Blood vessel segmentation method, device and storage medium

Abstract

The present application relates to a blood vessel segmentation method, device and storage medium. The method includes: obtaining a blood vessel mask image in the initial medical image according to an initial medical image and a preset segmentation model; inputting the initial medical image and the blood vessel mask image into a preset regression model to obtain an arteriovenous distance map; the arteriovenous distance map is used to characterize the distance between each arterial blood vessel point and the arterial blood vessel boundary in the initial medical image and the distance between each venous blood vessel point and the venous blood vessel boundary in the initial medical image; obtaining a blood vessel segmentation image corresponding to the initial medical image according to the blood vessel mask image, the arteriovenous distance map and a preset fusion model. The present method can use deep learning to extract the blood vessel mask image and the blood vessel arteriovenous distance map, effectively segment the blood vessel from the initial medical image, and then perform feature fusion based on the arteriovenous distance map and the initial medical image to obtain a more accurate arteriovenous segmentation result.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a blood vessel segmentation method, device and storage medium. Background Art

[0002] In recent years, the incidence of cancer has been increasing year by year, such as lung cancer. The treatment of lung cancer generally requires lesion removal surgery, so preoperative planning is particularly important during the operation. For lung cancer surgery planning, organs such as the lungs need to be reconstructed in three dimensions, and the key part of the three-dimensional reconstruction of each organ is to accurately segment the blood vessels in each organ.

[0003] In traditional technology, the segmentation of blood vessels in various organs usually adopts the threshold separation method, such as using the distance map of grayscale boundary features or using the morphological directional thickness to separate the arteriovenous blood vessel boundaries, thereby obtaining a blood vessel segmentation image.

[0004] However, current blood vessel segmentation methods suffer from low segmentation accuracy. Summary of the invention

[0005] Based on this, it is necessary to provide a blood vessel segmentation method, device and storage medium that can accurately segment arteries and veins in order to address the above technical problems.

[0006] In a first aspect, the present application provides a blood vessel segmentation method. The method comprises:

[0007] According to the initial medical image and the preset segmentation model, a blood vessel mask image in the initial medical image is obtained;

[0008] The initial medical image and the vascular mask image are input into a preset regression model to obtain an arteriovenous distance map; the arteriovenous distance map is used to represent the distance between each arterial vascular point in the initial medical image and the arterial vascular boundary and the distance between each venous vascular point in the initial medical image and the venous vascular boundary;

[0009] According to the vascular mask image, the arteriovenous distance map and the preset fusion model, a vascular segmentation image corresponding to the initial medical image is obtained.

[0010] In one embodiment, obtaining a blood vessel segmentation image corresponding to the initial medical image according to the blood vessel mask image, the arteriovenous distance map and a preset fusion model includes:

[0011] Determine the target path of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map;

[0012] The vascular mask image, the arteriovenous distance map, and the target path of each blood vessel are input into the preset fusion model to obtain a vascular segmentation image corresponding to the initial medical image.

[0013] In one embodiment, determining a target path of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map includes:

[0014] Determine the starting point of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map;

[0015] According to the blood vessel mask image, determining the end point of each blood vessel in the blood vessel mask image;

[0016] Normalizing each distance value in the arteriovenous distance map to obtain a processed distance map;

[0017] Determine the path between the starting point of each blood vessel and the end point of each blood vessel according to the processed distance graph and the shortest path search algorithm;

[0018] The target path of each blood vessel is determined based on the path between the starting point of each blood vessel and the end point of each blood vessel.

[0019] In one embodiment, the starting point of each blood vessel includes the starting point of a vein and the starting point of an artery; and determining the path between the starting point of each blood vessel and the end point of each blood vessel according to the processed distance graph and the shortest path search algorithm includes:

[0020] For the end point of each blood vessel, the shortest path search algorithm is used to search for the path between the end point of each blood vessel and the starting point of the vein and determine it as the first path of each blood vessel;

[0021] as well as,

[0022] For the end point of each blood vessel, a shortest path search algorithm is used to search for a path between the start point of the arterial blood vessel and the end point of each blood vessel and determine it as the second path of each blood vessel.

[0023] In one embodiment, the path between the starting point of each blood vessel and the end point of each blood vessel, and determining the target path of each blood vessel, includes:

[0024] Derivative the first path of each blood vessel to obtain a first slope corresponding to the first path;

[0025] Deriving the second path of each blood vessel to obtain a second slope corresponding to the second path;

[0026] The first slope, the second slope, and the preset slope threshold are used to determine the target path of each blood vessel.

[0027] In one embodiment, the segmentation model includes a first network model and a second network model; obtaining a blood vessel mask image in the initial medical image according to the initial medical image and the preset segmentation model includes:

[0028] Inputting the initial medical image into the first network model for rough segmentation to obtain a segmented image of the organ in the initial medical image;

[0029] According to the segmented image of the organ, the position of the organ is determined in the initial medical image, and the minimum circumscribed cube of the organ is obtained;

[0030] The minimum circumscribed cube is input into the second network model for fine segmentation to obtain a blood vessel mask image; wherein the sampling resolution of the second network model is smaller than the sampling resolution of the first network model.

[0031] In one embodiment, determining the starting point of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map includes:

[0032] The blood vessel mask image is processed using the segmented image of the organ to obtain the main blood vessels of the arterial blood vessels and the main blood vessels of the venous blood vessels;

[0033] According to the main blood vessels of the arterial blood vessels, a connected domain corresponding to the main blood vessels of the arterial blood vessels in the arteriovenous distance map is determined, and the point with the largest distance value in the connected domain corresponding to the main blood vessels of the arterial blood vessels is determined as the starting point of the arterial blood vessels;

[0034] According to the main blood vessels of the veins, the connected domain corresponding to the main blood vessels of the veins in the arteriovenous distance map is determined, and the point with the largest distance value in the connected domain corresponding to the main blood vessels of the veins is determined as the starting point of the veins.

