A blood vessel centerline classification method, a classification device, an apparatus and a medium
By combining convolutional neural networks and graph neural networks, target pixel maps and graph structure features are generated, solving the problem of low efficiency in vascular centerline classification in existing technologies and achieving fast and accurate vascular tree classification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN RAYSIGHT INTELLIGENT MEDICAL TECH CO LTD
- Filing Date
- 2023-03-01
- Publication Date
- 2026-07-21
Smart Images
Figure CN116309421B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical image processing technology, and more specifically, to a method, device, equipment, and medium for classifying the center line of blood vessels. Background Technology
[0002] As people's living standards improve, the incidence of vascular diseases is also on the rise. Therefore, early screening for vascular diseases is crucial to preventing their progression. Midline classification involves categorizing each point on the midline, such as which blood vessel it belongs to, the type of lesion, and the degree of lesion severity.
[0003] Medical image analysis is an important tool for screening vascular diseases, but analyzing large amounts of image data requires significant medical resources. This is where automated analysis methods play a crucial role. The complex structure of the vascular tree in images makes automated analysis of this tree challenging.
[0004] Existing methods for classifying blood vessel centerlines, including lesion classification at each point and prediction of blood vessel diameter, mostly employ traditional convolutional neural networks for feature extraction, followed by feature analysis using fully connected layers. The advantage of this most widely used method is that it can extract local features well, but it ignores the connectivity of data with strong topological structures, such as vascular trees. Summary of the Invention
[0005] In view of this, the purpose of this application is to provide a classification method, classification device, equipment and medium for vascular centerlines, which utilizes a vascular tree with a topological structure composed of vascular centerlines to perform fast and accurate conversion from pixel images to graphs, and fully combines the advantages of convolutional neural networks and graph neural networks to extract and fuse node and edge features, and finally perform multi-task classification of centerlines.
[0006] In a first aspect, embodiments of this application provide a method for classifying the centerline of blood vessels, the classification method comprising:
[0007] Obtain the vascular tree centerline pixel map corresponding to the original vascular tree medical image, and generate a target pixel map based on the vascular tree centerline pixel map; wherein, the target pixel map includes multiple nodes and the number of each node, and each node in the target pixel map corresponds one-to-one with each pixel on the vascular centerline of the vascular tree centerline pixel map;
[0008] Based on the original vascular tree medical image and each node in the target pixel image, a graph structure feature corresponding to the target pixel image is generated; wherein, the graph structure feature includes node data of each node and edge features of each edge;
[0009] The node data of each node is input into the convolutional neural network to obtain the first node feature of each node;
[0010] The target pixel image, the first node feature of each node in the target pixel image, and the edge feature of each edge in the target pixel image are input into the structure update network layer of a pre-trained graph neural network model to obtain the updated graph structure features; wherein, the updated graph structure features include the second node feature of each node;
[0011] For each node in the target pixel image, the second node feature of the node is input into the fully connected network layer in the graph neural network model to determine the classification result to which the node belongs, so as to determine the classification result of the corresponding pixel in the vascular tree centerline pixel image based on the classification result to which the node belongs.
[0012] Furthermore, generating the target pixel map based on the vascular tree centerline pixel map includes:
[0013] Create a blank image with the same dimensions as the pixel map of the vascular tree centerline;
[0014] For each pixel on the blood vessel centerline in the blood vessel tree centerline pixel image, determine the coordinates of that pixel in the blood vessel tree centerline pixel image;
[0015] Based on the coordinates of each pixel in the centerline pixel map of the blood vessel tree, the node corresponding to each pixel is determined in the blank image, and multiple nodes are numbered according to a preset order to generate the target pixel map.
[0016] Furthermore, the step of generating the graph structure features corresponding to the target pixel image based on each node of the original vascular tree medical image and the target pixel image includes:
[0017] For each preset node connection relationship, determine the corresponding convolution kernel for that preset node connection relationship when performing global convolution calculation;
[0018] The target pixel map is subjected to global convolution calculation using the convolution kernel, and at least one starting node with the preset node connection relationship and a termination node connected to each starting node with the preset node connection relationship are determined in the target pixel map.
[0019] Based on the number of each starting node in the target pixel image and the number of the ending node connected to each starting node in the target pixel image, the edge connection relationship corresponding to the preset node connection relationship is generated.
[0020] For each edge in the aforementioned edge connection relationship, the edge is encoded to obtain its edge features;
[0021] For each node in the target pixel image, a pre-sized image block is taken from the original vascular tree medical image as the node data, centered on the coordinates of the node in the target pixel image.
[0022] Furthermore, the graph neural network model is trained iteratively through the following steps:
[0023] Obtain the centerline sample pixel image corresponding to the vascular tree medical sample image; wherein, the centerline sample pixel image records the classification label of each pixel point on the sample centerline;
[0024] A sample pixel map is generated based on the centerline sample pixel map, and a sample map structure feature corresponding to the sample pixel map is generated based on the vascular tree medical sample map and each sample node in the sample pixel map; wherein, the sample pixel map includes multiple sample nodes and the number of each sample node, each sample node in the sample pixel map corresponds one-to-one with each pixel on the sample centerline in the centerline sample pixel map, and the sample map structure feature includes the node data of each sample node and the edge feature of each sample edge;
[0025] The node data of each sample node is input into the convolutional neural network to obtain the sample features of each sample node;
[0026] The sample pixel image, the sample features of each sample node in the sample pixel image, the edge features of each sample edge, and the classification label of each sample node are input into the graph neural network primitive model, and the graph neural network primitive model is iteratively trained to obtain the graph neural network model.
