CN106650818A - Resting state function magnetic resonance image data classification method based on high-order super network - Google Patents

Resting state function magnetic resonance image data classification method based on high-order super network Download PDF

Info

Publication number
CN106650818A
CN106650818A CN201611251416.6A CN201611251416A CN106650818A CN 106650818 A CN106650818 A CN 106650818A CN 201611251416 A CN201611251416 A CN 201611251416A CN 106650818 A CN106650818 A CN 106650818A
Authority
CN
China
Prior art keywords
super
magnetic resonance
node
represent
brain area
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
Application number
CN201611251416.6A
Other languages
Chinese (zh)
Other versions
CN106650818B (en
Inventor
郭浩
曹锐
杨艳丽
邓红霞
相洁
李海芳
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.)
Taiyuan University of Technology
Original Assignee
Taiyuan University of Technology
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 Taiyuan University of Technology filed Critical Taiyuan University of Technology
Priority to CN201611251416.6A priority Critical patent/CN106650818B/en
Publication of CN106650818A publication Critical patent/CN106650818A/en
Application granted granted Critical
Publication of CN106650818B publication Critical patent/CN106650818B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

The present invention relates to the image processing technology, and concretely provides a resting state function magnetic resonance image data classification method based on a high-order super network. The problem is solved that the traditional magnetic resonance image data classification method is low in classification accuracy. The resting state function magnetic resonance image data classification method based on the high-order super network comprises the following steps: the step S1: performing preprocessing of the resting state function magnetic resonance image; the step S2: performing time window segment of the average time sequence of each brain region; the step S3: calculating the Pearson's correlation coefficients between each two average time sequences of each brain region; the step S4: extracting the values of corresponding elements in the Pearson's correlation matrix; the step S5: employing a sparse linear regression model to construct a high-order super network; the step S6: calculating the local attributes of the high-order super network; the step S7: selecting the classification features and constructing a classifier; and the step S8: performing quantification of the importance degree and the redundancy degree of the selected features. The resting state function magnetic resonance image data classification method based on the high-order super network is suitable for the classification of the magnetic resonance image data.

