CN113450921A - Brain development data analysis method, system, equipment and storage medium - Google Patents

Brain development data analysis method, system, equipment and storage medium Download PDF

Info

Publication number
CN113450921A
CN113450921A CN202110706290.1A CN202110706290A CN113450921A CN 113450921 A CN113450921 A CN 113450921A CN 202110706290 A CN202110706290 A CN 202110706290A CN 113450921 A CN113450921 A CN 113450921A
Authority
CN
China
Prior art keywords
graph
layer
coding model
brain development
sample
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
Application number
CN202110706290.1A
Other languages
Chinese (zh)
Inventor
乔琛
胡鑫钰
许发明
刘岳晨
任鑫
黄崎
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN202110706290.1A priority Critical patent/CN113450921A/en
Publication of CN113450921A publication Critical patent/CN113450921A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

The invention discloses a brain development data analysis method, a system, equipment and a storage medium, wherein the method comprises the following steps: constructing a graph regular sparse depth self-coding model, wherein an implicit layer in the graph regular sparse depth self-coding model is formed by stacking N sparse self-codes with graph Laplace regularities, and a graph Laplace regularity item is added in a loss function of the graph regular sparse depth self-coding model; training and fine-tuning a regular sparse depth self-coding model of the image; the method, the system, the equipment and the storage medium can effectively solve the over-fitting problem when the high-latitude small-sample brain development data is processed.

