CN110298364A - Based on the feature selection approach of multitask under multi-threshold towards functional brain network - Google Patents
Based on the feature selection approach of multitask under multi-threshold towards functional brain network Download PDFInfo
- Publication number
- CN110298364A CN110298364A CN201910591933.5A CN201910591933A CN110298364A CN 110298364 A CN110298364 A CN 110298364A CN 201910591933 A CN201910591933 A CN 201910591933A CN 110298364 A CN110298364 A CN 110298364A
- Authority
- CN
- China
- Prior art keywords
- network
- brain
- feature
- under
- threshold
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
Based on the feature selection approach of multitask under multi-threshold towards functional brain network, multi-level network characterization is extracted using multi-threshold mode, extracts the further classification processing of multi-level features in the way of multicore multi-task learning for the network after thresholding.Existing methods deficiency is overcome, and then is learnt out with more judgement index and explanatory feature.The gk-MTFS method is using the feature learning under each threshold value as a task, retain each task the structured message of network using the kernel of graph (core of building on the diagram), and using the internal association between multi-task learning exploration task, and then learn to have more judgement index and explanatory feature out.Finally verified on true cerebral disease data set, the experimental results showed that, compared to method at this stage, the method for proposition has better sort feature to cerebral disease.
Description
Technical field
The invention belongs to machine learning and medical image analysis field, and in particular to the multi-threshold towards functional brain network
Under the feature selection approach based on multitask.
Background technique
With the fast development of present biotechnology, brain image technology, such as modern Magnetic resonance imaging (magnetic
Resonance imaging, MRI) technology, including Magnetic resonance imaging (functional MRI, fMRI), it provides
A kind of mode of non-intrusion type explores human brain, the mechanism for the brain structure and function that can not be recognized before disclosing.Brain network
Analysis can portray interaction between brain brain area on hierarchical link, become in medical image analysis and neuroimaging one it is new
Research hotspot.
Recently, the method for machine learning has been used in the analysis and classification of brain network.For example, researcher utilizes brain net
Network carries out the diagnosis and classification of early stage cerebral disease, obtains good performance.In these researchs, typical way is from brain net
The local measurement (such as cluster coefficients) that brain is extracted in network is used for the classification of disease as feature.And feature selecting be then filter out it is more
Remaining and unessential feature, to improve classification performance.For example, Chen et al. uses the weight on side as feature for AD
The classification of (Alzheimer ' s disease) and MCI (mild cognitive impairment).Wee et al. is from functional brain
The classification that cluster coefficients are used for MCI as feature is extracted in network.Zanin et al. uses 16 kinds of network measures to be used for as feature
The classification of MCI and normal person.Since locality measures the feature of only network partial structurtes, in assorting process, lose
The topological structure of the globality of network, so as to will affect classification performance.
In brain network analysis, the most frequently used two kinds of feature selection approach are t-test method and Lasso method.?
In t-test method, their identification is measured each characteristic use standard t-test first, and according to identification pair
Feature is ranked up, the character subset of one group of final choice most judgement index.Existing research shows that the t- under Small Sample Size
The performance that test method usually can obtain.It is not both that the method for Lasso is by minimizing a target with t-test method
Function completes feature selecting, research shows that when a large amount of incoherent features and Lasso method is very when only a small amount of sample
Effectively.Currently, most feature selection approach cannot be directly used to the complicated structuring number of processing mainly for vector data
According to such as brain network data.
Performance of the feature selecting since classifier can not only be improved, and can help to find some pairs of diseases sensitive
Biological marker.Existing method is usually that local measurement (weight or cluster coefficients on such as side) is extracted from network data as special
Sign, and is combined into a long feature vector, for subsequent feature selecting and classification, and some useful network structure informations
(overall topology of such as network) is lost, this may be decreased final classification performance.In addition, functional brain network is general
All it is full connection weighted network, needs to pre-process using thresholding to portray the structural features of network.However, on the one hand,
None good standard goes to select specific threshold value, and on the other hand, different threshold values typically result in different network structures, this
A little structures may contain complementary information, it is possible to further promote network analysis performance.
