CN104794708A - Atherosclerosis plaque composition dividing method based on multi-feature learning - Google Patents

Abstract

The invention discloses an atherosclerosis plaque composition dividing method based on multi-feature learning. The atherosclerosis plaque composition dividing method includes the following steps: 1), acquiring multi-sequence arterial blood vessel plaque images; 2), preprocessing the images; 3), extracting and describing plaque features; 4), training through classifiers, wherein the model of each classifier is a supportive vector machine (SVM); generating four feature sets by four sequences of MRI arterial plaque images, training each feature set to generate a classifier, and finally acquiring four classifiers including a lipid classifier, a fibrous classifier, a calcium classifier and a hemorrhage classifier; 5), optimizing a trainer; 6), using the four classifiers to calculate subordinate degrees of a certain pixel of plaque tissue to different tissue, competitively ordering acquired subordinate degrees, and taking a composition highest in subordinate degree as a final result. The atherosclerosis plaque composition dividing method is high in accuracy and robustness.

Description

A kind of atherosclerotic plaque composition dividing method based on multiple features study

Technical field

The invention belongs to field of medical image processing, refer more particularly to the application in medical image analysis in area of computer aided and artificial intelligence technology.

Background technology

Current, the framework still ununified for atherosclerotic plaque composition dividing method and technological means, its main cause is that the imaging technique of blood vessels present patch is ripe not enough, and this makes the image quality of vascular plaque (contrast of the image of such as patch, the sharpness etc. at edge) not high.Blood vessels present image generally has gray level image or pseudo color image in addition, and this makes computing machine be difficult to characterize Patches information by quantitative digital quantity.In order to address these problems, Chinese scholars has done a large amount of research work, wherein most is representational, is the patch dividing method based on bayesian theory that the people such as F.Liu propose, the method mainly utilizes the gray-scale value of patch, patch position in the blood vessel and shape information are as feature, then a Gauss model is gone out based on these features training, utilize Bayes posterior probability formulae discovery to go out patch and be under the jurisdiction of the probability of certain composition to complete the segmentation of composition, but the noise of the method to image is very sensitive, train a model can portraying Patch properties in addition and need a large amount of training samples.Some researchers are also had to utilize the method segmentation plaque component of (Fuzzy-clustering) of fuzzy clustering.The probability that the method utilizes the gray-scale value of artery plaque then to utilize the mode of fuzzy clustering to calculate patch to be under the jurisdiction of heterogeneity as feature, this method shortcoming clearly, first be that the Patch properties that the method utilizes is single, the information of patch can not be given full expression to, secondly the method robustness is lower, and the result of fuzzy clustering is easily subject to the impact of plaque component.Studies in China person money Huaming etc. proposes the algorithm of region growing in conjunction with time-domain information, this algorithm utilizes gray difference between image as growth conditions, and in conjunction with half-tone information as growth conditions, then utilize morphologic relevant affected method revise segmentation result and then extract the partitioning boundary of heterogeneity, but the method exists Patch properties equally express not enough and that Algorithm robustness is not high defect.

Summary of the invention

In order to overcome the deficiency that accuracy is poor, robustness is not high of existing atherosclerotic plaque composition dividing method, the invention provides the atherosclerotic plaque composition dividing method based on multiple features study that a kind of accuracy is good, robustness is higher.

The technical solution adopted for the present invention to solve the technical problems is:

Based on an atherosclerotic plaque composition dividing method for multiple features study, described dividing method comprises the following steps:

1) multisequencing arteries patch image is obtained;

2) Image semantic classification

Choose ROI centered by arteries inwall as survey region, to window width, the adjustment of window position that the ROI chosen does gray-scale value, then denoising is done to ROI;

3) Patch properties extracts and describes

The feature of patch is described respectively from the gray-scale value of patch image, Gaussian Blur feature, Grad, patch to the minimum distance of Ink vessel transfusing outer wall, Laplce's feature of image and the textural characteristics of patch;

4) sorter training: sorter model is support vector machines, and training process is as follows:

(4.1) collection of training sample and demarcation

In training sample, T1 sequence image is as reference image, and it is operate on T1 image that doctor manually demarcates atherosclerotic plaque composition, and the plaque component of other four kinds of sequence T1GD, T2, PD and STIR is that the result of demarcating according to T1 maps thereon;

(4.2) sample preprocessing

First gray-scale value stretched operation is done to sample image, make its grey value profile scope [0,255] interval range, then be to image denoising sonication, finally that doctor done binary conversion treatment to the calibration result of plaque component, in the two values matrix that each composition of the patch namely doctor demarcated is stored in;

(4.3) training sample is classified and characteristic quantification process further

The further classification of training sample is that sample is subdivided into lipid sample set, fibrous sample set, calcium sample set and hemorrhage sample set;

Characteristic quantification process is that the Patch properties extracted is done normalized, and all features form a proper vector after normalization, this proper vector shown in (10).

$\stackrel{\→}{F}=[\stackrel{\→}{V},\stackrel{\→}{G},\stackrel{\‾}{G}\stackrel{\→}{M},\stackrel{\→}{L},\stackrel{\→}{T}]---\left(10\right)$

In formula, V represents the gray-scale value of patch, and G represents the Gaussian Blur value of patch, and GM represents the gray feature after gray-scale value Gaussian Blur, and L represents Laplce's feature, and D represents patch locus feature, and T represents the textural characteristics of patch.

(4.4) sorter training, the symbiosis of last four kinds of sequence MRI artery plaque images becomes 4 feature sets, each feature set training generation sorter, finally obtain 4 sorters, i.e. lipid sorter, fibrous sorter, calcium sorter, hemorrhage sorter;

5) training aids optimization

Sample set is divided into training set and test set two parts, then the sample training sorter of training set is used, with the discrimination of the sorter verification test collection obtained, judge that circulation is gone down successively, until the training aids classification accuracy rate obtained reaches maximal value, optimum classifier thought by the sorter now obtained;

6) for some pixels of plaque, by four classifier calculated, it is under the jurisdiction of the degree of different tissues respectively, and process is as follows:

(6.1) feature extraction first T1GD, T2, PD and STIR plaque image pointed out in each pixel out;

(6.2) Patch properties extracted in (6.1) is put in four sorters after optimizing in step (5) respectively, four kinds of sorters calculate its degree being under the jurisdiction of its composition respectively;

(6.3) according to the degree of membership competition sequence in (6.2) that obtain, using composition the highest for degree of membership as net result.

Further, described dividing method is further comprising the steps of:

7) morphological operation of result, process is as follows:

(7.1) result of composition segmentation is stored in 4 bianry images, in bianry image, 1 represents structural constituent, and 0 represents background, is lipid composition, fibrous composition, calcium composition respectively, hemorrhage composition;

(7.2) then erosion operation is done to each bianry image, isolated block filtering;

(7.3) then do morphology opening operation, fill up the cavity in component block and be communicated with close isolated block;

(7.4) finally do closing operation, obtain atherosclerotic plaque composition.

Further, described step 3) in,

A () gray-scale value is at the gray-scale value of image after denoising;

(b) Gaussian Blur feature

Described Gaussian Blur feature refers to the gray-scale value of plaque a bit after Gaussian Blur, and leaching process is that the gaussian kernel of an employing 5*5 and plaque do convolution, shown in (2).

$G(x,\mathrm{\σ})=\frac{1}{\sqrt{2\mathrm{\π\σ}}}{e}^{-\frac{{\left|\right|x\left|\right|}^2}{2{\mathrm{\σ}}^2}}*I---\left(2\right)$

In formula, σ represents the yardstick of gaussian kernel, and x represents the gray-scale value of image, and I, G distinguish the image after representative image primitive vessel patch image and Gaussian Blur, and " * " is convolution symbol;

(c) Gradient Features

The Gradient Features GM of plaque:

$\mathrm{GM}=\sqrt{{\left(\frac{\∂\left(G(x,\mathrm{\σ})\right)}{\∂x}\right)}^2+{\left(\frac{\∂\left(G(x,\mathrm{\σ})\right)}{\∂y}\right)}^2}---\left(3\right)$

(d) Laplce feature L:

$L=\frac{{\∂}^2\left(G(x,\mathrm{\σ})\right)}{\∂x^2}+\frac{{\∂}^2\left(I(x,\mathrm{\σ})\right)}{\∂x^2}---\left(4\right)$

(e) patch locus feature (distance between inside and outside wall)

D _min(I)＝min||C(I(x))-Lumen(I)|| (5)

D _max(I)＝max||C(I(x))-Lumen(I)|| (6)

(f) textural characteristics

Be extracted first order statistic and gray level co-occurrence matrixes feature two class textural characteristics, totally seven first order statistic, comprising patch interior intensity average, standard deviation, coefficient of variation, the degree of bias, kurtosis, entropy of histogram and brightness entropy; EtHis such as formula (7) and brightness entropy EtBrt such as formula (8):

$\mathrm{EtHis}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{p}_i{\log }_2\left(p_i\right)---\left(7\right)$

$\mathrm{EtBrt}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}\frac{{\mathrm{\ω}}_i{\log }_2{\mathrm{\ω}}_i}{{\log }_{2}N}---\left(8\right)$

Wherein, P _i(i=0,1 ..., 255) be the probability that GTG i occurs in the picture, ω _ibe after normalization be not 0 image pixel intensities, N is the sum of all pixels of original image.

Beneficial effect of the present invention is mainly manifested in: utilize multisequencing magnetic resonance image (MRI) (multiple sequence MRI) as image source, then the feature of atherosclerotic plaque is extracted from many aspects, utilize the method training classifier of machine learning, finally complete the segmentation of plaque component; Accuracy is good, robustness is higher.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the atherosclerotic plaque composition dividing method based on multiple features study.

Fig. 2 is that ROI chooses and denoising result figure, and wherein, (a) chooses schematic diagram for ROI, and (b) is denoising result figure.

Fig. 3 is the registration result figure of multisequencing magnetic resonance image (MRI), wherein, a () is the spatial diversity figure of T1 before registration (left side figure) and PD (the right figure), b () is registration result, T1 (left side figure) and PD (the right figure) schematic diagram.

Fig. 4 is the training process flow diagram of sorter.

Fig. 5 is classifier optimization procedure chart.

Fig. 6 is the process flow diagram of plaque component identification.

Embodiment

Below in conjunction with accompanying drawing, the invention will be further described.

With reference to Fig. 1 ~ Fig. 6, a kind of atherosclerotic plaque composition dividing method based on multiple features study, comprises the steps:

1) multisequencing arteries patch image is obtained

Arteria carotis (artery vessel) magnetic resonance image (MRI) of the T1 (T1 weighted imaging) gathered under 1.5T magnetic field intensity, T1GD (T1 contrast agent imaging), T2 (T2 weighted imaging), PD (proton density imaging), STIR (short upset flip-flop transition reverts to picture) sequence is example;

2) Image semantic classification

Owing to collecting the cross-sectional view picture that image is whole artery position, wherein the position at arteries place only accounts for about 2% of whole image, tissue around arteries not only can be caused the impact of internal blood vessel tissue if do pre-service to entire image, also can increase the efficiency that operand reduces algorithm process simultaneously.Then choose the ROI of 60*60 pixel centered by arteries inwall (Region of Interest) as survey region in the Image semantic classification first step for this reason.The ROI chosen is done to window width, the adjustment of window position of gray-scale value, makes its tonal range be distributed in (0--255) such as formula (1),

$I=\frac{I\left(x\right)-I_{\min }}{I_{\max }-I_{\min }}\×255---\left(1\right)$

Then do denoising to ROI, the denoising that the present invention adopts is at last based on the denoising method that the adaptive threshold of wavelet decomposition regulates.As shown in Figure 2, Fig. 2 (a) represents T1 sequence image to its result, rectangle frame be wherein extract ROI region, Fig. 2 (b) be to ROI region denoising after result.