[0035] In one embodiment, the blood vessel segmentation method further includes:

[0036] According to the shortest path search method, the bifurcation point of each blood vessel in the blood vessel segmentation image is determined;

[0037] According to the preset segmentation and classification rules, the blood vessel segments between adjacent bifurcation points in each blood vessel are segmented and classified to obtain the structural diagram of each blood vessel.

[0038] In a second aspect, the present application also provides a blood vessel segmentation device. The device comprises:

[0039] A segmentation module, used for obtaining a blood vessel mask image in the initial medical image according to the initial medical image and a preset segmentation model;

[0040] A regression module, used for inputting the initial medical image and the vascular mask image into a preset regression model to obtain an arteriovenous distance map; the arteriovenous distance map is used for representing the distance between each arterial vascular point in the initial medical image and the arterial vascular boundary and the distance between each venous vascular point in the initial medical image and the venous vascular boundary;

[0041] The fusion module is used to obtain a blood vessel segmentation image corresponding to the initial medical image according to the blood vessel mask image, the arteriovenous distance map and the preset fusion model.

[0042] In a third aspect, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method in any one of the embodiments of the first aspect are implemented.

[0043] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method in any one of the embodiments of the first aspect are implemented.

[0044] In a fifth aspect, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the method in any one of the embodiments of the first aspect are implemented.

[0045] The above-mentioned blood vessel segmentation method, device and storage medium obtain a blood vessel mask image in the initial medical image based on the initial medical image and a preset segmentation model, input the initial medical image and the blood vessel mask image into a preset regression model to obtain an arteriovenous distance map, and obtain a blood vessel segmentation image corresponding to the initial medical image based on the blood vessel mask image, the arteriovenous distance map and a preset fusion model. The blood vessel mask image and the blood vessel arteriovenous distance map can be extracted using deep learning, and the blood vessels can be effectively segmented from the background of the initial medical image. Then, feature fusion is performed based on the arteriovenous distance map and the initial medical image to obtain a more accurate artery and vein segmentation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a diagram of an application environment of a blood vessel segmentation method in an embodiment;

[0047] Figure 2 is a schematic flow chart of a blood vessel segmentation method in one embodiment;

[0048] Figure 3 is a schematic flow chart of a blood vessel segmentation method in another embodiment;

[0049] Figure 4 is a schematic flow chart of a blood vessel segmentation method in another embodiment;

[0050] Figure 5 is a schematic flow chart of a blood vessel segmentation method in another embodiment;

[0051] Figure 6 is a schematic flow chart of a blood vessel segmentation method in another embodiment;

[0052] Figure 7 is a schematic flow chart of a blood vessel segmentation method in another embodiment;

[0053] Figure 8 is a schematic flow chart of a blood vessel segmentation method in another embodiment;

[0054] Fig. 9 FIG. 4 is a structural block diagram of a blood vessel segmentation device in an embodiment. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] In recent years, the incidence of cancer has been increasing year by year. Among cancers that cause death in men, lung cancer is the leading cause. The treatment of lung cancer generally requires lesion resection surgery. The five-year survival rate after surgery is closely related to the quality of the surgery. Preoperative planning before surgery is particularly important during the operation. For the surgical planning of lung cancer, it is necessary to reconstruct the lung organ tissues and other organs in three dimensions using image reconstruction methods to involve surgical approaches, surgical treatment plans, etc. In three-dimensional reconstruction, a more critical part is vascular reconstruction. The spatial position relationship between blood vessels and surgical pathways, the spatial proximity relationship between blood vessels and lesions, and the surgical plan for intraoperative vascular ligation are all closely related to the success of the operation.

[0057] The pulmonary vascular tree has abundant branches. After entering from the hilum of the lung, the main arteries and veins fork and intersect many times until they reach the edge of the lung. Due to the long path of the vascular tree, the abundant branches, and the many branch variations, it is difficult to segment the complete and accurate pulmonary blood vessels using traditional algorithms based on filtering or region growing. Deep learning can adapt to a variety of vascular tree forms due to its good feature extraction and expression capabilities. However, the pulmonary arteries and veins have various forms of adhesion in the main trunk part and in the interspersed process in the lungs, and the grayscale enhancement of the increase and decrease data obtained by scanning the target object using medical imaging equipment is inconsistent. Using a deep learning neural network to directly segment the arteries and veins in the initial image scanned by the medical imaging equipment is prone to segmentation and classification errors.

[0058] Based on this, the present application embodiment provides a blood vessel segmentation method, which can be applied to Figure 1The computer device shown in the figure can be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, after obtaining the initial medical image of the target object scanned by the medical imaging device, the initial medical image is inserted into a preset segmentation model to obtain a vascular mask image. And according to the vascular mask image and the initial medical image, an arteriovenous distance map is obtained, and the vascular mask image and the arteriovenous distance map are fused to obtain a vascular segmentation image corresponding to the initial medical image. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a key, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0059] It should be noted that the blood vessel segmentation method provided in the embodiment of the present application can be executed by a computer tomography (CT) imaging device, which can be implemented as part or all of a computer device through software, hardware, or a combination of software and hardware. In the following method embodiments, the execution subject is a computer device as an example for explanation.

[0060] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0061] In one embodiment, Figure 2 As shown, a blood vessel segmentation method is provided, which is applied to Figure 1 The computer device in the example is used to illustrate, including the following steps:

[0062] S202: Obtain a blood vessel mask image in the initial medical image according to the initial medical image and a preset segmentation model.

