Retinal vascular reconstruction method based on directed vascular tree
The retinal vascular skeleton is extracted through convolutional neural network and convolution algorithm to construct a directed vascular tree, which solves the problem of blurring and vascular interlacing of retinal images, and realizes the accurate segmentation and analysis of retinal blood vessels.
Patent Information
- Application Number
- CN202111370474.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-18
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2041-11-18
AI Technical Summary
The prior art is prone to blur after amplification of the retinal image, and the crisscrossing complex parts of the retinal blood vessels are difficult to segment, resulting in difficulty in detection and analysis.
Convolutional neural network is used to segment blood vessels from retinal images, and the convolution algorithm is used to extract vascular skeleton lines. The end nodes and branch nodes are extracted through geometric relationships to construct a directed blood vessel tree, and the topological structure diagram of retinal blood vessels is reconstructed.
Effective segmentation and distinction of retinal blood vessels reduce image blur, improve detection accuracy and reliability, and simplify the vascular analysis process.
Smart Images

Figure CN113902763B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of retinal blood vessel vectorization, and relates to a retinal blood vessel reconstruction method based on a directed blood vessel tree. Background Art
[0002] The retina is the inner layer of the eyeball wall, which is divided into the blind part and the visual part of the retina. To examine the retinal blood vessels, the image information of the retina is usually obtained first, and then the retinal image is magnified for examination; however, the magnified retinal image will become blurred, and the retinal blood vessels have complex parts that crisscross. If the crisscrossed parts and blood vessel branches are not distinguished, it is easy to cause misjudgment, which is not convenient for detecting and analyzing retinal blood vessels. Summary of the invention
[0003] In view of this, an object of the present invention is to provide a retinal vascular reconstruction method based on a directed vascular tree.
[0004] In order to achieve the above object, the present invention provides the following technical solutions:
[0005] A retinal vascular reconstruction method based on a directed vascular tree comprises the following steps:
[0006] S1, obtaining retinal image information;
[0007] S2, using convolutional neural network to segment retinal blood vessels from retinal image information;
[0008] S3, using convolution algorithm to extract the skeleton line of retinal blood vessels;
[0009] S4, cutting off the optic disc area in the retinal image information, and extracting the end nodes and branch nodes from the skeleton line according to the geometric relationship;
[0010] S5, segmenting the retinal blood vessels into blood vessel segments through branch nodes, and extracting skeleton line information and outer contour information of each blood vessel segment;
[0011] S6, constructing a directed vascular tree with a topological structure through skeleton line information and outer contour information of end nodes, branch nodes and vascular segments;
[0012] S7. Setting a type of a directed vascular tree, wherein the type of the directed vascular tree includes an arterial vascular tree and a venous vascular tree;
[0013] S8. Reconstruct the skeleton lines of the retinal blood vessels and / or the image information of the blood vessels in the form of a topological graph using the directed blood vessel tree.
[0014] Furthermore, in the step S3, when the skeleton line of the retinal blood vessels is extracted using the convolution algorithm, the convolution algorithm adopts the 8-neighboring point method, and the convolution kernel is as follows:
[0015] mask2=[1 1 1; 1 0 1; 1 1 1].
[0016] Furthermore, in the step S4, the method of extracting the end nodes and branch nodes from the skeleton line according to the geometric relationship is:
[0017] The number of adjacent points of all points in the skeleton line of the blood vessel is calculated, and the points with 1 and 3 adjacent points are selected as nodes. The nodes with 1 adjacent point are defined as end nodes, and the nodes with 3 adjacent points are defined as branch nodes.
[0018] Furthermore, in step S5, the method for extracting the skeleton information and the outer contour information of the blood vessel segment is:
[0019] A curvature threshold is preset, and the skeleton line of the blood vessel segment and the pixel points on the outer contour curves on both sides are extracted in turn. Then, three adjacent pixel points on the same curve are connected in turn to form a curve model, and the curvature of the curve at the middle pixel point is calculated as the curvature at the pixel point. The pixel points at both ends of the skeleton line and each outer contour curve and the pixel points with a curvature greater than the curvature threshold are screened out as the key points of the skeleton line and the outer contour curve; then, serial numbers are assigned to the key points of the skeleton line and the outer contour curve from one end point to another along the direction of the skeleton line and the outer contour curve; the coordinates and serial numbers of all the key points on the skeleton line and the outer contour curve of the blood vessel segment are used as the skeleton line information and outer contour information of the blood vessel segment.
