CN108920887A - A kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization - Google Patents
A kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization Download PDFInfo
- Publication number
- CN108920887A CN108920887A CN201810585444.4A CN201810585444A CN108920887A CN 108920887 A CN108920887 A CN 108920887A CN 201810585444 A CN201810585444 A CN 201810585444A CN 108920887 A CN108920887 A CN 108920887A
- Authority
- CN
- China
- Prior art keywords
- brain
- metanetwork
- network
- brain network
- development
- 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
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of sequential organization brain network analysis methods based on Non-negative Matrix Factorization in field of neural networks, include the following steps：1）Non-negative sequential organization brain network is constructed, corresponding each time point constructs the correlation between a network representation Different brain region；2）Using Non-negative Matrix Factorization as basic model by sequential organization brain network decomposition at multiple metanetworks, it is desirable that the development track of metanetwork and metanetwork after decomposition all meets nonnegativity restrictions；3）Low biasing is carried out to true timing brain network by addition nuclear norm regular terms to rebuild；4）Timing flatness regular terms is applied to development track corresponding to the metanetwork after decomposition；5）Orthogonality constraint is applied to the metanetwork after decomposition, so that not overlapped each other between metanetwork, for disclosing different brain sub-network development models, i.e. collaboration development models between the set of Different brain region, the present invention provides a benchmark for the development of healthy brain network, can be used in brain development research.
Description
Technical field
The present invention relates to a kind of sequential organization network, in particular to a kind of sequential organization brain network analysis method.
Background technique
In past ten years, educational circles of neurology department has reached common understanding, and the development of human brain is no matter to go back in structure
Being all is functionally nonlinear process.The understanding that people develop Normal brain is for disclosing structure-function relationship and reason
Neurodevelopmental disorder, such as self-closing disease, schizophrenia and attention deficit hyperactivity disorder are solved, is all vital.
Early stage is concentrated mainly on the development track of each function system the research of brain growth.For example, basic vision,
The feeling region of movement and the sense of hearing is developed at first, followed by early stage language skill and higher cognitive function.In recent years, have
The research for closing each brain area diencephalon connection sexual development of brain is of increasing concern.The science and technology of Rapid development is but also collect brain
Link information become increasingly easy, it is therefore an objective to quantify the correlation between any pair of brain region.Magnetic resonance imaging
(MRI) and diffusion tensor (DTI) respectively by Pearson correlation coefficient between the brain area attribute based on multiple measurands and
The link information in structure is provided based on the tracking of the white matter fiber of single measurand.Functional magnetic resonance imaging (fMRI), brain
Electrograph (EEG) is then based respectively on interrelated, mutual information with synchronous likelihood to quantify the function between brain area with magneticencephalogram (MEG)
It can connection.In the measurement of different types of brain area degree of communication, the deep understanding to the development models of structural connectivity is heavy to closing
It wants, it is close with the performance of every cerebral function and behavior because the structure organization of brain network is the neural basal of function connects
It is related.Understand that the growth course of brain structure connection can explain how cerebral function state generates from fabric,
And certain clue can be provided for the potential cause of various cerebral disorders.In recent years, neuroscientist has found the network of brain
Structure is that balance minimizes connection cost and maximizes the economic product for the ability that is suitable for, but people are to various economic factors
(such as connection cost and adaptability) is how mutually coordinated still to know little about it during brain development.
