CN117809108A - Autism classification method based on maximum entropy weighted independent set pooling - Google Patents
Autism classification method based on maximum entropy weighted independent set pooling Download PDFInfo
- Publication number
- CN117809108A CN117809108A CN202311859339.2A CN202311859339A CN117809108A CN 117809108 A CN117809108 A CN 117809108A CN 202311859339 A CN202311859339 A CN 202311859339A CN 117809108 A CN117809108 A CN 117809108A
- Authority
- CN
- China
- Prior art keywords
- graph
- node
- autism
- imaging
- brain
- 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
- 206010003805 Autism Diseases 0.000 title claims abstract description 37
- 208000020706 Autistic disease Diseases 0.000 title claims abstract description 37
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000011176 pooling Methods 0.000 title claims abstract description 19
- 238000003384 imaging method Methods 0.000 claims abstract description 41
- 210000004556 brain Anatomy 0.000 claims abstract description 27
- 230000000694 effects Effects 0.000 claims abstract description 9
- 239000011159 matrix material Substances 0.000 claims description 18
- 238000002599 functional magnetic resonance imaging Methods 0.000 claims description 15
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 238000002610 neuroimaging Methods 0.000 claims description 10
- 238000003745 diagnosis Methods 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 7
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 4
- 230000001054 cortical effect Effects 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 230000002739 subcortical effect Effects 0.000 claims description 3
- 238000010200 validation analysis Methods 0.000 claims description 3
- 230000003925 brain function Effects 0.000 abstract 1
- 230000004927 fusion Effects 0.000 abstract 1
- 238000005096 rolling process Methods 0.000 abstract 1
- 238000002595 magnetic resonance imaging Methods 0.000 description 5
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 208000029560 autism spectrum disease Diseases 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000002059 diagnostic imaging Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000000284 resting effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000002598 diffusion tensor imaging Methods 0.000 description 2
- 230000008506 pathogenesis Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000000547 structure data Methods 0.000 description 2
- 208000024827 Alzheimer disease Diseases 0.000 description 1
- 208000014644 Brain disease Diseases 0.000 description 1
- 206010051290 Central nervous system lesion Diseases 0.000 description 1
- 208000012902 Nervous system disease Diseases 0.000 description 1
- 208000025966 Neurological disease Diseases 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 239000000090 biomarker Substances 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000007177 brain activity Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000004195 computer-aided diagnosis Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000013399 early diagnosis Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000001123 neurodevelopmental effect Effects 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 230000003989 repetitive behavior Effects 0.000 description 1
- 208000013406 repetitive behavior Diseases 0.000 description 1
- 201000000980 schizophrenia Diseases 0.000 description 1
- 230000003997 social interaction Effects 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a method for classifying autism based on maximum entropy weighted independent set pooling, which can simultaneously utilize brain function connection characteristics and non-imaging characteristics to classify autism. The method first uses maximum entropy weighted independent set pooling to downsample the constructed brain graph into a substructure, and then uses flattened substructure features and real labels to train the multi-layer perceptron to extract higher-order features. A crowd graph is then constructed in which non-imaging information similarity between subjects is represented as edge weights of the crowd graph and a threshold is set for the edge weights to form edge connections. The high-order features extracted by the multi-layer perceptron are embedded into the nodes of the crowd graph, and the image rolling network learning nodes are embedded, so that the effective fusion and utilization of the imaging and non-imaging information of the subject are realized, and finally, the node classification is carried out on the output crowd graph by using a linear classifier, so that the identification of the autism is realized. The present invention enables learning of functional connections and regional activities of the whole brain network while integrating non-imaging data of the subject to achieve classification of autism and to achieve excellent performance in this task.
Description
Technical Field
The invention belongs to the field of medical image processing and analysis, and particularly relates to a resting state functional magnetic resonance imaging (resting state fMRI, rs-fMRI) analysis and deep learning technology such as a graph neural network, and the like, wherein the classification of autism is realized by utilizing rs-fMRI data and non-imaging data.
