Vessel recognition system, apparatus and storage medium

Abstract

The application discloses a blood vessel identification system, device and storage medium. The system comprises a processor configured to execute the following blood vessel identification method, which comprises: inputting a blood vessel segmentation template into a trained first image processing model to obtain a global feature map; determining a structure point set corresponding to the blood vessel in the blood vessel segmentation template, determining an adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and feature data of each structure point in the global feature map, and the structure point set comprising position coordinates of at least two structure points used for representing a blood vessel direction; inputting the structure point set, the adjacency matrix and the feature data corresponding to each structure point in the structure point set into a trained second image processing model to obtain an identification result; and determining a target class identification of each blood vessel in the blood vessel segmentation template according to the identification result and the blood vessel segmentation template. The application solves the problem that the existing liver vein and portal vein distinguishing method is low in robustness.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, in particular to a blood vessel identification system, device and storage medium. BACKGROUND

[0002] The blood vessels of the liver have two hepatic arteries and one hepatic vein. The hepatic arteries are the hepatic artery and the portal vein, and the hepatic vein is the hepatic vein. The hepatic artery branches into the proper hepatic artery, and the portal vein enters the liver through the hepatic portal and repeatedly branches into interlobular arteries and interlobular veins, which further branch into liver lobules and converge into blood sinuses. In the blood sinuses, arterial blood and venous blood are mixed, and after material exchange with hepatocytes, they converge into central veins and then into sublobular veins, and finally into the hepatic vein to be discharged into the inferior vena cava. Due to the development, the hepatic artery is relatively bright in the arterial phase, but the hepatic vein and the portal vein are basically invisible, and only the relatively obvious veins and portal veins can be seen in the venous phase or the parenchymal phase. When CT or MRI (magnetic resonance imaging) enhancement scanning is performed, due to the different blood supply of the hepatic artery and the portal vein, four phases will appear: arterial phase, portal vein phase, parenchymal phase and delay phase. SUMMARY

[0003] The present application provides a blood vessel identification system, device and storage medium to solve the problem of low robustness in the existing method of distinguishing the hepatic vein and the portal vein.

[0004] According to one aspect of the present application, a blood vessel identification system is provided, the system comprising a processor configured to execute the following blood vessel identification method, comprising:

[0005] inputting a pre-created blood vessel segmentation template into a trained first image processing model to obtain a global feature map, wherein the blood vessel segmentation template comprises at least two types of blood vessels;

[0006] determining a structure point set corresponding to the blood vessels in the blood vessel segmentation template, determining an adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and feature data in the global feature map, and the structure point set comprises position coordinates of at least two structure points for representing the direction of the blood vessels;

[0007] inputting the structure point set, the adjacency matrix and the feature data corresponding to each structure point in the structure point set into a trained second image processing model to obtain an identification result, each structure point in the identification result is identified with an initial class identification for representing the class of the blood vessels;

[0008] determining the target class identification of each blood vessel in the blood vessel segmentation template according to the identification result and the blood vessel segmentation template.

[0009] According to another aspect of the present application, there is provided a blood vessel recognition device, comprising:

[0010] a feature map module configured to input a pre-created blood vessel segmentation template into a trained first image processing model to obtain a global feature map, wherein the blood vessel segmentation template comprises at least two types of blood vessels;

[0011] an input data determination module configured to determine a structure point set corresponding to the blood vessels in the blood vessel segmentation template, determine an adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and feature data in the global feature map, and the structure point set comprises position coordinates of at least two structure points used to represent a blood vessel direction;

[0012] a first identification module configured to input the structure point set, the adjacency matrix and the feature data corresponding to each structure point in the structure point set into a trained second image processing model to obtain an identification result, and each structure point in the identification result is identified with an initial class identification used to represent a blood vessel class;

[0013] a second identification module configured to determine a target class identification of each blood vessel in the blood vessel segmentation template according to the identification result and the blood vessel segmentation template.