Description

Brain development data analysis method, system, equipment and storage medium
Technical Field
The invention belongs to the field of data processing, and relates to a brain development data analysis method, a brain development data analysis system, brain development data analysis equipment and a storage medium.
Background
High-dimensional small sample data is commonly used in the biomedical field, such as genomic data, medical image data, protein data and the like, and the data has the characteristics of small sample size but huge sample characteristics, particularly brain development data. This feature presents certain challenges to the process of data processing and analysis. When the sample size to sample feature ratio is small, classical machine learning algorithms tend to fail because irrelevant and redundant features may be contained in the high dimensional data. At present, deep learning is proved to be one of the strongest tools in big data analysis, but the application of the traditional deep learning algorithm in bioinformatics is still very limited, mainly because when the sample size of data is far smaller than the feature number, the model is often trapped in overfitting, further the data processing is inaccurate, and effective information is difficult to extract from the data, therefore, a method needs to be designed aiming at the data with the characteristics of high latitude and small samples, so as to solve the overfitting problem of the existing deep learning model in processing the data of the high latitude and small samples.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a brain development data analysis method, a brain development data analysis system, a brain development data analysis device and a storage medium, wherein the method, the system, the brain development data analysis device and the storage medium can effectively solve the overfitting problem when high latitude small sample brain development data are processed.
In order to achieve the above object, the method for analyzing brain development data according to the present invention comprises:
constructing a graph regular sparse depth self-coding model, wherein an implicit layer in the graph regular sparse depth self-coding model is formed by stacking N sparse self-codes with graph Laplace regularities, and a graph Laplace regularity item is added in a loss function of the graph regular sparse depth self-coding model;
training and fine-tuning a regular sparse depth self-coding model of the image;
and analyzing brain development data by using the trained image regular sparse depth self-coding model.
The specific process of training the regular sparse depth self-coding model of the graph is as follows:
collecting the recorded brain development data;
summarizing each individual and the corresponding data characteristics and the change value thereof into a unit data in the acquired brain development data, and constructing a data matrix by using the unit data corresponding to each individual;
dividing the data matrix into a training set and a testing set;
and training the regular sparse depth self-coding model of the graph by using the training set and the test set.
The hidden layer in the graph regular sparse depth self-coding model is formed by stacking six sparse self-codes with Laplace regular.
The graph laplacian regularization term is:
Figure BDA0003131413820000021
wherein the data set is X ═ { X ═ X1,x2,L,xN}，xp∈Rm，xpFor the p-th sample, N is the sample size, m is the characteristic dimension of the sample, phipq＝exp(-||xp-xq||2/σ) represents the weight of the connected edge, σ is the width of the gaussian kernel, H ═ H1,h2,L hN]T,hp∈Rm′For each sample xpCorresponding to the low-dimensional representation, Tr (-) represents the transpose of the matrix, D is the symmetric matrix,
Figure BDA0003131413820000022
Figure BDA0003131413820000023
in order to be a laplacian matrix,
Figure BDA0003131413820000024
the optimization target of the graph regular sparse depth self-coding model in the training process is as follows:
Figure BDA0003131413820000031
wherein X ═ { X ═ X1,x2,L,xN},xp∈Rm，xpIs the p sample, N is the sample size, m is the characteristic dimension of the sample, the number of neurons in the l layer is Nl,l＝1,2,3,
Figure BDA0003131413820000033
Represents the activation value of the ith neuron of the p-th sample of the output layer,
Figure BDA0003131413820000034
representing a connection weight matrix between the l layer and the l +1 layer; KL (-) represents KL divergence, rho is the sparse parameter of KL divergence term,
Figure BDA0003131413820000035
representing the average value of the activation values of the N training samples in the jth hidden layer unit;
Figure BDA0003131413820000036
wherein each one
Figure BDA0003131413820000037
Represents the activation value of the p sample of the l layer;
Figure BDA0003131413820000038
is the Laplace regularization term of the graph, phipq＝exp(-||xp-xq||2σ) σ is the nuclear width, λ1,λ2,λ3The weighting penalty parameters are KL, weight attenuation and graph regularization respectively.
The loss function of the graph canonical sparse depth self-coding model after the graph Laplace regularization term is added is as follows:
Figure BDA0003131413820000039
wherein X ═ { X ═ X1,x2,L,xN},xp∈Rm，xpIs the p sample, N is the sample size, m represents the characteristic dimension of the sample, the number of neurons in the l layer is Nl,l＝1,2,3；
Figure BDA00031314138200000311
Represents the activation value of the ith neuron of the p-th sample of the output layer,
Figure BDA00031314138200000312
represents a connection weight matrix between the l layer and the l +1 layer, KL (-) represents KL divergence, rho is a sparse parameter of KL divergence,
Figure BDA00031314138200000313
averaging the activation values of the N training samples in the jth hidden layer unit;
Figure BDA0003131413820000041
each one of which is
Figure BDA0003131413820000042
Represents the activation value of the p sample of the l layer;
Figure BDA0003131413820000043
