CN110263880A - Construction method, device and the intelligent terminal of cerebral disease disaggregated model - Google Patents
Construction method, device and the intelligent terminal of cerebral disease disaggregated model Download PDFInfo
- Publication number
- CN110263880A CN110263880A CN201910628897.5A CN201910628897A CN110263880A CN 110263880 A CN110263880 A CN 110263880A CN 201910628897 A CN201910628897 A CN 201910628897A CN 110263880 A CN110263880 A CN 110263880A
- Authority
- CN
- China
- Prior art keywords
- expressed
- cerebral disease
- image data
- brain
- data
- 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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
Abstract
The present invention provides construction method, device and the intelligent terminals of a kind of cerebral disease disaggregated model, this method comprises: obtaining Rs-fMRI image data from multiple cerebral disease data sets；Rs-fMRI image data is pre-processed based on multiple preset brain templates, obtains average time sequence；Feature is extracted according to average time sequence construct function connects network corresponding with each cerebral disease data set and brain template, and from function connects network, obtains eigenmatrix；According to preset sparse learning model, regularization term and eigenmatrix, multiple initialization cerebral disease disaggregated models corresponding with each cerebral disease data set are constructed；Integrated study is carried out to multiple initialization cerebral disease disaggregated models, determines the corresponding target cerebral disease disaggregated model of each cerebral disease data set.The present invention can make the brain function connection network of building have more physiological significance, and the big problem of heterogeneity present in data can be effectively relieved, improve the universality of cerebral disease disaggregated model.
Description
Technical field
The present invention relates to cerebral disease diagnostic techniques field, more particularly, to a kind of cerebral disease disaggregated model construction method,
Device and intelligent terminal.
Background technique
ASD (Autism spectrum disorder, autism spectrum disorder) is that one group fast-developing and height is heterogeneous
Neurodevelopmental disorder, proposed at present by machine learning techniques applied to ASD diagnose, but in the prior art building brain
There is stronger limitation when classification of diseases model, such as cerebral disease disaggregated model is constructed based on single brain template, based on single
Imaging center constructs cerebral disease disaggregated model or constructs cerebral disease disaggregated model based on the multiple brain templates of single imaging center, leads to
Crossing cerebral disease disaggregated model that these methods obtain has lower universality, so as to cause by cerebral disease disaggregated model application
There can be apparent heterogeneity when other imaging centers.
Summary of the invention
In view of this, the purpose of the present invention is to provide construction method, device and the intelligence of a kind of cerebral disease disaggregated model
Terminal can be effectively relieved the heterogeneous big problem of data, improve cerebral disease classification accuracy, and improve cerebral disease classification mould
The universality of type.
In a first aspect, the embodiment of the invention provides a kind of construction methods of cerebral disease disaggregated model, comprising: from multiple brains
Disease data set obtains Rs-fMRI (the Resting-state functional Magnetic Resonance
Imaging, tranquillization state functional mri) image data；Based on multiple preset brain templates to the Rs-fMRI picture number
According to being pre-processed, average time sequence is obtained；According to the average time sequence construct and each cerebral disease data set and
The corresponding function connects network of the brain template, and feature is extracted from the function connects network, obtain eigenmatrix；Wherein,
The function connects network is fitted with data item, sparse constraint is associated with order constraint item；According to preset sparse learning model,
Regularization term and the eigenmatrix, building multiple initialization cerebral diseases classification moulds corresponding with each cerebral disease data set
Type；Integrated study is carried out to multiple initialization cerebral disease disaggregated models, determines the corresponding mesh of each cerebral disease data set
Mark cerebral disease disaggregated model.
With reference to first aspect, the embodiment of the invention provides the first possible embodiments of first aspect, wherein institute
The step of pre-processing to the Rs-fMRI image data based on multiple preset brain templates, obtain average time sequence is stated,
Include: that processing is corrected to the Rs-fMRI image data, obtains the first image data；Wherein, the correction process includes
Time gradation correction and the dynamic correction of head；Spatial normalization processing is carried out to the first image data, obtains the second image data；
Recurrence processing is carried out to the signal of creating disturbances to of second image data, obtains third image data；Based on multiple preset brain moulds
The third image data is divided into multiple brain areas by plate；Extract the BOLD (Blood-Oxygen-Level of each brain area
Dependent, Blood oxygen level dependence) signal, and the average value of the BOLD signal of each brain area is calculated, obtain the third figure
As the corresponding average time sequence of data.
With reference to first aspect, the embodiment of the invention provides second of possible embodiments of first aspect, wherein institute
State function connects network representation are as follows:
Wherein, A is expressed as the side right value matrix of the function connects network, and T is expressed as the Rs-fMRI image data
Time series,It is expressed as A-TTThe Frobenius norm of T, λ1It is expressed as the first coefficient, λ2It is expressed as the second system
Number,It is expressed as the data fit term, λ1||A||1It is expressed as the sparse constraint, λ2||A||*It is expressed as institute
State order constraint item.
The possible embodiment of second with reference to first aspect, the embodiment of the invention provides the third of first aspect
Possible embodiment, wherein the method also includes: using data fit term described in gradient decline formula optimization, utilization is soft
Sparse constraint described in threshold operation formula optimization, and optimize the order constraint item using operational formula is shunk.
With reference to first aspect, the embodiment of the invention provides the 4th kind of possible embodiments of first aspect, wherein institute
State the expression of cerebral disease disaggregated model are as follows:
Wherein, W is expressed as weight coefficient matrix, and S is expressed as belonging to RN×NSimilarity matrix, M is expressed as the brain template
Number, y is expressed as label vector, XmIt is expressed as the corresponding eigenmatrix of m-th of brain template, wmIt is expressed as described in m-th
The corresponding weight coefficient vector of brain template, γ are expressed as regularization parameter, | | W | |1,1It is expressed as the sum of 1 norm of the row of W, β table
It is shown as non-negative parameter,It is expressed as XmI-th column,It is expressed as XmJth column, SuvThe u row v column of S are expressed as,Table
Adjustment parameter is shown as,It is expressed as the sparse learning model,It is expressed as the regularization term.
With reference to first aspect, the embodiment of the invention provides the 5th kind of possible embodiments of first aspect, wherein institute
State the expression of cerebral disease disaggregated model are as follows:
Wherein,W is expressed as weight coefficient matrix, and S is expressed as belonging to RN ×NSimilarity matrix, Z is expressed as class oriental matrix, and M is expressed as the number of the brain template, and y is expressed as label vector, XmTable
It is shown as the corresponding eigenmatrix of m-th of brain template, wmIt is expressed as the corresponding weight coefficient vector of m-th of brain template,
γ is expressed as regularization parameter, | | W | |1,1It is expressed as the sum of 1 norm of the row of W, Tr (ZTLSZ) it is expressed as ZTLSThe mark of Z, LSTable
It is shown as Laplacian Matrix, μmIt is expressed as the corresponding weight of m-th of brain template, β is expressed as non-negative parameter,It is expressed as Xm
I-th column,It is expressed as XmJth column, SuvThe u row v column of S are expressed as,Adjustment parameter is expressed as,It is expressed as the sparse learning model,It is expressed as the regularization term.
