CN106485707B - Multidimensional characteristic classification method based on brain magnetic resonance imaging image - Google Patents
Multidimensional characteristic classification method based on brain magnetic resonance imaging image Download PDFInfo
- Publication number
- CN106485707B CN106485707B CN201610886193.4A CN201610886193A CN106485707B CN 106485707 B CN106485707 B CN 106485707B CN 201610886193 A CN201610886193 A CN 201610886193A CN 106485707 B CN106485707 B CN 106485707B
- Authority
- CN
- China
- Prior art keywords
- characteristic
- feature
- roi
- optimal
- subset
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Abstract
The present invention discloses the multidimensional characteristic classification method based on brain MR image, carries out region division to brain MR image and extracts several ROI features；A kind of marker characteristic is selected, establishes and forms correlated characteristic about the correlation between several ROI features of the marker characteristic；Several brains MR image zooming-out goes out several ROI features and several correlated characteristics form ROI feature set and correlated characteristic set；Optimal ROI feature subset and optimal correlated characteristic subset are selected respectively to ROI feature set and correlated characteristic set by composite character algorithm；The weight factor of ROI feature function ratio in classifier is set, multi-core classifier is formed by weight factor and the optimal ROI feature subset of multicore SVM model integration, optimal correlated characteristic subset.The present invention obtains high dimensional feature, can analyze part change caused by related disease and function connects change simultaneously, and classification accuracy is high, can auxiliary diagnosis various disease.
Description
Technical field
The present invention relates to technical field of image processing, it is more particularly related to which a kind of be based on brain magnetic resonance imaging figure
The multidimensional characteristic classification method of picture.
Background technique
Medical image refers to for medical treatment or medical research, to human body or human body part, obtained with non-intruding mode in
The technology and treatment process of tissue image, portion.It includes the relatively independent research direction of following two: medical image system
(medical imaging system) and Medical Image Processing (medical image processing).The former refers to image
Row process, including to imaging mechanism, imaging device, imaging system analyze the problems such as research；The latter refers to having obtained
The image obtained further processes, the purpose is to either make original not enough clearly image restoration, or for protrusion
Certain characteristic informations in image, or pattern classification etc. is done to image.
In recent years, according to brain magnetic resonance imaging image, the auxiliary diagnosis of related disease and pre- is carried out using machine learning algorithm
Survey is a research hotspot now.The morphology that the existing feature based on brain magnetic resonance imaging image is mainly based upon voxel is special
Sign is changed using the structure that the sorting algorithm that this feature is established can only find out privileged site, can not be to privileged site function
The change of connection is analyzed, and there is presently no a kind of sorting algorithms well for clinic can be in brain magnetic resonance imaging image
On the basis of simultaneously analyze the change that organization of human body is connected with the structure function, this requires extract from brain magnetic resonance imaging image
Out more high-dimensional feature and establish classifier realize related disease sorting algorithm.
Summary of the invention
Shortcoming present in view of the above technology, it is special that the present invention provides a kind of multidimensional based on brain magnetic resonance imaging image
Classification method is levied, ROI feature is carried out to the extraction of correlated characteristic, setting ROI feature relative to related to brain magnetic resonance imaging image
Feature in classifier the weight factor of function ratio, multi-core classifier formed by multicore SVM model integration, realize that higher-dimension is special
Part change caused by related disease and function connects change are analyzed in the acquisition of sign simultaneously, and the accuracy of classification of diseases is high, can
For the auxiliary diagnosis of various disease, there is stronger applicability.
In order to realize these purposes and other advantages according to the present invention, the invention is realized by the following technical scheme:
Multidimensional characteristic classification method of the present invention based on brain magnetic resonance imaging image, comprising the following steps:
Region division is carried out to extract several ROI features, a ROI feature packet to a brain magnetic resonance imaging image
Include the marker characteristic for the several species that a brain magnetic resonance imaging image is marked；
A kind of marker characteristic is selected, is established about between a kind of several described ROI features of marker characteristic
Correlation, formed correlated characteristic；
Several described brain magnetic resonance imaging image zooming-outs go out several described ROI features and several are described related special
Sign, is respectively formed ROI feature set and correlated characteristic set；
Feature selecting is carried out respectively to the ROI feature set and the correlated characteristic set by composite character algorithm,
Select optimal ROI feature subset and optimal correlated characteristic subset；
Weight factor of the ROI feature relative to correlated characteristic function ratio in classifier is set, institute is passed through
Optimal ROI feature subset described in weight factor and multicore SVM model integration, the optimal correlated characteristic subset are stated, is formed more
Kernel classifier.
