A method and system for coronary angiogram image vessel branch recognition and stenosis detection

Abstract

The application discloses a kind of coronary angiography image blood vessel branch identification and stenosis detection method and system, comprising the following steps: A: the extraction of key frame image is carried out to the continuous image sequence of the obtained coronary angiography;B: the extracted key frame image is carried out blood vessel enhancement segmentation processing, and the binary graph after enhancement segmentation is obtained;C: the skeleton graph of blood vessel is obtained using morphological erosion method;D: blood vessel skeleton graph is used to carry out blood vessel branch identification and blood vessel stenosis detection respectively, according to blood vessel branch identification result is marked, and according to blood vessel stenosis detection result, stenosis site is positioned and marked.The application can accurately identify the blood vessel branch and stenosis position in coronary angiography image, and provide data support for medical staff to carry out stenosis detection.

Description

Technical Field

[0001] This invention relates to the field of medical image processing, and more particularly to a method and system for identifying blood vessel branches and detecting stenosis in coronary angiography images. Background Technology

[0002] Coronary artery disease (CAD) is one of the leading causes of death and disability worldwide. Diagnostic methods for CAD include myocardial enzyme testing, electrocardiogram (ECG), echocardiography, and coronary angiography. Among these, coronary angiography is widely recognized as the gold standard for diagnosing coronary artery disease. The acquisition of coronary angiography images is an ongoing process accompanying the angiography procedure, and therefore the results are ultimately presented as a continuous sequence of images.

[0003] In the current technology, there are three main difficulties in accurately diagnosing coronary angiography images:

[0004] 1. Because the coronary angiography process is affected by the patient's heartbeat and slight movements of the angiography platform, the resulting angiographic images will contain more texture and noise;

[0005] 2. Due to the influence of many factors such as the degree of lesion diffusion, the diameter of the reference vessel, and the different anatomical locations, the accuracy of coronary artery stenosis assessment is not good and there is a large fluctuation.

[0006] 3. Making accurate diagnoses based on coronary angiography images requires medical professionals to possess strong professional knowledge and long-term experience.

[0007] Computer image processing technology provides a feasible solution to the above problems. When using machine learning methods to process coronary angiography images to assist doctors in diagnosis, the focus is mainly on two aspects: first, identifying the branches of the coronary arteries, and second, judging the stenosis of the coronary arteries.

[0008] Real-time analysis systems for coronary angiography image branch recognition and stenosis detection, capable of assisting in diagnosis, are not yet widely used in current diagnostics. This is because the algorithms currently used for branch recognition and stenosis detection are mostly deep learning algorithms, requiring large amounts of data for training. Therefore, the accuracy of data annotation largely determines the quality of the results. Furthermore, due to the significant differences in coronary angiography images among different patients, manual annotation cannot cover all situations and is prone to errors. In addition, deep learning algorithms are large in size and have high hardware requirements, increasing costs and preventing large-scale adoption by medical systems. Moreover, there is currently no research on coronary artery recognition using traditional algorithms, and research on stenosis detection generally suffers from low accuracy due to insufficient precision in vessel segmentation and key vessel localization.

[0009] Chinese invention patent application number 202310193935.5, entitled "A Method and Module for Enhancing Foreground Features of DSA Images and a Stenosis Detection Device," discloses the following technical features: receiving an initial feature map of a DSA image, which is pre-extracted from the DSA image by a Faster-RCNN network; and using a channel attention mechanism to obtain a foreground feature enhancement map corresponding to the initial feature map, so that the Faster-RCNN network can locate the vascular stenosis location in the DSA image based on the foreground feature enhancement map. This technical solution can effectively enhance the foreground features of DSA images used for vascular stenosis detection, thereby improving the automation and intelligence of vascular stenosis location in DSA images, and effectively improving the accuracy and effectiveness of using the Faster-RCNN algorithm for vascular stenosis location in DSA images. This technical solution utilizes a channel attention mechanism to obtain a foreground feature enhancement map corresponding to the initial feature map, and then uses the Faster-RCNN network to locate the vascular stenosis location in the DSA image based on this foreground feature enhancement map. However, the above method does not fully consider the influence of vascular branches, resulting in the judgment result needing improvement. Summary of the Invention

[0010] The purpose of this invention is to provide a method and system for identifying blood vessel branches and detecting stenosis in coronary angiography images, which can accurately identify blood vessel branches and stenosis locations in angiography images, providing data support for medical personnel to detect stenosis.

[0011] The present invention adopts the following technical solution:

[0012] A method for identifying blood vessel branches and detecting stenosis in coronary angiography images, comprising the following steps:

[0013] A: Extract keyframe images from the acquired continuous image sequence of coronary angiography;

[0014] B: Perform vessel enhancement segmentation processing on the extracted keyframe images. The vessel enhancement segmentation processing includes image enhancement, image segmentation, and fusion processing of the segmentation results, and finally obtains the enhanced segmented binary image.

[0015] C: Using the enhanced segmented binary image obtained in step B, a morphological erosion method is used to obtain the vascular skeleton map;

[0016] D: Use the vascular skeleton map to identify vascular branches and detect vascular stenosis. Mark the vascular branches according to the results of vascular branch identification, and locate and mark the stenosis sites according to the results of vascular stenosis detection.

[0017] In step A, in the continuous image sequence of coronary angiography, the average pixel gray value of each frame is first calculated, and then the difference between the average pixel gray values ​​of any two adjacent frames is calculated. If the total number of image sequences is odd, the pixel gray value of the middle frame is selected as the reference pixel gray value. If the total number of image sequences is even, the average pixel gray value of the two middle frames is selected as the reference pixel gray value. Then, the image in the continuous image sequence whose difference between the pixel gray value and the reference pixel gray value is within the set pixel difference threshold range and whose average difference between the pixel gray values ​​and the two adjacent frames is the smallest is selected as the keyframe image.

[0018] Step B includes the following specific steps:

[0019] B1: Based on the Frangi blood vessel similarity function, an angle-modified blood vessel similarity function is used for image enhancement to obtain the enhanced keyframe image;

[0020] B2: Perform image segmentation on the enhanced keyframe images;

[0021] B3: For the keyframe images after image segmentation, a region growing method based on geodesic dilatation is used to fuse the segmentation results to obtain an enhanced binary image.

[0022] In step B1, the following angle-modified blood vessel similarity function is used for keyframe image enhancement:

[0023]

[0024]

[0025]

[0026] Where v(λ1,λ2) represents a measure of the degree of blood vessel similarity, λ1 represents the eigenvalues ​​of the Hessian matrix along the blood vessel axis, λ2 represents the eigenvalues ​​of the Hessian matrix perpendicular to the blood vessel axis, and R... C Let β be the arctangent function of the ratio of λ1 to λ2, β be the noise suppression sensitivity parameter, S be the square root of the sum of squares of eigenvalues, α be the vascular structure discrimination parameter, and γ be the noise cancellation parameter.

[0027] Step C includes the following specific steps:

[0028] C1: Perform image thinning on the enhanced segmented binary image to obtain a thinned binary image;

[0029] C2: The area of ​​isolated connected regions in the refined binary graph is denoted as S. zz represents the label of the connected component; the length of the branch in the thinned binary image is denoted as l. The image is denoised according to the following rules to obtain the denoised thinned binary image, and the skeleton image of the blood vessel is obtained by extracting the skeleton.