is the Laplace regularization term of the graph, phipq＝exp(-||xp-xq||2σ) σ is the nuclear width; lambda [ alpha ]1,λ2,λ3The weighting penalty parameters are KL, weight attenuation and graph regularization respectively.
The loss function of the graph regular sparse depth self-coding model in the fine tuning process is as follows:
Figure BDA0003131413820000044
wherein the content of the first and second substances,
Figure BDA0003131413820000047
denotes the L-th layer xpjThe result of the reconstruction of (a) is,
Figure BDA0003131413820000048
φpq＝exp(-||xp-xq||2/σ) and σ denotes the gaussian kernel width,
Figure BDA0003131413820000049
represents the response value of the jth neuron of the p-th sample in the l-th layer network,
Figure BDA00031314138200000410
denotes z(l)＝a(l-1)·W(l-1)The value at (p, j). f is any differentiable activation function,
Figure BDA00031314138200000411
represents the average activation value, beta, of the jth neuron at the l layer1,β2,β3Penalty parameters that are regular terms.
A brain development data analysis system, comprising:
the device comprises a construction module and a prediction module, wherein the construction module is used for constructing a graph regular sparse depth self-coding model, an implicit layer in the graph regular sparse depth self-coding model is formed by stacking N sparse self-codes with graph Laplace regularities, and a graph Laplace regularity item is added in a loss function of the graph regular sparse depth self-coding model;
the training and fine-tuning module is used for training and fine-tuning the regular sparse depth self-coding model of the image;
and the analysis module is used for analyzing the brain development data by utilizing the trained image regular sparse depth self-coding model.
A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the brain development data analysis method when executing the computer program.
A computer-readable storage medium, storing a computer program which, when executed by a processor, implements the steps of the brain development data analysis method.
The invention has the following beneficial effects:
the brain development data analysis method, the brain development data analysis system, the brain development data analysis equipment and the storage medium introduce the graph regular sparse depth self-coding model into the brain development data analysis during specific operation, meanwhile, a hidden layer in the graph regular sparse depth self-coding model is formed by stacking six sparse self-codes with Laplace regularization, and a graph Laplace regularization item is added into a loss function of the graph regular sparse depth self-coding model so as to consider the prior knowledge of an inherent structure in data, improve the learning capacity and the expression capacity of unsupervised sparse depth self-coding, and effectively solve the over-fitting problem occurring when high-latitude small sample brain development data is processed.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the invention and together with the description serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a schematic structural view of the present invention;
FIG. 2a is a diagram illustrating a significantly different dynamic functional link selection;
FIG. 2b is a diagram illustrating a classification comparison of the deep self-coding model;
FIG. 3 is a schematic illustration of the elbow rule;
FIG. 4a is a distribution plot of the 264 ROIs and their corresponding selected functional connections in the 13 RSNs at State 1;
FIG. 4b is a distribution plot of the 264 ROIs and their corresponding selected functional connections in the 13 RSNs at State 2;
FIG. 4c is a distribution plot of the 264 ROIs and their corresponding selected functional connections in the 13 RSNs at State 3;
FIG. 4d is a distribution plot of the 264 ROIs and their corresponding selected functional connections in the 13 RSNs at State 4;
FIG. 4e is a graph of FT mean distribution for children and adolescents for four states;
FIG. 4f is a DT mean distribution plot for children and adolescents in four states;
FIG. 4g is a graph of the statistical difference between FT and DT in four states for two sets of samples;
FIG. 5 is a diagram illustrating a distribution of selected functional connections in the fifth embodiment.
Detailed Description
The present invention will be described in detail below with reference to the embodiments with reference to the attached drawings. It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict.
The following detailed description is exemplary in nature and is intended to provide further details of the invention. Unless otherwise defined, all technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention.
Example one
Referring to fig. 1, the brain development data analysis method according to the present invention includes the following steps:
1) collecting the recorded brain development data;
2) in brain development data, summarizing data characteristics corresponding to each individual and variation values thereof into a unit of data to form a structured data matrix, wherein the data matrix comprises a sample volume N and a sample characteristic p, and the sample volume N is smaller than the characteristic p, namely N < < p;
3) dividing brain development data forming a structured data matrix into a training set and a testing set;
the traditional deep learning algorithm comprises a Pre-training process (Pre-training) and a Fine-tuning process (Fine-tuning), and for the Fine-tuning process, the invention utilizes a gradient calculation formula of a loss function with a graph Laplacian regularization term for parameter updating in the training process, and performs optimization operation on the data through a sparse depth self-coding model with graph Laplacian regularization, wherein the specific optimization process is as follows:
31) setting a deep learning basic framework, and establishing a data model comprising an input layer, six hidden layers and the input layer according to data characteristics for data in a training set, wherein the input layer comprises a plurality of nodes with data characteristics, namely the number of the nodes in the input layer is a sample characteristic p in the training set, the output layer is reconstructed input data, and each hidden layer comprises a plurality of nodes with mapping corresponding relations with the input value of the previous layer.
32) For each node of each layer of Sparse self-coding, a mathematical equation is adopted, a data model of the node is established, relevant parameter values in the mathematical equation are preset, in the deep network, 6 graph Sparse self-encodings (GSAE) are selected for stacking, namely the number of hidden layers is 5, and the number of neurons in each layer is 15000, 7000,6700, 7000 and 15000 respectively. The number of nodes of the input layer and the output layer is the training set sample characteristic and is marked as p. According to the above setting, the stacked deep neural network model of the present invention includes 8 neural layers, which are an input layer, 6 hidden layers and an output layer.
33) In the network training pre-training process, for each GSAE, setting a weight attenuation parameter, a KL divergence regularization term and a graph Laplace regularization term thereof to be 0.03, 0.1 and 0.01 respectively, wherein the kernel width sigma of the graph Laplace regularization term is 0.5; in the fine adjustment process, the weight attenuation parameter, the KL divergence regularization term and the graph Laplace regularization term are set to be 0.01, 0.3 and 0.01 respectively. In addition, ρ is 0.1, sigmoid is selected in model training as an activation function f of a network layer, and in a fine tuning process, a small batch (minibatch sampling) sampling strategy and a small batch random gradient descent (SGD with mini-batch) method are adopted for parameter updating.
34) Initializing parameter values Ai of each layer of network, including weight Wi connected with the network and corresponding bias bi, inputting training set data into an input layer for learning the network structure, comparing output values of each node in an output layer with corresponding original input data, repeatedly correcting parameter values of each layer of neural network, and sequentially circulating to obtain parameter values of each layer of neural network node corresponding to the output value which enables each node of the output layer to have the highest similarity with the original input data.
35) And inputting the test set data into the trained deep neural network model for determining the effectiveness of the selected dynamic functional connections with the differences, thereby further carrying out corresponding brain development analysis and research.
It should be noted that the mathematical equation is a parametric mathematical equation, and may be a linear model or a neuron model, for example, a sigmoid activation function or a convolution operation model, and the mathematical model is set in the following manner:
y＝g(X)＝fn(fn-1(fn-2(...f1(X))
wherein y is the reconstructed training set data, the dimensionality is p, X is the training set data, the characteristic dimensionality is p, f1To fnFor each set layer of operational equations, each layer of equations fiFor example, from the first layer input layer to the first hidden layer, the training data with the dimension p is converted into the output with the dimension 200, and the analogy is performed according to the number of neurons in each layer, wherein each layer of model f isiThere is a matching set of parameters Ai, including the connection weights Wi and the offsets bi of the various layers.
In order to simultaneously consider the prior knowledge of a sparse mode and an inherent structure in data and improve the learning capacity and the expression capacity of unsupervised deep sparse self-coding, the invention introduces a graph Laplace regularization term, and the specific form is
Figure BDA0003131413820000091
Wherein phi ispq＝exp(-||xp-xq||2σ), σ represents the width of the gaussian kernel, and the objective function after introducing the graph laplace regularization term is:
Figure BDA0003131413820000092
p is the sparse parameter of the KL divergence term,
Figure BDA0003131413820000094
is the average value of the activation values of the N training samples in the jth hidden layer unit, lambda1,λ2,λ3The weight penalty parameters of KL, weight attenuation and graph regularization are respectively used, the pre-training of GSAE is completed by using the functions, then the pre-training is stacked into GSDAE, the training is completed by fine tuning, and the loss function of the GSDAE in the fine tuning process is as follows:
Figure BDA0003131413820000095
β1,β2,β3penalty parameters that are regular terms.
Example two
The brain development data analysis system of the present invention comprises:
the device comprises a construction module and a prediction module, wherein the construction module is used for constructing a graph regular sparse depth self-coding model, an implicit layer in the graph regular sparse depth self-coding model is formed by stacking N sparse self-codes with graph Laplace regularities, and a graph Laplace regularity item is added in a loss function of the graph regular sparse depth self-coding model;
the training and fine-tuning module is used for training and fine-tuning the regular sparse depth self-coding model of the image;
and the analysis module is used for analyzing the brain development data by utilizing the trained image regular sparse depth self-coding model.
EXAMPLE III
A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the brain development data analysis method when executing the computer program.
Example four
A computer-readable storage medium, storing a computer program which, when executed by a processor, implements the steps of the brain development data analysis method.
EXAMPLE five
