzheimer’s disease prediction via brain structural-functional deep fusing network | |
| CN112529063A (en) | Depth domain adaptive classification method suitable for Parkinson voice data set | |
| CN115496732A (en) | Semi-supervised heart semantic segmentation algorithm | |
| CN118196153A (en) | Cross-modal deformable medical image registration method and system | |
| CN113450363B (en) | Meta-learning cell nucleus segmentation system and method based on label correction | |
| CN113705670A (en) | Brain image classification method and device based on magnetic resonance imaging and deep learning | |
| CN114400086A (en) | Articular disc forward movement auxiliary diagnosis system and method based on deep learning | |
| CN112750131B (en) | Pelvis nuclear magnetic resonance image musculoskeletal segmentation method based on scale and sequence relation | |
| CN115409743B (en) | Model construction method for brain magnetic resonance image processing based on deep learning |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200331 |
|
| RJ01 | Rejection of invention patent application after publication |