A method, apparatus, device and medium for vascular segmentation of coronary angiography images

Abstract

The present application provides a method, apparatus, device and medium for vascular segmentation of coronary angiography images. The method includes: inputting the original coronary angiography image into a segmentation model to obtain a binary segmentation image and multiple feature maps; selecting a plurality of reference points based on the vascular skeleton, and determining a central selection block corresponding to each reference point based on the multiple feature maps; for each target point, splicing the central selection block corresponding to the target point and the central selection block corresponding to the target auxiliary point to obtain the spliced input data corresponding to the target point; inputting the spliced input data corresponding to the target point into a regression network model to obtain the vascular boundary corresponding to the target point; mapping the vascular boundary corresponding to each target point into the binary segmentation image to obtain the vascular segmentation result corresponding to the binary segmentation image. According to the method and apparatus, the accuracy and robustness of vascular segmentation of coronary angiography images are improved.

Description

Technical Field

[0001] The present application relates to the technical field of medical image processing, and particularly relates to a method, device, equipment and medium for vascular segmentation of coronary angiography images. Background Technique

[0002] In the field of modern medical technology, vascular image technology can help doctors understand the vascular status of patients, and angiography technology for imaging tubular structures such as blood vessels has been widely used. Therefore, a vascular image segmentation method with high accuracy becomes particularly important.

[0003] Vascular segmentation in coronary angiography is a challenging task. Coronary angiography images are interfered by various factors, such as noise, scattering, calcified plaques, etc. These factors will lead to a decline in image quality, making the vascular boundaries blurred or unclear, increasing the difficulty of segmentation. Moreover, coronary blood vessels have diverse morphologies, including tortuosity, branching, stenosis, etc. This makes it difficult to determine the boundaries of blood vessels. Therefore, how to improve the accuracy of vascular segmentation in coronary angiography has become an urgent technical problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of the present application is to provide a method, device, equipment and medium for vascular segmentation of coronary angiography images, which use the vascular boundaries predicted by a regression network model to correct the binary segmentation image output by the segmentation model, obtain a more accurate vascular segmentation result, can accurately segment blood vessels of different thicknesses in coronary angiography, fully consider the characteristics of vascular stenosis, fully grasp the details of each position of the blood vessel, and improve the accuracy and robustness of vascular segmentation for coronary angiography images.

[0005] In a first aspect, an embodiment of the present application provides a method for vascular segmentation of coronary angiography images, and the vascular segmentation method includes:

[0006] Obtain the original coronary angiography image, and input the original coronary angiography image into a pre-trained segmentation model to obtain a binary segmentation image and multiple feature maps output by different network layers of the segmentation model;

[0007] Extract the vascular skeleton from the binary segmentation image, select multiple reference points from the binary segmentation image based on the vascular skeleton, and determine the central selection block corresponding to each reference point based on the multiple feature maps; wherein, the multiple reference points include multiple target points and the auxiliary points corresponding to each target point;

[0008] For each target point, use the auxiliary point corresponding to the target point as the target auxiliary point, and splice the central selection block corresponding to the target point and the central selection block corresponding to the target auxiliary point to obtain the spliced input data corresponding to the target point;

[0009] Input the splicing input data corresponding to the target point into the pre-trained regression network model to obtain the blood vessel boundary corresponding to the target point;

[0010] Map the blood vessel boundary corresponding to each target point to the binary segmentation image to correct the segmentation result of the segmentation model, so as to obtain the blood vessel segmentation result corresponding to the binary segmentation image.

[0011] Further, the selecting a plurality of reference points from the binary segmentation image based on the blood vessel skeleton includes:

[0012] For each skeleton point on the blood vessel skeleton, with this skeleton point as the center, determine the filtering kernel corresponding to this skeleton point from the binary segmentation image based on a preset radius;

[0013] Perform a convolution operation on the binary segmentation image and the filtering kernel corresponding to this skeleton point to obtain the diameter eigenvalue corresponding to this skeleton point;

[0014] According to the diameter eigenvalue corresponding to each skeleton point, determine the skeleton point with the largest diameter eigenvalue from multiple skeleton points as the coronary artery starting point;

[0015] Take the coronary artery starting point as the target point, and select the pixel points around the target point in the binary segmentation image as the auxiliary points corresponding to the target point;

[0016] Take the auxiliary points corresponding to the target point as the target point, and return to execute the step of selecting the pixel points around the target point in the binary segmentation image as the auxiliary points corresponding to the target point until the target point is the end point of the blood vessel skeleton.

[0017] Further, the determining the central selection block corresponding to each reference point based on the multiple feature maps includes:

[0018] Perform feature splicing on the multiple feature maps to obtain a comprehensive feature map;

[0019] Use each reference point to perform feature extraction on the comprehensive feature map to obtain the central selection block corresponding to each reference point.

[0020] Further, the mapping the blood vessel boundary corresponding to each target point to the binary segmentation image to correct the segmentation result of the segmentation model to obtain the blood vessel segmentation result corresponding to the binary segmentation image includes:

[0021] Based on the blood vessel boundary corresponding to each target point, determine the blood vessel area and a plurality of first blood vessel pixel points located in the blood vessel area from the binary segmentation image;

[0022] Determine multiple second vascular pixel points with pixel values of 1 in the binary segmentation image;

[0023] For each different target abscissa among the coordinate values corresponding to the multiple second vascular pixel points, screen out multiple reference pixel points from the multiple first vascular pixel points based on the target abscissa, and screen out multiple to-be-corrected pixel points from the multiple second vascular pixel points based on the target abscissa. Among them, the abscissa values of the reference pixel points and the to-be-corrected pixel points are both the target abscissa;

[0024] Based on the ordinate values of the multiple reference pixel points, correct the ordinate values of the multiple to-be-corrected pixel points to obtain multiple target vascular pixel points corresponding to the target abscissa;

[0025] Based on the multiple target vascular pixel points corresponding to the multiple target abscissas, obtain the vascular segmentation result corresponding to the binary segmentation image.

