CN106485203A - Carotid ultrasound image Internal-media thickness measuring method and system - Google Patents

Abstract

The invention relates to the field of carotid ultrasound image processing, and aims to provide a reliable and automatic measurement method for carotid ultrasound images, using a computer-aided measurement method to avoid the defects of the manual measurement method. The technical scheme adopted in the present invention is a method and system for measuring intima-media thickness in carotid artery ultrasound images, the steps are as follows: 1) Image cropping, clipping removes irrelevant information, and only retains the blood vessel part; 2) Automatically extracts the region of interest (ROI (Region of Interest); 3) image filtering, to filter out the speckle noise caused by echo reflection; 4) level set segmentation; 5) initial marker field; 6) re-segmentation; In addition to misclassifying small areas, the final segmented image and LII and MAI were obtained, and the intima-media thickness was measured. The invention is mainly applied to carotid ultrasound image processing.

Description

Carotid artery ultrasound image intima-media thickness measurement method and system

technical field

The invention relates to the field of carotid ultrasound image processing, in particular to an intima-media thickness measurement algorithm for carotid ultrasound images. Specifically, it relates to a method and system for measuring intima-media thickness in carotid ultrasound images.

Background technique

Intima Media Thickness (IMT) of the carotid artery wall in ultrasound imaging can be used as an important index to evaluate the early lesion degree of cardiovascular diseases, and it is of great value in the diagnosis and prevention of sudden myocardial infarction and stroke. Traditionally, the intima-media thickness of carotid ultrasound images is usually measured manually, and the measurer manually draws the lumen-intima interface (LII) and media-adventitia boundary (Media-adventitia) in the image. Adventitia Interface, MAI), and then obtain the IMT by calculating the distance between the two boundaries. However, this measurement method requires a lot of work, is very time-consuming, and the final result is affected by the tester's training, personal experience and subjective judgment. The shortcomings of manual measurement have promoted the development of computer-aided IMT measurement methods. The present invention proposes an algorithm for measuring intima-media thickness in carotid ultrasound images combined with level set segmentation and a Markov Random Field (MRF) model.

The level set method based on geometric deformation model was proposed by Osher and Sethian in 1982. The method is to use partial differential equations as a numerical analysis technique and is widely used in the motion tracking of contour surfaces or contour lines. This method transforms the problem of evolution of low-dimensional closed curves into an implicit way of solving the overall evolution of functions in high-dimensional space, so it can be adapted to the processing of topology changes. Central to the idea of the level set method is the Hamilton-Jacobi equation, which performs the numerical solution of a time-based equation for an implicit surface of motion. The basic principle of this method is to embed a curve or surface as a zero level set into a higher one-dimensional level set function, and express the evolution process of a low-dimensional curve or surface through a higher-dimensional function. The zero level set is embedded in the level set function, as long as the position of the zero level set is determined, the result of the motion curve or surface evolution can be determined. The evolution process of the curve model is briefly analyzed below. Plane closed curve C(s,t)=(m(s,t),n(s,t)):0≤s≤1 (s is the arc length parameter, t represents time, m and n are the coordinates of the plane curve ) is implicitly expressed as a three-dimensional continuous function surface φ(m,n,t) by the level set method: An equivalent curve with the same function value, let the function φ be a signed distance function, and the change of the curve C can be attributed to the corresponding change of the function. Therefore, the time-varying closed curve can be expressed as the change of the level set function with time:

The above formula transforms the evolution of the curve into the analysis of the level set function.

The evolution process of curve C along the normal direction can be represented by the following partial differential equation:

The above formula is called the level set equation (Level Set Equation), the function V is called the speed function (SpeedFunction), which represents the evolution speed of a point on the curve, and N is the normal direction of the curve. Taking the total differential of formula (2), we get:

In the formula is the gradient of φ, in the same direction as the normal to the curve. Assuming Ω is the whole image, the part of the level set function inside the curve C is negative, and the part outside is positive, then the inward unit normal vector of the level set is

Substituting formula (2) and formula (4) into formula (3), we can get

In this way, the curve evolution process is expressed by the level set equation, that is, the Euclidean expression of the curve evolution equation, which is a partial differential equation of the Hamilton-Jacobi type. Given the partial differential equation of the geometric active contour model and the initial level set function φ(m,n):φ(C)=0, formula (5) can ensure that the level set function φ(m,n,t) follows The evolution of time satisfies the condition of {φ(C(s,t),t)=0}, that is, the zero level set of φ(m,n,t) is always the contour line C(s,t).