MRI has been widely used for early detection, diagnosis and disease treatment as an important medical imaging mode, and such biomedical data generally has the characteristics of high-dimensional small samples, i.e. has a large number of features and a low sample number, and may have an overfitting problem when a deep learning method is applied to perform corresponding brain structure and brain function abnormality on such data. In order to verify that the algorithm proposed by us can effectively avoid overfitting, the valid features are selected. We used philiadelphia Neuro-graded cowort (PNC) data from large scale experimental data from cooperative studies of brain behaviour by pennsylvania university and Philadelphia children hospital, including fMRI data for 878 adolescents aged 8-22, with a standard pre-treatment process SPM 12.
The invention utilizes the regular sparse depth self-coding of the image to select the functional connection with significant difference in the brain development process. The data subsets of the complete data set were selected according to age (in months), wherein subjects with an age above 216 months belong to the first category and subjects with an age below 144 months belong to the second category, and the pearson correlation coefficient between them was first calculated from the regional average oxygenation level correlation (BOLD) signals of 264 brain regions of interest (ROIs). Thus, the input data has a dimension of 34716 and a total sample size of 397. And verifying the performance of the network by using the classification capability of the selected function connection, and then performing biological information analysis after connecting the selected function connection.
From the two groups of samples (children and adolescents), 70% of the subjects were selected as training set and the remaining 30% were selected as test set. Training data is utilized to realize network architecture learning of GSDAE, further intrinsic differences of the two groups of dFC are determined, test data is used for determining effectiveness of dFC with the differences, frequency statistics is carried out on the original 34716 connections according to reconstruction results of the training data, and the high-frequency connections are selected as dFC with significant differences in the two groups of samples. The whole selection process is performed in an unsupervised fashion. To avoid missing important connections, the above process was repeated 10 times. Furthermore, to test the effective performance of the GSDAE model in sparsity and feature selection, we compared the model to several other deep layers, including DAE, depth self-coding model with weight attenuation and KL regularization (sDAE), depth self-coding with Dropout (DP-DAE), depth self-coding with DropConnect (DPC-DAE), depth self-coding with pruning (Pr-DAE), and singular value decomposition based (SVD-DAE). The above models all share parameters with the GSDAE. All the above models and GSDAE were repeated 10 times on the test data, and the results are shown in fig. 2a and fig. 2b, and it can be seen from fig. 2a and fig. 2b that GSDAE can accurately identify the significant difference between the two groups dFC. Finally, 10 unsupervised dFC differences selected according to GSDAE are considered, the combination of the differences and the occurrence frequency of each difference in 10 results are considered, and the most significant feature is determined according to the average classification precision of the differences, the relation between the selected feature and the average classification precision of the test data is shown in figure 3, 1729 features are finally reserved as the most significant dynamic functional connection in the development process in 34716 functional connections, and the feature dimension of all samples is finally reduced to 1729, namely only dFC with the most significant sample is reserved.
Analysis of biological information contained in features
1. Dynamic functional ligation state analysis and its time-variability
First, using k-means clustering to identify different functional connection patterns, the optimal FC state number of two groups is determined to be 4 by elbow rule based on square error (the basis for state number determination is shown in FIG. 3)
For 4 different states, the difference between the significant groups of the functional network in the different states is studied respectively, and in order to further quantify the importance of the four states, the time occupation of the four states of the subject is studied by selecting three indexes of NS, FT and DT. The results are shown in fig. 4a to 4g, where a in fig. 4a to 4g represents the different distribution of 1729 functional links selected from the 264 ROIs and their corresponding 13 RSNs in four FC states, the dark line in each subgraph represents the functional links that increase during development, and the remaining lines represent the functional links that decrease during development. In section B, a represents the FT mean distribution of children and adolescents in four states. b represents the mean distribution of DT for the two groups of samples in the four states, and c represents the statistical difference in FT and DT for the two groups of samples in the four states.
2. Connectino omics difference analysis
Of the 1729 functional connections selected with significant developmental differences, 1160 functional connections were located in or between 12 RSNs with brain-specific functions, including a significant increase in age of 162 years, and 998 connections were significantly weakened with increasing age, and fig. 5 depicts the distribution of functional connections between and within 12 RSNs with brain-specific functions, where the lower half represents the reduced functional connections, the upper half represents the increased functional connections, the color of the color block represents the strength of the functional connections, and the deeper the color block, the stronger the functional connections.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.