[0027] Furthermore, the step of inputting the sample graph structural features and the classification label of each sample node into the graph neural network model, and iteratively training the graph neural network model to obtain the graph neural network model, includes:
[0028] The sample pixel image, the sample features of each sample node in the sample pixel image, and the edge features of each sample edge are input into the structure update original network layer in the graph neural network model. Feature fusion is performed on the sample features of each sample node and the edge features of each sample edge to determine the updated sample graph structure.
[0029] The updated sample graph structure is input into the fully connected network layer in the original graph neural network model, and classification prediction is performed on each sample node in the updated sample graph structure to obtain the predicted classification label of each sample node.
[0030] The classification label of each sample node is compared with the predicted classification label, and the loss value is calculated.
[0031] If the loss value is greater than the preset loss threshold, the training parameters of the original structure update network layer are adjusted, and the original structure update network layer with the adjusted parameters is determined as the original structure update network layer. Then, the step of inputting the sample graph structure into the original structure update network layer in the graph neural network model is returned to be executed until the loss value is less than the loss threshold, and the original structure update network layer is obtained.
[0032] The original structure update network layer in the original graph neural network model is replaced with the structure update network layer to obtain the graph neural network model.
[0033] Furthermore, after determining the classification result to which the node belongs, the classification method further includes:
[0034] Based on the coordinates of the node in the target pixel image, the pixel corresponding to the node is determined in the center line pixel image of the blood vessel tree;
[0035] In the pixel map of the vascular tree center line, the pixel corresponding to the node is displayed in a preset color corresponding to the classification result to which the node belongs.
[0036] Secondly, embodiments of this application also provide a classification device for the central line of blood vessels, the classification device comprising:
[0037] The pixel image generation module is used to obtain the vascular tree centerline pixel image corresponding to the original vascular tree medical image, and generate a target pixel image based on the vascular tree centerline pixel image; wherein, the target pixel image includes multiple nodes and the number of each node, and each node in the target pixel image corresponds one-to-one with each pixel on the vascular centerline in the vascular tree centerline pixel image;
[0038] The graph structure feature generation module is used to generate graph structure features corresponding to the target pixel image based on each node of the original vascular tree medical image and the target pixel image; wherein, the graph structure features include node data of each node and edge features of each edge;
[0039] The node feature determination module is used to input the node data of each node into the convolutional neural network to obtain the first node feature of each node;
[0040] The feature fusion module is used to input the target pixel image, the first node feature of each node in the target pixel image, and the edge feature of each edge in the target pixel image into the structure update network layer of the pre-trained graph neural network model to obtain the updated graph structure features; wherein, the updated graph structure features include the second node feature of each node;
[0041] The classification module is used to input the second node feature of each node in the target pixel image into the fully connected network layer of the graph neural network model to determine the classification result of the node, so as to determine the classification result of the corresponding pixel in the vascular tree centerline pixel image based on the classification result of the node.
[0042] Furthermore, when the pixel image generation module generates a target pixel image based on the vascular tree centerline pixel image, the pixel image generation module is also used for:
[0043] Create a blank image with the same dimensions as the pixel map of the vascular tree centerline;
[0044] For each pixel on the blood vessel centerline in the blood vessel tree centerline pixel image, determine the coordinates of that pixel in the blood vessel tree centerline pixel image;
[0045] Based on the coordinates of each pixel in the centerline pixel map of the blood vessel tree, the node corresponding to each pixel is determined in the blank image, and multiple nodes are numbered according to a preset order to generate the target pixel map.
[0046] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the vascular centerline classification method described above.
[0047] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the blood vessel centerline classification method described above.
[0048] This application provides a method, apparatus, device, and medium for classifying vascular centerlines. It utilizes a vascular tree with topological relationships composed of vascular centerlines to perform a fast and accurate conversion from pixel images to a graph. It fully combines the advantages of convolutional neural networks and graph neural networks to extract and fuse node and edge features, and finally performs multi-task classification of the centerlines. Employing an efficient method for converting between pixel images and graph structures, it fully utilizes the local advantages of traditional convolutional neural networks and the connectivity advantages of graph neural networks. While ensuring the extraction of local features, it performs connectivity fusion. The classification method provided in this application has a fast processing speed, high information utilization rate, and can accurately and reliably complete the vascular tree classification task.
[0049] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description
[0050] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0051] Figure 1 A flowchart illustrating a method for classifying the centerline of blood vessels provided in an embodiment of this application;
[0052] Figure 2(A) is a node field illustration provided in an embodiment of this application;
[0053] Figure 2(B) is a schematic diagram of a convolution kernel used for determining the lower left corner connection according to an embodiment of this application;
[0054] Figure 2(C) is a schematic diagram of graph feature fusion provided in an embodiment of this application;
[0055] Figure 3 This is a schematic diagram of the structure of a blood vessel centerline classification device provided in an embodiment of this application;
[0056] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0057] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.
[0058] First, the applicable scenarios for this application will be introduced. This application can be applied to the field of medical image processing technology.
[0059] As people's living standards improve, the incidence of vascular diseases is also on the rise. Therefore, early screening for vascular diseases is crucial to preventing their progression. Midline classification involves categorizing each point on the midline, such as which blood vessel it belongs to, the type of lesion, and the degree of lesion severity.
[0060] Medical image analysis is an important tool for screening vascular diseases, but analyzing large amounts of image data requires significant medical resources. This is where automated analysis methods play a crucial role. The complex structure of the vascular tree in images makes automated analysis of this tree challenging.
[0061] Research has found that existing tasks for classifying blood vessel centerlines, including lesion classification at each point and prediction of blood vessel diameter, mostly use traditional convolutional neural networks for feature extraction, followed by feature analysis using fully connected layers. The advantage of this most widely used method is that it can extract local features well, but it ignores the connectivity of data with strong topological structures, such as vascular trees.
[0062] Based on this, embodiments of this application provide a method for classifying vascular centerlines to improve the efficiency and accuracy of vascular centerline classification.