Background
Autism spectrum disorder (Autism Spectrum Disorder, ASD), also known as autism, is a disorder of neurodevelopmental development that affects language, social interaction and is accompanied by repetitive behavior. Autism symptoms usually appear in childhood and accompany the life of the patient, causing huge harm to the patient themselves and also bringing a huge burden to society. In addition, the number of patients diagnosed with autism has been increasing and the prevalence has continued to increase. At present, the medical imaging technology has been widely applied to research of various brain diseases, such as autism, alzheimer's disease, schizophrenia and the like, and common neuroimaging diagnosis technologies related to autism spectrum disorders include magnetic resonance imaging (Magnetic Resonance Imaging, MRI), diffusion tensor imaging (Diffusion Tensor Imgaing, DTI) and the like. The development of medical imaging technology creates conditions for studying the pathogenesis of autism from different angles. Non-invasive multi-modality magnetic resonance imaging is a powerful and safe technique that is widely used in brain research. The structural magnetic resonance imaging (sMRI) has the characteristics of clarity and high resolution, can provide high-quality three-dimensional images of brain structures and detailed structural information, and is a means for clinically assisting doctors in quantitatively analyzing certain brain lesions or specific areas. Functional magnetic resonance imaging (fMRI) is considered as one of the most promising imaging techniques for studying neurological diseases such as autism, and reflects brain activity by detecting blood oxygen levels to specific brain regions, and researchers can obtain data in different states, such as resting-state functional magnetic resonance imaging (resting state fMRI, rs-fMRI) and task-state functional magnetic resonance imaging (task-based fMRI, t-fMRI), as required. The above neuroimaging techniques can study the pathogenesis of autism from different angles, and the acquired data can provide valuable information for detection of potential neurobiomarkers to aid in early diagnosis and treatment of autism. However, it is still difficult for a doctor lacking specialized training to make a correct diagnosis by only relying on images. Therefore, computer-aided diagnosis based on neuroimaging is particularly important.
In recent years, a large number of researchers have explored the identification of autism and its potential biomarkers based on neuroimaging data using machine learning methods including deep learning, and have achieved more reasonable results. However, conventional deep learning methods have limited ability to process non-euclidean structure data, such as graph structure data. Extracting data features based on a graph structure has the advantage that, first, the graph structure has a strong information integration capability, which can integrate different types of information and correlate the information with each other. Secondly, the graph structure can be easily expanded, and the method is suitable for processing large-scale data. Furthermore, since nodes and edges in the graph structure have interdependencies, features can be extracted by considering the context information of the nodes and edges. Finally, the feature extraction method based on the graph structure can add different nodes and edges on the graph so as to extract useful features, thus having higher flexibility and interpretability. In addition, to further improve diagnostic accuracy, it is necessary to leverage complex medical multimodal data, typically including medical imaging and corresponding non-imaging data (e.g., demographic information), to extract the valid information hidden therein. In one aspect, imaging data may provide the most intuitive features for subjects in various diagnostic tasks to observe pathology and ultimately determine disease states. On the other hand, non-imaging information may reveal associations between different subjects in the population that are complementary to imaging data for disease prediction. It is therefore important to study computational models that can complementarily combine imaging data with non-imaging data to take advantage of the growing multi-modal medical data to achieve a more comprehensive characterization and more accurate disease prediction.
However, current research focuses mostly on selecting functional connections between brain regions while ignoring the effects of regional activity signals when using imaging data for autism diagnosis, and furthermore, exploring and learning from imaging and non-imaging data of a population in a unified deep learning model remains challenging.
Disclosure of Invention
The invention aims to provide a novel autism classification method based on maximum entropy weighted independent set pooling (Maximum Entropy Weighted Independent Set Pooling, MEWISPool), which effectively solves the defect that the existing method only focuses on selecting functional connections between brain areas and ignores the influence of regional activity signals, and simultaneously utilizes the functional connections of the brain areas and the regional activity signals and integrates non-imaging data to realize the classification of autism.
