CN109993230A - A kind of TSK Fuzzy System Modeling method towards brain function MRI classification - Google Patents

A kind of TSK Fuzzy System Modeling method towards brain function MRI classification Download PDF

Info

Publication number
CN109993230A
CN109993230A CN201910269679.7A CN201910269679A CN109993230A CN 109993230 A CN109993230 A CN 109993230A CN 201910269679 A CN201910269679 A CN 201910269679A CN 109993230 A CN109993230 A CN 109993230A
Authority
CN
China
Prior art keywords
tst
feature
indicate
vector
fuzzy
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910269679.7A
Other languages
Chinese (zh)
Other versions
CN109993230B (en
Inventor
王骏
张春香
邓赵红
石争浩
张嘉旭
祝继华
王士同
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangnan University
Original Assignee
Jiangnan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangnan University filed Critical Jiangnan University
Priority to CN201910269679.7A priority Critical patent/CN109993230B/en
Publication of CN109993230A publication Critical patent/CN109993230A/en
Application granted granted Critical
Publication of CN109993230B publication Critical patent/CN109993230B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Abstract

The present invention relates to a kind of TSK Fuzzy System Modeling methods towards brain function MRI classification, belong to technical field of image processing.This method comprises: step S1: being pre-processed to brain function MRI；Step S2: calculating the Pearson correlation coefficient between each brain area, obtains symmetrical matrix, take thereon triangle be unfolded by row, obtain sampling feature vectors, each column of sampling feature vectors represents the data of a picture；Step S3: feature extraction is carried out to sampling feature vectors；Step S4: structural classification device classifies to brain function MRI, and solves model used in classifier using alternative optimization algorithm, completes the classification of image.The present invention is based on TSK fuzzy systems to construct Nonlinear Classifier, indicates the correlation between feature using non-directed graph, can accurately classify to brain function MRI.