Contents of the invention

In order to overcome the deficiencies of the prior art, the present invention aims to propose a reliable and automatic carotid artery ultrasound image measurement method, which uses computer-aided measurement to avoid the defects of manual measurement. The technical scheme adopted in the present invention is a method for measuring intima-media thickness in carotid ultrasound images, the steps of which are as follows:

1) Image cropping, cropping removes irrelevant information, and only retains the blood vessel part;

2) Automatically extract ROI (Region of Interest);

3) Image filtering to filter out the speckle noise caused by echo reflection;

4) Level set segmentation, using the level set segmentation algorithm to track the lumen-intima boundary LII (Lumen-Intima Interface,) of the vessel wall;

5) The initial labeling field, according to the approximately parallel characteristics of each layer of the carotid artery wall, in the ROI, the intima boundary of the vessel wall is translated downward by a certain pixel distance, which is used as the media-adventitia boundary media-adventitia boundary MAI( Media-AdventitiaInterface,), and then divide the ROI into three parts: lumen, intima-media, and adventitia according to these two boundaries, and use the obtained segmented image as the initial label field of the Markov model;

6) re-segmentation, introducing the Markov model, and re-segmenting the image according to the initial label field and the gray field;

7) Post-processing, post-processing the segmented image, removing misclassified small areas, obtaining the final segmented image and LII, MAI, and measuring the intima media thickness IMT (Intima Media Thickness).

Step 2) The specific steps are: according to the statistical law of the gray scale of the ultrasonic image, adopt the maximum inter-class difference method to adaptively obtain the gray threshold value, and binarize the ultrasonic image; For the hole in the image, use the principle of larger connected domain area and larger ordinate of the centroid to select the desired white area, and the upper boundary where it intersects with the black area is considered to be the approximate position of LII; finally, the approximate position of LII is used as the standard, Take 40 pixels up and 20 pixels down, respectively, as the upper and lower boundaries of the ROI, and obtain two ROIs corresponding to the original image.

Step 3) Image filtering is based on the filtering of distance similarity and gray similarity between pixels, and the filtering output is

where g(r) is the filtered image, is the similarity function of pixel ξ and pixel r, is a normalization function, σ ₁ is the domain variance, σ ₂ is the range variance;

Step 4) is further refined into:

Step1: Initialization, assuming that the level set function φ(m,n,t _i ) at time t _i is known, m and n are the coordinates of the plane curve;

Step2: Solve the level set equation: obtain the value of the level set function φ(m,n,t _{i +1} ) in the entire image area at time t i+1, and the evolution curve recorded as Γ _i+1 at this time is φ(m ,n,t _i+1 ), that is, Γ _i+1 ＝{φ(m,n,t _i+1 )＝0}, at this time φ(m,n,t _i+1 ) is no longer a signed distance function;

Step3: Re-initialize, use φ(m,n,t _i+1 ) to replace the initial level set function φ ₀ in the following formula, iteratively solve to a stable solution, which is still recorded as φ(m,n,t _i+1 ):

Where φ _t is the level set function at time t, and φ(m,n) is the initial level set function;

Step4: Combining the value of φ(m,n,t _i+1 ), solve the governing equation of the physical quantity, and obtain the value of the physical quantity at time t _i+1 ;

Step5: Repeat the steps to enter the calculation at the next moment until it stops.

Step 6) set up the Markov model of image to be segmented, for the solution of Markov model, flow process is as follows:

Step1: input the original image to be divided;

Step2: Initialize the image to obtain the initial mark field of the image;

Step3: On the basis of the label field in the previous step, each pixel in the image is traversed to obtain a label that satisfies the iteration condition, and the original label field is updated;

Iterate Step3 until the convergence condition.