[0030] If S z If the number of pixels is less than 20, then set all pixels in the isolated connected region to zero and change the length of all branches to l-3.

[0031] Step D includes the following specific steps:

[0032] D1: Using the obtained vascular skeleton map, branch points are found based on the morphological features of the skeleton using a branch-finding algorithm to obtain all pixels with branch point morphological features; after removing small branches and ring-shaped branch points, the branch points are obtained and labeled.

[0033] D2: Using the enhanced segmented binary image, the direction of blood vessels is extracted by the eight-neighbor definition method based on the vascular skeleton, and the diameter of blood vessels is obtained by the pixel thresholding method based on the segmentation boundary and the direction of blood vessels. Finally, the vascular stenosis is detected by comparing the upstream and downstream diameters of blood vessels, and the stenosis sites are located and labeled according to the detection results.

[0034] Step D1 includes the following specific steps:

[0035] D11: Based on the morphological features of the skeleton diagram, define the central pixel among the eight neighboring pixels of a certain pixel in the image as P1. Then, the eight neighboring pixels of pixel P1, starting from the top left pixel and proceeding clockwise, are P9, P2, P3, ..., P7 and P8.

[0036] D12: Using a branch-finding algorithm, all pixels in the vascular skeleton diagram are judged sequentially according to the set branch point judgment rules to obtain all pixels with branch point morphological features.

[0037] The rule for determining branch points is as follows: if the sum of the values ​​of a pixel and its eight neighboring pixels is 4, and the sum of the absolute values ​​of the alternating differences of the outermost pixel values ​​of the pixel is 12, then the pixel possesses the morphological characteristics of a branch point. In other words, by traversing the pixels in the vascular skeleton diagram, pixels that satisfy the following mathematical relationship are considered to possess the morphological characteristics of a branch point:

[0038]

[0039] Where P2 to P9 represent the pixel values ​​of pixels P2 to P9 in the eight neighborhood of pixel P1, respectively, and abs(*) represents the absolute value; Pi represents the pixel value of the i-th pixel.

[0040] D13: Remove small branches and loop branches from all obtained pixels with branch point morphological features;

[0041] For small branch points, the lengths of branches extending in different directions from pixels that possess branch point morphological characteristics are denoted as Len. j If j represents different directions of the branch, then points that meet the following conditions are judged as small branches, and the pixel values ​​in that direction are set to zero.

[0042] Len j <50;

[0043] For loop-shaped branch points, loops with a length of less than 10 pixels are eliminated using a small hole filling algorithm. For loop-shaped vessels with a length greater than or equal to 10 pixels and less than or equal to 30 pixels, if there are two or more branch points within a 15-pixel range, and there are two connected pathways between the existing branch points within a 15-pixel range, then the branch points that meet the above conditions are determined to be branch points within the loop-shaped vessel, and the pixel values ​​at the connection points of the loop-shaped vessel are set to zero.

[0044] D14: All pixels with branch point morphological characteristics that have had small branches and loop branches removed are used as the determined branch points;

[0045] D15: Using the identified branch points, the feature points are matched with the main branches of the coronary arteries in different angiography positions based on morphological characteristics under different angiography positions, and the branch points are finally obtained and labeled.

[0046] In step D15, for the foot-position angiography image of the left coronary artery, a rectangular coordinate system is established with the upper left corner as the origin (0,0), the positive x-direction to the right, and the positive y-direction downwards. The coordinates of the lower right corner pixel are marked as (512,512), and the main branch point corresponding to the left circumflex artery is denoted as Point. r The major branch point corresponding to the left anterior descending branch is denoted as Point. d The remaining branch points are denoted as Point. k Iterate through all Points k If Point k The following conditions must be met:

[0047]

[0048] When Point k When a point is located at the top left corner of the image and is the leftmost branch among all branch points, it is the branch point corresponding to the left rotation branch; that is, this point is Point. r ;wherein, Point kx Pointk The value of the coordinate in the x-direction, Point ky Point k The value of the coordinate in the y-direction, Pointx represents the set of x-coordinates of all branch points;

[0049] If Point k The following conditions must be met:

[0050]

[0051] That is, when Point k When a point is located at the top right corner of the image and is the highest of all branch points, it is the branch point corresponding to the left anterior descending branch; that is, this point is Point. d Pointy represents an array consisting of the y-coordinates of all branch points;

[0052] For head-view images of the left coronary artery, based on the aforementioned coordinate system, the major branch point corresponding to the left circumflex artery is denoted as Point. r The major branch point corresponding to the left anterior descending branch is denoted as Point. d The remaining branch points are denoted as Point. l For the characteristics of the branch points under the head position, traverse all Points. l If Point l The following conditions must be met:

[0053]

[0054] That is, when Point l When a point is located at the top of the image and is the rightmost branch among all branch points satisfying this condition, it is the branch point corresponding to the left rotation branch; that is, this point is Point. r ;

[0055] If Point l The following conditions must be met:

[0056]

[0057] That is, when Point l Located at the top of the image and being the only branch point other than Point that satisfies this condition. r When the rightmost branch point is selected, that point corresponds to the branch point of the left anterior descending branch, i.e., that point is Point. d ;

[0058] Similarly, for images of the left coronary artery in the spider, right shoulder, left shoulder, and liver positions, parameters were designed based on observations of the positions of the main branch points in the angiographic images, as shown in the formula below:

[0059]

[0060] Where, Point mx Point represents the x-coordinate of the main branch point corresponding to the required body position. my The ordinate represents the ordinate of the main branch point corresponding to the required body position; h and w correspond to the abscissa marker parameter and ordinate marker parameter obtained in the parameter design, respectively; f() represents the corresponding functional relationship satisfied by the branch point.

[0061] Step D2 includes the following specific steps:

[0062] D21: Based on the morphological features of the vascular skeleton image, the correspondence of the blood vessel angles of the eight neighboring pixels of a certain pixel in the image is defined as follows; let the central pixel be C, then the blood vessel angles of the eight neighboring pixels of pixel C, starting from the top left pixel and proceeding clockwise, are 225°, 180°, 135°, 90°, 45°, 360°, 315° and 270° respectively; calculate θ for each pixel of the vascular skeleton image to obtain the feature information of the blood vessel direction in the skeleton image;

[0063] For any central pixel C, let θ1 and θ2 denote the angles corresponding to the two pixels in its eight-neighborhood. Then, the angular directions within the eight-neighborhood of the central pixel C are...

[0064]

[0065] The range of θ is from 0 to 180°. Angle values ​​exceeding 180° are transformed by subtracting 180° to obtain the correct range of θ.

[0066] D22: Determine the position of the pixel based on the angular and morphological features of the isolated pixel, and update the skeleton map by pixel displacement. Then, extract the direction of the blood vessels again after the skeleton is updated.

[0067] Here, the isolated pixel is denoted as point O, and the points in its eight neighborhoods (top left, top right, bottom left, and bottom right) are denoted as points A, B, C, and D, respectively. The direction of the blood vessel at the isolated pixel is θ. Then, the isolated pixel has the following characteristics:

[0068]

[0069] The isolated pixel and its eight neighboring pixels are translated to the same straight line to complete the skeleton map update; after the skeleton is updated, the blood vessel direction is extracted again.