Description

Tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network
Technical field
The present invention relates to image processing techniques, specifically a kind of tranquillization state functional magnetic resonance imaging based on high-order super-network Data classification method.
Background technology
Human brain is an extremely complex information processing system, and in neuroscience field, an important challenge is exactly to take off Show its internal function and structure organization pattern.As the combination of multi-modal mr imaging technique and Complex Networks Theory, magnetic Resonance image data sorting technique currently has become one of the focus in brain science field.However, conventional magnetic resonance image data Thus sorting technique causes its classification accuracy due to itself principle and the limitation of feature, the restriction of generally existing methodology It is low, so as to have a strong impact on its using value.
In traditional tranquillization state functional mri analysis, it is assumed that function connects are in time static, are ignored The nervous activity that may occur within sweep time or interaction.Function connects related in time, due to nerve The dynamic change of effect, may affect the correlation intensity between brain area.Therefore, the research of dynamic function connects has important The meaning.Meanwhile, two pairwise correlations that traditional function connection network is normally based between brain zones of different build, so as to ignore Their higher order relationship, the loss of these order of information is probably important for medical diagnosis on disease.Simultaneously based on related net Network many false connections because any selected threshold has.Therefore, it is necessary to invent a kind of brand-new nuclear magnetic resonance image data Sorting technique, to solve the problems referred to above of conventional magnetic resonance image data sorting technique presence.The present invention is dividing time window base Super-network is built using sparse representation method on plinth, the feature with regard to brain region is then extracted from super-network and is examined for disease It is disconnected.This method preferably reflects the time-varying characteristics of function connects.Meanwhile, by building high-order super-network, can not draw In the case of entering too many parameter, the interaction between higher level and more complicated brain region is presented.
The content of the invention
The present invention is low in order to solve the problems, such as conventional magnetic resonance image data sorting technique classification accuracy, there is provided a kind of Tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network.
The present invention adopts the following technical scheme that realization：
Based on the tranquillization state functional magnetic resonance imaging data classification method of high-order super-network, the method is to adopt following steps Realize：
Step S1：Tranquillization state functional magnetic resonance imaging is pre-processed, and according to selected standardization brain map to pre- Tranquillization state functional magnetic resonance imaging after process carries out region segmentation, and then each brain area to being split carries out average time sequence Extraction；
Step S2：The sliding window that designated length is fixed, and the average time sequence of each brain area is entered according to a fixed step size Row time window is split；
Step S3：The average time sequence of each brain area under each time window Pearson correlation coefficient between any two is calculated, Thus Pearson relevance matrix is obtained；
Step S4：The value of corresponding element in Pearson relevance matrix is extracted, High order correletion matrix is thus obtained；
Step S5：Using sparse linear regression model, the line of each element and other elements in High order correletion matrix is calculated Property combination represent, super side is thus set up, then according to super side structure high-order super-network；
Step S6：Calculate the local attribute of high-order super-network；The local attribute includes：Each element in high-order super-network Degree and cluster coefficients；
Step S7：Using support vector cassification algorithm, the local attribute of high-order super-network is selected as characteristic of division, by This carries out the structure of grader, then the grader for building is tested using cross validation method；
Step S8：Using mutual information analysis method, the importance degree and the redundancy amount of carrying out to selected feature in grader Change, then postsearch screening is carried out to selected feature according to quantized result, thus high-order super-network is optimized.
Compared with conventional magnetic resonance image data sorting technique, the tranquillization state work(based on high-order super-network of the present invention Energy nuclear magnetic resonance image data classification method is by using time window dividing method, Pearson came correlation technique, sparse linear recurrence side Method, support vector cassification algorithm, cross validation method, mutual information analysis method, realize the description of high-order super-network, thus The time-varying characteristics of function connects are reflected well, so as to classification accuracy greatly improved (as shown in figure 1, the present invention's divides Classification accuracy of the class accuracy rate apparently higher than conventional magnetic resonance image data sorting technique), and then cause using value higher.
The present invention efficiently solves the problems, such as that conventional magnetic resonance image data sorting technique classification accuracy is low, it is adaptable to magnetic Resonance image data classification.
Description of the drawings
Fig. 1 is the contrast schematic diagram of the present invention and conventional magnetic resonance image data sorting technique.
Specific embodiment
Based on the tranquillization state functional magnetic resonance imaging data classification method of high-order super-network, the method is to adopt following steps Realize：
Step S1：Tranquillization state functional magnetic resonance imaging is pre-processed, and according to selected standardization brain map to pre- Tranquillization state functional magnetic resonance imaging after process carries out region segmentation, and then each brain area to being split carries out average time sequence Extraction；
