A retinal arteriovenous vessel identification method and system based on image recognition

Abstract

This invention belongs to the field of vascular recognition technology, specifically relating to a method and system for retinal arteriovenous vessel recognition based on image recognition. The invention includes determining the structural features of a vascular path, including at least one of branch points and bending structures. In response to the determination of these structural features, the vascular path is classified into main vessels or branch vessels based on preset classification rules utilizing these features. This invention divides vascular nodes into a set of main control vascular points and a set of secondary control vascular points. Branch points are determined by calculating the offset of adjacent nodes of the main control vascular points, and bending structures are determined by calculating the introduced force based on the distribution of the secondary control vascular point set. This utilizes the different characteristics of the main control vascular point set and the secondary control vascular point set to differentiate between these two key topological structures—branch points and bending structures—avoiding the misjudgment of structures caused by treating all vascular nodes equally, thus improving the accuracy of structural feature recognition.

Description

Technical Field

[0001] This invention belongs to the field of vascular recognition technology, specifically relating to a method and system for retinal arteriovenous vessel recognition based on image recognition. Background Technology

[0002] Retinal imaging plays a crucial role in the monitoring, diagnosis, and management of various ophthalmic and systemic diseases. Specifically, as the only part of the body where blood vessels and nerve tissue can be directly observed, the morphology, distribution, and health status of the retina's vascular system are key windows for assessing the risk of diseases such as diabetic retinopathy, glaucoma, and hypertension. Accurate analysis of retinal vascular images and precise identification and localization of early pathological features such as retinal edema are of great significance for early warning and timely intervention of diseases.

[0003] However, existing technologies generally rely on simple and unstable image features for classifying arteries and veins, especially the color of the vessels. In fact, the color of retinal vessels varies greatly due to the influence of imaging equipment, lighting conditions, individual patient differences, and the pathological state itself, resulting in poor robustness of color-based classification algorithms and a tendency to make misjudgments.

[0004] To address the aforementioned problems, this invention proposes a method and system for identifying retinal arteries and veins based on image recognition. Summary of the Invention

[0005] The purpose of this invention is to provide a retinal arteriovenous vessel identification method based on image recognition, which can introduce more explicit judgment conditions into the classification process during the classification of retinal vessels, thereby improving classification accuracy while reducing the system's execution complexity.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A classification method for retinal vascular pathways based on image recognition includes the following steps: Identify vascular pathways including multiple vascular nodes from retinal images; determine the structural features of the vascular pathways, including at least one of branch points and bends. In response to the determination of structural features, the vascular pathways are classified into main vessels or branch vessels based on the preset classification rules that utilize the structural features. The process of identifying a vascular path including multiple vascular nodes from a retinal image includes: preprocessing the retinal image to generate initial image data; extracting vascular morphological features based on the initial image data; matching the vascular morphological features with the initial image data to generate vascular image data; and identifying the vascular path and obtaining multiple vascular nodes in the vascular path based on the vascular image data.

[0007] Preferably, determining the branch points of the vascular pathway includes: Multiple vascular nodes are divided into a set of primary control vascular nodes and a set of secondary control vascular nodes; adjacent nodes of the primary control vascular nodes in the set of primary control vascular nodes are extracted as potential branch points; Based on the preset branch formation conditions, the offset of potential branch points is calculated; and when the offset meets the preset branch determination conditions, the potential branch point is determined as a branch point.

[0008] Preferably, the tortuous structure for determining the vascular pathway includes: The distribution of secondary control vessel points in the secondary control vessel point set within the vessel path is statistically analyzed, and the corresponding attraction force is calculated. And when the introduced force is greater than the preset bending threshold, the blood vessel path is determined to be a bending structure, and the blood vessel nodes within the bending structure are marked as bending points.

[0009] Preferably, the preset classification rule based on structural features is as follows: When a bent structure is determined, calculate the distance between the bend point and the branch point; If the distance is less than a preset distance threshold, the blood vessel path containing the branch point is identified as a branch vessel; if the distance is greater than or equal to the preset distance threshold, the blood vessel path containing the branch point is identified as a main vessel.

[0010] Preferably, the preset classification rules further include: When the existence of a bending structure is not determined, the matching score between the branch point and the preset standard sample is calculated; the matching scores are sorted, and the blood vessel path with the higher score in the sorting results is identified as the main blood vessel, and the blood vessel path with the lower score is identified as the branch blood vessel.

[0011] Preferably, calculating the matching score between the branch point and the preset standard sample includes: Obtain the location and orientation distribution of branch points; based on the location and orientation distribution, calculate the branch proportion of the vascular path associated with the branch point; compare the branch proportion with the preset standard sample to generate a matching score.

[0012] This invention also discloses a retinal arteriovenous vessel recognition system based on image recognition, comprising the following modules: A vascular path recognition module is used to identify vascular paths, including multiple vascular nodes, from retinal images; The structural feature determination module is used to determine the structural features of the vascular path based on multiple vascular nodes. The structural features include at least one of branch points and bending structures. In addition, a blood vessel classification module is used to classify blood vessel paths into main vessels or branch vessels based on the pre-defined classification rules using the structural features, in response to the structural feature determination module's determination of structural features.