Description

A kind of TSK Fuzzy System Modeling method towards brain function MRI classification
Technical field
The invention belongs to technical field of image processing, and in particular to a kind of TSK towards brain function MRI classification Fuzzy System Modeling method.
Background technique
Tranquillization state functional mri (Resting-state Functional Magnetic Resonance Imaging, rs-fMRI) technology can under the conditions of hurtless measure, radiationless, by detection blood oxygen level obtain it is high-resolution Therefore image is increasingly becoming the important means that human brain function detects with is connected to Journal of Sex Research.FMRI is generally referred to based on blood oxygen level The magnetic resonance imaging of (blood oxygen level-dependent, BOLD) is relied on, it is caused by measuring by nervous activity Brain blood flow and the composition transfers such as brain blood oxygen and caused by magnetic resonance signal change to reflect cerebration.
In recent years, various machine learning methods are applied to image classification field by people.Wherein, based on the mould of clustering Inference system is pasted because its powerful uncertainties model ability, excellent interpretation and outstanding generalization ability are in data-driven Uncertain system modeling problem in have successful application.The core concept of this method is by the defeated of training data Enter/output set carries out clustering and extracts " IF-THEN " fuzzy rule, construct fuzzy inference system on this basis to dig Dig the mapping relations between input data and output data.Although fuzzy inference system has had more in image classification field Successfully application, but these methods towards be all relatively simple data scene, and in brain function MRI Complex scene, fuzzy inference system application it is very few.On the other hand, from there is significant between the feature extracted in brain image Correlation, brain function MRI classification performance can be effectively improved by effectively utilizing these relevant informations.However, existing When some machine learning method research image classification predictions, assumes that each feature is independent from each other mostly, it is special to fail effective use Correlation properties between sign.
Summary of the invention
In order to overcome the shortcomings of prior art, the technical problem to be solved by the present invention is to design a kind of interpretation it is strong, The high brain functional magnetic resonance image classification of nicety of grading.
Technical solution of the present invention:
A kind of TSK Fuzzy System Modeling method towards brain function MRI classification, comprising the following steps:
Step S1: brain function MRI is pre-processed；
The step S1 includes: the data at 10 time points before 1. removing brain function MRI sequence；2. time horizon Correction and head movement correction；3. data use t1 weighted image to divide and normalized to MNI152 by unified In (Montreal Neurological Institute 152) normed space；4. using Anatomical Automatic Brain is divided into 116 brain areas, each region resampling 3*3*3mm by Labeling (AAL) template3Voxel；5. using Full width at half maximum Gaussian kernel carries out space smoothing processing；6. removing noise using bandpass filtering (0.01-0.1Hz)；7. going linearly to float It moves and carries out overall signal's correction and go disturbance variable；8. calculating the average time sequence of each brain area.
Step S2: calculating the Pearson correlation coefficient between each brain area, obtains symmetrical matrix, take thereon triangle by row Expansion, obtains sampling feature vectors, and each column of sampling feature vectors represents a magnetic resonance image data, i.e. a sample；It crosses Journey is as shown in Figure 1；
In the step S2, the Pearson correlation coefficient s between vector a, b is defined as follows:
Wherein, ai、biRespectively i-th of element of vector a, b, d are the number of element in vector a, b,Indicate to Measure the mean value of a, b.
Step S3: feature extraction is carried out to sampling feature vectors, as shown in Figure 1；
The step S3 includes: step S31: by all samples and its label, proportionally random division is training set, tests Card collection and test set, training set are used to training pattern, and verifying collection is used to parameter optimization, and test set is used to prediction result.Wherein, it marks Label are classification belonging to sample, with the positive class and negative class of digital representation sample；Step S32: calculate training set in each column feature with Pearson correlation coefficient between label vector y, and arranged by its descending, wherein label vector y is all sample labels composition Vector；Step S33: the corresponding characteristic series index set of P Pearson coefficient before retaining extracts corresponding spy in training set Sign completes feature extraction；Step S34: indexing according to characteristic series obtained in step S33 and gather, and extracts verifying collection, test respectively The feature of concentration completes the feature extraction to verifying collection and test set.
Step S4: structural classification device classifies to brain function MRI on training set, and uses alternative optimization Algorithm solves model used in classifier, completes the classification of image.
The step S4 includes: step S41: utilizing TSK Fuzzy System Modeling；Step S42: the related letter between feature is extracted Cease simultaneously structural classification model；Step S43: using alternate optimization method solving model, completes the classification to image.
TSK Fuzzy System Modeling is utilized described in the step S41, detailed process is as follows:
Firstly, carrying out fuzzy division to training set feature using FCM clustering algorithm, the mean value of Gauss subordinating degree function is calculated crpAnd variances sigmarp:
Wherein, h indicates adjustable scale parameter, can be obtained by mesh parameter optimization；xnpIndicate the pth Wei Te of n-th of sample Sign；Indicate n-th of sampleThe degree of membership for belonging to the r articles fuzzy rule is gathered by FCM Class algorithm obtains.
Secondly, determining crpAnd σrpLater, input sample x is calculatednPth dimensional feature xnPThe r articles corresponding fuzzy rule Fuzzy subset ArpDegree of membership
Finally, according toWithIt calculatesWherein, φr=diag (φr(x1), φr(x2) ..., φr(xN)) (1, X it) indicates to be mapped to the data in new feature space by the r articles fuzzy rule, r=1,2 ..., k ... R, X=(x1, x2..., xN)TFor training set eigenmatrix；μr(xn) indicate the corresponding subordinating degree function of the r articles fuzzy rule, φr(xn) indicate normalization Fuzzy membership function afterwards, φ indicate the data by TSK FUZZY MAPPING to new feature space.