In the research of structure brain network, there are some researchers to think the structure associative mode of brain, compared to white matter fibre
Connection mode is tieed up, it may be closer to functional connection mode, because white matter may not be the sole mode of interaction between brain area.With
The preceding analysis and research to sequential organization brain network are broadly divided into two classes, and one kind uses the different methods based on graph theory, such as
Worldlet attribute and network efficiency, to determine the distinctive correlated characteristic of particular point in time；Another kind of is the timing for studying brain network
Modular organisation is divided with disclosing the sub-network structure evolved and community, and leads to the potential mechanism of its institutional framework.It is these two types of
Strategy otherwise indicate brain network with several topological attribute or with one group of lesser module.But these two types of strategies all do not have
Have and analyzed using complete brain network, therefore inevitably leads to the loss of information.In addition, second of strategy will be sub
Network regards that inner tight connects and external mutually disjoint module as, but has ignored a fact, i.e., one specific
Brain region may participate in multiple concurrent perception and cognitive function circuit.So far, researcher is to structure brain network
Timing collaborative variation development models research it is also seldom, more rarely have people to the variation of the weight of each metanetwork (i.e. development track)
Carry out quantification.
Complete timing brain network is analyzed, traditional matrix disassembling method include principal component analysis (PCA) and
Independent component analysis (ICA).Although these methods are already used to the brain network of analysis dynamic function, the spy that they are generated
The weight for levying vector or independent element allows negative, there is the complicated effect cancelled out each other, thus is not easy to explain.In order to true
The interpretation of brain connection mode and their development track is protected, a relatively good solution is using nonnegative matrix point
It solves (NMF).NMF can be regarded as a kind of version of principal component analysis, the difference is that it require decompose two because
Submatrix does not include negative.This is an important constraint, as it means that people can be (right according to obtained metanetwork is decomposed
Answer basic matrix) to explain the connection weight in metanetwork, and temporal expression matrix (coefficient of correspondence matrix) is construed to metanetwork
The dynamic of mode is contributed.But even Non-negative Matrix Factorization can not also be fully solved non-negative timing brain Crosslinking Structural
The problem of, such as the noise problem introduced in data-gathering process, the plyability problem of time smoothing problem and metanetwork.
Summary of the invention
The object of the present invention is to provide a kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization, Ke Yiwei
Sequential organization brain network provides consistent and compact expression；The development for analyzing brain structure network as the result is shown is one smooth
Process, it has merged the metanetwork of multiple spatial isomerisms；Each metanetwork is made of the collaborative variation correlation of multiple brain areas,
And as time change has corresponding development track；The brain timing metanetwork of this method discovery discloses brain network tissue
The potential association mode of dynamic change, and the different development tracks of each metanetwork have then quantified their moving in growth course
State contribution, so that the development for healthy brain network provides a benchmark.
The object of the present invention is achieved like this：A kind of sequential organization brain network analysis side based on Non-negative Matrix Factorization
Method includes the following steps：
1) non-negative sequential organization brain network is constructed, corresponding each time point constructs between a network representation Different brain region
Correlation, indicates uncorrelated between brain area with 0 in each network, indicates the correlation intensity between brain area with the real number value greater than 0
Value；
2) using Non-negative Matrix Factorization (NMF) as basic model by sequential organization brain network decomposition at multiple metanetworks,
It is required that the development track of metanetwork and metanetwork after decomposing all meets nonnegativity restrictions, to guarantee its interpretation；
3) it carries out low biasing to true timing brain network by addition nuclear norm regular terms to rebuild, so that it is dry to reduce noise
It disturbs；
4) timing flatness regular terms is applied to development track corresponding to the metanetwork after decomposition, in order to ensure
The bonding strength of metanetwork is smooth smooth at any time；
5) orthogonality constraint is applied to the metanetwork after decomposition, so that not overlapping each other between metanetwork, for disclosing difference
Brain sub-network development models, i.e., Different brain region set between collaboration development models.
It is further limited as of the invention, correlation matrix between brain area of the building based on cortical thickness, uses in step 1)
Pearson correlation coefficient absolute value between each pair of brain area, and it is different from 0/1 change pretreatment of previous processing brain network, retain
All absolute correlation values, to use more quantitative informations for analyzing the development models of brain.