Step S2：The sliding window that designated length is fixed, and the average time sequence of each brain area is entered according to a fixed step size Row time window is split；
Step S3：The average time sequence of each brain area under each time window Pearson correlation coefficient between any two is calculated, Thus Pearson relevance matrix is obtained；
Step S4：The value of corresponding element in Pearson relevance matrix is extracted, High order correletion matrix is thus obtained；
Step S5：Using sparse linear regression model, the line of each element and other elements in High order correletion matrix is calculated Property combination represent, super side is thus set up, then according to super side structure high-order super-network；
Step S6：Calculate the local attribute of high-order super-network；The local attribute includes：Each element in high-order super-network Degree and cluster coefficients；
Step S7：Using support vector cassification algorithm, the local attribute of high-order super-network is selected as characteristic of division, by This carries out the structure of grader, then the grader for building is tested using cross validation method；
Step S8：Using mutual information analysis method, the importance degree and the redundancy amount of carrying out to selected feature in grader Change, then postsearch screening is carried out to selected feature according to quantized result, thus high-order super-network is optimized.
In step S1, pre-treatment step is specifically included：Time horizon correction, the dynamic correction of head, joint registration, space criteria Change, low frequency filtering；Standardization brain map adopts AAL templates；The extraction step of average time sequence is specifically included：Extract AAL moulds BOLD intensity of all voxels that each brain area is included in plate in different time points, and by each voxel in different time points BOLD intensity carry out arithmetic average, thus obtain the average time sequence of each brain area.
In step S3, computing formula is specifically expressed as follows：
In formula (1)：rijRepresent the element of the i-th row jth row in Pearson relevance matrix, i.e., i-th brain area and j-th brain Pearson correlation coefficient between area；N represents time point number；xiT () represents the time series of i-th brain area；Represent i-th The seasonal effect in time series mean value of individual brain area；xjT () represents the time series of j-th brain area；Represent the time sequence of j-th brain area The mean value of row；The dimension of Pearson relevance matrix is 90 × 90.
In step S4, the dimension of High order correletion matrix is time window number × 4005.
In step S5, sparse linear regression model is specifically expressed as follows：
xm=Amαm+τm(2)；
In formula (2)：xmRepresent the time series for selecting brain area；Am=[x1,...,xm-1,0,xm+1,...,xM], its bag Time series containing all brain areas in addition to selected brain area；αmRepresent other brain areas to select the weight of brain area influence degree to Amount；τmRepresent noise item；αmThe corresponding brain area of middle nonzero element is the brain area interacted with selected brain area.
In step S6, computing formula is specifically expressed as follows：
In formula (3)：HCC1V () represents first kind cluster coefficients；U, t, v represent node；V represents set of node, and ε represents side collection, and e represents super side；N (v) is represented and is included node v The set of other nodes that contains of super side；IfSuch as u, t ∈ eiBut,ThenIt is no Then
In formula (4)：HCC2V () represents Equations of The Second Kind cluster coefficients；U, t, v represent node；V represents set of node, and ε represents side collection, and e represents super side；N (v) is represented and is included node v The set of other nodes that contains of super side；IfSuch as u, t, v ∈ ei, then I ' (u, t, v)=1, otherwise I ' (u, T, v)=0；
In formula (5)：HCC3 (v) represents the 3rd class cluster coefficients；V represents node；| e | represents that super side includes nodes Amount；V represents set of node, and ε represents side collection, and e represents super side；N (v) is represented comprising section The set of other nodes that the super side of point v is contained；S (v)={ ei∈ε:v∈ei, v represents node, eiRepresent super side；S (v) tables Show the set on the super side comprising node v；
In formula (6)：D (v) represents the node degree of vertex v；V represents node；E represents super side；ε represents side collection；H(v,e) Represent corresponding element in High order correletion matrix；
In formula (7)：H (v, e) represents corresponding element in High order correletion matrix；The row element of High order correletion matrix is section Point, column element is super side；If node v belongs to super side e, H (v, e)=1；If node v is not belonging to super side e, H (v, e) =0.
In step S7, the construction step of grader is specifically included：Using RBF kernel functions, after selecting non-parametric test Thus local attribute with notable group difference carries out the structure of grader as characteristic of division；
Checking procedure is specifically included：The sample of random selection 90% is used as training set, remaining 10% sample from sample set Thus this carry out class test and obtain classification accuracy as test set；Would be repeated for what is obtained after the test of 100 subseries Classification accuracy carries out arithmetic average, then using arithmetic mean of instantaneous value as grader classification accuracy.
In step S8, quantitative formula is specifically expressed as follows：
In formula (8)：D represents importance degree of the selected feature in grader；S represents the set of all features；| S | is represented The number of feature in S；xiRepresent selected feature；C represents the class label of sample；I(xi, c) represent the class of selected feature and sample The mutual information of distinguishing label c；
In formula (9)：R represents redundancy of the selected feature in grader；S represents the set of all features；| S | is represented The number of feature in S；xiRepresent selected feature；xjRepresent further feature；I(xi, xj) represent that selected feature is mutual with further feature Information；
Postsearch screening step is specifically included：Respectively selected feature is arranged according to importance degree size and redundancy size Name, then filters out that importance degree is larger and the less feature of redundancy.