Detailed process is as follows by the step S42:
Firstly, calculating φrIn covariance two-by-two between feature column vector obtain covariance matrixΩr= ∑r -1Indicate concentration matrix；Then, using Graphical Lasso algorithm to ΩrLS-SVM sparseness is carried out to obtainFinally, right In r=1,2 ..., R, constructionIt indicatesIt is not the set of line number i where 0 element in jth column,It indicatesIn the column of the i-th row element, calculateGatherThe number of middle element.
The step S43 obtains detailed process are as follows:
Firstly, initialization coefficient vector w=0；
Secondly, iteration optimization w and v according to the following formularp:
Wherein, vrpIndicate wrPth column in breakdown, ρ are a great parameter, wrIndicate that r rule is corresponding in w Coefficient vector；Y indicates training sample label vector,Indicate positive weights coefficient,It indicates for SrpInitial estimate, as vrpRelated coefficient between y.Parameter lambda and α are equal Greater than 0, obtained by mesh parameter optimizing.Finally, the w updated is classifier after completing k iteration on training set Coefficient vector.
Step S5: when carrying out prediction result on test set, the test set that division obtains is carried out according to step S3 first Feature extraction obtains X(tst)；Enable y(tst)=sigmoid (φ(tst)W) presentation class is as a result, φ(tst)=(φ1 (tst), φ2 (tst)..., φR (tst)),φr (tst)Calculation method and above φrIt is identical；Wherein sigmoid function is defined as:Threshold value is 0.5, it may be assumed that as S (z) >=0.5, be positive class, otherwise the class that is negative, z φ(tst)Each of w Element.
In step S31, by all samples and its label, random division is that training set, verifying collect and survey according to a certain percentage Examination collection, the positive class and negative class of sample are indicated with 1,0 or 1, -1
Beneficial effects of the present invention:
1, a kind of TSK Fuzzy System Modeling method towards brain function MRI classification provided by the invention, passes through It extracts the relevant information between feature, effectively improve model in conjunction with the methods of TSK Fuzzy System Modeling, alternative optimization solution The accuracy of interpretation and classification
2, a kind of TSK Fuzzy System Modeling method towards brain function MRI classification provided by the invention, can The block structure of relevant information and the output of TSK fuzzy system between feature is sufficiently combined, a kind of nonlinear model is constructed Classification method is pasted, the deficiency of traditional brain functional magnetic resonance image classification linear classification less effective is compensated for, is able to carry out Accurately classification.
Detailed description of the invention
Fig. 1 is a kind of TSK Fuzzy System Modeling towards brain function MRI classification according to the embodiment of the present invention The characteristic extraction procedure figure of method.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific embodiment, and reference Attached drawing, the present invention is described in more detail.
The present invention is to provide a kind of TSK Fuzzy System Modeling methods towards brain function MRI classification.The party Method first to brain function MRI carry out pretreatment and feature extraction, then by between brain area relevant information and TSK obscure The nonlinear classification method of system combined structure finally verifies the classification performance of proposed classification method using the method that reserves.Tool Body implementation steps are as follows:
Step S1: pretreatment is carried out to brain function MRI and brain area is divided, and calculates the mean time of each brain area Between sequence；
1. removing the data at 10 time points before brain function MRI sequence；2. time horizon correction and head movement school Just；3. data use t1 weighted image to divide and normalized to MNI152 (Montreal Neurological by unified Institute 152) in normed space；4. brain is drawn using Anatomical Automatic Labeling (AAL) template It is divided into 116 brain areas, each region resampling 3*3*3mm3Voxel；5. carrying out space smoothing using full width at half maximum Gaussian kernel Processing；6. removing noise using bandpass filtering (0.01-0.1Hz)；7. remove linear drift and carry out overall signal correction go to interfere Variable；8. calculating the average time sequence of each brain area.
Step S2: the Pearson correlation coefficient between brain area feature is calculated, function connects matrix is obtained, takes triangle thereon It is unfolded by row, obtains sampling feature vectors, each column of sampling feature vectors represents a magnetic resonance image data, i.e. a sample； Process is as shown in Figure 1；
Step S21: calculating the Pearson correlation coefficient between each brain area feature vector, obtains function connects matrix M, In, Pearson correlation coefficient is defined as follows:
ai、biRespectively i-th of element of vector a, b, d are the number of element in vector a, b,Indicate vector a, b Mean value.
Step S22: the upper triangle for the function connects matrix for taking step S21 to obtain simultaneously is unfolded by row, obtains 6670 dimension samples Feature vector；
Step S3: feature extraction is carried out to sampling feature vectors, process is as shown in Figure 1；
Step S31: all samples and its label are training set, verifying collection according to the ratio random division of 7:2:1 and surveyed Examination collection, wherein classification belonging to label, that is, sample usually indicates the positive class and negative class of sample with 1,0 or 1, -1；
Step S32: the Pearson correlation coefficient in training set between each column feature and training label vector y is calculated, and is pressed The arrangement of its descending；Step S33: the feature of P correlation maximum before retaining, and the corresponding column index of P feature before journal Set, according to column index set to training sample carry out feature extraction, the present invention in, enable P=180；
Step S34: feature extraction is carried out to verifying collection, training set respectively according to column index set obtained in step S33.
Step S4: structural classification device classifies to brain function MRI, and solves mould using alternate optimization method Type, it is final to use the classification performance for reserving method verifying classifier.
Step S41: TSK Fuzzy System Modeling is utilized；