[0026] Further, train the segmentation model and the regression network model through the following steps:

[0027] Obtain coronary angiography sample images; among them, label the sample tag value of the pixel points in the vascular region of the coronary angiography sample image as 1, and label the sample tag value of the pixel points in the non-vascular region of the coronary angiography sample image as 0;

[0028] Input the coronary angiography sample images into the original segmentation model to obtain a binary segmentation prediction image and multiple feature prediction maps output by different network layers of the original segmentation model;

[0029] Compare the predicted tag value corresponding to each pixel point in the binary segmentation prediction image with the sample tag value corresponding to each pixel point in the coronary angiography sample image to determine the first loss function of the original segmentation model in the current state;

[0030] Based on the binary segmentation prediction image and the multiple feature prediction maps, determine the predicted input data corresponding to each predicted target point located in the binary segmentation prediction image;

[0031] Input the predicted input data corresponding to each predicted target point into the original regression network model to obtain the predicted vascular boundary corresponding to each predicted target point, and based on the predicted vascular boundary corresponding to each predicted target point, respectively determine the predicted vascular regions from the binary segmentation prediction image and the coronary angiography sample image;

[0032] Compare the predicted label value corresponding to each pixel point in the predicted blood vessel region of the binary segmentation prediction image with the sample label value corresponding to each pixel point in the predicted blood vessel region of the coronary angiography sample image to determine the second loss function of the original regression network model in the current state;

[0033] Add the first loss function and the second loss function to obtain the total loss function, and continuously update the model parameters of the original segmentation model and the model parameters of the original regression network model based on the total loss function, and perform iterative training on the original segmentation model and the original regression network model until the total loss function converges to obtain the segmentation model and the regression network model.

[0034] In a second aspect, an embodiment of the present application further provides a blood vessel segmentation device for coronary angiography images, and the blood vessel segmentation device includes:

[0035] An image determination module, configured to obtain a coronary angiography original image, and input the coronary angiography original image into a pre-trained segmentation model to obtain a binary segmentation image and multiple feature maps output by different network layers of the segmentation model;

[0036] A center selection block determination module, configured to extract a blood vessel skeleton from the binary segmentation image, select multiple reference points from the binary segmentation image based on the blood vessel skeleton, and determine a center selection block corresponding to each reference point based on the multiple feature maps; wherein, the multiple reference points include multiple target points and an auxiliary point corresponding to each target point;

[0037] An input data determination module, configured to, for each target point, use the auxiliary point corresponding to the target point as a target auxiliary point, and splice the center selection block corresponding to the target point and the center selection block corresponding to the target auxiliary point to obtain the spliced input data corresponding to the target point;

[0038] A blood vessel boundary determination module, configured to input the spliced input data corresponding to the target point into a pre-trained regression network model to obtain the blood vessel boundary corresponding to the target point;

[0039] A segmentation result correction module, configured to map the blood vessel boundary corresponding to each target point to the binary segmentation image to correct the segmentation result of the segmentation model to obtain the blood vessel segmentation result corresponding to the binary segmentation image.

[0040] Further, when the center selection block determination module is used to select multiple reference points from the binary segmentation image based on the blood vessel skeleton, the center selection block determination module is further used for:

[0041] For each skeleton point on the blood vessel skeleton, with this skeleton point as the center, a filtering kernel corresponding to this skeleton point is determined from the binary segmentation image based on a preset radius;

[0042] Perform a convolution operation on the binary segmentation image and the filtering kernel corresponding to this skeleton point to obtain a diameter eigenvalue corresponding to this skeleton point;

[0043] According to the diameter eigenvalues corresponding to each skeleton point, determine the skeleton point with the largest diameter eigenvalue from multiple skeleton points as the coronary artery starting point;

[0044] Take the coronary artery starting point as the target point, and select the pixel points in the binary segmentation image around the target point as the auxiliary points corresponding to the target point;

[0045] Take the auxiliary points corresponding to the target point as the target point, and return to execute the step of selecting the pixel points in the binary segmentation image around the target point as the auxiliary points corresponding to the target point until the target point is the end point of the blood vessel skeleton.

[0046] Further, when the center selection block determination module is used to determine the center selection block corresponding to each reference point based on multiple feature maps, the center selection block determination module is further used for:

[0047] Perform feature splicing on multiple feature maps to obtain a comprehensive feature map;

[0048] Use each reference point to perform feature block extraction on the comprehensive feature map to obtain the center selection block corresponding to each reference point.

[0049] In a third aspect, an embodiment of the present application further provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, the steps of the blood vessel segmentation method for coronary angiography images as described above are executed.

[0050] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is run by a processor, the steps of the blood vessel segmentation method for coronary angiography images as described above are executed.

[0051] A method, device, equipment and medium for vascular segmentation of coronary angiography images provided by an embodiment of the present application. First, obtain the original coronary angiography image, and input the original coronary angiography image into a pre-trained segmentation model to obtain a binary segmentation image and multiple feature maps output by different network layers of the segmentation model. Then, extract the vascular skeleton from the binary segmentation image, select multiple reference points from the binary segmentation image based on the vascular skeleton, and determine the central selection block corresponding to each reference point based on the multiple feature maps. For each target point, use the auxiliary point corresponding to the target point as the target auxiliary point, and splice the central selection block corresponding to the target point and the central selection block corresponding to the target auxiliary point to obtain the spliced input data corresponding to the target point. Input the spliced input data corresponding to the target point into a pre-trained regression network model to obtain the vascular boundary corresponding to the target point. Finally, map the vascular boundary corresponding to each target point to the binary segmentation image to correct the segmentation result of the segmentation model, so as to obtain the vascular segmentation result corresponding to the binary segmentation image.

[0052] The present application uses the vascular boundary predicted by the regression network model to correct the binary segmentation image output by the segmentation model, obtaining a more accurate vascular segmentation result. It can accurately segment blood vessels of different thicknesses in coronary angiography, fully consider the characteristics of blood vessel stenosis, fully grasp the details of each position of the blood vessel, and improve the accuracy and robustness of vascular segmentation for coronary angiography images.

[0053] To make the above objects, features and advantages of the present application more obvious and understandable, the following specifically gives preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0055] Figure 1 It is a flowchart of a method for vascular segmentation of coronary angiography images provided by an embodiment of the present application;

[0056] Figure 2 It is a structural diagram of a segmentation model provided by an embodiment of the present application;

[0057] Figure 3 It is a schematic structural diagram of a device for vascular segmentation of coronary angiography images provided by an embodiment of the present application;

[0058] Figure 4 A schematic structural diagram of an electronic device provided by an embodiment of the present application. Specific implementation manners

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Components of the embodiments of the present application described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application to be protected, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, every other embodiment obtained by those skilled in the art without creative efforts belongs to the scope of protection of the present application.

