CN102800089B - Main carotid artery blood vessel extraction and thickness measuring method based on neck ultrasound images - Google Patents

Abstract

The invention discloses a method for extracting and measuring the thickness of the main carotid artery based on the ultrasound image of the neck, specifically: reading the ultrasound three-dimensional volume data of the neck, and marking the central axis of the main carotid artery based on the bifurcation point of the main carotid artery; According to the three-view direction, the three-dimensional volume data of the internal ultrasound is sequentially projected and segmented to obtain two-dimensional cross-sectional, coronal and sagittal plane sequence images; preprocessing, segmentation, and Reconstruction or intima-media thickness statistics to obtain the final cervical ultrasound aortocarotid artery wall inner and outer contours and vessel wall thickness and other related information. The present invention overcomes the disadvantages of the existing computer-aided diagnosis blood vessel segmentation method, such as the large calculation complexity, the inability to accurately measure the thickness of the blood vessel wall, and the subjective factors that easily cause errors, etc., and can completely, quickly and accurately obtain the inside and outside of the main carotid artery wall by neck ultrasound. Contour and vessel wall thickness. Compared with the manual segmentation method, the invention has quick operation and can be used for auxiliary diagnosis and prevention of cervical atherosclerosis and cardiovascular diseases.

Description

Extraction and Thickness Measurement of Aorta Carotid Arteries Based on Ultrasound Images of the Neck

technical field

The invention belongs to the cross field of biomedical engineering and image processing, and in particular relates to a method for extracting and measuring the thickness of a main carotid artery based on ultrasound images of the neck.

Background technique

According to the latest statistics from the World Health Organization, cardiovascular disease (Cardiovascular Diseases, CVDs) is one of the three leading causes of death in the world. Arterial wall thickening, plaque formation, and narrowing of blood vessels is one of the typical symptoms of atherosclerosis. Carotid atherosclerosis is a cardiovascular and cerebrovascular disease that can cause heart disease and stroke, and seriously endangers human health. Therefore, its early prevention, diagnosis, treatment and monitoring are of great significance.

On January 20, 2010, the American Heart Association (AHA) Strategic Planning Working Committee released the ten-year health strategic goals, that is, the AHA 2020 Health Strategy - "Defining and Developing National Targets for Promoting Cardiovascular Health and Reducing Diseases" ", for the first time put forward "to improve the health level as the main goal", which marks that the American Heart Association (AHA) has further moved forward the prevention front of cardiovascular disease CVDs, not only for high-risk groups and diseased groups with risk factors, but also to improve the common The health level of the population. Studies have shown that the key to the prevention and control of cardiovascular disease CVDs is early detection and early treatment. Therefore, the early detection and prevention of atherosclerosis has extremely important clinical significance for reducing the mortality rate of cardiovascular diseases and CVDs.

The thickness of the vessel wall of the common carotid artery (CCA) can be used as an important index to measure the lesion. The unique advantages of "real-time, economical, reliable, and safe" ultrasound imaging technology make intima-media thickness (IMT) based on this technology one of the commonly used indicators for evaluating the degree of carotid atherosclerosis. Ultrasound image vessel extraction and thickness measurement has become a research hotspot in recent years.

First of all, as far as blood vessel extraction is concerned, two patents with Chinese patent application numbers of 200910106119.6 and 201010297322.9 propose a method for blood vessel extraction from ultrasound images. The former is aimed at non-sequential single 2D ultrasound blood vessel grayscale images, and the latter is applied to intravascular ultrasound sequence images , both of which lack final detailed vascular information. For example, segmentation and method evaluation were not done in a targeted manner for sequence images, the closest real blood vessel thickness value could not be obtained according to the internal anatomy of the aorta carotid artery CCA, and it was difficult to apply industrially in the upgrading of domestic ultrasound machine software.

Secondly, in terms of thickness measurement, CULEX (Completely User-independent Layer Extraction) and CALEX (Completely Automatic Layer Extraction) are currently the most innovative and intelligent thickness measurement methods. Both can achieve fully automatic thickness measurement, but the implementation is difficult and the calculation complexity is high. CULEX and CALEX use different image features and adopt completely different ideas - the former uses the local statistics of image pixels to distinguish blood vessel lumen pixels and tissue pixels, and the segmentation method is based on the combination of gradient and activity models; while the latter integrates Feature extraction, linear fitting and classification all in one. Comparing the segmentation results of the two methods, it can be seen that CALEX is incomplete in the segmentation of the vascular lumen-intima LI contour, but better than CULEX in the segmentation of the media-adventitia MA contour; at the same time, CULEX is affected by image noise and image artifacts Larger, and the execution efficiency of CALEX is much higher than that of CULEX.

Contents of the invention

The purpose of the present invention is to provide an all-round, multi-angle, fast, accurate, easy-to-implement and easy-to-operate aortocarotid artery CCA blood vessel extraction and thickness measurement method.

The method for extracting and measuring the thickness of the main carotid artery based on the ultrasonic image of the neck comprises the following steps:

Read the neck ultrasound three-dimensional volume data, and mark the central axis of the main carotid artery based on the bifurcation point of the main carotid artery;

According to the central axis, according to the projection of the three-view direction, segment the cervical ultrasound three-dimensional volume data, and obtain two-dimensional cross-sectional, sagittal and coronal sequential images;

In each image of the two-dimensional cross-sectional sequence images, the region of interest of each aorta carotid artery is selected respectively, and the region of interest of each aorta carotid artery is preprocessed; In the region, the inner and outer contours of the main carotid vessels were obtained by segmentation respectively; according to the inner and outer contours of the regions of interest of the main carotid vessels in each two-dimensional cross-sectional image, the three-dimensional reconstruction was performed according to their spatial position relationship, and the three-dimensional main carotid artery was obtained. arterial vessel outline;

In each image of the two-dimensional sagittal plane sequence images, the regions of interest of the aorta carotid vessels are selected respectively, and the regions of interest of the aorta carotid vessels are preprocessed; In the region of interest, the inner and outer contours of each main carotid artery were segmented separately; the thickness between the inner and outer contours of each main carotid artery was calculated, that is, the intima-media thickness; the inner and outer contours of each two-dimensional sagittal image were counted. Median thickness mean;

In each image of the two-dimensional coronal sequence images, select the region of interest of each aorta carotid artery respectively, and preprocess each region of interest of the aorta carotid artery; In the region, the inner and outer contours of each main carotid artery were obtained by segmentation respectively; the thickness between the inner and outer contours of each main carotid blood vessel was calculated, that is, the intima-media thickness; the inner and outer contours of each two-dimensional coronal image were statistically calculated. Average film thickness;

Further, in the region of interest of the main carotid artery in each image of the two-dimensional cross-sectional sequence images, the inner and outer contours of each main carotid blood vessel are obtained by segmentation as follows:

A1. Extract the inner contour of the main carotid artery from the region of interest of the main carotid artery after the preprocessing of the two-dimensional cross-sectional image, and perform smoothing processing;

A2. Carry out ellipse fitting according to the inner contour of the main carotid artery vessel, and enlarge the fitted ellipse as the initial outer contour of the main carotid artery vessel;

A3. According to the initial outer contour of the main carotid artery, evolve to obtain the final outer contour of the main carotid artery.

The step A1 specifically includes: extracting the center position and radius of the aorta carotid vessel in the region of interest of the aorta carotid vessel after the preprocessing of the two-dimensional cross-sectional image, and extracting the inner contour according to the center position and radius.

The ellipse enlargement ratio coefficient in the step A2 is 1.02-1.08.

The evolution method adopted in the step A3 is any one of Active Contour Model, GVF-Snake, FastMarching Method, Level Set, and Sparse Field Algorithm.

Further, in the region of interest of the main carotid artery in each image of the two-dimensional sagittal plane sequence images, the inner and outer contours of each main carotid blood vessel are obtained by segmentation as follows:

B1. For each row of points in the region of interest of the main carotid artery after the preprocessing of the two-dimensional sagittal plane image, the serial number is sequentially marked from top to bottom as the index value of each row of points;

B2. Coarse segmentation of the region of interest:

B21. Obtaining the lower boundary contour DB: from left to right, sequentially scan each column of the main carotid artery region of interest in the two-dimensional sagittal plane image, and obtain the maximum gray value, minimum gray value and threshold condition of each column The lower boundary candidate points of the lower boundary candidate points; among the lower boundary candidate points, the point with the largest index value is taken as the only lower boundary contour point; all the lower boundary contour points are connected in sequence to form the lower boundary contour DB; the gray value of the lower boundary candidate point GV _candi should satisfy The gray threshold condition of GV _candidate ≥ a×ΔGV+GV _min = a×(GV _max -GV _min )+GV _min , the value range of weight coefficient a is 0.87~0.93;

B22. Obtain the upper boundary contour UB: move the lower boundary contour DB in the region of interest of the main carotid artery in the two-dimensional sagittal image in parallel upwards by 20-30 pixels as the upper boundary temporary contour UB_t; select the size as X×X The template, X takes 8~15 pixels, use the template to traverse the area between the upper boundary temporary contour UB_t and the lower boundary contour DB from left to right and from top to bottom, and calculate the pixel mean value EX and variance DX of the template coverage area ; If the mean variance threshold condition EX≤b and DX≤c is satisfied, the value range of b is 0.05~0.09, and the value range of c is 0.11~0.15, then the center of the template coverage area is the upper boundary candidate point; if the mean variance is not satisfied Threshold condition, the corresponding UB_t contour point is the upper boundary candidate point; among the upper boundary candidate points in each column, the one with the largest index value is used as the only upper boundary contour point; sequentially connect all the upper boundary contour points to form the upper boundary contour UB;

B3. Subdivision of the region of interest: Subdividing the area between the lower boundary contour DB and the upper boundary contour UB to segment the vessel lumen, intima and media, adventitia and tissues other than the adventitia;

B4. Extracting the interface between the lumen and the intima as the inner contour, and extracting the interface between the media and the adventitia as the outer contour.

Any one of C-means, fuzzy C-means, support vector machine SVM, and AdaBoost algorithm is used in the fine segmentation step B3 of the region of interest.

Further, in the region of interest of the main carotid artery in each image of the two-dimensional coronal sequence images, the inner and outer contours of each main carotid blood vessel are obtained by segmentation as follows:

C1. Using the mathematical morphology method to initialize the inner contour in the region of interest of the main carotid artery after the preprocessing of the two-dimensional coronal image, and generate the final inner contour according to the evolution of the initialized inner contour;

C2. The initial outer contour is obtained by moving down the initial inner contour, and the final outer contour is generated according to the evolution of the initial outer contour;

The evolution method used in the steps C1 and C2 is any one of the classic snake algorithm, GVF-Snake algorithm, level set, MS model, and CV model evolution method.

Beneficial effects of the present invention:

Compared with traditional blood vessel extraction and thickness measurement methods, the aorotid CCA blood vessel extraction and thickness measurement method based on neck ultrasound images provided by the present invention has four differences:

(1) Different from the traditional method of blood vessel extraction and thickness measurement on a certain plane, the present invention divides the neck ultrasound three-dimensional volume data according to the three-view direction projection, and obtains two-dimensional cross-section, sagittal plane and coronal plane respectively sequence images, and process them separately on each projection surface; unlike the traditional method of blood vessel extraction and thickness measurement on a single frame image of a certain surface, the present invention processes the sequence images of three surfaces in a targeted manner After blood vessel segmentation, extraction, measurement and calculation, the original complete three-dimensional data information can be fully utilized to obtain a full range of blood vessel parameters, such as: blood vessel thickness, area, volume, and plaque size, quantity, etc.;

(2) In the region of interest of the main carotid artery in each image of the two-dimensional cross-sectional sequence image, when the inner contour of each main carotid vessel is obtained by segmentation, Hough transform circle detection is introduced to obtain the center position and radius of the vessel circle parameters, so as to guide the size change of structural primitives in the subsequent mathematical morphology and optimize the inner contour segmentation results; in the region of interest of the aorta carotid artery in each image of the two-dimensional cross-sectional sequence images, each main carotid artery is evolved to obtain Before the outer contour of the carotid artery, the ellipse fitting strategy is introduced as prior knowledge, and the fitted ellipse is used as the initial outer contour, which is more in line with the physiological shape of the blood vessel; in each image of the two-dimensional cross-sectional sequence images In the area of interest of the main carotid artery, when the outer contour of each main carotid artery is evolved, the active contour model (ACM) method is introduced to solve the problem of weak boundaries and difficult distinctions of the outer contour of the blood vessels;

(3) In the region of interest of the main carotid artery in each image of the two-dimensional sagittal plane sequence image, when the inner and outer contours of each main carotid artery are obtained by segmentation, the "coarse-thin" two-layer segmentation is used structure; in the "coarse segmentation", the gray threshold condition and the mean variance threshold condition are used to constrain; in the "fine segmentation", the fuzzy C-means FCM clustering algorithm is first introduced; secondly, the vascular tissue is divided into multiple categories; finally Then merge them into three categories - "vascular lumen", "intima-media" and "adventitia and outer tissue" to obtain the outline of each interface to improve the accuracy of sagittal plane intima-media thickness measurement;

(4) In the region of interest of the main carotid artery in each image of the two-dimensional coronal sequence image, when the inner and outer contours of the main carotid blood vessels are obtained by segmentation, the initial inner and outer contours of the main carotid blood vessels are obtained by using the mathematical morphology method. Outer contour; then the final inner and outer contours are obtained through evolution; in the intima-media segmentation of the two-dimensional coronal image, on the basis of the Snake algorithm, the GVF-Snake algorithm is introduced again to solve the problem of deep depression, so as to improve the intima-media segmentation of the coronal image. Accuracy of media thickness measurements.

To sum up, the method for extracting and measuring the thickness of the aorta carotid artery CCA based on ultrasound images of the neck provided by the present invention can effectively deal with the diversity of ultrasound images, and achieve the purpose of accurately and quickly segmenting blood vessels and measuring thickness; secondly, through Quantitative analysis and comparison, this method is equivalent to the error of manual segmentation and measurement methods, and its application can directly reduce the workload of manual labeling of a large number of images of medical workers; finally, based on the method of the present invention, the clinical parameters (such as: blood vessel thickness, area , volume and plaque size, number, etc.), not only can intuitively and quantitatively reflect vascular lesions, but also provide more guiding significance for the early prevention and treatment of carotid atherosclerosis.

Description of drawings

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

Fig. 2 is the flowchart of two-dimensional sequential image processing among the present invention;

Fig. 3 is the flowchart of concrete steps of the present invention;

Fig. 4 is a schematic diagram of the bifurcation point and the central axis direction of the marked three-dimensional volume data 3D_Data;

Figure 5 is a schematic diagram of division of the main carotid artery according to the three-view projection of the central axis;

Figure 6 is a two-dimensional cross-sectional sequence image (2D_Data_Z);

Figure 7 is the 13th original image (2D_Data_Z_k, k=13) in the two-dimensional cross-sectional sequence image;

Figure 8 is a two-dimensional sagittal sequence image (2D_Data_Y);

Figure 9 is the second original image (2D_Data_Y, j=2) in the two-dimensional sagittal sequence image;

Figure 10 is a two-dimensional coronal sequence image (2D_Data_X);

Figure 11 is the first original image (2D_Data_X_i, i=1) in the two-dimensional coronal sequence image (2D_Data_X);

Fig. 12 is a flowchart of image segmentation of two-dimensional cross-sectional sequence;

Fig. 13 is the ROI diagram of the region of interest of the two-dimensional cross-sectional image;

Figure 14 is the step-by-step result diagram of ROI preprocessing in the region of interest: Figure 14-(a) is the segmented linear stretching result diagram; Figure 14-(b) is the SRAD nonlinear filtering result diagram;

Fig. 15 is the result figure of Hough transform circle detection to the preprocessing region of interest;

Fig. 16 is the result figure of Canny operator edge detection to preprocessing area of interest;

Fig. 17 is a graph of mathematical morphology processing results;

Fig. 18 is the extracted single-pixel inner contour result map;

Fig. 19 is an inner contour expanded point set diagram;

Fig. 20 is a fitting curve diagram of inner contour expansion point set;

Fig. 21 is a point set diagram of inner contour recovery;

Fig. 22 is a diagram of the segmentation result of the inner contour of the two-dimensional cross-section: wherein the solid circle point represents the manual segmentation contour; the solid line represents the segmentation contour of the present invention;

Fig. 23 is a figure of inner contour ellipse fitting result;

Fig. 24 is a diagram of the segmentation result of the outer contour of the two-dimensional cross-section: wherein the solid diamond point represents the manual segmentation contour; the solid square point represents the segmentation contour of the present invention;

Figure 25 is the result of two-dimensional sagittal plane preprocessing: Figure 25-(a) is the result of normalization; Figure 25-(b) is the result of filtering;

Figure 26 is a two-dimensional sagittal region of interest ROI preprocessing results;

Figure 27 is the result of rough segmentation of ROI in the two-dimensional sagittal region of interest: Figure 27-(a) is the original size; Figure 27-(b) is a double-magnified display; Figure 27-(c) is double-magnified display graph;

Figure 28 is the result of fine segmentation of ROI in the two-dimensional sagittal region of interest: Figure 28-(a) is the original size; Figure 28-(b) is the double-magnified display image; Figure 28-(c) is the double-magnified display display graph;

Figure 29 is the ROI result map of the two-dimensional coronal surface extraction region of interest: Figure 29-(a) is the original image; Figure 29-(b) is the corresponding ROI map;

Fig. 30 is the ROI preprocessing result figure of the region of interest of the two-dimensional coronal plane;

Fig. 31 is a graph of the thresholding result of the region of interest;

FIG. 32 is a result diagram of filling holes in the region of interest;

Fig. 33 is the deburring result diagram of the intima surface of the region of interest;

Figure 34 is a diagram of the result of intima correction in the region of interest;

Figure 35 is the results of the initial contour (dotted line) and final contour (solid line) of the intima: Figure 35-(a) is the original size; Figure 35-(b) is the double-magnified display; Figure 35-(c) Show graph for double magnification;

Figure 36 is the results of the initial contour (dotted line) and final contour (solid line) of the adventitia: Figure 36-(a) is the original size; Figure 36-(b) is the double-magnified display; Figure 36-(c) Show graph for double magnification;

Figure 37 is the initial contour (dotted line) and final contour (solid line) results of the intima (upper) and adventitia (lower): Figure 37-(a) is the original size; Figure 37-(b) is doubled Display diagram; Figure 37-(c) is a double-magnified display diagram;

Fig. 38 is a diagram of the gold standard results of manual measurement of intima-media thickness on the two-dimensional coronal plane.

Detailed ways

In the following, the present invention will be further described in detail in conjunction with specific implementation examples and descriptions of the drawings, and the solution verification of the present invention will be given.

A method for extracting and measuring the thickness of the main carotid artery CCA vessel based on ultrasound images of the neck, including the following five steps and scheme verification, as shown in Figure 1; specifically, the flow chart for each two-dimensional sequence image processing, as shown in FIG. As shown in Figure 2; the overall flow chart is shown in Figure 3.

Step (1) Read the neck ultrasound three-dimensional volume data 3D_Data, mark the main carotid bifurcation point "BF" and the central axis, as shown in Figure 4.

(1.1) Locate the bifurcation point "BF"

In the three-dimensional ultrasound volume data 3D_Data, identify the main carotid artery CCA, internal carotid artery ICA and external carotid artery ECA, and determine the position of the carotid artery bifurcation point "BF" (see Figure 4).

(1.2) Locate the central axis of the aortocarotid CCA

Referring to the position of the bifurcation point "BF", according to the physiological and anatomical structure of the aortacarotid artery CCA, the position of the central axis is preliminarily estimated, and two points are arbitrarily selected on the estimated central axis - "point 1" and "point 2" (see Figure 4 ), the line connecting "point 1" and "point 2" is used as the central axis of the aorta carotid artery CCA.

Step (2) According to the central axis, project according to the three-view direction, as shown in Figure 5, segment the cervical ultrasound three-dimensional volume data 3D_Data, and obtain the two-dimensional cross-sectional sequence image (2D_Data_Z) and the two-dimensional sagittal plane sequence image ( 2D_Data_Y) and two-dimensional coronal sequence images (2D_Data_X);

(2.1) Two-dimensional cross-sectional sequence image (2D_Data_Z)

Segmentation was performed perpendicular to the central axis of the carotid artery to obtain two-dimensional cross-sectional images. Generally, there are at most 13-15 two-dimensional cross-sectional images, as shown in FIG. 6 ; in the present invention, the 13th image is taken as an example, as shown in FIG. 7 .

(2.2) Two-dimensional sagittal sequence image (2D_Data_Y)

According to the direction parallel to the central axis of the carotid artery and segmented from the side projection, two-dimensional sagittal plane sequence images were obtained. Generally, there are at most 4 to 6 images in the two-dimensional sagittal plane sequence, as shown in FIG. 8 ; in the present invention, the second image is taken as an example, as shown in FIG. 9 .

(2.3) Two-dimensional coronal sequence image (2D_Data_X)

According to the direction parallel to the central axis of the carotid artery, it is segmented from the frontal projection to obtain two-dimensional coronal sequence images. Generally, there are at most three sequential images of the two-dimensional coronal plane, as shown in FIG. 10 ; in the present invention, the first one is taken as an example, as shown in FIG. 11 .

Step (3) extracting the three-dimensional aortic carotid vessel outline based on the two-dimensional cross-sectional sequence images. In this step, the inner and outer blood vessel contours of each two-dimensional cross-sectional sequence image are segmented, and the flow chart of the specific segmentation steps is shown in FIG. 12 .

In this step, select one of the two-dimensional cross-sectional images (2D_Data_Z) as an example to illustrate this step (see Figure 7). The internal and external contours of blood vessels are used to calculate parameters such as area and volume, provide clinical diagnosis and treatment indicators, and qualitatively measure clinical conditions. In view of the overall poor quality of ultrasound sequence images, it needs to be improved through preprocessing; the key to blood vessel extraction is the segmentation of inner and outer contours. In the present invention, combined with the shape of blood vessels, circles and ellipses are introduced as shape prior information to more effectively segment blood vessels. ; In order to better present the segmentation results, finally perform targeted post-processing on the inner contour.

(3.1) In the two-dimensional cross-sectional sequence images (2D_Data_Z), select the region of interest (2D_Data_Z_k_ROI) of each aortocarotid artery respectively, and preprocess the region of interest of each aortocarotid artery. The technical idea of preprocessing is: change the gray level first, and then filter and reduce noise.

The original two-dimensional cross-sectional image has dark gray scale, low contrast, and high noise, so it needs to be preprocessed. The present invention is to extract blood vessels, so it is only necessary to preprocess the sub-region containing blood vessels in the two-dimensional cross-sectional image as the region of interest ROI. The ROI of the region of interest in Figure 7 is shown in Figure 13, and the results of each preprocessing step As shown in Figure 14, the preprocessing process is as follows:

(3.1.1) Gray scale transformation (Figure 14-a)

Grayscale transformation is to improve the dynamic range of grayscale during image processing, so as to improve the brightness and contrast of the image. Methods such as linear stretching, nonlinear stretching, and image enhancement can be used. This example uses the simplest piecewise linear transformation function for illustration, which consists of two basic operations:

(3.1.1.1) Determine the two inflection points for grayscale stretching of the image;

The histogram statistics are made on the two-dimensional cross-sectional image to obtain an L grayscale image. The gray value before and after the gray scale transformation process is defined by r and s respectively, and the gray scale transformation function of the L gray level image is expressed as s=T(r). Assume that P1 and P2 are two inflection points of the piecewise linear transformation function T(r), and the gray values before and after transformation are (r ₁ , s ₁ ) and (r ₂ , s ₂ ) respectively.

(3.1.1.2) Gray scale transformation

In this example, the three-dimensional volume data of 10 anonymous patients were counted, and a total of 300 two-dimensional cross-sectional images were obtained. After statistical analysis, the gray scale range of the original image concerned is set to [r _min , r _max ], then P1, The two inflection points of P2 divide the transformation function T(r) into three sections: r _min ≤r<r ₁ , r ₁ ≤r≤r ₂ , r ₂ <r≤r _max . The expression for T(r) is as follows:

$\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ {\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ {\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>\frac{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}\end{array}\right.$

In this way, the gray level is linearly stretched from the original concerned range to the saturation range [0, L-1], where L is the transformed gray level. The grayscale transformation result of Figure 13 is shown in Figure 14-a.

(3.1.2) Filtering and noise reduction (Figure 14-b)

Non-local mean value, Gaussian filter, speckle noise anisotropic diffusion model SRAD and other methods can be used for filtering noise reduction. Among them, the SRAD nonlinear filter has the advantages of enhancing contrast and preserving details due to the use of different diffusion coefficients in different diffusion directions. The SRAD filter is proposed based on local statistical filters and anisotropic smoothing. Compared with other traditional local statistical filters and anisotropic smoothing filters, it has better homogeneous area smoothing performance and better retention Advantages of detail and border enhancement. The filtering and denoising results of Figure 14-a are shown in Figure 14-b.

(3.2) In the region of interest of each main carotid artery after preprocessing, segment the inner and outer contours of each main carotid artery.

The technical idea of extracting the inner and outer contours of the main carotid artery is as follows: first, the inner contour is extracted; second, the initial outer contour is obtained by fitting; finally, the final outer contour of the main carotid artery is obtained by evolution.

(3.2.1) In the region of interest of each main carotid artery after preprocessing, the inner contour of each main carotid vessel was extracted and smoothed.

The technical idea of extracting the inner contour of the aorta carotid artery is as follows: firstly extract the blood vessel parameter information, secondly, extract the single pixel inner contour of the blood vessel according to the blood vessel parameter information; finally, smooth the inner contour.

(3.2.1.1) Extract blood vessel parameter information from each preprocessed region of interest in the aorta carotid artery.

This step is to extract the parameter information of the blood vessel, that is, the position and radius of the center of the circle. In this example, the Hough transform circle detection method is used to illustrate (the method is not limited). O(centre, R _max , R _min ) can be obtained in Hough transform circle detection, which are the coordinates of the center of the circle, the maximum radius, and the minimum radius, so as to provide a reference for the size of the primitive in the subsequent steps. The essence of Hough transform is to cluster the pixels with a certain relationship in the image space, establish a parameter space that can connect these pixels with a certain analytical form, and find the cumulative corresponding points. In the Hough transform detection circle, the boundary point obtained by the Canny operator is the pixel, and the three parameters of the circle center coordinates and the circle radius are the parameters of its corresponding analytical form. In addition, in this example, in order to reduce the running time of the program, a certain circle radius range and the center position of the circle are artificially selected. Figure 15 shows the results of transformed circle detection on Figure 14-b.

(3.2.1.2) According to the blood vessel parameter information, extract the inner contour of the single pixel point of the blood vessel

Based on the mathematical morphology method, active contour model, level set method or Fast Marching, the inner contour of the aorta carotid artery in the region of interest after preprocessing in the two-dimensional cross-sectional image can be segmented. This example uses the mathematical morphology method to illustrate.

(3.2.1.2.1) Extract the edge in the region of interest. In this example, the Canny operator edge is used. The edge extraction result in Figure 13-b is shown in Figure 16;

(3.2.1.2.2) Set the size and shape of structural primitives

in In the size of the structural element SE_si, R _max R _m is the maximum radius and minimum radius of the circle calculated in (3.2.1.1) respectively (see Figure 15); and the size of the structural element SE_size and the shape SE shape are adjustable. For example: commonly used sizes are 3, 5, 7, etc.; commonly used shapes are circle, cross, square, etc. In the present invention, the shape SE_shape of the structural primitive adopts an octagonal disk, and its size is an integer multiple of 3.

(3.2.1.2.3) Close operation of mathematical morphology

Apply the structural primitives to Figure 16 to perform mathematical morphology closing operations to obtain the local area where the closed inner contour is located. Figure 17 shows the results after the closing operation in Figure 16 .

(3.2.1.2.4) Extract single-pixel intravascular contour

The outline of a blood vessel can be described by a series of outline points. Therefore, after morphologically segmenting the inner contour of a blood vessel, a single pixel point set is extracted for the local area, which can be used as the inner contour of the two-dimensional cross-sectional blood vessel. The single-pixel contour points in Fig. 17 are shown in Fig. 18 .

(3.2.1.3) Smoothing of intravascular contours (Fig. 19, Fig. 20 and Fig. 21)

The smoothing of the blood vessel contour includes the smoothing of the inner and outer contours. In the present invention, the "bread roll" expansion is adopted, and after smoothing and polynomial fitting, a new single-pixel contour point set is obtained by restoration. The specific process is as follows:

(3.2.1.3.1) Calculate the coordinates of the center point of the contour

Calculate the average value of the horizontal and vertical coordinates of all points in the contour point set, and record this point as the center point (x ₀ ,y ₀ );

(3.2.1.3.2) Sorting contour points

Choose a contour point, and use this point as the starting point, sort the contour counterclockwise;

(3.2.1.3.3) Calculate the relative distance

Sequentially calculate the relative distance Dis between the center point (x ₀ , y ₀ ) and the sorted contour point ( _xi , y _i );

${\mathrm{<span\; class="notranslate">\; dis</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

(3.2.1.3.4) Expand the inner contour point set (Figure 19)

Take the sorting number as the abscissa and each corresponding relative distance as the ordinate to obtain the original contour expansion point set;

(3.2.1.3.5) Polynomial fitting (Figure 20)

Curve polynomial fitting is performed on the above point set to remove burrs and make the contour expansion fitting curve smooth. Wherein, the polynomial order n ₀ is adjustable, generally 5~10, preferably, n ₀ =9;

(3.2.1.3.6) Restoring the contour point set (Figure 21)

The smoothed curve is used as a new contour to expand the curve, and according to the sequence number, the coordinates of the new contour point (x _j , y _j ) are inversely calculated, that is, the contour point set is restored.

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; dis</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; angle</span>}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; dis</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; sin</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; angle</span>}<span\; class="notranslate">\; )</span>\end{array}\right.$

Where: angle is the inclination angle of the line connecting the original contour center point (x ₀ , y ₀ ) and the original contour point ( _xi , y _i ). Finally, the result of the inner contour is shown in Figure 22, where the solid circle points represent the manually segmented contour; the solid line represents the segmented contour of the present invention.

(3.2.2) According to the inner contour of each main carotid vessel, ellipse fitting is performed on it, and each fitted ellipse is enlarged, and used as the initial outer contour of each main carotid vessel;

(3.2.2.1) Carry out ellipse fitting according to the inner contour of the blood vessel;

In the present invention, the inner contour single-pixel point set determined in step (3.2.1.2.4) is used as a given point set, and the ellipse is a fitting function, and the parameters E (centre, p, q, θ) of the ellipse are obtained by fitting: respectively represent the ellipse The coordinates of the center of the circle, the length of the semi-major axis of the ellipse, the length of the semi-minor axis of the ellipse, and the angle between the major axis of the ellipse and the horizontal line.

In the present invention, the result of fitting the inner contour ellipse is shown in FIG. 23 .

(3.2.2.2) The initial outer contour of the vessel

Based on the prior knowledge of the shape of the blood vessel, using the consistency and symmetry of the inner and outer contours of the blood vessel, the obtained inner contour of the blood vessel is fitted with an ellipse, and then enlarged as the initial contour of the outer contour of the blood vessel.

Enlarge (3.2.2.1) the fitted ellipse as the initial outline of the vessel. Preferably, the ellipse enlargement factor factor is between 1.02 and 1.08. The meaning of enlargement in the present invention is that the semi-major axis and the semi-minor axis of the ellipse are multiplied by the enlargement scale coefficient, and the center and inclination angle of the ellipse remain unchanged.

(3.2.3) According to the initial outer contour of each main carotid vessel, the active contour model ACM evolution is used to obtain the (final) outer contour of each main carotid vessel

Based on (3.2.2) the initial outer contour of the blood vessel, the active contour model (ACM) is used to evolve to obtain the final outer contour of the blood vessel, as shown in Figure 24, where the solid diamond points represent the manual segmentation contour; the solid square points represent the segmentation contour of the present invention.

Since the outer contour of the blood vessel in the present invention is obtained by the active contour model, and the inner contour is obtained by mathematical morphology, the outer contour has better smoothness than the inner contour, so only the inner contour can be smoothed according to the above (3.2.1.3) method deal with.

(3.3) According to the inner and outer contours of each region of interest, three-dimensional reconstruction is carried out according to their spatial position relationship to obtain the three-dimensional aortocarotid vessel contour. The vessel contour can assist clinical application.

After sequentially processing all the two-dimensional cross-sectional images according to step (3), the inner and outer contours of each blood vessel are obtained. The regions of interest marked with the aorta carotid artery are stacked in order according to the spatial position relationship, and the three-dimensional reconstruction is performed to obtain the three-dimensional aorta carotid artery; after the three-dimensional reconstruction, the parameters such as the area and volume of the aorta carotid artery CCA can be obtained, so as to provide doctors with Clinical diagnosis provides guidance, such as: drug evaluation; if a comparative study is introduced, it can be used for surgical effect evaluation and so on.

Step (4) Calculate the intima-media thickness in the sagittal plane based on the two-dimensional sagittal plane sequence images.

The main carotid CCA vessel wall, starting from the lumen (Lumen), is composed of three layers of membranes from inside to outside—intima, media, and adventitia. Ultrasound is the use of different acoustic impedance interfaces formed by "vascular lumen (L)-intima (I)-media (M)-adventitia (A)" to perform imaging to diagnose the disease.

In the present invention, the extracted intima contour (referred to as the inner contour) is located at the junction of the vascular lumen and the intima, which is recorded as LIB (Lumen Intima Boundary); the extracted adventitia contour (abbreviated as the outer contour) is located at the junction of the media and the adventitia The measured intima-media thickness (IMT) is the distance between the lumen-intima interface LIB and the media-adventitia interface MAB.

The main technical idea of this step is: extract the intima and adventitia of the two-dimensional sagittal plane sequence images, measure the intima-media thickness IMT, calculate the mean value of the sagittal plane intima-media thickness IMT, and provide blood vessel quantitative analysis measurement results. Take one of the two-dimensional sagittal images (see Figure 9) as this example, and the other images are processed in the same way, and finally the thickness measurement results are obtained;

(4.1) In the two-dimensional sagittal plane sequence image (2D_Data_Y), each aortic carotid artery region of interest (2D_Data_Y_j_ROI) was selected respectively, and each aortocarotid artery region of interest was preprocessed. The technical idea of preprocessing is: first normalize, then filter and denoise.

The selected two-dimensional sagittal sequence images include the aorta carotid artery CCA, bifurcation point BF, vessel lumen, distal vessel wall adventitia, and distal vessel wall adventitia tissue, from which extracts include the distal vessel wall adventitia tissue The sub-region of is used as the region of interest ROI, and it is preprocessed by normalization and filtering. The normalization function can adopt linear function conversion, logarithmic function conversion, inverse cotangent function conversion and so on. Filtering can use low-pass filtering, mean filtering, median filtering, frequency-domain filtering, band-pass filtering, and the like. The following is an example of linear function normalization and low-pass filtering, as shown in Figure 25-(a) and Figure 25-(b) respectively, and the ROI result after preprocessing is shown in Figure 26.

(4.1.1) Linear function normalization (Figure 25-(a))

Convert the gray value of the two-dimensional sagittal plane image to a linear function, so that it can be normalized in [0, 1]. The expression is as follows:

GV_aft=(GV_pre-MinValue)/(MaxValue-MinValue)

Among them: GV_pre, GV_aft are the gray value before and after conversion respectively, MaxValue, MinValue are the maximum gray value and the minimum gray value in the figure respectively.

(4.1.2) Low-pass filtering (Figure 25-(b))

Using Gaussian filtering, the specific operation is: use a template with a mean value of 0 and a standard deviation of 10, scan each pixel in the image, and calculate the weighted average of all pixels in the neighborhood where the template is located as the gray value of the pixel in the center of the template. Generally, the template size is 5, 7, or 9, and in the present invention, 5×5 is used. Finally, the preprocessed ROI results are shown in Figure 26.

(4.2) Segment the intima and adventitia of each aortic carotid vessel in the preprocessed regions of interest, and calculate the intima-media thickness of each region of interest;

According to the intima-media thickness defined in step (4), extract the intima and adventitia of the vessel in the region of interest of the preprocessed two-dimensional sagittal plane image, and then measure the intima-media of the two-dimensional sagittal plane image Thickness imt. The technical idea of blood vessel extraction is as follows: first, roughly segment the ROI of the region of interest, and locate the approximate position of the intima-media; secondly, finely segment the ROI of the region of interest, and obtain the intima and adventitia respectively; finally, calculate the thickness of the intima-media.

(4.2.1) Coarse segmentation of the region of interest ROI (Figure 27)

For each column of points on the image in the ROI area, mark the serial numbers 0, 1, 2, ... from top to bottom as the index value of each column of points. In the ROI search, in order to reduce redundant searches, it is necessary to further narrow the potential ROI area of the intima and media, so the region of interest ROI is roughly segmented to obtain the lower boundary contour DB and upper boundary contour UB. The rough segmentation result of the region of interest ROI is shown in Figure 27.

(4.2.1.1) Lower boundary contour DB acquisition

First, set the threshold condition that the gray value of the lower boundary candidate point GV _candidate should satisfy: GV _candidate ≥ a×ΔGV+GV _min = a×(GV _max -GV _min )+GV _min , (generally, the weight coefficient a The range is 0.87~0.93, preferably a=0.9); then, from left to right, scan each column of the ROI sequentially, and obtain the maximum gray value, the minimum gray value of each column, and the lower boundary candidates that meet the threshold condition point. Among the lower boundary candidate points, the point with the largest index value is used as the only lower boundary contour point; finally, all the lower boundary contour points in the ROI are sequentially connected to form the lower boundary contour DB.

(4.2.1.2) Acquisition of upper boundary contour UB;

First, move the lower boundary contour DB upward in ROI by Δ _index pixel distance in parallel, and Δ _index generally takes a value of 20 to 30 pixels as the upper boundary temporary contour UB_t, preferably Δ _index = 25; secondly, take the template size X×X, from left to right, from top to bottom, sequentially calculate the mean value EX and variance DX of the template where each pixel in the area between the upper boundary temporary contour UB_t and the lower boundary contour DB is located. If the threshold condition is satisfied: EX≤b and DX≤c, then the center of the template is the upper boundary candidate point; if the threshold condition is not satisfied, the corresponding UB_t contour point is the upper boundary candidate point. (Generally, the template size X is 8~15; b is 0.05~0.09; c is 0.11~0.15. Preferably, X=10; b=0.08; c=0.14.) Among the upper boundary candidate points of each column, Mark the one with the largest index value as the unique upper boundary contour point; finally, connect all the upper boundary contour points in the ROI in sequence to form the upper boundary contour UB.

(4.2.2) Subdivision of ROI in the region of interest (Figure 28)

The subdivision of the region of interest ROI is essentially completed between the lower boundary contour DB and the upper boundary contour UB. The image area obtained by the rough segmentation is the area where the intima and media are located. In order to correctly divide DB and UB into vascular lumen, intima, media, adventitia, and tissues other than adventitia, methods such as C-means, fuzzy C-means, support vector machine SVM, and AdaBoost algorithm can be used. In the present invention, the fuzzy C-means-based clustering method is used to illustrate the extraction of the inner and outer contours of the blood vessel wall. The fine segmentation results of the region of interest ROI are shown in Figure 28.

According to the characteristics of the distal vascular wall, the vascular part can be intuitively divided into three categories: "black, gray, and white", which can roughly correspond to "vascular lumen, intima-media, adventitia and other tissues". In the final classification result of the present invention, set a class of pixels with the lowest gray value as "vascular lumen"; a class of pixels with the highest gray value as "adventitia and other tissues"; pixels with other gray values The point is "intima-media".

Fuzzy division is adopted in the clustering division of the present invention, so that each given data point uses the degree of membership whose value is in the interval [0, 1] to determine the degree to which it belongs to each category. The sum of the membership degrees of a data set is always equal to 1. Let the data set X={x ₁ ,x ₂ ,...,x _n }, each sample has s characteristic attributes x _j ={x _j1 ,x _j2 ,...,x _js }(j=1, 2,...,n). In the present invention, the pixel gray value is used as the only characteristic attribute, ie s=1. For each column in the ROI, FCM clustering is used to achieve segmentation. Here we take a certain column as an example, and assume that the column has a total of n pixel samples. Specific steps are as follows:

(4.2.2.1) Parameter initialization: Given the number of clustering categories C (we take C=10), set the iteration stop threshold ε (we take ε=1e-5), and set the maximum number of iterations N (we take N= 100);

(4.2.2.2) Initialize the clustering prototype distribution matrix P ⁰ : randomly generate a matrix of size n×C, the jth row corresponds to the distribution of the membership value of the jth pixel sample in the column, that is, the jth pixel sample belongs to to all kinds of possibilities, and to satisfy

(4.2.2.3) Calculate or update the partition matrix U ^(b) with the following formula, for If d _ik ≠0 then:

${\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; ik</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; ik</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; rk</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ]</span>}^{\frac{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}$

If d _ik =0, then: And for j≠k,

In the formula, d _ik represents the distance between the pixel sample x _k in the i-th class and the cluster center of the i-th class, m is the weighting index, and b is the number of iterations.

(4.2.2.4) Use (Formula 1) to update the cluster distribution matrix P ^(b+1) :

$P_i^{(b+1)}=\frac{{\mathrm{\&Sigma;}}_{k=1}^{\mathrm{no}}{\left({\mathrm{\&mu;}}_{\mathrm{ik}}^{(b+1)}\right)}^m*x_k}{{\mathrm{\&Sigma;}}_{k=1}^{\mathrm{no}}{\left({\mathrm{\&mu;}}_{\mathrm{ik}}^{(b+1)}\right)}^m},i=\mathrm{1,2},...,C$ (Formula 1)

(4.2.2.5) If before and after a certain iteration, the clustering distribution matrix satisfies |P ^(b) -P ^(b+1) |<ε or b=N, then the algorithm stops and outputs the partition matrix U and the clustering distribution matrix P , otherwise let b=b+1, return to step (4.2.2.3). Finally, all the pixels of "intima-media" are obtained.

(4.2.3) Calculate the intima-media thickness of the two-dimensional sagittal ROI

After (4.2.2) clustering and segmentation, the corresponding index values of the intima and adventitia of each column of the ROI in the two-dimensional sagittal plane can be obtained, the difference is calculated, and the average value of each column is calculated, which is the two-dimensional vector The shape surface corresponds to the thickness of the intima-media.

(4.3) According to the intima-media thickness of each region of interest, the mean value is statistically calculated, which is the intima-media thickness in the sagittal plane

For each two-dimensional sagittal sequence image, after the above steps (4.1) and (4.2), perform statistical analysis of the results to obtain the mean value and variance of the sagittal vascular intima-media thickness IMT as parameters of the blood vessel evaluation index. IMT value has become a common and effective indicator for predicting cardiovascular and cerebrovascular diseases.

Step (5) Calculate the coronal intima-media thickness based on the two-dimensional coronal sequence images:

The definition of intima-media thickness in this step is the same as in step (4), that is, the measured intima-media thickness (IMT) is the difference between the lumen-intima interface LIB and the media-adventitia interface MAB. distance between.

The main technical idea of this step is: extract the intima and adventitia of the two-dimensional coronal sequence images, measure the intima-media thickness IMT, calculate the mean value of the coronal intima-media thickness IMT, and provide the blood vessel quantitative analysis measurement results. Take one of the two-dimensional coronal images (see Figure 11) as this example, and the other images are processed in the same way, and finally the thickness measurement results are obtained;

(5.1) In the two-dimensional coronal sequence images (2D_Data_X), each region of interest of the main carotid artery (2D_Data_X_i_ROI) was selected respectively, and each region of interest of the main carotid artery was preprocessed. The technical idea of preprocessing is: first enhance, then filter.

The selected two-dimensional coronal sequence images include the aorta carotid artery CCA, vessel lumen, distal vessel wall adventitia, and distal vessel wall adventitia tissue, from which the sub-region containing the distal vessel wall adventitia tissue is extracted as the subregion of interest Regional ROIs. Figure 11 extracts the region of interest ROI, and the result is shown in Figure 29.

Perform image enhancement and filter preprocessing on it. Image enhancement can use adaptive histogram equalization, spatial domain regulation, local statistical methods, etc.; filtering can use low-pass filtering, mean filtering, median filtering, etc. The following uses adaptive histogram equalization and adaptive mean filtering as examples for illustration. Obtain the preprocessed ROI, as shown in Figure 30.

(5.1.1) Image Enhancement

Image enhancement using Adaptive Histogram Equalization (AHE, Adaptive Histogram Equalization). Compared with the traditional histogram equalization, AHE pays more attention to the local features of the image. AHE determines the contrast enhancement method according to the local statistical characteristics of pixels. The gray value of each pixel is obtained through an equalization transformation function, and the transformation function is obtained from a local word image histogram centered on the pixel, which is called a local contrast enhancement method. The formula is:

x' _i,j =m _i,j +k×(x _i,j -m _i,j )

Among them, k is the adaptive reference quantity, and the expression is is the gray variance in the window W, is the gray variance of the entire image, k' is the proportional coefficient; x _{i, j} , x' _{i, j} are the gray values before and after transformation respectively; m _{i, j} is the average value of pixel gray in the window W.

(5.1.2) Adaptive Mean Filtering

Mean filtering is also called linear filtering, and the area averaging method is mainly used to remove noise. The basic principle of linear filtering is to replace each pixel value in the original image with the mean value, that is, the gray value x _{i, j} of the current pixel point to be processed, select a template, the template is composed of several pixels in its neighborhood, and find the value in the template The mean value of all pixel gray levels, and then assign the mean value to the current pixel point as the gray level of the processed image at this point, that is, x′ _{i, j} . Finally, the preprocessed ROI is obtained, as shown in FIG. 30 .

(5.2) Segment the intima and adventitia of each aortic carotid vessel in the preprocessed regions of interest, and calculate the intima-media thickness of each region of interest

According to the intima-media thickness defined in step (4), extract the intima and adventitia of the blood vessel in the region of interest of the preprocessed two-dimensional coronal image, and then measure the intima-media thickness IMT of the two-dimensional coronal plane image . The technical idea of its blood vessel extraction is: first initialize and extract the contour of the intima of the blood vessel; secondly initialize and extract the contour of the adventitia of the blood vessel; finally calculate the thickness of the intima and media.

(5.2.1) Initialize the inner contour

In view of the structural characteristics of the main carotid artery CCA, the two-dimensional coronal sequence images are extremely susceptible to the artifacts of the bifurcation point and the external carotid artery ECA, resulting in only about three two-dimensional coronal sequence images at most. This increases the difficulty of 2D coronal plane segmentation. In order to quickly obtain an inner membrane close to the real one, the method based on mathematical morphology is used in the present invention to obtain the initialized inner contour.

(5.2.1.1) Image Thresholding

There are four main categories of thresholding methods: point-based global thresholding methods, region-based global thresholding methods, local thresholding methods, and multi-thresholding methods. The embodiment of the present invention adopts the Otsu algorithm OTSU (adaptive thresholding), so that the white area includes the intima, media, adventitia, extra-adventitia tissue and part of the noise, and the black area is the vascular lumen, as shown in Figure 31;

(5.2.1.2) filling holes

In order to fill the holes in the white area shown in FIG. 27, template matching method, point-by-point scanning method, etc. can be used. In view of the characteristics of the binary image, the present invention adopts the morphological closing operation, and the result is shown in FIG. 32 .

(5.2.1.3) Deburring

In order to make the surface of the vascular intima smooth and conform to the physiological characteristics, the mathematical morphology opening operation is used to remove the burrs generated by the additional noise on the surface of the white area, and the result is shown in Figure 33;

(5.2.1.3) Correction of initial intimal contour

For the edge points of the intima surface shown in Figure 33, perform gap extraction on the upper edge pixels of the white area, and use the extracted pixels to extract pixels with a distance of 3 to 5 pixels from them to obtain a series of The initial coordinate points of the intima edge, the series of initial coordinate points of the intima edge are used as the initial contour of the intima, as shown in Figure 34;

(5.2.2) Extract inner contour (Figure 35)

According to the gradient information of the region of interest, the classic snake method is used to calculate the internal and external forces of the initial contour of the intima. Compare the internal and external forces, and evolve the contour according to the comparison results until the internal and external forces are equal, and the final contour is the extracted internal contour, as shown in Figure 35, where the initial contour of the inner membrane is a dotted line; the inner membrane The final contour of is a solid line. In addition to the classic snake (Snake) method, evolutionary methods such as level set, CV model, and GVF-Snak can also be used.

(5.2.3) Initialize the adventitia contour

The initial adventitia contour can be obtained by moving the initial contour of the intima obtained in (5.2.1) downward by Δ _index pixel distance in parallel. The general Δ _index is 15~30, and the Δ _index is preferably 25.

(5.2.4) Extract the outer membrane contour (Figure 36)

According to the gradient information and field information of the region of interest, the gradient vector field method GVF-Snake is used to calculate the internal and external forces of the initial contour of the adventitia. Compare the internal and external forces, and evolve the contour according to the comparison results until the internal and external forces are equal, and the finally determined contour is the contour of the adventitia, as shown in Figure 36, where the initial contour of the adventitia is a dotted line; The final contour is a solid line. In addition to the GVF-Snake algorithm, evolutionary methods such as classic snake (Snake), level set, and CV models can also be used.

(5.2.5) Calculate the intima-media thickness of the two-dimensional coronal ROI (Figure 37)

After (5.2.1)-(5.2.4) segmentation, the corresponding pixel positions of the inner and outer membranes in each column of the total ROI of the two-dimensional coronal plane can be obtained, and the pixel spacing between the corresponding pixels of the inner and outer membrane contours can be calculated , and convert it into the actual distance, and calculate the average value of each column, that is, the intima-media thickness corresponding to the two-dimensional coronal plane. As shown in Figure 37, the distance between the final contour of the upper intima (solid line) and the final contour of the lower adventitia (solid line) is the desired intima-media thickness.

(5.3) According to the intima-media thickness of each region of interest, statistically calculate its mean value, which is the intima-media thickness of the coronal plane;

For each of the two-dimensional coronal sequence images, after the above steps (5.1) and (5.2), perform statistical analysis of the results, and obtain parameters such as the mean value and variance of coronal vascular intima-media thickness IMT as vascular evaluation indicators . IMT value has become a common and effective indicator for predicting cardiovascular and cerebrovascular diseases.

Scheme verification:

Compare the segmentation results of (3), (4) and (5) with the manual segmentation results, and calculate the DSC (Dice Similarity Coefficient), MAD (Mean Absolute Distance) and MAXD (Maximum Absolute Distance) of the three-view segmentation results for algorithm verification Evaluation and Analysis. As shown in Figure 22, Figure 24, Figure 38 and Table 1:

A. Similarity coefficient DSC:

The similarity coefficient DSC is a standard for measuring the segmentation algorithm, which is used to evaluate the similarity of the area.

$\mathrm{<span\; class="notranslate">\; DSC</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 2</span>}\frac{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; \&cap;</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; |</span>}$

Among them, R _M and R _A represent the area surrounded by the manual contour and the area surrounded by the automatic contour, respectively. The closer the similarity coefficient DSC is to 1, the closer the two methods are.

B. Average absolute distance MAD and maximum absolute distance MAXD

Both the average absolute distance MAD and the maximum absolute distance MAXD are used to measure the segmentation algorithm and are used to evaluate the distance similarity.

Assuming that the manual contour point is {m _i : i=1, the contour point shall prevail, and the corresponding automatic contour point is {a _i : i=1, the contour point shall be . but

$\mathrm{<span\; class="notranslate">\; MAD</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$

$\mathrm{<span\; class="notranslate">\; MAXD</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; \}</span>}{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; \{</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \}</span>$

Among them, d(m _i ,A) is the relative distance between the corresponding points of the manual contour and the automatic contour. The smaller the mean absolute distance MAD and the maximum absolute distance MAXD, the closer the two methods are.

Table 1 The present invention compares with manual segmentation result

DSC (%) MAD(mm) MAXD(mm) Two-dimensional cross-sectional inner contour extraction blood vessels 94.1±3.6 0.32±0.21 0.77±0.47 Two-dimensional cross-sectional outline extraction of blood vessels 92.8±2.5 0.37±0.19 0.84±0.39 Two-dimensional sagittal plane IMT thickness measurement 0.73±0.47 1.01±0.75 Two-dimensional coronal IMT thickness measurement 1.02±0.76 1.89±0.34

In the above table, the mean value and standard deviation of the measured thickness are in mm.

It can be seen from the table that (1) the segmentation method used in the present invention is equivalent to the expert manual segmentation method, and its feasibility and robustness have been verified; (2) in the sequence image segmentation of two-dimensional cross-section, The method of the present invention has a higher similarity with the expert manual segmentation method, the similarity coefficient DSC is greater than 90%, the average absolute distance MAD and the maximum absolute distance MAXD are all within the error range; (3) in the two-dimensional sagittal plane and coronal plane In the IMT measurement, the two methods are equivalent to the gold standard of manual IMT measurement, and the average error is about 1 mm; (4) The measurement results of the two-dimensional sagittal plane are better than the two-dimensional coronal plane, which is related to the quantity and quality of the sequence images. , The specific segmentation method has a certain relationship. In addition, in terms of operation time, the present invention can complete the measurement of a three-dimensional volume data in 3 minutes; while experienced doctors usually need 8 minutes to complete the manual measurement.

The implementation examples listed above only express several implementation modes of the present invention, and the description thereof is relatively specific and detailed, but should not be construed as limiting the patent scope of the invention. It should be pointed out and emphasized that those skilled in the art can make various modifications and improvements without departing from the present invention, and these should belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be based on the appended claims.

It is easy for those skilled in the art to 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 (6)

1. the main carotid artery vascular based on neck ultrasonoscopy extracts and method for measuring thickness, comprises the following steps:

Read neck ultrasonic three-dimensional volume data, mark main carotid artery vascular central shaft based on main carotid bifuracation point;

According to central shaft, by three-view diagram direction projection, cutting neck ultrasonic three-dimensional volume data, obtains two-dimentional transversal section, sagittal plane and coronal-plane sequence image;

In each image of two-dimentional transversal section sequence image, choose each main carotid artery vascular area-of-interest respectively, and to the pre-service of each main carotid artery vascular area-of-interest; In each main carotid artery vascular area-of-interest after the pre-treatment, segmentation obtains the inside and outside profile of each main carotid artery vascular respectively; According to the inside and outside profile of the main carotid artery vascular area-of-interest of each two-dimentional cross-sectional image, by its spatial relation three-dimensional reconstruction, obtain three-dimensional main carotid artery vascular profile;

In each image of two-dimentional sagittal plane sequence image, choose each main carotid artery vascular area-of-interest respectively, and to the pre-service of each main carotid artery vascular area-of-interest; In each main carotid artery vascular area-of-interest after the pre-treatment, segmentation obtains the inside and outside profile of each main carotid artery vascular respectively; Thickness between the inside and outside profile calculating each main carotid artery vascular respectively and Internal-media thickness; Add up the Internal-media thickness average of each two-dimentional sagittal view picture;

In each image of two-dimensional coronal face sequence image, choose each main carotid artery vascular area-of-interest respectively, and to the pre-service of each main carotid artery vascular area-of-interest; In each main carotid artery vascular area-of-interest after the pre-treatment, segmentation obtains the inside and outside profile of each main carotid artery vascular respectively; Thickness between the inside and outside profile calculating each main carotid artery vascular respectively and Internal-media thickness; Add up the Internal-media thickness average of each two-dimensional coronal face image;

In main carotid artery vascular area-of-interest in each image of described two-dimentional transversal section sequence image, segmentation obtains the inside and outside profile of each main carotid artery vascular in the following manner:

A1, in the pretreated main carotid artery vascular area-of-interest of two-dimentional cross-sectional image, extract the Internal periphery of main carotid artery vascular, and do smoothing processing;

A2, carry out ellipse fitting according to the Internal periphery of main carotid artery vascular, after ellipse matching obtained amplifies, as the initial outline of main carotid artery vascular;

A3, initial outline according to main carotid artery vascular, develop and obtain the final outline of main carotid artery vascular;

In main carotid artery vascular area-of-interest in each image of described two-dimentional sagittal plane sequence image, segmentation obtains the inside and outside profile of each main carotid artery vascular in the following manner:

B1, the point arranged as each in pretreated main carotid artery vascular area-of-interest two-dimentional sagittal view, carry out sequence number mark from top to bottom successively, as the index value of each row point;

The coarse segmentation of B2, area-of-interest:

B21, acquisition lower boundary profile DB: from left to right, order scans each row of the main carotid artery vascular area-of-interest of two-dimentional sagittal view picture, obtains maximum gradation value of each row, minimum gradation value and meets the lower boundary candidate point of gray threshold condition; In lower boundary candidate point, the maximum point of marked index value is as unique lower boundary point; Connect all lower boundary point in turn and form lower boundary profile DB; Wherein, lower boundary candidate point gray-scale value GV _candidatethe gray threshold condition that should meet is: GV _candidate>=a × Δ GV+GV _min=a × (GV _max-GV _min)+GV _min, the span of weight coefficient a is 0.87 ~ 0.93;

B22, acquisition coboundary profile UB: lower boundary profile DB is upwards moved in parallel 20 ~ 30 pixels as coboundary temporary profile UB_t in the main carotid artery vascular area-of-interest of two-dimentional sagittal view picture; Choose the template that size is X × X, X gets 8 ~ 15 pixels, utilizes template from left to right, travels through from top to bottom, the pixel average EX of calculation template overlay area and variance DX to the region between coboundary temporary profile UB_t and lower boundary profile DB; If meet mean variance threshold condition EX≤b and DX≤c, b span is 0.05 ~ 0.09, c span is 0.11 ~ 0.15, then the center of template overlay area is coboundary candidate point; If do not meet mean variance threshold condition, then corresponding UB_t point is coboundary candidate point; In the coboundary candidate point of each row, marked index value the maximum is as unique coboundary point; Connect all coboundaries point in turn and form coboundary profile UB;

The segmentation of B3, area-of-interest is cut: the region segmentation between lower boundary profile DB and coboundary profile UB cuts out beyond lumen of vessels, inner membrance and middle film, adventitia and adventitia and organizes;

The interface of B4, extraction lumen of vessels and inner membrance is Internal periphery, and in extraction, the interface of film and adventitia is outline;

In main carotid artery vascular area-of-interest in each image of described two-dimensional coronal face sequence image, segmentation obtains the inside and outside profile of each main carotid artery vascular in the following manner:

Initialization Internal periphery in C1, the main carotid artery vascular area-of-interest of employing Mathematical Morphology Method after the Image semantic classification of two-dimensional coronal face, develops according to initialization Internal periphery and generates final Internal periphery;

C2, to be moved down obtain initial outline by initial Internal periphery, developing according to initial outline generates final outline.

2. the main carotid artery vascular based on neck ultrasonoscopy according to claim 1 extracts and method for measuring thickness, it is characterized in that, described steps A 1 is specially: in the pretreated main carotid artery vascular area-of-interest of two-dimentional cross-sectional image, extract home position and the radius of main carotid artery vascular, extract Internal periphery according to home position and radius.

3. the main carotid artery vascular based on neck ultrasonoscopy according to claim 1 extracts and method for measuring thickness, and it is characterized in that, in described steps A 2, oval magnification ratio coefficient is 1.02 ~ 1.08.

4. the main carotid artery vascular based on neck ultrasonoscopy according to claim 1 extracts and method for measuring thickness, it is characterized in that, the evolution method that described steps A 3 adopts be Active Contour Model, GVF-Snake, Fast Marching Method, Level Set, Sparse Field Algorithm any one.

5. main carotid artery vascular according to claim 1 extracts and method for measuring thickness, it is characterized in that, adopt in C average, fuzzy C-mean algorithm, Support Vector Machine SVM, BP, self-adaptive BP, AdaBoost algorithm in the thin segmentation step of described area-of-interest any one.

6. main carotid artery vascular according to claim 1 extracts and method for measuring thickness, it is characterized in that, the evolution method that described step C1 and C2 adopt is any one in classical S-Shaped Algorithm, GVF-Snake algorithm, level set, MS model, CV model evolution method.