Claims (10)

1. A method of analyzing brain development data, comprising:
constructing a graph regular sparse depth self-coding model, wherein an implicit layer in the graph regular sparse depth self-coding model is formed by stacking N sparse self-codes with graph Laplace regularities, and a graph Laplace regularity item is added in a loss function of the graph regular sparse depth self-coding model;
training and fine-tuning a regular sparse depth self-coding model of the image;
and analyzing brain development data by using the trained image regular sparse depth self-coding model.
2. The brain development data analysis method according to claim 1, wherein the specific process of training the graph canonical sparse depth self-coding model is as follows:
collecting the recorded brain development data;
summarizing each individual and the corresponding data characteristics and the change value thereof into a unit data in the acquired brain development data, and constructing a data matrix by using the unit data corresponding to each individual;
dividing the data matrix into a training set and a testing set;
and training the regular sparse depth self-coding model of the graph by using the training set and the test set.
3. The brain development data analysis method of claim 1, wherein the hidden layer in the graph canonical sparse depth self-coding model is stacked by six sparse self-codes with laplacian canonical.
4. The brain development data analysis method of claim 1, wherein the graph laplacian regularization term is:
Figure FDA0003131413810000011
wherein the data set is X ═ { X ═ X1,x2,L,xN}，xp∈Rm，xpFor the p-th sample, N is the sample size, m is the characteristic dimension of the sample, phipq＝exp(-||xp-xq||2/σ) represents the weight of the connected edge, σ is the width of the gaussian kernel, H ═ H1,h2,L hN]T,hp∈Rm′For each sample xpCorresponding to the low-dimensional representation, Tr (-) represents the transpose of the matrix, D is the symmetric matrix,
Figure FDA0003131413810000021
Figure FDA0003131413810000022
in order to be a laplacian matrix,
Figure FDA0003131413810000023
5. the brain development data analysis method according to claim 1, wherein the optimization goal of the graph canonical sparse depth self-coding model in the training process is:
Figure FDA0003131413810000024
wherein X ═ { X ═ X1,x2,L,xN},xp∈Rm，xpIs the p sample, N is the sample size, m is the characteristic dimension of the sample, the number of neurons in the l layer is Nl,l＝1,2,3,
Figure FDA0003131413810000025
Represents the activation value of the ith neuron of the p-th sample of the output layer,
Figure FDA0003131413810000026
representing a connection weight matrix between the l layer and the l +1 layer; KL (-) represents KL divergence, rho is the sparse parameter of KL divergence term,
Figure FDA0003131413810000027
representing the average value of the activation values of the N training samples in the jth hidden layer unit;
Figure FDA0003131413810000028
wherein each one
Figure FDA0003131413810000029
Represents the activation value of the p sample of the l layer;
Figure FDA00031314138100000210
is the Laplace regularization term of the graph, phipq＝exp(-||xp-xq||2σ) σ isWidth of nucleus, λ1,λ2,λ3The weighting penalty parameters are KL, weight attenuation and graph regularization respectively.
6. The brain development data analysis method according to claim 1, wherein the loss function of the graph regularized sparse depth self-coding model after adding the graph laplacian regularization term is:
Figure FDA00031314138100000211
wherein X ═ { X ═ X1,x2,L,xN},xp∈Rm，xpIs the p sample, N is the sample size, m represents the characteristic dimension of the sample, the number of neurons in the l layer is Nl,l＝1,2,3；
Figure FDA00031314138100000212
Represents the activation value of the ith neuron of the p-th sample of the output layer,
Figure FDA0003131413810000031
represents a connection weight matrix between the l layer and the l +1 layer, KL (-) represents KL divergence, rho is a sparse parameter of KL divergence,
Figure FDA0003131413810000032
averaging the activation values of the N training samples in the jth hidden layer unit;
Figure FDA0003131413810000033
each one of which is
Figure FDA0003131413810000034
Represents the activation value of the p sample of the l layer;
Figure FDA0003131413810000035
is the Laplace regularization term of the graph, phipq＝exp(-||xp-xq||2σ) σ is the nuclear width; lambda [ alpha ]1,λ2,λ3The weighting penalty parameters are KL, weight attenuation and graph regularization respectively.
7. The brain development data analysis method according to claim 1, wherein the loss function of the graph canonical sparse depth self-coding model in the fine tuning process is:
Figure FDA0003131413810000036
wherein the content of the first and second substances,
Figure FDA0003131413810000037
denotes the L-th layer xpjThe result of the reconstruction of (a) is,
Figure FDA0003131413810000038
φpq＝exp(-‖xp-xq‖2/σ) and σ denotes the gaussian kernel width,
Figure FDA0003131413810000039
represents the response value of the jth neuron of the p-th sample in the l-th layer network,
Figure FDA00031314138100000310
denotes z(l)＝a(l-1)·W(l-1)The value at (p, j). f is any differentiable activation function,
Figure FDA00031314138100000311
represents the average activation value, beta, of the jth neuron at the l layer1,β2,β3Penalty parameters that are regular terms.
8. A brain development data analysis system, comprising:
the device comprises a construction module and a prediction module, wherein the construction module is used for constructing a graph regular sparse depth self-coding model, an implicit layer in the graph regular sparse depth self-coding model is formed by stacking N sparse self-codes with graph Laplace regularities, and a graph Laplace regularity item is added in a loss function of the graph regular sparse depth self-coding model;
the training and fine-tuning module is used for training and fine-tuning the regular sparse depth self-coding model of the image;
and the analysis module is used for analyzing the brain development data by utilizing the trained image regular sparse depth self-coding model.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, carries out the steps of the brain development data analysis method according to any one of claims 1 to 7.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method for analyzing brain development data according to any one of claims 1 to 7.
CN202110706290.1A 2021-06-24 2021-06-24 Brain development data analysis method, system, equipment and storage medium Pending CN113450921A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110706290.1A CN113450921A (en) 2021-06-24 2021-06-24 Brain development data analysis method, system, equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110706290.1A CN113450921A (en) 2021-06-24 2021-06-24 Brain development data analysis method, system, equipment and storage medium