[0063] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for classifying the centerline of blood vessels, as provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the classification method includes:
[0064] S101, Obtain the vascular tree centerline pixel map corresponding to the original vascular tree medical image, and generate the target pixel map based on the vascular tree centerline pixel map.
[0065] It should be noted that the original vascular tree medical image refers to a medical image carrying the structure of the vascular tree. The vascular tree centerline pixel image is a binary image recording the vascular centerline from the original vascular tree medical image. The vascular centerline refers to the line connecting the center points of a vascular region within a 2D / 3D vascular image (angiography, CT, MR, etc.), which can be represented as a series of center point pixels. In the vascular tree centerline pixel image, pixels on the centerline have a grayscale value of 1, while other pixels have a grayscale value of 0. The target pixel image refers to the pixel image generated based on the individual pixels in the vascular tree centerline pixel image. The target pixel image includes multiple nodes and a node number; each node in the target pixel image corresponds one-to-one with each pixel on the vascular centerline in the vascular tree centerline pixel image.
[0066] Regarding step S101 above, in specific implementation, firstly, the pixel map of the vascular tree centerline corresponding to the original vascular tree medical image is obtained, and then the target pixel map is generated based on the obtained vascular tree centerline pixel map.
[0067] Specifically, regarding step S101 above, generating the target pixel map based on the vascular tree centerline pixel map includes:
[0068] Step 1011: Create a blank image with the same image size as the central line pixel map of the vascular tree.
[0069] Regarding step 1011 above, in specific implementation, first determine the image size corresponding to the pixel map of the vascular tree centerline, and then create a blank image with the same size as the pixel map of the vascular tree centerline.
[0070] Step 1012: For each pixel on the blood vessel centerline in the blood vessel tree centerline pixel image, determine the coordinates of that pixel in the blood vessel tree centerline pixel image.
[0071] Regarding step 1012 above, in specific implementation, since the vascular tree centerline pixel image is a binary image recording the vascular centerline, in the vascular tree centerline pixel image, the gray value of the pixels on the centerline is 1, and the gray value of other pixels is 0. First, the pixels with non-zero gray values are determined through the vascular tree centerline pixel image, that is, the pixels on the vascular centerline. Then, for each pixel on the vascular centerline in the vascular tree centerline pixel image, the coordinates of that pixel in the vascular tree centerline pixel image are determined.
[0072] Step 1013: Based on the coordinates of each pixel in the center line pixel map of the blood vessel tree, determine the node corresponding to each pixel in the blank image, and number the multiple nodes according to a preset order to generate the target pixel map.
[0073] It should be noted that the preset order refers to the order of the numbers set in advance. For example, the preset order could be to start numbering the nodes from the top left corner, and this application does not impose any specific limitations on this.
[0074] Regarding step 1013 above, in specific implementation, based on the coordinates of each pixel in the vascular tree centerline pixel map, the node corresponding to each pixel is determined in the blank image, and multiple nodes are numbered according to a preset order to generate the target pixel map. The obtained graph node information is as follows: nodes=[1,2,3,...,N], nodes_coordinate=[[x1,x2,...,xN], [y1,y2,...,yN]]. Here, nodes represents the node number, and nodes_coordinate represents the coordinates of the node in the target pixel map.
[0075] S102, Generate graph structure features corresponding to the target pixel map based on the original vascular tree medical image map and each node in the target pixel map.
[0076] It's important to note that in mathematics, a graph is a structure describing a set of objects, where certain pairs of objects are "related" in some sense. These objects correspond to mathematical abstractions called vertices (also known as nodes or points), and each related pair of vertices is called an edge (also known as a link or line). In other words, a graph consists of both nodes and edges. The graph structure features include the node data for each node in the target pixel graph and the edge features for each edge in the target pixel graph.
[0077] Regarding step S102 above, in specific implementation, after the target pixel image is determined in step S101, the graph structure features corresponding to the target pixel image are generated based on the original vascular tree medical image and each node in the generated target pixel image.
[0078] Specifically, regarding step S102 above, generating the graph structure features corresponding to the target pixel image based on each node of the original vascular tree medical image and the target pixel image includes:
[0079] Step 1021: For each preset node connection relationship, determine the convolution kernel corresponding to that preset node connection relationship when performing global convolution calculation.
[0080] It should be noted that the preset node connection relationship refers to the connection relationship between two adjacent nodes. Please refer to Figure 2(A), which is a node neighborhood illustration provided by an embodiment of this application. As shown in Figure 2(A), each node has 8 neighborhoods, namely the top left corner, top, top right corner, left side, right side, bottom left corner, bottom, and bottom right corner. Therefore, there are 8 connection methods for each point, and there are also 8 preset node connection relationships, namely top left corner connection, top connection, top right corner connection, left side connection, right side connection, bottom left corner connection, bottom connection, and bottom right corner connection. As an example, when the preset node connection relationship is bottom left corner connection, it means that there is another node in the bottom left corner neighborhood of a certain node. Please refer to Figure 2(B), which is a schematic diagram of a convolution kernel used for bottom left corner connection determination provided by an embodiment of this application. As shown in Figure 2(B), since the adjacent node is located in the bottom left corner when connecting at the bottom left corner, the parameter of the bottom left corner in the convolution kernel is set to 1. Similarly, when the preset node connection relationship is a bottom right corner connection, the bottom right corner parameter of the convolution kernel used is 1, and the other parameters are 0.
[0081] Regarding step 1021 above, in specific implementation, for each preset node connection relationship, the convolution kernel corresponding to that preset node connection relationship is determined when performing global convolution calculation.
[0082] Step 1022: Perform global convolution calculation on the target pixel map using the convolution kernel, and determine at least one starting node in the target pixel map that has the preset node connection relationship, and a termination node connected to each starting node by the preset node connection relationship.