[0070] D23: The blood vessel diameter is obtained using a pixel thresholding method based on segmentation boundaries and blood vessel direction. This involves extending the blood vessel skeleton along the normal blood vessel direction, and statistically analyzing the length value upon reaching the blood vessel boundary to obtain the blood vessel diameter. The blood vessel direction corresponding to each pixel is denoted as θ′, and the normal blood vessel direction is denoted as β′. The formula for the blood vessel skeleton extension direction is as follows:

[0071]

[0072] D24: Optimize the blood vessel diameter obtained in step D23 to obtain the optimized blood vessel diameter;

[0073] The diameter corresponding to each blood vessel pixel is denoted as diam(p,q), where p and q are the x and y coordinates of the blood vessel pixel, respectively. The average diameter of the blood vessels corresponding to pixels within a 10*10 range is ave_diam. The value of diam(p,q) is set to zero when the following condition is met:

[0074] diam(p,q)>2*ave_diam;

[0075] D25: Based on the optimized blood vessel diameter, the upstream and downstream diameters of the blood vessel are compared to achieve stenosis detection, and the stenosis site is located and marked according to the detection results;

[0076] For regions without branch points, the diameter value of the blood vessel pixel to be detected within the region is denoted as diam, and the diameter value of the upstream point is denoted as diam. up The diameter of the downstream point is denoted as diam. down The proportion of the blood vessel at the location of the pixel to be detected is denoted as the degree. When the diameter value corresponding to a pixel is less than the upstream and downstream diameter values ​​multiplied by the degree of stenosis, the pixel is considered to have stenosis. That is, pixels in the stenosis region have the following characteristics:

[0077]

[0078] The upstream and downstream diameter values ​​were obtained using the mean value method.

[0079] The point 15 pixels upstream of the connected component of the blood vessel pixel to be detected is denoted as the upstream point of the blood vessel. up The point located 15 pixels downstream of the connected component of the blood vessel pixel to be detected is denoted as the downstream point of the blood vessel. down The average diameter of the upstream and downstream points of the blood vessel within an eight-neighborhood is denoted as ave. up and ave down Then we have:

[0080]

[0081] For regions containing branch points, a recursive method is used to select upstream and downstream points on the main branch. That is, the points in the connected region within 15 pixels of the blood vessel pixel to be detected are traversed. If a branch point exists, the branch point is used as the reference point, and the average diameter of different branches on the branch point is recorded as diam1 and diam2 respectively. Their size relationship is compared, and the branch with the larger diameter is retained as the reference branch when comparing upstream and downstream diameters.

[0082] The system based on coronary angiography image vascular branch recognition and stenosis detection method includes a keyframe image extraction module, a vascular enhancement segmentation processing module, a skeleton map generation module, and a vascular branch recognition and vascular stenosis detection module.

[0083] The keyframe image extraction module is used to acquire keyframe images from a continuous image sequence of coronary angiography.

[0084] The aforementioned blood vessel enhancement and segmentation processing module is used to perform image enhancement, image segmentation, and fusion processing on keyframe images to finally obtain an enhanced and segmented binary image; the blood vessel enhancement and segmentation processing module includes an image enhancement unit, an image segmentation unit, and a fusion unit;

[0085] The skeleton map generation module is used to generate a skeleton map of blood vessels using an enhanced segmented binary map through a morphological erosion method.

[0086] The aforementioned vascular branch recognition and vascular stenosis detection module is used to perform vascular branch recognition and vascular stenosis detection respectively through the vascular skeleton diagram, to mark the vascular branch recognition results, and to locate and mark the stenosis sites according to the vascular stenosis detection results; the vascular branch recognition and vascular stenosis detection module includes a vascular branch recognition unit and a vascular stenosis detection unit.

[0087] This invention enables rapid and accurate identification of vascular branches in coronary angiography images through the extraction of major vascular branch points. It also effectively solves the problem of coronary artery stenosis detection. By extracting the vascular skeleton and vascular edges, the stenosis value is quantified, greatly improving the accuracy of stenosis detection. It can quickly and accurately detect potentially stenotic areas in coronary angiography images, thus providing data support for medical personnel in stenosis detection. Attached Figure Description

[0088] Figure 1 This is a flowchart illustrating the method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to the present invention.

[0089] Figure 2 This is a pixel distribution diagram of the eight neighboring pixels of a pixel in this invention;

[0090] Figure 3This is a schematic diagram showing the distribution of the main branches in a mid-foot view angiography image of the left coronary artery;

[0091] Figure 4 This is a schematic diagram showing the distribution of the main branches in a cephalic angiography image of the left coronary artery.

[0092] Figure 5 This is an angular distribution diagram of the eight neighborhoods of a pixel in this invention;

[0093] Figure 6 This is a schematic diagram of the coronary angiography image vascular branch recognition and stenosis detection system of the present invention. Detailed Implementation

[0094] The present invention will now be described in detail with reference to the accompanying drawings and embodiments:

[0095] like Figure 1 As shown, the method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to the present invention includes the following steps in sequence:

[0096] A: Extract keyframe images from the acquired continuous image sequence of coronary angiography;

[0097] In this invention, in a continuous image sequence of coronary angiography, the average pixel gray value of each frame is first calculated, and then the difference between the average pixel gray values ​​of any two adjacent frames is calculated. If the total number of image sequences is odd, the pixel gray value of the middle frame is selected as the reference pixel gray value. If the total number of image sequences is even, the average pixel gray value of the two middle frames is selected as the reference pixel gray value. Then, the image in the continuous image sequence whose difference between the pixel gray value and the reference pixel gray value is within a set pixel difference threshold range and whose average difference between the pixel gray values ​​and the two adjacent frames is the smallest is selected as the keyframe image.

[0098] In this invention, the above-mentioned keyframe image selection method has strong robustness. At the same time, it combines the image features of the coronary angiography image itself. The selected keyframes have relatively full contrast agent injection, higher image quality, which is beneficial for subsequent image processing and can improve the accuracy of vascular branch and stenosis location identification.

[0099] The pixel difference threshold can be set according to the grayscale value changes of the coronary angiography image. In this embodiment, the pixel difference threshold is 2.

[0100] In this invention, continuous images of coronary angiography can be acquired using a digital subtraction angiography (DSA) machine. The DSA machine is a commonly used machine in existing medical systems for coronary angiography, and the sequence of angiographic images is acquired by a physician operating the DSA machine.

[0101] The data output by a digital subtraction angiography (DSA) machine for each patient position is a collection of consecutive frames, and the image data is stored in a DICOM format file. The overall structure of the DICOM format file is shown in Table 1:

[0102] Table 1

[0103]

[0104] In a DICOM format file, the first part is a 128-byte Introduction section, followed by a 4-byte "DICM" string, and finally several "DataElement" elements arranged in sequence until the end of the file.

[0105] The “DataElement” element is the subject that stores information about this format file, and its overall structure is shown in Table 2.

[0106] Table 2

[0107]

[0108] The “DataElement” element first contains a 4-byte Tag, which consists of a 2-byte group number and a 2-byte element number. Next is a 2-byte VR, whose value specifies the meaning and storage cell of different data types. Then comes a 2-byte Len, which records the length of the data. Finally, there is a several-byte Value, which stores the value of the data.

[0109] The value of a tag determines the type of subsequent data value, which can be mainly divided into three types: file meta tag, ordinary tag, and pixel tag. File meta tags are used to define the transmission syntax and have no actual meaning; ordinary tags record various types of common data, including image width and height, data transmission format, patient name, patient's date of birth, medical record department, and description of the condition; pixel tags store the image information of the medical record.