[0014] According to another aspect of the present application, there is provided a computer readable storage medium storing computer instructions for causing a processor to implement the blood vessel recognition method according to any one of the embodiments of the present application when executed, comprising:

[0015] inputting a pre-created blood vessel segmentation template into a trained first image processing model to obtain a global feature map, wherein the blood vessel segmentation template comprises at least two types of blood vessels;

[0016] determining a structure point set corresponding to the blood vessels in the blood vessel segmentation template, determining an adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and feature data in the global feature map, and the structure point set comprises position coordinates of at least two structure points used to represent a blood vessel direction;

[0017] inputting the structure point set, the adjacency matrix and the feature data corresponding to each structure point in the structure point set into a trained second image processing model to obtain an identification result, and each structure point in the identification result is identified with an initial class identification used to represent a blood vessel class;

[0018] determining a target class identification of each blood vessel in the blood vessel segmentation template according to the identification result and the blood vessel segmentation template.

[0019] Compared with the prior art, the technical scheme provided by the embodiment of the application has the following advantages: the global feature map carries the position information of each blood vessel in the blood vessel segmentation template; the structure point set can determine the direction of the blood vessels through the positions of the structure points; the corresponding feature data of each structure point in the global feature map realizes the association of each structure point with the global feature map; the second image processing model can extract the features of the structure points in the global feature map and determine the relative position relationship between the structure points based on the adjacency matrix of each structure point; therefore, by inputting the structure point set, the adjacency matrix of each structure point in the blood vessel segmentation template, and the feature data of each structure point in the global feature map into the trained second image processing model, the initial class identifier of each structure point in the at least two structure points can be determined by the trained second image processing model based on the position relationship and the features of the at least two structure points, so that an accurate identification result is obtained, and the target class identifier of each blood vessel in the blood vessel segmentation template can be accurately determined based on the identification result and the blood vessel segmentation template, which helps to improve the accuracy of clinical diagnosis.

[0020] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the application, nor is it used to limit the scope of the application. Other features of the application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can also be obtained by those skilled in the art without creative effort.

[0022] Figure 1 is a structural block diagram of a blood vessel recognition system provided according to an embodiment of the application;

[0023] Figure 2 is a flowchart of a blood vessel recognition method provided according to an embodiment of the application;

[0024] Figure 3 is a flowchart of a blood vessel recognition method provided according to another embodiment of the application;

[0025] Figure 4 is a flowchart of a model training method provided according to another embodiment of the application;

[0026] Figure 5 is a structural block diagram of a blood vessel recognition device for implementing the embodiments of the application. DETAILED DESCRIPTION

[0027] In the following, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work should fall within the protection scope of the present application.

[0028] It should be noted that the terms "first", "second" and the like in the description and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in other than the order illustrated or described herein. In addition, the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a list of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0029] Embodiments

[0030] Figure 1 The blood vessel recognition system provided by the embodiments of the present application includes a processor 101, a memory 102 and a computer program stored in the memory 102, and the processor 101 implements the blood vessel recognition method when executing the computer program. In an embodiment, the system includes the processor 101 and the memory 102, and further includes an input device 103 and an output device 104; wherein the number of the processor 101 can be one or more, Figure 1 In the embodiment, the processor 101 in the device is taken as an example; the processor 101, the memory 102, the input device 103 and the output device 104 in the device can be connected through a bus or other means, Figure 1 In the embodiment, the connection through the bus is taken as an example.

[0031] The memory 102 as a computer readable storage medium can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the blood vessel recognition method in the embodiments of the present application. The processor 101 executes various function applications and data processing of the device by running the software programs, instructions and modules stored in the memory 102, that is, implements the blood vessel recognition method described in any embodiment.

[0032] The memory 102 can include a program storage area and a data storage area, where the program storage area can store an operating system, application programs required by at least one function, and the data storage area can store data created according to the use of the terminal, etc. In addition, the memory 102 can include a high-speed random access memory, and can also include a non-volatile memory such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some examples, the memory 102 can further include a memory disposed remotely with respect to the processor 101, which can be connected to the device through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0033] The input device 103 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function controls of the device. The output device 104 can include a display device such as a display screen of the user terminal.

[0034] Figure 2 A flowchart of a blood vessel identification method provided by another embodiment of the present application, which can be applicable to the case of distinguishing each blood vessel category in a blood vessel segmentation template based on a trained first image processing model and a second image processing model. The method can be executed by a blood vessel identification device, which can be realized in the form of hardware and / or software and configured in a processor of a blood vessel identification system. As shown in the figure, the method includes: Figure 2

[0035] S110, inputting a pre-created blood vessel segmentation template into the trained first image processing model to obtain a global feature map, wherein the blood vessel segmentation template includes at least two types of blood vessels.