Because the breathing of patient, pulsation, swallowing act and trickle to tremble etc. all can cause multi-sequence image rigidity and nonrigid deformation in multi-sequence image imaging process, the interference that these uncontrollable " motions " bring is fatal for the Carotid Plaque Compositions identification of pixel scale.The such as movement of patient 1mm in realistic space can cause the skew of ten to tens pixels on image.Therefore be necessary to the registration work between sequence image.The open source software that present invention employs elastix V4.7 image registration completes registration work, its registration result as shown in Figure 3, Fig. 3 (a) represents the difference before T1 (left side) and PD (right side) image sequence alignment, wherein two irregular circle representatives is the Ink vessel transfusing outer wall of reference picture, Fig. 3 (b) represents the result after registration, obviously can see that T1 and PD sequence Ink vessel transfusing outer wall achieves and be identical;

3) Patch properties extracts and describes

In order to the specificity of the atherosclerotic plaque of description that can be accurate and effective, this method describes the feature of patch respectively respectively from the gray-scale value of patch image, Gaussian Blur feature, Grad, patch to the minimum distance of Ink vessel transfusing outer wall, Laplce's feature of image and the textural characteristics aspect of patch.The extracting method of these features is below described.

(a) gray-scale value

Gray-scale value namely at the gray-scale value of image after denoising, this feature be multi-sequence image to the specific direct embodiment of patch, be most valuable feature to a certain extent, also because it obtains simple, effect is directly perceived and be widely used.

(b) Gaussian Blur feature

Gaussian Blur feature in this method refers to the gray-scale value of plaque a bit after Gaussian Blur, eliminate the interference of imaging noise to a certain extent, its leaching process is that the gaussian kernel of an employing 5*5 and plaque do convolution, shown in (2).

$G(x,\mathrm{\σ})=\frac{1}{\sqrt{2\mathrm{\π\σ}}}{e}^{-\frac{{\left|\right|x\left|\right|}^2}{2{\mathrm{\σ}}^2}}*I---\left(2\right)$

Wherein, σ represents the yardstick of gaussian kernel, and x represents the gray-scale value of image, and I, G distinguish the image after representative image primitive vessel patch image and Gaussian Blur, and " * " is convolution symbol;

(c) Gradient Features

The Gradient Features GM of plaque:

$\mathrm{GM}=\sqrt{{\left(\frac{\∂\left(G(x,\mathrm{\σ})\right)}{\∂x}\right)}^2+{\left(\frac{\∂\left(G(x,\mathrm{\σ})\right)}{\∂y}\right)}^2}---\left(3\right)$

(d) Laplce feature L:

$L=\frac{{\∂}^2\left(G(x,\mathrm{\σ})\right)}{\∂x^2}+\frac{{\∂}^2\left(I(x,\mathrm{\σ})\right)}{\∂x^2}---\left(4\right)$

(e) patch locus feature (distance between inside and outside wall)

D _min(I)＝min||C(I(x))-Lumen(I)|| (5)

D _max(I)＝max||C(I(x))-Lumen(I)|| (6)

(f) textural characteristics

The textural characteristics of atherosclerotic plaque extracted and analyzes, contributing to objectively classifying to plaque type, thus making the diagnosis of being correlated with more objective and accurate.In order to reflect the statistical law that patch interior intensity distributes, this method is extracted first order statistic and gray level co-occurrence matrixes (gray level co-occurrence matrix, GLCM) feature two class textural characteristics.Totally seven first order statistic, comprising patch interior intensity average (Mean), standard deviation (STD), coefficient of variation (coefficient of variance, CoV), the degree of bias (skewness, Skew), kurtosis (kurtosis, Kurtf), also have entropy of histogram (EtHis) in addition such as formula (7), brightness entropy (EtBrt) such as formula (8), their computing formula is respectively:

$\mathrm{EtHis}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{p}_i{\log }_2\left(p_i\right)---\left(7\right)$

$\mathrm{EtBrt}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}\frac{{\mathrm{\ω}}_i{\log }_2{\mathrm{\ω}}_i}{{\log }_{2}N}---\left(8\right)$

Wherein, P _i(i=0,1 ..., 255) be the probability that GTG i occurs in the picture, ω _ibe after normalization be not 0 image pixel intensities, N is the sum of all pixels of original image;

4) sorter training

The object of sorter training is a kind of approaching problem true model, the sorter model that this method is used is support vector machine (SVM), the method is a kind of machine learning method of classics, that the VC being based upon Statistical Learning Theory ties up on theoretical and Structural risk minization basis, between the complicacy (namely to the study precision of specific training sample) and learning ability (namely identifying the ability of arbitrary sample error-free) of model, optimal compromise is sought, to obtaining best Generalization Ability (or claiming generalization ability) according to limited sample information.