Based on this, the present invention extracts multi-level network characterization using multi-threshold mode, for the network benefit after thresholding
The further classification processing of multi-level features is extracted with multicore multi-task learning.Existing methods deficiency is overcome, and then learns
With more judgement index and explanatory feature.The gk-MTFS method of proposition is appointed the feature learning under each threshold value as one
Business is retained the structured message of network using the kernel of graph (core of building on the diagram) to each task, and is visited using multi-task learning
Internal association between rope task, and then learn to have more judgement index and explanatory feature out.Finally in true cerebral disease number
According to being verified on collection, the experimental results showed that, compared to method at this stage, the method for proposition has preferably point cerebral disease
Class feature.
Summary of the invention
The present invention aiming at the shortcomings in the prior art, provides under the multi-threshold towards functional brain network based on multitask
Feature selection approach.The gk-MTFS method is first with L21Normal form group rarefaction item, enables feature with more identification
It is selected.The Laplacian regularization term under multi-threshold based on the kernel of graph is further used, remains functional brain network connection certainly
The topology information of body.Finally learnt using multicore characteristic binding, objective function is carried out using approximate Speed gradient algorithm
Optimization Solution.
To achieve the above object, the invention adopts the following technical scheme:
Based on the feature selection approach of multitask under multi-threshold towards functional brain network, which is characterized in that including such as
Lower step:
Step 1: being pre-processed to fMRI data, constructing function brain network；
Step 2: carrying out thresholding processing to the functional brain network of building simultaneously using R threshold value；
Step 3: the part that the cluster coefficients for extracting brain area to each thresholding network are used to measure network as feature is opened up
Flutter structure；
Step 4: utilizing the similitude of overall topology kernel of graph calculating network to each thresholding network；
Step 5: being based on step 3 and step 4, gk-MTFS feature selection approach under the multi-threshold towards brain network is established
Objective function；
Step 6: using accelerating approximate gradient algorithm to optimize the objective function of proposition.
To optimize above-mentioned technical proposal, the concrete measure taken further include:
Further, in the step 1, brain space is divided into 116 brain areas, the time series of brain area is obtained, makes
With Pearson correlation coefficient constructing function brain network, constructed brain network is the fully-connected network of weighting, for subsequent
Brain network analysis.
Further, it in the step 2, to the fully-connected network with weight constructed in step 1, utilizes simultaneously
R given threshold value converts weighting network on the network of multiple binaryzations, portrays multi-level topological structure, is used for subsequent spy
Sign is extracted and structured features selection.
Further, in the step 3, for the brain network of thresholding each in step 2, the office of each brain area is extracted
Portion's cluster coefficients are as feature, and the feature from all brain areas constitutes a feature vector together, for portraying brain network
Local topology.
Further, in the step 4, the similitude of two networks is defined using the kernel of graph, for reserved functional company
Network data topology information is connect, the whole similitude in the structure of two network datas is directly defined using the kernel of graph, i.e.,
For two brain networks under r-th of threshold valueWithIts similitude is defined as:
Wherein,Indicate two brain networks under r-th of threshold valueWithSimilitude,It is the kernel of graph of definition,
The corresponding kernel of graph is constructed using the method for Weisfeiler-Lehman subtree.
