CN106485707A - Multidimensional characteristic sorting algorithm based on brain magnetic resonance imaging image - Google Patents
Multidimensional characteristic sorting algorithm based on brain magnetic resonance imaging image Download PDFInfo
- Publication number
- CN106485707A CN106485707A CN201610886193.4A CN201610886193A CN106485707A CN 106485707 A CN106485707 A CN 106485707A CN 201610886193 A CN201610886193 A CN 201610886193A CN 106485707 A CN106485707 A CN 106485707A
- Authority
- CN
- China
- Prior art keywords
- feature
- roi
- features
- magnetic resonance
- 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.)
- Granted
Links
- 210000004556 brain Anatomy 0.000 title claims abstract description 54
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 44
- 238000002595 magnetic resonance imaging Methods 0.000 title 1
- 239000003550 marker Substances 0.000 claims abstract description 21
- 238000005481 NMR spectroscopy Methods 0.000 claims description 35
- 238000002790 cross-validation Methods 0.000 claims description 30
- 238000007635 classification algorithm Methods 0.000 claims description 24
- 238000012706 support-vector machine Methods 0.000 claims description 19
- 238000001914 filtration Methods 0.000 claims description 16
- 238000000034 method Methods 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 6
- 238000004806 packaging method and process Methods 0.000 claims description 5
- 230000008030 elimination Effects 0.000 claims description 4
- 238000003379 elimination reaction Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 238000010187 selection method Methods 0.000 claims description 4
- 238000005538 encapsulation Methods 0.000 claims description 3
- 201000010099 disease Diseases 0.000 abstract description 13
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 13
- 238000003745 diagnosis Methods 0.000 abstract description 5
- 230000008859 change Effects 0.000 abstract description 3
- 230000002596 correlated effect Effects 0.000 abstract 6
- 239000002131 composite material Substances 0.000 abstract 1
- 239000000284 extract Substances 0.000 abstract 1
- 230000010354 integration Effects 0.000 abstract 1
- 230000006870 function Effects 0.000 description 13
- 230000001054 cortical effect Effects 0.000 description 12
- 230000000052 comparative effect Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000002059 diagnostic imaging Methods 0.000 description 3
- 210000004884 grey matter Anatomy 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 238000002372 labelling Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 208000018737 Parkinson disease Diseases 0.000 description 2
- 210000001175 cerebrospinal fluid Anatomy 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 208000010877 cognitive disease Diseases 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 210000004885 white matter Anatomy 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000003925 brain function Effects 0.000 description 1
- 210000003710 cerebral cortex Anatomy 0.000 description 1
- 230000003920 cognitive function Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013399 early diagnosis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000007917 intracranial administration Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000926 neurological effect Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000002739 subcortical effect Effects 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Radiology & Medical Imaging (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Quality & Reliability (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The open multidimensional characteristic sorting algorithm based on brain MR image of the present invention, carries out region division to brain MR image and extracts several ROI feature；Select a kind of marker characteristic, set up and form correlated characteristic with regard to the dependency between several ROI feature of this marker characteristic；Several brains MR image zooming-out goes out several ROI feature and several correlated characteristics form ROI feature set and correlated characteristic set；Optimum ROI feature subset and optimum correlated characteristic subset are selected respectively by composite character algorithm to ROI feature set and correlated characteristic set；The weight factor of setting ROI feature function ratio in grader, forms multi-core classifier by weight factor and multinuclear SVM model integration optimum ROI feature subset, optimum correlated characteristic subset.The present invention obtains high dimensional feature, can analyze this change of local and the function connects change that relevant disease causes simultaneously, and classification accuracy is high, can auxiliary diagnosis various disease.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a multi-dimensional feature classification algorithm based on brain nuclear magnetic resonance images.
Background
Medical imaging refers to the technique and process of obtaining images of internal tissues of a human body or a part of the human body in a non-invasive manner for medical treatment or medical research. It contains the following two relatively independent directions of study: medical imaging systems (medical imaging systems) and medical image processing (medical image processing). The former refers to the process of image formation, including the research on the problems of imaging mechanism, imaging equipment, imaging system analysis and the like; the latter refers to further processing of the acquired images, either to restore the original less sharp image, to highlight some feature information in the image, to classify the pattern of the image, or the like.