[0060] First, an applicable application scenario of the present application is introduced. The present application can be applied to the field of medical image processing technology.

[0061] In the field of modern medical technology, vascular image technology can help doctors understand the vascular status of patients, and angiography technology for imaging tubular structures such as blood vessels has been widely used. Therefore, a high-accuracy vascular image segmentation method becomes particularly important.

[0062] Vascular segmentation in coronary angiography is a challenging task. Through research, it is found that it mainly includes the following difficulties:

[0063] Image quality: Coronary angiography images are interfered by various factors, such as noise, scattering, calcified plaques, etc. These factors will cause the image quality to decline, making the vascular boundaries blurred or unclear, and increasing the difficulty of segmentation.

[0064] Vascular morphological diversity: Coronary blood vessels have diverse morphologies, including tortuosity, branching, stenosis, etc. These morphological changes increase the complexity of the segmentation algorithm because the algorithm needs to be able to accurately identify and segment blood vessels of various morphologies.

[0065] Crossing and overlapping: In coronary angiography images, blood vessels may cross and overlap with each other. This makes it difficult to determine the boundaries of blood vessels because the segmentation algorithm needs to be able to distinguish different blood vessels in the crossing and overlapping regions.

[0066] Lesions and lesion regions: Lesions such as plaques, stenosis, or occlusion may exist in coronary images. The boundaries between these lesion regions and normal blood vessels are blurred, and precise segmentation algorithms are required to identify and segment blood vessels and lesion regions.

[0067] Therefore, how to improve the accuracy of blood vessel segmentation in coronary angiography has become a technical problem that urgently needs to be solved.

[0068] Based on this, the embodiments of the present application provide a blood vessel segmentation method for coronary angiography images, which can accurately segment blood vessels of different thicknesses in coronary angiography, fully consider the characteristics of blood vessel stenosis, fully grasp the details of each position of the blood vessels, and improve the accuracy and robustness of blood vessel segmentation for coronary angiography images.

[0069] Please refer to Figure 1 , Figure 1 which is a flowchart of a blood vessel segmentation method for coronary angiography images provided by the embodiments of the present application. As Figure 1 shown in

[0070] S101, obtain the original coronary angiography image, and input the original coronary angiography image into a pre-trained segmentation model to obtain a binary segmentation image and multiple feature maps output by different network layers of the segmentation model.

[0071] It should be noted that the original coronary angiography image is the original medical image taken after injecting a contrast agent into the patient. The binary segmentation image refers to the segmentation image obtained by binarizing the original coronary angiography image by the segmentation model. The feature map refers to the feature map output by different network layers in the segmentation model.

[0072] For the above step S101, in specific implementation, first obtain the original coronary angiography image obtained by coronary angiography, and then input the original coronary angiography image into a pre-trained segmentation model to obtain the binary segmentation image output by the segmentation model and multiple feature maps output by different network layers of the segmentation model. Please refer to Figure 2 , Figure 2 which is a structural diagram of a segmentation model provided by the embodiments of the present application. As Figure 2 shown in

[0073] S102, extract the blood vessel skeleton from the binary segmentation image, select multiple reference points from the binary segmentation image based on the blood vessel skeleton, and determine the central selection block corresponding to each reference point based on the multiple feature maps.

[0074] Regarding the above step S102, in specific implementation, after obtaining the binary segmentation image output by the segmentation model, the blood vessel skeleton is extracted from the binary segmentation image. Here, the two-dimensional coordinates of the blood vessel skeleton are represented as skele(i, j) = {(i, j)|S(i, j) = 1}, where S represents the blood vessel skeleton, and (i, j) are the coordinates of each point on the blood vessel skeleton. Specifically, the following several conventional schemes can be adopted for the skeletonization algorithm: (1) Thinning Algorithm: The thinning algorithm is an iterative skeletonization algorithm that obtains the skeleton line by repeatedly deleting foreground pixels. The most famous thinning algorithms include the Zhang-Suen algorithm, the Guo-Hall algorithm, and the Rosenfeld algorithm, etc. (2) Medial Axis Transform: The medial axis transform is based on the boundary information between the object pixels and the surrounding background pixels, and obtains the skeleton line by tracing the light rays on the boundary. This method can preserve the connectivity and geometric shape of the object. (3) Distance Transform: The distance transform is based on the distance from each pixel in the image to the nearest background pixel, and obtains the skeleton line through threshold operation and morphological processing. The distance transform can preserve the shape information of the object and add additional topological information to the skeleton line. (4) Morphological Skeletonization: Morphological skeletonization is a skeletonization method based on morphological operations, which uses erosion and dilation operations to extract the central axis of the object. This method is simple and efficient and is suitable for objects with simple shapes. Here, after extracting the blood vessel skeleton from the binary segmentation image, multiple reference points are selected from the binary segmentation image based on the blood vessel skeleton. Specifically, the multiple reference points include multiple target points and the auxiliary points corresponding to each target point. Then, the central selection blocks corresponding to each reference point are determined from the multiple feature maps obtained in step S101.

[0075] As an optional embodiment, regarding the above step S102, the selecting multiple reference points from the binary segmentation image based on the blood vessel skeleton includes:

[0076] A: For each skeleton point on the blood vessel skeleton, with this skeleton point as the center, a filter kernel corresponding to this skeleton point is determined from the binary segmentation image based on a preset radius.

[0077] Here, the preset radius can be preset to 10 pixels. The skeleton points are each pixel point on the blood vessel skeleton.

[0078] For the above step A, in specific implementation, for each skeleton point on the blood vessel skeleton, with this skeleton point as the center, a filter kernel corresponding to this skeleton point is determined from the binary segmentation image based on a preset radius. Here, continuing the above example, when the preset radius is 10 pixels, then with the skeleton point as the center and a radius of 10 pixels to select the filter kernel. Specifically, the filter kernel is expressed as:

[0079] kernel(x, y) = {(x - 10) 2 + (y - 10) 2 ≤ 10 2 | 0 ≤ x ≤ 20, 0 ≤ y ≤ 20}

[0080] where kernel(x, y) represents the filter kernel.

[0081] B: Perform a convolution operation on the binary segmentation image and the filter kernel corresponding to this skeleton point to obtain the diameter eigenvalue corresponding to this skeleton point.