Preferably, after carrying out region division to a brain magnetic resonance imaging image to extract several ROI features, further include
Step:
The several species marker characteristic in each ROI feature is normalized respectively.
Preferably, a kind of marker characteristic is selected, several described ROI about a kind of marker characteristic are established
Correlation between feature forms correlated characteristic；Specifically includes the following steps:
By vector that a kind of N number of ROI feature with marker characteristic is formed by calculating related coefficient be established as N ×
The correlation matrix of N, the phase in the correlation matrix between a kind of each element representation two ROI features with marker characteristic
Guan Xing；
By i-th ROI feature and a kind of j-th of ROI with marker characteristic with a kind of marker characteristic
Irrelevance between feature is defined as: d (i, j)=[t (i)-t (j)]2, wherein t (i) and t (j) respectively indicate i-th of ROI
A kind of characteristic value of the marker characteristic in feature, a kind of characteristic value of the marker characteristic in j-th of ROI feature；
Then, i-th of ROI feature and a kind of j-th of ROI with marker characteristic with a kind of marker characteristic
Correlation between feature is defined as:Wherein, δiAnd δjRespectively indicate i-th
The standard deviation of the marker characteristic characteristic value between a ROI feature and j-th of ROI feature；That is, s (i, j) is correlated characteristic.
Preferably, the composite character algorithm include to the ROI feature set and the correlated characteristic set successively
Carry out the first filtering characteristic selection algorithm, the second filtering characteristic selection algorithm and the package feature selection algorithm of feature selecting；
The first filtering characteristic selection algorithm is for reducing feature quantity；
The second filtering characteristic selection algorithm is minimal redundancy maximal correlation feature selection approach, obtains optimal characteristics
Collection；
The package feature selection algorithm is the recursive feature elimination algorithm based on support vector machines, and acquisition advanced optimizes
Optimal feature subset.
Preferably, by optimal ROI feature subset described in the weight factor and multicore SVM model integration, described
Optimal correlated characteristic subset forms multi-core classifier, comprising the following steps:
By being built respectively based on Radial basis kernel function to the optimal ROI feature subset and the optimal correlated characteristic subset
Vertical nuclear matrix；
N training sample is defined, defining the weight factor is βm；The then feature vector of i-th of sample are as follows:Wherein, M is the type of marker characteristic；The corresponding label of each feature vector is yi={ -1,1 }；
So, mixed nuclear matrix are as follows:Wherein,
And as 0≤aiWhen≤C,Φ () indicates the mapping function of kernel function guidance,
Indicate training sampleWithThe nuclear matrix in feature in m, a indicate Lagrange multiplier,<,>indicate inner product fortune
It calculates, C indicates the number of constraint condition in model parameter；
Therefore, multi-core classifier is
Preferably, the weight factor is βmValue range be 0.3-0.6.
It preferably, further include step to two layers of multi-core classifier progress of nested cross validation, including following step
It is rapid:
First layer cross validation is carried out to the multi-core classifier；
Second layer cross validation is carried out to the multi-core classifier；
A nested third layer cross validation outside the first layer cross validation with the second layer cross validation.
The present invention is include at least the following beneficial effects:
1) ROI feature is carried out to the extraction of correlated characteristic, setting ROI feature relative to related special to brain magnetic resonance imaging image
It levies the weight factor of function ratio in classifier, form multi-core classifier, realization high dimensional feature by multicore SVM model integration
Acquisition, analyze caused by related disease the part change simultaneously and function connects change, the accuracy of classification of diseases is high, can use
In the auxiliary diagnosis of various disease, there is stronger applicability；
2) individual difference for considering variety classes marker characteristic, to the several species marker characteristic in each ROI feature
It is normalized respectively, eliminates individual difference；
3) composite character algorithm carries out feature selecting to ROI feature set and correlated characteristic set respectively, and dimensionality reduction avoids tieing up
Number disaster, selects optimal ROI feature subset and optimal correlated characteristic subset；
4) the nested cross validation that two layers is carried out to multi-core classifier, further obtains optimal classification model.