[0020] Furthermore, in the step S6, the method of constructing a directed vascular tree of topological structure through the skeleton line information of the end nodes, branch nodes and vascular segments and the key points of the outer contour is:
[0021] S61, finding a branch node connected to two end nodes in the skeleton line; if the angle formed by the skeleton line connecting the two end nodes and the branch node at the branch node is less than a preset angle threshold, executing step S62; otherwise, returning to execute step S61;
[0022] S62, taking the branch node as the root node, taking the two end nodes as leaf nodes, and combining the skeleton line information and the outer contour information of the corresponding two blood vessel segments to form a subtree;
[0023] S63, integrating the information of the subtrees as a supernode, setting the supernode as an end node and replacing the branch node in the subtree;
[0024] S64, repeat steps S61 to S63 until all remaining nodes are end nodes;
[0025] S65 , using the non-super node of the two mutually connected end nodes as the starting root node and the super node as the leaf node to form a directed vascular tree with a topological structure, and storing information of the directed vascular tree.
[0026] Furthermore, the information of the subtree includes information of the branch node and information of two end nodes, as well as skeleton line information and outer contour information of two vascular segments segmented by the branch node and the two end nodes; the information of the branch node includes the coordinates of the branch point, when the end node is a super node, the information of the end node is the information of the subtree stored by the super node, and when the end node is a non-super node, the information of the end node is the coordinates of the end node; the information of the vascular tree includes information of two end nodes, as well as skeleton line information and outer contour information of the corresponding vascular segments; when the end node is a super node, the information of the end node is the information of the subtree stored by the super node, and when the end node is a starting root node, the information of the end node is the coordinates of the end node.
[0027] Furthermore, in the step S8, the method of reconstructing the image of the retinal blood vessels in the form of a topological map using the directed blood vessel tree is as follows: based on the stored information of each directed blood vessel tree, the skeleton line information of each directed blood vessel tree and / or the image information of the directed blood vessel tree are restored; for areas where the image information of the directed blood vessel trees overlap with each other, the image information of the venous blood vessel tree in the area is displayed, and the image information of the arterial blood vessel tree in the area is hidden.
[0028] Furthermore, a method for restoring skeleton line information of each directional vascular tree and / or image information of the directional vascular tree according to the stored information of each directional vascular tree is:
[0029] Extracting information of two end nodes and key point information of the skeleton line and / or outer contour of the corresponding blood vessel segment from the stored information of the blood vessel tree; using the coordinates of the starting root node as one end of the blood vessel segment and the coordinates of the leaf node as the other end of the blood vessel segment, and reconstructing the skeleton line of the blood vessel segment and / or the image information of the blood vessel segment according to the key point information of the skeleton line and / or outer contour;
[0030] When a leaf node is a super node, the leaf node is restored to a subtree; the method is: extracting the coordinates of the branch node, the coordinates of the two end segments, and the key point information of the skeleton lines and / or outer contours of the corresponding two blood vessel segments from the super node; using the coordinates of the branch node as one end of the two blood vessel segments, and using the coordinates of the two leaf nodes as the other ends of the two blood vessel segments, respectively reconstructing the skeleton lines of the two blood vessel segments and / or the image information of the two blood vessel segments according to the key point information of the skeleton lines and / or outer contours of the two blood vessel segments;
[0031] When the leaf nodes of a subtree include a supernode, the leaf nodes are continuously restored to the subtree until all the supernodes are restored to the subtree.
[0032] Furthermore, the method for reconstructing the image information of the blood vessel segment according to the key point information of the outer contour is:
[0033] The coordinates and serial numbers of each key point of the outer contour curve are extracted from the key point information of the outer contour, and the spline curves connecting the key points with adjacent serial numbers on the outer contour curve are drawn in sequence using the quadratic interpolation spline curve construction method to generate the outer contour of the blood vessel segment and reconstruct the blood vessel segment.