Step 6) is further refined as:

Image segmentation is the process of assigning a label to each pixel based on its feature attributes and region attributes. In the Markov Random Field model (Markov Random Field, MRF), two random fields are used to describe the image to be segmented, one is the label field X, often called the hidden random field, described by the prior distribution P(X) The local correlation of the label field; the other is the gray field or feature field Y, which is conditioned on the label field and uses the distribution function P(Y|X) to describe the distribution of the observed data or feature vectors, where the label field X is the Markov field, which satisfies the Markov property, and its prior distribution is expressed as

Where A={1,…a} is the set of pixels in the image, a is the total number of pixels in the image, x _i ∈ {1,…,q} is the category label of pixel i, q is the total number of pixel categories in the image, N _i is the neighborhood of pixel i, according to the Hammersley-Clifford equivalence theorem, the joint probability distribution of all pixels is

where Z is the normalization constant and U(x) is the prior energy

c is a group, that is, a set containing several positions, x _i and x _j are the category labels of pixel points i and j, in extreme cases, all subsets of the neighborhood structure are considered to be groups, and C represents a set of groups , V _c (x) is the potential function defined on the group c, which is only related to the neighboring pixels of the pixel, where V ₁ ( _xi ) corresponds to the neighborhood correlation function, and the potential group model chooses the second-order Potts The model is estimated, that is, only the interaction between two points in the second-order neighborhood system is considered, and the prior probability approximated by the pseudo-likelihood is expressed as:

Potts model:

Among them, β is a function to control the interaction strength between neighborhoods, N _i and n _i are the neighboring pixels of pixel i;

Another random field Y refers to the image to be segmented, which is an observable random process, that is, P(Y) is fixed, then according to Bayes' theorem:

P(X|Y)∝P(X)P(Y|X) (13)

Transform the model into a solution to the likelihood function P(Y|X) and the prior probability P(X), where P(X|Y) is the posterior probability of the image, where the conditional distribution function P(Y|X), Usually described by a Gaussian distribution:

in is the variance of the pixel gray level of the same region marked as _xi , is the mean value of the neighborhood gray value of pixel i;

Introduce the Markov model, and solve the model according to formula (11) and formula (14).

Select the Potts model as the potential group expression function in formula (10), and use the conditional iterative algorithm ICM (Iterative Conditional Mode) to solve the Markov model.

The carotid ultrasound image intima-media thickness measurement system consists of an ultrasound device and a computer. The image information measured by the ultrasound device is sent to the computer for processing. The computer is equipped with the following modules:

1) Image cropping module, which crops and removes irrelevant information, and only retains the blood vessel part;

2) Automatically extract ROI (Region of Interest) module;

3) Image filtering module, which filters out the speckle noise caused by echo reflection;

4) The level set segmentation module uses the level set segmentation algorithm to track the lumen-intima boundary LII (Lumen-Intima Interface,) of the vessel wall;

5) The initial marker field module, according to the approximately parallel characteristics of each layer of the carotid artery wall, in the ROI, translate the intima boundary of the vessel wall downward by a certain pixel distance, and use it as the media-adventitia boundary media-adventitia boundary MAI (Media-Adventitia Interface,), and then divide the ROI into three parts: lumen, intima-media, and adventitia according to these two boundaries, and use the obtained segmented image as the initial label field of the Markov model;

6) Segmentation module again, introduces Markov model, divides image again to image according to initial label field and gray scale field;

7) The post-processing module is used to post-process the segmented image, remove misclassified small regions, obtain the final segmented image and LII, MAI, and measure the intima media thickness IMT (Intima Media Thickness).

Features and beneficial effects of the present invention are:

The essence of the level set algorithm is to use the zero contour line of the higher one-dimensional surface as the target boundary to segment the lower one-dimensional target. If one looks at low latitudes from high dimensions, topological changes at low latitudes appear as morphological changes of curved surfaces at high latitudes, and will not cause changes in the topological structure of curved surfaces. So the level set algorithm has quite strong low-latitude topological variability. The invention introduces the level set method as the initial segmentation algorithm of the ROI, and the method is stable and helps to obtain reliable results. The Markov model can effectively describe the spatial information of the neighboring pixels in the image, and can make the measurement results of the computer more accurate. The present invention strongly supports the measurement of intima-media thickness in clinical carotid ultrasound images, provides a reference for the further optimization and development of IMT computer-aided measurement technology, and is a good supplement to manual measurement by experts.