Further, it in the step 5, enablesR=1,2 ..., R, XrIt indicates
The eigenmatrix extracted under R threshold value of N number of sample in step 3,I=1,2 ... N is indicated from i-th of sample in r
The feature vector extracted under a threshold value, d are every kind of intrinsic dimensionalities；
Enable Y=[y1, y2..., yN]∈RN, the corresponding response vector of the N number of sample of Y expression, yi, i=1,2 ... N indicates sample
Class label, to two class classification problems, yi∈ {+1, -1 }, can be expressed as patient and normal person；
Based on this, propose that the objective function of gk-MTFS feature selection approach under the multi-threshold towards brain network is as follows:
Wherein, W=[w1, w2..., wR]∈Rd*NIt is a weight matrix, wrIndicate the weight under r-th of threshold value, Mr=
Cr-SrIt is Laplacian matrix, SrExpression is defined on the similar matrix under r-th of threshold value (i.e. task), is defined using formula (1)
The similitude of two networks, evenI=1,2 ... N, j=1,2 ... N, CrIt is diagonal matrix, and Indicate i-th it is diagonal on element；
The objective function includes three, and first item is loss function item, and using quadratic loss function, Section 2 is that group is sparse
Change (group-sparsity) regularization term, for selecting public feature from different task, Section 3 be Laplacian just
Then change item, for retaining the structured message of network and the distributed intelligence of network data, λ and β are for balancing between three
The constant for being greater than 0 of relative contribution.
The beneficial effects of the present invention are: exploring different threshold value (task) lower threshold values by the way of multitask for multi-threshold
The complementary information of network, is sufficiently extracted sample characteristics, remains the overall structure information of brain network data and the topology of itself
Information, and verified on two true public data collection (attention deficit hyperactivity disorder data set and senile dementia data set)
Proposed method gk-MTFS has a better classification performance.
Detailed description of the invention
Fig. 1 a to Fig. 1 c indicates that nicety of grading result is with the change of different regularization parameter λ and β value in three classification tasks
Change curve, wherein Fig. 1 a indicates 1MCI vs.eMCI classification, and Fig. 1 b indicates eMCI vs.HC classification, and Fig. 1 c indicates ADHD
Vs.HC classification.
Specific embodiment
In conjunction with the accompanying drawings, the present invention is further explained in detail.
Feature selection approach based on multitask under multi-threshold proposed by the present invention towards functional brain network, including such as
Lower step:
Step 1: being pre-processed to fMRI data, constructing function brain network；
Step 2: carrying out thresholding processing to the functional brain network of building simultaneously using R threshold value；
Step 3: the part that the cluster coefficients for extracting brain area to each thresholding network are used to measure network as feature is opened up
Flutter structure；
Step 4: utilizing the similitude of overall topology kernel of graph calculating network to each thresholding network；
Step 5: establishing the objective function of gk-MTFS feature selection approach under the multi-threshold towards brain network；
Step 6: using accelerating approximate gradient algorithm to optimize the objective function of proposition.
It enablesR=1,2 ..., R are indicated in step 3 from R threshold of N number of sample
The lower eigenmatrix extracted of value, wherein d is every kind of intrinsic dimensionality, enables Y=[y1, y2..., yN]∈RNIndicate that N number of sample is corresponding
Response vector, wherein yiThe class label for indicating sample, to two class classification problems, i.e. yi∈ {+1, -1 }, can be expressed as disease
People and normal person；Based on this, propose that the objective function of gk-MTFS feature selection approach under the multi-threshold towards brain network is as follows:
Wherein, W=[w1, w2..., wR]∈Rd*NIt is a weight matrix,It is a L2,1Model
Formula, it is therefore an objective to encourage the vector that multirow value is 0 in weight matrix W, feature corresponding to non-zero element will be selected.Using L2,1Model
Formula will make a small amount of feature from multiple threshold tasks by Combination selection.Mr=Cr-SrIt is Laplacian matrix, enables SrIt indicates
The similar matrix being defined under r-th of threshold value (i.e. task) defines the similitude of two networks using formula (1), evenI=1,2 ... N, j=1,2 ... N, CrIt is diagonal matrix, and
The objective function includes three, and first item is loss function item, uses quadratic loss function here, Section 2 is group
Rarefaction (group-sparsity) regularization term, for selecting public feature from different task, Section 3 is
Laplacian regularization term, for retaining the structured message of network and the distributed intelligence of network data.λ and β is for balancing
The constant for being greater than 0 of relative contribution between three.
In order to remain functional connection network data itself topology information, the kernel of graph is introduced, directly defines two
Similitude in the structure of network data, i.e., for two brain networks under r-th of threshold valueWithIts similitude is defined as:
Wherein, k indicates kernel function,It is the kernel of graph of definition, uses the method for Weisfeiler-Lehman subtree
To construct the corresponding kernel of graph.