Further advantage, target and feature of the invention will be partially reflected by the following instructions, and part will also be by this
The research and practice of invention and be understood by the person skilled in the art.
Detailed description of the invention
Fig. 1 is the multidimensional characteristic classification method flow chart of the present invention based on brain magnetic resonance imaging image；
Fig. 2 is the method flow schematic diagram that the brain MR Image Classifier that the embodiment of the present invention provides is established.
Specific embodiment
Present invention will be described in further detail below with reference to the accompanying drawings, to enable those skilled in the art referring to specification text
Word can be implemented accordingly.
It should be appreciated that such as " having ", "comprising" and " comprising " term used herein do not allot one or more
The presence or addition of a other elements or combinations thereof.
As depicted in figs. 1 and 2, the multidimensional characteristic sorting algorithm provided by the invention based on brain magnetic resonance imaging image, including
Following steps:
S10 carries out region division to a brain magnetic resonance imaging image to extract several ROI (region of
Interest, area-of-interest) feature, a ROI feature include a brain magnetic resonance imaging image is marked it is several
The marker characteristic of class；
S20 selects a kind of marker characteristic, establishes about the correlation between a kind of several ROI features of marker characteristic,
Form correlated characteristic；
S30, several brain magnetic resonance imaging image zooming-outs go out several ROI features and several correlated characteristics, respectively shape
At ROI feature set and correlated characteristic set；
S40 carries out feature selecting to ROI feature set and correlated characteristic set by composite character algorithm respectively, selection
Optimal ROI feature subset and optimal correlated characteristic subset out；
Weight factor of the ROI feature relative to correlated characteristic function ratio in classifier is arranged in S50, by weight because
The son and optimal ROI feature subset of multicore SVM (Support Vector Machine, support vector machines) model integration, optimal
Correlated characteristic subset forms multi-core classifier.
In above embodiment, brain magnetic resonance imaging image, i.e. brain MR image.Compared to only extracting ROI feature, correlated characteristic
It is a kind of new, more higher-dimension feature, the meaning that correlated characteristic extracts is for calculating in brain magnetic resonance imaging image not same district
Morphology correlation between domain.Therefore, extraction, the setting ROI of ROI feature and correlated characteristic are carried out to brain magnetic resonance imaging image
Feature relative to correlated characteristic in classifier the weight factor of function ratio, multicore point formed by multicore SVM model integration
Class device realizes the acquisition of high dimensional feature, can analyze the change of partial structurtes caused by related disease and function connects change simultaneously.It is logical
The classifier that brain magnetic resonance imaging Image Classifier method for building up provided by the invention is established is crossed, to brain magnetic resonance imaging image dependent part
The classification that the structure and related disease of position cause structure function connection to change, accuracy with higher, for not
With the auxiliary diagnosis of disease, there is stronger applicability.
In above embodiment, it is contemplated that the individual difference of variety classes marker characteristic, in step S10, to a brain core
After magnetic resonance image carries out region division to extract several ROI features, further comprise the steps of: to several in each ROI feature
Category flag feature is normalized respectively.Normalized eliminates the individual difference between each category flag feature,
Improve the accuracy rate of feature extraction.
In above embodiment, step S20, specifically includes the following steps:
S21, by vector that a kind of N number of ROI feature with marker characteristic is formed by calculating related coefficient be established as N ×
The correlation matrix of N, the correlation in the correlation matrix between a kind of each element representation two ROI features with marker characteristic
Property；
S22, by i-th ROI feature and a kind of j-th of ROI feature with marker characteristic with a kind of marker characteristic
Between irrelevance is defined as: d (i, j)=[t (i)-t (j)]2, wherein t (i) and t (j) respectively indicate i-th of ROI feature
A kind of characteristic value of marker characteristic in the characteristic value of middle marker characteristic a kind of, j-th of ROI feature；
S23, then, and i-th ROI feature and a kind of j-th of ROI feature with marker characteristic with a kind of marker characteristic
Between correlation is defined as:Wherein, δiAnd δjIt respectively indicates i-th
The standard deviation of marker characteristic characteristic value between ROI feature and j-th of ROI feature；That is, s (i, j) is correlated characteristic.