It is further limited as of the invention, step 2) is using Non-negative Matrix Factorization as basic model, it is desirable that after decomposition
Metanetwork and the development track of metanetwork all meet nonnegativity restrictions：
Wherein, X=[x1, x2..., xT] be non-negative correlation coefficient two-by-two between recording brain area timing brain network, T is the time
The number of point, xiIt is the brain network vector representation at i-th of time point, U=[u1, u2..., ur] it is metanetwork set, r
For the number of metanetwork, V=[v1, v2..., vr] it is that U develops track accordingly.
Further limited as of the invention, step 3) by addition nuclear norm regular terms to true timing brain network into
The low biasing of row is rebuild, and is specially：
s.t.X0=UVT, U > 0, V >=0
Wherein, X0The true timing brain network to be restored is indicated, so the first item of formula indicates true timing brain network
Approximate error, the Section 2 of formula carries out low-rank constraint by nuclear norm regular terms, to true timing brain network, because high
Order part is often the λ X as caused by noise0Order weight, the bound term of the second row indicates that we will be from true timing
It is decomposed in brain network and obtains metanetwork and its develop track, the provable above objective function is equivalent to following form：
Wherein equality constraint is utilized in first item, that is, uses UVTInstead of X0, Section 2 be based on minimize two nonnegative matrixes multiply
Long-pending nuclear norm is equivalent to minimize the existing theory of their F norm, i.e.,
It is further limited as of the invention, step 4) applies timing flatness canonical to the metanetwork development track of decomposition
, so that it is guaranteed that the bonding strength of metanetwork smooth change at any time, is specially：
Wherein, the Section 3 of objective function is the slickness regular terms of the development track to all metanetworks, because according to
Existing brain network model, the connection mode of brain network be slowly it is changed, L is defined in the Laplacian Matrix on V,
wijIndicate the neighbor relationships at time point, β is the weight of slickness regular terms.
It is further limited as of the invention, step 5) applies orthogonality constraint to metanetwork, does not overlap each other to generate
Metanetwork, for disclosing different brain sub-network development models, objective function is：
s.t.UTU=Ir
Since the Frobenius norm of U is fixed under the orthogonality constraint of above formula, it is equivalent to final objective function
For：
s.t.UTU=Ir
By using Lagragian Multiplier Method, the multiplication that can be derived by U and V updates rule as follows：
Compared with prior art, the beneficial effects of the present invention are：
1) present invention is compared with other NMF algorithms, it is contemplated that the low-dimensional of truthful data manifold, to unknown true number
Be applied with low-rank constraint according to (i.e. muting data), and indicated with nuclear norm, thus can not only decompose to obtain two because
Submatrix (corresponding metanetwork and development track matrix), but also original low-rank truthful data can be restored；
2) present invention in the past with the method for graph theory or Modularity analysis timing brain network compared with, avoid one or
Multiple regions and other regions separately carry out independent module analysis, but carry out global analysis to whole network, to make brain
The relationship two-by-two in section is completely retained, to carry out agonic analysis；This point is very important, because each big
Brain area domain ceaselessly develops all in growth course to adapt to the environment of entire brain；This also explains why we allow
There is the brain region of overlapping in different metanetworks, this is more more meaningful than Modularity analysis biologically；
3) present invention is different with the research of 0/1 mode is only carried out to sequential organization brain network in the past, and this method is analyzing number
According to when remain all correlations two-by-two between brain region, thus reduce the loss of information, can obtain more meaningful
Analysis result.
Detailed description of the invention
Fig. 1 is flow diagram of the present invention.
Fig. 2 is the metanetwork that the present invention is applied to that 3-20 years old sequential organization brain network decomposition obtains
Fig. 3 is the development track of metanetwork in Fig. 2.
Fig. 4 be the present invention with the development track roughness ratio of PCA compared with.
Specific embodiment
The present invention will be further described combined with specific embodiments below.
A kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization as shown in Figure 1, steps are as follows：
One, non-negative sequential organization brain network is constructed：Firstly, collecting the test object of each age (such as 3-20 years old)
Each brain area skin thickness is then based on the brain area skin thickness of multiple measurands, calculates the Pearson phase relation between brain area
Number, and takes its absolute value, i.e., indicates uncorrelated between brain area with 0, indicate the correlation intensity between brain area with the real number value for being greater than 0
Value finally obtains one group of timing brain structural network based on brain area cortical thickness；
Two, by sequential organization brain network decomposition at multiple metanetworks：Assuming that X=[x1, x2..., xT] it is that there is absolute brain area
Between related coefficient timing brain network,It is that the vectorization of timing brain network that length is n i-th time point shows shape
Formula, T are the numbers at time point, and metanetwork is U=[u1, u2..., ur], it is V=[v that they develop track accordingly1, v2...,
vr],It is the metanetwork of j-th of vector quantization,It isDevelopment track, r is the number of metanetwork, vijIt is
Weight of j-th of metanetwork in i-th of time point building sequential network；Therefore, the brain network at i-th of time point can pass through
Following expression re-formation：
xi≈u1×vi1+u2×vi2+…+ur×vir
Then, we carry out nonnegativity restrictions to U and V using Non-negative Matrix Factorization as basic model：
But the decomposition of nonnegative matrix also results in the sparse of the metanetwork after decomposing, and the positive input of metanetwork can indicate to deposit
, and zero entry representation does not have correlation, then we need to introduce noise, x0Indicate true timing network of relation, E generation
Table noise：
X=X0+E.rank(X0)≤r
The data low-rank representation of noise reduction, we have following objective function：
It consists of two parts：First part is approximate error, and second part is nuclear norm regularization term, and λ is the order of X0
Weight, in order to combine it with the basic model of NMF, by x0NMF indicate as the constraint to objective function：
s.t.X0=UVT, U > 0, V >=0
Use UVTInstead of X0, we write above formula as succinct form：
The nuclear norm for minimizing two nonnegative matrix products is equal to the F norm for minimizing them：
Nuclear norm in above formula is replaced with into F norm, is further simplified optimization：
Then, it would be desirable to which the problem of considering time Smoothness, our target is to ensure that with cranial nerve network
Constantly during development, development track can smoothly change, and in order to assess the smoothness of development track, allow V (vi=[vi1,
vi2..., vir]) the i-th row become all metanetworks i-th of time node co-variation value, andIt is brain
Neighbor relationships of the network on ith and jth time point,Indicate the institute in all different time points
There is the flatness of development track, we can be written as follow form：
L=D-W is defined in the Laplacian Matrix on each row of V, and W is that element is wijPairs of similar matrix, D is W
Diagonal degree matrix, in order to solve the problems, such as time Smoothness, we can apply timing flatness regular terms, to obtain new
Objective function, wherein the Section 3 of objective function is the slickness regular terms of the development track to all metanetworks, because according to
Existing brain network model, its connection mode are slowly changed：
Finally, due to which our model is built upon in the analysis to connection matrix, therefore model itself has solved
The same brain region may participate in the problem of different metanetworks, still, in order to avoid identical connection has different development
Track, we are applied with to Orthonormality constraints on metanetwork matrix, to ensure not overlapping each other for metanetwork, therefore target letter
Number can be write as：
s.t.UTU=Ir
Since the Frobenius norm of U is fixed under the orthogonality constraint of above formula, final objective function can be write as simpler
Clean form：
s.t.UTU=Ir
By using Lagragian Multiplier Method, the multiplication that can be derived by U and V updates rule as follows：
⊙ represents the product of element, these updates can be obtained by carrying out derivation to objective function as above, in order to keep away
Exempt from local minimum, we use the initialization strategy on U and V similar with kmeans algorithm, that is, pass through random initializtion
Repetition multiple (100) is secondary, selects optimal data.