Description of drawings:

Figure 1 algorithm flow chart;

Figure 2 Initial carotid ultrasound image;

Figure 3 cropped image;

Figure 4 ROI;

Figure 5 filtered image;

Figure 6 Initial LII boundary;

Figure 7 initial LII, MAI boundaries;

Figure 8 initial labeling field;

Figure 9 Label field after iteration;

Fig. 10 final segmentation image;

Figure 11 Final LII, MAI boundaries.

detailed description

The purpose of the present invention is to propose a reliable and automatic carotid artery ultrasonic image measurement algorithm, which uses computer-aided measurement to avoid the defects of manual measurement. Based on this purpose, aiming at the current research status at home and abroad, the present invention adopts the level set algorithm to initially segment the measurement image, and introduces the Markov model to transform the image segmentation problem into a process of assigning a label to each pixel.

The invention combines the characteristics of the ultrasonic carotid artery image, and can effectively complete the work of segmenting the carotid artery image and measuring the IMT during the test process of the given image library, and has good theoretical and application value.

In order to achieve the above object, the present invention adopts the following technical solutions:

1) Image cropping. Information about the patient and the ultrasound instrument is distributed around the initial ultrasound image. The existence of this information will affect the subsequent image processing steps. Therefore, it is necessary to crop and remove these parts, and only keep the blood vessel part.

2) Region of Interest (ROI) extraction. Automatic extraction of ROI can avoid the tedious manual segmentation of ROI and realize the purpose of automatic measurement of IMT.

3) Image filtering. With the advantages of real-time, repeatability, non-invasiveness and low cost, ultrasound images have become the preferred imaging method for carotid artery examination. However, the speckle noise brought by echo reflection will not only reduce the imaging quality of the image, but also increase the difficulty of computer processing. By filtering the image, the influence of noise can be reduced, and more reliable results can be obtained.

4) Level set segmentation. The lumen-intima boundary ((Lumen-Intima Interface, LII)) of the vessel wall was traced using a level set segmentation algorithm.

5) Initial label field. According to the approximately parallel characteristics of each layer of the carotid artery wall, in the ROI, the intima boundary of the vessel wall is translated downward by a certain pixel distance as the Media-Adventitia Interface (MAI) initial state. Then, according to these two boundaries, the ROI was segmented into lumen, intima-media, and adventitia, and the obtained segmented image was used as the initial label field of the Markov model.

6) Re-division. The Markov model is introduced to segment the image again according to the initial label field and the gray field.

7) post-processing. After processing the segmented image, remove the misclassified small area, obtain the final segmented image and LII, MAI, and measure the IMT.

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

1) Image cropping. Under the condition of ensuring the existence of the distal vessel wall, a region with a size of 320×300 pixels in the lower part of the ultrasound image was intercepted.

2) Region of Interest (ROI) extraction.

Automatically pick a threshold and binarize it. Ideally, the gray levels of the lumen and wall of the blood vessel in the ultrasound image are dark and bright, respectively. Therefore, according to the statistical law of the gray level of the image, the maximum inter-class difference method is used to adaptively obtain the gray level threshold, and the binary value image.

Find the approximate location of the distal vessel wall. Due to the different positions of blood vessels in ultrasound images and the existence of plaque noise, there may be multiple white areas in the binarized image. In order to find the white area corresponding to the distal vessel wall, the binary image morphological closed operation is performed to fill the holes in the image, and according to the characteristics that the distal vessel wall is located below the ultrasonic image, the large area of the connected domain and the vertical coordinate of the centroid are used The larger principle is to select the desired white area, and the upper boundary where it intersects with the black area can be considered as the approximate position of LII.

Extract ROIs. The IMT of normal carotid vessels is in the range of 0.5mm to 1mm, corresponding to about 8-16 pixels in the ultrasound image. Considering that there may be bends and lesions in the blood vessels, we take the approximate position of the LII as the standard, and take 40 pixels up and 20 pixels down as the upper and lower boundaries of the ROI, respectively. In this step, two ROIs respectively corresponding to the original image can be obtained.

3) Image filtering.

The present invention selects a bilateral filtering algorithm to filter two ROI images respectively. The bilateral filtering algorithm is a nonlinear filtering method with a small amount of calculation. It has a better performance in the denoising processing of medical images, and can better maintain the edge information of the image while removing the noise. The algorithm takes into account the distance similarity and gray value similarity between image pixels at the same time, and can be described by Gaussian distribution. For the image f(r), the filtering output based on the distance similarity and gray similarity between pixels is:

where g(r) is the filtered image, is the similarity function of pixels ξ and r, is the normalization function. σ ₁ is the domain variance, σ ₂ is the range variance;

4) Level set segmentation.