In above embodiment, composite character algorithm includes successively carrying out spy to ROI feature set and correlated characteristic set
Levy the first filtering characteristic selection algorithm, the second filtering characteristic selection algorithm and package feature selection algorithm of selection；First mistake
Filter feature selecting algorithm is for reducing feature quantity；Second filtering characteristic selection algorithm is minimal redundancy maximal correlation feature selecting
Method obtains optimal feature subset；Package feature selection algorithm is the recursive feature elimination algorithm based on support vector machines, is obtained
The optimal feature subset advanced optimized.Because ROI feature and correlated characteristic are all high dimensional features, composite character algorithm is to ROI
Characteristic set and correlated characteristic set carry out feature selecting respectively, and dimensionality reduction avoids dimension disaster, select optimal ROI feature
Collection and optimal correlated characteristic subset.
In step S50, pass through weight factor and the optimal ROI feature subset of multicore SVM model integration, optimal correlated characteristic
Subset forms multi-core classifier, comprising the following steps:
By establishing nuclear moment respectively to optimal ROI feature subset and optimal correlated characteristic subset based on Radial basis kernel function
Battle array；
N training sample is defined, definition weight factor is βm；The then feature vector of i-th of sample are as follows:Wherein, M is the type of marker characteristic；The corresponding label of each feature vector is yi={ -1,1 }；
So, mixed nuclear matrix are as follows:Wherein,
And as 0≤aiWhen≤C,Φ () indicates the mapping function of kernel function guidance,
Indicate training sampleWithThe nuclear matrix in feature in m, a indicate Lagrange multiplier,<,>indicate inner product fortune
It calculates, C indicates the number of constraint condition in model parameter；
Therefore, multi-core classifier is
In above embodiment, weight factor βmBigger, function of the ROI feature in multi-core classifier is bigger, as excellent
Choosing, weight factor βmValue range be 0.3-0.6.
Multidimensional characteristic classification method provided by the invention based on brain magnetic resonance imaging image, further comprises the steps of: S60, to more
Kernel classifier carries out two layers of nested cross validation.The following steps are included:
First layer cross validation is carried out to multi-core classifier；Second layer cross validation is carried out to multi-core classifier；First
A layer cross validation third layer cross validation nested with second layer cross validation outside.
In above embodiment, first layer cross validation recycles the hyper parameter for determining SVM model from training set,
Second layer cross validation circulation assesses the replicability of SVM model with independent verifying set.Two are carried out to multi-core classifier
The nested cross validation of layer further obtains optimal classification model.It is showed in the third layer cross validation of nested cross validation
Best SVM model is then optimal models, and hyper parameter will be used to test new data.
<embodiment 1>
On the basis of the multidimensional characteristic classification method for the brain magnetic resonance imaging image that above embodiment provides, the present embodiment
Provide the example of the multidimensional characteristic classification method based on brain MR image.
Firstly, carrying out region division to a brain MR image to extract several ROI features, a ROI feature includes pair
The marker characteristic for the several species that one brain MR image is marked, such as grey matter volume, white matter volume, cerebrospinal fluid volume, brain skin
The marker characteristic of the types such as thickness degree and cortex surface area.In order to eliminate individual difference, to above-mentioned various types of marker characteristic into
Row normalized, accordingly, the normalization of grey matter volume, white matter volume and cerebrospinal fluid volume are by by the body of each ROI
Product minimizes individual difference divided by encephalic total volume；Cortex thickness be by by the average skin thickness of each ROI divided by
Its standard deviation for corresponding to ROI is normalized；Cortex surface area be by by the cortex surface area of each ROI divided by
Full brain surface area eliminates individual difference.It should be noted that region division is using Montreal Neurological
Anatomical Automatic Labeling (AAL) Partition Mask that Institute (MNI) mechanism provides；ALL subregion mould
Brain is divided into 90 brain areas, left and right each 45, half brain by plate, and the structure below cerebral cortex is not studied because its structure is complicated, institute
With only 78 ROI for establishing feature vector, ROI under 12 cortexes is not studied.