[0036] The blood vessel segmentation template is created based on a blood vessel segmentation result, which is a binary image with pixel values of 1 for the blood vessel part and 0 for the background part. The blood vessel segmentation result is obtained by performing blood vessel segmentation on the to-be-analyzed image data using an existing image segmentation method. It can be understood that after obtaining the blood vessel segmentation result, a binary processing is performed on the blood vessel segmentation result to obtain the blood vessel segmentation template, where the binary processing process includes assigning a value of 1 to the pixels of the blood vessel part and a value of 0 to the pixels of the background part. The to-be-analyzed image data is existing clinical blood vessel diagnosis image data, such as contrast-enhanced CT (Computed Tomography) image data.

[0037] In one embodiment, the blood vessel segmentation template includes at least two types of blood vessels, such as the hepatic vein and the portal vein.

[0038] ​The first image processing model can be a neural network model or a three-dimensional semantic segmentation network. The global feature map can be the feature map output by the last layer of the up-sampling network of the trained first image processing model, and the size of the global feature map is the same as the size of the blood vessel segmentation template received by the trained first image processing model.

[0039] In one embodiment, the blood vessel segmentation template is down-sampled to update the blood vessel segmentation template, and the updated blood vessel segmentation template is input into the trained first image processing model to obtain the global feature map. Compared with the original image data, the binary blood vessel segmentation template as the model input data can reduce the data operation amount of the trained first image processing model. Down-sampling can reduce the resolution of the blood vessel segmentation template, but does not change the relative position relationship of each pixel in the blood vessel segmentation template. Therefore, down-sampling the blood vessel segmentation template can further reduce the data operation amount of the trained first image processing model, thereby greatly improving the data processing speed of the trained first image processing model. In this embodiment, the size of the updated blood vessel segmentation template can be 128x128x128, and the size of the global feature map is also 128x128x128.

[0040] In S120, a structure point set corresponding to the blood vessels in the blood vessel segmentation template is determined, and an adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and feature data of each structure point in the global feature map are determined. The structure point set includes position coordinates of at least two structure points used to represent the blood vessel direction.

[0041] The structure point set is a set of structure points, which includes position coordinates of at least two structure points used to represent the blood vessel direction. Specifically, the structure point set includes an identifier of each structure point and the position coordinates corresponding to the identifier.

[0042] The structure point set of the blood vessels in the blood vessel segmentation template is determined by the following steps: determining the center line of each blood vessel in the blood vessel segmentation template; and determining at least two structure points of each blood vessel according to the center line of each blood vessel to obtain the structure point set containing structure points of all blood vessels. By taking the pixel points in the center line of each blood vessel that meet the set condition as at least two structure points that can reflect the direction of each blood vessel, the accuracy of structure point determination can be improved, because each blood vessel has a unique center line regardless of the occurrence of variant blood vessels or the occurrence of sticky blood vessels.

[0043] In one embodiment, all pixel points on the center line of each blood vessel that meet the set condition are taken as the structure points of the corresponding blood vessel. The set condition can be that if any pixel on the center line is a bifurcation point or an end point of the blood vessel, the pixel is taken as a structure point; and the first pixel in the continuous N pixels that do not contain the bifurcation point or the end point is taken as a structure point, where N is greater than 1.

[0044] wherein the adjacency matrix is used to represent the adjacency relationship between points. In one embodiment, the adjacency matrix of each structure point in the structure point set in the vessel segmentation template is determined based on a 26-neighborhood method. Specifically, the structure point is regarded as a cube, and thus the neighborhood is 6 neighborhoods along the face of the cube, 8 neighborhoods along the corner of the cube, and 12 neighborhoods connected to the edge of the cube. The adjacency matrix can be used to determine the spatial position of the structure point in the vessel segmentation template.

[0045] In one embodiment, the reference plane position of each structure point in the structure point set in the global feature map is determined, and the feature data with the same reference plane position is taken as the feature data of the corresponding structure point. Wherein the reference plane position can be selected as the x-y plane position. For example, if the position coordinates of the structure point are (x1, y1, z1), the feature data with the x-y plane coordinates (x1, y1) in the global feature map is taken as the feature data of the structure point. This embodiment is used to establish the association relationship between the structure point and the global feature map.