Second aspect, the embodiment of the present invention also provide a kind of construction device of cerebral disease disaggregated model, comprising: data acquisition
Module, for obtaining Rs-fMRI image data from multiple cerebral disease data sets；Preprocessing module, for based on multiple preset
Brain template pre-processes the Rs-fMRI image data, obtains average time sequence；Characteristic extracting module is used for basis
Average time sequence construct function connects network corresponding with each cerebral disease data set and the brain template, and from institute
It states function connects network and extracts feature, obtain eigenmatrix；Wherein, the function connects network be fitted with data item, it is sparse about
Beam item is associated with order constraint item；Model construction module, for according to preset sparse learning model, regularization term and the spy
Levy matrix, building multiple initialization cerebral disease disaggregated models corresponding with each cerebral disease data set；Ensemble classifier module is right
Multiple initialization cerebral disease disaggregated models carry out integration trainingt, determine the corresponding target brain disease of each cerebral disease data set
Sick disaggregated model.
In conjunction with second aspect, the embodiment of the invention provides the first possible embodiments of second aspect, wherein institute
It states preprocessing module to be also used to: processing being corrected to the Rs-fMRI image data, obtains the first image data；Wherein, institute
Stating correction process includes time gradation correction and the dynamic correction of head；Spatial normalization processing is carried out to the first image data, is obtained
To the second image data；Recurrence processing is carried out to the signal of creating disturbances to of second image data, obtains third image data；It is based on
The third image data is divided into multiple brain areas by multiple preset brain templates；The BOLD signal of each brain area is extracted, and
The average value for calculating the BOLD signal of each brain area obtains the corresponding average time sequence of the third image data.
The third aspect, the embodiment of the present invention also provide a kind of intelligent terminal, and the intelligent terminal includes memory and place
Device is managed, the memory is used to store the 5th kind of possible embodiment for supporting processor to execute first aspect to first aspect
The program of any one the method, the processor is configured to for executing the program stored in the memory.
Fourth aspect, the embodiment of the present invention also provide a kind of computer storage medium, for being stored as first aspect to
Computer software instructions used in any one of 5th kind of possible embodiment of one side the method.
The embodiment of the present invention bring it is following the utility model has the advantages that
Construction method, device and the intelligent terminal of a kind of cerebral disease disaggregated model provided in an embodiment of the present invention, first from
Rs-fMRI image data is obtained in multiple cerebral disease data sets, and using multiple preset brain templates to the Rs-fMRI picture number
According to carrying out pretreatment to obtaining average time sequence, then according to average time sequence construct and each cerebral disease data set and in advance
If the corresponding function connects network of brain template (being fitted with data item, sparse constraint is associated with order constraint item), from the function
It connects network and extracts feature, eigenmatrix is obtained, so that root is preset according to sparse learning model, regularization term and this feature square
Battle array constructs multiple initialization cerebral disease disaggregated models corresponding with each cerebral disease data set, to multiple initialization cerebral diseases point
Class model carries out integration trainingt, finally obtains the corresponding target cerebral disease disaggregated model of each cerebral disease data set.The present invention
Based on the Rs-fMRI image data and the obtained average time sequence of multiple preset brain templates in multiple cerebral disease data sets,
It can effectively improve the diversity of average time sequence, and then more complementary characteristic information can be provided；Based on preset
Sparse learning model, regularization term and eigenmatrix building initialization cerebral disease disaggregated model, can effectively improve initialization brain
The learning ability of classification of diseases model, in addition, each institute can be obtained by carrying out integration trainingt to initialization cerebral disease disaggregated model
The corresponding target cerebral disease disaggregated model of cerebral disease data set is stated, can alleviate that data are heterogeneous biggish to ask by the above method
Topic, the comprehensive universality for improving cerebral disease disaggregated model.
Other features and advantages of the present invention will illustrate in the following description, also, partly become from specification
It obtains it is clear that understand through the implementation of the invention.The objectives and other advantages of the invention are in specification, claims
And specifically noted structure is achieved and obtained in attached drawing.
To enable the above objects, features and advantages of the present invention to be clearer and more comprehensible, preferred embodiment is cited below particularly, and cooperate
Appended attached drawing, is described in detail below.
Detailed description of the invention
It, below will be to specific in order to illustrate more clearly of the specific embodiment of the invention or technical solution in the prior art
Embodiment or attached drawing needed to be used in the description of the prior art be briefly described, it should be apparent that, it is described below
Attached drawing is some embodiments of the present invention, for those of ordinary skill in the art, before not making the creative labor
It puts, is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of flow diagram of the construction method of cerebral disease disaggregated model provided in an embodiment of the present invention；
Fig. 2 is the flow diagram of the construction method of another cerebral disease disaggregated model provided in an embodiment of the present invention；
Fig. 3 is the flow diagram of the construction method of another cerebral disease disaggregated model provided in an embodiment of the present invention；
Fig. 4 is a kind of structural schematic diagram of the construction device of cerebral disease disaggregated model provided in an embodiment of the present invention；
Fig. 5 is a kind of structural schematic diagram of intelligent terminal provided in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with embodiment to this hair
Bright technical solution is clearly and completely described, it is clear that and described embodiments are some of the embodiments of the present invention, without
It is whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not before making creative work
Every other embodiment obtained is put, shall fall within the protection scope of the present invention.
The cardinal symptom of ASD shows as social interactions obstacle, communication obstacle, abnormal behavior and interest, and these
Symptom often starts from the Childhood and continues all one's life.According to CDC (Centers for Disease Control and
Prevention, Disease Control and Prevention Center) publication newest report, every about just thering is one to be diagnosed as in 59 children
With ASD, therefore it is badly in need of developing a kind of effective ASD diagnostic method.Currently, having when constructing cerebral disease disaggregated model relatively strong
Limitation, such as based on single brain template building cerebral disease disaggregated model, based on single imaging center building cerebral disease classification
Model constructs cerebral disease disaggregated model, the brain disease obtained by these methods based on the multiple brain templates of single imaging center
Sick disaggregated model has lower universality, so as to cause cerebral disease disaggregated model is being applied to other imaging centers Shi Huicun
In apparent heterogeneity.Wherein, for the method based on single brain template building cerebral disease disaggregated model, it is based on single brain
The feature that template obtains cannot disclose the group difference between ASD patient and normal person (Normal Controls, NC) completely；
For the method based on single imaging center building cerebral disease disaggregated model, since different imaging centers are filled using different imagings
It sets and imaging parameters, and the range of age of subject and sex ratio are different, therefore is based only upon an imaging center building brain disease
Sick disaggregated model, it is unreasonable for then being directly applied to other imaging centers and carrying out the mode of cerebral disease classification.It is another
Data from multiple imaging centers can be combined into the data set for being used for joint training by aspect.Although multicenter data
Data set is extended, but the heterogeneity of different imaging center data is not resolved yet.In addition, inventor also found the prior art
The structural information of data is not furtherd investigate when establishing cerebral disease disaggregated model, the cerebral disease disaggregated model caused
Accuracy it is lower.
Based on this, present invention implementation provides construction method, device and the intelligent terminal of a kind of cerebral disease disaggregated model, can be with
The physiological significance for effectively improving brain function connection network model alleviates the heterogeneous big problem of data, also involvement data structure letter
Breath improves the universality of cerebral disease disaggregated model.