Aiming at the purpose, the invention provides a method for classifying autism based on maximum entropy weighted independent set pooling, which comprises the following steps:
performing brain graph downsampling by adopting maximum entropy weighted independent set pooling, including representing brain imaging as a graph structure, dividing different brain regions according to Harvard-Oxford (HO) atlas in the graph structure to define nodes, distributing time sequences representing the activity of the brain region regions for each node, then calculating local variation metrics of node characteristics, calculating node entropy to obtain a graph with node entropy values, inputting the graph into a graph neural network, adopting a conditional expectation algorithm, selecting nodes with the maximum weighted independent set loss function values, reconstructing the graph according to the pooled nodes to obtain a graph substructure, and constructing a graph substructure characteristic matrix;
extracting high-order features by using a Multi-layer Perceptron (MLP), wherein the Multi-layer Perceptron (MLP) comprises extracting features of a graph sub-structure feature matrix by using a Perceptron with two hidden layers to obtain a high-order feature matrix;
the classification of the autism is realized by adopting a graph convolution network to fuse imaging characteristics and non-imaging characteristics, including constructing a crowd graph by utilizing gender in non-imaging data and acquisition site information, distributing subject high-order imaging characteristics extracted by MLP to each corresponding node in the crowd graph, then learning node embedding on the constructed crowd graph by using a graph convolution neural network with two layers of convolution layers, and finally outputting a predictive diagnosis result of each node through a linear classifier, thereby realizing the classification of the autism.
The method comprises the following steps of a1, representing brain imaging data into graph structures, dividing the graph structures into cortical and subcortical probability graphs according to HO graphs, filtering each graph by adopting a 25% threshold, dividing the graph structures into left and right hemispheres at a midline, dividing the brain into 110 functional areas, wherein each area is provided with a digital label, so that automatic analysis can be conveniently carried out, calculating the mean value of fMRI time sequence data of each brain area in the HO graph in the ROI, forming the average time sequence of the ROI area, and converting original 4D data into a 2D data structure;
step a2, measuring the smoothness of the adjacent signals of the nodes by calculating the local variation measurement of the node characteristics, modeling the probability distribution of the node signals based on the local variation of the nodes, and further calculating the node entropy to obtain a graph with the node entropy value;
step a3, inputting the graph with the node entropy as the node characteristic into a graph neural network to generate a probability score vector Z, wherein each element represents the membership degree of the node pair maximum weighted independent set, and then applying a conditional expectation algorithm to the probability score vector of the graph and an adjacency matrix of the input graph to extract the maximum entropy weighted independent set;
and a4, constructing an adjacency matrix of the pooled nodes according to the extracted independent node subset of the input graph.
In addition, the method adopts the multi-layer perceptron to extract the high-order characteristics, and the specific implementation steps are as follows, step b1, the graph substructure characteristic matrix is input into the MLP, the MLP comprises an input layer, two layers of hidden layers and an output layer, the input layer is used for receiving input data and transmitting the input data to the hidden layers, the neurons of the hidden layers calculate and transmit the input data to the output layer, and the neurons of the output layer output the final result. Specifically, flattened features of the substructure of the graph are input to the multi-layer perceptron, and the data is trained using a back propagation algorithm, with the entire training process being supervised. In the training stage of the multi-layer perceptron, in order to avoid over fitting, ten-fold nested cross validation is performed, optimal parameters of a model are obtained by using a validation set, and then high-order features of each subject are extracted by using the optimal model.
The graph convolution network fuses imaging features and non-imaging features to realize classification of autism, and the specific implementation steps include the following steps that step c1, sex and acquisition site information of a subject are subjected to feature digitization through One-Hot coding, cosine similarity between digitization feature vectors of non-imaging information of each two subjects is calculated, similarity scores larger than 0.5 are used as connection conditions of edges between two nodes, and subject high-order features extracted by MLP are distributed to each corresponding node;
step c2, learning node embedding on the constructed crowd graph by using a graph convolution neural network with two convolution layers;
and c3, outputting a predictive diagnosis result of each node through a linear classifier by the crowd graph obtained through two-layer convolution operation, thereby realizing classification of autism.
Compared with the prior art, the invention has the remarkable advantages that: the method can simultaneously utilize brain region functional connection and region activity signals and integrate non-imaging data, and from the perspective of information theory, aims at maximizing mutual information between input nodes and pooling nodes, selects a subset with the maximum entropy nodes, and reconnects the subset, so that an input graph structure can be downsampled into a substructure, the substructure retains important node and side information in the input graph, and the flattened substructure features are input to a multi-layer perceptron to extract high-order features of each subject. The method has better performance on the classification task of the autism than other comparison methods.
