CN107610107B - A fractal-based method for 3D vascular plaque ultrasound image feature description - Google Patents

Abstract

The invention discloses a fractal dimension-based three-dimensional vascular plaque ultrasonic image feature description method. The method comprises the following steps: (1) acquiring a three-dimensional blood vessel ultrasound image; (2) segmenting a plaque outline from the collected three-dimensional blood vessel ultrasound image; (3) obtaining a three-dimensional blood vessel plaque ultrasound image from step (2). The fractal features of plaques are calculated in images, including: 3D box counting method and 3D blanket covering method, which describe the surface roughness and 3D texture changes of plaques, respectively. This method is the first to apply fractal theory to three-dimensional vascular ultrasound images to extract the surface changes and internal characteristics of plaques, and provides a quantitative analysis method for the prediction of atherosclerosis.

Description

A fractal-based method for 3D vascular plaque ultrasound image feature description

technical field

The invention belongs to the intersecting field of computer technology and medical images, and in particular relates to a feature extraction method of blood vessel plaques in ultrasonic images.

Background technique

In the previous studies on ultrasound images of vascular plaques, most of them analyzed the vulnerable characteristics of plaques by extracting the statistical features of two-dimensional B-ultrasound images, including: gray-scale median (GSM), mean, standard deviation, etc. In recent years, many scholars have proposed to use texture features to describe the characteristics of plaques, and have achieved certain results in distinguishing symptomatic and asymptomatic plaques. The most common texture features include: Spatial Gray Level Dependency Matrix (SGLDM), Gray Level Co-occurrence Matrix (GLCM), Gray Level Difference Statistics (GLDS), Law's Texture, Local Binary Pattern (LBP), Fourier Spectral Analysis, etc.

Three-dimensional ultrasound provides a more efficient, repeatable, and reliable means of detection and analysis of vascular plaques. Impact. The rupture of carotid plaque is one of the main factors in the occurrence of cerebrovascular diseases. The quantitative characterization of carotid plaque is of great significance in distinguishing vulnerable plaques and evaluating the efficacy of treatment. Landry and Fenster proposed to obtain the total plaque volume (Total plaque volume, TPV) from three-dimensional ultrasound images as a sign of atherosclerosis. Wannarong and parrage also verified the intima-media thickness (Intima-mediathickness, IMT), total plaque area (Total plaque area, TPA) and the law of TPV on the growth of atherosclerosis in clinical for the first time. In order to overcome the limitation of TPV, Egger and Spence proposed to describe atherosclerosis with vessel wall volume (Vessel wall volume, VWV).

The acquisition of features such as IMT, TPA, TPV, and VWV of vascular plaques in three-dimensional ultrasound images often requires manual participation, which is not only time-consuming and laborious, but also depends on the subjectivity of doctors, and the repeatability is poor. In previous studies, texture features were mainly obtained from two-dimensional B-ultrasound images, and some scholars used three-dimensional vascular ultrasound images to extract plaque texture features, but after extracting features from two-dimensional images of each section, they were combined to obtain 3D features, rather than true 3D features, do not consider the relationship between frames. On the one hand, it cannot obtain the change information of the three-dimensional surface; on the other hand, it obtains the texture inside the two-dimensional image, but cannot obtain the change relationship between the axial pixels, which is incomplete.

Contents of the invention

In view of the above defects or improvement needs of the prior art, the present invention provides a fractal-based three-dimensional vascular plaque ultrasonic image feature description method, the purpose of which is to use fractal theory to directly calculate the fractal feature of the three-dimensional vascular plaque ultrasonic image, To describe the surface roughness and internal texture changes of the plaque, thereby solving the problems of strong subjectivity and poor repeatability in the existing vascular plaque ultrasound image feature description method, and at the same time, considering the connection between the three-dimensional ultrasound image frames, It is more descriptive and robust.