The training sample of sorter be from population sample random selecting wherein 1/2, remaining 1/2 be used to inspection-classification performance and Optimum Classification device.

As shown in Figure 4, this process is mainly divided into following step to the training flow process of sorter:

(4.1) collection of training sample and demarcation

Five kinds of MRI sequences such as mainly T1 (T1 weighted imaging), T1GD (T1 contrast agent imaging) that training sample adopts, T2 (T2 weighted imaging), PD (proton density imaging), STIR (short upset flip-flop transition reverts to picture), wherein T1 sequence image is as reference image (T1GD, T2, PD, STIR sequence registration take T1 as reference), in addition, doctor manually demarcates atherosclerotic plaque composition and also operates on T1 image, and other four kinds of sequence plaque component are that the result of demarcating according to T1 maps thereon.

(4.2) work that sample preprocessing will complete is to the quantization means of sample calibration result to sample image and doctor, this process first gray-scale value stretched operation is done to sample image, make its grey value profile scope [0,255] interval range, then be to image denoising sonication, finally that doctor done binary conversion treatment to the calibration result of plaque component, in the two values matrix that each composition of the patch namely doctor demarcated is stored in.

(4.3) training sample is classified and characteristic quantification process further

The further classification of training sample is that sample is subdivided into lipid (fat) sample set, fibrous (fiber) sample set, calcium (calcification) sample set and hemorrhage (thrombus) sample set.

Characteristic quantification process is that the Patch properties extracted is done normalized, its objective is that the distribution range after all characteristic quantifications making to extract is [-1,1] interval, because the feature object of different characteristic components does not coexist, the order of magnitude there is bigger difference, but characteristic component large in the linear classification function of SVM is larger than the impact of little characteristic component, but this can not illustrate that large feature is more important, be necessary is normalized unification to feature, shown in (9).

$x_i=\frac{N(x_i-\stackrel{\‾}{x})}{\underset{i}{\overset{N}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})}---\left(9\right)$

Wherein, x _icharacteristic component numerical value, the mean value of representative feature component x, N is the number of training sample;

All features form a proper vector after normalization, this proper vector shown in (10).

$\stackrel{\→}{F}=[\stackrel{\→}{V},\stackrel{\→}{G},\stackrel{\‾}{G}\stackrel{\→}{M},\stackrel{\→}{L},\stackrel{\→}{T}]---\left(10\right)$

In formula, V represents the gray-scale value of patch, and G represents the Gaussian Blur value of patch, and GM represents the gray feature after gray-scale value Gaussian Blur, and L represents Laplce's feature, and D represents patch locus feature, and T represents the textural characteristics of patch.

(4.4) sorter training, the symbiosis of last four kinds of sequence MRI artery plaque images becomes 4 feature sets, and each feature set can train generation sorter, finally obtain 4 sorters, i.e. lipid sorter, fibrous sorter, calcium sorter, hemorrhage sorter.

The training of sorter is actually convex optimization problem, by training sample optimization lineoid H:

ω·x+b＝0 (11)

Wherein, ω is weight vectors, and x is proper vector, and b is a parameter, and this lineoid with the form of maximum boundary by positive and negative sample area separately.

5) classifier optimization

4 sorters (lipid, fibrous, calcium, hemorrhage) that classifier optimization process obtains respectively are done and are optimized, its optimize flow process as shown in Figure 5, below with to the optimizing process of lipid sorter be example illustrate optimizing process.

First, sample set is divided into training set and test set two parts, this two-part choosing is random, both respectively account for 50%, then use the sample training sorter of training set, with the discrimination of the sorter verification test collection obtained, judge that circulation is gone down successively, until the lipid training aids classification accuracy rate obtained reaches maximal value, lipid optimum classifier thought by the lipid sorter now obtained.The optimization that uses the same method of other component classifier obtains.