[0063] The initial medical image is a pre-processed image of DICOM data obtained by scanning the target part of the target object using a CT imaging device. The DICOM data can be pre-processed and converted into volume data, and the volume data is normalized with a unified window width and window position. For example, an enhanced window can be used, and [40,400] can be used as the normalized mean and standard deviation. When the voxels in the volume data exceed the preset maximum range, the exceeded voxels are adjusted to the closest maximum range boundary value. For example, [40,400] is used as the normalized mean and standard deviation, that is, the voxels in the volume data need to be adjusted to -360 to +440. If the voxel is less than -360, it is adjusted to -360, and if the voxel is greater than +440, it is adjusted to +440. The target object can include a human body or an animal body. The target parts can include: lungs, heart, liver, kidneys, pancreas, etc. The preset segmentation model may include convolutional neural network models such as VB-Net neural network model, V-net neural network model, U-net neural network model, etc., which are not limited here. Among them, the preset segmentation model may adopt Adam training optimizer, and the loss function may adopt DiceLoss loss function, cross entropy loss function, etc.

[0064] Specifically, the initial medical image can be input into a preset segmentation model, and the blood vessel segmentation can be performed on the initial medical image to obtain a blood vessel mask image in the initial medical image. The initial medical image can also be input into a preset first network model using a first sampling resolution for rough segmentation to determine the target part in the initial medical image, and the image of the target part is intercepted in the initial medical image, and the image of the target part is input again into a preset second network model using a second sampling resolution for fine segmentation to obtain a blood vessel mask image. The first sampling resolution is greater than the second sampling resolution; and the difference between the first sampling resolution and the second sampling resolution is greater than or equal to a preset difference threshold. For example, if the preset difference threshold is 4.95, the first sampling resolution is 5mm, and the second sampling resolution is a maximum of 0.05mm.

[0065] S204, inputting the initial medical image and the vascular mask image into a preset regression model to obtain an arteriovenous distance map; the arteriovenous distance map is used to characterize the distance between each arterial vascular point in the initial medical image and the arterial vascular boundary and the distance between each venous vascular point in the initial medical image and the venous vascular boundary.

[0066] The preset regression model may include a dual-channel VB-Net neural network model, a V-net neural network model, a U-net neural network model or other convolutional neural network models, which are not limited here. The arteriovenous distance map is a distance map of two types of blood vessels, arteries and veins.

[0067] Specifically, the initial medical image and the vascular mask image are input into a preset regression model, the vascular mask image is used as an enhanced information constraint, the characteristic information of the blood vessels and non-blood vessels is extracted, the distance transformation is performed, and the arteriovenous distance map is output.

[0068] Optionally, during the training process of the preset regression model, the setting of the initial parameters can adopt the training parameters of the preset segmentation model for transfer learning, and the loss function adopts the L1Loss loss function; wherein, the regression gold standard of the preset regression model is the standard arteriovenous distance map obtained by distance transformation of the standard arteriovenous blood vessel map.

[0069] S206, obtaining a blood vessel segmentation image corresponding to the initial medical image according to the blood vessel mask image, the arteriovenous distance map and the preset fusion model.

[0070] Specifically, after obtaining the vascular mask image and the arteriovenous distance map, the vascular mask image and the arteriovenous distance map can be input into the fusion model to extract the features of the arterial and venous vessels, and obtain the vascular segmentation image corresponding to the fused initial medical image. Among them, the vascular mask image can provide the vascular boundary information of the blood vessel without distinguishing between arteries and veins, and the arteriovenous distance map can provide the vascular branch structure, diameter information and boundary information. The fusion model determines the structure of the artery and vein according to the vascular branch structure, determines the boundaries of each artery and vein using the diameter information and the vascular boundary information of the artery and vein, and obtains the segmentation results of the arteriovenous classification. Among them, the fusion model can be a convolutional neural network model such as a multi-channel VB-Net neural network model, a V-net neural network model, and a U-net neural network model.

[0071] Optionally, the initialization parameters used in the fusion model training process may be the training parameters of the preset segmentation model, and the preset segmentation model training parameters are used to perform transfer learning to complete the training of the fusion model.

[0072] In the above-mentioned blood vessel segmentation method, a blood vessel mask image in the initial medical image is obtained according to the initial medical image and a preset segmentation model, the initial medical image and the blood vessel mask image are input into a preset regression model to obtain an arteriovenous distance map, and a blood vessel segmentation image corresponding to the initial medical image is obtained according to the blood vessel mask image, the arteriovenous distance map and a preset fusion model. The blood vessel mask image and the blood vessel arteriovenous distance map can be extracted by deep learning, and the blood vessels can be effectively segmented from the background of the initial medical image. Then, feature fusion is performed based on the arteriovenous distance map and the initial medical image to obtain more accurate artery and vein segmentation results.

[0073] The above embodiment describes the blood vessel segmentation method. Now, an embodiment is used to further describe how to obtain a blood vessel segmentation image corresponding to an initial medical image based on a blood vessel mask image, an arteriovenous distance map, and a preset fusion model. Figure 3 As shown, according to the vascular mask image, the arteriovenous distance map and the preset fusion model, a vascular segmentation image corresponding to the initial medical image is obtained, including:

[0074] S302: Determine a target path of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map.

[0075] Specifically, after obtaining the vascular mask image, the vascular boundary information of each blood vessel without distinguishing between arteries and veins can be determined, and the arteriovenous distance map can provide the vascular branch structure, diameter information and boundary information of the arteries and veins. According to the vascular boundary information of the arteries and veins in the vascular mask image, the vascular branch structure, diameter information and boundary information in the arteriovenous distance map, the path of each blood vessel is determined, that is, the target path of each blood vessel in the vascular mask image is determined.

[0076] Further, in one embodiment, if Figure 4 As shown, according to the vascular mask image and the arteriovenous distance map, the target path of each blood vessel in the vascular mask image is determined, including:

[0077] S402: Determine the starting point of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map.