[0083] Since each point has eight possible connection methods, calculating the connection relationships point by point would be computationally intensive and slow. Therefore, the embodiments provided in this application employ global convolution. Regarding step 1022 above, in specific implementation, after determining the convolution kernel corresponding to the preset node connection relationship, the kernel is used to perform global convolution calculation on the target pixel image. At least one starting node with the preset node connection relationship and a termination node connected to each starting node by the preset node connection relationship are identified in the target pixel image. Thus, only eight global convolutions are needed for each preset node connection relationship.
[0084] Step 1023: Generate the edge connection relationship corresponding to the preset node connection relationship based on the number of each starting node in the target pixel image and the number of the ending node connected to each starting node in the target pixel image.
[0085] Regarding step 1023 above, in specific implementation, since each node in the target pixel image has its own number, the number of each starting node in the target pixel image and the number of the ending node connected to each starting node through the preset node connection relationship in the target pixel image are determined. The starting node and the ending node form an edge. The edge connection relationship can be expressed as edges=[[u1,u2,...un], [v1,v2,...vn]], with a data shape of 2*E, where E represents the number of edges, u represents the number of the starting node of the edge in the target pixel image, and v represents the number of the ending node of the edge in the target pixel image.
[0086] Step 1024: For each edge in the edge connection relationship, encode the edge to obtain the edge feature.
[0087] Regarding step 1024 above, in specific implementation, since the connection direction between the starting node and the ending node is known in step 1023, after the edge connection relationship of each edge is determined, each edge in the edge connection relationship is encoded to obtain the edge feature. For example, when two nodes in the edge connection relationship are connected at the bottom left corner, the edge corresponding to these two nodes is encoded as 1; when two nodes in the edge connection relationship are connected at the bottom right corner, the edge corresponding to these two nodes is encoded as 7. This application does not make specific limitations on this.
[0088] Step 1025: For each node in the target pixel image, take the coordinates of the node in the target pixel image as the center, and take an image block of a preset size in the original vascular tree medical image as the node data of the node.
[0089] Regarding step 1025 above, in specific implementation, for each node in the target pixel image, firstly, the coordinates of that node in the target pixel image are determined. Since the target pixel image and the vascular tree centerline pixel image are the same size, and the vascular tree centerline pixel image is a binary image corresponding to the original vascular tree medical image, the target pixel image and the original vascular tree medical image are also the same size. Using the coordinates of that node in the target pixel image as the center, a pre-sized image block is taken from the original vascular tree medical image as the node data for that node. For example, a 32*32 image block is taken as the node data.
[0090] S103, input the node data of each node into the convolutional neural network to obtain the first node feature of each node.
[0091] Here, node features can be understood as local features, and convolutional neural networks have a natural advantage in extracting local features. Therefore, before the data enters the neural network model, it first goes through the convolutional neural network to extract features from each node.
[0092] In specific implementation of step S103, for each node in the target pixel image, the node data of that node is input into a convolutional neural network to obtain the first node feature of each node.
[0093] S104, the target pixel image, the first node feature of each node in the target pixel image, and the edge feature of each edge in the target pixel image are input into the structure update network layer of the pre-trained graph neural network model to obtain the updated graph structure features.
[0094] It should be noted that Graph Neural Network (GNN) refers to a general term for algorithms that use neural networks to learn graph-structured data, extract and discover features and patterns in graph-structured data, and meet the needs of graph learning tasks such as clustering, classification, prediction, segmentation, and generation.
[0095] Regarding step S104 above, in specific implementation, the target pixel image, the first node feature of each node in the target pixel image, and the edge feature of each edge in the target pixel image are input into the structure update network layer of the pre-trained graph neural network model to obtain the updated graph structure features. The updated graph structure features include the second node feature of each node in the target oak tree graph.
[0096] Specifically, according to the classification method provided in the embodiments of this application, the graph neural network model is iteratively trained through the following steps:
[0097] A: Obtain the centerline sample pixel image corresponding to the vascular tree medical sample image.
[0098] It should be noted that the vascular tree medical sample image is a medical sample image carrying an image of the vascular tree structure. The centerline sample pixel image is a binary image recording the centerline of the sample in the vascular tree medical sample image. The centerline sample pixel image also records the classification label of each pixel on the sample centerline. For example, the classification label can be which blood vessel the pixel belongs to, the lesion type corresponding to the pixel, etc.
[0099] Regarding step A above, in specific implementation, the centerline sample pixel image corresponding to the vascular tree medical sample image is obtained.
[0100] B: Generate a sample pixel map based on the centerline sample pixel map, and generate the sample map structure features corresponding to the sample pixel map based on the vascular tree medical sample map and each sample node in the sample pixel map.
[0101] The sample pixel map includes multiple sample nodes and the number of each sample node. Each sample node in the sample pixel map corresponds one-to-one with each pixel on the center line of the sample pixel map. The structural features of the sample map include the node data of each sample node and the edge features of each sample edge.
[0102] Regarding step B above, in specific implementation, a sample pixel image is generated based on the centerline sample pixel image, and the sample image structure features corresponding to the sample pixel image are generated based on the vascular tree medical sample image and each sample node in the sample pixel image. Here, the method of generating the sample pixel image based on the centerline sample pixel image is the same as the method of generating the target pixel image based on the vascular tree centerline pixel image in step S101 above, and can achieve the same technical effect, so it will not be described again here. The method of generating the sample image structure features corresponding to the sample pixel image based on the vascular tree medical sample image and each sample node in the sample pixel image is the same as the method of generating the target pixel image structure features based on the original vascular tree medical image and each node in the target pixel image in step S102 above, and can achieve the same technical effect, so it will not be described again here.
[0103] C: Input the node data of each sample node into the convolutional neural network to obtain the sample features of each sample node.