[0046] S130, input the structure point set, the adjacency matrix and the feature data corresponding to each structure point in the structure point set into the trained second image processing model to obtain an identification result, and each structure point in the identification result is identified with an initial class label for representing a vessel class.

[0047] The second image processing model can be selected as a graph attention network model or a graph convolution network model, which can extract the features of the structure points in the global feature map, determine the relative position relationship between the structure points based on the adjacency matrix of each structure point, and thus can determine the initial class label of each structure point in the at least two structure points by comprehensively considering the position relationship and the features of the at least two structure points.

[0048] In one embodiment, the vessel segmentation template includes a hepatic vein and a portal vein, the class label of the hepatic vein is 1, and the class label of the portal vein is 2. The initial class label is used to represent the vessel class to which the vessel where the structure point is located belongs.

[0049] In one embodiment, the structure point set and the adjacency matrix corresponding to each structure point in the structure point set are spliced to obtain a target adjacency matrix; the structure point set and the feature data corresponding to each structure point in the structure point set are spliced to obtain target feature data, and then the structure point set, the target adjacency matrix and the target feature data are input into the trained second image processing model to obtain an identification result.

[0050] S140, determining the target class label of each vessel in the vessel segmentation template according to the identification result and the vessel segmentation template.

[0051] The target category identifier is used to represent the blood vessel category to which the corresponding blood vessel belongs. It should be noted that for any blood vessel, the initial category identifier of the structure point thereof can be the same as or different from the target category identifier thereof. In an example one, the initial category identifier of the structure point of the hepatic vein is 1, and the initial category identifier of the structure point of the portal vein is 2; the target category identifier of the hepatic vein is A1, and the target category identifier of the portal vein is A2. In an example two, the target category identifier of the hepatic vein is the same as the initial category identifier of the structure point thereof, and the target category identifier of the portal vein is the same as the initial category identifier of the structure point thereof.

[0052] After the identification result is obtained, the target category identifier of each blood vessel in the blood vessel segmentation template is determined according to the identification result and the blood vessel segmentation result, so as to distinguish the categories of the blood vessels in the blood vessel segmentation template.

[0053] Compared with the prior art, the technical scheme provided by the embodiment of the present application has the following advantages: the global feature map carries the position information of each blood vessel in the blood vessel segmentation template; the structure point set can determine the blood vessel direction through the positions of the structure points; the corresponding feature data of each structure point in the global feature map realizes the association between the structure point and the global feature map; the second image processing model can extract the features of the structure points in the global feature map, and determine the relative position relationship between the structure points based on the adjacency matrix of each structure point; therefore, by inputting the structure point set, the adjacency matrix of each structure point in the blood vessel segmentation template, and the feature data of each structure point in the global feature map into the trained second image processing model, the initial category identifier of each structure point in at least two structure points can be determined by the trained second image processing model based on the position relationship and the features of the at least two structure points, so as to obtain an accurate identification result, and the target category identifier of each blood vessel in the blood vessel segmentation template can be accurately determined according to the identification result and the blood vessel segmentation template, which helps to improve the robustness of blood vessel recognition and the accuracy of clinical diagnosis.

[0054] Figure 3 The flowchart of the blood vessel recognition method provided by another embodiment of the present application is used to further refine the “determining the target category identifier of each blood vessel in the blood vessel segmentation template according to the identification result and the blood vessel segmentation template” in the above-mentioned embodiment. As shown in FIG. 8, the method comprises the following steps. Figure 3

[0055] S210, inputting a pre-created blood vessel segmentation template into a trained first image processing model to obtain a global feature map, wherein the blood vessel segmentation template comprises at least two types of blood vessels.

[0056] ​S220, determine a structure point set corresponding to the blood vessels in the blood vessel segmentation template, determine an adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and feature data in the global feature map, and the structure point set includes position coordinates of at least two structure points used to represent the blood vessel direction.

[0057] S230, input the structure point set, the adjacency matrix and the feature data corresponding to each structure point in the structure point set into the trained second image processing model to obtain a labeling result, and each structure point in the labeling result is labeled with an initial class label used to represent a blood vessel class.

[0058] S2401, based on a set label determination rule, unify the initial class labels of the structure points of each blood vessel in the labeling result to update the labeling result.