[0034] In the present invention, retinal images are used as data sources, vascular segments are used as the main research objects, and the identification and relationship construction of vascular segments are used as the ultimate goal. A convolutional neural network is used to segment the blood vessels, and a convolution algorithm is used to extract the skeleton of the blood vessels. The vascular skeleton is segmented according to geometric relationships, nodes are extracted, and node categories are divided. Finally, a directed vascular tree is used to reconstruct the vascular structure diagram; by cutting off the optic disc area in the retinal image information, it is convenient to find the starting point of the blood vessel from the overlapping arteries and veins in the optic disc area, thereby forming a directed vascular tree, and the directed vascular tree is used to reconstruct the topological structure diagram of the blood vessels and the image information of the blood vessels, so that different blood vessels can be distinguished. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:
[0036] Figure 1 The flowchart is a preferred embodiment of the retinal vascular reconstruction method based on the directed vascular tree of the present invention.
[0037] Figure 2 Schematic diagram of the acquired retinal image information.
[0038] Figure 3 Schematic diagram of retinal blood vessels segmented from retinal image information.
[0039] Figure 4 This is a schematic diagram of the retinal blood vessels after the optic disc area is removed.
[0040] Figure 5 Schematic diagram of the extracted nodes.
[0041] Figure 6 A schematic diagram for calculating the angle formed by the skeleton line connecting two end nodes and the branch node at the branch node.
[0042] Figure 7 Schematic diagram of constructing a directed vascular tree.
[0043] Figure 8 Schematic diagram of the reconstructed topological structure of retinal blood vessels. DETAILED DESCRIPTION
[0044] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.
[0045] like Figure 1 As shown, a preferred embodiment of the retinal vascular reconstruction method based on the directed vascular tree of the present invention includes the following steps:
[0046] S1. Obtain retinal image information; Obtain retinal image information such as Figure 2 shown.
[0047] S2, such as Figure 3 As shown, a convolutional neural network is used to segment retinal blood vessels from retinal image information. A convolutional neural network can achieve semantic segmentation of images, so this embodiment uses a convolutional neural network to segment retinal blood vessels as the basis for blood vessel description.
[0048] S3. Extract the skeleton line of the retinal blood vessels using a convolution algorithm; the convolution algorithm preferably adopts an 8-neighboring point method, and the convolution kernel is as follows:
[0049] mask2=[1 1 1; 1 0 1; 1 1 1].
[0050] S4, first cut off the optic disc area in the retinal image information, and the retinal image information after cutting off the optic disc area is as follows: Figure 4 As shown; then, end nodes and branch nodes are extracted from the skeleton line according to the geometric relationship, laying the foundation for the segmentation and reconstruction of blood vessels. The method for extracting end nodes and branch nodes from the skeleton line according to the geometric relationship is:
[0051] Calculate the number of adjacent points of all points in the skeleton line of the blood vessel, and select the points with 1, 3, and more than 3 adjacent points as nodes. The selected nodes are as follows: Figure 5 As shown in the figure, nodes with 1 neighboring point are defined as end nodes, nodes with 3 neighboring points are defined as branch nodes, and nodes with more than 3 neighboring points are defined as cross nodes. Cross nodes are removed, and end nodes and branch nodes are retained.
[0052] S5. Segment the retinal blood vessels into blood vessel segments through the branch nodes, and extract the skeleton line information and outer contour information of each blood vessel segment; the method for extracting the skeleton line information and outer contour information of the blood vessel segment is:
[0053] A curvature threshold is preset, and the skeleton line of the vascular segment and the pixel points on the outer contour curves on both sides are extracted in turn. Then, three adjacent pixel points on the same curve are connected by parabolas to form a curve model. The curvature of the curve at the middle pixel point is calculated as the curvature at the pixel point. The pixel points at both ends of the skeleton line and each outer contour curve and the pixel points with a curvature greater than the curvature threshold are screened out as the key points of the skeleton line and the outer contour curve; then, serial numbers are assigned to the key points of the skeleton line and the outer contour curve from one end point to another along the direction of the skeleton line and the outer contour curve; the coordinates and serial numbers of all the key points on the skeleton line and the outer contour curve of the vascular segment are used as the skeleton line and outer contour information of the vascular segment. Since the skeleton line and the outer contour curve include many pixels, if each pixel is stored, the amount of information to be stored is too large, and the pixels are limited by the resolution and are distorted after being magnified. The coordinates of the pixel points cannot accurately describe the shape of the outer contour curve of the blood vessel tree. Therefore, this embodiment sets a curvature threshold to remove points that have little impact on the direction of the curve and only retains key points, thereby greatly reducing the amount of stored information and facilitating blood vessel reconstruction.