Firstly, extracting TSK fuzzy rule former piece: carrying out fuzzy division to data set D using FCM clustering algorithm, calculate high The mean value and variance of this subordinating degree function:
Wherein, h indicates adjustable scale parameter, can be obtained by grid optimization；xnpIndicate p-th of feature of n-th of sample,Indicate n-th of input variable xn=(xn1, xn2..., xnP)T∈RPBelong to the degree of membership of r-th of cluster.；
Determine crpAnd σrpLater, degree of membership is calculated
Wherein, xpIndicate pth column feature.Finally, according toWith Calculate φ=(φ1, φ2..., φr... φR)∈RN×R(P+1), wherein φr=diag (φr(x1), φr(x2) ..., φr(xN)) (1, X) indicate to be mapped to the data in new feature space by the r articles fuzzy rule, r=1,2 ..., K ... R, X=(x1, x2..., xN)TFor training set eigenmatrix；μr(xn) indicate the corresponding degree of membership letter of the r articles fuzzy rule Number, φr(xn) indicate that the fuzzy membership function after normalization, φ indicate the number by TSK FUZZY MAPPING to new feature space According to.
Step S42: the relevant information and structural classification device between feature are extracted:
Firstly, extracting φrRelevant information between (r=1,2 ..., R) feature.Calculate φrIn two-by-two between column vector Covariance obtains covariance matrix ∑r∈RP×P, enable Ωr=∑r -1Indicate its corresponding concentration matrix.Utilize Graphical Lasso algorithm is to ΩrLS-SVM sparseness is carried out to obtainFinally, for r=1,2 ..., R, constructionTable ShowIt is not the set of line number i where 0 element in jth column,It indicatesIn the i-th row jth column element, calculate GatherThe number of middle element.
Alternative optimization solving model.Initialize w=0；
When kth time iteration, w and vrpIt is respectively as follows:
Wherein, y indicates the label of training sample,Indicate positive weights coefficient,It indicates for SrpInitial estimate, in the present invention, we willIt is expressed as vrpRelated coefficient between y.Parameter λ and α is obtained by grid optimizing.After k iteration, final w is returned.
Step S5: the prediction result on test set:
When carrying out prediction result on test set, feature is carried out according to step S3 to the test set that division obtains first and is mentioned It takes, obtains X(tst).Enable y(tst)=sigmoid (φ(tst)W) presentation class is as a result, φ(tst)=(φ1 (tst), φ2 (tst)... φr (tst)..., φR (tst)) in φr (tst)Calculation method and above φrIt is identical.Wherein, y(tst)Indicate prediction result, Sigmoid function are as follows:Threshold value is 0.5, it may be assumed that as S (z) >=0.5, be positive class, otherwise the class that is negative, z are φ(tst)Each of w element.
Performance Evaluation: according to the ratio of 7:2:1, being divided into training set, verifying collection and test set for data set and class label, The training pattern on training set collects upper adjustment parameter, the assessment models on test set in verifying with the method for mesh parameter optimizing Classification performance, repeat the above process 20 times, record result simultaneously take mean value as final classification performance evaluation index.
A kind of classification effect of TSK Fuzzy System Modeling method towards brain function MRI classification of the present invention Fruit can be illustrated by true ASD data:
(1) ASD data experiment process
To show effect of the invention, in embodiments using from ABIDE (Autism Brain Imaging Data Exchange) database (http://preprocessed-connectomes-project.org/abide/ download.html) in three different imaging centers data (NYU, UCLA, UM), relevant information is as shown in table 1.
The particularly relevant information of 1 ASD medical data collection of table
(2) experimental result
The classification performance of five kinds of algorithms of different is as shown in table 2 on true ASD medical data, best result in figure with Runic is shown.Wherein, the present invention is measured each using classification accuracy (ACC), susceptibility (SEN), specific (SPE), AUC value The classification performance of a algorithm:
Wherein, TP, FN, FP, TN respectively indicate real example, false counter-example, false positive example, true counter-example.As shown in table 2, method GSUG-TSK is achieved in tri- data centers of NYU, UCLA, UM with 0.7639,0.7318,0.7500 average nicety of grading Best classifying quality.For equally used the WE algorithm of TSK fuzzy system, L2-TSKFS algorithm, TSFS-SVM algorithm and GENFIS3 algorithm, WE algorithm will be substantially better than remaining four kinds of algorithm, this shows that the relevant information between brain image feature is combined to have Conducive to the diagnosis capability for improving self-closing disease brain function MRI.
The classification method of 2 embodiment of the present invention of table compares other existing sorting algorithms in the classification at different data center Energy comparison diagram, wherein best result overstriking is shown；
The classification performance of all kinds of algorithms of table 2
In order to find out the brain region most beneficial for ASD medical diagnosis on disease, we pick out the feature of most taste in rs- Corresponding brain area pair in fMRI.Therefore, in same imaging center, the sum of the feature weight of each imaging center is calculated, and from 30 weights feature corresponding in function connects matrix M and 60 interregional functional characters before middle selection.Table 3 is shown 30 pairs of interregional functional characters of the most taste that GSUG-TSK method is selected from rs-fMRI, the digital representation in bracket The index of functional area in AAL-Label template.
Table 3 is 30 pairs of interregional function of the most taste that the classification method of the embodiment of the present invention is chosen from rs-fMRI Can feature, the index of functional area in the digital representation AAL-Label template in bracket；
The interregional functional character of 3 preceding 30 pairs of table most taste
The experiment results show that a kind of TSK fuzzy system towards brain function MRI classification of the present invention is built Mould method achieves preferable classifying quality on ASD data set, effectively improves the classification of brain function MRI Energy.
Particular embodiments described above has carried out further in detail the purpose of the present invention, technical scheme and beneficial effects It describes in detail bright, it should be understood that the above is only a specific embodiment of the present invention, is not limited to the present invention, it is all Within the spirit and principles in the present invention, any modification, equivalent substitution, improvement and etc. done should be included in guarantor of the invention Within the scope of shield.