Drawings
Fig. 1 is a general block diagram of the present invention.
Detailed Description
The invention discloses a classification method based on maximum entropy weighted independent set pooling, which comprises three designs of adopting MEWISPool to perform brain map downsampling, extracting high-order features and realizing classification of autism by fusing imaging features and non-imaging features:
performing brain graph downsampling by adopting maximum entropy weighted independent set pooling, including representing brain imaging data as a graph structure, dividing different brain regions according to Harvard-Oxford (HO) atlas in the graph structure to define nodes, distributing time sequences representing the activities of the brain region regions for each node, then calculating local variation metrics of node characteristics, calculating node entropy to obtain a graph with node entropy values, inputting the graph into a graph neural network, adopting a conditional expectation algorithm, selecting nodes with maximum weighted independent set loss function values, reconstructing the graph according to the pooled nodes to obtain a graph substructure, and constructing a graph substructure characteristic matrix;
extracting high-order features by adopting a multi-layer perceptron, wherein the extracting of features of the feature matrix of the graph substructure by using the perceptron with two hidden layers is carried out to obtain a high-order feature matrix;
the classification of the autism is realized by adopting a graph convolution network to fuse imaging characteristics and non-imaging characteristics, including constructing a crowd graph by utilizing gender in non-imaging data and acquisition site information, distributing subject high-order imaging characteristics extracted by MLP to each corresponding node in the crowd graph, then learning node embedding on the constructed crowd graph by using a graph convolution neural network with two layers of convolution layers, and finally outputting a predictive diagnosis result of each node through a linear classifier, thereby realizing the classification of the autism.
The present invention is described in detail below with reference to fig. 1 for implementation.
1. The specific steps of performing brain map downsampling by adopting maximum entropy weighted independent set pooling are as follows:
step a1, brain imaging data are expressed as a graph structure, the graph structure is divided into cortical and subcortical probability graphs according to HO graphs, each graph is filtered by adopting a 25% threshold value, then the graph is divided into a left hemisphere and a right hemisphere at a midline, the brain is divided into 110 functional areas, each area is provided with a digital label, automatic analysis can be conveniently carried out, and the mean value of fMRI time sequence data of each brain area in the HO graph in the ROI is calculated, so that the average time sequence of the ROI area is formed, and the conversion of original 4D data into a 2D data structure is realized.
And a2, measuring the smoothness of the adjacent signals of the nodes by calculating the local change measurement of the node characteristics, modeling the probability distribution of the node signals based on the local change of the nodes, and further calculating the node entropy to obtain a graph with the node entropy value.
And a step a3, inputting the graph with the node entropy as the node characteristic into a graph neural network to generate a probability score vector Z, wherein each element represents the membership degree of the node to the maximum weighted independent set, and then applying a conditional expectation algorithm to the probability score vector of the graph and the adjacency matrix of the input graph to extract the maximum entropy weighted independent set.
And a4, constructing an adjacency matrix of the pooled nodes according to the extracted independent node subset of the input graph.
2. The specific steps of extracting the high-order features by adopting the multilayer perceptron are as follows:
step b1, inputting the characteristic matrix of the graph sub-structure into an MLP, wherein the MLP comprises an input layer, two hidden layers and an output layer, the input layer is used for receiving input data and transmitting the input data to the hidden layers, the neurons of the hidden layers calculate and transmit the input data to the output layer, and the neurons of the output layer output a final result. Specifically, flattened features of the substructure of the graph are input to the multi-layer perceptron, and the data is trained using a back propagation algorithm, with the entire training process being supervised. In the training stage of the multi-layer perceptron, in order to avoid over fitting, ten-fold nested cross validation is performed, optimal parameters of a model are obtained by using a validation set, and then high-order features of each subject are extracted by using the optimal model.
In this step, each hidden layer contains a linear layer, a ReLU layer, and a Dropout layer.
3. The method adopts a graph convolution network to fuse imaging and non-imaging information, and comprises the following specific implementation steps of:
step c1, performing feature digitization on gender and acquisition site information of a subject through One-Hot coding, then calculating cosine similarity between digitization feature vectors of non-imaging information of each two subjects, using a similarity score larger than 0.5 as a connection condition of edges between two nodes, and distributing subject high-order features extracted by MLP to each corresponding node.