[0104] Regarding step C above, in specific implementation, after obtaining the node data of each sample node, the node data of each sample node is input into the convolutional neural network to obtain the sample features of each sample node.
[0105] D: Input the sample pixel image, the sample features of each sample node in the sample pixel image, the edge features of each sample edge, and the classification label of each sample node into the graph neural network model, and iteratively train the graph neural network model to obtain the graph neural network model.
[0106] It should be noted that the graph neural network primitive model refers to a pre-built primitive model used to predict the classification results of blood vessels at various points in the blood vessel centerline.
[0107] Regarding step D above, in specific implementation, the sample pixel map, the sample features of each sample node in the sample pixel map, the edge features of each sample edge, and the classification label of each sample node are input into the graph neural network model, and the graph neural network model is iteratively trained to obtain the graph neural network model.
[0108] Specifically, regarding step D above, the step of inputting the sample graph structural features and the classification label of each sample node into the graph neural network model, and iteratively training the graph neural network model to obtain the graph neural network model, includes:
[0109] a: Input the sample pixel map, the sample features of each sample node in the sample pixel map, and the edge features of each sample edge into the structure update original network layer in the graph neural network model, perform feature fusion on the sample features of each sample node and the edge features of each sample edge, and determine the updated sample graph structure.
[0110] Here, graph feature fusion is accomplished through a graph neural network. During the iterative training of the graph neural network, the feature information of the graph (nodes and edges) continuously changes. The information of an edge is related to the vertices it connects to, and the information of a vertex is related to the edge it connects to and the node at that end, resulting in a cyclical iterative update. Please refer to Figure 2(C), which is a schematic diagram of graph feature fusion provided in an embodiment of this application. As shown in Figure 2(C), assuming that after passing through the convolutional neural network, the features of each node are extracted as n1, n2, n3, and n4, if we aggregate the features of the node containing n1, we first need to update the edge features. Here, the initial input of the edge features is the edge feature of the edge starting from node n1. The edge feature update is shown in the following formula:
[0111] ;
[0112] in, t+1 and t They represent the first t+1 and the t The next iteration. edata Representing edge features, ndata Represents node characteristics. fe This represents the edge feature update function, which is represented here using a graph neural network (graph convolution, etc.). i and j This indicates the index of the sample node in the sample pixel map. For example, edata 13 This represents the edge characteristics of an edge that starts at node n1 and ends at node n3.
[0113] After the edge features are updated, the node update is related to the edges connected to it. The node feature update is shown in the following formula:
[0114] ;
[0115] in, N Indicates and i The total number of edges connected to a node.fn The node update function can be represented using a graph neural network (graph convolution, etc.).
[0116] b: Input the updated sample graph structure into the fully connected network layer in the original graph neural network model, perform classification prediction on each sample node in the updated sample graph structure, and obtain the predicted classification label of each sample node.
[0117] It should be noted that the fully connected network layer is used to predict the classification result corresponding to each node based on the characteristics of each node.
[0118] Regarding step b above, in specific implementation, the updated sample graph structure is input into the fully connected network layer in the original graph neural network model, and classification prediction is performed on each sample node in the updated sample graph structure to obtain the predicted classification label of each sample node.
[0119] c: Compare the classification label of each sample node with the predicted classification label and calculate the loss value.
[0120] It should be noted that the loss function is a function that maps the values of a random event or its related random variables to non-negative real numbers to represent the "risk" or "loss" of that random event. In applications, the loss function is often used as a learning criterion in relation to optimization problems; that is, the model is solved and evaluated by minimizing the loss function.
[0121] Regarding step c above, in practice, the classification label of each sample node is compared with the predicted classification label. If the predicted classification label of a sample node is different from the predicted classification label, a loss value needs to be calculated. The method for calculating the loss value is explained in detail in existing technologies and will not be elaborated further here.
[0122] d: If the loss value is greater than the preset loss threshold, the training parameters of the original structure update network layer are adjusted, and the original structure update network layer with the adjusted parameters is determined as the original structure update network layer. Then, the step of inputting the sample graph structure into the original structure update network layer in the graph neural network model is returned to be executed until the loss value is less than the loss threshold, and the original structure update network layer is obtained.
[0123] e: Replace the original structure update network layer in the original graph neural network model with the structure update network layer to obtain the graph neural network model.
[0124] In the embodiments provided in this application, the loss threshold refers to a pre-set standard. As an optional implementation, the minimum threshold can be set to the second derivative of the loss value being close to 0. This is because when the second derivative is close to 0, the slope of the loss value is the smallest, meaning that the change in the loss value between two iterations of the graph neural network model is already very small. When the loss value is close to this loss threshold, the graph neural network model is considered to have reached a convergence state, and the prediction of the graph neural network model at this time is relatively accurate.
[0125] Regarding step d above, in specific implementation, if the calculated loss value is greater than the preset loss threshold, the training parameters of the original structure update network layer are continuously adjusted. The original structure update network layer with adjusted parameters is then determined as the original structure update network layer. The process returns to step a above, where the sample graph structure is input into the original structure update network layer of the graph neural network model, until the loss value is less than the loss threshold. At this point, iterative training stops, and the structure update network layer is obtained. Specifically, the loss is continuously minimized through iteration. At each iteration step, the loss value is calculated. When the loss value cannot reach the loss threshold, the training parameters are continuously updated. The new parameters calculate a new loss value, resulting in a fluctuating downward trend in the loss value during iteration. Finally, when the loss value reaches a smooth state, i.e., when the loss value is no greater than the loss threshold (i.e., the decrease in loss value compared to the previously calculated loss value is not significant), the model is considered to have converged, and training ends. Then, the original structure update network layer in the graph neural network model is replaced with the structure update network layer to obtain the graph neural network model.