[0054] S6, constructing a directed vascular tree with a topological structure through the skeleton line information and outer contour information of the end nodes, branch nodes and vascular segments; specifically, the following sub-steps may be included:
[0055] S61, find the branch node connected to the two end nodes in the skeleton line. At this time, an angle threshold can also be preset (for example, the angle threshold can be set to 90°). If the angle formed by the skeleton line connecting the two end nodes and the branch node at the branch node is less than the preset angle threshold, it is considered that the branch node and the two end nodes constitute a subtree, and step S62 is executed; otherwise, it is considered that the branch node and the two end nodes are not a subtree, and step S61 is returned to continue to search for the subtree.
[0056] like Figure 6 As shown, the method for calculating the angle formed by the skeleton line connecting the two end nodes and the branch node at the branch node is preferably to draw a circle with the branch node as the center, the circle and the corresponding two skeleton lines intersect at point a and point b respectively, connect point a and point b to the branch node through straight line segments respectively, and calculate the angle between the two straight line segments as the angle formed by the skeleton line connecting the two end nodes and the branch node at the branch node. For example, Figure 7As shown, branch node 4 is connected to end node 6 and end node 7, and the angle of the corresponding skeleton lines is less than the angle threshold, so branch node 4 and end node 6 and end node 7 constitute a subtree; although branch node 1 is connected to end node 0 and end node 2, the angle of the corresponding skeleton lines is greater than the angle threshold, so branch node 1 and end node 0 and end node 2 do not constitute a subtree.
[0057] S62. Taking the branch node as the root node and the two end nodes as leaf nodes, a subtree is formed in combination with the outer contour information of the corresponding two blood vessel segments; the information of the subtree includes the information of the branch node and the information of the two end nodes, as well as the skeleton line information and outer contour information of the corresponding two blood vessel segments segmented by the branch node and the two end nodes; the information of the branch node includes the coordinates of the branch point. When the end node is a super node, the information of the end node is the information of the subtree stored in the super node. When the end node is a non-super node, the information of the end node is the coordinates of the end node.
[0058] S63, merge the information of the subtrees together as a supernode, set the supernode as an end node and replace the branch node in the subtree; for example, Figure 7 As shown, branch node 4 is used as the root node, and end node 6 and end node 7 are used as leaf nodes to form a subtree, and the information of the subtree is fused together as a super node, as shown in FIG. Figure 7 As shown by the dotted line in ; at this time, the branch node 4 is replaced by the end node 4 composed of the super node.
[0059] S64, repeat steps S61 to S63 until all remaining nodes are end nodes. Figure 7 As shown, after node 4 is replaced by end node 4 composed of super nodes, branch node 2, end node 4, and end node 4 form a subtree, and then the super node formed by the subtree replaces branch node 2 to form end node 2. After that, branch node 1, end node 2, and end node 3 form a subtree, and a super node is formed to replace branch node 1, so that only end node 0 and end node 1 are left.
[0060] S65, using the non-super node of the two mutually connected end nodes as the starting root node of the directed vascular tree, and using the super node as the leaf node to form a directed vascular tree with a topological structure, and storing the information of the directed vascular tree. The information of the vascular tree includes the information of the two end nodes, and the skeleton line information and outer contour information of the corresponding vascular segment; when the end node is a super node, the information of the end node is the information of the subtree stored in the super node, and when the end node is the starting root node, the information of the end node is the coordinates of the end node. Figure 7 As shown, finally, the non-super node end node 0 and the super node end node 1 are left, so that a directed vessel tree is formed with the end node 0 as the starting root node.
[0061] S7. Set the type of directed vascular tree, which includes an arterial vascular tree and a venous vascular tree. At this time, the color of the directed vascular tree after reconstruction can also be set. In order to distinguish different blood vessels after reconstruction, each directed vascular tree can be assigned a different color.