Secondly, selecting a kind of marker characteristic, establish about the correlation between a kind of several ROI features of marker characteristic
Property, form correlated characteristic.The marker characteristic that the present embodiment selects is cortex thickness characteristics, and cortex thickness characteristics are capable of providing
The relevant information of brain function connection, therefore, correlated characteristic is established on the basis of cortex thickness characteristics.At this point, by N=
The Correlation Moment that the vector that 78 ROI features with cortex thickness characteristics are formed is established as 78 × 78 by calculating related coefficient
Gust, the correlation in the correlation matrix between each element representation two ROI features with cortex thickness characteristics；By i-th
Irrelevance between a ROI feature and j-th of ROI feature with cortex thickness characteristics with cortex thickness characteristics
Is defined as: d (i, j)=[t (i)-t (j)]2, wherein t (i) and t (j) respectively indicate i-th of ROI feature midbrain skin thickness
Value, j-th of ROI feature midbrain skin thickness value；Then, there is the ROI feature of cortex thickness characteristics and have for j-th for i-th
Correlation between the ROI feature of cortex thickness characteristics is defined as:Its
In, δiAnd δjRespectively indicate the standard deviation of cortex thickness value between i-th of ROI feature and j-th of ROI feature；That is, s (i, j)
For correlated characteristic.
Next, carrying out feature selecting respectively to ROI feature set and correlated characteristic set by composite character algorithm, select
Select out optimal ROI feature subset and optimal correlated characteristic subset.Composite character algorithm includes successively carrying out the first mistake of feature selecting
Filter feature selecting algorithm, the second filtering characteristic selection algorithm and package feature selection algorithm.First filtering characteristic selection algorithm
For reducing feature quantity；Second filtering characteristic selection algorithm is minimal redundancy maximal correlation feature selection approach, is further dropped
The dimension of low feature obtains optimal feature subset；Package feature selection algorithm is that the recursive feature based on support vector machines is eliminated
Algorithm is tieed up by one character subset of selection, and then obtains the optimal ROI feature subset and optimal phase advanced optimized
Close character subset.
Finally, weight factor of the setting ROI feature relative to correlated characteristic function ratio in classifier, by weight because
Son and the optimal ROI feature subset of multicore SVM model integration, optimal correlated characteristic subset, formation multi-core classifier, which is used, to be based on
Diameter integrates optimal ROI feature subset to the multicore SVM model of base kernel function (radial basis Function, RBF)
With optimal correlated characteristic subset to establish classifier.Specifically, by based on Radial basis kernel function to optimal ROI feature subset and
Optimal correlated characteristic subset establishes nuclear matrix respectively；N training sample is defined, definition weight factor is βm；Then i-th sample
Feature vector are as follows:Wherein, M is the type of marker characteristic；The corresponding label of each feature vector is yi
={ -1,1 }；So, mixed nuclear matrix are as follows:Wherein,And as 0≤aiWhen≤C,Φ () indicates that kernel function is drawn
The mapping function led,Indicate training sampleWithThe nuclear matrix in feature in m, a indicate glug
Bright day multiplier,<,>indicate that inner product operation, C indicate the number of constraint condition in model parameter；Therefore, multi-core classifier is
As the optimization of classifier, nested cross validation is carried out to the classifier of brain MR image, for example, first layer intersects
100 foldings are verified as, second layer cross validation is 2 foldings, and nested cross validation third layer cross validation is 10 foldings.In nested friendship
The SVM model to behave oneself best in fork verifying is then optimal models, and hyper parameter will be used to test new data.
<comparative example 1>
On the basis of embodiment 1, this comparative example 1 is illustrated logical based on the analysis of cognitive function in disturbances in patients with Parkinson disease
It crosses and obtains each performance indicator and to analyze respectively using multidimensional characteristic classification method and single features classification method.
Cognition dysfunction is a kind of complication of Parkinson's disease, and illness rate is higher, using multidimensional ROI feature and engineering
Algorithm is practised to study cognition dysfunction caused by Parkinson's disease, doctor, patient, family members and researcher can be helped to find more
Good method carries out early diagnosis and therapy.After table 1 gives single ROI feature and multidimensional ROI feature difference classification processing
Each performance indicator.
Each performance indicator after the single ROI feature of table 1 and multidimensional ROI feature difference classification processing
As shown in Table 1, be respectively 92.35% based on classification accuracy of the multidimensional ROI feature in three groups, 83.95%,
80.84%, when being compared due to Parkinson's group with normal group, two groups of subject diversity ratios are larger, so being divided using each category feature
The classification accuracy that class obtains is above the classification accuracy that single ROI feature is classified, it is seen that special based on multidimensional ROI
Sign has good resolution capability.