[0059] The more the number of blood vessels, the more the structure points included in the blood vessels, and when the blood vessel distribution is relatively complex, the determination of the initial class label of the individual structure point may have errors. Therefore, based on a set label determination rule, the initial class labels of the structure points of each blood vessel in the labeling result are unified to update the labeling result, so as to improve the accuracy of the structure point determination.

[0060] In one embodiment, the mode of the initial class labels of the structure points of each blood vessel in the labeling result is determined, and the mode is taken as the initial class label of the structure points of each blood vessel to update the labeling result. Specifically, the following steps are included:

[0061] Step a1, divide the blood vessels corresponding to each structure point in the labeling result into at least two blood vessel segments, and the blood vessel segment is the blood vessel between adjacent bifurcation points and end points, or the blood vessel between two adjacent bifurcation points.

[0062] Since the blood vessel segment is the blood vessel between adjacent bifurcation points and end points, or the blood vessel between two adjacent bifurcation points, each blood vessel segment is usually linear or close to linear, does not include a blood vessel bifurcation part, and has a relatively simple structure, so the determination of the initial class labels of the at least two structure points corresponding thereto is relatively uniform.

[0063] Step a2, determine the mode of the initial class labels of the structure points of each blood vessel segment in the labeling result, and take the mode as the initial class label of the structure points of the corresponding blood vessel segment to update the labeling result.

[0064] Since the mode is the data with the highest frequency in a group of data, in this embodiment, the mode of the initial class labels of the structure points of each blood vessel segment in the labeling result is determined, and the mode is taken as the initial class label of the structure points of the corresponding blood vessel segment. For example, blood vessel segment A includes five structure points, and the initial class labels of the five structure points are 1, 0, 0, 1 and 1 respectively. Since the mode of the five initial class labels is 1, the initial class labels of the five structure points are all updated to 1.

[0065] Since any blood vessel segment belongs to the same type of blood vessel, the initial class label of each structure point of the blood vessel segment that can be spliced is unified, and the initial class label of each structure point of each type of blood vessel is unified, so that the accuracy of the unified initial class label of the structure point is improved.

[0066] S2402, determining the target class label of each blood vessel in the blood vessel segmentation template according to the updated identification result and the blood vessel segmentation template.

[0067] Since the accuracy of the initial class label of the structure point of the blood vessel in the updated identification result is higher, the target class label of each blood vessel in the blood vessel segmentation template determined according to the updated identification result and the blood vessel segmentation template has higher accuracy.

[0068] In one embodiment, according to the updated identification result and the blood vessel segmentation template, the at least two types of blood vessels are reconstructed in the blood vessel segmentation template based on a region growing algorithm to obtain a target blood vessel segmentation template, and the at least two types of blood vessels in the target blood vessel segmentation template are labeled with corresponding target class labels. Through the region growing algorithm, the region growing of the structure points with the same initial class label in the identification result can be quickly and accurately performed to obtain the blood vessels of the corresponding type, and the purpose of quickly and accurately distinguishing the blood vessel classification in the blood vessel segmentation template is achieved.

[0069] Compared with the prior art, by unifying the initial class label of each structure point of the blood vessel segment that can be spliced, the initial class label of each structure point of each blood vessel is unified, which helps to improve the accuracy of the initial class label of each structure point, thereby improving the accuracy of the target blood vessel segmentation template determined according to the updated identification result and the blood vessel segmentation template, that is, improving the accuracy and robustness of distinguishing each type of blood vessel in the target blood vessel segmentation template.

[0070] Figure 4 The flowchart of the model training method provided by the embodiment of the present application is shown in FIG. 1. Figure 4 As shown in the figure, the method comprises the following steps:

[0071] S310, inputting the blood vessel segmentation template carrying the target class label into the first image processing model to complete the training of the first image processing model, wherein the blood vessel segmentation template comprises at least two types of blood vessels.

[0072] The first image processing model can be a neural network model or a three-dimensional semantic segmentation network model.

[0073] The blood vessel segmentation template carrying the target category identifier is down-sampled to a set size to update the blood vessel segmentation template, such as 128x128x128, and the updated blood vessel segmentation template is input into the first image processing model to start training of the first image processing model, and network parameters of the first image processing model are adjusted according to the target category identifier and a predicted category identifier determined by the first image processing model during the training until a difference between the predicted category identifier and the target category identifier is within a set error threshold range.

[0074] The first image processing model includes an up-sampling network. After the first image processing model is trained, a feature map output by the up-sampling network in the trained first image processing model is taken as a global feature map. The global feature map has the same size as the blood vessel segmentation template received by the first image processing model.

[0075] The loss function in the first image processing model can be a cross-entropy loss function or a Dice loss function.

[0076] S320, a structure point set corresponding to the blood vessels in the blood vessel segmentation template is determined, and an adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and feature data of each structure point in the global feature map are determined. The structure point set includes position coordinates of at least two structure points used to represent the blood vessel direction.

[0077] S330, the structure point set carrying the initial category identifier of each structure point, the adjacency matrix and the feature data corresponding to each structure point in the structure point set are input into the second image processing model to obtain an identification result, and each structure point in the identification result is identified with the initial category identifier used to represent the blood vessel category.

[0078] The loss function used by the second image processing model during model training can be a cross-entropy loss function or other existing classification loss function.

[0079] The structure point set carrying the initial category identifier of each structure point, the adjacency matrix and the feature data corresponding to each structure point in the structure point set are input into the second image processing model to start training of the second image processing model, and network parameters of the second image processing model are adjusted according to the known category identifier and a predicted category identifier determined by the second image processing model during the training until a difference between the predicted category identifier and the target category identifier is within a set error threshold range.

[0080] The model training method provided by the embodiment of the present application simply and effectively completes the training of the deep learning network fusing the first image processing model and the second image processing model, the first image processing model can extract features of a blood vessel segmentation template to obtain a global feature map, the second image processing model can extract features of structure points in the global feature map, the relative position relationship between the structure points is determined based on an adjacency matrix of each structure point, so that the initial class identifier of each structure point in at least two structure points can be determined by comprehensively considering the position relationship and the features of the at least two structure points.

[0081] Figure 5 The structure diagram of the blood vessel recognition device provided by another embodiment of the present application is shown in the figure. Figure 5 The device comprises:

[0082] The feature map module 11 is configured to input a pre-created blood vessel segmentation template into the trained first image processing model to obtain a global feature map, wherein the blood vessel segmentation template comprises at least two types of blood vessels.

[0083] The input data determination module 12 is configured to determine a structure point set corresponding to the blood vessels in the blood vessel segmentation template, determine an adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and feature data in the global feature map, and the structure point set comprises position coordinates of at least two structure points used to represent the direction of the blood vessels.

[0084] The first identification module 13 is configured to input the structure point set, the adjacency matrix and the feature data corresponding to each structure point in the structure point set into the trained second image processing model to obtain an identification result, and each structure point in the identification result is identified with an initial class identifier used to represent the class of the blood vessels.

[0085] The second identification module 14 is configured to determine the target class identifier of each blood vessel in the blood vessel segmentation template according to the identification result and the blood vessel segmentation template.

[0086] Optionally, the at least two types of blood vessels comprise hepatic veins and portal veins.

[0087] Optionally, the feature map module 11 is configured to down-sample the blood vessel segmentation template to update the blood vessel segmentation template, and input the updated blood vessel segmentation template into the trained first image processing model to obtain the global feature map.

[0088] Optionally, the input data determination module 12 is configured to determine the center line of each blood vessel in the blood vessel segmentation template, and determine at least two structure points of each blood vessel according to the center line of each blood vessel to obtain the structure point set corresponding to the blood vessels in the blood vessel segmentation template.

[0089] Optionally, the input data determination module 12 is configured to determine, based on a 26-neighborhood method, an adjacency matrix of each structure point in the set of structure points in the blood vessel segmentation template.

[0090] Optionally, the input data determination module 12 is configured to determine a reference plane position of each structure point in the set of structure points in the global feature map, and take feature data with the same reference position as feature data of the corresponding structure point.

[0091] Optionally, the second identification module 14 comprises:

[0092] a category identification updating unit configured to uniformly update initial category identifications of structure points of each blood vessel in the identification result based on a set label determination rule to update the identification result.

[0093] a category identification determination unit configured to determine target category identifications of each blood vessel in the blood vessel segmentation template according to the updated identification result and the blood vessel segmentation template.

[0094] Optionally, the category identification updating unit is configured to determine a mode of the initial category identifications of the structure points of each blood vessel in the identification result, and take the mode as the category identification of the structure points of the corresponding blood vessel to update the identification result.

[0095] Optionally, the category identification updating unit is specifically configured to divide a blood vessel corresponding to each structure point in the identification result into at least two vessel segments, the vessel segments being vessels between adjacent bifurcation points and end points, or vessels between two adjacent bifurcation points; determine a mode of the initial category identifications of the structure points of each vessel segment in the identification result, and take the mode as the initial category identification of the structure points of the corresponding vessel segment to update the identification result.

[0096] Optionally, the category identification determination unit is configured to reconstruct the at least two types of blood vessels in the blood vessel segmentation template based on a region growing algorithm according to the updated identification result and the blood vessel segmentation template to obtain a target blood vessel segmentation template, and each of the at least two types of blood vessels in the target blood vessel segmentation template is identified with a corresponding target category identification.

[0097] Optionally, the first image processing model is a neural network model or a three-dimensional semantic segmentation network model, and the second image processing model is a graph attention network model or a graph convolution network model.

[0098] Compared with the prior art, the technical scheme provided by the embodiment of the present application has the following advantages: the global feature map carries the position information of each blood vessel in the blood vessel segmentation template; the structure point set can determine the direction of the blood vessels through the positions of the structure points; the corresponding feature data of each structure point in the global feature map realizes the association of each structure point with the global feature map; the second image processing model can extract the features of the structure points in the global feature map and determine the relative position relationship between the structure points based on the adjacency matrix of each structure point; therefore, by inputting the structure point set, the adjacency matrix of each structure point in the blood vessel segmentation template, and the feature data of each structure point in the global feature map into the trained second image processing model, the initial class identifier of each structure point in the at least two structure points can be determined by the trained second image processing model based on the position relationship and the features of the at least two structure points, so that an accurate identification result is obtained, and the target class identifier of each blood vessel in the blood vessel segmentation template can be accurately determined based on the identification result and the blood vessel segmentation template, which helps to improve the accuracy of clinical diagnosis.

[0099] The blood vessel identification device provided by the embodiment of the present application can execute the blood vessel identification method provided by any embodiment of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0100] In some embodiments, the blood vessel identification method can be implemented as a computer program which is tangibly embodied in a computer readable storage medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the processor, one or more steps of the blood vessel identification method described above can be performed. Alternatively, in other embodiments, the processor can be configured to perform the blood vessel identification method by any other appropriate means (for example, by means of firmware). The blood vessel identification method includes the following steps:

[0101] inputting a pre-created blood vessel segmentation template into a trained first image processing model to obtain a global feature map, wherein the blood vessel segmentation template includes at least two types of blood vessels;

[0102] determining a structure point set corresponding to the blood vessels in the blood vessel segmentation template, determining the adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and the feature data of each structure point in the global feature map, and the structure point set includes the position coordinates of at least two structure points for representing the direction of the blood vessels;

[0103] inputting the structure point set, the adjacency matrix and the feature data corresponding to each structure point in the structure point set into a trained second image processing model to obtain an identification result, and each structure point in the identification result is identified with an initial class identifier for representing the class of the blood vessels;

[0104] According to the identification result and the blood vessel segmentation template, a target class of each blood vessel in the blood vessel segmentation template is determined.

[0105] Computer programs for implementing the methods of the present application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions / acts specified in the flow diagrams and / or block diagrams to be implemented. The computer programs can be executed in whole on a machine, partially on a machine, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0106] In the context of the present application, a computer readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, a computer readable storage medium can be a machine readable signal medium. More specific examples of the machine readable storage medium will include one or more lines of a program of instructions in a transitory signal form, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0107] It should be understood that various forms of flow shown above can be used with orders of the steps re-arranged, added or deleted. For example, the steps recited in the present application can be performed in parallel, in series, or in different orders, as long as the desired results of the technical solutions of the present application can be achieved, which are not limited herein.

[0108] The above detailed description does not constitute a limitation on the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A blood vessel recognition system, characterized by, The system comprises a processor configured to perform the following blood vessel identification method, comprising: inputting a pre-created blood vessel segmentation template into a trained first image processing model to obtain a global feature map, wherein the blood vessel segmentation template comprises at least two types of blood vessels and is created based on a blood vessel segmentation result, and is a binary graph; determining a structure point set corresponding to the blood vessels in the blood vessel segmentation template, determining an adjacency matrix of each structure point in the structure point set in the blood vessel segmentation template and feature data of each structure point in the global feature map, and the structure point set comprises position coordinates of at least two structure points used to represent the blood vessel direction; inputting the structure point set, the adjacency matrix and the feature data of each structure point in the structure point set into a trained second image processing model to obtain an identification result, and each structure point in the identification result is identified with an initial class label used to represent the blood vessel class; determining a target class label of each blood vessel in the blood vessel segmentation template according to the identification result and the blood vessel segmentation template.

2. The system of claim 1, wherein, The at least two types of blood vessels comprise hepatic veins and portal veins.

3. The system of claim 1, wherein, The inputting a pre-created blood vessel segmentation template into a trained first image processing model to obtain a global feature map comprises: down-sampling the blood vessel segmentation template to update the blood vessel segmentation template; inputting the updated blood vessel segmentation template into the trained first image processing model to obtain the global feature map.

4. The system of claim 1, wherein, The determining a structure point set corresponding to the blood vessels in the blood vessel segmentation template comprises: determining a center line of each blood vessel in the blood vessel segmentation template; determining at least two structure points of each blood vessel according to the center line of each blood vessel to obtain the structure point set corresponding to the blood vessels in the blood vessel segmentation template.

5. The system of claim 1, wherein, The blood vessel segmentation template is determined by the following steps, comprising: obtaining user-inputted to-be-analyzed image data of a target object; performing blood vessel segmentation on the to-be-analyzed image data to obtain a blood vessel segmentation result; performing binary processing on the blood vessel segmentation result to obtain the blood vessel segmentation template.

6. The system of claim 1, wherein, The feature data of each structure point in the structure point set in the global feature map is determined by the following steps, comprising: determining a reference plane position of each structure point in the structure point set in the global feature map, and taking feature data with the same reference position as the feature data of the corresponding structure point.

7. The system of claim 1, wherein, The determining a target class label of each blood vessel in the blood vessel segmentation template according to the identification result and the blood vessel segmentation template comprises: unifying the initial class labels of the structure points of each blood vessel in the identification result based on a set label determination rule to update the identification result; determining the target class label of each blood vessel in the blood vessel segmentation template according to the updated identification result and the blood vessel segmentation template.

8. The system of claim 7, wherein, The unifying the initial class labels of the structure points of each blood vessel in the identification result based on a set label determination rule to update the identification result comprises: determining a mode of the initial class labels of the structure points of each blood vessel in the identification result, and taking the mode as the class label of the structure points of the corresponding blood vessel to update the identification result.

9. The system of claim 8, wherein, The method comprises the following steps of: dividing each vessel in the identification result into at least two vessel segments, wherein the vessel segment is a vessel between adjacent branch points and end points, or a vessel between two adjacent branch points; determining the mode of the initial class identification of the structure points of each vessel segment in the identification result, and taking the mode as the initial class identification of the structure points of the corresponding vessel segment to update the identification result.

10. The system of claim 7, wherein, The method comprises the following steps of: reconstructing the at least two types of vessels in the vessel segmentation template based on a region growing algorithm according to the updated identification result and the vessel segmentation template, to obtain a target vessel segmentation template, and the at least two types of vessels in the target vessel segmentation template are all identified with corresponding target class identifications.

11. The system of claim 1, wherein, The first image processing model is a neural network model or a three-dimensional semantic segmentation network model, and the second image processing model is a graph attention network model or a graph convolution network model.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing the processor to implement the following vessel identification method when executed, comprising: inputting a pre-created vessel segmentation template into a trained first image processing model to obtain a global feature map, wherein the vessel segmentation template comprises at least two types of vessels, and is created based on a vessel segmentation result and is a binary graph; determining a structure point set corresponding to the vessels in the vessel segmentation template, determining an adjacency matrix of each structure point in the structure point set in the vessel segmentation template and feature data in the global feature map, and the structure point set comprises position coordinates of at least two structure points used to represent vessel orientations; inputting the structure point set, the adjacency matrix and the feature data corresponding to each structure point in the structure point set into a trained second image processing model to obtain an identification result, and each structure point in the identification result is identified with an initial class identification used to represent a vessel class; determining target class identifications of each vessel in the vessel segmentation template according to the identification result and the vessel segmentation template.

Images