Claims (10)

1. a kind of TSK Fuzzy System Modeling method towards brain function MRI classification, which is characterized in that including following step It is rapid:
Step S1: brain function MRI is pre-processed；
Step S2: calculating the Pearson correlation coefficient between each brain area, obtains symmetrical matrix, take thereon triangle be unfolded by row, Sampling feature vectors are obtained, each column of sampling feature vectors represents a magnetic resonance image data, i.e. a sample；
Step S3: feature extraction is carried out to sampling feature vectors；
Step S4: structural classification device classifies to brain function MRI on training set, specifically includes: step S41: benefit With TSK Fuzzy System Modeling；Step S42: the relevant information and structural classification model between feature are extracted；Step S43: using alternating Optimization Method model completes the classification to image；
Step S5: the prediction data on test set.
2. a kind of TSK Fuzzy System Modeling method towards brain function MRI classification according to claim 1, It is characterized in that, the step S1 includes: the data at 10 time points before 1. removing brain function MRI sequence；2. time horizon Correction and head movement correction；3. data use t1 weighted image to divide and normalized to MNI152 by unified In (Montreal Neurological Institute152) normed space；4. using Anatomical Automatic Brain is divided into 116 brain areas, each region resampling 3*3*3mm by Labeling (AAL) template3Voxel；5. using Full width at half maximum Gaussian kernel carries out space smoothing processing；6. removing noise using bandpass filtering (0.01-0.1 Hz)；7. going linearly to float It moves and carries out overall signal's correction and go disturbance variable；8. calculating the average time sequence of each brain area.
3. a kind of Fuzzy System Modeling side TSK towards brain function MRI classification according to claim 1 or 2 Method, which is characterized in that in the step S2, the Pearson correlation coefficient s between vector a, b is defined as follows:
Wherein, ai、biRespectively i-th of element of vector a, b, d are the number of element in vector a, b,Indicate vector a, b Mean value.
4. a kind of Fuzzy System Modeling side TSK towards brain function MRI classification according to claim 1 or 2 Method, which is characterized in that the step S3 includes: step S31: by all samples and its label, proportionally random division is training Collection, verifying collection and test set, training set are used to training pattern, and verifying collection is used to parameter optimization, and test set is used to prediction result；Its In, classification belonging to label, that is, sample, with the positive class and negative class of digital representation sample；Step S32: it is special to calculate each column in training set Pearson correlation coefficient between sign and label vector y, and arranged by its descending, wherein label vector y is all training samples The vector of label composition；Step S33: the corresponding characteristic series index set of P Pearson coefficient before retaining is extracted in training set Corresponding feature completes feature extraction；Step S34: indexing according to characteristic series obtained in step S33 and gather, and extracts test respectively Feature in card collection, test set, completes the feature extraction to verifying collection and test set.
5. a kind of TSK Fuzzy System Modeling method towards brain function MRI classification according to claim 3, Be characterized in that, the step S3 includes: step S31: by all samples and its label, proportionally random division is training set, tests Card collection and test set, training set are used to training pattern, and verifying collection is used to parameter optimization, and test set is used to prediction result；Wherein, it marks Label are classification belonging to sample, with the positive class and negative class of digital representation sample；Step S32: calculate training set in each column feature with Pearson correlation coefficient between label vector y, and arranged by its descending, wherein label vector y is all sample labels composition Vector；Step S33: the corresponding characteristic series index set of P Pearson coefficient before retaining extracts corresponding spy in training set Sign completes feature extraction；Step S34: indexing according to characteristic series obtained in step S33 and gather, and extracts verifying collection, test respectively The feature of concentration completes the feature extraction to verifying collection and test set.
6. a kind of according to claim 1, Fuzzy System Modeling side TSK towards brain function MRI classification described in 2 or 5 Method, which is characterized in that
TSK Fuzzy System Modeling is utilized described in the step S41, detailed process is as follows:
Firstly, carrying out fuzzy division to training set feature using FCM clustering algorithm, the mean value c of Gauss subordinating degree function is calculatedrpWith Variances sigmarp:
Wherein, h indicates adjustable scale parameter, can be obtained by mesh parameter optimization；xnpIndicate the pth dimensional feature of n-th of sample； Indicate n-th of sampleThe degree of membership for belonging to the r articles fuzzy rule, by FCM clustering algorithm It obtains；
Secondly, determining crpAnd σrpLater, input sample x is calculatednPth dimensional feature xnpThe mould of the r articles corresponding fuzzy rule Paste subset ArpDegree of membership
Finally, according toWithIt calculatesWherein, φr=diag (φr(x1), φr(x2) ..., φr(xN)) (1, X it) indicates to be mapped to the data in new feature space by the r articles fuzzy rule, r=1,2 ..., k ... R, X=(x1, x2..., xN)TFor training set eigenmatrix；μr(xn) indicate the corresponding subordinating degree function of the r articles fuzzy rule, φr(xn) indicate normalization Fuzzy membership function afterwards, φ indicate the data by TSK FUZZY MAPPING to new feature space；
Detailed process is as follows by the step S42:
Firstly, calculating φrIn covariance two-by-two between feature column vector obtain covariance matrixΩr=∑r -1Table Show concentration matrix；Then, using Graphical Lasso algorithm to ΩrLS-SVM sparseness is carried out to obtainFinally, for r=1, 2 ..., R, constructionIt indicatesIt is not the set of line number i where 0 element in jth column,It indicates In the i-th row jth column element, calculateGatherThe number of middle element；
The step S43 obtains detailed process are as follows:
Firstly, initialization coefficient vector w=0；
Secondly, iteration optimization w and v according to the following formularp:
Wherein, vrpIndicate wrPth column in breakdown, ρ are a great parameter, wrIndicate the corresponding system of r rule in w Number vector；Y indicates the vector of all sample label compositions,Indicate positive weights system Number, It indicates for SrpInitial estimate, as vrpWith the related coefficient between label y；Parameter lambda It is all larger than 0 with α, is obtained by mesh parameter optimizing；Finally, the w updated is point after completing k iteration on training set The coefficient vector of class device；
Step S5: when carrying out prediction result on test set, feature is carried out according to step S3 to the test set that division obtains first It extracts, obtains X(tst)；Enable y(tst)=sigmoid (φ(tst)W) presentation class is as a result, φ(tst)=(φ1 (tst), φ2 (tst)..., φR (tst)), φr (tst)Calculation method and above φrIt is identical；Wherein sigmoid function is defined as:Threshold value It is 0.5, it may be assumed that as S (z) >=0.5, be positive class, otherwise the class that is negative, z φ(tst)Each of w element.