<comparative example 2>
On the basis of embodiment 1, this comparative example 2 is illustrated based on the analysis of self-respect degree and brain structure by making
It obtains each performance indicator and to analyze respectively with multidimensional characteristic classification method and single features classification method.
In view of the difference of brain structure will lead to the difference of brain function, subject is divided into two groups by this experiment: high level of self respect
Group and low level of self respect group study the brain architectural difference between two groups with multidimensional ROI feature and machine learning algorithm.Table 2 provides
Each performance indicator after single ROI feature and multidimensional ROI feature difference classification processing.
Each performance indicator of the 2 two groups of subjects of table after single ROI feature and multidimensional ROI feature difference classification processing
As shown in Table 2, when only being classified with grey matter volume, classification accuracy is minimum, is 65.63%, with related special
When sign is classified, obtained classification accuracy is 80.71%, and classifying quality is better than using single ROI feature, and uses
When multidimensional characteristic, classification accuracy highest reaches 87.33%, illustrates that multidimensional characteristic preferably can be used to distinguish two groups of subjects.
Although the embodiments of the present invention have been disclosed as above, but its is not only in the description and the implementation listed
With.It can be applied to various suitable the field of the invention completely.It for those skilled in the art can be easily
Realize other modification.Therefore without departing from the general concept defined in the claims and the equivalent scope, the present invention is simultaneously unlimited
In specific details and legend shown and described herein.
Claims (6)
1. a kind of multidimensional characteristic classification method based on brain magnetic resonance imaging image, which comprises the following steps:
Region division is carried out to a brain magnetic resonance imaging image to extract several ROI features, a ROI feature includes pair
The marker characteristic for the several species that one brain magnetic resonance imaging image is marked；
A kind of marker characteristic is selected, is established about the phase between a kind of several described ROI features of marker characteristic
Guan Xing forms correlated characteristic；
Several described brain magnetic resonance imaging image zooming-outs go out several described ROI features and several described correlated characteristics, point
It Xing Cheng not ROI feature set and correlated characteristic set；
By composite character algorithm feature selecting is carried out to the ROI feature set and the correlated characteristic set respectively, selected
Optimal ROI feature subset and optimal correlated characteristic subset out；
Weight factor of the ROI feature relative to correlated characteristic function ratio in classifier is set, the power is passed through
Optimal ROI feature subset described in repeated factor and multicore SVM model integration, the optimal correlated characteristic subset form multicore point
Class device；
Wherein, the composite character algorithm includes successively carrying out feature to the ROI feature set and the correlated characteristic set
The first filtering characteristic selection algorithm, the second filtering characteristic selection algorithm and the package feature selection algorithm of selection；
The first filtering characteristic selection algorithm is for reducing feature quantity；
The second filtering characteristic selection algorithm is minimal redundancy maximal correlation feature selection approach, obtains optimal feature subset；
The package feature selection algorithm is the recursive feature elimination algorithm based on support vector machines, obtains and advanced optimizes most
Excellent character subset.
2. as described in claim 1 based on the multidimensional characteristic classification method of brain magnetic resonance imaging image, which is characterized in that one
After brain magnetic resonance imaging image carries out region division to extract several ROI features, further comprise the steps of:
The several species marker characteristic in each ROI feature is normalized respectively.
3. as described in claim 1 based on the multidimensional characteristic classification method of brain magnetic resonance imaging image, which is characterized in that selection one
The kind marker characteristic is established about the correlation between a kind of several described ROI features of marker characteristic, forms phase
Close feature；Specifically includes the following steps:
The vector that a kind of N number of ROI feature with marker characteristic is formed is established as N × N's by calculating related coefficient
Correlation matrix, the correlation in the correlation matrix between a kind of each element representation two ROI features with marker characteristic
Property；
By i-th ROI feature and a kind of j-th of ROI feature with marker characteristic with a kind of marker characteristic
Between irrelevance is defined as: d (i, j)=[t (i)-t (j)]2, wherein t (i) and t (j) respectively indicate i-th of ROI feature
A kind of characteristic value of the marker characteristic in the characteristic value of middle marker characteristic a kind of, j-th of ROI feature；
Then, i-th of ROI feature and a kind of j-th of ROI feature with marker characteristic with a kind of marker characteristic
Between correlation is defined as:Wherein, δiAnd δjIt respectively indicates i-th
The standard deviation of the marker characteristic characteristic value between ROI feature and j-th of ROI feature；That is, s (i, j) is correlated characteristic.