[0078] Specifically, after the vascular mask map is determined, the main vessels of the arterial vessels and the main vessels of the venous vessels can be determined according to the vascular mask map. After the main vessels of the arterial vessels and the main vessels of the venous vessels are determined, since there is a corresponding relationship between the vascular mask map and the arteriovenous distance map, the connected domain corresponding to the main vessels of the arterial vessels in the corresponding arteriovenous distance map can be determined according to the main vessels of the arterial vessels in the vascular mask map; and the connected domain corresponding to the main vessels of the venous vessels in the corresponding arteriovenous distance map can be determined according to the main vessels of the venous vessels in the vascular mask map. Then, the positions with the largest diameters in the connected domain corresponding to the main vessels of the arterial vessels and the connected domain corresponding to the main vessels of the venous vessels can be respectively found to be the starting points of each vessel. For example, after the vascular mask map is obtained, the position area of ​​the main vessels of the arterial vessels in the vascular mask image and the position area of ​​the main vessels of the venous vessels can be first determined according to the vascular mask map. Then, according to the position area of ​​the main vessels of the arterial vessels, the corresponding position area in the arteriovenous distance map is found as the connected domain of the main vessels of the arterial vessels. Similarly, according to the location area of ​​the main blood vessels of the veins, find the corresponding location area in the arteriovenous distance map as the connected domain of the main blood vessels of the veins. Then find the place with the largest distance value, that is, the thickest blood vessel, in the connected domain of the main blood vessels of the arteries as the starting point of the arteries. Find the place with the largest distance value, that is, the thickest blood vessel, in the connected domain of the main blood vessels of the veins as the starting point of the veins.

[0079] Optionally, before determining the starting point of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map, the blood vessel mask image in the initial medical image may be obtained according to the initial medical image and a preset segmentation model. Figure 6 As shown, the segmentation model includes a first network model and a second network model; according to the initial medical image and the preset segmentation model, a blood vessel mask image in the initial medical image is obtained, including:

[0080] S602, inputting the initial medical image into the first network model for rough segmentation to obtain a segmented image of the organs in the initial medical image.

[0081] S604, determining the position of the organ in the initial medical image according to the segmented image of the organ, and obtaining the minimum circumscribed cube of the organ.

[0082] S606, inputting the minimum circumscribed cube into the second network model for fine segmentation to obtain a blood vessel mask image; wherein the sampling resolution of the second network model is smaller than the sampling resolution of the first network model.

[0083] When performing rough segmentation, the sampling resolution of the first network model can be set to 5 mm; when performing fine segmentation, the sampling resolution of the second network model can be set to 0.5 mm. Both the first network model and the second network model can be VB-net neural network models.

[0084] Specifically, the initial medical image is input into the first network model, segmented, background information is removed, and only the foreground information containing the position of the organ is retained, that is, the segmented image of the organ in the initial medical image is obtained. After the segmented image of the organ is obtained, the segmented image of the organ can be used to determine the position of the organ in the initial medical image, and based on the position of the organ, the minimum circumscribed cube of the organ is taken. Then the minimum circumscribed cube is input into the second network model to determine the blood vessels of various levels in the organ and obtain a blood vessel mask image.

[0085] In this embodiment, the initial medical image is input into the first network model for rough segmentation to obtain a segmented image of the organ in the initial medical image, and the position of the organ is determined in the initial medical image according to the segmented image of the organ, and the minimum circumscribed cube of the organ is obtained, and the minimum circumscribed cube is input into the second network model for fine segmentation to obtain a vascular mask image. The position area of ​​the organ is accurately determined in the first network model based on a larger sampling resolution, and the target area for vascular segmentation is determined according to the position area of ​​the organ, and the image of the target area is input into the second network model based on a smaller sampling resolution to obtain an accurate vascular segmentation result in the organ, thereby improving the accuracy of vascular segmentation.

[0086] Based on the above determined blood vessel mask image, further, in one embodiment, Figure 7 As shown, according to the vascular mask image and the arteriovenous distance map, the starting point of each blood vessel in the vascular mask image is determined, including:

[0087] S702, using the segmented image of the organ to process the blood vessel mask image, to obtain the main blood vessels of the arterial blood vessels and the main blood vessels of the venous blood vessels.

[0088] Specifically, the segmented image of the organ is the mask image of the organ, and the vascular mask image can be subtracted from the vascular mask image by the intersection area of ​​the segmented image and the vascular mask image to obtain the filtering result, namely the main blood vessels of the arteries and the main blood vessels of the veins.

[0089] S704, determining a connected domain corresponding to the main blood vessel of the artery in the arteriovenous distance map according to the main blood vessel of the artery, and determining the point with the largest distance value in the connected domain corresponding to the main blood vessel of the artery as the starting point of the artery.

[0090] Specifically, according to the position area of ​​the main blood vessels of the arterial blood vessels in the vascular mask map, the connected domain corresponding to the main blood vessels of the arterial blood vessels in the same position area of ​​the arteriovenous distance map is determined, and the point with the largest diameter distance value in the connected domain corresponding to the main blood vessels of the arterial blood vessels is determined as the starting point of the arterial blood vessels.

[0091] S706, determining the connected domain corresponding to the main blood vessel of the venous blood vessel in the arteriovenous distance map according to the main blood vessel of the venous blood vessel, and determining the point with the largest distance value in the connected domain corresponding to the main blood vessel of the venous blood vessel as the starting point of the venous blood vessel.

[0092] Specifically, according to the position area of ​​the main blood vessels of the venous blood vessels in the vascular mask map, the connected domain corresponding to the main blood vessels of the venous blood vessels in the same position area of ​​the arteriovenous distance map is determined, and the point with the largest distance value in the connected domain corresponding to the main blood vessels of the venous blood vessels is determined as the starting point of the venous blood vessels.

[0093] In this embodiment, the vascular mask image is filtered through the segmented image of the organ to obtain the trunk area of ​​the arterial blood vessels and the trunk area of ​​the venous blood vessels of the organ, and the connected domain corresponding to the trunk blood vessels of the arterial blood vessels in the arteriovenous distance map is determined according to the trunk area of ​​the arterial blood vessels, and the connected domain corresponding to the trunk blood vessels of the venous blood vessels in the venous distance map is determined according to the trunk area of ​​the venous blood vessels, and the point with the largest distance value in the connected domain corresponding to the trunk blood vessels of the arterial blood vessels is determined as the starting point of the arterial blood vessels, and the point with the largest distance value in the connected domain corresponding to the trunk blood vessels of the venous blood vessels is determined as the starting point of the venous blood vessels, so that the trunk area of ​​the arterial blood vessels and the trunk area of ​​the venous blood vessels can be accurately determined, and the starting points of the arterial blood vessels and the venous blood vessels can be found.