For convenient for understanding the present embodiment, first to a kind of cerebral disease disaggregated model disclosed in the embodiment of the present invention
Construction method describe in detail, a kind of flow diagram of the construction method of cerebral disease disaggregated model shown in Figure 1,
This method may comprise steps of:
Step S102 obtains Rs-fMRI image data from multiple cerebral disease data sets.
Cerebral disease data set namely imaging center are subordinated to ABIDE (Autism Brain Imaging Data
Exchange, self-closing disease Brian Imaging data exchange) database, Rs-fMRI image is stored in each cerebral disease imaging center
Data；Rs-fMRI image data is the brain imaging image obtained based on Rs-fMRI technology.It, can be in a kind of embodiment
4 cerebral disease imaging centers, such as NYU (New York University Langone are chosen in ABIDE database
Medical Center) imaging center, UM_1 (University of Michigan:Sample 1) imaging center, UCLA_1
(University of California, Los Angeles:Sample 1) imaging center and YALE (Yale Child
Study Center) imaging center, obtain the Rs-fMRI image data of all subjects in above-mentioned 4 imaging centers；Wherein,
NYU imaging center includes 171 subjects (35 people of 136 people of male and women), and UM_1 imaging center includes 82 subjects (male
Property 59 people and women 23 people), UCLA_1 imaging center includes 55 subjects (6 people of 49 people of male and women), in YALE imaging
Pericardium includes 48 subjects (14 people of 34 people of male and women).
Step S104 pre-processes Rs-fMRI image data based on multiple preset brain templates, obtains average time
Sequence.
Wherein, brain template can be AAL (Anatomical Automatic Labeling, automatic anatomical landmarks) template,
Dos 160 (Dosenbach 160) template, CC 200 (Craddock 200) template are used for above-mentioned each Rs-fMRI picture number
According to being divided into several brain areas.In one embodiment, it can use DPARSF (the Data Processing
Assistant for Resting-State fMRI, rs-fMRI data processing assistant) Rs-fMRI image data is carried out in advance
Processing, wherein pretreatment may include correction process, Spatial normalization processing, recurrence processing, and brain area divides processing and signal
Extraction process, to obtain average time sequence corresponding with multiple brain templates and Rs-fMRI image data.
Step S106, according to average time sequence construct function connects net corresponding with each cerebral disease data set and brain template
Network, and feature is extracted from function connects network, obtain eigenmatrix.
Function connects (Functional Connectivity, FC) network is fitted with data item, sparse constraint and order about
Beam Xiang Guanlian；Wherein, for the side right value matrix in FC network, sparse constraint is used to assess the degree of rarefication of side right value matrix,
Order constraint item is used to indicate the order of side right value matrix.It in the specific implementation, can be with by adjusting sparse constraint and order constraint item
The modular construction of brain network is realized, so that the function connects network made has stronger physiological significance.
Step S108, according to sparse learning model, regularization term and eigenmatrix, building and each cerebral disease data set pair
The multiple initialization cerebral disease disaggregated models answered.
Since each eigenmatrix is obtained based on different brain templates, in order to identify Rs-fMRI picture number
According to the feature for being directed to each brain template most discrimination property, the present invention uses preset sparse learning model.Although furthermore, it is contemplated that
Sparse learning model remains the correlation between label and Rs-fMRI image data, but without retaining Rs-fMRI figure
As the interdependency between data, therefore the present invention is by introducing regularization term on the basis of sparse learning model, thus
Retain the interdependency between Rs-fMRI image data, obtains multiple initialization brains corresponding with each cerebral disease data set
Classification of diseases model, and the initialization cerebral disease disaggregated model has the advantages that weight multi-template automatically and neighborhood is adaptive.
Step S110 carries out integrated study to multiple initialization cerebral disease disaggregated models, determines each cerebral disease data set pair
The target cerebral disease disaggregated model answered.
In order to preferably distinguish property feature, sheet using the multiple groups selected in each cerebral disease imaging center and each brain template
Invention uses multi-template multicenter Integrated Strategy, i.e., carries out integration trainingt to multiple initialization cerebral disease disaggregated models, obtains each
The corresponding optimal target cerebral disease disaggregated model of cerebral disease imaging center.
The construction method of a kind of cerebral disease disaggregated model provided in an embodiment of the present invention, first from multiple cerebral disease data sets
Middle acquisition Rs-fMRI image data, and using multiple preset brain templates to the Rs-fMRI image data carry out pretreatment to
Average time sequence is obtained, it is then corresponding with each cerebral disease data set and preset brain template according to average time sequence construct
Function connects network (is fitted with data item, sparse constraint is associated with order constraint item), extracts feature from the function connects network,
Eigenmatrix is obtained, thus according to the building of sparse learning model, regularization term and this feature matrix and each cerebral disease data set
Corresponding multiple initialization cerebral disease disaggregated models finally carry out integration trainingt to multiple initialization cerebral disease disaggregated models, most
The corresponding target cerebral disease disaggregated model of each cerebral disease data set is obtained eventually.The present invention is based in multiple cerebral disease data sets
The average time sequence that Rs-fMRI image data and multiple preset brain templates obtain, can effectively improve average time sequence
The diversity of value, and then more complementary characteristic information can be provided；Based on sparse learning model, regularization term and feature square
Battle array building initialization cerebral disease disaggregated model, can effectively improve the learning ability of initialization cerebral disease disaggregated model, in addition, logical
It crosses and integration trainingt is carried out to initialization cerebral disease disaggregated model, the corresponding target cerebral disease of each cerebral disease data set can be obtained
Disaggregated model can alleviate heterogeneity present in cerebral disease data by the above method, comprehensive to improve cerebral disease classification
The universality of model.
To understand convenient for the construction method of the cerebral disease disaggregated model provided previous embodiment, the embodiment of the present invention
Additionally provide the construction method of another cerebral disease disaggregated model, the structure of another cerebral disease disaggregated model shown in Figure 2
The flow diagram of construction method, this method may comprise steps of:
Step S202 obtains Rs-fMRI image data from multiple cerebral disease data sets.
In the specific implementation, every 2 to 3 seconds run-downs, and a volume (i.e. time series) is automatically generated, obtained
Rs-fMRI image data, at this point, needing to abandon the Rs- of each subject in order to which the intensity of magnetization of retention time sequence is equal
Preceding ten time serieses of fMRI image data.
Step S204 is corrected processing to Rs-fMRI image data, obtains the first image data.
Wherein, correction process includes time gradation correction (slice timing correction) and the dynamic correction (head of head
motion correction)。
Step S206 carries out Spatial normalization processing to the first image data, obtains the second image data.
In one embodiment, all first image datas can be normalized to MNI (the Montreal
Neurological Institute, standardize Montreal neurology research institute) space, wherein the resolution ratio in the space MNI be
3x 3x 3mm3。
Step S208 carries out recurrence processing to the signal of creating disturbances to of the second image data, obtains third image data.
Mix change due to caused by the drift of physiology course (heartbeat and breathing), head movement and low-frequency sweep instrument to remove
Change (confounding variation), the present invention also carries out recurrence processing to the signal of creating disturbances to of above-mentioned second image data
(nuisance variable regression), wherein creating disturbances to signal may include white matter signal, cerebrospinal fluid signal and full brain
Average signal etc..
Step S210 is based on multiple preset brain templates, third image data is divided into multiple brain areas.
By taking the default brain template different based on three as an example, AAL template by third image data be divided into 116 brain areas,
Third image data is divided into 160 brain areas, 200 template of CC and third image data is divided into 200 by 160 template of Dos
Brain area, wherein brain area namely ROI (regions of interest, area-of-interest).
It, can also be to institute in each brain area after obtaining the corresponding multiple brain areas of third image in a kind of implementation normal form
The time series obtained carries out bandpass filtering, removes high-frequency signal and low frequency signal.
Step S212, extracts the BOLD signal of each brain area, and calculates the average value of the BOLD signal of each brain area, obtains third
The corresponding average time sequence of image data.
Step S214, according to average time sequence construct function connects net corresponding with each cerebral disease data set and brain template
Network, and feature is extracted from function connects network, obtain eigenmatrix.
For ease of understanding, the embodiment of the invention also provides a kind of construction methods of FC network.Wherein, it is utilized in the present invention
Bold capital letter representing matrix, bold case lower case letters indicate that vector, common tilted letter indicate scalar.Base of the embodiment of the present invention
In PC (Pearson ' s correlation, Pearson correlation coefficients), it is method the simplest when assessing FC network,
It is defined as formula (1):
Wherein, apqThe FC being expressed as between p-th of ROI and q-th of ROI, uses tpIndicate the time series of p-th of ROI, then
The time series of entire brain can be expressed as T=[t1,t2,...,tr]∈Rt×r, vectorIndicate time series tpMean value
Vector.
To tpCentralization is carried out to obtainIt is right againFurther progress normalizes to obtainThe transposition operation of T representing matrix, above-mentioned formula (1) is reduced to a at this timepq=tp TtqOr A=
TTT, and A=TTT is the solution of following regression formulas (2):
Wherein, A indicates the side right value matrix of FC network.
Formula (2) is equivalent to following matrix form:
Since brain is distributed in modularization, the application binding modules priori construction one on the basis of formula (3)
A new FC network, is defined as:
Wherein, A is expressed as the side right value matrix of function connects network, and T is expressed as the time of single Rs-fMRI image data
Sequence,It is expressed as A-TTThe Frobenius norm of T, λ1It is expressed as the first coefficient, λ2It is expressed as the second coefficient,It is expressed as data fit term, for being fitted input data, λ1||A||1It is expressed as sparse constraint, for assessing
The degree of rarefication of the side right value matrix of FC network, rank (A) indicate order constraint item, for indicating the side right value matrix of FC network
Order.By adjusting the first parameter lambda1With the second parameter lambda2The modular construction of FC network may be implemented, however in view of rank function is
Non-convex function, therefore rank function is relaxed as trace norm | | A | |*, function connects network representation at this time are as follows:
Wherein, formula (5) has compared with prior art there are two advantage: (1) with existing based on SR (Sparse
Representation, rarefaction representation) method compare, statistically have stronger robustness because SR is related to association side
Poor inverse of a matrix operation, there are ill-posed problems for this, especially when time series dimension is less than ROI number.And the present invention mentions
The formula (5) of confession can export result without inverse operation directly from covariance matrix, therefore robustness is stronger；(2) formula
(5) prior information (that is, FC network has modular construction) is also introduced, the FC network of building is made to have more physiological significance.
Further, the embodiment of the invention also provides a kind of optimization methods of FC network, although formula (5) is convex function
But non-differentiability, therefore formula (5) is handled using proximal end method (proximal method), when it is implemented, utilizing gradient
Decline formula (6) and optimize data fit term, optimizes sparse constraint using soft-threshold operational formula (7), and utilize contraction operation
Formula (8) optimized rank bound term, optimization is sequentially iterated according to this, until convergence.Specifically, considering data fit term firstThe data fit term can be micro-, and derivative is expressed as 2 (A-TTT), therefore gradient decline formula (6) is obtained:
Ak=Ak-1-θkf(T,Ak-1) (6)
Wherein, k indicates iteration index, θkIt identifies gradient and declines step-length.
Then, according to the definition for closing on operator (proximal operator), λ1||A||1Close on operator and can indicate
For soft-threshold operational formula (7):
proxλ1||·||1(A)=[sgn (apq)×max{|apq|-λ1,0}]r×r (7)
λ2||A||*The contraction operational formula (8) for closing on the singular value that operator can be expressed as about A:
proxλ2||·||*(A)=Udiag (max { σ1-λ2,0},...,{σr-λ2,0})VT (8)
Wherein, Udiag (σ1,...,σr)VTIt is expressed as the singular value decomposition to A.
In the specific implementation, each element in A can be considered as a feature.Since matrix A is symmetrically, only
The feature of triangular portions is needed, and is a vector, the side obtained to each Rs-fMRI image data by the remodeling of these features
Weight matrix is performed both by the operation, to obtain the eigenmatrix in corresponding different templates space.
Step S216, according to sparse learning model, regularization term and eigenmatrix, building and each cerebral disease data set pair
The multiple initialization cerebral disease disaggregated models answered.
For ease of understanding, the embodiment of the invention also provides a kind of construction methods of cerebral disease disaggregated model.Assuming that there are M
A brain template, then useIndicate the corresponding eigenmatrix of m-th of preset brain template, wherein
N is Rs-fMRI image data number (i.e. number of samples),Indicate i-th of Rs-fMRI image of m-th of preset brain template
The corresponding feature vector of data, d are expressed as characteristic dimension.Obviously, from same Rs-fMRI image data based on different brain moulds
The feature of plate class label having the same.In addition, present invention y=[y1,y2,...yn]T∈RNIndicate corresponding label vector,
yi∈ { -1 ,+1 } indicates the class label of i-th of Rs-fMRI image data.Since each eigenmatrix is by different brain moulds
What plate obtained, therefore the eigenmatrix in order to identify most discrimination property, the present invention use the sparse learning model of linear multitask,
As shown in formula (9):
Wherein, W is expressed as weight coefficient matrix, and M is expressed as the number of preset brain template, and y is expressed as label vector, Xm
It is expressed as the corresponding eigenmatrix of m-th of preset brain template, wmBe expressed as the corresponding weight coefficient of m-th of default brain template to
Amount, γ are expressed as third parameter, | | W | |1,1It is expressed as 1 norm (namely the l of the row of W1Norm) sum, control matrix W in certain
A little elements are 0, help to identify the distinctive feature specific to different brain templatespaces.
In view of in the prior art, the process of similar matrix construction and feature selecting be it is separated, lead to similarity matrix
It always generates from the primitive character comprising much noise and remains unchanged in the follow-up process, this can reduce similarity matrix
Accuracy to further destroy local manifolds structure；In addition, by conventional method obtain similarity matrix there is also only
There is the problem of connected component.Therefore, the present invention is by optimizing similarity matrix.In one embodiment,
Similarity matrix, S are indicated using SuvIt is expressed as the u row v column of S.Present invention assumes that eachWith probability SuvWith it is every other
Rs-fMRI image data is connected, which is considered as the similitude size between Rs-fMRI image data.Due to Rs-fMRI
The distance between image data is inversely proportional with similitude size, therefore the present invention obtains similarity matrix using following formula (10)
S:
Wherein, S is expressed as belonging to RN×NSimilarity matrix,It is expressed as adjustment parameter.For avoiding trifling solution.
However, only including a connected component by the similarity matrix that formula (10) obtain.According to Laplacian Matrix
Property, as rank (LSWhen)=n-c, similar matrix has c connected component, and c is the other quantity of tag class, LSFor Laplce
Matrix, and LS=D- (ST+ S)/2), D is corresponding rank matrix, D (duu=∑v(suv+svu)/2).Therefore, the present invention by order about
Beam is added in formula (10), obtains following formula (11):
Formula (11) is adaptively each data point distribution neighborhood, therefore S is updated always, until it includes suitable number
The connected component of amount.Further, in the case of above-mentioned formula (11) being expanded to multi-template, it is shared that multiple brain template datas are obtained
Similarity matrix, as shown in formula (12):
μ=[μ1, μ2..., μm] indicate the corresponding weight vectors of each brain template,It indicates non-negative scalar, makes weight point
Cloth is smoothened,It is expressed as XmI-th column,It is expressed as XmJth column.
If the feature of two Rs-fMRI image datas is closely similar, the distance between they are also answered in Label space
The very little.In above-mentioned sparse learning model (formula 9), we pass through linear mapping function f (x)=xTW is by feature from initial
High-dimensional feature space is mapped to one-dimensional response space.Although it remain label and sample (that is, Rs-fMRI image data) it
Between correlation, but these interdependencies between sample are ignored.In order to solve this problem and model is improved
Performance needs to introduce a new Laplace regularization item on the basis of formula (9), wherein new Laplce's canonical
Change item to obtain based on above-mentioned formula (12), and new Laplace regularization item is expressed as formula (13):
Above-mentioned regularization term can keep the local proximity structure of the same category data in projection.
Cerebral disease categorization module is obtained at this time.And cerebral disease disaggregated model is expressed as formula (14):
Wherein, β is expressed as non-negative parameter
In order to reduce number of parameters, the present invention provides a kind of new automatic Weighted adaptive neighbor learning based on multi-template
Cerebral disease disaggregated model is also defined as formula (15) by method:
Because of rank (LS)=n-c relies on S, it is therefore desirable to optimize to formula (15).Utilize σi(LS) indicate LS?
I the smallest characteristic values, according to LSPositive semidefinite property, available σi(LS) >=0, so rank (LS)=n-c is equivalent toAccording to Ky Fan ' s theorem, formula (16) are obtained:
Wherein, Z=[z1,z2,...,zN] it is class oriental matrix, therefore above-mentioned formula (15) can be converted to formula (17):
When γ is sufficiently large, Tr (ZTLSZ it) will be close to zero, to obtain
For the ease of optimizing and finding the optimal solution of variable W, S and Z, above-mentioned formula (17) conversion is as follows formula
(18) and formula (19):
In one embodiment, cerebral disease disaggregated model can be optimized using the method for iteration.The embodiment of the present invention is also
A kind of optimization method of cerebral disease disaggregated model is provided, referring to following steps a-c:
Step a, when above-mentioned S, Z and it is μm fixed when, W is updated using formula (20).Wherein, the following institute of formula (20)
Show:
Specifically, above-mentioned formula (18) is converted to formula (21) first:
According to an equation in spectrum analysis (spectral analysis)Above-mentioned formula (21) equivalence is converted into above-mentioned formula (20).
Step b, when above-mentioned S and W are fixed, using formula (19) to μm be updated and using formula (22) to Z into
Row updates, wherein formula (22) is as follows:
Wherein, the optimal solution of Z is by LSC minimal eigenvalue corresponding feature vector composition.
Step c, when above-mentioned W, Z and it is μm fixed when, S is updated using formula (23), wherein the following institute of formula (23)
Show:
Specifically, the embodiment of the invention provides a kind of acquisition process of formula (23), above-mentioned formula (18) are turned first
It is changed to formula as follows (24):
To obtain convenient for calculatingWithIt at this time will be above-mentioned
Formula (24) conversion is as follows formula (25):
Since the similarity vectors of each sample are independent, available formula (26) as follows:
Further, it utilizesIndicate vector di∈Rd, and then formula (26) is equivalent to above-mentioned formula
(23).The value of parameter alpha can determine that adaptive neighborhood means each data neighborhood of a point by the quantity of adaptive neighborhood
It is unfixed, because of the d between every two data point in each iterationijIt is to constantly update, so adaptive neighborhood is
It constantly updates.It observes formula (18), when disaggregated model algorithmic statement, the form of Laplace regularization item be can be regarded as
The popular linear combination of multiple brain templates, μmIt is corresponding weight.Because formula (18) is originated from formula (15), we can be true
Recognizing it is entirely the process weighted automatically.Moreover, according to formula (18), if m-th of brain template is more suitable for diagnosing, phase
Answer weight mumIt will be bigger.In contrast, weaker template will be assigned lesser weight.This means that automatic weighting of the invention
Multi-template learning model is effective.
Step S218 carries out integrated study to multiple initialization cerebral disease disaggregated models, determines each cerebral disease data set pair
The target cerebral disease disaggregated model answered.
In order to preferably utilize the multiple groups selected in each brain template of each imaging center of correspondence discrimination property feature, the present invention
Further use multi-template multicenter Integrated Strategy.Specific step is as follows, 1) instruction of corresponding each imaging center disaggregated model
Practice: learning (ASL, adaptive structural using adaptive structure on the training data of each imaging center
Learning after) method is trained, a corresponding weight coefficient matrix W can be respectively obtained, the matrix is by corresponding to M
The weight coefficient vector of brain template forms.2) single centre multi-template integrated study: the weight obtained for some imaging center
Coefficient matrix W, we obtain M character subset for corresponding to M brain template.Then, pass through linear function y=xmwmIt opens respectively
M classifier is sent out, and balances the output of M different classifications device using majority vote rule, with the class for new test sample
Distinguishing label.3) the classification mould to multiple centers multicenter multi-template Ensemble classifier: is carried out on the basis of multi-template integrated study
Type carries out majority vote rule, to obtain the target cerebral disease disaggregated model of each imaging center.
For the ease of understanding method provided by the above embodiment, the embodiment of the invention also provides another brain diseases
The creation method of sick disaggregated model, firstly, being that each sample of each imaging center constructs based on different preset brain templates
Multiple FC networks, and extract feature from these FC networks to generate multiple eigenmatrixes, wherein the quantity of eigenmatrix and pre-
If brain template number be identical；Secondly, corresponding to the eigenmatrix of multiple brain templates based on each imaging center, ASL is utilized
Method creates multiple cerebral disease disaggregated models, and determines that obtaining final each cerebral disease data set corresponds to using integrated study
Target cerebral disease disaggregated model, and ASD is diagnosed using target cerebral disease disaggregated model.Shown in Figure 3 is another
The flow chart of the construction method of kind cerebral disease model, it is assumed that there are four imaging centers, center 1, center 2, center 3 and center 4,
Three preset brain templates, template 1, template 2 and template 3 are existed simultaneously, and these three brain templates and each imaging center are right
It answers, and the corresponding FC network in each center is obtained based on above three template, by taking center 1 as an example, center 1 is incited somebody to action at this time
To and the corresponding and corresponding FC network 2 of template 2 and FC network 3 corresponding with template 3 of FC network 1 of template 1, and then obtain with
The corresponding eigenmatrix 1 of the template 1 and corresponding eigenmatrix 2 of template 2 and eigenmatrix 3 corresponding with template 3, utilizes these
Eigenmatrix obtains multiple initialization cerebral disease disaggregated models corresponding with center 1 by ASL method.What final utilization integrated
Method obtains target cerebral disease disaggregated model to the end, determines the class label of each sample in each center.
In conclusion the embodiment of the present invention can achieve at least one following features:
(1) a kind of network establishing method is proposed, the FC network of multiple and different sizes is generated for each sample.The FC network
Based on the building of sparse and low-rank regularization constraint, the modular construction of brain is encoded.
(2) the cerebral disease disaggregated model based on multi-template data is proposed.Based on multi-task learning frame, can introduce every
Structural information present in a brain template data, so that selection corresponds to the feature of the most discrimination of each templatespace.
(3) integrated approach is further used to the corresponding cerebral disease disaggregated model of each imaging center, makes ASD diagnostic result
It is more acurrate.
For the construction method for the cerebral disease disaggregated model that previous embodiment provides, the embodiment of the invention also provides one kind
The construction device of cerebral disease disaggregated model, a kind of structural representation of the construction device of cerebral disease disaggregated model shown in Figure 4
Figure, the apparatus may include following parts:
Data acquisition module 402, for obtaining Rs-fMRI image data from multiple cerebral disease data sets；
Preprocessing module 404 is obtained for being pre-processed based on multiple preset brain templates to Rs-fMRI image data
To average time sequence；
Characteristic extracting module 406, for according to average time sequence construct and each cerebral disease data set and preset brain mould
The corresponding function connects network of plate, and feature is extracted from function connects network, obtain eigenmatrix；Wherein, function connects network
It is fitted with data item, sparse constraint is associated with order constraint item；
Model construction module 408, for according to preset sparse learning model, regularization term and eigenmatrix, building
Multiple initialization cerebral disease disaggregated models corresponding with each cerebral disease data set；
Ensemble classifier module 410 determines each brain disease for carrying out integrated study to multiple initialization cerebral disease disaggregated models
The corresponding target cerebral disease disaggregated model of sick data set.
The present invention is based in multiple cerebral disease data sets Rs-fMRI image data and multiple preset brain templates obtain
Average time sequence can effectively improve the diversity of average time sequential value, and then can provide more complementary feature
Information；Initialization cerebral disease disaggregated model is constructed based on sparse learning model, regularization term and eigenmatrix, can effectively improve
The learning ability of cerebral disease disaggregated model is initialized, in addition, by carrying out integrated study to initialization cerebral disease disaggregated model, it can
The target cerebral disease disaggregated model for being most suitable for cerebral disease data set is obtained, cerebral disease data can be effectively relieved by the above method
Existing heterogeneity, the comprehensive universality for improving cerebral disease disaggregated model.
In one embodiment, above-mentioned preprocessing module is also used to: processing is corrected to Rs-fMRI image data,
Obtain the first image data；Wherein, correction process includes time gradation correction and the dynamic correction of head；First image data is carried out empty
Between standardization, obtain the second image data；Recurrence processing is carried out to the signal of creating disturbances to of the second image data, obtains third figure
As data；Based on multiple preset brain templates, third image data is divided into multiple brain areas；Extract the BOLD letter of each brain area
Number, and the average value of the BOLD signal of each brain area is calculated, obtain the corresponding average time sequence of third image data.
The technical effect and preceding method embodiment phase of device provided by the embodiment of the present invention, realization principle and generation
Together, to briefly describe, Installation practice part does not refer to place, can refer to corresponding contents in preceding method embodiment.
The equipment is a kind of intelligent terminal, specifically, the intelligent terminal includes processor and storage device；On storage device
It is stored with computer program, computer program executes any one institute of embodiment as described above when being run by the processor
The method stated.
Fig. 5 is a kind of structural schematic diagram of intelligent terminal provided in an embodiment of the present invention, which includes: place
Device 50 is managed, memory 51, bus 52 and communication interface 53, the processor 50, communication interface 53 and memory 51 pass through bus
52 connections；Processor 50 is for executing the executable module stored in memory 51, such as computer program.
Wherein, memory 51 may include high-speed random access memory (RAM, Random Access Memory),
It may further include non-labile memory (non-volatile memory), for example, at least a magnetic disk storage.By extremely
A few communication interface 53 (can be wired or wireless) is realized logical between the system network element and at least one other network element
Letter connection, can be used internet, wide area network, local network, Metropolitan Area Network (MAN) etc..
Bus 52 can be isa bus, pci bus or eisa bus etc..The bus can be divided into address bus, data
Bus, control bus etc..Only to be indicated with a four-headed arrow convenient for indicating, in Fig. 5, it is not intended that an only bus or
A type of bus.
Wherein, memory 51 is for storing program, and the processor 50 executes the journey after receiving and executing instruction
Sequence, method performed by the device that the stream process that aforementioned any embodiment of the embodiment of the present invention discloses defines can be applied to handle
In device 50, or realized by processor 50.
Processor 50 may be a kind of IC chip, the processing capacity with signal.During realization, above-mentioned side
Each step of method can be completed by the integrated logic circuit of the hardware in processor 50 or the instruction of software form.Above-mentioned
Processor 50 can be general processor, including central processing unit (Central Processing Unit, abbreviation CPU), network
Processor (Network Processor, abbreviation NP) etc.；It can also be digital signal processor (Digital Signal
Processing, abbreviation DSP), specific integrated circuit (Application Specific Integrated Circuit, referred to as
ASIC), ready-made programmable gate array (Field-Programmable Gate Array, abbreviation FPGA) or other are programmable
Logical device, discrete gate or transistor logic, discrete hardware components.It may be implemented or execute in the embodiment of the present invention
Disclosed each method, step and logic diagram.General processor can be microprocessor or the processor is also possible to appoint
What conventional processor etc..The step of method in conjunction with disclosed in the embodiment of the present invention, can be embodied directly in hardware decoding processing
Device executes completion, or in decoding processor hardware and software module combination execute completion.Software module can be located at
Machine memory, flash memory, read-only memory, programmable read only memory or electrically erasable programmable memory, register etc. are originally
In the storage medium of field maturation.The storage medium is located at memory 51, and processor 50 reads the information in memory 51, in conjunction with
Its hardware completes the step of above method.
The computer program product of readable storage medium storing program for executing provided by the embodiment of the present invention, including storing program code
Computer readable storage medium, the instruction that said program code includes can be used for executing previous methods side as described in the examples
Method, specific implementation can be found in preceding method embodiment, and details are not described herein.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) it performs all or part of the steps of the method described in the various embodiments of the present invention.
And storage medium above-mentioned includes: that USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
The various media that can store program code such as reservoir (RAM, Random Access Memory), magnetic or disk.
Finally, it should be noted that embodiment described above, only a specific embodiment of the invention, to illustrate the present invention
Technical solution, rather than its limitations, scope of protection of the present invention is not limited thereto, although with reference to the foregoing embodiments to this hair
It is bright to be described in detail, those skilled in the art should understand that: anyone skilled in the art
In the technical scope disclosed by the present invention, it can still modify to technical solution documented by previous embodiment or can be light
It is readily conceivable that variation or equivalent replacement of some of the technical features；And these modifications, variation or replacement, do not make
The essence of corresponding technical solution is detached from the spirit and scope of technical solution of the embodiment of the present invention, should all cover in protection of the invention
Within the scope of.Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. a kind of construction method of cerebral disease disaggregated model characterized by comprising
Rs-fMRI image data is obtained from multiple cerebral disease data sets；
The Rs-fMRI image data is pre-processed based on multiple preset brain templates, obtains average time sequence；
According to average time sequence construct function connects net corresponding with each cerebral disease data set and the brain template
Network, and feature is extracted from the function connects network, obtain eigenmatrix；Wherein, the function connects network is fitted with data
Item, sparse constraint are associated with order constraint item；
According to preset sparse learning model, regularization term and the eigenmatrix, building and each cerebral disease data set
Corresponding multiple initialization cerebral disease disaggregated models；
Integrated study is carried out to multiple initialization cerebral disease disaggregated models, determines the corresponding mesh of each cerebral disease data set
Mark cerebral disease disaggregated model.
2. the method according to claim 1, wherein described be based on multiple preset brain templates to the Rs-
The step of fMRI image data is pre-processed, and average time sequence is obtained, comprising:
Processing is corrected to the Rs-fMRI image data, obtains the first image data；Wherein, the correction process includes
Time gradation correction and the dynamic correction of head；
Spatial normalization processing is carried out to the first image data, obtains the second image data；
Recurrence processing is carried out to the signal of creating disturbances to of second image data, obtains third image data；
Based on multiple preset brain templates, the third image data is divided into multiple brain areas；
The BOLD signal of each brain area is extracted, and calculates the average value of the BOLD signal of each brain area, obtains the third
The corresponding average time sequence of image data.
3. the method according to claim 1, wherein the function connects network representation are as follows:
Wherein, A is expressed as the side right value matrix of the function connects network, and T is expressed as the time of the Rs-fMRI image data
Sequence,It is expressed as A-TTThe Frobenius norm of T, λ1It is expressed as the first coefficient, λ2It is expressed as the second coefficient,It is expressed as the data fit term, λ1||A||1It is expressed as the sparse constraint, λ2||A||*It is expressed as described
Order constraint item.
4. according to the method described in claim 3, it is characterized in that, the method also includes:
Using data fit term described in gradient decline formula optimization, optimize the sparse constraint using soft-threshold operational formula,
And optimize the order constraint item using operational formula is shunk.
5. the method according to claim 1, wherein the cerebral disease disaggregated model indicates are as follows:
Wherein, W is expressed as weight coefficient matrix, and S is expressed as belonging to RN×NSimilarity matrix, M be expressed as the brain template
Number, y are expressed as label vector, XmIt is expressed as the corresponding eigenmatrix of m-th of brain template, wmIt is expressed as m-th of brain mould
The corresponding weight coefficient vector of plate, γ are expressed as regularization parameter, | | W | |1,1It is expressed as the sum of 1 norm of the row of W, β is expressed as
Non-negative parameter,It is expressed as XmI-th column,It is expressed as XmJth column, SuvThe u row v column of S are expressed as,It is expressed as
Adjustment parameter,It is expressed as the sparse learning model,It is expressed as the regularization term.
6. the method according to claim 1, wherein the cerebral disease disaggregated model indicates are as follows:
Wherein,
W is expressed as weight coefficient matrix, and S is expressed as belonging to RN×NSimilarity matrix, Z is expressed as class oriental matrix, and M is expressed as
The number of the brain template, y are expressed as label vector, XmIt is expressed as the corresponding eigenmatrix of m-th of brain template, wmIt indicates
For the corresponding weight coefficient vector of m-th of brain template, γ is expressed as regularization parameter, | | W | |1,1It is expressed as the 1 of the row of W
The sum of norm, Tr (ZTLSZ) it is expressed as ZTLSThe mark of Z, LSIt is expressed as Laplacian Matrix, μmIt is expressed as m-th of brain template
Corresponding weight, β are expressed as non-negative parameter,It is expressed as XmI-th column,It is expressed as XmJth column, SuvIt is expressed as S's
U row v column,Adjustment parameter is expressed as,It is expressed as the sparse study
Model,It is expressed as the regularization term.
7. a kind of construction device of cerebral disease disaggregated model characterized by comprising
Data acquisition module, for obtaining Rs-fMRI image data from multiple cerebral disease data sets；
Preprocessing module is put down for being pre-processed based on multiple preset brain templates to the Rs-fMRI image data
Equal time series；
Characteristic extracting module, for according to the average time sequence construct and each cerebral disease data set and the brain template
Corresponding function connects network, and feature is extracted from the function connects network, obtain eigenmatrix；Wherein, the function connects
Connect that network is fitted with data item, sparse constraint is associated with order constraint item；
Model construction module, for constructing and each according to preset sparse learning model, regularization term and the eigenmatrix
The corresponding multiple initialization cerebral disease disaggregated models of the cerebral disease data set；
Ensemble classifier module determines each brain for carrying out integrated study to multiple initialization cerebral disease disaggregated models
The corresponding target cerebral disease disaggregated model of disease data set.
8. device according to claim 7, which is characterized in that the preprocessing module is also used to:
Processing is corrected to the Rs-fMRI image data, obtains the first image data；Wherein, the correction process includes
Time gradation correction and the dynamic correction of head；
Spatial normalization processing is carried out to the first image data, obtains the second image data；
Recurrence processing is carried out to the signal of creating disturbances to of second image data, obtains third image data；
Based on multiple preset brain templates, the third image data is divided into multiple brain areas；
The BOLD signal of each brain area is extracted, and calculates the average value of the BOLD signal of each brain area, obtains the third
The corresponding average time sequence of image data.
9. a kind of intelligent terminal, which is characterized in that the intelligent terminal includes memory and processor, and the memory is used for
Storage supports processor perform claim to require the program of any one of 1 to 6 the method, the processor is configured to for executing
The program stored in the memory.