[0082] For the above step B, in specific implementation, after obtaining the filter kernel corresponding to this skeleton point, perform a convolution operation on the binary segmentation image and the filter kernel corresponding to this skeleton point to obtain the diameter eigenvalue corresponding to this skeleton point. Specifically, the diameter eigenvalue is calculated through the following formula:

[0083]

[0084] where Seg represents the binary segmentation image output by the segmentation model; represents the convolution operation, that is, the pixels at each position are multiplied correspondingly, and then the sum of all multiplication results is calculated; sum(i, j) represents the diameter eigenvalue calculated when the skeleton point is (i, j).

[0085] C: According to the diameter eigenvalue corresponding to each skeleton point, determine the skeleton point with the largest diameter eigenvalue among multiple skeleton points as the coronary artery starting point.

[0086] For the above step C, in specific implementation, after the diameter eigenvalue corresponding to each skeleton point is determined, based on the diameter eigenvalue corresponding to each skeleton point, determine the skeleton point with the largest diameter eigenvalue among multiple skeleton points as the coronary artery starting point. Specifically, the coordinates of the coronary artery starting point start are expressed as start = argmax skele(i，j) [sum(i, j)].

[0087] D: Use the coronary artery starting point as the target point, and select the pixel points in the binary segmentation image around the target point as the auxiliary points corresponding to the target point.

[0088] E: Use the auxiliary point corresponding to the target point as the target point, and return to execute the step of selecting the pixel points around the target point in the binary segmented image as the auxiliary point corresponding to the target point until the target point is the end point of the blood vessel skeleton.

[0089] For the above steps D - E, in specific implementation, use the coronary artery starting point determined in step C as the target point, and select the pixel points around this target point in the binary segmented image as the auxiliary points corresponding to this target point. Specifically, the auxiliary points can be expressed as:

[0090] assist(i, j) = {(i - x) 2 + (j - y) 2 ≤ 1 | x 2 + y 2 = R 2 , (i, j) ∈ skele(i, j)}

[0091] where (i, j) represents the coordinates of the current target point, assist(i, j) represents the set of auxiliary points corresponding to the target point (i, j), and R is the set radius, defaulting to R = 3.

[0092] After determining the auxiliary points corresponding to the target point through the above step D, in the above step E, these auxiliary points will also be used as the target points for the next selection. Use the auxiliary points corresponding to the target point as the target point, and return to execute the step of selecting the pixel points around the target point in the binary segmented image as the auxiliary points corresponding to the target point in the above step D, and continue to select forward until the target point is the end point of the blood vessel skeleton. Here, the role of selecting the auxiliary points is to extract their features to assist the target point in boundary regression because the features of the surrounding points will affect the current target point.

[0093] As an optional embodiment, for the above step S102, the determining the central selection block corresponding to each reference point based on the multiple feature maps includes:

[0094] a: Perform feature splicing on the multiple feature maps to obtain a comprehensive feature map.

[0095] For the above step a, in specific implementation, perform feature splicing on the multiple feature maps, perform multi - layer feature splicing of the segmentation model to obtain a comprehensive feature map. Specifically, obtain the comprehensive feature map through the following formula:

[0096] F3 = concat(UpSample(C4), U3)

[0097] F2 = concat(UpSample(F3), U2)

[0098] F1 = concat(UpSample(F2), U1)

[0099] Among them, C4, U3, U2, and U1 respectively represent the feature maps output by different network layers in the segmentation model. F3 represents the feature map obtained by feature concatenation of C4 and U3. F3 represents the feature map obtained by feature concatenation of F3 and U2. F1 represents the comprehensive feature map obtained by feature concatenation of F2 and U1. UpSample() represents the upsampling process, and the method of transposed convolution is used here. concat represents the concatenation operation.

[0100] b: Use each of the reference points to perform feature block extraction on the comprehensive feature map to obtain the central selection block corresponding to each reference point.

[0101] Regarding the above step b, in specific implementation, after the comprehensive feature map is determined, use each reference point to perform feature block extraction on the comprehensive feature map to obtain the central selection block corresponding to each reference point. Specifically, the central selection block is expressed as:

[0102] patch(i, j) = F1[i - 32:i + 32, j - 32:j + 32]

[0103] Among them, patch(i, j) represents the central selection block corresponding to the reference point (i, j).

[0104] S103. For each target point, use the auxiliary point corresponding to this target point as the target auxiliary point, and perform concatenation on the central selection block corresponding to this target point and the central selection block corresponding to the target auxiliary point to obtain the concatenated input data corresponding to this target point.

[0105] Regarding the above step S103, in specific implementation, for each target point, use the auxiliary point corresponding to this target point as the target auxiliary point, and perform concatenation on the central selection block corresponding to this target point and the central selection block corresponding to the target auxiliary point to obtain the concatenated input data corresponding to this target point as the input of the regression network model. Specifically, the concatenated input data is expressed as:

[0106] input(i, j) = concat{patch(i, j)|(i, j) ∈ {(i, j), assist(i, j)}}

[0107] Among them, input(i, j) represents the concatenated input data of the regression network model.

[0108] S104. Input the concatenated input data corresponding to this target point into the pre-trained regression network model to obtain the blood vessel boundary corresponding to this target point.

[0109] For the above-mentioned step S104, in specific implementation, for each target point, the splicing input data corresponding to the target point is input into a pre-trained regression network model to obtain the blood vessel boundary corresponding to the target point. Specifically, the regression network module provided in the embodiments of the present application is composed of four sub-modules, namely three convolutional modules (D3, D2, D1) and a fully connected layer (FC1). Among them, each convolutional module is composed of two convolutional layers connected in series and a downsampling layer. The output of the regression network model is a 1x2 vector, indicating the number of pixels of the left boundary and the right boundary in the case of the target point (i, j), so as to represent the blood vessel boundary corresponding to the target point.

[0110] As an alternative embodiment, the segmentation model and the regression network model are trained through the following steps:

[0111] I: Obtain coronary angiography sample images.

[0112] Among them, the sample label value of the pixel points in the blood vessel area of the coronary angiography sample image is marked as 1, and the sample label value of the pixel points in the non-blood vessel area of the coronary angiography sample image is marked as 0.

[0113] For the above-mentioned step I, in specific implementation, coronary angiography sample images are obtained. In the coronary angiography sample images, the pixel points with a label value of 1 are the pixel points in the blood vessel area, and the pixel points with a label value of 0 are the pixel points in the blood vessel area.

[0114] II: Input the coronary angiography sample images into the original segmentation model to obtain a binary segmentation prediction image and multiple feature prediction maps output by different network layers of the original segmentation model.