[0062] S8. Reconstruct the skeleton line of the retinal blood vessels and / or the image information of the blood vessels in the form of a topological map using the directed blood vessel tree. Preferably, the skeleton line information and / or the image information of the directed blood vessel tree are reconstructed according to the stored information of each directed blood vessel tree, so as to reconstruct the structural diagram of the retinal blood vessels. At this time, only the topological map of the skeleton line of the directed blood vessel tree can be reconstructed as needed, or only the image information of the directed blood vessel tree can be reconstructed, or the topological map of the skeleton line of the directed blood vessel tree and the image information of the directed blood vessel tree can be reconstructed at the same time. Each directed blood vessel tree is displayed in a specified color for easy distinction. For areas where the image information of the directed blood vessel trees overlap, the image information of the venous blood vessel tree in the area is displayed, and the image information of the arterial blood vessel tree in the area is hidden.
[0063] Restoring the skeleton lines of each directional vascular tree and the image information of each directional vascular tree according to the stored information of each directional vascular tree may specifically include the following sub-steps:
[0064] S81, extracting information of two end nodes and key point information of the skeleton line and outer contour of the corresponding blood vessel segment from the stored information of the blood vessel tree; taking the coordinates of the starting root node as one end of the blood vessel segment and the coordinates of the leaf node as the other end of the blood vessel segment, reconstructing the skeleton line and image information of the blood vessel segment according to the key point information of the skeleton line and outer contour. Figure 7 For the directed vascular tree in , first extract the information of end node 0 and end node 1 (super node at this time), as well as the skeleton line and key point information of the outer contour of the vascular segment connecting end node 0 and end node 1 (the same position as branch node 1), and reconstruct the skeleton line of the vascular segment and the image information of the vascular segment.
[0065] The method for reconstructing the skeleton line and image information of the blood vessel segment according to the key point information of the skeleton line and the outer contour is as follows:
[0066] The coordinates and serial numbers of the key points of the skeleton line and the outer contour curve are extracted from the key point information of the skeleton line and the outer contour of the blood vessel segment, and the spline curves connecting the key points with adjacent serial numbers on the skeleton line and the outer contour curve are drawn in sequence using the construction method of the quadratic interpolation spline curve to generate the skeleton line and the outer contour of the blood vessel segment, and reconstruct the skeleton line of the blood vessel segment and the image of the blood vessel segment. Since the skeleton line and the outer contour curve are generated by regenerating the intermediate connection points according to the coordinates of the key points using the construction method of the quadratic interpolation spline curve, they can be arbitrarily enlarged or reduced without distortion.
[0067] S82, when the leaf node is a super node, restore the leaf node to a subtree; the method is: extract the coordinates of the branch node, the coordinates of the two end segments, and the key point information of the skeleton lines and outer contours of the corresponding two blood vessel segments from the super node; use the coordinates of the branch node as one end of the two blood vessel segments, and use the coordinates of the two leaf nodes as the other ends of the two blood vessel segments, and reconstruct the skeleton lines of the two blood vessel segments and the image information of the two blood vessel segments according to the key point information of the skeleton lines and outer contours of the two blood vessel segments. For example, Figure 7 As shown, after reconstructing the skeleton line and blood vessel segment image information between the end node 0 and the end node 1, the information of the branch node 1, the end node 2 (the super node at this time) and the end node 3 are extracted from the end node 1 (the super node at this time), and the skeleton line and blood vessel segment image information between the branch node 1 and the end node 2 (the same position as the branch node 2), as well as the skeleton line and blood vessel segment image information between the branch node 1 and the end node 3 are reconstructed.
[0068] S83: When the leaf node of the subtree includes a supernode, continue to execute step S82 to restore the leaf node to a subtree until all supernodes are restored to subtrees. Figure 7 As shown, the information of branch node 2, end node 4 (super node) and end node 5 is extracted from end node 2 (super node at this time), and the corresponding skeleton line and blood vessel segment image information are reconstructed; finally, the information of branch node 4, end node 6 and end node 7 is extracted from end node 4 (super node at this time), and the corresponding skeleton line and blood vessel segment image information is reconstructed, thereby completing Figure 7 Reconstruction of directed vascular tree in .