Publications (1)

Publication Number Publication Date
CN113450921A true CN113450921A (en) 2021-09-28

Family

ID=77812465

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110706290.1A Pending CN113450921A (en) 2021-06-24 2021-06-24 Brain development data analysis method, system, equipment and storage medium

Country Status (1)

Country Link
CN (1) CN113450921A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170213000A1 (en) * 2016-01-25 2017-07-27 Shenzhen University Metabolic mass spectrometry screening method for diseases based on deep learning and the system thereof
CN107145836A (en) * 2017-04-13 2017-09-08 西安电子科技大学 Hyperspectral image classification method based on stack boundary discrimination self-encoding encoder
CN111916204A (en) * 2020-07-08 2020-11-10 西安交通大学 Brain disease data evaluation method based on self-adaptive sparse deep neural network

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170213000A1 (en) * 2016-01-25 2017-07-27 Shenzhen University Metabolic mass spectrometry screening method for diseases based on deep learning and the system thereof
CN107145836A (en) * 2017-04-13 2017-09-08 西安电子科技大学 Hyperspectral image classification method based on stack boundary discrimination self-encoding encoder
CN111916204A (en) * 2020-07-08 2020-11-10 西安交通大学 Brain disease data evaluation method based on self-adaptive sparse deep neural network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
CHEN QIAO等: "A deep autoencoder with sparse and graph Laplacian regularization for characterizing dynamic functional connectivity during brain development", 《NEUROCOMPUTING》 *
胡聪;吴小俊;舒振球;陈素根;: "拉普拉斯阶梯网络", 软件学报 *

Similar Documents

Publication Publication Date Title
CN110335261B (en) CT lymph node detection system based on space-time circulation attention mechanism
Nan et al. Application of improved som network in gene data cluster analysis
Wu et al. Complex system fault diagnosis based on a fuzzy robust wavelet support vector classifier and an adaptive Gaussian particle swarm optimization
CN109389171B (en) Medical image classification method based on multi-granularity convolution noise reduction automatic encoder technology
CN111128380A (en) Method and system for constructing chronic disease health management model for simulating doctor diagnosis and accurate intervention strategy
CN111009324A (en) Mild cognitive impairment auxiliary diagnosis system and method based on brain network multi-feature analysis
Ding et al. Tensor sliced inverse regression
CN111513717A (en) Method for extracting brain functional state
Sorić et al. Using convolutional neural network for chest X-ray image classification
CN115474939A (en) Autism spectrum disorder recognition model based on deep expansion neural network
Qiao et al. Log-sum enhanced sparse deep neural network
Wang et al. A novel restricted Boltzmann machine training algorithm with fast Gibbs sampling policy
CN111916204A (en) Brain disease data evaluation method based on self-adaptive sparse deep neural network
Sun et al. A fuzzy brain emotional learning classifier design and application in medical diagnosis
CN114037014A (en) Reference network clustering method based on graph self-encoder
Das et al. Missing value imputation–A review
CN113450921A (en) Brain development data analysis method, system, equipment and storage medium
CN116580848A (en) Multi-head attention mechanism-based method for analyzing multiple groups of chemical data of cancers
CN115661498A (en) Self-optimization single cell clustering method
Ragab et al. Mathematical Modelling of Quantum Kernel Method for Biomedical Data Analysis.
Zhou et al. Kernel principal components based cascade forest towards disease identification with human microbiota
Lin et al. A neuronal morphology classification approach based on deep residual neural networks
CN111738298A (en) Data classification method based on depth-width-variable multi-core learning
Wu et al. An RBF-LVQPNN model and its application to time-varying signal classification
CN114862834B (en) Resting state functional magnetic resonance image data classification method

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