[0115] For the above-mentioned step II, in specific implementation, the coronary angiography sample images are input into the original segmentation model to obtain a binary segmentation prediction image and multiple feature prediction maps output by different network layers of the original segmentation model.

[0116] III: Compare the predicted label value corresponding to each pixel point in the binary segmentation prediction image with the sample label value corresponding to each pixel point in the coronary angiography sample image to determine the first loss function of the original segmentation model in the current state.

[0117] For the above-mentioned step III, in specific implementation, the predicted label value corresponding to each pixel point in the binary segmentation prediction image is compared with the sample label value corresponding to each pixel point in the coronary angiography sample image to determine the first loss function of the original segmentation model in the current state. Specifically, the first loss function is calculated through the following formula:

[0118]

[0119] Among them, L seg represents the first loss function. The first loss function adopts the conventional binary cross-entropy loss. Here, l(x, y) represents the sample coordinate value of the pixel point with coordinates (x, y) in the coronary angiography sample image, l(x, y) ∈ {0, 1}, and p(x, y) represents the predicted label value of the pixel point with coordinates (x, y) in the binary segmentation prediction image, that is, the foreground probability result, and p(x, y) ∈ [0, 1].

[0120] IV: Determine the predicted input data corresponding to each predicted target point located within the binary segmentation prediction image based on the binary segmentation prediction image and the multiple feature prediction maps.

[0121] Regarding the above step IV, in specific implementation, determine the predicted input data corresponding to each predicted target point located within the binary segmentation prediction image based on the binary segmentation prediction image output by the original segmentation model and the multiple feature prediction maps. Here, the description of determining the predicted input data can refer to the description of S102 to S103, and the same technical effects can be achieved, so it will not be elaborated here.

[0122] V: Input the predicted input data corresponding to each predicted target point into the original regression network model to obtain the predicted blood vessel boundary corresponding to each predicted target point, and determine the predicted blood vessel region from the binary segmentation prediction image and the coronary angiography sample image respectively based on the predicted blood vessel boundary corresponding to each predicted target point.

[0123] Regarding the above step V, in specific implementation, input the predicted input data corresponding to each predicted target point into the original regression network model to obtain the predicted blood vessel boundary corresponding to each predicted target point, and determine the predicted blood vessel region from the binary segmentation prediction image based on the predicted blood vessel boundary corresponding to each predicted target point. Here, the predicted blood vessel boundary represents the number of pixels of the left boundary and the right boundary of the predicted target point. Determine the position of each predicted target point in the binary segmentation prediction image, and then the predicted blood vessel region can be determined from the binary segmentation prediction image based on the predicted blood vessel boundary corresponding to each predicted target point. Since the binary segmentation prediction image has the same size as the coronary angiography sample image, the predicted blood vessel region in the coronary angiography sample image can be determined in the same way as above.

[0124] VI: Compare the predicted label value corresponding to each pixel point within the predicted blood vessel region in the binary segmentation prediction image with the sample label value corresponding to each pixel point within the predicted blood vessel region in the coronary angiography sample image to determine the second loss function of the original regression network model in the current state.

[0125] For the above-mentioned step VI, in specific implementation, the predicted label value corresponding to each pixel point in the predicted blood vessel area of the binary segmentation prediction image is compared with the sample label value corresponding to each pixel point in the predicted blood vessel area of the coronary angiography sample image to determine the second loss function of the original regression network model in the current state. Specifically, the second loss function is calculated by the following formula:

[0126]

[0127] where L reg represents the second loss function. The second loss function adopts the conventional L1-normal loss function, where r(x, y) represents the sample coordinate value of the pixel point with coordinates (x, y) in the predicted blood vessel area of the coronary angiography sample image, and d(x, y) represents the sample coordinate value of the pixel point with coordinates (x, y) in the predicted blood vessel area of the binary segmentation prediction image.

[0128] VII: Add the first loss function and the second loss function to obtain the total loss function, and continuously update the model parameters of the original segmentation model and the model parameters of the original regression network model based on the total loss function, and perform iterative training on the original segmentation model and the original regression network model until the total loss function converges to obtain the segmentation model and the regression network model.

[0129] For the above-mentioned step VII, in specific implementation, add the first loss function of the original segmentation model and the second loss function of the original regression network model to obtain the total loss function. Specifically, the total loss function L is expressed as L = L seg + L reg . After the total loss function is determined, update the model parameters of the original segmentation model and the model parameters of the original regression network model according to the calculated total loss function. After the model parameters are updated, if the current total loss function does not converge, continue the iterative training of the next training round. At each step of the iteration, a new total loss function will be calculated. When the total loss function does not converge, continuously update the model parameters of the original segmentation model and the model parameters of the original regression network model, and the new weights will calculate a new total loss function, so that the total loss function shows a fluctuating downward trend during the iteration process. Finally, when the total loss function reaches convergence, that is, when the total loss function does not decrease significantly compared with the total loss function calculated last time, it is considered that the original segmentation model and the original regression network model reach the convergence state. At this time, the predictions of the original segmentation model and the original regression network model are relatively accurate, and the training is ended to obtain the segmentation model and the regression network model.

[0130] S105. Map the blood vessel boundary corresponding to each target point into the binary segmentation image, and correct the segmentation result of the segmentation model to obtain the blood vessel segmentation result corresponding to the binary segmentation image.

[0131] For the above step S105, in specific implementation, map the blood vessel boundary corresponding to each obtained target point into the binary segmentation image, and correct the segmentation result of the segmentation model to obtain the blood vessel segmentation result corresponding to the binary segmentation image.

[0132] As an optional embodiment, for the above step S105, the step of mapping the blood vessel boundary corresponding to each target point into the binary segmentation image, correcting the segmentation result of the segmentation model to obtain the blood vessel segmentation result corresponding to the binary segmentation image includes:

[0133] Step 1051. Determine the blood vessel region and a plurality of first blood vessel pixel points located in the blood vessel region from the binary segmentation image based on the blood vessel boundary corresponding to each target point.

[0134] For the above step 1051, in specific implementation, determine the blood vessel region and a plurality of first blood vessel pixel points located in the blood vessel region from the binary segmentation image based on the blood vessel boundary corresponding to each target point. Here, the blood vessel boundary represents the number of pixels of the left boundary and the right boundary of the target point. Determine the position of each target point in the binary segmentation image, and then the blood vessel region can be determined from the binary segmentation image based on the blood vessel boundary corresponding to each target point.