7. a kind of TSK Fuzzy System Modeling method towards brain function MRI classification according to claim 3, It is characterized in that,
TSK Fuzzy System Modeling is utilized described in the step 541, detailed process is as follows:
Firstly, carrying out fuzzy division to training set feature using FCM clustering algorithm, the mean value c of Gauss subordinating degree function is calculatedrpWith Variances sigmarp:
Wherein, h indicates adjustable scale parameter, can be obtained by mesh parameter optimization；xnpIndicate the pth dimensional feature of n-th of sample； Indicate n-th of sampleThe degree of membership for belonging to the r articles fuzzy rule, by FCM clustering algorithm It obtains；
Secondly, determining crpAnd σrpLater, input sample x is calculatednPth dimensional feature xnpThe mould of the r articles corresponding fuzzy rule Paste subset ArpDegree of membership
Finally, according toWithIt calculatesWherein, φr=diag (φr(x1), φr(x2) ..., φr(xN)) (1, X it) indicates to be mapped to the data in new feature space by the r articles fuzzy rule, r=1,2 ..., k ... R, X=(x1, x2..., xN)TFor training set eigenmatrix；μr(xn) indicate the corresponding subordinating degree function of the r articles fuzzy rule, φr(xn) indicate normalization Fuzzy membership function afterwards, φ indicate the data by TSK FUZZY MAPPING to new feature space；
Detailed process is as follows by the step S42:
Firstly, calculating φrIn covariance two-by-two between feature column vector obtain covariance matrixΩr=∑r -1Table Show concentration matrix；Then, using Graphical Lasso algorithm to ΩrLS-SVM sparseness is carried out to obtainFinally, for r=1, 2 ..., R, constructionIt indicatesIt is not the set of line number i where 0 element in jth column,It indicates In the i-th row jth column element, calculateGatherThe number of middle element；
The step S43 obtains detailed process are as follows:
Firstly, initialization coefficient vector w=0；
Secondly, iteration optimization w and v according to the following formularp:
Wherein, vrpIndicate wrPth column in breakdown, ρ are a great parameter, wrIndicate the corresponding system of r rule in w Number vector；Y indicates the vector of all sample label compositions,Indicate positive weights system Number, It indicates for SrpInitial estimate, as vrpWith the related coefficient between label y；Parameter lambda It is all larger than 0 with α, is obtained by mesh parameter optimizing；Finally, the w updated is point after completing k iteration on training set The coefficient vector of class device；
Step S5: when carrying out prediction result on test set, feature is carried out according to step S3 to the test set that division obtains first It extracts, obtains x(tst)；Enable y(tst)=sigmoid (φ(tst)W) presentation class is as a result, φ(tst)=(φ1 (tst), φ2 (tst)..., φR (tst)), φr (tst)Calculation method and above φrIt is identical；Wherein sigmoid function is defined as:Threshold value It is 0.5, it may be assumed that as S (z) >=0.5, be positive class, otherwise the class that is negative, z φ(tst)Each of w element.
8. a kind of TSK Fuzzy System Modeling method towards brain function MRI classification according to claim 4, It is characterized in that,
TSK Fuzzy System Modeling is utilized described in the step 541, detailed process is as follows:
Firstly, carrying out fuzzy division to training set feature using FCM clustering algorithm, the mean value c of Gauss subordinating degree function is calculatedrpWith Variances sigmarp:
Wherein, h indicates adjustable scale parameter, can be obtained by mesh parameter optimization；xnpIndicate the pth dimensional feature of n-th of sample； Indicate n-th of sampleThe degree of membership for belonging to the r articles fuzzy rule, by FCM clustering algorithm It obtains；
Secondly, determining crpAnd σrpLater, input sample x is calculatednPth dimensional feature xnpThe mould of the r articles corresponding fuzzy rule Paste subset ArpDegree of membership
Finally, according toWithIt calculatesWherein, φr=diag (φr(x1), φr(x2) ..., φr(xN)) (1, X it) indicates to be mapped to the data in new feature space by the r articles fuzzy rule, r=1,2 ..., k ... R, X=(x1, x2..., xN)TFor training set eigenmatrix；μr(xn) indicate the corresponding subordinating degree function of the r articles fuzzy rule, φr(xn) indicate normalization Fuzzy membership function afterwards, φ indicate the data by TSK FUZZY MAPPING to new feature space；
Detailed process is as follows for the step 542:
Firstly, calculating φrIn covariance two-by-two between feature column vector obtain covariance matrixΩr=∑r -1Table Show concentration matrix；Then, LS-SVM sparseness is carried out to Ω r using Graphical Lasso algorithm to obtainFinally, for r=1, 2 ..., R, constructionIt indicatesIt is not the set of line number i where 0 element in jth column,It indicates In the i-th row jth column element, calculateGatherThe number of middle element；
The step 543 obtains detailed process are as follows:
Firstly, initialization coefficient vector w=0；
Secondly, iteration optimization w and v according to the following formularp:
Wherein, vrpIndicate wrPth column in breakdown, ρ are a great parameter, wrIndicate the corresponding system of r rule in w Number vector；Y indicates the vector of all sample label compositions,Indicate positive weights system Number, Indicate the initial estimate for Srp, as vrpWith the related coefficient between label y；Parameter lambda It is all larger than 0 with α, is obtained by mesh parameter optimizing；Finally, the w updated is point after completing k iteration on training set The coefficient vector of class device；
Step S5: when carrying out prediction result on test set, feature is carried out according to step S3 to the test set that division obtains first It extracts, obtains X(tst)；Enable y(tst)=sigmoid (φ(tst)W) presentation class is as a result, φ(tst)=(φ1 (tst), φ2 (tst)..., φR (tst)), φr (tst)Calculation method and above φrIt is identical；Wherein sigmoid function is defined as:Threshold Value is 0.5, it may be assumed that as S (z) >=0.5, be positive class, otherwise the class that is negative, z φ(tst)Each of w element.
9. a kind of Fuzzy System Modeling side TSK towards brain function MRI classification according to claim 5,7 or 8 Method, which is characterized in that in step S31, proportionally random division is that training set, verifying collect and survey by all samples and its label Examination collection, the positive class and negative class of sample are indicated with 1,0 or 1, -1.
10. a kind of TSK Fuzzy System Modeling method towards brain function MRI classification according to claim 6, It is characterized in that, in step S31, by all samples and its label, proportionally random division is training set, verifying collects and test Collection, the positive class and negative class of sample are indicated with 1,0 or 1, -1.
CN201910269679.7A 2019-04-04 2019-04-04 TSK fuzzy system modeling method for brain function magnetic resonance image classification Active CN109993230B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910269679.7A CN109993230B (en) 2019-04-04 2019-04-04 TSK fuzzy system modeling method for brain function magnetic resonance image classification

