CN102509123B - Brain function magnetic resonance image classification method based on complex network - Google Patents

Abstract

The invention relates to a brain functional magnetic resonance image classification method based on a complex network, which comprises the following steps: pre-processing training sample images and test sample images, carrying out region segmentation, and extracting an average time sequence from each region; calculating the partial correlation coefficient among the average time sequences, carrying out matrix binarization on the partial correlation coefficient to obtain a complex network model, and calculating the feature path length, cost and clustering degree of the complex network model to respectively obtain network features of the training sample images and the test sample images; training to obtain an adaboost classifier; and by using the adaboost classifier obtained by training, classifying the test sample images. By using information in the brain functional magnetic resonance images as much as possible, the method can accurately classify the brain functional magnetic resonance images.

Description

A kind of brain function MRI sorting technique based on complex network

Technical field

The invention belongs to technical field of image processing, be specifically related to a kind of brain function MRI sorting technique based on complex network.

Background technology

With its high-spatial and temporal resolution, the characteristics such as non-intrusion type are being widely applied aspect the sacred disease diagnoses and treatment functional mri (functional Magnetic Resonance Imaging, fMRI).FMRI refers generally to rely on (blood oxygen level-dependent based on blood oxygen level, BOLD) magnetic resonance imaging, it changes the magnetic resonance signal that causes and changes to react cerebration by measuring the compositions such as the brain blood flow that caused by nervous activity and brain blood oxygen.Brain is the system of a complexity, and corresponding variation can occur the magnetic resonance image (MRI) of brain when being upset conditioned disjunction experience pathology.Utilize image classification method, calculate the possibility size that brain function MRI has certain attribute, perhaps the category attribute of automatic discrimination image is an important application of computer-aided analysis.

Traditional functional MRI sorting technique mainly contains area-of-interest (ROI) mode and two kinds of sorting techniques of voxel (voxel) mode.The sorting technique of area-of-interest mode becomes a plurality of target areas with sample, and accordingly target is classified according to the priori of object construction with Target Segmentation; The sorting technique of voxel mode adopts complicated non-linear registration, to realize to greatest extent the accurate correspondence between individuality, then with each mikey (voxel) of image as classification foundation.These the two kinds of methods all internal organizational structure of hypothetical target and sample are one to one.The former thinks that the image-region of priori is present in the middle of each target image, and can accurately cut apart; The latter supposes that the voxel behind the non-linear registration is one to one.Yet such hypothesis is also unreasonable under many circumstances.The brain function MRI of people under different conditions can be subject to the interference of many factors, and traditional sorting technique is not according to the build-in attribute of brain brain function MRI to be classified, and therefore all can cause the decline of classification performance.

Summary of the invention

The technical matters that (one) will solve

In order to overcome the deficiency of prior art, technical matters to be solved by this invention is the brain function MRI sorting technique that a kind of classification accuracy of design is high, Generalization Capability is strong.

(2) technical scheme

For achieving the above object, the present invention proposes a kind of brain function MRI sorting technique based on complex network, may further comprise the steps:

Step Sa: training sample image and test sample image are carried out pre-service, then carry out the brain differentiation and cut, and extract sequence averaging time in each brain district;

Step Sb: calculate the partial correlation coefficient between each of sequence averaging time, obtain the partial correlation coefficient matrix;

Step Sc: with described partial correlation coefficient matrix binaryzation, obtain complex network model;

Step Sd: calculate Path length, cost and the cluster degree of this complex network model as the feature of functional MRI;

Step Se: utilize the network parameter of training sample image as the feature of the training sample image in the feature of this functional MRI, train a self-adaptation to improve (adaboost) sorter;

Step Sf: utilize this self-adaptation that trains to improve (adaboost) sorter test sample image is classified.

(3) beneficial effect

The present invention is directed to the brain function MRI classification problem, by making up brain network model, computational grid characteristic parameter, the training self-adaptation Stability and veracity of Images Classification that improved the method Effective Raises such as (adaboost) sorter.

The present invention can utilize information as much as possible in the brain function MRI, the brain network parameter can be from reacting the activity of brain in essence, remedied the deficiency that the traditional classification method can not embody the cerebration build-in attribute, can classify to brain function MRI accurately.

Description of drawings

Fig. 1 is method flow diagram of classifying based on the brain function MRI of complex network provided by the invention;

Fig. 2 uses sorting technique of the present invention (method A) contrast existing sorting technique based on local feature (method B), the correlation curve of classification experimenter's operating characteristic (ROC) according to the embodiment of the invention.

Embodiment

For making the purpose, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, the present invention is described in more detail.

Brain function MRI classification based on complex network is a kind of brand-new brain function MRI sorting technique.The method model complex brain network model, Path length and the cluster degree of calculating brain network are in order to characterize different image models; Then utilize this Path length and cluster degree to train a self-adaptation to improve (adaboost) sorter; Utilizing at last this self-adaptation that trains to improve (adaboost) sorter classifies to test sample image.

With reference to Fig. 1, according to a kind of human brain function magnetic resonance imaging image classification method of the present invention, can determine according to training sample image the classification of test sample image, the implementation step is as follows:

Step Sa carries out pre-service to training sample image and test sample image, then carries out the brain differentiation and cuts, and extract sequence averaging time in each brain district;

1. the pre-service of brain function MRI

Because the impact of various noises in the magnetic resonance imaging process, tested individual self exists yardstick and locational difference, is necessary very much before analyzing data data to be done certain pre-service.In the data acquisition of whole experiment, main noise information source has: (1) physics head is moving; (2) interlayer difference sweep time in the image; (3) unevenness of exterior magnetic field etc.The pretreated common step of brain function MRI has: section alignment sweep time, image sequence alignment, associating registration, standardization (or claiming homogenization), space smoothing filtering and time smoothing filtering etc.

2. brain function MRI cuts apart

Adopt international structure tag template (AAL), full brain is divided into 90 brain districts.The structure tag template is the most widely used brain stay in place form of brain function MRI research field.

3. extract sequence averaging time in each brain district

Data according to pretreated brain function MRI, extraction is contained in time series Y (the matrix dimension D * N) of each voxel activation value on different time points of inside, corresponding brain district, wherein D is the voxel number that is contained in spheroid inside, and N counts the time.Described activation value refers to that the blood oxygen level of each voxel on different time points relies on (BOLD) intensity.

Step Sb: calculate the partial correlation coefficient between average each time series.This step Sb specifically comprises the steps:

1. calculate the covariance coefficient between sequence averaging time

The time series in each brain district that extracts according to step Sa is calculated the covariance matrix S between each of sequence, each element s of S averaging time _{I, j}Be the covariance coefficient between i and j the time series,

$s_{i,j}=\frac{1}{M}\underset{t=1}{\overset{M}{\mathrm{\Σ}}}(x_i\left(t\right)-\stackrel{\‾}{x_i})(x_j\left(t\right)-\stackrel{\‾}{x_j})$

Wherein, M is the time point number, x _i(t) (i=1 ..., M) be i time series,

Be i seasonal effect in time series mean value, Be j seasonal effect in time series mean value.

2. calculate the partial correlation coefficient between sequence averaging time

According to the covariance coefficient matrix S between time series (matrix dimensionality is 90 * 90), the partial correlation coefficient matrix R (matrix dimensionality is 90 * 90) between the computing time sequence, each element r of R _{I, j}For:

$r_{i,j}=-\frac{s_{i,j}^{-1}}{\sqrt{s_{i,i}^{-1}{s}_{j,j}^{-1}}}$

Wherein,

{ i, j} element for the inverse matrix of covariance matrix S (matrix dimensionality is 90 * 90).

3. partial correlation coefficient is carried out the Fisher conversion

According to partial correlation coefficient matrix R (matrix dimensionality is 90 * 90), calculate the partial correlation coefficient matrix F (matrix dimensionality is 90 * 90) through the Fisher conversion, each element f of F _IjFor:

$f_{i,j}=\frac{1}{2}\left(\frac{1+r_{i,j}}{1-r_{i,j}}\right),$

Wherein, f _IjBe { i, j} element, the r through the partial correlation coefficient matrix F (matrix dimensionality is 90 * 90) after the Fisher conversion _Ij{ i, j} element for partial correlation coefficient matrix R (matrix dimensionality is 90 * 90).

Step Sc: with partial correlation coefficient matrix binaryzation, obtain complex network model;

Setting threshold T ', order is 1 through the value more than or equal to T ' in the partial correlation coefficient matrix F (matrix dimensionality is 90 * 90) after the Fisher conversion, is 0 less than the value of T ', obtains complex network model.1 expression has connection between two brain districts in the matrix after the binaryzation, and namely the limit between two nodes exists in the network, and 0 expression does not connect between two brain districts, does not namely have the limit between two nodes in the network.The method that threshold value is chosen is: making the quantity in esse limit in the network is the quantity on the limit that may exist in the network

Wherein N is the number of nodes) 1/10th.The process of binaryzation can be described as order

$w_{i,j}=\left\{\begin{array}{c}1,\left|f_{i,j}\right|\&GreaterEqual;T^{\&prime;}\\ 0,\left|f_{i,j}\right|<T^{\&prime;}\end{array}\right.,$

Wherein, w _Ij{ i, j} element, f for the network after the binaryzation _IjFor through the partial correlation coefficient matrix F (matrix dimensionality is 90 * 90) of Fisher conversion the i, j} element, T ' are the threshold value of choosing, || be the absolute value compute sign.

Step Sd: calculate Path length, cost and the cluster degree of this complex network model as the feature of functional MRI;

According to complex network model, calculate Path length, cost and the cluster degree of this complex network model, as the feature of functional MRI.

Path length provides the information of a certain node in the network to arrive the optimal path of another node.We can use any two node i in the Path length matrix description network, the Path length l of j _IjNetwork average characteristics path L has described the mean value of the Path length of any two nodes in the network, namely

$L=\frac{1}{N(N-1)}\underset{i,j\&Element;V,i\&NotEqual;j}{\mathrm{\&Sigma;}}{l}_{\mathrm{ij}}$

Wherein, N is the number of nodes, the brain district several 90 of namely cutting apart; l _IjBe node i, the Path length between the j, V is the set of all nodes in the network.

Cost is an important parameter of tolerance network character, is used for weighing the required overall cost of paying of structure network.Computing method are with the quantity in esse all limits in the network quantity than the limit that may exist at most in the upper network, that is:

$K=\frac{\frac{\mathrm{\&Sigma;}{K}_i}{2}}{\frac{N(N-1)}{2}}=\frac{1}{N(N-1)}\mathrm{\&Sigma;}{K}_i,$

Wherein, N is the number of nodes, K _iBe the quantity on the limit that is connected to node i in the network, K is the cost of network.

The cluster degree is another key character of tolerance network character, and the adjacent node that is used for measuring a certain node is neighbours' possibility each other.The cluster degree C of a certain node i _iValue equal the number on the limit that exists between its adjacent node and the ratio of all possible limit number between them, namely

$C_i=\frac{e_i}{\frac{k_i(k_i-1)}{2}}=\frac{2e_i}{k_i(k_i-1)}$

Wherein, e _iThe limit number that exists between the adjoint point of expression node i, k _iThe number of the adjoint point of expression node i,

Just represent the limit number that may exist between the adjoint point of node i.

Step Se: utilize the network parameter of training sample image as the feature of the training sample image in the feature of this functional MRI, train a self-adaptation to improve (adaboost) sorter.

After obtaining the feature of training sample image, at first with Path length, cost and cluster degree as three linear classifiers, weighted sum with these three linear classifiers forms new self-adaptation raising (adaboost) sorter, and the weight of each sorter is made as at first

(m is the number of sample image), self-adaptation improves (adaboost) sorter is adjusted three linear classifiers gradually in training process weight, and the self-adaptation that obtains at last an optimum improves (adaboost) sorter.The implementation step is as follows:

To given sample (x ₁, y ₁) ..., (x _m, y _m), x wherein _i∈ X, y _i∈ Y=(1,1), X are the network characterization of training sample image, and Y is image category, and the weight of at first setting the initialization sorter is

Carry out T time afterwards iteration, iterative process is as follows:

Variable t is increased to T since 1, and each iteration is at first calculated each feature h _tThe error in classification ε that training sample image is classified and obtained _t, then calculate new sample weights,

${\mathrm{\&alpha;}}_t=\frac{1}{2}\ln \left(\frac{1-{\mathrm{\&epsiv;}}_t}{{\mathrm{\&epsiv;}}_t}\right),$

At last, upgrade the weight of each linear classifier,

$D_{t+1}\left(i\right)=\frac{D_t\left(i\right)}{Z_t}\left\{\begin{array}{c}{e}^{-{\mathrm{\&alpha;}}_t},h_t\left(x_i\right)=y_i\\ e^{{\mathrm{\&alpha;}}_t},h_t\left(x_i\right)\&NotEqual;y_i\end{array}\right.,$

Z wherein _tBe normalized factor.

Obtain optimum self-adaptation after circulation finishes and improve (adaboost) sorter:

$H\left(x\right)=\mathrm{sign}\left(\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{h}_t\left(x\right)\right).$

Step Sf: the optimum self-adaptation of utilizing training to obtain improves (adaboost) sorter test sample image is classified.

Test sample book is inputted optimum self-adaptation raising (adaboost) sorter that above-mentioned steps obtains, test sample image is classified, classification results is by classification accuracy rate, True Positive Rate and false positive rate output.

The effect of the brain function MRI sorting technique based on complex network of the present invention can be illustrated by real brain function magnetic resonance brain imaging data:

(1) True Data experimentation

For showing effect of the present invention, adopt in embodiments the True Data collection to test, totally 39 have testedly participated in experiment, 20 men, 19 woman.Tested age bracket and clinical dementia rating information see Table lattice 1.BOLD fMRI tranquillization data behind experiment employing T2* weighted gradient echo-planar imaging (Echo-Planar Imaging, EPI) the retrieval acupuncture stimulation.

Employing Statistical Parametric Mapping (SPM) software ( Http:// www.fil.ion.ucl.ac.uk/spm/) data are carried out pre-service, comprise section alignment sweep time, image sequence alignment, associating registration, standardization (or claiming homogenization), space smoothing filtering.Use the method for the invention (method A) contrast existing sorting technique based on local feature (method B), obtain experimenter's operating characteristic (ROC) curve and the area under curve (AUC) thereof of sorting technique, and with ROC curve and the AUC tolerance as classifier performance.

Form 1 tested information

(2) experimental result

The classification ROC curve of two kinds of methods shows in Fig. 2 respectively on true experiment data set, wherein, True Positive Rate among Fig. 2 refers to actual positive and correctly be judged to positive number percent by the standard of this Screen test, and false positive rate refers to actual negative and is judged to mistakenly positive number percent by the standard of this Screen test.As shown in Figure 2, the ROC curve of method A is higher than method B in most of threshold range; AUC value contrast situation: the AUC value of method A is 0.85, and the AUC value of method B is 0.78.Area under curve (AUC) can be measured overall classification performance, posterior probability and ordering performance, and the AUC value is larger, and then the overall performance of this sorting technique is better.Thus, method A effect is better than method B.

Experimental result explanation, the brain function MRI sorting technique based on complex network of the present invention has improved the classification performance of brain function MRI effectively.

The above; only be the embodiment among the present invention, but protection scope of the present invention is not limited to this, anyly is familiar with the people of this technology in the disclosed technical scope of the present invention; can understand conversion or the replacement expected, all should be encompassed within the protection domain of claims of the present invention.

Claims (2)

1. the brain function MRI sorting technique based on complex network is characterized in that, may further comprise the steps:

Step Sa: training sample image and test sample image are carried out pre-service, then carry out the brain differentiation and cut, and extract sequence averaging time in each brain district, the step of wherein extracting brain district sequence averaging time is:

At first according to standard brain stay in place form full brain is divided into 90 brain districts, extracts respectively the activation value of inner each voxel in each brain district on different time points, the activation value with each voxel averages again, obtains brain district sequence averaging time;

Step Sb: calculate the partial correlation coefficient between each of sequence averaging time, obtain the partial correlation coefficient matrix, this step is:

At first calculate the covariance matrix S between each of sequence, each element s that this covariance matrix dimension is 90 * 90, S averaging time _{I, j}Be the covariance coefficient between i and j the time series,

$s_{i,j}=\frac{1}{M}\underset{t=1}{\overset{M}{\mathrm{\&Sigma;}}}(x_i\left(t\right)-\stackrel{\&OverBar;}{x_i})(x_j\left(t\right)-\stackrel{\&OverBar;}{x_j}),$

Wherein, M is the time point number, x _i(t) (i=1 ..., M) be i time series,

Be i seasonal effect in time series mean value,

Be j seasonal effect in time series mean value;

Then, calculate the partial correlation coefficient matrix R between sequence averaging time, the dimension of this partial correlation coefficient matrix R is each element r of 90 * 90, R _{I, j}For:

$r_{i,j}=-\frac{s_{i,j}^{-1}}{\sqrt{s_{i,i}^{-1}{s}_{j,j}^{-1}}};$

Wherein,

{ i, j} element for the inverse matrix of covariance matrix S;

At last, partial correlation coefficient is carried out the Fisher conversion, obtaining through the partial correlation coefficient matrix dimensionality after this conversion of partial correlation coefficient matrix F after the Fisher conversion is 90 * 90;

Step Sc: with described partial correlation coefficient matrix binaryzation, obtain complex network model, this step is:

Selected threshold will be through the partial correlation coefficient matrix F binaryzation of Fisher conversion, partial correlation coefficient matrix dimensionality after this conversion is 90 * 90, between two brain districts of 1 expression connection is arranged after the binaryzation, be that the limit between two nodes exists in the network, 0 expression does not connect between two brain districts, does not namely have the limit between two nodes in the network;

The method that threshold value is chosen is: make the quantity of selecting this threshold value to carry out in esse limit in the network after the binaryzation be the limit that may exist at most in the network quantity 1/10th

Step Sd: calculate Path length, cost and the cluster degree of this complex network model as the feature of functional MRI, wherein,

The step of calculating the Path length of this complex network model is:

With any two node i in the Path length matrix description network, the Path length l of j _Ij, network average characteristics path L has described the mean value of the Path length of any two nodes in the network, namely

$L=\frac{1}{N(N-1)}\underset{i,j\&Element;V,i\&NotEqual;j}{\mathrm{\&Sigma;}}{l}_{\mathrm{ij}}$

Wherein, N is the number of nodes, the brain district several 90 of namely cutting apart; l _IjBe node i, the Path length between the j, V is the set of all nodes in the network;

The step of calculating the cost of this complex network model is:

With the quantity in esse all limits in the network quantity than the limit that may exist at most in the upper network, that is:

$K=\frac{\frac{\mathrm{\&Sigma;}{K}_i}{2}}{\frac{N(N-1)}{2}}=\frac{1}{N(N-1)}\mathrm{\&Sigma;}{K}_i,$

Wherein, N is the number of nodes, K _iQuantity for the limit that is connected to node i in the network;

The step of calculating the cluster degree of this complex network model is:

The cluster degree C of a certain node i _iValue equal the number on the limit that exists between its adjacent node and the ratio of all possible limit number between them, namely

$C_i=\frac{e_i}{\frac{k_i(k_i-1)}{2}}=\frac{{2e}_i}{k_i(k_i-1)}$

Wherein, e _iThe limit number that exists between the adjoint point of expression node i, k _iThe number of the adjoint point of expression node i,

Just represent the limit number that may exist between the adjoint point of node i;

Step Se: utilize Path length, cost and the cluster degree of training sample image as the feature of the training sample image in the feature of this functional MRI, train a self-adaptation to improve (adaboost) sorter, comprise the steps:

At first with Path length, cost and cluster degree as three linear classifiers, form a new self-adaptation with the weighted sum of these three linear classifiers and improve (adaboost) sorter, the weight of each sorter is made as at first

M is the number of sample image, and self-adaptation improves (adaboost) sorter is adjusted three linear classifiers gradually in training process weight, and the self-adaptation that obtains at last an optimum improves (adaboost) sorter, and the implementation step is as follows:

To given sample (x ₁, y ₁) ..., (x _m, y _m), x wherein _i∈ X, y _i∈ Y=(1,1), x are the network characterization of training sample image, and Y is image category, and the weight of at first setting the initialization sorter is

Carry out T time afterwards iteration, iterative process is as follows:

Variable t is increased to T since 1, and each iteration is at first calculated each feature h _tThe error in classification ε that training sample image is classified and obtained _t, then calculate new sample weights,

${\mathrm{\&alpha;}}_t=\frac{1}{2}\ln \left(\frac{1-{\mathrm{\&epsiv;}}_t}{{\mathrm{\&epsiv;}}_t}\right),$

At last, upgrade the weight of each linear classifier,

$D_{t+1}\left(i\right)=\frac{D_t\left(i\right)}{Z_t}\left\{\begin{array}{c}{e}^{{-\mathrm{\&alpha;}}_t},h_t\left(x_i\right)=y_i\\ e^{{\mathrm{\&alpha;}}_t},h_t\left(x_i\right)\&NotEqual;y_i\end{array}\right.$

Z wherein _tBe normalized factor;

Obtain optimum self-adaptation after circulation finishes and improve (adaboost) sorter:

$H\left(x\right)=\mathrm{sign}\left(\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{h}_t\left(x\right)\right);$

Step Sf: utilize this self-adaptation that trains to improve (adaboost) sorter test sample image is classified.

2. the brain function MRI sorting technique based on complex network as claimed in claim 1, it is characterized in that, described training sample image and test sample image are carried out pre-service, when keeping the brain function image detail, use Brain mapping picture and standard form to carry out the pre-service of affine registration mapping mode, and improve the signal to noise ratio (S/N ratio) of Brain mapping picture.

Images