Three, the number of metanetwork is determined：
The number of metanetwork is an important parameter in the method for the present invention.Very little metanetwork possibly can not capture big
Some secondary development models of brain network, there may be unstable results for too many metanetwork.Therefore, the suitable member of selection
Network number has important influence to the stability and reliability of result.The present invention mainly determines suitable member in terms of two
Network number.On the one hand, we have studied under different metanetwork numbers, the repeatability of decomposition result；On the other hand, Wo Menjian
It has tested under different metanetwork numbers, decomposition result is reconstitution to timing brain network.
When carrying out repeatability experiment, sequential organization brain network is divided into two according to the odd and even number age by us
Point, i.e. it odd number age (3,5,7,9,11,13,15,17,19) and even number age (4,6,8,10,12,14,16,18,20), carries out
Different number of metanetwork decomposes.By calculating the similarity between the corresponding metanetwork of two parts (or development track), to knot
The repeatability of fruit is quantified.Specifically, we used normalized inner product, i.e. cosine similarity, carry out metrics match
Metanetwork (or development track) between similarity.Since two subsets of sequential organization brain network have 1 year time interval,
Therefore cosine similarity is advantageous in the evaluation to development track repeatability, because it is the judgement to direction, without
It is that stringent size compares.If two development tracks are parallel, the two most like (maximum value 1), if they are
Vertical, then two development tracks are most different (minimum values 0)；
When carrying out failtests, the present invention evaluates different number of metanetwork clock synchronization using the standard of mean square error
The reconstructed error of sequence structure brain network.
By combining reproducibility and the result of fail-safe analysis, we select most suitable metanetwork number to solve
Release the reproducibility of most of vertical structure network of relation.
By application it is proposed that the timing brain network analysis model based on Non-negative Matrix Factorization, we were from 3-20 years old
5 metanetworks are decomposited in timing cortex-thickness structure brain network, as Figure 2-3.Each metanetwork corresponds to difference
Brain development subpattern, and its develop track then quantified the variation that each metanetwork changes over time importance；In fact,
Mathematical model proposed by the invention also can be applied in other kinds of sequential network analysis.The present invention and PCA algorithm phase
Than the result not only generated has interpretation, but also roughness is lower, as shown in figure 4, flatness is more preferably；In the present invention,
A time smoothing constraint is applied with to punish the change dramatically of metanetwork development track, can effectively be weakened when certain in this way
Between put some unexpected influence of noises of introducing, so that it is guaranteed that this method robustness is higher.
The present invention is not limited to the above embodiments, on the basis of technical solution disclosed by the invention, the skill of this field
For art personnel according to disclosed technology contents, one can be made to some of which technical characteristic by not needing creative labor
A little replacements and deformation, these replacements and deformation are within the scope of the invention.
Claims (6)
1. a kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization, which is characterized in that include the following steps：
1) non-negative sequential organization brain network is constructed, corresponding each time point constructs the correlation between a network representation Different brain region
Property, it indicates uncorrelated between brain area with 0 in each network, indicates the correlation intensity value between brain area with the real number value greater than 0；
2) using Non-negative Matrix Factorization as basic model by sequential organization brain network decomposition at multiple metanetworks, it is desirable that after decomposition
Metanetwork and the development track of metanetwork all meet nonnegativity restrictions；
3) low biasing is carried out to true timing brain network by addition nuclear norm regular terms to rebuild；
4) timing flatness regular terms is applied to development track corresponding to the metanetwork after decomposition；
5) orthogonality constraint is applied to the metanetwork after decomposition, it is different big for disclosing so that do not overlapped each other between metanetwork
Collaboration development models between brain network development models, i.e. Different brain region set.
2. a kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization according to claim 1, feature
It is, correlation matrix between brain area of the building based on cortical thickness, uses the Pearson correlation coefficient between each pair of brain area in step 1)
Absolute value, and it is different from 0/1 change pretreatment of previous processing brain network, retain all absolute correlation values, to use more
More quantitative informations is used to analyze the development models of brain.