[0135] Step 1052. Determine a plurality of second blood vessel pixel points with pixel values of 1 in the binary segmentation image.

[0136] Since each pixel in a binary image has only two colors, black and white, and the pixel has only two values, 0 and 1, generally 0 is used to represent black and 1 is used to represent white. For the above step 1052, in specific implementation, determine a plurality of second blood vessel pixel points with pixel values of 1 in the binary segmentation image.

[0137] Step 1053. For each different target abscissa among the coordinate values corresponding to the plurality of second blood vessel pixel points, screen out a plurality of reference pixel points from the plurality of first blood vessel pixel points based on the target abscissa, and screen out a plurality of pixels to be corrected from the plurality of second blood vessel pixel points based on the target abscissa.

[0138] For step 1053 above, in specific implementation, for each different target abscissa among the coordinate values corresponding to multiple second blood vessel pixel points, multiple reference pixel points are screened out from multiple first blood vessel pixel points based on the target abscissa, and multiple pixel points to be corrected are screened out from multiple second blood vessel pixel points based on the target abscissa. Here, the abscissa values of the reference pixel points and the pixel points to be corrected are both the target abscissa.

[0139] Step 1054, correct the ordinate values of multiple pixel points to be corrected based on the ordinate values of multiple reference pixel points to obtain multiple target blood vessel pixel points corresponding to the target abscissa.

[0140] For step 1054 above, in specific implementation, correct the ordinate values of multiple pixel points to be corrected according to the ordinate values of multiple reference pixel points to obtain multiple target blood vessel pixel points corresponding to the target abscissa. Specifically, correct the ordinate values of multiple pixel points to be corrected through the following formula:

[0141] Seg(x, Y) = 1

[0142] Y = {y - d(x, y)[0]: y + d(x, y)[1]}

[0143] Wherein, Seg(x, Y) = 1 indicates that the pixel point with coordinates (x, Y) in the binary segmentation image is 1, y represents the ordinate value of the pixel point to be corrected, x represents the target abscissa, is a number, Y represents a set, is several numbers. d(x, y) is the prediction result of the regression network model, that is, the reference pixel point, is a vector of two elements, [0] indicates selecting the first element among them, and [1] indicates selecting the second element among them.

[0144] Step 1055, obtain the blood vessel segmentation result corresponding to the binary segmentation image based on multiple target blood vessel pixel points corresponding to multiple target abscissas.

[0145] For step 1055 above, in specific implementation, after multiple target blood vessel pixel points corresponding to each target abscissa are determined, merge multiple target blood vessel pixel points corresponding to multiple target abscissas, and the blood vessel segmentation result corresponding to the binary segmentation image can be determined.

[0146] The vascular segmentation method for coronary angiography images provided by the embodiments of the present application. First, obtain the original coronary angiography image, and input the original coronary angiography image into a pre-trained segmentation model to obtain a binary segmentation image and multiple feature maps output by different network layers of the segmentation model; then, extract the vascular skeleton from the binary segmentation image, select multiple reference points from the binary segmentation image based on the vascular skeleton, and determine the central selection block corresponding to each reference point based on the multiple feature maps; for each target point, use the auxiliary point corresponding to the target point as the target auxiliary point, and splice the central selection block corresponding to the target point and the central selection block corresponding to the target auxiliary point to obtain the spliced input data corresponding to the target point; input the spliced input data corresponding to the target point into a pre-trained regression network model to obtain the vascular boundary corresponding to the target point; finally, map the vascular boundary corresponding to each target point to the binary segmentation image to correct the segmentation result of the segmentation model, so as to obtain the vascular segmentation result corresponding to the binary segmentation image.

[0147] The present application uses the vascular boundary predicted by the regression network model to correct the binary segmentation image output by the segmentation model, obtains a more accurate vascular segmentation result, can accurately segment blood vessels of different thicknesses in coronary angiography, fully considers the characteristics of blood vessel stenosis, fully grasps the details of each position of the blood vessel, and improves the accuracy and robustness of vascular segmentation for coronary angiography images.

[0148] Please refer to Figure 3 , Figure 3 which is a schematic structural diagram of a vascular segmentation device for coronary angiography images provided by the embodiments of the present application. As Figure 3 shown in

[0149] The image determination module 301 is configured to obtain the original coronary angiography image, and input the original coronary angiography image into a pre-trained segmentation model to obtain a binary segmentation image and multiple feature maps output by different network layers of the segmentation model;

[0150] The central selection block determination module 302 is configured to extract the vascular skeleton from the binary segmentation image, select multiple reference points from the binary segmentation image based on the vascular skeleton, and determine the central selection block corresponding to each reference point based on the multiple feature maps; where the multiple reference points include multiple target points and the auxiliary point corresponding to each target point;

[0151] The input data determination module 303 is configured to, for each target point, use the auxiliary point corresponding to the target point as the target auxiliary point, and splice the central selection block corresponding to the target point and the central selection block corresponding to the target auxiliary point to obtain the spliced input data corresponding to the target point;

[0152] The blood vessel boundary determination module 304 is configured to input the spliced input data corresponding to the target point into a pre-trained regression network model to obtain the blood vessel boundary corresponding to the target point.

[0153] The segmentation result correction module 305 is configured to map the blood vessel boundary corresponding to each target point into the binary segmentation image, and correct the segmentation result of the segmentation model to obtain the blood vessel segmentation result corresponding to the binary segmentation image.

[0154] Further, when the center selection block determination module 302 is used to select a plurality of reference points from the binary segmentation image based on the blood vessel skeleton, the center selection block determination module 302 is further configured to:

[0155] For each skeleton point on the blood vessel skeleton, with the skeleton point as the center, a filtering kernel corresponding to the skeleton point is determined from the binary segmentation image based on a preset radius;

[0156] A convolution operation is performed on the binary segmentation image and the filtering kernel corresponding to the skeleton point to obtain a diameter eigenvalue corresponding to the skeleton point;

[0157] According to the diameter eigenvalue corresponding to each skeleton point, the skeleton point with the largest diameter eigenvalue is determined from the plurality of skeleton points as the coronary artery starting point;

[0158] Taking the coronary artery starting point as the target point, and selecting the pixel points around the target point in the binary segmentation image as the auxiliary points corresponding to the target point;

[0159] Taking the auxiliary points corresponding to the target point as the target point, and returning to execute the step of selecting the pixel points around the target point in the binary segmentation image as the auxiliary points corresponding to the target point until the target point is the end point of the blood vessel skeleton.