Step1: Initialization, assuming that the level set function φ(m,n,t _i ) at time t _i is known, m and n are the coordinates of the plane curve;

Step2: Solve the level set equation: obtain the value of the level set function φ(m,n,t _{i +1} ) in the entire image area at time t i+1, and the evolution curve recorded as Γ _i+1 at this time is φ(m ,n,t _i+1 ), that is, Γ _i+1 ＝{φ(m,n,t _i+1 )＝0}, at this time φ(m,n,t _i+1 ) is no longer a signed distance function;

Step3: Re-initialize, use φ(m,n,t _i+1 ) to replace the initial level set function φ ₀ in the following formula, iteratively solve to a stable solution, which is still recorded as φ(m,n,t _i+1 ):

Among them, φ _ti+1 is the level set function at time t _i+1 .

Step4: Combining the value of φ(m,n,t _i+1 ), solve the governing equation of the physical quantity, and obtain the value of the physical quantity at time t _i+1 ;

Step5: Repeat the steps to enter the calculation at the next moment until it stops.

5) Initial label field.

Create an initial label field. According to the vertical edge map of the ROI image, the initial LII is translated down a certain distance (the maximum position of the average gradient value in the horizontal direction) as the initial mark of the MAI. The ROI image is divided into three parts by LII and MAI, from top to bottom: lumen (marked as 1), intima-media (marked as 2) and adventitia (marked as 3).

6) Re-division.

Image segmentation is the process of assigning a label to each pixel based on its feature attributes and region attributes. In the Markov Random Field model (Markov Random Field, MRF), two random fields are commonly used to describe the image to be segmented, one is the label field X, often called the hidden random field, described by the prior distribution P(X) Local correlation of label fields. The other is the gray field or feature field Y, which is often conditioned on the label field, and the distribution function P(Y|X) is used to describe the distribution of observed data or feature vectors. The label field X is a Markov field, which satisfies the Markov property, and its prior distribution can be expressed as

Where A={1,…a} is the set of pixels in the image, a is the total number of pixels in the image, x _i ∈ {1,…,q} is the category label of pixel i, q is the total number of pixel categories in the image, N _i is the neighborhood of pixel i. According to the Hammersley-Clifford equivalence theorem, the joint probability distribution of all pixels is

where Z is the normalization constant and U(x) is the prior energy

c is a group, that is, a set containing several positions, x _i and x _j are the category labels of pixel points i and j, in extreme cases, it can be considered that all subsets of the neighborhood structure are groups, and C represents the group gather. V _c (x) is the potential function defined on the group c, which is only related to the neighboring pixels of the pixel. In formula (10), V ₁ ( _xi ) corresponds to the neighborhood correlation function. The potential group model chooses two The first-order Potts model is estimated, that is, only the interaction between two points in the second-order neighborhood system is considered. Then the prior probability approximated by the pseudo-likelihood is expressed as:

Potts model:

Among them, β is a function to control the interaction strength between neighborhoods, and N _i and n _i are the neighboring pixels of pixel i.

Another random field Y refers to the image to be segmented, which is an observable random process, that is, P(Y) is fixed. Then according to Bayes' theorem:

P(X|Y)∝P(X)P(Y|X) (13)

Transform the model into a solution to the likelihood function P(Y|X) and the prior probability P(X). Here P(X|Y) is the posterior probability of the image, where the conditional distribution function P(Y|X) is usually described by a Gaussian distribution:

in is the variance of the pixel gray level of the same label (marked as _xi ), is the mean value of the neighborhood gray value of pixel i.

Introduce the Markov model, and solve the model according to formula (11) and formula (14). The Potts model is selected as the potential group expression function in formula (10). In the present invention, a conditional iterative algorithm (Iterative Conditional Mode, ICM) is used to solve the Markov model. ICM is a deterministic algorithm capable of obtaining local optimal solutions. In the iterative process of each step of the algorithm, the maximization is required so that the solution process can converge too quickly to reach the local optimal solution. The calculation amount of ICM is greatly reduced.

7) post-processing.

We use the maximum connected domain criterion to remove discrete small regions to obtain the final segmented image. The boundary of the second part of the intima-media (marked 2) in the segmented image was extracted with a Sobel operator to obtain the final contours of LII and MAI. And calculate the average distance between boundaries, as the computer-aided carotid IMT distance.