[0094] S404: Determine the end point of each blood vessel in the blood vessel mask image according to the blood vessel mask image.

[0095] Specifically, the center line of each blood vessel in the blood vessel mask image is extracted; each point on the center line of each blood vessel is traversed using a depth traversal search algorithm, and the point with 1 neighboring point is used as the end point of each blood vessel. The end point is the distal end point of each blood vessel.

[0096] S406, normalizing each distance value in the arteriovenous distance map to obtain a processed distance map.

[0097] The distance value is the distance from the foreground point of each blood vessel in the arteriovenous distance map to the blood vessel boundary, for example, the distance from the foreground point to the boundary of the blood vessel cross section where the foreground point is located, that is, the radius value of the cross section.

[0098] Specifically, each distance value in the arteriovenous distance map is input into the normalized scaling transformation formula, which is: Where D is the distance value from the foreground point of each blood vessel to the corresponding blood vessel boundary, max(D) is the maximum distance value among the distance values, and D' is the value after the distance value from the foreground point of each blood vessel to the corresponding blood vessel boundary is transformed by the normalized scaling transformation formula. After the distance values ​​of each distance value are obtained by the normalized scaling transformation formula, a processed distance map can be formed according to the multiple distance values ​​obtained by the normalized scaling transformation formula. The value distribution of the transformed distance map has the following characteristics: from the starting point of the blood vessel to the end point of the blood vessel, the value on the center line of each blood vessel changes from small to large; for the same cross-section blood vessel, the value changes from small to large from the center to the boundary. Using the characteristics of the processed distance map, by searching the path between the starting point of the two categories of artery and vein to each end point, the correct path value changes smoothly from small to large, and there will be a sudden change in value when the path crosses.

[0099] S408: Determine the path between the starting point of each blood vessel and the end point of each blood vessel according to the processed distance graph and the shortest path search algorithm.

[0100] Specifically, the shortest path search algorithm is used to search the center line of the processed distance graph to obtain the path between the starting point of each blood vessel and the end point of each blood vessel. In other words, the end point of each blood vessel can correspond to the starting point of the artery and the starting point of the vein, respectively, to obtain two paths.

[0101] S410, determining a target path of each blood vessel according to a path between a starting point of each blood vessel and an end point of each blood vessel.

[0102] Specifically, for the end point of each blood vessel, the shortest path search algorithm is used to search for the path between the end point of each blood vessel and the starting point of the venous blood vessel and determine it as the first path of each blood vessel; and for the end point of each blood vessel, the shortest path search algorithm is used to search for the path between the starting point of the arterial blood vessel and the end point of each blood vessel and determine it as the second path of each blood vessel. Optionally, the first path and the second path can be derived respectively to obtain the first slope of the first path and the second slope of the second path, and the target path can be determined according to the first slope of the first path and the second slope of the second path. Further, the first slope and the second slope can be compared, and the target path can be determined according to the comparison result.

[0103] In this embodiment, since the value distribution of the distance map after normalization and scaling transformation has the characteristics that the value on the center line of the blood vessel changes from small to large from the main thick blood vessel to the peripheral thin blood vessel; and the value of the same cross-section blood vessel changes from small to large from the center to the boundary. Then, the characteristics of the distance map after normalization and scaling transformation can be used to search for the path from the starting point of the two blood vessels of artery and vein to each peripheral endpoint (end point), determine the correct target path according to the slope of the first path and the second path, accurately determine whether the starting point corresponding to each peripheral endpoint is the starting point of the arterial blood vessel or the starting point of the venous blood vessel, and then determine the correct path between the peripheral endpoint and the target starting point, that is, the target path of the arterial blood vessel or the venous blood vessel, to achieve the classification of blood vessel categories.

[0104] Further, in one embodiment, if Figure 5 As shown, according to the path between the starting point of each blood vessel and the end point of each blood vessel, the target path of each blood vessel is determined, including:

[0105] S502, deriving the first path of each blood vessel to obtain a first slope corresponding to the first path.

[0106] Specifically, by taking the derivative of each point on the first path of each blood vessel, a plurality of first slopes corresponding to the first path can be obtained.

[0107] S504, deriving the second path of each blood vessel to obtain a second slope corresponding to the second path.

[0108] Specifically, by taking the derivative of each point on the second path of each blood vessel, a plurality of second slopes corresponding to the second path can be obtained.

[0109] S506: Determine a target path of each blood vessel by combining the first slope, the second slope, and a preset slope threshold.

[0110] Specifically, each first slope is compared with a preset slope threshold, the absolute value of the first slope greater than the preset slope threshold is taken, and the first slopes are sorted according to the absolute value of each first slope.

[0111] Each second slope is compared with a preset slope threshold, an absolute value of a second slope greater than the preset slope threshold is taken, and the second slopes are sorted according to their absolute values.

[0112] According to the sorting order, the first slope with the largest absolute value of the first slope is compared with the second slope with the largest absolute value of the second slope, and the path corresponding to the larger slope is taken as the target path. If the first slope with the largest absolute value of the first slope is equal to the second slope with the largest absolute value of the second slope, the first slope and the second slope with the second largest absolute value in the sorting order are compared, and the path corresponding to the larger slope is taken as the target path. If the first slope and the second slope with the second largest absolute value are equal, the magnitude of the first slope and the second slope with the third largest absolute value is determined, and so on, until the target path is determined. For example, the first slope is 0.1, 0.2, -0.3, 0.4; the second slope is 0.11, -0.21, 0.31, 0.41; the preset slope threshold is 0.15, then the first slopes greater than the preset slope threshold are 0.2 and 0.4, and the second slopes greater than the preset slope threshold are 0.31 and 0.41. The absolute value of the first slope is taken and sorted as 0.4 and 0.2; the absolute value of the second slope is taken and sorted as 0.41 and 0.31. At this time, the first slope and the second slope values ​​that are ranked first are compared. If 0.41 is greater than 0.4, the second path corresponding to the second slope is the target path.

[0113] In this embodiment, the first path of each blood vessel is differentiated to obtain the first slope corresponding to the first path, and the second path of each blood vessel is differentiated to obtain the second slope corresponding to the second path. The first slope, the second slope, and the preset slope threshold are used to determine the target path of each blood vessel. The correct path, that is, the target path, can be accurately determined by the slope value of the path.

[0114] S304, inputting the blood vessel mask image, the arteriovenous distance map, and the target path of each blood vessel into a preset fusion model to obtain a blood vessel segmentation image corresponding to the initial medical image.

[0115] Among them, the vascular mask image provides the vascular boundary information that does not distinguish between arteries and veins, the target path of each blood vessel provides the branch structure information of the arteriovenous blood vessels, and the arteriovenous distance map provides the rough diameter and boundary information of the branch structure of the arteriovenous blood vessels. By inputting the vascular mask image, the arteriovenous distance map, and the target path of each blood vessel into the preset fusion model, the vascular segmentation image corresponding to the initial medical image can be obtained.

[0116] Optionally, the segmentation probability map corresponding to the vascular mask image, the arteriovenous distance map, and the target path of each blood vessel can be input into a preset fusion model to obtain a vascular segmentation image corresponding to the initial medical image. The segmentation probability map corresponding to the vascular mask image is a probability map of each pixel being a foreground point obtained when the initial medical image is input into a preset segmentation model for segmentation. Further, after the probability map is activated using an activation function, the above-mentioned vascular mask image can be obtained.

[0117] In this embodiment, the target path of each blood vessel in the blood vessel mask image is determined according to the blood vessel mask image and the arteriovenous distance map, and the blood vessel mask image, the arteriovenous distance map, and the target path of each blood vessel are input into a preset fusion model to obtain a blood vessel segmentation image corresponding to the initial medical image. The path from each terminal end point to the starting point of the arterial and venous blood vessel trunks can be searched on the regressed arteriovenous distance map by the shortest path search algorithm, and according to the change characteristics of the distance path weight after the normalized scaling transformation, a reasonable target path can be determined to obtain a blood vessel segmentation image with an arterial blood vessel path and a venous blood vessel path, and an accurate blood vessel segmentation result can be obtained. Since blood vessels generally have a certain width, the algorithm based on path search has higher fault tolerance and better interpretability than the method based on threshold separation.

[0118] The above embodiment describes the blood vessel segmentation method. After obtaining the blood vessel segmentation image, it can be further segmented and graded to form a complete tree-level branch structure. Figure 8 As shown, the blood vessel segmentation method also includes:

[0119] S802: Determine the bifurcation point of each blood vessel in the blood vessel segmentation image according to the shortest path search method.

[0120] S804: Segment and classify the blood vessel segments between adjacent bifurcation points in each blood vessel according to a preset segment and classify rule to obtain a structural diagram of each blood vessel.

[0121] Specifically, the shortest path search algorithm is used to search the center line of the blood vessel segmentation image to obtain the bifurcation point of each blood vessel. Triggered from the starting point of the arterial and venous trunks, the classification rule is implemented when encountering the bifurcation point, and the arterial and venous blood vessels are divided into different levels from coarse to fine, and the segment between each two bifurcation points is regarded as a different segment to obtain the structural diagram of each blood vessel. Among them, the shortest path algorithm includes but is not limited to the Dijkstra algorithm, such as the Bellman-ford algorithm, the SPFA algorithm, the Floyd algorithm, etc.

[0122] In this embodiment, the bifurcation points of each blood vessel in the blood vessel segmentation image are determined according to the shortest path search method, and the blood vessel segments between adjacent bifurcation points in each blood vessel are segmented and graded according to preset segmentation and classification rules to obtain a structural diagram of each blood vessel, thereby obtaining a structural diagram of the blood vessels with a clear tree-level structure.

[0123] In order to facilitate the understanding of those skilled in the art, the blood vessel segmentation method is further described with an embodiment. In one embodiment, the blood vessel segmentation method includes:

[0124] S10, inputting the initial medical image into the first network model for rough segmentation to obtain a segmented image of the organs in the initial medical image.

[0125] S20, determining the position of the organ in the initial medical image according to the segmented image of the organ, and obtaining the minimum circumscribed cube of the organ.

[0126] S30, inputting the minimum circumscribed cube into the second network model for fine segmentation to obtain a blood vessel mask image; wherein the sampling resolution of the second network model is smaller than the sampling resolution of the first network model.

[0127] S40, inputting the initial medical image and the vascular mask image into a preset regression model to obtain an arteriovenous distance map; the arteriovenous distance map is used to characterize the distance between each arterial vascular point in the initial medical image and the arterial vascular boundary and the distance between each venous vascular point in the initial medical image and the venous vascular boundary.

[0128] S50, processing the blood vessel mask image using the segmented image of the organ to obtain the main blood vessels of the arterial blood vessels and the main blood vessels of the venous blood vessels.

[0129] S60, determining a connected domain corresponding to the main blood vessel of the artery in the arteriovenous distance map according to the main blood vessel of the artery, and determining the point with the largest distance value in the connected domain corresponding to the main blood vessel of the artery as the starting point of the artery.

[0130] S70, determining a connected domain corresponding to the main blood vessel of the venous blood vessel in the arteriovenous distance map according to the main blood vessel of the venous blood vessel, and determining the point with the largest distance value in the connected domain corresponding to the main blood vessel of the venous blood vessel as the starting point of the venous blood vessel.