In order to achieve the above object, according to one aspect of the present invention, a method for describing the characteristics of three-dimensional vascular plaque ultrasonic images is provided, which is characterized in that it includes the following steps:

(1) Obtaining a three-dimensional blood vessel ultrasound image;

(2) segment the plaque outline from the three-dimensional vascular ultrasound image obtained in step (1);

(3) Obtain the ultrasound image of the plaque area from the plaque outline obtained in step (2), and then obtain the three-dimensional surface roughness of the plaque from the ultrasound image of the plaque area according to the plaque outline and using the three-dimensional box counting method Fractal eigenvalues; according to the change of the gray value inside the plaque and using the three-dimensional blanket covering method to obtain the fractal eigenvalues representing the three-dimensional internal texture changes of the plaque;

The three-dimensional box counting method is as follows: by using cube boxes with different side lengths to cover the ultrasound image of the plaque area, a regression curve is obtained between the number of boxes and the logarithm of the side lengths of the boxes, and the slope of the curve is the three-dimensional plaque characteristic. Fractal eigenvalue of surface roughness; the side length of the box changes from 1 to the maximum value, and the maximum side length does not exceed the minimum side length of the 3D image;

The three-dimensional blanket covering method is as follows: the gray value of the ultrasound image voxel in the plaque area is set as a gray-scale fractal hypersurface in four-dimensional space, and then there are two hypersurfaces at a position distance from the fractal hypersurface ε, Calculate the volume of the region between the two hypersurfaces; by changing the value of ε from 1 to the maximum value, the logarithm of the volume value is obtained with the regression curve of the logarithm of the distance ε, and the grayscale fractal hypersurface is obtained. Fractal dimension value, the value of fractal dimension has a linear relationship with the slope of the curve; the maximum value of ε does not exceed the minimum side length and minimum gray value of the three-dimensional image; the fractal dimension value describes the change of the gray fractal hypersurface, that is The fractal eigenvalues that reflect the three-dimensional internal texture changes of plaques.

Preferably, the fractal feature extracted in the step (3) is the 3D surface feature and the internal texture feature of the 3D image directly extracted from the blood vessel 3D ultrasound image data, rather than the feature obtained by dividing the 3D ultrasound image into two-dimensional slices . It reflects the three-dimensional surface characteristics of the plaque, and at the same time, more comprehensively reflects the surface and internal characteristics of the plaque.

Preferably, the three-dimensional box counting method in the step (3) obtains the fractal eigenvalues representing the three-dimensional surface roughness of the plaque, and the steps are as follows:

(1) Cover the ultrasound image of the plaque area with a cube box whose side length is l, where l=1;

(2) Define the voxel box that only includes the inside of the plaque as BLACK; the voxel box that only includes the outside tissue of the plaque is defined as WHITE; the voxel box that includes both the inside of the plaque and the outside tissue is defined as GREY, GRAY Indicates that the patch outline passes through the box;

(3) Change the voxel box side length l, make l=2,3,4, ... M, M is the width of the ultrasound image of the plaque area; repeat step (2), until l is equal to the ultrasound image of the plaque area the width;

(4) Count the number of voxel boxes defining BLACK, WHITE, and GREY respectively;

(5) Record the number of voxel boxes counted in step (4) as N(l), where N(l) is the count of one or several voxel boxes of BLACK, WHITE, and GRAY, using the least squares method Fit the regression curve of log(N(l))/log(l);

(6) The slope of the regression curve in the step (5) corresponds to the fractal characteristic value of the three-dimensional surface roughness of the plaque.

Preferably, the box counting method in the step (3) counts the voxels inside (marked as BLACK) and border (marked as GREY) of the plaque in the three-dimensional ultrasound image.

Preferably, the three-dimensional blanket covering method obtains fractal eigenvalues representing the three-dimensional internal texture changes of plaques, and the steps are as follows:

(1) Set the gray value of the ultrasound image voxel in the plaque area as a gray hypersurface in the fourth-dimensional space independent of the three-dimensional voxel coordinates (x, y, z), then a voxel can be expressed as P(x,y,z,g(x,y,z));

(2) There are two upper and lower hypersurfaces at the position of the distance ε from the gray-scale fractal hypersurface, and the volume V(ε) of the ultrasound image in the plaque area is expressed as the hypervolume between the two hypersurfaces divided by 2ε, It is calculated from the following publicity:

Among them, u _ε (i, j, k) and b _ε (i, j, k) respectively represent the height of a point whose coordinates are (i, j, k) on the upper and lower hyper-surfaces of the fractal hyper-surface ε value, which is the gray value of the point in the ultrasound image of the plaque area;

The u _ε (i,j,k) can be obtained from the hypersurface of ε-1, and the approximate value of u _ε (i,j,k) Due to the consideration of the smoothness of the boundary and surface, the value of u _ε (i,j,k) is the maximum value of u _ε-1 (m,n,p) in the area where the distance point (i,j,k) is 1 value with the greater of the two;

The b _ε (i,j,k) is obtained from the hypersurface of ε-1, the approximate value of b _ε (i,j,k) Due to the consideration of the smoothness of the boundary and surface, the value of b _ε (i,j,k) is the minimum value of b _ε-1 (m,n,p) in the area where the distance point (i,j,k) is 1 value with the smaller of the two;

(3) Change different ε values, and use the least squares method to fit the curve of log(V(ε))/log(ε);

(4) Use the following formula to calculate the value of the three-dimensional fractal dimension of the ultrasound image:

in,

(5) D _E represents the object dimension in Euclidean space in the step (4) formula, for three-dimensional body D _E =3, namely: FD=3-p;

(6) The value of the three-dimensional fractal dimension calculated by the formulas in steps (4) and (5) is the fractal dimension characteristic value representing the three-dimensional internal texture change of the plaque.

Preferably, the plaque is carotid artery plaque, aortic plaque, coronary artery plaque or peripheral artery plaque.

Generally speaking, compared with the prior art, the above technical solution conceived by the present invention has the following characteristics and advantages:

(1) Using three-dimensional ultrasound images to extract the features of vascular plaques has more objective and rich information than the texture features extracted from two-dimensional B-ultrasound images. At the same time, it is not affected by the subjective factors of the collector and has better repeatability and robustness.

(2) For the first time, fractal features are used to describe three-dimensional vascular plaques.

(3) The method based on three-dimensional box counting provides a method for describing the surface roughness of three-dimensional vascular plaques. Rupture is often associated with plaque surface irregularities, and thus surface roughness features better describe plaque vulnerability.

(4) The blanket-covering based method provides a way to describe the texture variation inside the 3D vascular plaque. Combining this feature with the roughness feature not only expresses the surface characteristics of the plaque but also expresses the internal characteristics of the plaque, and describes the characteristics of the plaque more comprehensively.

(5) The method is simple and easy to operate, with a small amount of calculation and a fast processing speed, and is very suitable for extracting features of 3D ultrasound volume data.

Description of drawings

Fig. 1 is a flow chart of the present invention;

Fig. 2 is the plaque segmentation result of the three-dimensional carotid artery ultrasound image; (a) is the manual segmentation result of the carotid plaque intima and intima, (b) is the carotid plaque intima and intima obtained from the three-dimensional reconstruction;

Fig. 3 is the fitting curve that three-dimensional box counting method calculates fractal dimension value;

Fig. 4 is the distribution of fractal dimension changes and SVM classifier classification score results of two groups of patients taking atorvastatin and placebo;

Figure 5 is the ROC curve of the classification results of the SVM classifier for patients taking atorvastatin and placebo.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not constitute a conflict with each other.

Example 1

The three-dimensional carotid plaque ultrasound image feature description method provided by the present invention, as shown in Figure 1, comprises the following steps:

(1) Obtain the three-dimensional ultrasound volume data of the carotid artery. The actual three-dimensional carotid artery ultrasound images are derived from the clinic, including 28 patients with carotid artery stenosis of more than 60%. Among them, 14 patients in the test group (average age 68±8.6) took atorvastatin every day for 3 months, and 14 patients in the control group (mean age 70±9.4) taking placebo.

(2) Manually segment the three-dimensional carotid artery ultrasound image obtained in step (1) to obtain the contour boundary of the plaque. Figure 2 is a schematic diagram of an example of segmentation results, where (a) is the result of manual segmentation of carotid plaque intima, intima, and plaque contour, and (b) is the result of three-dimensional reconstruction.

(3) Using fractal theory to extract three-dimensional fractal features from carotid plaque three-dimensional volume data, including: plaque surface roughness (three-dimensional box counting method) and plaque internal texture features (three-dimensional blanket covering method).

The box counting method obtains the fractal dimension of the fractal object, which can be used to describe the surface roughness of the object. This method uses boxes with different side lengths to cover the image to obtain a regression curve of the number of boxes and the logarithm of the side length of the box, and the slope of the curve is the fractal eigenvalue;

The invention improves the traditional box counting method and is applied to the three-dimensional ultrasonic volume data of the carotid artery plaque to describe the surface roughness of the plaque. Specifically include the following steps:

(1) Generate voxel boxes with different side lengths, and the side length l of the box varies from 1 voxel to the size of the entire 3D volume data image.

(2) Define the voxel box contained inside the plaque as BLACK; define the voxel box including only the external tissue of the plaque as WHITE; define the voxel box including both the interior of the plaque and the external tissue as GREY, which means plaque The block boundary outline goes through this box.

(3) Count the number of voxel boxes defined as BLACK, WHITE, and GRAY respectively.

(4) Assuming that the number of marked boxes is N(l), the curve of log(N(l))/log(l) is fitted by the least square method. Figure 3 shows the regression curve results of the number of statistical boxes using BLACK and GRAY features.

(5) The slope of the above-mentioned regression curve promptly corresponds to the value of the fractal dimension;

The blanket covering method is used for multi-resolution texture analysis and object classification, which calculates a texture feature of a two-dimensional image grayscale.

Extend the blanket covering method to 3D voxel images, and use 3D fractal dimensions to represent the texture information of the gray surface of 3D voxel images, including the following steps:

(1) The voxel gray value of three-dimensional carotid plaque ultrasound is considered to form a fractal hypersurface in a fourth-dimensional space independent of the three-dimensional voxel coordinates (x, y, z), then a voxel can be Expressed as P(x,y,z,g(x,y,z)).

(2) Assuming that the upper and lower hypersurfaces are formed at a point with a distance of ε from the gray hypersurface, which is equivalent to being covered with a blanket with a thickness of 2ε, then the volume V(ε) of the three-dimensional image is expressed as the distance between the two hypersurfaces The hypervolume in between is divided by 2ε.

Let g(x,y,z) be a 3D voxel image matrix, and u _ε (i,j,k) and b _ε (i,j,k) denote the grayscale surfaces of the above two 3D images.

(3) Change the width ε of the covering blanket area from 1 to max, and use the least square method to fit the curve of log(V _ε )/log(ε), and the slope of the curve is used to estimate the size of the three-dimensional fractal dimension.

(4) Then the Minkowski dimension is equal to

where D _E is the Euclidean dimension of the object, and D _E is 3 for a three-dimensional image.

Test the described three-dimensional carotid plaque ultrasonic image feature description method, and use clinical data to evaluate the discrimination and effectiveness of the extracted features, as follows:

First, the t-hypothesis test was used to evaluate the discrimination of the extracted features. Table 1 compares the t-test p-values of the two extracted fractal features and traditional texture features. The smaller the p-value, the higher the discrimination of the features.

Table 1 t hypothesis test results

Then, a support vector machine (SVM) was used to classify whether the plaque belonged to patients taking atorvastatin or placebo to test the effect of our proposed 3D fractal feature on plaque characterization.

During testing, we combined the proposed 3D fractal features with plaque volume (TPV), co-occurrence matrix texture features, a total of 7-dimensional features, and input them into the SVM classifier for classification. Due to the limitation of the number of cases, we use the "leave one out method" for testing, which can ensure the independence of training samples and test samples, making the classification effect more robust. Specifically, one of the 38 patch data is first selected as a test sample, and the remaining 37 patch data are used as training samples to obtain classification results, and then the same test is performed on each sample in turn, repeating 38 times in total.

Calculate the precision, accuracy, sensitivity, and specificity of the classification results, and draw the ROC curve.

Table II shows the final classification results. It can be seen that the precision (94.7%), accuracy (92.1%), sensitivity (90.0%) and specificity (94.4%) of the feature classification have good results.

Table 2 SVM classification results

Figure 4 shows the distribution relationship between the scoring results of the SVM classifier and the variation of fractal features. It can be seen that the distribution of fractal dimension changes within three months of the cases taking atorvastatin and placebo is significantly different, and is consistent with the results of the SVM classifier.

Figure 5 shows the ROC curve of the SVM classification results, and it can be seen that the area of the ROC curve is 0.90278. It shows that the classifier designed with the extracted features has high sensitivity and specificity, and has good robustness.

Those skilled in the art can easily understand that the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, All should be included within the protection scope of the present invention.

Claims (5)

1. A three-dimensional blood vessel plaque ultrasonic image feature description method, is characterized in that, comprises the following steps: (1) Obtaining a three-dimensional blood vessel ultrasound image; (2) segment the plaque outline from the three-dimensional vascular ultrasound image obtained in step (1); (3) Obtain the ultrasound image of the plaque area from the plaque outline obtained in step (2), and then obtain the three-dimensional surface roughness of the plaque from the ultrasound image of the plaque area according to the plaque outline and using the three-dimensional box counting method Fractal eigenvalues; according to the change of the gray value inside the plaque and using the three-dimensional blanket covering method to obtain the fractal eigenvalues representing the three-dimensional internal texture changes of the plaque; The three-dimensional box counting method is as follows: by using cube boxes with different side lengths to cover the ultrasound image of the plaque area, a regression curve is obtained between the number of boxes and the logarithm of the side length of the box, and the slope of the curve is the three-dimensional surface representing the plaque. Roughness fractal eigenvalues; The three-dimensional blanket covering method is as follows: the gray value of the ultrasound image voxel in the plaque area is set as a gray-scale fractal hypersurface in four-dimensional space, and then there are two hypersurfaces at a position distance from the fractal hypersurface ε, Calculate the volume of the region between the two hypersurfaces; by changing the value of ε from 1 to the maximum value, the maximum value of ε does not exceed the minimum side length and minimum gray value of the three-dimensional image, and obtain the logarithm of the volume value For the regression curve that changes logarithmically with the distance ε, the fractal dimension value of the grayscale fractal hypersurface is linearly related to the slope of the regression curve; the fractal dimension value describes the change of the grayscale fractal hypersurface, which is the three-dimensional Fractal eigenvalues of internal texture variations. 2. the three-dimensional vascular plaque ultrasound image feature description method as claimed in claim 1, is characterized in that, the three-dimensional box counting method in the described step (3) obtains the fractal eigenvalue of the three-dimensional surface roughness representing plaque, Proceed as follows: (1) Cover the ultrasound image of the plaque area with a cube box whose side length is l, where l=1; (2) Define the voxel box that only includes the inside of the plaque as BLACK; the voxel box that only includes the outside tissue of the plaque is defined as WHITE; the voxel box that includes both the inside of the plaque and the outside tissue is defined as GREY, GRAY Indicates that the patch outline passes through the box; (3) Change the voxel box side length l, make l=2,3,4, ... M, M is the width of the ultrasound image of the plaque area; repeat step (2), until l is equal to the ultrasound image of the plaque area the width; (4) Statistically count the number of voxel boxes defined as BLACK, WHITE and GREY when l takes different values; (5) Record the number of voxel boxes counted in step (4) as N(l), where N(l) is the count of one or several voxel boxes of BLACK, WHITE, and GRAY, using the least squares method Fit the regression curve of log(N(l))/log(l); (6) The slope of the regression curve in the step (5) corresponds to the fractal characteristic value of the three-dimensional surface roughness of the plaque. 3. The three-dimensional vascular plaque ultrasound image feature description method according to claim 2, wherein in the step (5), the number of voxel boxes is the number of plaque internal tissue and border tissue in the three-dimensional blood vessel ultrasound image The total number of voxel boxes, the border tissue refers to the tissue that includes both the inside and outside of the plaque. 4. The three-dimensional vascular plaque ultrasonic image feature description method according to claim 1, wherein the three-dimensional blanket covering method obtains the fractal eigenvalues representing the three-dimensional internal texture changes of the plaque, and the steps are as follows: (1) Set the gray value of the ultrasound image voxel in the plaque area as a gray fractal hypersurface in the fourth dimension space independent of the three-dimensional voxel coordinates (x, y, z), then a voxel can represent is P(x,y,z,g(x,y,z)); (2) There are two upper and lower hypersurfaces at the position of the distance ε from the gray-scale fractal hypersurface, and the volume V(ε) of the ultrasound image in the plaque area is expressed as the hypervolume between the two hypersurfaces divided by 2ε, It is calculated from the following publicity: Wherein, the u _ε (i, j, k) and b _ε (i, j, k) respectively represent points with coordinates (i, j, k) on the upper and lower hypersurfaces of the fractal hypersurface ε The height value of is the gray value of the point in the ultrasound image of the plaque area; The value of u _ε (i,j,k) is the same as the maximum value of u _ε-1 (m,n,p) in the area where the distance point (i,j,k) is 1 and u _ε-1 (i ,j,k)+1 whichever is larger; The value of b _ε (i, j, k) is the minimum value of b _ε-1 (m, n, p) in the area where the distance point (i, j, k) is 1 and u _ε-1 (i ,j,k)+1 the smaller value of the two; (3) Change different ε values, and use the least squares method to fit the curve of log(V(ε))/log(ε); (4) Use the following formula to calculate the value of the three-dimensional fractal dimension of the ultrasound image: in, The D _E represents the dimension of the object in Euclidean space, and the D _E of the three-dimensional ultrasonic image is equal to 3, namely: FD=3-p; (5) The value of the three-dimensional fractal dimension calculated by the formula in step (4) is the fractal dimension characteristic value representing the three-dimensional internal texture change of the plaque. 5. The three-dimensional vascular plaque ultrasonic image feature description method according to claim 1, wherein the plaque is carotid plaque, aortic plaque, coronary plaque or peripheral artery plaque.