[0110] For the files output by the digital subtraction angiography machine in the above format, the present invention can use the dicomread function in MATLAB to process the DICOM format. This function can read the DICOM file and convert the DICOM format file into .mat format data, which facilitates the subsequent extraction of keyframe images.

[0111] B: Perform vessel enhancement segmentation processing on the extracted keyframe images. The vessel enhancement segmentation processing includes image enhancement, image segmentation, and fusion processing of the segmentation results, and finally obtains the enhanced segmented binary image.

[0112] In this invention, a multi-scale image enhancement method based on the Frangi blood vessel similarity function and with angle improvement is used to enhance the image. Then, a threshold segmentation method based on maximum inter-class variance is used for image segmentation. Finally, a region growing method based on geodesic dilatation is used to fuse the segmentation results, and the enhanced segmented binary image is obtained.

[0113] Step B includes the following specific steps:

[0114] B1: Based on the Frangi blood vessel similarity function, an angle-modified blood vessel similarity function is used for image enhancement to obtain the enhanced keyframe image;

[0115] The Frangi vessel similarity function is a multi-scale image enhancement method based on the Hessian matrix. By filtering the image data, it enhances the linear structure of blood vessels to obtain a higher quality image. The Frangi vessel similarity function method is shown below:

[0116]

[0117]

[0118]

[0119] Where v(λ1,λ2) represents a measure of the degree of blood vessel similarity, λ1 represents the eigenvalues ​​of the Hessian matrix along the blood vessel axis, λ2 represents the eigenvalues ​​of the Hessian matrix perpendicular to the blood vessel axis, and R... B It is the ratio of λ1 to λ2. When in a linear structure, R B Approaching 0 makes the response value approach 1; when in a patchy structure, R B The value of is very large, making the response value close to 0. Therefore, the response value is close to 1 only when the pixel is located in a linear structure, thereby achieving the enhancement of blood vessels; α is the blood vessel structure discrimination parameter, used to control the discrimination of blood vessel structures; β is the noise suppression sensitivity parameter, used to control the sensitivity of noise suppression; S is the square root of the sum of squares of the feature values; in this embodiment, the value of α is 450 and the value of β is 0.5.

[0120] Although the existing Frangi vessel similarity function can achieve image enhancement, it mainly focuses on the magnitude relationship of feature values ​​and does not involve the direction of feature vectors. However, the direction of feature vectors has a stronger geometric meaning and helps to make more accurate discrimination of linear structures. Therefore, this invention uses the following angle-improved vessel similarity function based on the Frangi vessel similarity function to enhance the image of key frames.

[0121]

[0122]

[0123]

[0124] Where v(λ1,λ2) represents the measure of blood vessel similarity, γ is a noise reduction parameter used for further noise reduction. Since the absolute values ​​of the feature values ​​in the noise region are all small, the response value approaches 0, thus better suppressing the noise and compensating for the limitations of single parameter adjustment; R C The arctangent function of the ratio of λ1 to λ2 represents the angle between the vector corresponding to the end of the eigenvector corresponding to eigenvalue λ2 pointing to the end of the eigenvector corresponding to eigenvalue λ1, and the eigenvector corresponding to eigenvalue λ1. When in a linear structure, the response value approaches 1, while when in a patchy structure, the response value approaches 0; from R... B To R C The improvements make the function's response to linear structures more accurate.

[0125] In this embodiment, the value of α is 450, the value of β is 0.5, and the value of γ is 1.

[0126] B2: Perform image segmentation on the enhanced keyframe images;

[0127] The chosen segmentation method is a thresholding method based on maximum inter-class variance (MOV). This method is an adaptive approach based on a global threshold among traditional thresholding methods, also known as the OTUS algorithm. It uses MOV as the criterion to determine the optimal segmentation threshold, achieving fast image segmentation, minimal susceptibility to noise, and high accuracy.

[0128] B3: For the keyframe images after image segmentation, a region growing method based on geodesic dilatation is used to fuse the segmentation results to obtain an enhanced binary image after segmentation.

[0129] In this embodiment, a region growing algorithm is used to fuse the segmentation results in order to further supplement the vascular details;

[0130] The idea behind region growing is to group pixels with similar properties together to form a whole region. By continuously growing the seed region with reference to the mask image, the points within the region are merged into a whole.

[0131] The mask image is obtained by thresholding the image enhancement result. Since the mask image needs to encompass the details of blood vessels, a threshold of 4 / 255 is selected in this embodiment, which yields a good mask image.

[0132] The region growing algorithm fuses the segmentation results. The seed region is selected directly from the OTUs segmentation result in step B2, and the mask image is selected according to the parameters set above. The growth process is based on the difference between the seed region and the mask image, that is, the difference between the grayscale images is the criterion for seed region growth. The stopping condition for region growth is the zero difference condition, that is, if the image before and after a certain growth does not change, the region growth stops.

[0133] C: Using the enhanced segmented binary image obtained in step B, a morphological erosion method is used to obtain the vascular skeleton map;

[0134] Step C includes the following specific steps:

[0135] C1: Perform image thinning on the enhanced segmented binary image to obtain a thinned binary image;

[0136] Skeleton extraction is similar to image thinning; both involve removing a subset of points from a binary image while preserving the original shape of the remaining points—that is, the image skeleton is well-preserved. This embodiment uses existing image thinning methods to obtain the thinned binary image. The thinning algorithm proposed by Zhang-Suen in 1984 can be used, which not only maintains image connectivity but also offers fast processing speed.

[0137] C2: Denoise the thinned binary image to obtain a denoised thinned binary image, and extract the skeleton image of the blood vessels through skeleton extraction;

[0138] When extracting the skeleton from the thinned binary image, noise interference such as small branches and small spots in the thinned binary image was found, which seriously affected the subsequent extraction of branch points. Therefore, in this invention, the thinned binary image is denoised to ensure the accuracy of subsequent branch point extraction.

[0139] In this example, the area of ​​isolated connected components in the refined binary graph is denoted as S. z z represents the label of the connected component; the length of the branch in the thinned binary image is denoted as l. The image is denoised according to the following rules to obtain the denoised thinned binary image, and the skeleton image of the blood vessel is obtained by extracting the skeleton.

[0140] If S zIf the number of pixels is less than 20, then set all pixels in the isolated connected component to zero and change the length of all branches to l-3.

[0141] If S z If the number of pixels is ≥20, then no processing will be performed.

[0142] The above denoising method can remove small branches and spots from the refined binary image, making the main trunk of the blood vessel clearer and laying the groundwork for the next step of finding branch points.

[0143] D: Use the vascular skeleton map to identify vascular branches and detect vascular stenosis. Mark the vascular branches according to the results of vascular branch identification, and locate and mark the stenosis sites according to the results of vascular stenosis detection.

[0144] Step D includes the following specific steps:

[0145] D1: Using the obtained vascular skeleton map, branch points are found based on the morphological features of the skeleton using a branch-finding algorithm to obtain all pixels with branch point morphological features; after removing small branches and ring-shaped branch points, the branch points are obtained and labeled.

[0146] D2: Using the enhanced segmented binary image, the direction of blood vessels is extracted by the eight-neighbor definition method based on the vascular skeleton, and the diameter of blood vessels is obtained by the pixel thresholding method based on the segmentation boundary and the direction of blood vessels. Finally, the vascular stenosis is detected by comparing the upstream and downstream diameters of blood vessels, and the stenosis is located and labeled according to the detection results.

[0147] Step D1 includes the following specific steps:

[0148] D11: Based on the morphological features of the skeleton map, define the eight neighboring pixels of a pixel in the image as follows: Figure 2 As shown; define the central pixel as P1, then the eight neighboring pixels of pixel P1, starting from the top left pixel and proceeding clockwise, are P9, P2, P3, ..., P7 and P8.

[0149] D12: Using a branch-finding algorithm, all pixels in the vascular skeleton diagram are judged sequentially according to the set branch point judgment rules to obtain all pixels with branch point morphological features.

[0150] The branch point determination rule is as follows: the sum of the pixel value and its eight neighboring pixel values ​​is 4, and the sum of the absolute values ​​of the alternating differences of the outer ring pixel values ​​is 12. Therefore, the pixel possesses the branch point morphological characteristics. In other words, by traversing the pixels in the blood vessel skeleton diagram, pixels that satisfy the following mathematical relationship are considered to possess the branch point morphological characteristics:

[0151]

[0152] Where P2 to P9 represent the pixel values ​​of pixels P2 to P9 in the eight neighborhood of pixel P1, respectively, and abs(*) represents the absolute value; Pi represents the pixel value of the i-th pixel.

[0153] D13: Remove small branches and loop branches from all obtained pixels with branch point morphological features;

[0154] The branch-finding algorithm determines the positions of all pixels with branch point morphological features. However, due to the complexity of blood vessel morphology, the number of pixels with branch point morphological features is enormous, including many small branches or loop-shaped branches, which interfere with the selection of the main branch points. The interference from loop-shaped branches and small branches increases the complexity of finding the main branch points. Therefore, it is necessary to remove small branches and loop-shaped branches to improve the efficiency of finding the main branch points.

[0155] For small branch points: In this invention, the branch lengths extending in different directions at pixels with branch point morphological characteristics are denoted as Len. j If j represents different directions of the branch, then points that meet the following conditions are judged as small branches, and the pixel values ​​in that direction are set to zero.

[0156] Len j <50;

[0157] If the length of a branch in a certain direction is less than 50 pixels, then such a branch point is divided into a small branch, and the pixel value in that direction is set to zero, thereby eliminating the small branches in the skeleton diagram.

[0158] For circular branch points:

[0159] To address the issue of pixels potentially forming closed loops in the skeleton diagram, this invention employs a connectivity determination method to avoid this problem.

[0160] The removal of loop branch points targets loop blood vessels with a loop length between 10 and 30 pixels. Loop blood vessels with a loop length greater than 30 pixels have a negligible impact on branch point finding. For small loops with a loop length less than 10 pixels, the small hole filling algorithm can be used to eliminate them. Loop blood vessels with a loop length greater than or equal to 10 pixels and less than or equal to 30 pixels are judged and processed according to the following method:

[0161] If there are two or more branch points within a 15-pixel range, and there are two connected pathways between the branch points within a 15-pixel range, then the branch points that meet the above conditions are determined to be branch points within the annular blood vessel. The pixel values ​​at the connection points of the annular blood vessel are then set to zero, which means that the annular blood vessel is destroyed into a linear blood vessel, thereby reducing the influence of the annular blood vessel.

[0162] D14: All pixels with branch point morphological characteristics that have had small branches and loop branches removed are used as the determined branch points;

[0163] D15: Using the determined branch points, the feature points are matched with the main branches of the coronary arteries in different angiography positions based on the morphological features under different angiography positions, and the branch points are finally obtained and labeled.

[0164] Common angiographic positions are summarized as follows: there are six common angiographic positions for the left coronary artery: mid-foot view, head view, spider view, right shoulder view, left shoulder view, and liver view; there are two common positions for the right coronary artery: left anterior oblique view and right anterior oblique view. The main branches of the right coronary artery are relatively simple, while those of the left coronary artery are more complex. Therefore, the diagnosis of the left coronary artery is significantly affected by the complexity of its vascular branches, posing a greater challenge to physicians and making it a key area for auxiliary diagnosis. Therefore, this embodiment uses mid-foot and head views of the left coronary artery as examples for detailed explanation.

[0165] The distribution of major branches in the left coronary artery mid-foot and head-up angiography images are as follows: Figure 3 and Figure 4 As shown.

[0166] For the foot position image, the main branch points are located on the left and right sides of the upper half of the image, exhibiting obvious morphological features. A Cartesian coordinate system is established with the upper left corner as the origin (0,0), with the positive x-direction to the right and the positive y-direction downwards. The coordinates of the lower right pixel are denoted as (512,512). The main branch point corresponding to the left gyral branch 1 of the foot position image is denoted as Point. r The main branch point corresponding to the left anterior descending branch 2 in the foot position image is denoted as Point. d The remaining branch points are denoted as Point. k The characteristics of the branch points in foot position can be summarized into the following mathematical expression:

[0167] Traverse all Points k If Point k The following conditions must be met:

[0168]

[0169] Point kx Point k The value of the coordinate in the x-direction, Point ky Point k The value of the coordinate in the y-direction, i.e., when Point k When a point is located at the top left corner of the image and is the leftmost branch among all branch points, it corresponds to the left rotation branch of the foot position image; that is, this point is Point. r Pointx represents the set of x-coordinates of all branch points;

[0170] Similarly, if Point k The following conditions must be met:

[0171]

[0172] That is, when Point k When a point is located in the upper right corner of the image and is the highest of all branch points, it corresponds to the left anterior descending branch of the foot position image; that is, this point is Point. d Pointy represents an array consisting of the y-coordinates of all branch points;

[0173] Using the coordinate system established above, for head-view images, the distribution of major branch points is relatively concentrated in the x-direction, while exhibiting significant elevation differences from other points in the y-direction, demonstrating prominent morphological features. Similarly, the major branch point corresponding to the left gyral branch 4 in the head-view image is denoted as Point. r The main branch point corresponding to the left anterior descending branch 3 in the head position image is denoted as Point. d The remaining branch points are denoted as Point. l The characteristics of the branch points in the head position can be summarized into the following mathematical expression:

[0174] Traverse all Points l If Point l The following conditions must be met:

[0175]

[0176] That is, when Point l When a point is located at the top of the image and is the rightmost branch among all branches that satisfy this condition, it is the branch corresponding to the left rotation branch of the head-view image; that is, this point is Point. r ;

[0177] Similarly, if Point l The following conditions must be met:

[0178]

[0179] That is, when Point l Located at the top of the image and being the only branch point other than Point that satisfies this condition. r When the rightmost branch point is selected, this point corresponds to the branch point of the left anterior descending branch in the head-view image; that is, this point is Point. d ;

[0180] The branching of the spider, right shoulder, left shoulder, liver, and left anterior oblique views is determined using the same method. Parameters are designed by observing the positions of the main branching points in the contrast images, as shown in the formula below:

[0181]

[0182] Where Point mx Point represents the x-coordinate of the main branch point corresponding to the required body position. my The ordinate represents the ordinate of the main branch point corresponding to the required body position; h and w correspond to the abscissa marker parameter and ordinate marker parameter obtained in the parameter design, respectively; f() represents the corresponding functional relationship satisfied by the branch point; Pointx represents the set of abscissas of all branch points; and Pointy represents the array of ordinates of all branch points.

[0183] That is, when the main branch points meet the formula under the above parameter design, the corresponding main branch points can be screened out, thereby realizing the identification of the main branch points of blood vessels.

[0184] Once a branch point is determined, the branch below that point is the corresponding main branch.

[0185] Step D2 includes the following specific steps:

[0186] D21: Based on the morphological features of the skeleton map, define the correspondence of blood vessel angles among the eight neighboring pixels of a given pixel in the image, as follows: Figure 5 As shown; define the central pixel as C, then among the eight neighboring pixels of pixel C, starting from the top left pixel, the angles of the blood vessels in a clockwise direction are 225°, 180°, 135°, 90°, 45°, 360°, 315° and 270° respectively.

[0187] For a complete vascular skeleton image, all pixels are divided into three categories: edge points, branch points, and midpoints.

[0188] Within the eight neighboring regions, they each have distinct characteristics: edge points have pixels in only one direction, located at the edge of the blood vessel, and are single pixels with no stenosis; branch points have pixels in three directions, located at the intersection of blood vessels, and have almost no stenosis; ordinary points have pixels in two directions, located in the middle region of blood vessel branches, where stenosis mainly occurs.

[0189] Therefore, the extraction of blood vessel direction mainly focuses on the midpoint, and edge points and branch points are not counted, and their blood vessel direction is set to zero.

[0190] That is, for any intermediate point C, the angles corresponding to the two pixels in its eight-neighborhood are denoted as θ1 and θ2, respectively. The angular directions in the eight-neighborhood of this point are defined as follows.

[0191]

[0192] The range of θ is from 0 to 180°. Angle values ​​exceeding 180° need to be transformed by subtracting 180° to obtain the correct range of θ.

[0193] By calculating θ for each pixel of the vascular skeleton image, feature information about the direction of blood vessels in the skeleton image can be obtained.

[0194] D22: The position of a pixel is determined based on the angular and morphological features of the isolated pixel, and the skeleton map is updated by pixel displacement. After the skeleton is updated, the direction of the blood vessels is extracted again.

[0195] Because the process of vascular skeletonization is a continuous erosion process, the result of erosion can lead to isolated pixels, causing the skeleton to not accurately reflect the direction of the blood vessels. When the position of an isolated pixel deviates from the overall direction of the blood vessels, it will have a certain impact on the extraction of the blood vessel orientation, resulting in a deviation in the blood vessel orientation between two pixels in the neighborhood of the isolated point.

[0196] Therefore, in the case of isolated pixels, this invention determines the position of the isolated pixel based on its angular and morphological features, and updates the skeleton map by pixel displacement.

[0197] Let the isolated pixel be denoted as point O, and let the points in its eight neighborhoods (top left, top right, bottom left, and bottom right) be denoted as points A, B, C, and D respectively. Let the direction of the blood vessel at the isolated pixel be θ. Figure 5 This diagram illustrates the pixel distribution of an isolated pixel within its eight neighborhoods.

[0198] Isolated pixels then possess the following characteristics:

[0199]

[0200] Based on the distribution of pixels within their eight neighborhoods, isolated pixels are translated to the same straight line as other pixels within their eight neighborhoods, thus completing the skeleton map update. After the skeleton is updated, the blood vessel direction is extracted again to prepare for the subsequent extraction of blood vessel diameter.

[0201] D23: The diameter of blood vessels is obtained by using a pixel thresholding method based on segmentation boundaries and blood vessel orientation;

[0202] Obtaining the diameter involves extending the vascular skeleton along the normal direction of the blood vessel and then calculating the length value after reaching the blood vessel boundary.

[0203] Let θ′ be the direction of the blood vessel corresponding to the pixel, and β′ be the normal direction of the blood vessel. Then, the following formula is satisfied for the extension direction of the blood vessel skeleton:

[0204]

[0205] Extending along the above direction, the diameter value can be initially obtained. The diameter value obtained at this time is basically accurate, but there is a large deviation at the junction of blood vessel branches, which will interfere with the subsequent comparison of upstream and downstream diameters. Therefore, the diameter value at the junction of blood vessel branches is also optimized in this invention.

[0206] D24: Optimize the blood vessel diameter obtained in step D23 to obtain the optimized blood vessel diameter;

[0207] Because the flow rate of blood vessels at branch junctions is high, their diameter can easily be overestimated, which may lead to the vessels near the branch junctions being identified as narrowed. Therefore, in this application, the diameter of blood vessels at branch junctions is further optimized according to the following method.

[0208] The diameter corresponding to each blood vessel pixel is denoted as diam(p,q), where p and q are the x and y coordinates of the blood vessel pixel, respectively. The average diameter of the blood vessels corresponding to pixels within a 10*10 range is ave_diam. The value of diam(p,q) is set to zero when the following condition is met:

[0209] diam(p,q)>2*ave_diam;

[0210] That is, when the diameter value is abnormally large, the diameter value is set to 0 and is not included in the upstream and downstream diameter comparison, so as not to affect the subsequent experimental results.

[0211] D25: Based on the optimized blood vessel diameter, the upstream and downstream diameters of the blood vessel are compared to achieve stenosis detection, and the stenosis site is located and marked according to the detection results;

[0212] The task of stenosis detection is completed by comparing the upstream and downstream diameters of the blood vessel, which means using a quantitative method to identify the location that is smaller than the diameter of the surrounding blood vessels.

[0213] The difficulty of comparing upstream and downstream diameters varies depending on the location of the blood vessel. Blood vessels within a region consisting entirely of midpoints experience less interference and are simpler to compare. However, blood vessels with branching points within the region present the challenge of selecting the main branches for comparison.

[0214] In this invention, for regions without branch points, the diameter value of the blood vessel pixel to be detected within the region is denoted as diam, and the diameter value of the upstream point is denoted as diam. up The diameter of the downstream point is denoted as diam. down The proportion of blood vessels at the location of the pixel of the blood vessel to be detected is denoted as degree;

[0215] Narrow region pixels have the following characteristics:

[0216]

[0217] That is, when the diameter value corresponding to a pixel is less than the diameter values ​​of the upstream and downstream points multiplied by the narrowing degree (1-degree), the pixel is considered to be narrow. Therefore, the key problem to be solved is how to obtain the upstream and downstream diameter values ​​to ensure good generalization. This invention uses the mean value method to obtain the upstream and downstream diameter values:

[0218] The point 15 pixels upstream of the connected component of the blood vessel pixel to be detected is denoted as the upstream point of the blood vessel. up The point located 15 pixels downstream of the connected component of the blood vessel pixel to be detected is denoted as the downstream point of the blood vessel. down The average diameter of the upstream and downstream points of the blood vessel within an eight-neighborhood is denoted as ave. up and ave down Then we have:

[0219]

[0220] This allows for a comparison of the upstream and downstream diameters of pixels in this type of region.

[0221] For regions containing branch points, the following complex issues arise when comparing upstream and downstream diameters: When searching for upstream and downstream points, there are multiple branches. If upstream and downstream points are selected in the direction of a small branch, narrow parts may be missed, which greatly hinders diagnosis.

[0222] In this invention, for regions containing branch points, a method of recursively selecting upstream and downstream points on the main branch is adopted to solve the above-mentioned missed detection problem.