Step c2, using a graph convolutional neural network with two convolutional layers to learn node embedding on the constructed crowd graph.
And c3, outputting a predictive diagnosis result of each node through a linear classifier by the crowd graph obtained through two-layer convolution operation, thereby realizing classification of autism.
The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments disclosed herein. It will be understood by those skilled in the art that various changes in detail may be effected therein without departing from the scope of the invention as defined by the claims appended hereto.
Claims (4)
1. An autism classification method based on maximum entropy weighted independent set pooling (Maximum Entropy Weighted Independent Set Pooling, mewisboost), characterized by:
performing brain graph downsampling by adopting maximum entropy weighted independent set pooling, including representing brain imaging data as a graph structure, dividing different brain regions according to Harvard-Oxford (HO) atlas in the graph structure to define nodes, distributing time sequences representing the activities of the brain region regions for each node, then calculating local variation metrics of node characteristics, calculating node entropy to obtain a graph with node entropy values, inputting the graph into a graph neural network, adopting a conditional expectation algorithm, selecting nodes with maximum weighted independent set loss function values, reconstructing the graph according to the pooled nodes to obtain a graph substructure, and constructing a graph substructure characteristic matrix;
extracting high-order features by using a Multi-layer Perceptron (MLP), wherein the Multi-layer Perceptron (MLP) comprises extracting features of a graph sub-structure feature matrix by using a Perceptron with two hidden layers to obtain a high-order feature matrix;
the classification of the autism is realized by adopting a graph convolution network to fuse imaging characteristics and non-imaging characteristics, including constructing a crowd graph by utilizing gender in non-imaging data and acquisition site information, distributing subject high-order imaging characteristics extracted by MLP to each corresponding node in the crowd graph, then learning node embedding on the constructed crowd graph by using a graph convolution neural network with two layers of convolution layers, and outputting a predictive diagnosis result of each node through a linear classifier, thereby realizing the classification of the autism.
2. The method for classifying autism based on maximum entropy weighted independent set pooling according to claim 1, wherein: the method adopts the maximum entropy weighted independent set pooling to carry out brain map downsampling, and comprises the following specific implementation steps,
step a1, brain imaging data are expressed as graph structures, the graph structures are divided into cortical and subcortical probability graphs according to HO graphs, each graph is filtered by adopting a 25% threshold value, then the graph structures are divided into left and right hemispheres at a midline, the brain is divided into 110 functional areas, each area is provided with a digital label, automatic analysis can be conveniently carried out, and the mean value of fMRI time sequence data of each brain area in the HO graph in the ROI is calculated, so that the mean time sequence of the ROI area is formed, and the original 4D data are converted into a 2D data structure;
step a2, measuring the smoothness of the adjacent signals of the nodes by calculating the local variation measurement of the node characteristics, modeling the probability distribution of the node signals based on the local variation of the nodes, and further calculating the node entropy to obtain a graph with the node entropy value;
step a3, inputting the graph with the node entropy as the node characteristic into a graph neural network to generate a probability score vector Z, wherein each element represents the membership degree of the node pair maximum weighted independent set, and then applying a conditional expectation algorithm to the probability score vector of the graph and an adjacency matrix of the input graph to extract the maximum entropy weighted independent set;
and a4, constructing an adjacency matrix of the pooled nodes according to the extracted independent node subset of the input graph.
3. The method for classifying autism based on maximum entropy weighted independent set pooling according to claim 2, wherein: the method adopts a multi-layer perceptron to extract high-order characteristics, and comprises the following specific implementation steps,
step b1, inputting the characteristic matrix of the graph sub-structure into an MLP, wherein the MLP comprises an input layer, two hidden layers and an output layer, the input layer is used for receiving input data and transmitting the input data to the hidden layers, the neurons of the hidden layers calculate and transmit the input data to the output layer, and the neurons of the output layer output a final result. Specifically, flattened features of the substructure of the graph are input to the multi-layer perceptron, and the data is trained using a back propagation algorithm, with the entire training process being supervised. In the training stage of the multi-layer perceptron, in order to avoid over fitting, ten-fold nested cross validation is performed, optimal parameters of a model are obtained by using a validation set, and then high-order features of each subject are extracted by using the optimal model.