According to Laplacian regular terms defined above it is found that when two samples are closely similar, after mapping as far as possible
It is close, and then the similitude of network data can be calculated by the kernel of graph.This guarantees the structuring letters for saving network
Breath, the space distribution information of network data of having withed a hook at the end entirety.
Solution for the objective function of definition is optimized using the acceleration approximate gradient being widely used at present.It is true at two
Real public data collection (i.e. ADHD (Attention Deficit Hyperactivity Disorder) data set and ADNI (the
Alzheimer ' s Disease Neuroimaging Initiative)) on the results show propose the effective of method
Property.
Technical solution of the present invention is described in further details below in conjunction with application example:
A specific embodiment of the invention lists and evaluates the effective of proposition method on two open fMRI data sets
Property.Table 1 provides the characteristic of these data sets.
Table 1: the statistical information of the sample of two datasets
MMSE=Mini-Mental State Examination
For ADHD data set, when having used preprocessed good from NYU (New York University) website
Between sequence data, detailed pre-treatment step can be in http://www.nitrc.org/plugins/mwiki/
Index.php/neurobureau:Athena is found.Pretreated data are according to AAL (Automated Anatomical
Labeling brain) is divided into 90 brain areas, each brain area is contained 172 time point datas, constructed using Pearson correlation coefficient
Functional brain network.
To ADNI data set, the standard that uses pre-processes pipeline, including timeslice (correction and the dynamic correction of head).Utilize SPM8
(Statistical Parametric Mapping software package) (http: //
Www.fil.ion.ucl.ac.uk.spm) image preprocessing is completed.For each sample, preceding 10 width fMRI image is abandoned, with true
Protect magnetization balance.The correction being sliced in an acquisition time delay is carried out first to image is retained, and is and then carried out head movement and is rectified
Just, the influence of head movement is eliminated.Since the ventricles of the brain region (ventricle) and the region white matter (white matter, WM) include
Relatively high noise extracts Blood oxygen level dependence (the blood oxygenation using grey matter (gray matter, GM)
Level dependent, BOLD) signal come constructing function connection network, in order to eliminate the possible influence of WM and CSF, each sample
This GM tissue is further used in their corresponding fMRI images of cover (mask).The first time scanning of fMRI time series is matched
Standard arrives the t1 weighted image of same sample, and estimated transformation is applied to identical sample other time sequence.After correction
FMRI image is registrated to same templatespace first with the deformable registration method of HAMMER, and using AAL template it
It is divided into 90 interested regions (region-of-interest, ROI).Finally, to each ROI, all voxels (voxel)
Upper averagely fMRI time series is by the time series as the ROI.Similarly also use Pearson correlation coefficient constructing function
Brain network.
The brain network data of original building is the weighted graph connected entirely, in order to portray topological structure at many levels, is used
Multi-threshold mode carries out thresholding to data to get the two-value network arrived after multiple thresholdings.Then for each two-value network,
According to classic map theory related algorithm, Local Clustering coefficient is extracted as feature.Finally, being held using the gk-MTFS method proposed
Row feature selecting uses multi-kernel support vector machine (multiple kernel support vector in classifying step
Machine, multi-SVM) classify.
Table 2, table 3, table 4 respectively show experimental result of the gk-MTFS method of proposition in three kinds of classification tasks, and with
Other three kinds of prevailing characteristics selection methods be not carried out the method (Baseline) of feature selecting with making comparisons.Specifically,
It is proposed that precision of the method in three classification tasks of the 1MCI with eMCI, eMCI and HC, ADHD and HC be respectively
76.5%, 76.9% and 68.0%, AUC value 0.81,0.79 and 0.70.Further demonstrate the validity of proposition method.