Claims (8)

1. a kind of tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network, it is characterised in that：The method It is to be realized using following steps：
Step S1：Tranquillization state functional magnetic resonance imaging is pre-processed, and according to selected standardization brain map to pretreatment Tranquillization state functional magnetic resonance imaging afterwards carries out region segmentation, and then each brain area to being split carries out carrying for average time sequence Take；
Step S2：The sliding window that designated length is fixed, and when carrying out to the average time sequence of each brain area according to a fixed step size Between window segmentation；
Step S3：The average time sequence of each brain area under each time window Pearson correlation coefficient between any two is calculated, thus Obtain Pearson relevance matrix；
Step S4：The value of corresponding element in Pearson relevance matrix is extracted, High order correletion matrix is thus obtained；
Step S5：Using sparse linear regression model, linear group of each element and other elements in High order correletion matrix is calculated Close and represent, thus set up super side, then high-order super-network is built according to super side；
Step S6：Calculate the local attribute of high-order super-network；The local attribute includes：In high-order super-network the degree of each element and Cluster coefficients；
Step S7：Using support vector cassification algorithm, select the local attribute of high-order super-network as characteristic of division, thus enter The structure of row grader, is then tested using cross validation method to the grader for building；
Step S8：Using mutual information analysis method, the importance degree and redundancy to selected feature in grader quantifies, so Afterwards postsearch screening is carried out to selected feature according to quantized result, thus high-order super-network is optimized.
2. the tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network according to claim 1, its It is characterised by：In step S1, pre-treatment step is specifically included：Time horizon correction, the dynamic correction of head, joint registration, space mark Standardization, low frequency filtering；Standardization brain map adopts AAL templates；The extraction step of average time sequence is specifically included：Extract AAL BOLD intensity of all voxels that each brain area is included in template in different time points, and by each voxel in different time points On BOLD intensity carry out arithmetic average, thus obtain the average time sequence of each brain area.
3. the tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network according to claim 1, its It is characterised by：In step S3, computing formula is specifically expressed as follows：
r i j = Σ i , j = 1 n ( x i ( t ) - x i ‾ ) ( x j ( t ) - x j ‾ ) Σ i = 1 n ( x i ( t ) - x i ‾ ) 2 Σ j = 1 n ( x j ( t ) - x j ‾ ) 2 - - - ( 1 ) ;
In formula (1)：rijRepresent the element of the i-th row jth row in Pearson relevance matrix, i.e., i-th brain area and j-th brain area it Between Pearson correlation coefficient；N represents time point number；xiT () represents the time series of i-th brain area；Represent i-th brain The seasonal effect in time series mean value in area；xjT () represents the time series of j-th brain area；Represent the seasonal effect in time series of j-th brain area Mean value；The dimension of Pearson relevance matrix is 90 × 90.
4. the tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network according to claim 1, its It is characterised by：In step S4, the dimension of High order correletion matrix is time window number × 4005.
5. the tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network according to claim 1, its It is characterised by：In step S5, sparse linear regression model is specifically expressed as follows：
xm=Amαm+τm(2)；
In formula (2)：xmRepresent the time series for selecting brain area；Am=[x1,...,xm-1,0,xm+1,...,xM], it is included except choosing Determine the time series of all brain areas outside brain area；αmRepresent other brain areas to selecting the weight vectors of brain area influence degree；τmTable Show noise item；αmThe corresponding brain area of middle nonzero element is the brain area interacted with selected brain area.
6. the tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network according to claim 1, its It is characterised by：In step S6, computing formula is specifically expressed as follows：
In formula (3)：HCC1V () represents first kind cluster coefficients；U, t, v represent node；V represents set of node, and ε represents side collection, and e represents super side；N (v) is represented and is included node v The set of other nodes that contains of super side；IfSuch as u, t ∈ eiBut,Then, it is no Then
HCC 2 ( v ) = 2 Σ u , t ∈ N ( v ) I ′ ( u , t , v ) | N ( v ) | ( | N ( v ) | - 1 ) - - - ( 4 ) ;
In formula (4)：HCC2V () represents Equations of The Second Kind cluster coefficients；U, t, v represent node；V represents set of node, and ε represents side collection, and e represents super side；N (v) is represented and is included node v The set of other nodes that contains of super side；IfSuch as u, t, v ∈ ei, then I ' (u, t, v)=1, otherwise I ' (u, T, v)=0；
HCC 3 ( v ) = 2 Σ e ∈ S ( v ) ( | e | - 1 ) - | N ( v ) | | N ( v ) | ( | S ( v ) | - 1 ) - - - ( 5 ) ;
In formula (5)：HCC3V () represents the 3rd class cluster coefficients；V represents node；| e | represents that super side includes number of nodes；V represents set of node, and ε represents side collection, and e represents super side；N (v) is represented and is included node v The set of other nodes that contains of super side；S (v)={ ei∈ε:v∈ei, v represents node, eiRepresent super side；S (v) represents bag The set on the super side containing node v；
d ( v ) = Σ e ∈ ϵ H ( v , e ) - - - ( 6 ) ;
In formula (6)：D (v) represents the node degree of vertex v；V represents node；E represents super side；ε represents side collection；H (v, e) is represented Corresponding element in High order correletion matrix；
H ( v , e ) = 1 , i f v ∈ e 0 , i f v ∉ e - - - ( 7 ) ;
In formula (7)：H (v, e) represents corresponding element in High order correletion matrix；The row element of High order correletion matrix is node, Column element is super side；If node v belongs to super side e, H (v, e)=1；If node v is not belonging to super side e, H (v, e)=0.
7. the tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network according to claim 1, its It is characterised by：In step S7, the construction step of grader is specifically included：Using RBF kernel functions, after selecting non-parametric test Thus local attribute with notable group difference carries out the structure of grader as characteristic of division；
Checking procedure is specifically included：The sample of random selection 90% used as training set, make by remaining 10% sample from sample set For test set, thus carry out class test and obtain classification accuracy；Would be repeated for the classification obtained after the test of 100 subseries Accuracy rate carries out arithmetic average, then using arithmetic mean of instantaneous value as grader classification accuracy.
8. the tranquillization state functional magnetic resonance imaging data classification method based on high-order super-network according to claim 1, its It is characterised by：In step S8, quantitative formula is specifically expressed as follows：
D = 1 | S | Σ x i ∈ S I ( x i , c ) - - - ( 8 ) ;
In formula (8)：D represents importance degree of the selected feature in grader；S represents the set of all features；| S | is represented in S The number of feature；xiRepresent selected feature；C represents the class label of sample；I(xi, c) represent the classification of selected feature and sample The mutual information of label c；
R = 1 | S | 2 Σ x i , x j ∈ S I ( x i , x j ) - - - ( 9 ) ;
In formula (9)：R represents redundancy of the selected feature in grader；S represents the set of all features；| S | is represented in S The number of feature；xiRepresent selected feature；xjRepresent further feature；I(xi, xj) mutual trust of feature and further feature selected by expression Breath；
Postsearch screening step is specifically included：Respectively ranking is carried out to selected feature according to importance degree size and redundancy size, so After filter out that importance degree is larger and the less feature of redundancy.
CN201611251416.6A 2016-12-30 2016-12-30 Resting state function magnetic resonance image data classification method based on high-order hyper-network Active CN106650818B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611251416.6A CN106650818B (en) 2016-12-30 2016-12-30 Resting state function magnetic resonance image data classification method based on high-order hyper-network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611251416.6A CN106650818B (en) 2016-12-30 2016-12-30 Resting state function magnetic resonance image data classification method based on high-order hyper-network