6) plaque component identification and morphological operation

Plaque component identification is the core of the method, and for some pixels of plaque, by four classifier calculated, it is under the jurisdiction of the degree of different tissues respectively in the method, and its process as shown in Figure 6.

(6.1) first T1, T1GD, PD, out, the textural characteristics such as corresponding to pixel P comprises P point at T1, T1GD in the feature extraction that STIR plaque image is pointed out in each pixel, textural characteristics T (P)=[T in PD, STIR tetra-kinds of sequences ₁(P), T ₂(P), T ₃(P), T ₄(P)] extract respectively.Same method extracts other five kinds of features.

(6.2) Patch properties extracted in (6.1) is put in four sorters after optimizing in step (5) respectively, four kinds of sorters calculate its degree being under the jurisdiction of its composition respectively.

(6.3) according to the degree of membership competition sequence in (6.2) that obtain, using composition the highest for degree of membership as net result.

7) morphological operation of result

Object is to get rid of the impact of noise on segmentation result, gets rid of isolated to be organized in branch and segmentation result cavity etc.Concrete operations are as follows:

(7.1) result of composition segmentation is stored in 4 bianry images, in bianry image, 1 represents structural constituent, and 0 represents background, is lipid composition, fibrous composition, calcium composition respectively, hemorrhage composition.

(7.2) then erosion operation is done to each bianry image, isolated block (number of pixels is less than or equal to ten) filtering, because too little component block is a kind of interference to diagnosis, this composition is very likely that picture noise causes in addition

(7.3) then do morphology opening operation, the object of opening operation is the cavity filled up in component block and is communicated with close isolated block, increases that it is connective.

(7.4) finally do closing operation, this step is to offset the expansion of opening operation to structural constituent.

Claims (3)

1., based on an atherosclerotic plaque composition dividing method for multiple features study, it is characterized in that: described dividing method comprises the following steps:

1) multisequencing arteries patch image is obtained;

2) Image semantic classification

Choose ROI centered by arteries inwall as survey region, to window width, the adjustment of window position that the ROI chosen does gray-scale value, then denoising is done to ROI;

3) Patch properties extracts and describes

The feature of patch is described respectively from the gray-scale value of patch image, Gaussian Blur feature, Grad, patch to the minimum distance of Ink vessel transfusing outer wall, Laplce's feature of image and the textural characteristics of patch;

4) sorter training: sorter model is support vector machines, and training process is as follows:

(4.1) collection of training sample and demarcation

In training sample, T1 sequence image is as reference image, and it is operate on T1 image that doctor manually demarcates atherosclerotic plaque composition, and the plaque component of other four kinds of sequence T1GD, T2, PD and STIR is that the result of demarcating according to T1 maps thereon;

(4.2) sample preprocessing

First gray-scale value stretched operation is done to sample image, make its grey value profile scope [0,255] interval range, then be to image denoising sonication, finally that doctor done binary conversion treatment to the calibration result of plaque component, in the two values matrix that each composition of the patch namely doctor demarcated is stored in;

(4.3) training sample is classified and characteristic quantification process further

The further classification of training sample is that sample is subdivided into lipid sample set, fibrous sample set, calcium sample set and hemorrhage sample set;

Characteristic quantification process is that the Patch properties extracted is done normalized, and all features form a proper vector after normalization, this proper vector shown in (10):

$\stackrel{\→}{F}=[\stackrel{\→}{V},\stackrel{\→}{G},\stackrel{\‾}{G}\stackrel{\→}{M},\stackrel{\→}{L},\stackrel{\→}{T}]---\left(10\right)$

In formula, V represents the gray-scale value of patch, and G represents the Gaussian Blur value of patch, and GM represents the gray feature after gray-scale value Gaussian Blur, and L represents Laplce's feature, and D represents patch locus feature, and T represents the textural characteristics of patch.；

(4.4) sorter training, the symbiosis of last four kinds of sequence MRI artery plaque images becomes 4 feature sets, each feature set training generation sorter, finally obtain 4 sorters, i.e. lipid sorter, fibrous sorter, calcium sorter, hemorrhage sorter;