[0223] The points within a 15-pixel range of the blood vessel pixel to be detected are traversed. If there are branch points, the branch points are used as reference points. The average diameters of different branches on the branch points are denoted as diam1 and diam2, respectively. Their size relationship is compared, and the branch with the larger diameter is retained as the reference branch when comparing the upstream and downstream diameters.

[0224] This method allows for a comprehensive assessment of narrow areas within the branch point region, and enables the location and marking of narrow sections based on the detection results.

[0225] like Figure 6 As shown, the coronary angiography image vascular branch recognition and stenosis detection system of the present invention includes a keyframe image extraction module, a vascular enhancement segmentation processing module, a skeleton map generation module, and a vascular branch recognition and vascular stenosis detection module.

[0226] The keyframe image extraction module is used to acquire keyframe images from a continuous image sequence of coronary angiography.

[0227] The extraction process of the keyframe image extraction module is described in detail in step A above, and will not be repeated here;

[0228] The aforementioned blood vessel enhancement and segmentation processing module is used to perform image enhancement, image segmentation, and fusion processing on keyframe images to finally obtain an enhanced and segmented binary image; the blood vessel enhancement and segmentation processing module includes an image enhancement unit, an image segmentation unit, and a fusion unit;

[0229] The processing procedure of the blood vessel enhancement segmentation module is described in detail in step B and steps B1 to B3 above, and will not be repeated here.

[0230] The skeleton map generation module is used to generate a skeleton map of blood vessels using an enhanced segmented binary map through a morphological erosion method.

[0231] The process of generating the skeleton diagram is described in detail in step C and steps C1 to C2 above, and will not be repeated here.

[0232] The aforementioned vascular branch recognition and vascular stenosis detection module is used to perform vascular branch recognition and vascular stenosis detection respectively through the vascular skeleton diagram, to mark the vascular branch recognition results, and to locate and mark the stenosis sites according to the vascular stenosis detection results; the vascular branch recognition and vascular stenosis detection module includes a vascular branch recognition unit and a vascular stenosis detection unit.

[0233] The identification and detection process of the vascular branch identification and vascular stenosis detection module is described in detail in steps D, D1, D11 to D16, D2 and D21 to D25 above, and will not be repeated here.

Claims

1. A method for identifying blood vessel branches and detecting stenosis in coronary angiography images, characterized in that, The steps are as follows: A: Extract keyframe images from the acquired continuous image sequence of coronary angiography; B: Perform vessel enhancement segmentation processing on the extracted keyframe images. The vessel enhancement segmentation processing includes image enhancement, image segmentation, and fusion processing of the segmentation results, and finally obtains the enhanced segmented binary image. C: Using the enhanced segmented binary image obtained in step B, a morphological erosion method is used to obtain the vascular skeleton map; D: Use the vascular skeleton map to identify vascular branches and detect vascular stenosis. Mark the vascular branches according to the results of vascular branch identification, and locate and mark the stenosis sites according to the results of vascular stenosis detection. For regions without branch points, the diameter value of the blood vessel pixel to be detected within the region is denoted as... The diameter of the upstream point is denoted as The diameter of the downstream point is denoted as The proportion of blood vessels at the location of the pixel to be detected is denoted as When the diameter value corresponding to a pixel is less than the diameter values ​​of the upstream and downstream nodes multiplied by the degree of narrowness, the pixel is considered to be narrow; that is, pixels in narrow regions have the following characteristics: The upstream and downstream diameter values ​​were obtained using the mean value method. Points located 15 pixels upstream of the connected component of the blood vessel pixel to be detected are denoted as upstream points of the blood vessel. Points located 15 pixels downstream of the connected component of the blood vessel pixel to be detected are denoted as downstream points of the blood vessel. The average diameter of the upstream and downstream points of the blood vessel within an eight-neighborhood is denoted as... and Then we have: For regions containing branch points, a recursive method is used to select upstream and downstream points on the main branch. This involves traversing the connected components within a 15-pixel range of the blood vessel pixel to be detected. If a branch point exists, it is used as the reference point, and the average diameter of different branches at that branch point is denoted as... By comparing their size relationships, the branch with the larger diameter is retained as the reference branch when comparing the diameters of the upstream and downstream branches.

2. The method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to claim 1, characterized in that: In step A, in the continuous image sequence of coronary angiography, the average pixel gray value of each frame is first calculated, and then the difference between the average pixel gray values ​​of any two adjacent frames is calculated. If the total number of image sequences is odd, the pixel gray value of the middle frame is selected as the reference pixel gray value. If the total number of image sequences is even, the average pixel gray value of the two middle frames is selected as the reference pixel gray value. Then, the image in the continuous image sequence whose difference between the pixel gray value and the reference pixel gray value is within the set pixel difference threshold range and whose average difference between the pixel gray values ​​and the two adjacent frames is the smallest is selected as the keyframe image.

3. The method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to claim 1, characterized in that, Step B includes the following specific steps: B1: Based on the Frangi blood vessel similarity function, an angle-modified blood vessel similarity function is used for image enhancement to obtain the enhanced keyframe image; B2: Perform image segmentation on the enhanced keyframe images; B3: For the keyframe images after image segmentation, a region growing method based on geodesic dilatation is used to fuse the segmentation results to obtain an enhanced binary image.

4. The method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to claim 3, characterized in that, In step B1, the following angle-modified blood vessel similarity function is used for keyframe image enhancement: ； in, A measure of the degree of similarity in blood vessels. This represents the eigenvalues ​​corresponding to the eigenvectors of the Hessian matrix along the vessel axis. This represents the eigenvalues ​​of the Hessian matrix corresponding to the eigenvectors perpendicular to the vessel axis. To and The ratio takes the arctangent function. It is a noise suppression sensitivity parameter. It is the square root of the sum of squares of the eigenvalues. Parameters for identifying vascular structures. To eliminate noise parameters.

5. The method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to claim 1, characterized in that, Step C includes the following specific steps: C1: Perform image thinning on the enhanced segmented binary image to obtain a thinned binary image; C2: The area of ​​isolated connected regions in the refined binary graph is denoted as... , The label represents the connected component; the length of the branch in the refined binary graph is denoted as... The image is denoised according to the following rules to obtain a denoised and thinned binary image, and the skeleton image of the blood vessels is obtained by extracting the skeleton. if If there are 1 pixel, then set all pixels in the isolated connected component to zero, and change the length of all branches to 1. .

6. The method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to claim 1, characterized in that, Step D includes the following specific steps: D1: Using the obtained vascular skeleton map, branch points are found based on the morphological features of the skeleton using a branch-finding algorithm to obtain all pixels with branch point morphological features. After removing small branches and circular branch points, the branch points are obtained and labeled; D2: Using the enhanced segmented binary image, the direction of blood vessels is extracted by the eight-neighbor definition method based on the vascular skeleton, and the diameter of blood vessels is obtained by the pixel thresholding method based on the segmentation boundary and the direction of blood vessels. Finally, the vascular stenosis is detected by comparing the upstream and downstream diameters of blood vessels, and the stenosis sites are located and labeled according to the detection results.