Claims (7)

1. a carotid ultrasound image intima-media thickness measurement method, is characterized in that, the steps are as follows: 1) Image cropping, cropping removes irrelevant information, and only retains the blood vessel part; 2) Automatically extract ROI (Region of Interest); 3) Image filtering to filter out the speckle noise caused by echo reflection; 4) Level set segmentation, using a level set segmentation algorithm to track the lumen-intima boundary LII (Lumen-IntimaInterface,) of the vessel wall; 5) The initial labeling field, according to the approximately parallel characteristics of each layer of the carotid artery wall, in the ROI, the intima boundary of the vessel wall is translated downward by a certain pixel distance, which is used as the media-adventitia boundary media-adventitia boundary MAI( Media-AdventitiaInterface,), and then divide the ROI into three parts: lumen, intima-media, and adventitia according to these two boundaries, and use the obtained segmented image as the initial label field of the Markov model; 6) re-segmentation, introducing the Markov model, and re-segmenting the image according to the initial label field and the gray field; 7) Post-processing, post-processing the segmented image, removing misclassified small areas, obtaining the final segmented image and LII, MAI, and measuring the intima media thickness IMT (Intima Media Thickness). 2. the carotid ultrasound image intima-media thickness measurement method as claimed in claim 1, is characterized in that, step 2) specific steps are: according to the statistical law of ultrasound image grayscale, adopt maximum inter-class difference method adaptively to obtain grayscale degree threshold, and binarize the ultrasonic image; then, perform a morphological closed operation on the binarized ultrasonic image, and then fill the holes in the image, and select the desired white area by using the principle of larger connected domain area and larger vertical coordinate of the centroid, The upper boundary where it intersects with the black area is considered to be the approximate position of LII; finally, taking the approximate position of LII as the standard, take 40 pixels up and 20 pixels down as the upper and lower boundaries of the ROI respectively, and obtain two corresponding to the original Image ROIs. 3. carotid ultrasound image intima-media thickness measurement method as claimed in claim 1, is characterized in that, step 3) image filtering is based on the filtering of inter-pixel distance similarity and gray scale similarity, and filtering output is: $\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; k</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span>{<span\; class="notranslate">\; \&Integral;</span>}_{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&infin;</span>}}^{\mathrm{<span\; class="notranslate">\; \&infin;</span>}}{<span\; class="notranslate">\; \&Integral;</span>}_{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&infin;</span>}}^{\mathrm{<span\; class="notranslate">\; \&infin;</span>}}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&xi;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&xi;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&xi;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; \&xi;</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$ where g(r) is the filtered image, is the similarity function of pixel ξ and pixel r, is a normalization function, σ ₁ is the domain variance, and σ ₂ is the value domain variance. 4. carotid ultrasound image intima-media thickness measuring method as claimed in claim 1, is characterized in that, step 4) is further and refined as: Step1: Initialization, assuming that the level set function φ(m,n,t _i ) at time t _i is known, m and n are the coordinates of the plane curve; Step2: Solve the level set equation: obtain the value of the level set function φ(m,n,t _{i +1} ) in the entire image area at time t i+1, and the evolution curve recorded as Γ _i+1 at this time is φ(m ,n,t _i+1 ), that is, Γ _i+1 ＝{φ(m,n,t _i+1 )＝0}, at this time φ(m,n,t _i+1 ) is no longer a signed distance function; Step3: Re-initialize, use φ(m,n,t _i+1 ) to replace the initial level set function φ ₀ in the following formula, iteratively solve to a stable solution, which is still recorded as φ(m,n,t _i+1 ): $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&dtri;</span>\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; \&phi;</span>}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$ Where φ _t is the level set function at time t, and φ(m,n) is the initial level set function; Step4: Combining the value of φ(m,n,t _i+1 ), solve the governing equation of the physical quantity, and obtain the value of the physical quantity at time t _i+1 ; Step5: Repeat the steps to enter the calculation at the next moment until it stops. 5. carotid ultrasound image intima-media thickness measuring method as claimed in claim 1, is characterized in that, step 6) sets up the Markov model of image to be segmented, for the solution of Markov model, flow process is as follows: Step1: input the original image to be divided; Step2: Initialize the image to obtain the initial mark field of the image; Step3: On the basis of the label field in the previous step, each pixel in the image is traversed to obtain a label that satisfies the iteration condition, and the original label field is updated; Iterate Step3 until the convergence condition. 