10. a kind of computer storage medium, which is characterized in that for being stored as used in any one of claim 1 to 6 the method
Computer software instructions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910628897.5A CN110263880B (en) | 2019-07-12 | 2019-07-12 | Method and device for constructing brain disease classification model and intelligent terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910628897.5A CN110263880B (en) | 2019-07-12 | 2019-07-12 | Method and device for constructing brain disease classification model and intelligent terminal |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110263880A true CN110263880A (en) | 2019-09-20 |
| CN110263880B CN110263880B (en) | 2021-08-03 |
Family
ID=67925905
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910628897.5A Active CN110263880B (en) | 2019-07-12 | 2019-07-12 | Method and device for constructing brain disease classification model and intelligent terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110263880B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110956610A (en) * | 2019-10-31 | 2020-04-03 | 广州市番禺区中心医院(广州市番禺区人民医院、广州市番禺区心血管疾病研究所) | Method and system for predicting thrombolytic curative effect of lower limb deep venous thrombosis based on sparse representation |
| CN111012337A (en) * | 2019-12-13 | 2020-04-17 | 燕山大学 | Brain network and regularized discriminant analysis-based electroencephalogram analysis method |
| CN111062420A (en) * | 2019-11-26 | 2020-04-24 | 深圳大学 | Brain structure feature selection method, mobile terminal and computer-readable storage medium |
| CN112037914A (en) * | 2020-08-11 | 2020-12-04 | 深圳大学 | Construction method, system and equipment of obsessive-compulsive disorder risk assessment model |
| CN112289412A (en) * | 2020-10-09 | 2021-01-29 | 深圳市儿童医院 | Construction method of autism spectrum disorder classifier, device thereof and electronic equipment |
| CN113450426A (en) * | 2021-06-09 | 2021-09-28 | 深圳市铱硙医疗科技有限公司 | Magnetic resonance cerebral perfusion imaging data processing system, method, terminal and medium |
| CN114190884A (en) * | 2020-09-18 | 2022-03-18 | 深圳大学 | Longitudinal analysis method, system and device for brain disease data |
| CN117058471A (en) * | 2023-10-12 | 2023-11-14 | 之江实验室 | Disease brain image parting system based on normal brain image database |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108257657A (en) * | 2016-12-28 | 2018-07-06 | 复旦大学附属华山医院 | The data analysing method of magnetic resonance detection based on the prediction of disturbance of consciousness patient consciousness recovery |
| CN109344889A (en) * | 2018-09-19 | 2019-02-15 | 深圳大学 | A kind of cerebral disease classification method, device and user terminal |
| WO2019084697A1 (en) * | 2017-11-06 | 2019-05-09 | University Health Network | Platform, device and process for annotation and classification of tissue specimens using convolutional neural network |
| CN109770932A (en) * | 2019-02-21 | 2019-05-21 | 河北工业大学 | The processing method of multi-modal brain neuroblastoma image feature |
| CN109948740A (en) * | 2019-04-26 | 2019-06-28 | 中南大学湘雅医院 | A kind of classification method based on tranquillization state brain image |
-
2019
- 2019-07-12 CN CN201910628897.5A patent/CN110263880B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108257657A (en) * | 2016-12-28 | 2018-07-06 | 复旦大学附属华山医院 | The data analysing method of magnetic resonance detection based on the prediction of disturbance of consciousness patient consciousness recovery |
| WO2019084697A1 (en) * | 2017-11-06 | 2019-05-09 | University Health Network | Platform, device and process for annotation and classification of tissue specimens using convolutional neural network |
| CN109344889A (en) * | 2018-09-19 | 2019-02-15 | 深圳大学 | A kind of cerebral disease classification method, device and user terminal |
| CN109770932A (en) * | 2019-02-21 | 2019-05-21 | 河北工业大学 | The processing method of multi-modal brain neuroblastoma image feature |
| CN109948740A (en) * | 2019-04-26 | 2019-06-28 | 中南大学湘雅医院 | A kind of classification method based on tranquillization state brain image |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110956610A (en) * | 2019-10-31 | 2020-04-03 | 广州市番禺区中心医院(广州市番禺区人民医院、广州市番禺区心血管疾病研究所) | Method and system for predicting thrombolytic curative effect of lower limb deep venous thrombosis based on sparse representation |
| CN111062420A (en) * | 2019-11-26 | 2020-04-24 | 深圳大学 | Brain structure feature selection method, mobile terminal and computer-readable storage medium |
| CN111062420B (en) * | 2019-11-26 | 2023-04-25 | 深圳大学 | Brain structure feature selection method, mobile terminal and computer readable storage medium |
| CN111012337A (en) * | 2019-12-13 | 2020-04-17 | 燕山大学 | Brain network and regularized discriminant analysis-based electroencephalogram analysis method |
| CN112037914A (en) * | 2020-08-11 | 2020-12-04 | 深圳大学 | Construction method, system and equipment of obsessive-compulsive disorder risk assessment model |
| CN112037914B (en) * | 2020-08-11 | 2021-06-01 | 深圳大学 | Construction method, system and equipment of obsessive-compulsive disorder risk assessment model |
| CN114190884A (en) * | 2020-09-18 | 2022-03-18 | 深圳大学 | Longitudinal analysis method, system and device for brain disease data |
| CN112289412A (en) * | 2020-10-09 | 2021-01-29 | 深圳市儿童医院 | Construction method of autism spectrum disorder classifier, device thereof and electronic equipment |
| CN113450426A (en) * | 2021-06-09 | 2021-09-28 | 深圳市铱硙医疗科技有限公司 | Magnetic resonance cerebral perfusion imaging data processing system, method, terminal and medium |
| CN117058471A (en) * | 2023-10-12 | 2023-11-14 | 之江实验室 | Disease brain image parting system based on normal brain image database |
| CN117058471B (en) * | 2023-10-12 | 2024-01-09 | 之江实验室 | Disease brain image parting system based on normal brain image database |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110263880B (en) | 2021-08-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110263880A (en) | Construction method, device and the intelligent terminal of cerebral disease disaggregated model | |
| Goutte et al. | On clustering fMRI time series | |
| CN113616184B (en) | Brain network modeling and individual prediction method based on multi-mode magnetic resonance image | |
| US8861815B2 (en) | Systems and methods for modeling and processing functional magnetic resonance image data using full-brain vector auto-regressive model | |
| Wang et al. | Central and peripheral vision for scene recognition: A neurocomputational modeling exploration | |
| Veropoulos | Machine learning approaches to medical decision making | |
| CN111863244B (en) | Functional connection mental disease classification method and system based on sparse pooling graph convolution | |
| Seiffert | Self-organizing neural networks: recent advances and applications | |
| CN111090764A (en) | Image classification method and device based on multitask learning and graph convolution neural network | |
| CN108960289A (en) | Medical imaging sorter and method | |
| CN109192298A (en) | Deep brain medical diagnosis on disease algorithm based on brain network | |
| CN116503680B (en) | Brain image structured analysis and brain disease classification system based on brain atlas | |
| CN109344889A (en) | A kind of cerebral disease classification method, device and user terminal | |
| CN107437252A (en) | Disaggregated model construction method and equipment for ARM region segmentation | |
| CN110298364A (en) | Based on the feature selection approach of multitask under multi-threshold towards functional brain network | |
| CN115272295A (en) | Dynamic brain function network analysis method and system based on time domain-space domain combined state | |
| Bertrand | Hyper-parameter optimization in deep learning and transfer learning: applications to medical imaging | |
| Pakzad et al. | CIRCLe: Color invariant representation learning for unbiased classification of skin lesions | |
| CN111540467A (en) | Schizophrenia classification identification method, operation control device and medical equipment | |
| Wein et al. | Forecasting brain activity based on models of spatiotemporal brain dynamics: A comparison of graph neural network architectures | |
| CN113180695B (en) | Brain-computer interface signal classification method, system, equipment and storage medium | |
| Thirion et al. | Feature characterization in fMRI data: the Information Bottleneck approach | |
| Liu et al. | PPA: principal parcellation analysis for brain connectomes and multiple traits | |
| Castro et al. | Generation of synthetic structural magnetic resonance images for deep learning pre-training | |
| CN107256408B (en) | Method for searching key path of brain function network |
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 |