Institute's methodical performance when table 2:1MCI and eMCI classifies
Institute's methodical performance when table 3:eMCI vs.HC classifies
The classification performance of method therefor when table 4:ADHD vs.HC classifies
All these result verifications validity of the gk-MTFS method in feature selecting.From table 2~4, it can also see
Feature selecting is carried out (i.e. than no to the method (i.e. gk-MTFS, MMT-LASSO and MMT-LASSO) for carrying out feature selecting
Baseline method) can obtain preferable performance, show feature selecting to the significant contribution for improving cerebral disease classification performance.Separately
Outside, Fig. 1 a, Fig. 1 b, Fig. 1 c illustrate the precision variation obtained under two regularization parameter λ, β situations of change.It can be observed that
Lesser λ value indicates that the feature under more threshold values will show the classification performance of multitask feature selection approach by Combination selection
Largely influenced by λ value in three classification tasks.This means that being selected in gk-MTFS method proposed by the present invention
Best λ value is critically important, the possible reason is the sparsity solved in parameter lambda governing equation.
The above is only the preferred embodiment of the present invention, protection scope of the present invention is not limited merely to above-described embodiment,
All technical solutions belonged under thinking of the present invention all belong to the scope of protection of the present invention.It should be pointed out that for the art
For those of ordinary skill, several improvements and modifications without departing from the principles of the present invention should be regarded as protection of the invention
Range.
It should be noted that the term of such as "upper", "lower", "left", "right", "front", "rear" cited in invention, also
Only being illustrated convenient for narration, rather than to limit the scope of the invention, relativeness is altered or modified, in nothing
Under essence change technology contents, when being also considered as the enforceable scope of the present invention.
The above is only the preferred embodiment of the present invention, protection scope of the present invention is not limited merely to above-described embodiment,
All technical solutions belonged under thinking of the present invention all belong to the scope of protection of the present invention.It should be pointed out that for the art
For those of ordinary skill, several improvements and modifications without departing from the principles of the present invention should be regarded as protection of the invention
Range.
Claims (6)
1. based on the feature selection approach of multitask under the multi-threshold towards functional brain network, which is characterized in that including as follows
Step:
Step 1: being pre-processed to fMRI data, constructing function brain network；
Step 2: carrying out thresholding processing to the functional brain network of building simultaneously using R threshold value；
Step 3: extracting the cluster coefficients of brain area as feature for measuring the local topology knot of network to each thresholding network
Structure；
Step 4: utilizing the similitude of overall topology kernel of graph calculating network to each thresholding network；
Step 5: being based on step 3 and step 4, the mesh of gk-MTFS feature selection approach under the multi-threshold towards brain network is established
Scalar functions；
Step 6: using accelerating approximate gradient algorithm to optimize the objective function of proposition.
2. the feature selection approach based on multitask under the multi-threshold as described in claim 1 towards functional brain network,
It is characterized in that: in the step 1, brain space being divided into 116 brain areas, obtains the time series of brain area, use
Pearson correlation coefficient constructing function brain network, constructed brain network are the fully-connected networks of weighting.
3. the feature selection approach based on multitask under the multi-threshold as claimed in claim 2 towards functional brain network,
It is characterized in that: in the step 2, to the fully-connected network with weight constructed in step 1, and meanwhile it is given using R
Threshold value converts weighting network on the network of multiple binaryzations, for portraying multi-level topological structure.
4. the feature selection approach based on multitask under the multi-threshold as described in claim 1 towards functional brain network,
It is characterized in that: in the step 3, for the brain network of thresholding each in step 2, extracting the Local Clustering system of each brain area
Number is used as feature, and the feature from all brain areas constitutes a feature vector together, for portraying the local topology of brain network
Structure.
5. the feature selection approach based on multitask under the multi-threshold as described in claim 1 towards functional brain network,
Pair it is characterized in that: in the step 4, the whole similitude in the structure of two network datas is directly defined using the kernel of graph, i.e.,
Two brain networks under r-th of threshold valueWithIts similitude is defined as:
Wherein,Indicate two brain networks under r-th of threshold valueWithSimilitude,It is the kernel of graph of definition, uses
The method of Weisfeiler-Lehman subtree constructs the corresponding kernel of graph.