6. carotid artery ultrasound image intima-media thickness measuring method as claimed in claim 5, is characterized in that, step 6) is further refined as: image segmentation is based on the characteristic attribute of pixel and the region attribute to assign label to each pixel process. In the Markov Random Field model (Markov Random Field, MRF), two random fields are used to describe the image to be segmented, one is the label field X, often called the hidden random field, described by the prior distribution P(X) The local correlation of the label field; the other is the gray field or feature field Y, which is conditioned on the label field and uses the distribution function P(Y|X) to describe the distribution of the observed data or feature vectors, where the label field X is the Markov The field satisfies the Markov property, and its prior distribution is expressed as: $\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \}</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&ForAll;</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$ Where A={1,…a} is the set of pixels in the image, a is the total number of pixels in the image, x _i ∈ {1,…,q} is the category label of pixel i, q is the total number of pixel categories in the image, N _i is the neighborhood of pixel i, according to the Hammersley-Clifford equivalence theorem, the joint probability distribution of all pixels is: $\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; Z</span>}}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; x</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$ where Z is a normalization constant and U(x) is the prior energy: $\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; C</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\underset{<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; ...</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$ c is a group, that is, a set containing several positions, x _i and x _j are the category labels of pixel points i and j, in extreme cases, all subsets of the neighborhood structure are considered to be groups, and C represents a set of groups , V _c (x) is the potential function defined on the group c, which is only related to the neighboring pixels of the pixel, where V ₁ ( _xi ) corresponds to the neighborhood correlation function, and the potential group model chooses the second-order Potts The model is estimated, that is, only the interaction between two points in the second-order neighborhood system is considered, and the prior probability approximated by the pseudo-likelihood is expressed as: $\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; a</span>}}{\mathrm{<span\; class="notranslate">\; \&Pi;</span>}}}\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; a</span>}}{\mathrm{<span\; class="notranslate">\; \&Pi;</span>}}}\frac{\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&beta;n</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span>}{\underset{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; q</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&beta;n</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$ Potts model: Among them, β is a function to control the interaction strength between neighborhoods, N _i and n _i are the neighboring pixels of pixel i; Another random field Y refers to the image to be segmented, which is an observable random process, that is, P(Y) is fixed, then according to Bayes' theorem: P(X|Y)∝P(X)P(Y|X) (13) Transform the model into a solution to the likelihood function P(Y|X) and the prior probability P(X), where P(X|Y) is the posterior probability of the image, where the conditional distribution function P(Y|X), Usually described by a Gaussian distribution: $\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; A</span>}}{<span\; class="notranslate">\; \&Pi;</span>}\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; A</span>}}{<span\; class="notranslate">\; \&Pi;</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&pi;\&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>\frac{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 14</span>}<span\; class="notranslate">\; )</span>$ in is the variance of the pixel gray level of the same region marked as _xi , is the mean value of the neighborhood gray value of pixel i; Introduce the Markov model and solve the model according to formula (11) and formula (14); Select the Potts model as the potential group expression function in formula (10), and use the conditional iterative algorithm ICM (IterativeConditional Mode) to solve the Markov model. 7. A carotid ultrasound image intima-media thickness measurement system is characterized in that it is made up of an ultrasound device and a computer, and the image information measured by the ultrasound device is sent to a computer for processing, and the computer is provided with the following modules: 1) Image cropping module, which crops and removes irrelevant information, and only retains the blood vessel part; 2) Automatically extract ROI (Region of Interest) module; 3) Image filtering module, which filters out the speckle noise caused by echo reflection; 4) The level set segmentation module uses the level set segmentation algorithm to track the lumen-intima boundary LII (Lumen-Intima Interface,) of the vessel wall; 5) The initial marker field module, according to the approximately parallel characteristics of each layer of the carotid artery wall, in the ROI, the intima boundary of the vessel wall is translated downward by a certain pixel distance, and used as the media-adventitia boundary media-adventitia boundary MAI The initial state of (Media-AdventitiaInterface,), and then divide the ROI into three parts: lumen, intima-media, and adventitia according to these two boundaries, and use the obtained segmented image as the initial label field of the Markov model; 6) Segmentation module again, introduces Markov model, divides image again to image according to initial label field and gray scale field; 7) The post-processing module is used to post-process the segmented image, remove misclassified small areas, obtain the final segmented image and LII, MAI, and measure the intima media thickness IMT (Intima Media Thickness).