[0069] When all directed vascular trees are reconstructed, the skeleton topology diagram of the retinal blood vessels and the structure diagram of the blood vessels are reconstructed. Figure 8As shown in the figure, the skeleton topology of the retinal blood vessels finally reconstructed has a total of 11 directed blood vessel trees, each of which forms a connection relationship. By displaying each blood vessel tree in a different color, it is convenient to observe the blood vessel trees one by one. In addition, in order to avoid mistakenly truncating the same blood vessel tree in the process of generating the directed blood vessel tree, these directed blood vessel trees can be further merged manually according to the original image information to finally construct the arteriovenous blood vessel tree.
[0070] In this embodiment, the retinal image is used as the data source, the vascular segment is used as the main research object, and the identification and relationship construction of the vascular segment is used as the ultimate goal. The convolutional neural network is used to segment the blood vessels, the convolution algorithm is used to extract the skeleton of the blood vessels, the vascular skeleton is segmented according to the geometric relationship, the nodes are extracted, and the node categories are divided. Finally, the vascular structure diagram is reconstructed using a directed vascular tree; by cutting off the optic disc area in the retinal image information, it is convenient to find the starting point of the blood vessel from the overlapping arteries and veins in the optic disc area, thereby forming a directed vascular tree, and the construction method of the quadratic interpolation spline curve is used to reconstruct the outer contour curve of the vascular segment. After enlargement, there will be no distortion, which is convenient for accurately obtaining retinal vascular information and obtaining local information of fine blood vessels, and reduces the graphics storage space. Through the directed vascular tree, not only the structural diagram of the blood vessel can be reconstructed, but also the topological structure diagram of the vascular skeleton line can be reconstructed, which is convenient for distinguishing different blood vessels.
[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.
Claims
1. A retinal vascular reconstruction method based on a directed vascular tree, characterized in that: The following steps are involved: S1, obtaining retinal image information; S2, using convolutional neural network to segment retinal blood vessels from retinal image information; S3, using convolution algorithm to extract the skeleton line of retinal blood vessels; S4, cutting off the optic disc area in the retinal image information, and extracting the end nodes and branch nodes from the skeleton line according to the geometric relationship; S5, segmenting the retinal blood vessels into blood vessel segments through branch nodes, and extracting skeleton line information and outer contour information of each blood vessel segment; S6, constructing a directed vascular tree with a topological structure through skeleton line information and outer contour information of end nodes, branch nodes and vascular segments; S7. Setting a type of a directed vascular tree, wherein the type of the directed vascular tree includes an arterial vascular tree and a venous vascular tree; S8. Reconstructing the skeleton lines of retinal blood vessels and / or the image information of blood vessels in the form of a topological graph using the directed blood vessel tree; In the step S6, the method of constructing a directed vascular tree of topological structure through the skeleton line information of the end nodes, branch nodes and vascular segments and the key points of the outer contour is: S61, finding a branch node connected to two end nodes in the skeleton line; if the angle formed by the skeleton line connecting the two end nodes and the branch node at the branch node is less than a preset angle threshold, executing step S62; Otherwise, return to execute step S61; S62, taking the branch node as the root node, taking the two end nodes as leaf nodes, and combining the skeleton line information and the outer contour information of the corresponding two blood vessel segments to form a subtree; S63, integrating the information of the subtrees as a supernode, setting the supernode as an end node and replacing the branch node in the subtree; S64, repeat steps S61 to S63 until all remaining nodes are end nodes; S65 , using the non-super node of the two mutually connected end nodes as the starting root node and the super node as the leaf node to form a directed vascular tree with a topological structure, and storing information of the directed vascular tree.
2. The retinal vascular reconstruction method based on a directed vascular tree according to claim 1, characterized in that: In the step S3, when the skeleton line of the retinal blood vessels is extracted using the convolution algorithm, the convolution algorithm adopts the 8-neighboring point method, and the convolution kernel is as follows: mask2 = [111; 101; 111].
3. The retinal vascular reconstruction method based on a directed vascular tree according to claim 2, characterized in that: In the step S4, the method of extracting the end nodes and branch nodes from the skeleton line according to the geometric relationship is: The number of adjacent points of all points in the skeleton line of the blood vessel is calculated, and the points with 1 and 3 adjacent points are selected as nodes. The nodes with 1 adjacent point are defined as end nodes, and the nodes with 3 adjacent points are defined as branch nodes.