4. A method of classifying autism based on maximum entropy weighted independent set pooling according to claim 3, wherein: the method adopts a graph convolution network to fuse imaging and non-imaging information to realize classification of autism, comprises the following specific implementation steps,
step c1, carrying out feature digitization on gender and acquisition site information of a subject through One-Hot coding, then calculating cosine similarity between digitization feature vectors of non-imaging information of each two subjects, using similarity scores larger than 0.5 as a connection condition of edges between two nodes, and distributing subject high-order features extracted by MLP to each corresponding node;
step c2, learning node embedding on the constructed crowd graph by using a graph convolution neural network with two convolution layers;
and c3, outputting a predictive diagnosis result of each node through a linear classifier by the crowd graph obtained through two-layer convolution operation, thereby realizing classification of autism.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311859339.2A CN117809108A (en) | 2023-12-30 | 2023-12-30 | Autism classification method based on maximum entropy weighted independent set pooling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311859339.2A CN117809108A (en) | 2023-12-30 | 2023-12-30 | Autism classification method based on maximum entropy weighted independent set pooling |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117809108A true CN117809108A (en) | 2024-04-02 |
Family
ID=90434470
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311859339.2A Pending CN117809108A (en) | 2023-12-30 | 2023-12-30 | Autism classification method based on maximum entropy weighted independent set pooling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117809108A (en) |
-
2023
- 2023-12-30 CN CN202311859339.2A patent/CN117809108A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113040715B (en) | Human brain function network classification method based on convolutional neural network | |
| WO2023077603A1 (en) | Prediction system, method and apparatus for abnormal brain connectivity, and readable storage medium | |
| CN109376751B (en) | Human brain function network classification method based on convolutional neural network | |
| Hedayati et al. | Deep feature extraction method based on ensemble of convolutional auto encoders: Application to Alzheimer’s disease diagnosis | |
| Yang et al. | A deep neural network study of the ABIDE repository on autism spectrum classification | |
| CN110598793B (en) | Brain function network feature classification method | |
| CN112674720B (en) | Alzheimer disease pre-judgment method based on 3D convolutional neural network | |
| CN113693563B (en) | Brain function network classification method based on hypergraph attention network | |
| CN117218453B (en) | Incomplete multi-mode medical image learning method | |
| CN113951883B (en) | Gender difference detection method based on electroencephalogram signal emotion recognition | |
| Song et al. | Classification of mild cognitive impairment based on a combined high-order network and graph convolutional network | |
| CN115272295A (en) | Dynamic brain function network analysis method and system based on time domain-space domain combined state | |
| Zuo et al. | Prior-guided adversarial learning with hypergraph for predicting abnormal connections in Alzheimer’s disease | |
| CN115496953A (en) | Brain network classification method based on space-time graph convolution | |
| Alharthi et al. | Do it the transformer way: a comprehensive review of brain and vision transformers for autism spectrum disorder diagnosis and classification | |
| Zuo et al. | A prior guided adversarial representation learning and hypergraph perceptual network for predicting abnormal connections of Alzheimer's disease | |
| Meng et al. | Research on early diagnosis of Alzheimer's disease based on dual fusion cluster graph convolutional network | |
| Sasikala | A CAD system design using iteratively reweighted fuzzy c-means and deep tree training for Alzheimer's disease diagnosis | |
| Pallawi et al. | Study of Alzheimer’s disease brain impairment and methods for its early diagnosis: a comprehensive survey | |
| Subah et al. | A comprehensive study on atlas-based classification of autism spectrum disorder using functional connectivity features from resting-state functional magnetic resonance imaging | |
| Zong et al. | A new brain network construction paradigm for brain disorder via diffusion-based graph contrastive learning | |
| CN113723485B (en) | Hypergraph processing method for brain image of mild hepatic encephalopathy | |
| CN117809108A (en) | Autism classification method based on maximum entropy weighted independent set pooling | |
| CN114983341A (en) | Multi-modal feature fusion based multi-classification prediction system for Alzheimer's disease | |
| Vinutha et al. | A convolution neural network based classifier for diagnosis of Alzheimer’s disease |
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 |