[0131] S80, determining the end point of each blood vessel in the blood vessel mask image according to the blood vessel mask image.

[0132] S90, normalizing each distance value in the arteriovenous distance map to obtain a processed distance map.

[0133] S100, determining the path between the starting point of each blood vessel and the end point of each blood vessel according to the processed distance graph and the shortest path search algorithm.

[0134] S110, deriving the first path of each blood vessel to obtain a first slope corresponding to the first path.

[0135] S120, deriving the second path of each blood vessel to obtain a second slope corresponding to the second path.

[0136] S130, determining a target path of each blood vessel using the first slope, the second slope, and a preset slope threshold.

[0137] S140, inputting the blood vessel mask image, the arteriovenous distance map, and the target path of each blood vessel into a preset fusion model to obtain a blood vessel segmentation image corresponding to the initial medical image.

[0138] S150, determining the bifurcation point of each blood vessel in the blood vessel segmentation image according to the shortest path search method.

[0139] S160, segmenting and grading the blood vessel segments between adjacent bifurcation points in each blood vessel according to a preset segmentation and grading rule to obtain a structural diagram of each blood vessel.

[0140] In this embodiment, a vascular mask image in the initial medical image is obtained based on the initial medical image and a preset segmentation model, the initial medical image and the vascular mask image are input into a preset regression model to obtain an arteriovenous distance map, and a vascular segmentation image corresponding to the initial medical image is obtained based on the vascular mask image, the arteriovenous distance map and a preset fusion model. The vascular mask image and the vascular arteriovenous distance map can be extracted by deep learning, and the blood vessels can be effectively segmented from the background of the initial medical image. Then, feature fusion is performed based on the arteriovenous distance map and the initial medical image to obtain more accurate artery and vein segmentation results.

[0141] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0142] Based on the same inventive concept, the embodiment of the present application also provides a blood vessel segmentation device for implementing the blood vessel segmentation method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more blood vessel segmentation device embodiments provided below can refer to the limitations on the blood vessel segmentation method above, and will not be repeated here.

[0143] In one embodiment, Fig. 9 As shown, a blood vessel segmentation device is provided, comprising:

[0144] The segmentation module 901 is used to obtain a blood vessel mask image in the initial medical image according to the initial medical image and a preset segmentation model;

[0145] The regression module 902 is used to input the initial medical image and the vascular mask image into a preset regression model to obtain an arteriovenous distance map; the arteriovenous distance map is used to represent the distance between each arterial vascular point in the initial medical image and the arterial vascular boundary and the distance between each venous vascular point in the initial medical image and the venous vascular boundary;

[0146] The fusion module 903 is used to obtain a blood vessel segmentation image corresponding to the initial medical image according to the blood vessel mask image, the arteriovenous distance map and a preset fusion model.

[0147] The target object determination device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0148] In one embodiment, the fusion module includes:

[0149] A first determination unit, configured to determine a target path of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map;

[0150] The second determination unit is used to input the blood vessel mask image, the arteriovenous distance map, and the target path of each blood vessel into a preset fusion model to obtain a blood vessel segmentation image corresponding to the initial medical image.

[0151] In one embodiment, the first determination unit is specifically used to determine the starting point of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map; determine the end point of each blood vessel in the blood vessel mask image according to the blood vessel mask image; normalize each distance value in the arteriovenous distance map to obtain a processed distance map; determine the path between the starting point of each blood vessel and the end point of each blood vessel according to the processed distance map and the shortest path search algorithm; determine the target path of each blood vessel according to the path between the starting point of each blood vessel and the end point of each blood vessel.

[0152] In one embodiment, the starting point of each blood vessel includes the starting point of a venous blood vessel and the starting point of an arterial blood vessel; the first determination unit is specifically used to use the shortest path search algorithm for the end point of each blood vessel, search for the path between the end point of each blood vessel and the starting point of the venous blood vessel and determine it as the first path of each blood vessel; and use the shortest path search algorithm for the end point of each blood vessel, search for the path between the starting point of the arterial blood vessel and the end point of each blood vessel and determine it as the second path of each blood vessel.

[0153] In one embodiment, the first determination unit is specifically used to derive the first path of each blood vessel to obtain a first slope corresponding to the first path; derive the second path of each blood vessel to obtain a second slope corresponding to the second path; and determine the target path of each blood vessel by combining the first slope, the second slope, and a preset slope threshold.

[0154] In one embodiment, the segmentation model includes a first network model and a second network model; the segmentation module includes:

[0155] A coarse segmentation unit, used for inputting the initial medical image into the first network model for coarse segmentation to obtain a segmented image of the organ in the initial medical image;

[0156] A second determining unit is used to determine the position of the organ in the initial medical image according to the segmented image of the organ, and obtain the minimum circumscribed cube of the organ;

[0157] The fine segmentation unit inputs the minimum circumscribed cube into the second network model for fine segmentation to obtain a blood vessel mask image; wherein the sampling resolution of the second network model is smaller than the sampling resolution of the first network model.

[0158] In one embodiment, the first determination unit is specifically used to process the blood vessel mask image using the segmented image of the organ to obtain the main blood vessels of the arterial blood vessels and the main blood vessels of the venous blood vessels; according to the main blood vessels of the arterial blood vessels, determine the connected domain corresponding to the main blood vessels of the arterial blood vessels in the arteriovenous distance map, and determine the point with the largest distance value in the connected domain corresponding to the main blood vessels of the arterial blood vessels as the starting point of the arterial blood vessels; according to the main blood vessels of the venous blood vessels, determine the connected domain corresponding to the main blood vessels of the venous blood vessels in the arteriovenous distance map, and determine the point with the largest distance value in the connected domain corresponding to the main blood vessels of the venous blood vessels as the starting point of the venous blood vessels.