4. as described in claim 1 based on the multidimensional characteristic classification method of brain magnetic resonance imaging image, which is characterized in that pass through institute
Optimal ROI feature subset described in weight factor and multicore SVM model integration, the optimal correlated characteristic subset are stated, is formed more
Kernel classifier, comprising the following steps:
By establishing core respectively to the optimal ROI feature subset and the optimal correlated characteristic subset based on Radial basis kernel function
Matrix；
N training sample is defined, defining the weight factor is βm；The then feature vector of i-th of sample are as follows: xi={ xi (1)..., xi (M)}；Wherein, M is the type of marker characteristic；The corresponding label of each feature vector is yi={ -1,1 }；
So, mixed nuclear matrix are as follows:Wherein, k(m)(xi (m), xj (m))=< Φ (xi (m)),Φ(xj (m)) >；
And as 0≤aiWhen≤C,Φ () indicates the mapping function of kernel function guidance, k(m)(xi (m), xj (m)) table
Show training sample xi (m)And xj (m)The nuclear matrix in feature in m, a indicate Lagrange multiplier,<,>indicate inner product operation,
C indicates the number of constraint condition in model parameter；
Therefore, multi-core classifier is
5. as claimed in claim 4 based on the multidimensional characteristic classification method of brain magnetic resonance imaging image, which is characterized in that the power
Repeated factor is βmValue range be 0.3-0.6.
6. as described in claim 4 or 5 based on the multidimensional characteristic classification method of brain magnetic resonance imaging image, which is characterized in that also
Two layers of nested cross validation is carried out to the multi-core classifier including step, comprising the following steps:
First layer cross validation is carried out to the multi-core classifier；
Second layer cross validation is carried out to the multi-core classifier；
A nested third layer cross validation outside the first layer cross validation with the second layer cross validation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610886193.4A CN106485707B (en) | 2016-10-11 | 2016-10-11 | Multidimensional characteristic classification method based on brain magnetic resonance imaging image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610886193.4A CN106485707B (en) | 2016-10-11 | 2016-10-11 | Multidimensional characteristic classification method based on brain magnetic resonance imaging image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106485707A CN106485707A (en) | 2017-03-08 |
| CN106485707B true CN106485707B (en) | 2019-05-28 |
Family
ID=58270529
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610886193.4A Active CN106485707B (en) | 2016-10-11 | 2016-10-11 | Multidimensional characteristic classification method based on brain magnetic resonance imaging image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106485707B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106919930A (en) * | 2017-03-10 | 2017-07-04 | 成都智锗科技有限公司 | A kind of low resolution infrared image parahypnosis situation determination methods |
| CN107194417B (en) * | 2017-05-09 | 2020-06-05 | 北京积水潭医院 | Finger damage detection method and system based on MR image |
| CN108154924A (en) * | 2017-12-07 | 2018-06-12 | 广州市本真网络科技有限公司 | Alzheimer's disease tagsort method and system based on support vector machines |
| CN108764367B (en) * | 2018-06-07 | 2023-05-12 | 深圳大学 | Feature image extraction device and method based on relation regularization |
| CN109800812A (en) * | 2019-01-24 | 2019-05-24 | 山东大学第二医院 | CT image classification feature selection approach and system based on counterfeit filter |
| CN111797859A (en) * | 2019-04-09 | 2020-10-20 | Oppo广东移动通信有限公司 | Feature processing method, feature processing device, storage medium and electronic equipment |
| CN110428396A (en) * | 2019-06-25 | 2019-11-08 | 郑州大学第一附属医院 | A kind of feature vectors dimensional down method and system based on CT images |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104573742A (en) * | 2014-12-30 | 2015-04-29 | 中国科学院深圳先进技术研究院 | Medical image classification method and system |
| CN104834935A (en) * | 2015-04-27 | 2015-08-12 | 电子科技大学 | Stable brain tumor non-supervised nosology imaging method |
| CN104881687A (en) * | 2015-06-02 | 2015-09-02 | 四川理工学院 | Magnetic resonance image classification method based on semi-supervised Gaussian mixed model |