Publications (2)

Publication Number Publication Date
CN106650818A true CN106650818A (en) 2017-05-10
CN106650818B CN106650818B (en) 2020-01-03

Family

ID=58836651

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611251416.6A Active CN106650818B (en) 2016-12-30 2016-12-30 Resting state function magnetic resonance image data classification method based on high-order hyper-network

Country Status (1)

Country Link
CN (1) CN106650818B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107133651A (en) * 2017-05-12 2017-09-05 太原理工大学 The functional magnetic resonance imaging data classification method of subgraph is differentiated based on super-network
CN108846407A (en) * 2018-04-20 2018-11-20 太原理工大学 The nuclear magnetic resonance image classification method of brain network is not known based on independent element high order
CN109035265A (en) * 2018-09-26 2018-12-18 重庆邮电大学 A kind of novel structure brain connection map construction method
CN110211386A (en) * 2019-05-22 2019-09-06 东南大学 A kind of highway vehicle type classification method based on non-parametric test
CN111325288A (en) * 2020-03-17 2020-06-23 山东工商学院 Clustering idea-based multi-view dynamic brain network characteristic dimension reduction method
CN111754395A (en) * 2020-07-01 2020-10-09 太原理工大学 Robustness assessment method for brain function hyper-network model
CN113723485A (en) * 2021-08-23 2021-11-30 天津大学 Method for processing brain image hypergraph of mild hepatic encephalopathy

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104836711A (en) * 2015-03-29 2015-08-12 朱江 Construction method of command control network generative model

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104836711A (en) * 2015-03-29 2015-08-12 朱江 Construction method of command control network generative model

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
BIAO JIE,ETC: ""Brain Connectivity Hyper-Network for MCI Classification"", 《PROC OF THE INTERNATIONAL CONFERENCE》 *
曹静: ""基于最大相关最小冗余的特征选择算法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
郭浩，等: ""静息态功能脑网络差异指标分析及抑郁症分类应用"", 《计算机应用与软件》 *
郭浩: ""抑郁症静息态功能脑网络异常拓扑属性分析及分类研究"", 《中国博士学位论文全文数据库 医药卫生科技辑》 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107133651A (en) * 2017-05-12 2017-09-05 太原理工大学 The functional magnetic resonance imaging data classification method of subgraph is differentiated based on super-network
CN107133651B (en) * 2017-05-12 2018-03-16 太原理工大学 The functional magnetic resonance imaging data classification method of subgraph is differentiated based on super-network
CN108846407A (en) * 2018-04-20 2018-11-20 太原理工大学 The nuclear magnetic resonance image classification method of brain network is not known based on independent element high order
CN108846407B (en) * 2018-04-20 2022-02-08 太原理工大学 Magnetic resonance image classification method based on independent component high-order uncertain brain network
CN109035265A (en) * 2018-09-26 2018-12-18 重庆邮电大学 A kind of novel structure brain connection map construction method
CN110211386A (en) * 2019-05-22 2019-09-06 东南大学 A kind of highway vehicle type classification method based on non-parametric test
CN111325288A (en) * 2020-03-17 2020-06-23 山东工商学院 Clustering idea-based multi-view dynamic brain network characteristic dimension reduction method
CN111325288B (en) * 2020-03-17 2022-02-25 山东工商学院 Clustering idea-based multi-view dynamic brain network characteristic dimension reduction method
CN111754395A (en) * 2020-07-01 2020-10-09 太原理工大学 Robustness assessment method for brain function hyper-network model
CN111754395B (en) * 2020-07-01 2022-07-08 太原理工大学 Robustness assessment method for brain function hyper-network model
CN113723485A (en) * 2021-08-23 2021-11-30 天津大学 Method for processing brain image hypergraph of mild hepatic encephalopathy
CN113723485B (en) * 2021-08-23 2023-06-06 天津大学 Hypergraph processing method for brain image of mild hepatic encephalopathy

Also Published As

Publication number Publication date
CN106650818B (en) 2020-01-03

Similar Documents

Publication Publication Date Title
CN106650818A (en) Resting state function magnetic resonance image data classification method based on high-order super network
Khan et al. Intelligent fusion-assisted skin lesion localization and classification for smart healthcare
CN106548206B (en) Multi-modal nuclear magnetic resonance image data classification method based on minimum spanning tree
Amico et al. Mapping hybrid functional-structural connectivity traits in the human connectome
CN103886328B (en) Based on the functional magnetic resonance imaging data classification method of brain mixed-media network modules mixed-media architectural feature
CN107133651B (en) The functional magnetic resonance imaging data classification method of subgraph is differentiated based on super-network
Dimitriadis et al. Improving the reliability of network metrics in structural brain networks by integrating different network weighting strategies into a single graph
CN105046709A (en) Nuclear magnetic resonance imaging based brain age analysis method
CN112418337B (en) Multi-feature fusion data classification method based on brain function hyper-network model
CN108345903B (en) A kind of multi-modal fusion image classification method based on mode distance restraint
CN104715241B (en) Tensor decomposition-based fMRI feature extraction and identification method
CN103699904B (en) The image computer auxiliary judgment method of multisequencing nuclear magnetic resonance image
CN104715261A (en) fMRI dynamic brain function sub-network construction and parallel SVM weighted recognition method
CN107944490A (en) A kind of image classification method based on half multi-modal fusion feature reduction frame
CN109993230A (en) A kind of TSK Fuzzy System Modeling method towards brain function MRI classification
CN111754395B (en) Robustness assessment method for brain function hyper-network model
CN105117731A (en) Community partition method of brain functional network
CN104573742A (en) Medical image classification method and system
CN106127263A (en) The human brain magnetic resonance image (MRI) classifying identification method extracted based on three-dimensional feature and system
CN111090764A (en) Image classification method and device based on multitask learning and graph convolution neural network
CN105989336A (en) Scene recognition method based on deconvolution deep network learning with weight
CN115474939A (en) Autism spectrum disorder recognition model based on deep expansion neural network
CN104573745B (en) Fruit-fly classified method based on magnetic resonance imaging
Yeung et al. Pipeline comparisons of convolutional neural networks for structural connectomes: predicting sex across 3,152 participants
CN111612739A (en) Deep learning-based cerebral infarction 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
GR01 Patent grant
GR01 Patent grant