3. according to claim 1 or 2 establish a kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization,
It is characterized in that, step 2) is using Non-negative Matrix Factorization as basic model, it is desirable that the hair of metanetwork and metanetwork after decomposition
It educates track and all meets nonnegativity restrictions, specially：
Wherein, X=[x1, x2..., xT] be non-negative correlation coefficient two-by-two between recording brain area timing brain network, T is time point
Number, xiIt is the brain network vector representation at i-th of time point, U=[u1, u2..., ur] it is metanetwork set, r is member
The number of network, V=[v1, v2..., vr] it is that U develops track accordingly.
4. a kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization according to claim 1 or 2, special
Sign is that step 3) carries out low biasing to true timing brain network by addition nuclear norm regular terms and rebuilds, and is specially：
s.t.X0=UVT, U > 0, V >=O
Wherein, X0Indicate the true timing brain network to be restored, the first item of formula indicates that the approximate of true timing brain network is missed
Difference, the Section 2 of formula carry out low-rank constraint to true timing brain network, because high order part is past by nuclear norm regular terms
Toward being the λ X as caused by noise0Order weight, the bound term of the second row indicates that we will be from true timing brain network
Decomposition obtains metanetwork and its development track, the provable above objective function are equivalent to following form：
Wherein equality constraint is utilized in first item, that is, uses UVTInstead of X0, Section 2 is based on two nonnegative matrix products of minimum
Nuclear norm is equivalent to minimize the existing theory of their F norm, i.e.,
5. a kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization according to claim 1 or 2, special
Sign is that step 4) applies timing flatness regular terms to the metanetwork development track of decomposition, so that it is guaranteed that the connection of metanetwork
Intensity smooth change at any time is specially：
Wherein, the Section 3 of objective function is the slickness regular terms of the development track to all metanetworks, because according to existing
Brain network model, the connection mode of brain network be slowly it is changed, L is defined in the Laplacian Matrix on V,
wijIndicate the neighbor relationships at time point, β is the weight of slickness regular terms.
6. a kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization according to claim 1 or 2, special
Sign is, step 5) applies orthogonality constraint to metanetwork, different big for disclosing to generate the metanetwork not overlapped each other
Brain network development models, objective function are：
s.t.UTU=Ir
Since the Frobenius norm of U is fixed under the orthogonality constraint of above formula, it is equivalent to final objective function and is：
s.t.UTU=Ir
By using Lagragian Multiplier Method, the multiplication that can be derived by U and V updates rule as follows：
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810585444.4A CN108920887B (en) | 2018-06-08 | 2018-06-08 | Time sequence structure brain network analysis method based on non-negative matrix factorization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810585444.4A CN108920887B (en) | 2018-06-08 | 2018-06-08 | Time sequence structure brain network analysis method based on non-negative matrix factorization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108920887A true CN108920887A (en) | 2018-11-30 |
| CN108920887B CN108920887B (en) | 2021-10-15 |
Family
ID=64418549
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810585444.4A Active CN108920887B (en) | 2018-06-08 | 2018-06-08 | Time sequence structure brain network analysis method based on non-negative matrix factorization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108920887B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109920550A (en) * | 2018-12-25 | 2019-06-21 | 天津大学 | A method of teenager's lafora's disease is studied based on dMRI |
| CN110428614A (en) * | 2019-07-11 | 2019-11-08 | 银江股份有限公司 | A kind of traffic congestion temperature spatio-temporal prediction method based on non-negative tensor resolution |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107016261A (en) * | 2017-04-11 | 2017-08-04 | 曲阜师范大学 | Difference expression gene discrimination method based on joint constrained non-negative matrix decomposition |