4. The retinal vascular reconstruction method based on a directed vascular tree according to claim 1, characterized in that: In step S5, the method for extracting the skeleton information and outer contour information of the blood vessel segment is: A curvature threshold is preset, and the skeleton line of the blood vessel segment and the pixel points on the outer contour curves on both sides are extracted in turn. Then, three adjacent pixel points on the same curve are connected in turn to form a curve model, and the curvature of the curve at the middle pixel point is calculated as the curvature at the pixel point. The pixel points at both ends of the skeleton line and each outer contour curve and the pixel points with a curvature greater than the curvature threshold are screened out as the key points of the skeleton line and the outer contour curve; then, serial numbers are assigned to the key points of the skeleton line and the outer contour curve from one end point to another along the direction of the skeleton line and the outer contour curve; the coordinates and serial numbers of all the key points on the skeleton line and the outer contour curve of the blood vessel segment are used as the skeleton line information and outer contour information of the blood vessel segment.
5. The retinal vascular reconstruction method based on a directed vascular tree according to claim 1, characterized in that: The information of the subtree includes the information of the branch node and the information of the two end nodes, as well as the skeleton line information and the outer contour information of the two blood vessel segments divided by the branch node and the two end nodes; the information of the branch node includes the coordinates of the branch point, when the end node is a super node, the information of the end node is the information of the subtree stored by the super node, when the end node is a non-super node, the information of the end node is the coordinates of the end node; the information of the blood vessel tree includes the information of the two end nodes, as well as the skeleton line information and the outer contour information of the corresponding blood vessel segments; when the end node is a super node, the information of the end node is the information of the subtree stored by the super node, when the end node is a starting root node, the information of the end node is the coordinates of the end node.
6. The retinal vascular reconstruction method based on directed vascular tree according to claim 5, characterized in that: In the step S8, the method of reconstructing the image of the retinal blood vessels in the form of a topological map using the directed blood vessel tree is as follows: the skeleton line information of each directed blood vessel tree and / or the image information of the directed blood vessel tree are restored based on the stored information of each directed blood vessel tree, and for the area where the image information of the directed blood vessel trees overlap with each other, the image information of the venous blood vessel tree in the area is displayed, and the image information of the arterial blood vessel tree in the area is hidden.
7. The retinal vascular reconstruction method based on a directed vascular tree according to claim 6, characterized in that: The method for restoring the skeleton line information of each directional vascular tree and / or the image information of the directional vascular tree according to the stored information of each directional vascular tree is: Extracting information of two end nodes and key point information of the skeleton line and / or outer contour of the corresponding blood vessel segment from the stored information of the blood vessel tree; using the coordinates of the starting root node as one end of the blood vessel segment and the coordinates of the leaf node as the other end of the blood vessel segment, and reconstructing the skeleton line of the blood vessel segment and / or the image information of the blood vessel segment according to the key point information of the skeleton line and / or outer contour; When a leaf node is a super node, the leaf node is restored to a subtree; the method is: extracting the coordinates of the branch node, the coordinates of the two end segments, and the key point information of the skeleton lines and / or outer contours of the corresponding two blood vessel segments from the super node; using the coordinates of the branch node as one end of the two blood vessel segments, and using the coordinates of the two leaf nodes as the other ends of the two blood vessel segments, respectively reconstructing the skeleton lines of the two blood vessel segments and / or the image information of the two blood vessel segments according to the key point information of the skeleton lines and / or outer contours of the two blood vessel segments; When the leaf nodes of a subtree include a supernode, the leaf nodes are continuously restored to the subtree until all the supernodes are restored to the subtree.
8. The retinal vascular reconstruction method based on a directed vascular tree according to claim 7, characterized in that: The method for reconstructing the skeleton line of the blood vessel segment and / or the image information of the blood vessel segment according to the key point information of the skeleton line and / or the outer contour is: The coordinates and serial numbers of each key point of the skeleton line and / or outer contour curve are extracted from the key point information of the skeleton line and / or outer contour, and the spline curves connecting the key points with adjacent serial numbers on the skeleton line and / or outer contour curve are drawn in sequence using the quadratic interpolation spline curve construction method to generate the skeleton line and / or outer contour of the blood vessel segment, and reconstruct the skeleton line of the blood vessel segment and / or the image information of the blood vessel segment.