| CN105809175A (en) * | 2014-12-30 | 2016-07-27 | 深圳先进技术研究院 | Encephaledema segmentation method and system based on support vector machine algorithm |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8422756B2 (en) * | 2010-04-27 | 2013-04-16 | Magnetic Resonance Innovations, Inc. | Method of generating nuclear magnetic resonance images using susceptibility weighted imaging and susceptibility mapping (SWIM) |
| US8326010B2 (en) * | 2010-05-03 | 2012-12-04 | General Electric Company | System and method for nuclear magnetic resonance (NMR) temperature monitoring |
-
2016
- 2016-10-11 CN CN201610886193.4A patent/CN106485707B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104573742A (en) * | 2014-12-30 | 2015-04-29 | 中国科学院深圳先进技术研究院 | Medical image classification method and system |
| CN105809175A (en) * | 2014-12-30 | 2016-07-27 | 深圳先进技术研究院 | Encephaledema segmentation method and system based on support vector machine algorithm |
| CN104834935A (en) * | 2015-04-27 | 2015-08-12 | 电子科技大学 | Stable brain tumor non-supervised nosology imaging method |
| CN104881687A (en) * | 2015-06-02 | 2015-09-02 | 四川理工学院 | Magnetic resonance image classification method based on semi-supervised Gaussian mixed model |
Non-Patent Citations (1)
| Title |
|---|
| 基于磁共振图像的脑肿瘤自动识别与分析;高宏建等;《北京工业大学学报》;20120630;第38卷(第6期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106485707A (en) | 2017-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106485707B (en) | Multidimensional characteristic classification method based on brain magnetic resonance imaging image | |
| CN111833359B (en) | Brain tumor segmentation data enhancement method based on generation of confrontation network | |
| O’Donnell et al. | Automated white matter fiber tract identification in patients with brain tumors | |
| CN111047589B (en) | Attention-enhanced brain tumor auxiliary intelligent detection and identification method | |
| Ramzan et al. | Volumetric segmentation of brain regions from MRI scans using 3D convolutional neural networks | |
| CN107103612B (en) | Automate the quantitative calculation method of subregion brain atrophy | |
| CN111047594A (en) | Tumor MRI weak supervised learning analysis modeling method and model thereof | |
| CN104414636B (en) | Cerebral microbleeds computer-aided detection system based on MRI | |
| Reda et al. | A novel adcs-based cnn classification system for precise diagnosis of prostate cancer | |
| CN106127263A (en) | The human brain magnetic resonance image (MRI) classifying identification method extracted based on three-dimensional feature and system | |
| Zhang et al. | Analysis of structural brain MRI and multi-parameter classification for Alzheimer’s disease | |
| Ptak et al. | Visual object agnosia is associated with a breakdown of object-selective responses in the lateral occipital cortex | |
| Montagne et al. | 3D Local Binary Pattern for PET image classification by SVM, Application to early Alzheimer disease diagnosis | |
| Mareček et al. | Automated fusion of multimodal imaging data for identifying epileptogenic lesions in patients with inconclusive magnetic resonance imaging | |
| CN108520283A (en) | A kind of classification method of medical image based on brain area indirect relation network struction | |
| Wei et al. | Differential brain mechanisms for processing distracting information in task‐relevant and‐irrelevant dimensions in visual search | |
| Barton | Face processing in the temporal lobe | |
| Li et al. | Diffusion tensor-based fast marching for modeling human brain connectivity network | |
| Thome et al. | “I spy with my little eye, something that is a face…”: a brain network for illusory face detection | |
| Mattia et al. | Investigating the discrimination ability of 3d convolutional neural networks applied to altered brain mri parametric maps | |
| KR20020079742A (en) | Convolution filtering of similarity data for visual display of enhanced image | |
| Stough et al. | Thalamic parcellation from multi-modal data using random forest learning | |
| Liu et al. | MPS-FFA: a multiplane and multiscale feature fusion attention network for Alzheimer’s disease prediction with structural MRI | |
| KR102373992B1 (en) | Method and apparatut for alzheimer's disease classification using texture features | |
| Uma et al. | A comparative analysis of brain tumor segmentation techniques |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 |