[0160] Further, when the center selection block determination module 302 is used to determine the center selection block corresponding to each reference point based on the plurality of feature maps, the center selection block determination module 302 is further configured to:

[0161] Perform feature splicing on the plurality of feature maps to obtain a comprehensive feature map;

[0162] Use each reference point to perform feature extraction on the comprehensive feature map to obtain the center selection block corresponding to each reference point.

[0163] Further, when the segmentation result correction module 305 is used to map the blood vessel boundary corresponding to each target point into the binary segmentation image to correct the segmentation result of the segmentation model to obtain the blood vessel segmentation result corresponding to the binary segmentation image, the segmentation result correction module 305 is further used for:

[0164] Determine a blood vessel region and a plurality of first blood vessel pixel points located within the blood vessel region from the binary segmentation image based on the blood vessel boundary corresponding to each target point;

[0165] Determine a plurality of second blood vessel pixel points with a pixel value of 1 in the binary segmentation image;

[0166] For each different target abscissa in the coordinate values corresponding to the plurality of second blood vessel pixel points, screen out a plurality of reference pixel points from the plurality of first blood vessel pixel points based on the target abscissa, and screen out a plurality of pixel points to be corrected from the plurality of second blood vessel pixel points based on the target abscissa, wherein the abscissa values of the reference pixel points and the abscissa values of the pixel points to be corrected are both the target abscissa;

[0167] Correct the ordinate values of the plurality of pixel points to be corrected based on the ordinate values of the plurality of reference pixel points to obtain a plurality of target blood vessel pixel points corresponding to the target abscissa;

[0168] Obtain the blood vessel segmentation result corresponding to the binary segmentation image based on the plurality of target blood vessel pixel points corresponding to the plurality of target abscissas.

[0169] Further, the blood vessel segmentation device 300 further includes a model training module, and the model training module trains the segmentation model and the regression network model through the following steps:

[0170] Obtain a coronary angiography sample image; wherein, the sample label value of the pixel points within the blood vessel region in the coronary angiography sample image is marked as 1, and the sample label value of the pixel points outside the blood vessel region in the coronary angiography sample image is marked as 0;

[0171] Input the coronary angiography sample image into the original segmentation model to obtain a binary segmentation prediction image and a plurality of feature prediction maps output by different network layers of the original segmentation model;

[0172] Compare the predicted label value corresponding to each pixel point in the binary segmentation prediction image with the sample label value corresponding to each pixel point in the coronary angiography sample image to determine the first loss function of the original segmentation model in the current state;

[0173] Determine the predicted input data corresponding to each predicted target point located within the binary segmentation prediction image based on the binary segmentation prediction image and the plurality of feature prediction maps;

[0174] Input the prediction input data corresponding to each predicted target point into the original regression network model to obtain the predicted vascular boundary corresponding to each predicted target point, and determine the predicted vascular region from the binary segmentation prediction image and the coronary angiography sample image respectively based on the predicted vascular boundary corresponding to each predicted target point;

[0175] Compare the predicted label value corresponding to each pixel point in the predicted vascular region of the binary segmentation prediction image with the sample label value corresponding to each pixel point in the predicted vascular region of the coronary angiography sample image to determine the second loss function of the original regression network model in the current state;

[0176] Add the first loss function and the second loss function to obtain the total loss function, and continuously update the model parameters of the original segmentation model and the model parameters of the original regression network model based on the total loss function, and perform iterative training on the original segmentation model and the original regression network model until the total loss function converges to obtain the segmentation model and the regression network model.

[0177] Please refer to Figure 4 , Figure 4 which is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 4 shown in

[0178] the electronic device 400 includes a processor 410, a memory 420, and a bus 430. The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 runs, the processor 410 communicates with the memory 420 through the bus 430. When the machine-readable instructions are executed by the processor 410, they can execute the steps of the method for segmenting blood vessels in coronary angiography images in the method embodiment as described above. The specific implementation manner can refer to the method embodiment and will not be elaborated here. Figure 1

[0179] Figure 1 An embodiment of the present application further provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is run by a processor, it can execute the steps of the method for segmenting blood vessels in coronary angiography images in the method embodiment as described above. The specific implementation manner can refer to the method embodiment and will not be elaborated here.

[0180]

[0181] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated here.

[0181] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some communication interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical, or other forms.

[0182] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0183] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0184] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0185] Finally, it should be noted that the above-described embodiments are only specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present application can still modify the technical solutions recorded in the foregoing embodiments or can easily think of changes, or make equivalent replacements for some of the technical features; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for vascular segmentation of coronary angiography images, characterized in that, The vascular segmentation method includes: Obtain the original coronary angiography image, and input the original coronary angiography image into a pre-trained segmentation model to obtain a binary segmentation image and multiple feature maps output by different network layers of the segmentation model; Extract the vascular skeleton from the binary segmentation image, select multiple reference points from the binary segmentation image based on the vascular skeleton, and determine the central selection block corresponding to each reference point based on the multiple feature maps; wherein, the multiple reference points include multiple target points and the auxiliary points corresponding to each target point; For each target point, use the auxiliary point corresponding to the target point as the target auxiliary point, and splice the central selection block corresponding to the target point and the central selection block corresponding to the target auxiliary point to obtain the spliced input data corresponding to the target point; Input the spliced input data corresponding to the target point into a pre-trained regression network model to obtain the vascular boundary corresponding to the target point; Map the vascular boundary corresponding to each target point to the binary segmentation image to correct the segmentation result of the segmentation model, so as to obtain the vascular segmentation result corresponding to the binary segmentation image; The selecting multiple reference points from the binary segmentation image based on the vascular skeleton includes: For each skeleton point on the vascular skeleton, with the skeleton point as the center, determine the filtering kernel corresponding to the skeleton point from the binary segmentation image based on a preset radius; Perform a convolution operation on the binary segmentation image and the filtering kernel corresponding to the skeleton point to obtain the diameter eigenvalue corresponding to the skeleton point; According to the diameter eigenvalue corresponding to each skeleton point, determine the skeleton point with the largest diameter eigenvalue from multiple skeleton points as the coronary artery starting point; Use the coronary artery starting point as the target point, and select the pixel points around the target point in the binary segmentation image as the auxiliary points corresponding to the target point; Use the auxiliary point corresponding to the target point as the target point, and return to execute the step of selecting the pixel points around the target point in the binary segmentation image as the auxiliary points corresponding to the target point until the target point is the end point of the vascular skeleton.

2. The vascular segmentation method according to claim 1, wherein The determining the central selection block corresponding to each reference point based on the multiple feature maps includes: Perform feature splicing on the multiple feature maps to obtain a comprehensive feature map; Use each reference point to perform feature block extraction on the comprehensive feature map to obtain the central selection block corresponding to each reference point.

3. The vascular segmentation method according to claim 1, wherein The mapping the vascular boundary corresponding to each target point to the binary segmentation image to correct the segmentation result of the segmentation model, so as to obtain the vascular segmentation result corresponding to the binary segmentation image includes: Determine the vascular region and multiple first vascular pixel points located in the vascular region from the binary segmentation image based on the vascular boundary corresponding to each target point; Determine multiple second vascular pixel points with pixel values of 1 in the binary segmentation image; For each different target abscissa among the coordinate values corresponding to multiple second blood vessel pixel points, multiple reference pixel points are screened out from multiple first blood vessel pixel points based on the target abscissa, and multiple pixel points to be corrected are screened out from multiple second blood vessel pixel points based on the target abscissa. Among them, the abscissa values of the reference pixel points and the abscissa values of the pixel points to be corrected are both the target abscissa; Based on the ordinate values of multiple reference pixel points, the ordinate values of multiple pixel points to be corrected are corrected to obtain multiple target blood vessel pixel points corresponding to the target abscissa; Based on multiple target blood vessel pixel points corresponding to multiple target abscissas, the blood vessel segmentation result corresponding to the binary segmentation image is obtained.

4. The blood vessel segmentation method according to claim 1, characterized in that The segmentation model and the regression network model are trained through the following steps: Obtain a coronary angiography sample image; among them, the sample label value of the pixel points in the blood vessel area of the coronary angiography sample image is marked as 1, and the sample label value of the pixel points in the non-blood vessel area of the coronary angiography sample image is marked as 0; Input the coronary angiography sample image into the original segmentation model to obtain a binary segmentation prediction image and multiple feature prediction maps output by different network layers of the original segmentation model; Compare the predicted label value corresponding to each pixel point in the binary segmentation prediction image with the sample label value corresponding to each pixel point in the coronary angiography sample image to determine the first loss function of the original segmentation model in the current state; Based on the binary segmentation prediction image and multiple feature prediction maps, determine the predicted input data corresponding to each predicted target point located in the binary segmentation prediction image; Input the predicted input data corresponding to each predicted target point into the original regression network model to obtain the predicted blood vessel boundary corresponding to each predicted target point, and based on the predicted blood vessel boundary corresponding to each predicted target point, determine the predicted blood vessel area from the binary segmentation prediction image and the coronary angiography sample image respectively; Compare the predicted label value corresponding to each pixel point in the predicted blood vessel area in the binary segmentation prediction image with the sample label value corresponding to each pixel point in the predicted blood vessel area in the coronary angiography sample image to determine the second loss function of the original regression network model in the current state; Add the first loss function and the second loss function to obtain the total loss function, and continuously update the model parameters of the original segmentation model and the model parameters of the original regression network model based on the total loss function, and perform iterative training on the original segmentation model and the original regression network model until the total loss function converges to obtain the segmentation model and the regression network model.

5. An apparatus for segmenting blood vessels in coronary angiography images, characterized in that, The blood vessel segmentation device includes: An image determination module, configured to obtain a coronary angiography original image, and input the coronary angiography original image into a pre-trained segmentation model to obtain a binary segmentation image and multiple feature maps output by different network layers of the segmentation model; A central selection block determination module, configured to extract a blood vessel skeleton from the binary segmentation image, select a plurality of reference points from the binary segmentation image based on the blood vessel skeleton, and determine a central selection block corresponding to each reference point based on the plurality of feature maps; wherein, the plurality of reference points include a plurality of target points and an auxiliary point corresponding to each target point; An input data determination module, configured to, for each target point, use the auxiliary point corresponding to the target point as a target auxiliary point, and splice the central selection block corresponding to the target point and the central selection block corresponding to the target auxiliary point to obtain the spliced input data corresponding to the target point; A blood vessel boundary determination module, configured to input the spliced input data corresponding to the target point into a pre-trained regression network model to obtain the blood vessel boundary corresponding to the target point; A segmentation result correction module, configured to map the blood vessel boundary corresponding to each target point to the binary segmentation image, correct the segmentation result of the segmentation model to obtain the blood vessel segmentation result corresponding to the binary segmentation image; When the central selection block determination module is used to select a plurality of reference points from the binary segmentation image based on the blood vessel skeleton, the central selection block determination module is further configured to: For each skeleton point on the blood vessel skeleton, with the skeleton point as the center, determine a filter kernel corresponding to the skeleton point from the binary segmentation image based on a preset radius; Perform a convolution operation on the binary segmentation image and the filter kernel corresponding to the skeleton point to obtain a diameter eigenvalue corresponding to the skeleton point; According to the diameter eigenvalue corresponding to each skeleton point, determine the skeleton point with the largest diameter eigenvalue among the plurality of skeleton points as the coronary artery starting point; Use the coronary artery starting point as the target point, and select the pixel points in the binary segmentation image around the target point as the auxiliary points corresponding to the target point; Use the auxiliary points corresponding to the target point as the target point, and return to execute the step of selecting the pixel points in the binary segmentation image around the target point as the auxiliary points corresponding to the target point until the target point is the end point of the blood vessel skeleton.

6. The vascular segmentation device according to claim 5, characterized in that, When the central selection block determination module is used to determine a central selection block corresponding to each reference point based on the plurality of feature maps, the central selection block determination module is further configured to: Perform feature splicing on the plurality of feature maps to obtain a comprehensive feature map; Use each reference point to perform feature block extraction on the comprehensive feature map to obtain a central selection block corresponding to each reference point.

7. An electronic device, characterized in that, Including: A processor, a memory and a bus, the memory stores machine-readable instructions executable by the processor, when the electronic device runs, the processor communicates with the memory through the bus, and when the machine-readable instructions are run by the processor, the steps of the blood vessel segmentation method for coronary angiography images according to any one of claims 1 to 4 are executed.

8. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, the steps of the blood vessel segmentation method for coronary angiography images according to any one of claims 1 to 4 are executed.

Images