| CN107301382A (en) * | 2017-06-06 | 2017-10-27 | 西安电子科技大学 | The Activity recognition method of lower depth Non-negative Matrix Factorization is constrained based on Time Dependent |
| CN107424174A (en) * | 2017-07-15 | 2017-12-01 | 西安电子科技大学 | Motion marking area extracting method based on local restriction Non-negative Matrix Factorization |
-
2018
- 2018-06-08 CN CN201810585444.4A patent/CN108920887B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107016261A (en) * | 2017-04-11 | 2017-08-04 | 曲阜师范大学 | Difference expression gene discrimination method based on joint constrained non-negative matrix decomposition |
| CN107301382A (en) * | 2017-06-06 | 2017-10-27 | 西安电子科技大学 | The Activity recognition method of lower depth Non-negative Matrix Factorization is constrained based on Time Dependent |
| CN107424174A (en) * | 2017-07-15 | 2017-12-01 | 西安电子科技大学 | Motion marking area extracting method based on local restriction Non-negative Matrix Factorization |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109920550A (en) * | 2018-12-25 | 2019-06-21 | 天津大学 | A method of teenager's lafora's disease is studied based on dMRI |
| CN110428614A (en) * | 2019-07-11 | 2019-11-08 | 银江股份有限公司 | A kind of traffic congestion temperature spatio-temporal prediction method based on non-negative tensor resolution |
| CN110428614B (en) * | 2019-07-11 | 2021-02-05 | 银江股份有限公司 | Traffic jam heat degree space-time prediction method based on non-negative tensor decomposition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108920887B (en) | 2021-10-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Amico et al. | Mapping hybrid functional-structural connectivity traits in the human connectome | |
| Messé et al. | Predicting functional connectivity from structural connectivity via computational models using MRI: an extensive comparison study | |
| Kording et al. | How are complex cell properties adapted to the statistics of natural stimuli? | |
| Ryali et al. | Multivariate dynamical systems models for estimating causal interactions in fMRI | |
| Nakagawa et al. | Bottom up modeling of the connectome: linking structure and function in the resting brain and their changes in aging | |
| Gao et al. | Regularized matrix data clustering and its application to image analysis | |
| Mermillod et al. | Coarse scales are sufficient for efficient categorization of emotional facial expressions: Evidence from neural computation | |
| Glomb et al. | Resting state networks in empirical and simulated dynamic functional connectivity | |
| Qiao et al. | Ternary-task convolutional bidirectional neural turing machine for assessment of EEG-based cognitive workload | |
| Netzer et al. | Real-time EEG classification via coresets for BCI applications | |
| Ziraki et al. | Multiple-view flexible semi-supervised classification through consistent graph construction and label propagation | |
| Wein et al. | Brain Connectivity Studies on Structure‐Function Relationships: A Short Survey with an Emphasis on Machine Learning | |
| Ngo et al. | Beyond consensus: embracing heterogeneity in curated neuroimaging meta-analysis | |
| Wu et al. | Robust cluster validity indexes | |
| Litvak et al. | There’s no such thing as a ‘true’model: the challenge of assessing face validity | |
| Cao et al. | Semi-supervised tensor factorization for brain network analysis | |
| Qiang et al. | A deep learning method for autism spectrum disorder identification based on interactions of hierarchical brain networks | |
| CN108920887A (en) | A kind of sequential organization brain network analysis method based on Non-negative Matrix Factorization | |
| Betzel et al. | Living on the edge: network neuroscience beyond nodes | |
| Gao et al. | Non-linear manifold learning in fmri uncovers a low-dimensional space of brain dynamics | |
| Wein et al. | Forecasting brain activity based on models of spatiotemporal brain dynamics: A comparison of graph neural network architectures | |
| Ghosh et al. | A joint subspace mapping between structural and functional brain connectomes | |
| Rahimi et al. | Identifying nonlinear functional connectivity with EEG/MEG using nonlinear time-lagged multidimensional pattern connectivity (nTL-MDPC) | |
| Huang et al. | Connectivity analyses for task-based fMRI | |
| Tewarie et al. | Non-reversibility outperforms functional connectivity in characterisation of brain states in MEG data |
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 |