6. the feature selection approach based on multitask under the multi-threshold as claimed in claim 5 towards functional brain network,
It is characterized in that: in the step 5, enablingXrIt indicates in step 3
The eigenmatrix extracted under R threshold value of N number of sample,Expression is extracted under i-th of sample, r-th of threshold value
Feature vector, d is every kind of intrinsic dimensionality；
Enable Y=[y1, y2..., yN]∈RN, the corresponding response vector of the N number of sample of Y expression, yi, i=1,2 ... the class of N expression sample
Label, to two class classification problems, yi∈ {+1, -1 }；
Based on this, propose that the objective function of gk-MTFS feature selection approach under the multi-threshold towards brain network is as follows:
Wherein, W=[w1, w2..., wR]∈Rd*NIt is a weight matrix, wrIndicate the weight under r-th of threshold value, Mr=Cr-SrIt is
Laplacian matrix, SrIndicate the similar matrix being defined under r-th of threshold value, CrIt is diagonal matrix, and Indicate i-th it is diagonal on element；
The objective function include three, first item is loss function item, using quadratic loss function, Section 2 be group rarefaction just
Then change item, for selecting public feature from different task, Section 3 is Laplacian regularization term, for retaining network
Structured message and network data distributed intelligence, λ and β are for balancing the normal greater than 0 of the relative contribution between three
Number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910591933.5A CN110298364B (en) | 2019-06-27 | 2019-06-27 | Multi-threshold-value multi-task-based feature selection method for functional brain network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910591933.5A CN110298364B (en) | 2019-06-27 | 2019-06-27 | Multi-threshold-value multi-task-based feature selection method for functional brain network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110298364A true CN110298364A (en) | 2019-10-01 |
| CN110298364B CN110298364B (en) | 2021-06-15 |
Family
ID=68029908
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910591933.5A Active CN110298364B (en) | 2019-06-27 | 2019-06-27 | Multi-threshold-value multi-task-based feature selection method for functional brain network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110298364B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111627553A (en) * | 2020-05-26 | 2020-09-04 | 四川大学华西医院 | Method for constructing individualized prediction model of first-onset schizophrenia |
| CN113920123A (en) * | 2021-12-16 | 2022-01-11 | 中国科学院深圳先进技术研究院 | Addictive brain network analysis method and device |
| CN114190884A (en) * | 2020-09-18 | 2022-03-18 | 深圳大学 | Longitudinal analysis method, system and device for brain disease data |
| CN114376558A (en) * | 2022-03-24 | 2022-04-22 | 之江实验室 | Brain atlas individuation method and system based on magnetic resonance and twin map neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093087A (en) * | 2013-01-05 | 2013-05-08 | 电子科技大学 | Multimodal brain network feature fusion method based on multi-task learning |
| CN106650768A (en) * | 2016-09-27 | 2017-05-10 | 北京航空航天大学 | Gaussian image model-based brain network modeling and mode classification method |
| CN107944443A (en) * | 2017-11-16 | 2018-04-20 | 深圳市唯特视科技有限公司 | One kind carries out object consistency detection method based on end-to-end deep learning |
| CN108345903A (en) * | 2018-01-25 | 2018-07-31 | 中南大学湘雅二医院 | A kind of multi-modal fusion image classification method based on mode distance restraint |
| CN109065128A (en) * | 2018-09-28 | 2018-12-21 | 郑州大学 | A kind of sparse brain network establishing method of weighted graph regularization |
-
2019
- 2019-06-27 CN CN201910591933.5A patent/CN110298364B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093087A (en) * | 2013-01-05 | 2013-05-08 | 电子科技大学 | Multimodal brain network feature fusion method based on multi-task learning |
| CN106650768A (en) * | 2016-09-27 | 2017-05-10 | 北京航空航天大学 | Gaussian image model-based brain network modeling and mode classification method |
| CN107944443A (en) * | 2017-11-16 | 2018-04-20 | 深圳市唯特视科技有限公司 | One kind carries out object consistency detection method based on end-to-end deep learning |