In recent years, the auxiliary diagnosis and prediction of related diseases by using machine learning algorithm based on brain nuclear magnetic resonance images is a research focus nowadays. The existing features based on the brain nuclear magnetic resonance image are mainly based on the morphological features of voxels, a classification algorithm established by utilizing the features can only find out structural changes of specific parts and cannot analyze the changes of functional connections of the specific parts, and a good classification algorithm for clinical use cannot be found out at present, and can simultaneously analyze the changes of human body structures and the structural functional connections on the basis of the brain nuclear magnetic resonance image, so that the classification algorithm for realizing related diseases by extracting higher-dimensional features from the brain nuclear magnetic resonance image and establishing a classifier is required.
Disclosure of Invention
Aiming at the defects in the technology, the invention provides a multi-dimensional feature classification algorithm based on a brain nuclear magnetic resonance image, which is used for extracting ROI features and related features from the brain nuclear magnetic resonance image, setting weight factors of the ROI features relative to the function proportion of the related features in a classifier, and integrating through a multi-core SVM model to form the multi-core classifier, so that the high-dimensional features are obtained, local changes and function connection changes caused by related diseases are analyzed, the accuracy of disease classification is high, the multi-dimensional feature classification algorithm can be used for auxiliary diagnosis of different diseases, and the multi-core classifier has strong applicability.
To achieve these objects and other advantages in accordance with the present invention, the present invention is implemented by the following solutions:
the invention relates to a multidimensional feature classification algorithm based on brain nuclear magnetic resonance images, which comprises the following steps:
performing region division on a brain nuclear magnetic resonance image to extract a plurality of ROI features, wherein one ROI feature comprises a plurality of types of marking features for marking one brain nuclear magnetic resonance image;
selecting one of said marker features, establishing a correlation between a number of said ROI features with respect to one of said marker features, forming a correlation feature;
extracting a plurality of ROI features and a plurality of relevant features from the plurality of brain nuclear magnetic resonance images to respectively form an ROI feature set and a relevant feature set;
respectively selecting the features of the ROI feature set and the related feature set through a mixed feature algorithm, and selecting an optimal ROI feature subset and an optimal related feature subset;
and setting a weight factor of the ROI feature relative to the functional proportion of the related feature in the classifier, and integrating the optimal ROI feature subset and the optimal related feature subset through the weight factor and a multi-kernel SVM model to form the multi-kernel classifier.
Preferably, after performing region division on a brain nuclear magnetic resonance image to extract a plurality of ROI features, the method further includes the steps of:
and respectively carrying out normalization processing on the plurality of kinds of marked features in each ROI feature.
Preferably, one of said marker features is selected, correlation between a number of said ROI features with respect to one of said marker features is established, forming a correlation feature; the method specifically comprises the following steps:
establishing a vector formed by N ROI features with one kind of the marked feature as an N multiplied by N correlation matrix by calculating a correlation coefficient, wherein each element in the correlation matrix represents the correlation between two ROI features with one kind of the marked feature;
defining the irrelevance between the ith ROI feature having one of said marker features and the jth ROI feature having one of said marker features as: d (i, j) ═ t (i) -t (j)]2Wherein t (i) and t (j) respectively represent the feature value of one of the marking features in the i-th ROI feature and the feature value of one of the marking features in the j-th ROI feature;
then, the correlation between the ith ROI feature having one of the labeled features and the jth ROI feature having one of the labeled features is defined as:wherein,iandjrespectively representing the standard deviation of the mark feature values between the ith ROI feature and the jth ROI feature; that is, s (i, j) is a correlation feature.
Preferably, the mixed feature algorithm includes a first filtering feature selection algorithm, a second filtering feature selection algorithm and an encapsulation feature selection algorithm, which sequentially perform feature selection on the ROI feature set and the related feature set;
the first filtering feature selection algorithm is used for reducing the number of features;
the second filtering feature selection algorithm is a minimum redundancy maximum correlation feature selection method, and an optimal feature subset is obtained;
the packaging feature selection algorithm is a recursive feature elimination algorithm based on a support vector machine, and further optimized optimal feature subsets are obtained.
Preferably, the optimal ROI feature subset and the optimal relevant feature subset are integrated by the weighting factor and the multi-kernel SVM model to form the multi-kernel classifier, which includes the following steps:
respectively establishing kernel matrixes for the optimal ROI feature subset and the optimal related feature subset based on a radial basis kernel function;
defining n training samples, defining the weight factor as βm(ii) a Then the feature vector for the ith sample is: x is the number ofi＝{xi (1)，...，xi (M)}; wherein M is the type of the marking feature; the label corresponding to each feature vector is yi＝{-1,1}；
Then, the mixed kernel matrix is:wherein k is(m)(xi (m)，xj (m))＝＜Φ(xi (m)),Φ(xj (m))＞；
And when 0 is less than or equal to aiWhen the content is less than or equal to C,phi (-) represents a kernel-function-guided mapping function, k(m)(xi (m)，xj (m)) Representing a training sample xi (m)And xj (m)In the m, a represents a Lagrange multiplier, < - > represents inner product operation, and C represents the number of constraint conditions in the model parameters;
thus, the multi-core classifier is
Preferably, the weighting factor is βmThe value range of (A) is 0.3-0.6.
Preferably, the method further comprises the step of performing two-layer nested cross validation on the multi-core classifier, and comprises the following steps:
performing first-layer cross validation on the multi-class classifier;
performing second-layer cross validation on the multi-class classifier;
nesting a third level cross-validation outside of said first level cross-validation and said second level cross-validation.
The invention at least comprises the following beneficial effects:
1) extracting ROI (region of interest) characteristics and related characteristics of the brain nuclear magnetic resonance image, setting weight factors of the ROI characteristics relative to the function proportion of the related characteristics in a classifier, and integrating the ROI characteristics and the related characteristics through a multi-core SVM (support vector machine) model to form a multi-core classifier, so that high-dimensional characteristics are obtained, local changes and function connection changes caused by related diseases are analyzed, the accuracy of disease classification is high, the method can be used for auxiliary diagnosis of different diseases, and the method has high applicability;
2) considering the individual difference of different types of marking features, respectively carrying out normalization processing on a plurality of types of marking features in each ROI feature to eliminate the individual difference;
3) the mixed feature algorithm respectively selects features of the ROI feature set and the related feature set, reduces dimensions, avoids dimension disasters, and selects an optimal ROI feature subset and an optimal related feature subset;
4) and performing two-layer nested cross validation on the multi-core classifier to further obtain an optimal classification model.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Drawings
FIG. 1 is a flow chart of a multi-dimensional feature classification algorithm based on brain nuclear magnetic resonance images according to the present invention;
fig. 2 is a flowchart illustrating a method for establishing a brain MR image classifier according to an embodiment of the present invention.
Detailed Description
The present invention is further described in detail below with reference to the attached drawings so that those skilled in the art can implement the invention by referring to the description text.
It will be understood that terms such as "having," "including," and "comprising," as used herein, do not preclude the presence or addition of one or more other elements or groups thereof.
As shown in fig. 1 and fig. 2, the multidimensional feature classification algorithm based on brain nuclear magnetic resonance images provided by the present invention includes the following steps:
s10, performing region division on a brain nmr image to extract a plurality of ROI (region of interest) features, where one ROI feature includes a plurality of kinds of labeling features for labeling a brain nmr image;
s20, selecting a marking feature, establishing correlation among a plurality of ROI features related to the marking feature, and forming a correlation feature;
s30, extracting a plurality of ROI features and a plurality of relevant features from the brain nuclear magnetic resonance images to respectively form an ROI feature set and a relevant feature set;
s40, respectively selecting the ROI feature set and the related feature set through a mixed feature algorithm, and selecting an optimal ROI feature subset and an optimal related feature subset;
s50, setting a weight factor of the ROI feature relative to the function proportion of the related feature in the classifier, and integrating the optimal ROI feature subset and the optimal related feature subset through the weight factor and a multi-core SVM (Support Vector Machine) model to form the multi-core classifier.
In the above embodiments, the brain magnetic resonance image is a brain MR image. Compared with only extracting the ROI feature, the relevant feature is a new and higher-dimensional feature, and the significance of relevant feature extraction is to calculate morphological correlation between different regions in the brain nuclear magnetic resonance image. Therefore, the method comprises the steps of extracting ROI features and relevant features of the brain nuclear magnetic resonance images, setting weight factors of the ROI features relative to the relevant features in a classifier, and integrating the ROI features and the relevant features through a multi-kernel SVM model to form the multi-kernel classifier, so that high-dimensional features are obtained, and local structural changes and functional connection changes caused by relevant diseases can be analyzed simultaneously. The classifier established by the method for establishing the brain nuclear magnetic resonance image classifier provided by the invention has higher accuracy on the classification of structural function connection change caused by the structure of the relevant part of the brain nuclear magnetic resonance image and relevant diseases, so that the classifier is used for auxiliary diagnosis of different diseases and has stronger applicability.
In the above embodiment, in consideration of individual differences of different kinds of marker features, after performing region division on one brain nuclear magnetic resonance image to extract a plurality of ROI features in step S10, the method further includes the steps of: and respectively carrying out normalization processing on a plurality of kinds of marked features in each ROI feature. And (4) normalization processing is carried out, so that individual differences among the labeled features of each type are eliminated, and the accuracy of feature extraction is improved.
In the above embodiment, step S20 specifically includes the following steps:
s21, establishing a vector formed by N ROI features with one marker feature into an N multiplied by N correlation matrix by calculating a correlation coefficient, wherein each element in the correlation matrix represents the correlation between two ROI features with one marker feature;
s22, defining the irrelevance between the ith ROI feature with one marker feature and the jth ROI feature with one marker feature as: d (i, j) ═ t (i) -t (j)]2Wherein t (i) and t (j) respectively represent the characteristic value of one marking characteristic in the ith ROI characteristic and the characteristic value of one marking characteristic in the jth ROI characteristic;
s23, the correlation between the ith ROI feature with a marker feature and the jth ROI feature with a marker feature is defined as:wherein,iandjrespectively representing the standard deviation of the mark characteristic values between the ith ROI characteristic and the jth ROI characteristic; that is, s (i, j) is a correlation feature.
In the above embodiment, the mixed feature algorithm includes a first filtering feature selection algorithm, a second filtering feature selection algorithm, and an encapsulation feature selection algorithm, which sequentially perform feature selection on the ROI feature set and the relevant feature set; a first filtering feature selection algorithm is used for reducing the number of features; the second filtering feature selection algorithm is a minimum redundancy maximum correlation feature selection method, and an optimal feature subset is obtained; the packaging feature selection algorithm is a recursive feature elimination algorithm based on a support vector machine, and an optimal feature subset for further optimization is obtained. Because the ROI features and the related features are high-dimensional features, the mixed feature algorithm respectively selects and reduces the dimensions of the ROI feature set and the related feature set, avoids dimension disasters and selects an optimal ROI feature subset and an optimal related feature subset.
In step S50, integrating the optimal ROI feature subset and the optimal relevant feature subset by the weight factor and the multi-kernel SVM model to form a multi-kernel classifier, including the following steps:
respectively establishing a kernel matrix for the optimal ROI feature subset and the optimal related feature subset based on a radial basis kernel function;
defining n training samples and defining a weight factor of βm(ii) a Then the feature vector for the ith sample is: x is the number ofi＝{xi (1)，...，xi (M)}; wherein M is the type of the marking feature; the label corresponding to each feature vector is yi＝{-1,1}；
Then, the mixed kernel matrix is:wherein k is(m)(xi (m)，xj (m))＝＜Φ(xi (m)),Φ(xj (m))＞；
And when 0 is less than or equal to aiWhen the content is less than or equal to C,phi (-) represents a kernel-function-guided mapping function, k(m)(xi (m)，xj (m)) Representing a training sample xi (m)And xj (m)In the m, a represents a Lagrange multiplier, < - > represents inner product operation, and C represents the number of constraint conditions in the model parameters;
thus, the multi-core classifier is
In the above embodiment, the weighting factor βmThe larger the ROI feature, the more functional it is in the multi-core classifier, and preferably the weighting factor βmThe value range of (A) is 0.3-0.6.
The invention provides a multidimensional feature classification algorithm based on brain nuclear magnetic resonance images, which further comprises the following steps: and S60, performing two-layer nested cross validation on the multi-core classifier. The method comprises the following steps:
performing first-layer cross validation on the multi-class classifier; performing second-layer cross validation on the multi-class classifier; a third level cross-validation is nested outside the first level cross-validation and the second level cross-validation.
In the above embodiment, the first layer of cross validation loop is used to determine the hyper-parameters of the SVM model from the training set, and the second layer of cross validation loop uses an independent validation set to evaluate the generalizability of the SVM model. And performing two-layer nested cross validation on the multi-core classifier to further obtain an optimal classification model. The SVM model that performs best in the third level of cross-validation of nested cross-validation is then the optimal model, whose hyper-parameters will be used to test the new data.
<Example 1>
On the basis of the multi-dimensional feature classification algorithm of brain nuclear magnetic resonance images provided by the above embodiments, the present embodiment gives an example of a multi-dimensional feature classification algorithm based on brain MR images.
Firstly, a brain MR image is subjected to region division to extract a plurality of ROI features, wherein one ROI feature comprises a plurality of types of marking features for marking the brain MR image, such as marking features of gray matter volume, white matter volume, cerebrospinal fluid volume, cortex thickness, cortex surface area and the like. To eliminate individual differences, normalization of the various above-mentioned kinds of signature features was performed, and accordingly, gray matter volume, white matter volume, and cerebrospinal fluid volume were normalized by dividing the volume of each ROI by the total intracranial volume to minimize individual differences; cortical thickness is normalized by dividing the mean cortical thickness of each ROI by the standard deviation of its corresponding ROI; cortical surface area individual differences were eliminated by dividing the cortical surface area of each ROI by the whole brain surface area. It should be noted that, the regional division adopts the Atomic Automatic Labeling (AAL) partition template provided by Montreal Neurological Institute (MNI) organization; the ALL partition template divides the brain into 90 brain regions, 45 each for the left and right half-brains, and the structure under the cerebral cortex is not studied because of its complex structure, so that only 78 ROIs are used for establishing the feature vector, and 12 subcortical ROIs are not studied.
In this case, a vector formed by N78 ROI features having cortical thickness features is calculated to form a correlation matrix of 78 × 78, each element in the correlation matrix represents a correlation between two ROI features having cortical thickness features, and a non-correlation between the ith ROI feature having cortical thickness features and the jth ROI feature having cortical thickness features is defined as d (i, j) ═ t (j) — (t (i-t (j))]2Wherein t (i) and t (j) represent the cortical thickness value in the ith ROI feature and the cortical thickness value in the jth ROI feature respectively; then, the correlation between the ith ROI feature with cortical thickness features and the jth ROI feature with cortical thickness features is defined as:wherein,iandjrespectively representing the standard deviation of the cortex thickness value between the ith ROI feature and the jth ROI feature; that is, s (i, j) is a correlation feature.
And then, respectively selecting the features of the ROI feature set and the related feature set through a mixed feature algorithm, and selecting an optimal ROI feature subset and an optimal related feature subset. The mixed feature algorithm comprises a first filtering feature selection algorithm, a second filtering feature selection algorithm and a packaging feature selection algorithm which are sequentially performed. A first filtering feature selection algorithm is used for reducing the number of features; the second filtering feature selection algorithm is a minimum redundancy maximum correlation feature selection method, so that the feature dimensionality is further reduced, and an optimal feature subset is obtained; the packaging feature selection algorithm is a recursive feature elimination algorithm based on a support vector machine, and dimension is carried out by selecting a feature subset, so that an optimal ROI feature subset and an optimal related feature subset which are further optimized are obtained.
Finally, setting a weight factor of the ROI feature relative to the Function proportion of the related feature in a classifier, integrating an optimal ROI feature subset and an optimal related feature subset through the weight factor and a multi-kernel SVM model to form a multi-kernel classifier, and integrating the optimal ROI feature subset and the optimal related feature subset by using a multi-kernel SVM model based on a radial, i.e. Radial Basis Function (RBF) to establish the classifierm(ii) a Then the feature vector for the ith sample is: x is the number ofi＝{xi (1)，...，xi (M)}; wherein M is the type of the marking feature; the label corresponding to each feature vector is yi-1,1 }; then, the mixed kernel matrix is:wherein k is(m)(xi (m)，xj (m))＝＜Φ(xi (m)),Φ(xj (m)) >; and when 0 is less than or equal to aiWhen the content is less than or equal to C,phi (-) represents a kernel-function-guided mapping function, k(m)(xi (m)，xj (m)) Representing a training sample xi (m)And xj (m)In the m, a represents a Lagrange multiplier, < - > represents inner product operation, and C represents the number of constraint conditions in the model parameters; thus, the multi-core classifier is
As an optimization of the classifier, nested cross-validation is performed on the classifier of the brain MR images, for example, the first layer cross-validation is 100-fold, the second layer cross-validation is 2-fold, and the nested cross-validation third layer cross-validation is 10-fold. The SVM model that performs best in nested cross-validation is then the optimal model, whose hyper-parameters will be used to test the new data.
<Comparative example 1>
On the basis of example 1, in comparative example 1, each performance index obtained by using a multidimensional feature classification algorithm and a single feature classification algorithm was explained and analyzed based on the analysis of cognitive function in parkinson patients.
The cognitive dysfunction is a complication of the Parkinson's disease, the prevalence rate is high, the cognitive dysfunction caused by the Parkinson's disease is researched by applying a multi-dimensional ROI characteristic and a machine learning algorithm, and doctors, patients, family members and researchers can be helped to find a better method for early diagnosis and treatment. Table 1 shows the performance indexes after classification processing of the single ROI feature and the multi-dimensional ROI feature.
TABLE 1 Performance indices after respective classification of Single ROI features and Multi-dimensional ROI features
As can be seen from table 1, the classification accuracy rates of the three groups based on the multidimensional ROI features are 92.35%, 83.95% and 80.84%, respectively, and since the difference between the two tested groups is large when the parkinson group and the normal group are compared, the classification accuracy rate obtained by classifying with various features is higher than that obtained by classifying with a single ROI feature, and it can be seen that the three groups have good resolution capability based on the multidimensional ROI features.
<Comparative example 2>
On the basis of example 1, this comparative example 2 explains and analyzes each performance index obtained by using a multi-dimensional feature classification algorithm and a single feature classification algorithm, respectively, based on the analysis of the self-esteem degree and the brain structure.
Considering that differences in brain structure lead to differences in brain function, this experiment will be divided into two groups: high and low self-esteem level groups, and the difference in brain structure between the two groups was studied using multidimensional ROI features and machine learning algorithms. Table 2 shows the performance indexes of the single ROI feature and the multi-dimensional ROI feature after classification processing respectively.
TABLE 2 Performance indices of two groups of subjects after respective classification of Single ROI features and Multi-dimensional ROI features
As can be seen from table 2, when only the gray matter volume is used for classification, the classification accuracy is the lowest, which is 65.63%, when the classification is performed by using the relevant features, the obtained classification accuracy is 80.71%, the classification effect is better than that of using a single ROI feature, and when the multi-dimensional features are used, the classification accuracy is the highest, which reaches 87.33%, which indicates that the multi-dimensional features can be better used for distinguishing two groups of subjects.
While embodiments of the invention have been disclosed above, it is not intended to be limited to the uses set forth in the specification and examples. It can be applied to all kinds of fields suitable for the present invention. Additional modifications will readily occur to those skilled in the art. It is therefore intended that the invention not be limited to the exact details and illustrations described and illustrated herein, but fall within the scope of the appended claims and equivalents thereof.
Claims (7)
1. A multi-dimensional feature classification algorithm based on brain nuclear magnetic resonance images is characterized by comprising the following steps:
performing region division on a brain nuclear magnetic resonance image to extract a plurality of ROI features, wherein one ROI feature comprises a plurality of types of marking features for marking one brain nuclear magnetic resonance image;
selecting one of said marker features, establishing a correlation between a number of said ROI features with respect to one of said marker features, forming a correlation feature;
extracting a plurality of ROI features and a plurality of relevant features from the plurality of brain nuclear magnetic resonance images to respectively form an ROI feature set and a relevant feature set;
respectively selecting the features of the ROI feature set and the related feature set through a mixed feature algorithm, and selecting an optimal ROI feature subset and an optimal related feature subset;
and setting a weight factor of the ROI feature relative to the functional proportion of the related feature in the classifier, and integrating the optimal ROI feature subset and the optimal related feature subset through the weight factor and a multi-kernel SVM model to form the multi-kernel classifier.
2. The multi-dimensional feature classification algorithm based on brain nuclear magnetic resonance images according to claim 1, wherein after performing region division on one brain nuclear magnetic resonance image to extract a plurality of ROI features, the method further comprises the following steps:
and respectively carrying out normalization processing on the plurality of kinds of marked features in each ROI feature.
3. The multi-dimensional feature classification algorithm based on brain nuclear magnetic resonance images according to claim 1, characterized in that one of the labeled features is selected, correlation among a plurality of the ROI features related to one of the labeled features is established, and a correlation feature is formed; the method specifically comprises the following steps:
establishing a vector formed by N ROI features with one kind of the marked feature as an N multiplied by N correlation matrix by calculating a correlation coefficient, wherein each element in the correlation matrix represents the correlation between two ROI features with one kind of the marked feature;
defining the irrelevance between the ith ROI feature having one of said marker features and the jth ROI feature having one of said marker features as: d (i, j) ═ t (i) -t (j)]2Wherein t (i) and t (j) respectively represent the feature value of one of the marking features in the i-th ROI feature and the feature value of one of the marking features in the j-th ROI feature;
then, the ith toolThe correlation between the ROI feature having one of said marker features and the jth ROI feature having one of said marker features is defined as:wherein,iandjrespectively representing the standard deviation of the mark feature values between the ith ROI feature and the jth ROI feature; that is, s (i, j) is a correlation feature.
4. The multi-dimensional feature classification algorithm based on brain nuclear magnetic resonance images according to claim 1, wherein the mixed feature algorithm comprises a first filtering feature selection algorithm, a second filtering feature selection algorithm and an encapsulation feature selection algorithm which sequentially perform feature selection on the ROI feature set and the related feature set;
the first filtering feature selection algorithm is used for reducing the number of features;
the second filtering feature selection algorithm is a minimum redundancy maximum correlation feature selection method, and an optimal feature subset is obtained;
the packaging feature selection algorithm is a recursive feature elimination algorithm based on a support vector machine, and further optimized optimal feature subsets are obtained.
5. The multi-dimensional feature classification algorithm based on brain nuclear magnetic resonance images according to claim 1, wherein the optimal ROI feature subset and the optimal related feature subset are integrated through the weighting factors and a multi-kernel SVM model to form a multi-kernel classifier, comprising the following steps:
respectively establishing kernel matrixes for the optimal ROI feature subset and the optimal related feature subset based on a radial basis kernel function;
defining n training samples, defining the weight factor as βm(ii) a Then the feature vector for the ith sample is:wherein M is a markThe type of feature; the label corresponding to each feature vector is yi＝{-1,1}；
Then, the mixed kernel matrix is:wherein,
and when 0 is less than or equal to aiWhen the content is less than or equal to C,phi (-) represents a kernel-function-guided mapping function, k(m)(xi (m)，xj (m)) Representing a training sample xi (m)And xj (m)In the characteristic upper kernel matrix in the m, a represents a Lagrange multiplier, < - > represents inner product operation, and C represents the number of constraint conditions in the model parameters;
thus, the multi-core classifier is
6. The multi-dimensional feature classification algorithm based on brain nuclear magnetic resonance images according to claim 5, characterized in that the weighting factor is βmThe value range of (A) is 0.3-0.6.
7. The multi-dimensional feature classification algorithm based on brain nuclear magnetic resonance images according to claim 5 or 6, further comprising the step of performing two-layer nested cross validation on the multi-kernel classifier, comprising the steps of:
performing first-layer cross validation on the multi-class classifier;
performing second-layer cross validation on the multi-class classifier;
nesting a third level cross-validation outside of said first level cross-validation and said second level 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 true CN106485707A (en) | 2017-03-08 |
| CN106485707B 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) |
Cited By (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 |
| CN107194417A (en) * | 2017-05-09 | 2017-09-22 | 北京积水潭医院 | Injury of finger detection method and system based on MR images |
| 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 |
| CN108764367A (en) * | 2018-06-07 | 2018-11-06 | 深圳大学 | A kind of characteristic image extraction element and extracting method based on relationship 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 |
| 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 |
| 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 |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110268332A1 (en) * | 2010-05-03 | 2011-11-03 | General Electric Company | System and method for nuclear magnetic resonance (nmr) temperature monitoring |
| US20130195342A1 (en) * | 2010-04-27 | 2013-08-01 | Magnetic Resonance Innovations, Inc. | Method of Generating Nuclear Magnetic Resonance Images Using Susceptibility Weighted Imaging and Susceptibility Mapping (SWIM) |
| 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 |
-
2016
- 2016-10-11 CN CN201610886193.4A patent/CN106485707B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130195342A1 (en) * | 2010-04-27 | 2013-08-01 | Magnetic Resonance Innovations, Inc. | Method of Generating Nuclear Magnetic Resonance Images Using Susceptibility Weighted Imaging and Susceptibility Mapping (SWIM) |
| US20110268332A1 (en) * | 2010-05-03 | 2011-11-03 | General Electric Company | System and method for nuclear magnetic resonance (nmr) temperature monitoring |
| 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 |
|---|
| 高宏建等: "基于磁共振图像的脑肿瘤自动识别与分析", 《北京工业大学学报》 * |
Cited By (9)
| 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 |
| CN107194417A (en) * | 2017-05-09 | 2017-09-22 | 北京积水潭医院 | Injury of finger detection method and system based on MR images |
| 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 |
| CN108764367A (en) * | 2018-06-07 | 2018-11-06 | 深圳大学 | A kind of characteristic image extraction element and extracting method based on relationship regularization |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106485707B (en) | 2019-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106485707B (en) | Multidimensional characteristic classification method based on brain magnetic resonance imaging image | |
| CN107944490B (en) | Image classification method based on semi-multimodal fusion feature reduction framework | |
| Al-Tamimi et al. | TUMOR BRAIN DETECTION THROUGH MR IMAGES: A REVIEW OF LITERATURE. | |
| US20060104519A1 (en) | System and method for a contiguous support vector machine | |
| Fratello et al. | Multi-view ensemble classification of brain connectivity images for neurodegeneration type discrimination | |
| CN104376147A (en) | Image-based risk score-a prognostic predictor of survival and outcome from digital histopathology | |
| CN109086809A (en) | Based on the improved early stage Alzheimer disease classification method supervised and be locally linear embedding into | |
| EP3190542B1 (en) | Method and system for generating multiparametric nosological images | |
| Sørensen et al. | Dementia diagnosis using MRI cortical thickness, shape, texture, and volumetry | |
| Swaraja et al. | Segmentation and detection of brain tumor through optimal selection of integrated features using transfer learning | |
| Bansal et al. | An improved hybrid classification of brain tumor MRI images based on conglomeration feature extraction techniques | |
| Akselrod-Ballin et al. | An integrated segmentation and classification approach applied to multiple sclerosis analysis | |
| Ramanathan et al. | Alzheimer’s disease shape detection model in brain magnetic resonance images via whale optimization with kernel support vector machine | |
| Ahmed et al. | Early Alzheimer disease detection with bag-of-visual-words and hybrid fusion on structural MRI | |
| Tran et al. | High-dimensional MRI data analysis using a large-scale manifold learning approach | |
| Aderghal | Classification of multimodal MRI images using Deep Learning: Application to the diagnosis of Alzheimer’s disease. | |
| KR102373992B1 (en) | Method and apparatut for alzheimer's disease classification using texture features | |
| Vinutha et al. | Efficient morphometric techniques in Alzheimer’s disease detection: survey and tools | |
| Aggarwal et al. | Role of surfacelet transform in diagnosing Alzheimer’s disease | |
| CN113080929A (en) | anti-NMDAR encephalitis image feature classification method based on machine learning | |
| Khode et al. | Efficient brain tumor detection using wavelet transform | |
| Shishegar et al. | Hippocampal shape analysis in epilepsy using Laplace-Beltrami spectrum | |
| Sinthia et al. | An effective two way classification of breast cancer images: A detailed review | |
| Al-Majeed et al. | MR images classification of Alzheimer’s disease based on deep belief network method | |
| Chen et al. | A supervised hybrid classifier for brain tissues and white matter lesions on multispectral MRI |
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 |