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910269679.7A CN109993230B (en) 2019-04-04 2019-04-04 TSK fuzzy system modeling method for brain function magnetic resonance image classification

Publications (2)

Publication Number Publication Date
CN109993230A true CN109993230A (en) 2019-07-09
CN109993230B CN109993230B (en) 2023-04-18

Family

ID=67132461

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910269679.7A Active CN109993230B (en) 2019-04-04 2019-04-04 TSK fuzzy system modeling method for brain function magnetic resonance image classification

Country Status (1)

Country Link
CN (1) CN109993230B (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363165A (en) * 2019-07-18 2019-10-22 深圳大学 Multi-object tracking method, device and storage medium based on TSK fuzzy system
CN110797109A (en) * 2019-10-15 2020-02-14 山东朗润医疗系统有限公司 Automatic classification system for magnetic resonance images
CN111414579A (en) * 2020-02-19 2020-07-14 深圳市儿童医院 Method and system for acquiring brain area association information based on multi-angle correlation
CN111488934A (en) * 2020-04-13 2020-08-04 脑结构(南京)科技有限公司 Brain image data processing method, storage medium, computer device and apparatus
CN112561935A (en) * 2020-12-26 2021-03-26 广东工业大学 Method, device and equipment for identifying Alzheimer's disease
CN113052271A (en) * 2021-05-14 2021-06-29 江南大学 Biological fermentation data prediction method based on deep neural network
CN113723239A (en) * 2021-08-18 2021-11-30 广东工业大学 Magnetic resonance image classification method and system based on causal relationship
CN116188880A (en) * 2023-05-05 2023-05-30 中国科学院地理科学与资源研究所 Cultivated land classification method and system based on remote sensing image and fuzzy recognition
CN116898401A (en) * 2023-07-17 2023-10-20 燕山大学 Autism spectrum disorder subtype classification method and device based on combined recursion quantification