5) training aids optimization

Sample set is divided into training set and test set two parts, then the sample training sorter of training set is used, with the discrimination of the sorter verification test collection obtained, judge that circulation is gone down successively, until the training aids classification accuracy rate obtained reaches maximal value, optimum classifier thought by the sorter now obtained;

6) for some pixels of plaque, by four classifier calculated, it is under the jurisdiction of the degree of different tissues respectively, and process is as follows:

(6.1) feature extraction first T1GD, T2, PD and STIR plaque image pointed out in each pixel out;

(6.2) Patch properties extracted in (6.1) is put in four sorters after optimizing in step (5) respectively, four kinds of sorters calculate its degree being under the jurisdiction of its composition respectively;

(6.3) according to the degree of membership competition sequence in (6.2) that obtain, using composition the highest for degree of membership as net result.

2. as claimed in claim 1 a kind of based on multiple features study atherosclerotic plaque composition dividing method, it is characterized in that: described dividing method is further comprising the steps of:

7) morphological operation of result, process is as follows:

(7.1) result of composition segmentation is stored in 4 bianry images, in bianry image, 1 represents structural constituent, and 0 represents background, is lipid composition, fibrous composition, calcium composition respectively, hemorrhage composition;

(7.2) then erosion operation is done to each bianry image, isolated block filtering;

(7.3) then do morphology opening operation, fill up the cavity in component block and be communicated with close isolated block;

(7.4) finally do closing operation, obtain atherosclerotic plaque composition.

3. as claimed in claim 1 a kind of based on multiple features study atherosclerotic plaque composition dividing method, it is characterized in that: described step 3) in,

A () gray-scale value is at the gray-scale value of image after denoising;

(b) Gaussian Blur feature

Described Gaussian Blur feature refers to the gray-scale value of plaque a bit after Gaussian Blur, and leaching process is that the gaussian kernel of an employing 5*5 and plaque do convolution, shown in (2);

$G(x,\mathrm{\σ})=\frac{1}{\sqrt{2\mathrm{\π\σ}}}{e}^{-\frac{{\left|\right|x\left|\right|}^2}{2{\mathrm{\σ}}^2}}*I---\left(2\right)$

In formula, σ represents the yardstick of gaussian kernel, and x represents the gray-scale value of image, and I, G distinguish the image after representative image primitive vessel patch image and Gaussian Blur, and " * " is convolution symbol;

(c) Gradient Features

The Gradient Features GM of plaque:

$\mathrm{GM}=\sqrt{{\left(\frac{\∂\left(G(x,\mathrm{\σ})\right)}{\∂x}\right)}^2+{\left(\frac{\∂\left(G(x,\mathrm{\σ})\right)}{\∂y}\right)}^2}---\left(3\right)$

(d) Laplce feature L:

$L=\frac{{\∂}^2\left(G(x,\mathrm{\σ})\right)}{\∂x^2}+\frac{{\∂}^2(I,(x,\mathrm{\σ}))}{\∂x^2}---\left(4\right)$

(e) patch locus feature (distance between inside and outside wall)

D _min(I)＝min||C(I(x))-Lumen(I)|| (5)

D _max(I)＝max||C(I(x))-Lumen(I)|| (6)

(f) textural characteristics

Be extracted first order statistic and gray level co-occurrence matrixes feature two class textural characteristics, totally seven first order statistic, comprising patch interior intensity average, standard deviation, coefficient of variation, the degree of bias, kurtosis, entropy of histogram and brightness entropy; EtHis such as formula (7) and brightness entropy EtBrt such as formula (8):

$\mathrm{EtHis}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}{p}_i{\log }_2\left(p_i\right)---\left(7\right)$

$\mathrm{EtBrt}=-\underset{i=0}{\overset{255}{\mathrm{\Σ}}}\frac{{\mathrm{\ω}}_i{\log }_2{\mathrm{\ω}}_i}{{\log }_{2}N}---\left(8\right)$

Wherein, P _i(i=0,1 ..., 255) be the probability that GTG i occurs in the picture, ω _ibe after normalization be not 0 image pixel intensities, N is the sum of all pixels of original image.