7. The method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to claim 6, characterized in that, Step D1 includes the following specific steps: D11: Based on the morphological features of the skeleton diagram, define the center pixel among the eight neighboring pixels of a given pixel in the image as... Then the pixel In the eight neighboring pixels, starting from the top left pixel, in a clockwise direction, they are: , , … and ; D12: Using a branch-finding algorithm, all pixels in the vascular skeleton diagram are judged sequentially according to the set branch point judgment rules to obtain all pixels with branch point morphological features. The rule for determining branch points is as follows: if the sum of the values ​​of a pixel and its eight neighboring pixels is 4, and the sum of the absolute values ​​of the alternating differences of the outermost pixel values ​​of the pixel is 12, then the pixel possesses the morphological characteristics of a branch point. In other words, by traversing the pixels in the vascular skeleton diagram, pixels that satisfy the following mathematical relationship are considered to possess the morphological characteristics of a branch point: in, to Representing pixels pixels in the eight neighborhood to pixel values, This indicates taking the absolute value; Indicates the first The pixel value of each pixel; D13: Remove small branches and loop branches from all obtained pixels with branch point morphological features; For small branch points, the branch lengths extending in different directions from pixels that possess branch point morphological characteristics are denoted as... , Representing different directions of the branch, points that meet the following conditions are judged as small branches, and the pixel values ​​in that direction are set to zero. For loop-shaped branch points, loops with a length of less than 10 pixels are eliminated using a small hole filling algorithm. For loop-shaped vessels with a length greater than or equal to 10 pixels and less than or equal to 30 pixels, if there are two or more branch points within a 15-pixel range, and there are two connected pathways between the existing branch points within a 15-pixel range, then the branch points that meet the above conditions are determined to be branch points within the loop-shaped vessel, and the pixel values ​​at the connection points of the loop-shaped vessel are set to zero. D14: All pixels with branch point morphological characteristics that have had small branches and loop branches removed are used as the determined branch points; D15: Using the identified branch points, the feature points are matched with the main branches of the coronary arteries in different angiography positions based on morphological characteristics under different angiography positions, and the branch points are finally obtained and labeled.

8. The method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to claim 7, characterized in that: In step D15, for the foot-position angiographic image of the left coronary artery, a rectangular coordinate system is established with the upper left corner as the origin (0,0), the positive x-direction to the right, and the positive y-direction downwards. The coordinates of the lower right pixel are marked as (512,512), and the main branch point corresponding to the left circumflex artery is denoted as... The major branch point corresponding to the left anterior descending branch is denoted as The remaining branch points are denoted as traverse all ,if The following conditions must be met: ； when When a point is located at the top left corner of the image and is the leftmost branch among all branch points, it is the branch point corresponding to the left rotation branch, i.e., this point is... ;in, express The value of the coordinate in the x-direction, express The value of the coordinate in the y-direction, This represents the set of x-coordinates of all branch points; if The following conditions must be met: That is when When a point is located at the top right corner of the image and is the highest of all branch points, it is the branch point corresponding to the left anterior descending branch, i.e., this point is... ; This represents an array consisting of the y-coordinates of all branch points; For head-down images of the left coronary artery, based on the aforementioned coordinate system, the major branch points corresponding to the left circumflex artery are denoted as... The major branch point corresponding to the left anterior descending branch is denoted as The remaining branch points are denoted as For the characteristics of the branch points under the head position, traverse all... ,if The following conditions must be met: That is when When a point is located at the top of the image and is the rightmost of all branch points satisfying this condition, it is the branch point corresponding to the left rotation branch, i.e., this point is... ; if The following conditions must be met: ； That is when Located at the top of the image and being the only branch point among all branches that satisfy this condition. When the rightmost branch point is reached, that point corresponds to the branch point of the left anterior descending branch, i.e., that point is... ; Similarly, for images of the left coronary artery in the spider, right shoulder, left shoulder, and liver positions, parameters were designed based on observations of the positions of the main branch points in the angiographic images, as shown in the formula below: ； in, The x-coordinate represents the main branch point corresponding to the required body position. The ordinate of the main branch point corresponding to the required body position; and These correspond to the horizontal and vertical axis marker parameters obtained in the parameter design, respectively. This represents the functional relationship satisfied by the branch point.

9. The method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to claim 7, characterized in that, Step D2 includes the following specific steps: D21: Based on the morphological features of the skeleton map, the correspondence of blood vessel angles among the eight neighboring pixels of a certain pixel in the image is defined as follows; let the central pixel be... Then the pixel In the eight neighboring pixels, starting from the top left pixel, the angles of the blood vessels are arranged clockwise as follows: , , , , , , and ; Calculate for each pixel of the vascular skeleton map To obtain feature information about the direction of blood vessels in the skeleton diagram; For any centered pixel The angles corresponding to two pixels within the eight-neighborhood are denoted as follows: The centered pixel The angular direction within the eight neighboring regions is, in, The range is ,Exceed The angle value is obtained by subtracting The transformation yields the correct range. ; D22: Determine the position of the pixel based on the angular and morphological features of the isolated pixel, and update the skeleton map by pixel displacement. Then, extract the direction of the blood vessels again after the skeleton is updated. Here, the isolated pixel is denoted as point O, and the points in its eight neighborhoods (top left, top right, bottom left, and bottom right) are denoted as points A, B, C, and D, respectively. The direction of the blood vessel at the isolated pixel is... An isolated pixel has the following characteristics: The isolated pixel and its eight neighboring pixels are translated to the same straight line to complete the skeleton map update; after the skeleton is updated, the blood vessel direction is extracted again. D23: A pixel thresholding method based on segmentation boundaries and vessel direction is used to obtain the vessel diameter. This involves extending the vessel skeleton along the vessel direction, and statistically analyzing the length value upon reaching the vessel boundary to obtain the vessel diameter. The vessel direction corresponding to each pixel is denoted as... The direction of the blood vessel method is recorded as Then, the following formula applies to the direction of vascular skeleton extension: D24: Optimize the blood vessel diameter obtained in step D23 to obtain the optimized blood vessel diameter; The diameter corresponding to the blood vessel pixel is denoted as , and These are the x and y coordinates of the blood vessel pixels, respectively. The average diameter of the blood vessels corresponding to the pixels within the range is When the following conditions are met, Set the value to zero: D25: Based on the optimized blood vessel diameter, the upstream and downstream diameters of the blood vessel are compared to achieve stenosis detection, and the stenosis site is located and marked according to the detection results.

10. A system based on the method for identifying blood vessel branches and detecting stenosis in coronary angiography images according to any one of claims 1 to 9, characterized in that: It includes a keyframe image extraction module, a blood vessel enhancement and segmentation processing module, a skeleton map generation module, and a blood vessel branch recognition and blood vessel stenosis detection module; The keyframe image extraction module is used to acquire keyframe images from a continuous image sequence of coronary angiography. The aforementioned blood vessel enhancement and segmentation processing module is used to perform image enhancement, image segmentation, and fusion processing on keyframe images to finally obtain an enhanced and segmented binary image; the blood vessel enhancement and segmentation processing module includes an image enhancement unit, an image segmentation unit, and a fusion unit; The skeleton map generation module is used to generate a skeleton map of blood vessels using an enhanced segmented binary map through a morphological erosion method. The aforementioned vascular branch recognition and vascular stenosis detection module is used to perform vascular branch recognition and vascular stenosis detection respectively through the vascular skeleton diagram, to mark the vascular branch recognition results, and to locate and mark the stenosis sites according to the vascular stenosis detection results; the vascular branch recognition and vascular stenosis detection module includes a vascular branch recognition unit and a vascular stenosis detection unit.

Images