[0126] S105, for each node in the target pixel image, the second node feature of the node is input into the fully connected network layer in the graph neural network model to determine the classification result to which the node belongs, so as to determine the classification result of the corresponding pixel in the vascular tree centerline pixel image based on the classification result to which the node belongs.
[0127] The classification results here are the classification results for each point on the center line of the blood vessel, such as which blood vessel the point belongs to, the type of lesion at the point, and the degree of lesion.
[0128] Regarding step S105 above, in specific implementation, for each node in the target pixel image, the second node feature of that node is input into a fully connected network layer in the graph neural network model to determine the classification result to which the node belongs. Specifically, during classification, for example, if there are two categories, and the feature dimension of each node is 512, a 512x2 fully connected network layer is used to calculate the features to obtain an output vector of length 2. If the first digit of the vector is larger than the second digit, for example [0.8, 0.2], then the node is considered to belong to the first category; otherwise, it belongs to the second category. The principle for multiple classification results is as described above. For N-classification, a 512xN fully connected layer is used, with an output length of N. The largest digit in the vector indicates which category the point belongs to. Since each node in the target pixel image corresponds one-to-one with each pixel on the blood vessel centerline pixel image in the blood vessel tree centerline pixel image, the classification result of the node in the blood vessel tree centerline pixel image can be determined based on the classification result to which the node belongs.
[0129] According to the classification method provided in the embodiments of this application, as an optional implementation, after determining the classification result to which the node belongs, the classification method further includes:
[0130] (1) Based on the coordinates of the node in the target pixel map, determine the pixel point corresponding to the node in the center line pixel map of the blood vessel tree.
[0131] (2) In the pixel map of the center line of the blood vessel tree, the pixel corresponding to the node is displayed in the preset color corresponding to the classification result to which the node belongs.
[0132] It should be noted that the preset color refers to the display color corresponding to the category to which the node belongs. For example, when the node belongs to the first category, the preset color can be red, and when the node belongs to the second category, the preset color can be green. This application does not make specific limitations on this.
[0133] Regarding steps (1) and (2) above, in specific implementation, based on the coordinates of the node in the target pixel image, the pixel corresponding to the node is determined in the vascular tree centerline pixel image. Then, the pixel corresponding to the node is displayed in the vascular tree centerline pixel image with a preset color corresponding to the classification result of the node. In this way, the classification result of each node is converted to the vascular tree centerline pixel image, and the classification results of different nodes are returned to the vascular tree centerline pixel image according to their coordinates in the target pixel image. Furthermore, the pixels of different nodes are displayed with different colors, making it convenient for users to intuitively judge the classification result of the vascular tree centerline.
[0134] This application provides a method for classifying vascular centerlines. First, a vascular tree centerline pixel map corresponding to an original vascular tree medical image is obtained, and a target pixel map is generated based on the vascular tree centerline pixel map. Then, a graph structure feature corresponding to the target pixel map is generated based on each node in the original vascular tree medical image and the target pixel map. The node data of each node is input into a convolutional neural network to obtain the first node feature of each node. The target pixel map, the first node feature of each node in the target pixel map, and the edge feature of each edge in the target pixel map are input into the structure update network layer of a pre-trained graph neural network model to obtain the updated graph structure feature. Finally, for each node in the target pixel map, the second node feature of the node is input into the fully connected network layer of the graph neural network model to determine the classification result to which the node belongs, so as to determine the classification result of the pixel point corresponding to the node in the vascular tree centerline pixel map based on the classification result to which the node belongs.
[0135] This application utilizes a vascular tree with a topological structure composed of vascular centerlines to perform a fast and accurate conversion from pixel images to graphs. It fully combines the advantages of convolutional neural networks and graph neural networks to extract and fuse node and edge features, and finally performs multi-task classification of the centerlines. Employing an efficient method for converting between pixel images and graph structures, it fully leverages the local advantages of traditional convolutional neural networks and the connectivity advantages of graph neural networks. While ensuring the extraction of local features, it performs connectivity fusion. The classification method provided in this application has a fast processing speed, high information utilization rate, and can accurately and reliably complete the vascular tree classification task.
[0136] Please see Figure 3 , Figure 3 This is a schematic diagram of a blood vessel centerline classification device provided in an embodiment of this application. Figure 3 As shown, the sorting device 300 includes:
[0137] The pixel image generation module 301 is used to obtain the vascular tree centerline pixel image corresponding to the original vascular tree medical image, and generate a target pixel image based on the vascular tree centerline pixel image; wherein, the target pixel image includes multiple nodes and the number of each node, and each node in the target pixel image corresponds one-to-one with each pixel on the vascular centerline of the vascular tree centerline pixel image;
[0138] The graph structure feature generation module 302 is used to generate graph structure features corresponding to the target pixel image based on each node of the original vascular tree medical image and the target pixel image; wherein, the graph structure features include node data of each node and edge features of each edge;
[0139] The node feature determination module 303 is used to input the node data of each node into the convolutional neural network to obtain the first node feature of each node;
[0140] The feature fusion module 304 is used to input the target pixel image, the first node feature of each node in the target pixel image, and the edge feature of each edge in the target pixel image into the structure update network layer of the pre-trained graph neural network model to obtain the updated graph structure features; wherein, the updated graph structure features include the second node feature of each node;
[0141] The classification module 305 is used to input the second node feature of each node in the target pixel image into the fully connected network layer in the graph neural network model to determine the classification result of the node, so as to determine the classification result of the corresponding pixel in the vascular tree centerline pixel image based on the classification result of the node.
[0142] Furthermore, when generating a target pixel image based on the vascular tree centerline pixel image, the pixel image generation module 301 is also used to:
[0143] Create a blank image with the same dimensions as the pixel map of the vascular tree centerline;
[0144] For each pixel on the blood vessel centerline in the blood vessel tree centerline pixel image, determine the coordinates of that pixel in the blood vessel tree centerline pixel image;
[0145] Based on the coordinates of each pixel in the centerline pixel map of the blood vessel tree, the node corresponding to each pixel is determined in the blank image, and multiple nodes are numbered according to a preset order to generate the target pixel map.
[0146] Furthermore, when generating graph structure feature generation module 302 based on each node of the original vascular tree medical image and the target pixel image to generate graph structure features corresponding to the target pixel image, graph structure feature generation module 302 is also used to:
[0147] For each preset node connection relationship, determine the corresponding convolution kernel for that preset node connection relationship when performing global convolution calculation;
[0148] The target pixel map is subjected to global convolution calculation using the convolution kernel, and at least one starting node with the preset node connection relationship and a termination node connected to each starting node with the preset node connection relationship are determined in the target pixel map.
[0149] Based on the number of each starting node in the target pixel image and the number of the ending node connected to each starting node in the target pixel image, the edge connection relationship corresponding to the preset node connection relationship is generated.
[0150] For each edge in the aforementioned edge connection relationship, the edge is encoded to obtain its edge features;
[0151] For each node in the target pixel image, a pre-sized image block is taken from the original vascular tree medical image as the node data, centered on the coordinates of the node in the target pixel image.
[0152] Furthermore, the classification device 300 also includes a model training module, which iteratively trains the graph neural network model through the following steps:
[0153] Obtain the centerline sample pixel image corresponding to the vascular tree medical sample image; wherein, the centerline sample pixel image records the classification label of each pixel point on the sample centerline;
[0154] A sample pixel map is generated based on the centerline sample pixel map, and a sample map structure feature corresponding to the sample pixel map is generated based on the vascular tree medical sample map and each sample node in the sample pixel map; wherein, the sample pixel map includes multiple sample nodes and the number of each sample node, each sample node in the sample pixel map corresponds one-to-one with each pixel on the sample centerline in the centerline sample pixel map, and the sample map structure feature includes the node data of each sample node and the edge feature of each sample edge;
[0155] The node data of each sample node is input into the convolutional neural network to obtain the sample features of each sample node;
[0156] The sample pixel image, the sample features of each sample node in the sample pixel image, the edge features of each sample edge, and the classification label of each sample node are input into the graph neural network primitive model, and the graph neural network primitive model is iteratively trained to obtain the graph neural network model.
[0157] Furthermore, when the model training module inputs the sample graph structural features and the classification label of each sample node into the original graph neural network model and iteratively trains the original graph neural network model to obtain the graph neural network model, the model training module is also used for:
[0158] The sample pixel image, the sample features of each sample node in the sample pixel image, and the edge features of each sample edge are input into the structure update original network layer in the graph neural network model. Feature fusion is performed on the sample features of each sample node and the edge features of each sample edge to determine the updated sample graph structure.
[0159] The updated sample graph structure is input into the fully connected network layer in the original graph neural network model, and classification prediction is performed on each sample node in the updated sample graph structure to obtain the predicted classification label of each sample node.
[0160] The classification label of each sample node is compared with the predicted classification label, and the loss value is calculated.
[0161] If the loss value is greater than the preset loss threshold, the training parameters of the original structure update network layer are adjusted, and the original structure update network layer with the adjusted parameters is determined as the original structure update network layer. Then, the step of inputting the sample graph structure into the original structure update network layer in the graph neural network model is returned to be executed until the loss value is less than the loss threshold, and the original structure update network layer is obtained.
[0162] The original structure update network layer in the original graph neural network model is replaced with the structure update network layer to obtain the graph neural network model.
[0163] Furthermore, the classification device 300 also includes a display module, which, after determining the classification result to which the node belongs, is used to:
[0164] Based on the coordinates of the node in the target pixel image, the pixel corresponding to the node is determined in the center line pixel image of the blood vessel tree;
[0165] In the pixel map of the vascular tree center line, the pixel corresponding to the node is displayed in a preset color corresponding to the classification result to which the node belongs.
[0166] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 includes a processor 410, a memory 420, and a bus 430.
[0167] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, they can perform the operations described above. Figure 1The steps of the classification method for the vascular centerline in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.
[0168] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The steps of the classification method for the vascular centerline in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.
[0169] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0170] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.
[0171] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0172] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
[0173] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0174] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0175] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for classifying the centerline of blood vessels, characterized in that, The classification method includes: Obtain the vascular tree centerline pixel map corresponding to the original vascular tree medical image, and generate a target pixel map based on the vascular tree centerline pixel map; wherein, the target pixel map includes multiple nodes and the number of each node, and each node in the target pixel map corresponds one-to-one with each pixel on the vascular centerline of the vascular tree centerline pixel map; Based on the original vascular tree medical image and each node in the target pixel image, a graph structure feature corresponding to the target pixel image is generated; wherein, the graph structure feature includes node data of each node and edge features of each edge; The node data of each node is input into the convolutional neural network to obtain the first node feature of each node; The target pixel image, the first node feature of each node in the target pixel image, and the edge feature of each edge in the target pixel image are input into the structure update network layer of a pre-trained graph neural network model to obtain the updated graph structure features; wherein, the updated graph structure features include the second node feature of each node; For any one of the multiple nodes in the target pixel image, the second node feature of the node is input into the fully connected network layer in the graph neural network model to determine the classification result to which the node belongs, so as to determine the classification result of the corresponding pixel in the vascular tree centerline pixel image based on the classification result to which the node belongs.
2. The classification method according to claim 1, characterized in that, The generation of the target pixel map based on the vascular tree centerline pixel map includes: Create a blank image with the same dimensions as the pixel map of the vascular tree centerline; For each pixel on the blood vessel centerline in the blood vessel tree centerline pixel image, determine the coordinates of that pixel in the blood vessel tree centerline pixel image; Based on the coordinates of each pixel in the centerline pixel map of the blood vessel tree, the node corresponding to each pixel is determined in the blank image, and multiple nodes are numbered according to a preset order to generate the target pixel map.
3. The classification method according to claim 1, characterized in that, The process of generating graph structure features corresponding to the target pixel image based on each node of the original vascular tree medical image and the target pixel image includes: For each preset node connection relationship, determine the corresponding convolution kernel for that preset node connection relationship when performing global convolution calculation; The target pixel map is subjected to global convolution calculation using the convolution kernel, and at least one starting node with the preset node connection relationship and a termination node connected to each starting node with the preset node connection relationship are determined in the target pixel map. Based on the number of each starting node in the target pixel image and the number of the ending node connected to each starting node in the target pixel image, the edge connection relationship corresponding to the preset node connection relationship is generated. For each edge in the aforementioned edge connection relationship, the edge is encoded to obtain its edge features; For any one of the multiple nodes in the target pixel image, taking the coordinates of the node in the target pixel image as the center, a preset-sized image block is taken from the original vascular tree medical image as the node data of that node.
4. The classification method according to claim 1, characterized in that, The graph neural network model is trained iteratively through the following steps: Obtain the centerline sample pixel image corresponding to the vascular tree medical sample image; wherein, the centerline sample pixel image records the classification label of each pixel point on the sample centerline; A sample pixel map is generated based on the centerline sample pixel map, and a sample map structure feature corresponding to the sample pixel map is generated based on the vascular tree medical sample map and each sample node in the sample pixel map; wherein, the sample pixel map includes multiple sample nodes and the number of each sample node, each sample node in the sample pixel map corresponds one-to-one with each pixel on the sample centerline in the centerline sample pixel map, and the sample map structure feature includes the node data of each sample node and the edge feature of each sample edge; The node data of each sample node is input into the convolutional neural network to obtain the sample features of each sample node; The sample pixel image, the sample features of each sample node in the sample pixel image, the edge features of each sample edge, and the classification label of each sample node are input into the graph neural network primitive model, and the graph neural network primitive model is iteratively trained to obtain the graph neural network model.
5. The classification method according to claim 4, characterized in that, The process of inputting the sample pixel image, the sample features of each sample node in the sample pixel image, the edge features of each sample edge, and the classification label of each sample node into the graph neural network model, and iteratively training the graph neural network model to obtain the graph neural network model, includes: The sample pixel image, the sample features of each sample node in the sample pixel image, and the edge features of each sample edge are input into the structure update original network layer in the graph neural network model. Feature fusion is performed on the sample features of each sample node and the edge features of each sample edge to determine the updated sample graph structure. The updated sample graph structure is input into the fully connected network layer in the original graph neural network model, and classification prediction is performed on each sample node in the updated sample graph structure to obtain the predicted classification label of each sample node. The classification label of each sample node is compared with the predicted classification label, and the loss value is calculated. If the loss value is greater than the preset loss threshold, the training parameters of the original structure update network layer are adjusted, and the original structure update network layer with the adjusted parameters is determined as the original structure update network layer. Then, the step of inputting the sample pixel map, the sample features of each sample node in the sample pixel map, and the edge features of each sample edge into the original structure update network layer in the graph neural network model is returned until the loss value is less than the loss threshold, and the original structure update network layer is obtained. The original structure update network layer in the original graph neural network model is replaced with the structure update network layer to obtain the graph neural network model.
6. The classification method according to claim 1, characterized in that, After determining the classification result to which the node belongs, the classification method further includes: Based on the coordinates of the node in the target pixel image, the pixel corresponding to the node is determined in the center line pixel image of the blood vessel tree; In the pixel map of the vascular tree center line, the pixel corresponding to the node is displayed in a preset color corresponding to the classification result to which the node belongs.
7. A classification device for the central line of blood vessels, characterized in that, The sorting device includes: The pixel image generation module is used to obtain the vascular tree centerline pixel image corresponding to the original vascular tree medical image, and generate a target pixel image based on the vascular tree centerline pixel image; wherein, the target pixel image includes multiple nodes and the number of each node, and each node in the target pixel image corresponds one-to-one with each pixel on the vascular centerline in the vascular tree centerline pixel image; The graph structure feature generation module is used to generate graph structure features corresponding to the target pixel image based on each node of the original vascular tree medical image and the target pixel image; wherein, the graph structure features include node data of each node and edge features of each edge; The node feature determination module is used to input the node data of each node into the convolutional neural network to obtain the first node feature of each node; The feature fusion module is used to input the target pixel image, the first node feature of each node in the target pixel image, and the edge feature of each edge in the target pixel image into the structure update network layer of the pre-trained graph neural network model to obtain the updated graph structure features; wherein, the updated graph structure features include the second node feature of each node; The classification module is used to input the second node feature of any one of the multiple nodes in the target pixel image into the fully connected network layer of the graph neural network model to determine the classification result of the node, so as to determine the classification result of the corresponding pixel in the vascular tree centerline pixel image based on the classification result of the node.
8. The sorting device according to claim 7, characterized in that, When generating a target pixel image based on the vascular tree centerline pixel image, the pixel image generation module is further configured to: Create a blank image with the same dimensions as the pixel map of the vascular tree centerline; For each pixel on the blood vessel centerline in the blood vessel tree centerline pixel image, determine the coordinates of that pixel in the blood vessel tree centerline pixel image; Based on the coordinates of each pixel in the centerline pixel map of the blood vessel tree, the node corresponding to each pixel is determined in the blank image, and multiple nodes are numbered according to a preset order to generate the target pixel map.
9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. The machine-readable instructions are executed by the processor to perform the steps of the vascular centerline classification method as described in any one of claims 1 to 6.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method for classifying the vascular centerline as described in any one of claims 1 to 6.