[0159] In one embodiment, the blood vessel segmentation device further comprises:

[0160] A determination module, used for determining the bifurcation point of each blood vessel in the blood vessel segmentation image according to the shortest path search method;

[0161] The classification module is used to classify the blood vessel segments between adjacent bifurcation points in each blood vessel according to a preset classification rule to obtain a structural diagram of each blood vessel.

[0162] The target object determination device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0163] Each module in the above-mentioned blood vessel segmentation device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0164] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the method in any one of the above embodiments are implemented.

[0165] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method in any of the above embodiments are implemented.

[0166] In one embodiment, a computer program product is provided, including a computer program, which implements the steps of the method in any one of the above embodiments when executed by a processor.

[0167] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0168] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0169] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0170] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A blood vessel segmentation method, characterized in that: The method comprises: Obtaining a blood vessel mask image in the initial medical image according to the initial medical image and a preset segmentation model; Inputting the initial medical image and the vascular mask image into a preset regression model to obtain an arteriovenous distance map; the arteriovenous distance map is used to characterize the distance between each arterial vascular point in the initial medical image and the arterial vascular boundary and the distance between each venous vascular point in the initial medical image and the venous vascular boundary; Obtaining a blood vessel segmentation image corresponding to the initial medical image according to the blood vessel mask image, the arteriovenous distance map and a preset fusion model; Wherein, obtaining the blood vessel segmentation image corresponding to the initial medical image according to the blood vessel mask image, the arteriovenous distance map and the preset fusion model includes: determining a target path of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map; The blood vessel mask image, the arteriovenous distance map, and the target path of each of the blood vessels are input into a preset fusion model to obtain a blood vessel segmentation image corresponding to the initial medical image.

2. The method according to claim 1, characterized in that Determining the target path of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map includes: Determining the starting point of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map; Determining, according to the blood vessel mask image, the end point of each blood vessel in the blood vessel mask image; Normalizing each distance value in the arteriovenous distance map to obtain a processed distance map; Determining a path between a starting point of each of the blood vessels and an end point of each of the blood vessels according to the processed distance graph and a shortest path search algorithm; The target path of each blood vessel is determined according to the path between the starting point of each blood vessel and the end point of each blood vessel.

3. The method according to claim 2, characterized in that The starting points of the blood vessels include the starting points of the veins and the starting points of the arteries; and determining the paths between the starting points of the blood vessels and the end points of the blood vessels according to the processed distance graph and the shortest path search algorithm includes: For the end point of each of the blood vessels, the shortest path search algorithm is used to search for a path between the end point of each of the blood vessels and the starting point of the venous blood vessel and determine it as the first path of each of the blood vessels; as well as, For the end point of each of the blood vessels, the shortest path search algorithm is used to search for a path between the starting point of the arterial blood vessel and the end point of each of the blood vessels and determine it as the second path of each of the blood vessels.

4. The method according to claim 3, characterized in that Determining the target path of each blood vessel according to the path between the starting point of each blood vessel and the end point of each blood vessel includes: Deriving the first path of each of the blood vessels to obtain a first slope corresponding to the first path; Deriving the second path of each of the blood vessels to obtain a second slope corresponding to the second path; The first slope, the second slope, and a preset slope threshold are used to determine a target path of each of the blood vessels.

5. The method according to claim 2, characterized in that: The segmentation model includes a first network model and a second network model; the step of obtaining a blood vessel mask image in the initial medical image according to the initial medical image and the preset segmentation model includes: Inputting the initial medical image into the first network model for rough segmentation to obtain a segmented image of the organ in the initial medical image; According to the segmented image of the organ, determining the position of the organ in the initial medical image, and obtaining the minimum circumscribed cube of the organ; The minimum circumscribed cube is input into the second network model for fine segmentation to obtain the blood vessel mask image; wherein the sampling resolution of the second network model is smaller than the sampling resolution of the first network model.

6. The method according to claim 5, characterized in that The step of determining the starting point of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map comprises: Processing the blood vessel mask image using the segmented image of the organ to obtain the main blood vessels of the arterial blood vessels and the main blood vessels of the venous blood vessels; According to the main blood vessel of the artery, determining the connected domain corresponding to the main blood vessel of the artery in the arteriovenous distance map, and determining the point with the largest distance value in the connected domain corresponding to the main blood vessel of the artery as the starting point of the artery; According to the main blood vessel of the venous blood vessel, the connected domain corresponding to the main blood vessel of the venous blood vessel in the arteriovenous distance map is determined, and the point with the largest distance value in the connected domain corresponding to the main blood vessel of the venous blood vessel is determined as the starting point of the venous blood vessel.

7. The method according to any one of claims 1 to 6, characterized in that: The method further comprises: Determining the bifurcation point of each blood vessel in the blood vessel segmentation image according to the shortest path search method; According to the preset segmentation and classification rules, the blood vessel segments between adjacent bifurcation points in each of the blood vessels are segmented and classified to obtain the structural diagram of each of the blood vessels.

8. A blood vessel segmentation device, characterized in that: The device comprises: A segmentation module, used for obtaining a blood vessel mask image in the initial medical image according to the initial medical image and a preset segmentation model; A regression module, used for inputting the initial medical image and the vascular mask image into a preset regression model to obtain an arteriovenous distance map; the arteriovenous distance map is used for representing the distance between each arterial vascular point in the initial medical image and the arterial vascular boundary and the distance between each venous vascular point in the initial medical image and the venous vascular boundary; A fusion module, used for obtaining a blood vessel segmentation image corresponding to the initial medical image according to the blood vessel mask image, the arteriovenous distance map and a preset fusion model; Wherein, the fusion module includes: A first determining unit, configured to determine a target path of each blood vessel in the blood vessel mask image according to the blood vessel mask image and the arteriovenous distance map; The second determination unit is used to input the blood vessel mask image, the arteriovenous distance map, and the target path of each of the blood vessels into a preset fusion model to obtain a blood vessel segmentation image corresponding to the initial medical image.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images