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105389599A (en) * 2015-10-12 2016-03-09 上海电机学院 Feature selection approach based on neural-fuzzy network
CN108596228A (en) * 2018-04-13 2018-09-28 江南大学 A kind of brain functional magnetic resonance image classification based on unsupervised fuzzy system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105389599A (en) * 2015-10-12 2016-03-09 上海电机学院 Feature selection approach based on neural-fuzzy network
CN108596228A (en) * 2018-04-13 2018-09-28 江南大学 A kind of brain functional magnetic resonance image classification based on unsupervised fuzzy system

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363165B (en) * 2019-07-18 2023-04-14 深圳大学 Multi-target tracking method and device based on TSK fuzzy system and storage medium
CN110363165A (en) * 2019-07-18 2019-10-22 深圳大学 Multi-object tracking method, device and storage medium based on TSK fuzzy system
CN110797109A (en) * 2019-10-15 2020-02-14 山东朗润医疗系统有限公司 Automatic classification system for magnetic resonance images
CN111414579A (en) * 2020-02-19 2020-07-14 深圳市儿童医院 Method and system for acquiring brain area association information based on multi-angle correlation
CN111488934A (en) * 2020-04-13 2020-08-04 脑结构(南京)科技有限公司 Brain image data processing method, storage medium, computer device and apparatus
CN112561935A (en) * 2020-12-26 2021-03-26 广东工业大学 Method, device and equipment for identifying Alzheimer's disease
CN112561935B (en) * 2020-12-26 2022-09-16 广东工业大学 Intelligent classification method, device and equipment for brain images
CN113052271A (en) * 2021-05-14 2021-06-29 江南大学 Biological fermentation data prediction method based on deep neural network
CN113723239A (en) * 2021-08-18 2021-11-30 广东工业大学 Magnetic resonance image classification method and system based on causal relationship
CN113723239B (en) * 2021-08-18 2023-10-20 广东工业大学 Magnetic resonance image classification method and system based on causal relationship
CN116188880A (en) * 2023-05-05 2023-05-30 中国科学院地理科学与资源研究所 Cultivated land classification method and system based on remote sensing image and fuzzy recognition
CN116898401A (en) * 2023-07-17 2023-10-20 燕山大学 Autism spectrum disorder subtype classification method and device based on combined recursion quantification
CN116898401B (en) * 2023-07-17 2024-02-02 燕山大学 Autism spectrum disorder subtype classification method and device based on combined recursion quantification

Also Published As

Publication number Publication date
CN109993230B (en) 2023-04-18

Similar Documents

Publication Publication Date Title
CN109993230A (en) A kind of TSK Fuzzy System Modeling method towards brain function MRI classification
Liao et al. Deep facial spatiotemporal network for engagement prediction in online learning
CN110020623B (en) Human body activity recognition system and method based on conditional variation self-encoder
CN104063719B (en) Pedestrian detection method and device based on depth convolutional network
CN105975916B (en) Age estimation method based on multi output convolutional neural networks and ordinal regression
CN109635835A (en) A kind of breast lesion method for detecting area based on deep learning and transfer learning
CN109101938B (en) Multi-label age estimation method based on convolutional neural network
CN108229267A (en) Object properties detection, neural metwork training, method for detecting area and device
CN111090764B (en) Image classification method and device based on multitask learning and graph convolution neural network
CN108986140A (en) Target scale adaptive tracking method based on correlation filtering and color detection
CN102509123B (en) Brain function magnetic resonance image classification method based on complex network
CN103295031B (en) A kind of image object method of counting based on canonical risk minimization
CN113616184A (en) Brain network modeling and individual prediction method based on multi-mode magnetic resonance image
CN106778687A (en) Method for viewing points detecting based on local evaluation and global optimization
CN109886944B (en) A kind of white matter high signal intensity detection and localization method based on multichannel chromatogram
CN111863244B (en) Functional connection mental disease classification method and system based on sparse pooling graph convolution
CN109948740A (en) A kind of classification method based on tranquillization state brain image
CN104281835A (en) Face recognition method based on local sensitive kernel sparse representation
CN110264454A (en) Cervical cancer tissues pathological image diagnostic method based on more hidden layer condition random fields
CN112418337A (en) Multi-feature fusion data classification method based on brain function hyper-network model
CN106529377A (en) Age estimating method, age estimating device and age estimating system based on image
CN111507184B (en) Human body posture detection method based on parallel cavity convolution and body structure constraint
CN106780422A (en) A kind of notable figure fusion method based on Choquet integrations
CN114511759A (en) Method and system for identifying categories and determining characteristics of skin state images
CN108960005A (en) The foundation and display methods, system of subjects visual label in a kind of intelligent vision Internet of Things

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