| CN108345903A (en) * | 2018-01-25 | 2018-07-31 | 中南大学湘雅二医院 | A kind of multi-modal fusion image classification method based on mode distance restraint |
| CN109065128A (en) * | 2018-09-28 | 2018-12-21 | 郑州大学 | A kind of sparse brain network establishing method of weighted graph regularization |
Non-Patent Citations (3)
| Title |
|---|
| MI WANG: "Graph-Kernel Based Structured Feature Selection for Brain Disease Classification Using Functional Connectivity Networks", 《IEEE》 * |
| YANG LI: "Multimodal hyper-connectivity of functional networks using functionally-weighted LASSO for MCI classification", 《RESEARCHGATE》 * |
| 张道强: "基于机器学习的脑网络分析方法及应用", 《数据采集与处理》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111627553A (en) * | 2020-05-26 | 2020-09-04 | 四川大学华西医院 | Method for constructing individualized prediction model of first-onset schizophrenia |
| CN114190884A (en) * | 2020-09-18 | 2022-03-18 | 深圳大学 | Longitudinal analysis method, system and device for brain disease data |
| CN113920123A (en) * | 2021-12-16 | 2022-01-11 | 中国科学院深圳先进技术研究院 | Addictive brain network analysis method and device |
| CN114376558A (en) * | 2022-03-24 | 2022-04-22 | 之江实验室 | Brain atlas individuation method and system based on magnetic resonance and twin map neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110298364B (en) | 2021-06-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107506761B (en) | Brain image segmentation method and system based on significance learning convolutional neural network | |
| CN110223281B (en) | Pulmonary nodule image classification method when data set contains uncertain data | |
| CN107977671B (en) | Tongue picture classification method based on multitask convolutional neural network | |
| CN110298364A (en) | Based on the feature selection approach of multitask under multi-threshold towards functional brain network | |
| Xie et al. | Automatic segmentation of dermoscopy images using self-generating neural networks seeded by genetic algorithm | |
| CN113616184B (en) | Brain network modeling and individual prediction method based on multi-mode magnetic resonance image | |
| Wen et al. | A novel statistical cerebrovascular segmentation algorithm with particle swarm optimization | |
| CN107506797A (en) | One kind is based on deep neural network and multi-modal image alzheimer disease sorting technique | |
| CN108960341A (en) | A kind of structured features selection method towards brain network | |
| Lin et al. | A high resolution representation network with multi-path scale for retinal vessel segmentation | |
| CN110211671B (en) | Thresholding method based on weight distribution | |
| CN113610808A (en) | Individual brain atlas individualization method, system and equipment based on individual brain connection atlas | |
| CN107748889A (en) | A kind of Ultrasound Image of Breast Tumor automatic classification method | |
| Cui et al. | Bidirectional cross-modality unsupervised domain adaptation using generative adversarial networks for cardiac image segmentation | |
| CN109886346A (en) | A kind of cardiac muscle MRI image categorizing system | |
| Zhao et al. | Emds-6: Environmental microorganism image dataset sixth version for image denoising, segmentation, feature extraction, classification, and detection method evaluation | |
| Kumar et al. | Computer-aided automated discrimination of Alzheimer's disease and its clinical progression in magnetic resonance images using hybrid clustering and game theory-based classification strategies | |
| CN113723239B (en) | Magnetic resonance image classification method and system based on causal relationship | |
| CN110569880A (en) | Method for decoding visual stimulation by using artificial neural network model | |
| CN114937298A (en) | Micro-expression recognition method based on feature decoupling | |
| CN104751461B (en) | A kind of leukocyte cell core dividing method based on histogram thresholding and low-rank representation | |
| Meng et al. | Representation disentanglement for multi-task learning with application to fetal ultrasound | |
| CN113255734A (en) | Depression classification method based on self-supervision learning and transfer learning | |
| CN107590806A (en) | A kind of detection method and system based on brain medical imaging | |
| CN116759076A (en) | Unsupervised disease diagnosis method and system based on medical image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |