[ summary of the invention ]
Based on this, it is necessary to provide an automatic image-based intravascular membrane thickness extraction method that can reduce the influence of noise and is convenient to operate.
An image-based intravascular medium-thickness automatic extraction method comprises the following steps:
obtaining a blood vessel ultrasonic image, and selecting an interested area from the blood vessel ultrasonic image;
denoising the region of interest by adopting an empirical mode decomposition algorithm based on a non-uniform B spline;
classifying the pixel points in the denoised interesting region based on pixel gray scale by a K-means clustering method so as to separate three parts of a blood vessel cavity, a blood vessel wall adventitia and other regions except the blood vessel cavity and the blood vessel wall adventitia in the interesting region;
and extracting the thickness of the intravascular medium membrane from the other region parts except the blood vessel cavity and the adventitia of the blood vessel wall in the separated region of interest by mathematical morphology.
Preferably, the step of denoising the region of interest by using an empirical mode decomposition algorithm based on a non-uniform B-spline specifically includes the following steps:
extracting local maximum points and local minimum points in the region of interest;
and respectively carrying out interpolation fitting on the extreme value point and the minimum value point by a non-uniform B-spline surface fitting method to obtain an upper envelope surface and a lower envelope surface of the region of interest, and calculating the mean value of the upper envelope surface and the lower envelope surface to obtain a residual signal of the first-scale decomposition of the empirical mode decomposition algorithm, namely the denoised region of interest.
Preferably, after the step of extracting the local maximum point and the local minimum point in the region of interest, the method further includes the steps of:
dense grids are arranged in the longitudinal direction of the blood vessel wall of the region of interest, and sparse grids are arranged in the tangential direction of the blood vessel wall;
and respectively carrying out interpolation fitting on the extreme value points and the minimum value points by the non-uniform B-spline surface fitting method according to the dense grids and the sparse grids.
Preferably, before the step of denoising the region of interest by using the empirical mode decomposition algorithm based on the non-uniform B-spline, the method further comprises the steps of:
and preliminarily denoising the region of interest by adopting a Gaussian filter.
Preferably, the step of classifying the pixel points in the denoised region of interest based on pixel gray scale by using a K-means clustering method to separate the blood vessel lumen, the blood vessel wall adventitia and other regions in the region of interest except the blood vessel lumen and the blood vessel wall adventitia specifically comprises:
setting a characteristic vector consisting of gray values representing the gray levels of three types of bright pixels, gray pixels and dark pixels as an initial value of a clustering center;
and separating the region of interest into a first region, a second region and a third region according to the initial value of the clustering center, wherein the first region corresponds to the blood vessel cavity part, the second region corresponds to the blood vessel wall adventitia part, and the third region corresponds to other region parts except the blood vessel cavity and the blood vessel wall adventitia in the region of interest.
Preferably, the step of extracting the thickness of the intravascular medium membrane from the other regions except the blood vessel cavity and the adventitia of the blood vessel wall in the separated region of interest by mathematical morphology specifically comprises the following steps:
presetting form radiuses as a first variable parameter and a second variable parameter respectively;
and extracting the adventitia of the blood vessel wall, removing the part serving as a mask from the region of interest, then performing segmentation operation by taking the shape radius as a first variable parameter and a second variable parameter respectively by taking the other region parts except the blood vessel cavity and the blood vessel wall adventitia in the region of interest as a foreground to obtain a segmentation result of the intima-media region of the blood vessel, and measuring the segmentation result to obtain the intima-media thickness of the blood vessel.
In addition, it is necessary to provide an automatic image-based intravascular intima-media thickness extraction system that reduces the noise effect and is easy to operate.
An image-based automatic extraction system for intima-media thickness in blood vessels, comprising:
the image acquisition module is used for acquiring a blood vessel ultrasonic image and selecting an interested area from the blood vessel ultrasonic image;
the denoising processing module is used for denoising the region of interest by adopting an empirical mode decomposition algorithm based on a non-uniform B spline;
the separation module is used for classifying the pixel points in the denoised region of interest based on pixel gray scale through a K-means clustering method so as to separate three parts of a blood vessel cavity, a blood vessel wall adventitia and other regions except the blood vessel cavity and the blood vessel wall adventitia in the region of interest;
and the extraction module is used for extracting the thickness of the intravascular medium membrane from other regions except the blood vessel cavity and the adventitia of the blood vessel wall in the separated region of interest through mathematical morphology.
Preferably, the denoising processing module is further configured to extract a local maximum point and a local minimum point in the region of interest, perform interpolation fitting on the maximum point and the minimum point respectively by using a non-uniform B-spline surface fitting method to obtain an upper envelope surface and a lower envelope surface of the region of interest, and calculate an average value of the upper envelope surface and the lower envelope surface to obtain a residual signal of a first scale decomposition of the empirical mode decomposition algorithm, which is the denoised region of interest.
Preferably, the denoising processing module is further configured to set a dense grid in the vessel wall longitudinal direction of the region of interest and a sparse grid in the vessel wall tangential direction, and then perform interpolation fitting on the maximum value point and the minimum value point respectively through the non-uniform B-spline surface fitting method according to the grids.
Preferably, a gaussian filter is further included, and the gaussian filter is used for preliminarily denoising the region of interest.
Preferably, the separation module is further configured to set eigenvectors formed by gray values representing gray levels of three types of pixels, namely light, gray and dark, as initial values of a clustering center, and separate the region of interest into a first region, a second region and a third region according to the initial values of the clustering center, where the first region corresponds to the blood vessel cavity portion, the second region corresponds to the blood vessel wall adventitia portion, and the third region corresponds to other region portions of the region of interest except the blood vessel cavity and the blood vessel wall adventitia.
Preferably, the extraction module is further configured to preset shape radii as a first variable parameter and a second variable parameter respectively, extract the vessel wall adventitia, remove the part as a mask from the region of interest, perform segmentation operation on the other region parts except the blood vessel cavity and the vessel wall adventitia in the region of interest as a foreground by using the shape radii as the first variable parameter and the second variable parameter respectively, obtain a segmentation result of the blood vessel intima region, and measure the segmentation result to obtain the blood vessel intima-media thickness.
According to the method and the system for automatically extracting the thickness of the intravascular medium membrane based on the image, after the selected region of interest is denoised by the empirical mode decomposition algorithm based on the non-uniform B-spline, the noise influence can be reduced, pixel points of the region of interest are classified based on the pixel gray scale through the K-means clustering method so as to separate the region of interest, and the thickness of the intravascular medium membrane is extracted from other regions except a blood vessel cavity and a blood vessel wall adventitia in the separated region of interest.
[ detailed description ] embodiments
The following describes the method and system for automatically extracting the intravascular membrane thickness based on an image in detail with reference to specific embodiments and drawings.
As shown in fig. 1, in one embodiment, an image-based method for automatically extracting a thickness of an intravascular membrane includes the following steps:
step S110, obtaining a blood vessel ultrasonic image, and selecting an interested area from the blood vessel ultrasonic image.
An ultrasound probe may be used to acquire an image of the blood vessel and then select a region of interest from the acquired image. The region of interest is a rectangular region around the vessel wall containing the vessel wall, the intima-media lining of the vessel, and structures external to the vessel. Fig. 2 shows an area of interest selected from a carotid artery ultrasound image, wherein the black background at the upper part is inside a blood vessel cavity, the three thin strip-shaped bright-dark-bright stripes at the middle part are intima, media and adventitia tissues respectively, and the large area at the lower part is other areas (other tissues) and artifacts except the blood vessel cavity and the blood vessel wall adventitia in the area of interest. To measure IMT, the two luminal-intima and media-adventitia margins of the blood vessel are separated, and the two "+" marks in the figure indicate the markers for manual measurement of IMT by the physician.
And S120, denoising the region of interest by adopting an empirical mode decomposition algorithm based on non-uniform B-splines.
And denoising the image of the region of interest by an Empirical Mode Decomposition (EMD) algorithm of non-uniform B splines to remove noise. As shown in fig. 3, the method specifically comprises the following steps:
step S121 extracts a local maximum point and a local minimum point in the region of interest.
And extracting local maximum points and local minimum points in the region of interest by a four-neighborhood algorithm or an eight-neighborhood algorithm. As shown in fig. 4, the luminance is the extracted maximum point.
An image can be treated as a discrete two-dimensional array, and the process of screening the local extreme value is to compare the value of the position with four neighborhood values in the upper, lower, left and right directions or eight neighborhood values in the upper, lower, left, right, southeast, northwest, northeast and southwest directions, in this embodiment, a four-domain algorithm is adopted, and the set judgment condition is as follows:
(1) if the value of the central position is larger than all the adjacent values, the point is regarded as a local maximum value point;
(2) if the value of the central position is smaller than all the adjacent values, the point is regarded as a local minimum value point;
(3) if the value of the central position is smaller than one part of adjacent values and is larger than the other part of adjacent values, the central position is regarded as a non-extreme point;
(4) if a certain point is adjacent to the local extreme point and equal to the adjacent local extreme value, the two points are regarded as an area until the area satisfies the conditions (1) to (3) or the boundary of the area is found.
And S123, respectively carrying out interpolation fitting on the extreme value points and the minimum value points through a non-uniform B-spline surface fitting method to obtain an upper envelope surface and a lower envelope surface of the region of interest, and calculating the mean values of the upper envelope surface and the lower envelope surface to obtain a residual signal of the first scale decomposition of the empirical mode decomposition algorithm, namely the denoised region of interest.
The non-uniform B-spline surface fitting method, namely the B-spline surface fitting method of the scattered data, carries out interpolation fitting on the extreme value point and the minimum value point through the non-uniform B-spline surface fitting method to obtain an upper envelope I of the image of the region of interestmaxAnd a lower envelope IminCurved surface, calculating the mean value I of the twomean=(Imax+Imin) And/2, obtaining a residual signal of the first-scale decomposition of the EMD. Experimental verification shows that the residual signal of the first-scale decomposition of the EMD algorithm can best retain the hierarchical structure and details of the vascular wall, and has ideal performance in smoothness and detail resolution capability, so that the residual signal is used as the basis of subsequent segmentation. Fig. 5 is the result of the first-scale decomposition of the EMD algorithm, and it can be seen that compared to fig. 2, the structure and edges are preserved while the whole is much smoothed.
The specific process of fitting by adopting the B spline surface is as follows: in the rectangular image space Ω { (x, y) |0 ≦ x < m, 0 ≦ y < n }, there are some sets of scattered data P ≦ x ≦ n { (x, y) |
c,y
c,z
c) Wherein (x)
c,y
c) Is a point in Ω. The function f is approximated by a uniform B-spline functionAnd fitting the curved surface where the P is located. Let the grid of the function f be phi, phi
ijRepresenting the value of Φ at the (i, j) position, the corresponding fitting function f can be expressed as
Wherein i ═ x]-1,j=[y]-1,s=x-[x],t=y-[y]。B
kAnd B
lIs a unit third order B-spline basis function:
B0(t)=(1-t)3/6;B1(t)=(3t3-6t2+4)/6;B2(t)=(-3t3+3t2+3t+1)/6;B3(t)=t36; wherein, t is more than or equal to 0 and less than 1, the function of the method is to mark different control points, and the control points are represented as different weight coefficients according to different distances from (x, y).
The grid Φ is further determined below, taking as an example one of the scattered data points, the function f is in (x)c,yc) Is taken asWherein, Wkl=Bk(s)Bl(t),s=xc-1,t=yc-1. At a plurality of phi satisfying the above conditionklIn accordance with the least square criterionTo obtain
Then, extending to all data points in P, each data point affects 4 × 4 control points around it, corresponding to each data point (x)
c,y
c,z
c) All have a corresponding phi
ij:
At the same position, a plurality of different phi can be obtained
cTo obtain a phi
ijUnique solution, by minimizing the error e (phi)
ij)=∑
c(W
cφ
ij-W
cφ
c)
2To obtain
Thus, for a given mesh size, an optimal approximation fit equation can be obtained.
In addition, when the extreme value point and the minimum value point are subjected to interpolation fitting, the size of the grid, namely the density and the sparsity of the grid, has a direct influence on the fitting result. Fig. 6A-6D show the results of B-spline surface fitting for the same set of data at different mesh sizes. FIG. 6A is a scattered data point. Fig. 6B is the fitting result for a grid of 8 × 8. Fig. 6C is the fitting result for a grid of 16 × 16. Fig. 6D is the fitting result for a grid of 32 x 32. It can be seen that the larger the grid, such as fig. 6D, the higher the goodness of fit of the results to the raw data, but at the same time the more drastic the variation of the results, the relatively flat results of fitting the corresponding smaller grid, such as fig. 6B. Therefore, the proper grid size is selected according to the needs and the data characteristics in specific application.
Preferably, the dense grid and the tangential direction of the vessel wall are arranged in the longitudinal direction of the vessel wall of the region of interest, i.e. in the vertical direction, i.e. in the horizontal direction, and the sparse grid is arranged; and respectively carrying out interpolation fitting on the extreme value points and the minimum value points by a non-uniform B spline surface fitting method according to the dense grid and the sparse grid. In the longitudinal direction of the vascular wall of the region of interest, namely in the vertical direction, dense grids are adopted to have better detail resolution capability, so that better resolution of the hierarchical structure of the vascular wall is kept; in the blood vessel wall tangential of the region of interest, namely the blood horizontal direction, the sparse grid is adopted to have a better smoothing effect, so that the purposes of smoothing the structure of the blood vessel wall and removing noise are achieved. In the embodiment, the size of the grid adopted in the horizontal and vertical directions of the blood vessel wall is (X/12, Y/3), so that the grid is sparse in tangential direction and dense in longitudinal direction, a good denoising effect is achieved, and good resolution is kept. Where (X, Y) is the size of the region-of-interest image.
In a preferred embodiment, the step of denoising the region of interest by using the EMD algorithm of the non-uniform B-spline further comprises: and preliminarily denoising the region of interest by adopting a Gaussian filter. In order to reduce the error of selecting the local extreme point, the influence of point-like noise in the ultrasound image can be obviously weakened by using a 1-sigma Gaussian filter.
And S130, classifying the pixel points in the denoised interested region based on the pixel gray scale by a K-means clustering method so as to separate three parts of a blood vessel cavity, a blood vessel wall adventitia and other regions except the blood vessel cavity and the blood vessel wall adventitia in the interested region.
The specific steps of separating the interesting regions by the K-means clustering method are as follows: setting a characteristic vector consisting of gray values representing the gray levels of three types of bright pixels, gray pixels and dark pixels as an initial value of a clustering center; and separating the region of interest into a first region, a second region and a third region according to the initial value of the clustering center, wherein the first region corresponds to a vascular cavity part, the second region corresponds to a vascular wall adventitia part, and the third region corresponds to other region parts except the vascular cavity and the vascular wall adventitia in the region of interest, namely including the vascular intima, media and other tissues. When the K-means clustering method is used for calculation, an initial value of a clustering center is set, iteration is carried out, namely data are distributed to the closest clustering center according to a similarity criterion, the data are redistributed, and then the clustering center is updated, namely the average vector of each type is used as a new clustering center. Fig. 7 shows the result of K-means clustering, and the whole image is divided into three regions, the first region 71, the dark region in the figure, is the vessel lumen portion, and the second region 72, the most bright region in the figure, is the region containing adventitial tissue of the vessel wall. The third region 73, i.e., the second bright region (gray region) in the figure corresponds to the other regions of the region of interest except the blood vessel lumen and the adventitia of the blood vessel wall, wherein the uppermost narrow second bright region 731 (a part of the gray region) is the tissue portion of the intima-media wall of the blood vessel wall, and the maximum second bright region 732 (a part of the gray region) corresponds to the other tissues of the blood vessel lumen and the blood vessel wall. In this embodiment, the initial cluster center initial values are set to 0, 100, and 200.
The K-means clustering method only considers the gray value of the pixel point, does not consider the spatial position and neighborhood information, and is simple in algorithm.
In step S140, the intravascular intermediate membrane thickness is extracted from the separated regions of interest other than the blood vessel lumen and the vessel wall adventitia by mathematical morphology.
The step of extracting the thickness of the intravascular medium membrane by adopting mathematical morphology specifically comprises the following steps: presetting form radiuses as a first variable parameter and a second variable parameter respectively; and extracting the adventitia part of the blood vessel wall, removing the adventitia part of the blood vessel wall from the region of interest by taking the region of interest as a mask, then taking other regions except the blood vessel cavity and the blood vessel wall adventitia as a foreground, respectively carrying out segmentation operation by taking the form radius as a first variable parameter and a second variable parameter to obtain a segmentation result of the intima-media region of the blood vessel, and measuring the segmentation result to obtain the intima-media thickness of the blood vessel. And performing opening operation by taking the form radius as a first variable parameter to separate the intima-media tissue in the blood vessel, performing further smoothing by using the form radius as a second variable parameter to obtain a final segmentation result of the intima-media region in the blood vessel, and measuring the segmentation result to obtain the intima-media thickness in the blood vessel. In this embodiment, the first variable parameter is 2 and the second variable parameter is 4. Fig. 8 is a mathematical morphology of the resulting endomesenteric region. Fig. 9 is a mark of the final segmentation result on the image of the region of interest, which is a closed bright ring shape, and it can be seen that the result is relatively accurate.
In addition, the method for automatically extracting the intravascular middle membrane thickness based on the image further comprises the following steps: and providing an IMT segmentation software system for human-computer interaction and algorithm parameter setting. The IMT segmentation software system may provide a visualized image processing environment. FIG. 10 is an interface diagram of a software system for IMT segmentation.
As shown in fig. 11, in one embodiment, an automatic image-based extraction system for intima-media thickness in blood vessels includes an image acquisition module 100, a denoising module 200, a separation module 300, and an extraction module 400. Wherein,
the image acquisition module 100 is configured to acquire a blood vessel ultrasound image and select an area of interest from the blood vessel ultrasound image. An ultrasound probe may be used to acquire the blood vessel image. The image acquisition module 100 acquires the acquired vascular ultrasound image and then selects a region of interest from the acquired vascular ultrasound image. The region of interest is a rectangular region around the vessel wall containing the vessel wall, the intima-media lining of the vessel, and structures external to the vessel. Fig. 2 shows an area of interest selected from a carotid artery ultrasound image, wherein the black background at the upper part is inside a blood vessel cavity, the three thin strip-shaped bright-dark-bright stripes at the middle part are intima, media and adventitia tissues respectively, and the large area at the lower part is other areas and artifacts except the blood vessel cavity and the blood vessel wall adventitia in the area of interest. To measure IMT, the two luminal-intima and media-adventitia margins of the blood vessel are separated, and the two "+" marks in the figure indicate the markers for manual measurement of IMT by the physician.
The denoising processing module 200 is configured to denoise the region of interest by using an empirical mode decomposition algorithm based on non-uniform B-splines. And denoising the image of the region of interest by an Empirical Mode Decomposition (EMD) algorithm of non-uniform B splines to remove noise.
The denoising processing module 200 is further configured to extract a local maximum point and a local minimum point in the region of interest, perform interpolation fitting on the maximum point and the minimum point respectively by using a non-uniform B-spline surface fitting method to obtain an upper envelope surface and a lower envelope surface of the region of interest, and calculate an average value of the upper envelope surface and the lower envelope surface to obtain a residual signal of the first scale decomposition of the empirical mode decomposition algorithm, which is the denoised region of interest. The denoising processing module 200 extracts a local maximum point and a local minimum point in the region of interest through a four-neighbor region algorithm. The specific judgment conditions for extracting the local maximum value points and setting the local minimum value points by adopting the four-neighborhood algorithm are described in the method, and are not described herein again.
The non-uniform B-spline surface fitting method, namely the B-spline surface fitting method of the scattered data, carries out interpolation fitting on the extreme value point and the minimum value point through the non-uniform B-spline surface fitting method to obtain an upper envelope I of the image of the region of interestmaxAnd a lower envelope IminCurved surface, calculating the mean value I of the twomean(Imax+Imin) And/2, obtaining a residual signal of the first-scale decomposition of the EMD, namely the region of interest after denoising. Experimental verification shows that the residual signal of the first-scale decomposition of the EMD algorithm can best retain the hierarchical structure and details of the vascular wall, and has ideal performance in smoothness and detail resolution capability, so that the residual signal is used as the basis of subsequent segmentation. Fig. 5 is the result of the first-scale decomposition of the EMD algorithm, and it can be seen that compared to fig. 2, the structure and edges are preserved while the whole is much smoothed. The specific process of fitting the maximum value point and the minimum value point by using the B-spline surface is described in the method, and is not described herein again.
In addition, when the extreme value point and the minimum value point are subjected to interpolation fitting, the size of the grid, namely the density and the sparsity of the grid, has a direct influence on the fitting result. Fig. 6A to 6D show the same set of data, and the fitting results of the B-spline surface under different mesh sizes, and it can be seen that the larger and denser the mesh is, the higher the goodness of fit of the result to the original data is, but at the same time, the more drastic the change of the result is, the more sparse the corresponding smaller mesh is, and the relatively gentle the fitting result is, so that an appropriate mesh size should be selected according to the needs and the data characteristics in specific applications.
Preferably, the dense grids and the tangential direction of the vessel wall are arranged in the longitudinal direction, namely the vertical direction, of the vessel wall of the region of interest, namely the horizontal direction, and the sparse grids are arranged; and respectively carrying out interpolation fitting on the extreme value points and the minimum value points by a non-uniform B spline surface fitting method according to the dense grid and the sparse grid. In the longitudinal direction of the vascular wall of the region of interest, namely in the vertical direction, dense grids are adopted to have better detail resolution capability, so that better resolution of the hierarchical structure of the vascular wall is kept; in the tangential direction of the blood vessel wall of the region of interest, namely in the horizontal direction, the sparse grid is adopted to have a better smoothing effect, so that the purposes of smoothing the structure of the blood vessel wall and removing noise are achieved. In the implementation case, the size of the grid adopted in the horizontal and vertical directions of the blood vessel wall is (X/12, Y/3), so that the grid is sparse in tangential direction and dense in longitudinal direction, a good denoising effect is achieved, and good resolution is kept. Where (X, Y) is the size of the region-of-interest image.
The separation module 300 is configured to classify the pixel points in the denoised region of interest based on the pixel gray levels by a K-means clustering method, so as to separate three parts, namely a blood vessel cavity, a blood vessel wall adventitia, and other regions in the region of interest except the blood vessel cavity and the blood vessel wall adventitia. The separation module 300 is further configured to set a feature vector composed of gray values representing gray levels of three types of pixels, namely light, gray and dark, as an initial value of a clustering center, and separate the region of interest into a first region, a second region and a third region according to the initial value of the clustering center, where the first region corresponds to a portion of a lumen of a blood vessel, the second region corresponds to a portion of an adventitia of a blood vessel wall, and the third region corresponds to portions of other regions of the region of interest except the lumen of the blood vessel and the adventitia of the blood vessel wall. When the K-means clustering method is used for calculation, an initial value of a clustering center is set, iteration is carried out, namely data are distributed to the closest clustering center according to a similarity criterion, the data are redistributed, and then the clustering center is updated, namely the average vector of each type is used as a new clustering center. Fig. 7 shows the result of K-means clustering, and the whole image is divided into three regions, the first region 71, the dark region in the figure, is the vessel lumen portion, and the second region 72, the most bright region in the figure, is the region containing adventitial tissue of the vessel wall. The third region 73, i.e., the second bright region (gray region) in the figure corresponds to the other regions of the region of interest except the blood vessel lumen and the adventitia of the blood vessel wall, wherein the uppermost narrow second bright region 731 (a part of the gray region) is the tissue portion of the intima-media wall of the blood vessel wall, and the maximum second bright region 732 (a part of the gray region) corresponds to the other tissues of the blood vessel lumen and the blood vessel wall.
The K-means clustering method only considers the gray value of the pixel point, does not consider the spatial position and neighborhood information, and is simple in algorithm.
The extraction module 400 is used for extracting the thickness of the intravascular medium membrane from other regions of the separated region of interest except for the blood vessel cavity and the adventitia of the blood vessel wall through mathematical morphology. The extraction module 400 extracts the thickness of the tunica media in the blood vessel by adopting mathematical morphology, specifically: the method comprises the steps of presetting form radiuses as a first variable parameter and a second variable parameter respectively, extracting a blood vessel wall adventitia part, namely a part containing a blood vessel wall adventitia tissue region, removing the part from the region of interest as a mask, then taking other regions except a blood vessel cavity and the blood vessel wall adventitia in the region of interest as a foreground, performing segmentation operation by taking the form radiuses as the first variable parameter and the second variable parameter respectively to obtain a segmentation result of an intravascular midfilm region, and measuring the segmentation result to obtain the intravascular midfilm thickness. And performing opening operation by taking the form radius as a first variable parameter to separate the intima-media tissue in the blood vessel, performing opening operation by taking the form radius as a second variable parameter to further smooth the intima-media tissue, obtaining a segmentation result of the intima-media region in the blood vessel, and measuring the segmentation result to obtain the intima-media thickness in the blood vessel. In this embodiment, the first variable parameter is 2 and the second variable parameter is 4. Fig. 8 is a mathematical morphology of the resulting endomesenteric region. Fig. 9 is a mark of the final segmentation result on the image of the region of interest, which is a closed bright ring shape, and it can be seen that the result is relatively accurate.
In another embodiment, as shown in fig. 12, the above-mentioned automatic extraction system for image-based intima-media thickness includes a gaussian filter 500 in addition to the image acquisition module 100, the denoising processing module 200, the separation module 300, and the extraction module 400. A gaussian filter 500 is used to de-noise the region of interest. To reduce the error in selecting local extreme points, the use of a 1- σ gaussian filter 500 can significantly attenuate the effects of point-like noise in the ultrasound image.
In addition, the system also comprises an IMT segmentation software system for human-computer interaction and algorithm parameter setting. The IMT segmentation software system may also provide a visual image processing environment.
According to the method and the system for automatically extracting the thickness of the intravascular medium membrane based on the image, after the selected region of interest is denoised by the empirical mode decomposition algorithm based on the non-uniform B spline, the noise influence can be reduced, the region of interest is separated by the K-means clustering method, and the thickness of the intravascular medium membrane is extracted from other regions except the blood vessel cavity and the blood vessel wall adventitia in the separated region of interest.
In addition, when the empirical mode decomposition algorithm adopting the non-uniform B-spline is adopted for denoising, the details and the edge information of the vessel wall stripe structure are reserved, and the denoising method can be applied to extraction of the stripe structure in a strong noise environment.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.