[0013] Preferably, determining the branch points of the vascular pathway includes: Multiple vascular nodes are divided into a set of primary control vascular nodes and a set of secondary control vascular nodes; adjacent nodes of the primary control vascular nodes in the set of primary control vascular nodes are extracted as potential branch points; Based on the preset branch formation conditions, the offset of potential branch points is calculated; when the offset meets the preset branch determination conditions, the potential branch point is determined as a branch point.

[0014] Preferably, the tortuous structure for determining the vascular pathway includes: The distribution of secondary control vessel points in the secondary control vessel point set within the vessel path is statistically analyzed, and the corresponding attraction force is calculated. And when the introduced force is greater than the preset bending threshold, the blood vessel path is determined to be a bending structure, and the blood vessel nodes within the bending structure are marked as bending points.

[0015] Preferably, the blood vessel classification module is specifically configured as follows: In response to the determination of the existence of a bend, the distance between the bend point and the branch point is calculated, and the vascular path is classified as a main vessel or a branch vessel based on the comparison result of the distance with a preset distance threshold. In response to the absence of a bending structure, the matching score between the branch point and the preset standard sample is calculated, and the vascular path is classified as a main vessel or a branch vessel based on the ranking of the matching scores. Beneficial effects

[0016] This invention divides vascular nodes into a set of primary control vascular points and a set of secondary control vascular points. By calculating the offset of adjacent nodes of the primary control vascular point, branch points are determined. Based on the distribution of the secondary control vascular point set, the introduced force is calculated to determine the bending structure. By utilizing the different characteristics of the primary control vascular point set and the secondary control vascular point set, the invention differentiates the two key topological structures, namely branch points and bending structures, avoiding the misjudgment of structures caused by treating all vascular nodes the same, and improving the accuracy of structural feature recognition.

[0017] When a bend is identified in a vascular pathway, this invention classifies the pathway based on the distance between the bend and the branch point; when no bend is identified, it classifies the pathway based on the matching score between the branch point and a preset standard sample. This allows for adaptive selection of classification criteria based on the geometric complexity of the vascular pathway, ensuring both accuracy and universality of the classification. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention. Example

[0020] See Figure 1 This embodiment provides a classification method for retinal vascular pathways based on image recognition, including the following steps: S1. Acquire the retinal image of the object to be tested, and preprocess the retinal image to generate initial image data; This preprocessing is used to transform raw retinal images, which may contain noise and uneven illumination, into standardized data with uniform format and distinctive features, laying the foundation for subsequent accurate analysis. The specific preprocessing includes multi-level cropping and noise reduction. The multi-level cropping steps are as follows: Coarse cropping involves automatically identifying and removing obviously irrelevant black background areas at the image edges by analyzing the overall brightness and color distribution of the image. Fine cropping involves using key landmarks in the retinal anatomy as references, such as the optic disc, which serves as the center of blood vessel convergence and appears as a bright circular area in the image. This allows for the precise definition of the effective analysis area containing the complete vascular network, thereby eliminating interference from non-vascular structures such as the optic disc and macula for subsequent blood vessel extraction.

[0021] The noise reduction process applies a preset image smoothing workflow to effectively suppress random noise introduced by imaging equipment or environmental factors and enhance the contrast between blood vessels and the background.

[0022] After processing, the pixel information of the image is converted into a standardized numerical matrix. For example, the gray value of each pixel is linearly mapped to the range of 0 to 1 to form the initial image data. This data format facilitates subsequent unified numerical calculations.

[0023] The image smoothing process involves replacing the current pixel value with the median or weighted average intensity value within the pixel's neighborhood.

[0024] S2. Based on the initial image data, extract the vascular morphology features of the blood vessels and match the vascular morphology features with the initial image data to generate vascular image data; the main purpose is to accurately segment the complete vascular network from the background.

[0025] The process of extracting vascular morphological features includes: By using preset pixel filtering rules, a set of pixels that conform to the characteristics of blood vessels is initially identified in the initial image data, forming a connected blood vessel region; The preferred pixel filtering rule is a judgment logic based on pixel intensity threshold or color channel difference.

[0026] A multi-stage boundary refinement process is performed on the vascular region to obtain robust features that are insensitive to changes in illumination; this process specifically includes: The intensity gradient of each point in the image is calculated, and then non-maximum gradient points are suppressed. Through double threshold connection processing, the single-pixel width structural boundary representing the blood vessel contour is accurately delineated, which is the blood vessel morphology feature.

[0027] The steps for matching vascular morphological features with initial image data include: The extracted structural boundaries are used as a verification template and superimposed on the initial image data. The feature similarity between each point on the structural boundary and its corresponding position in the original image is calculated to generate a quantified consistency score. This score is calculated based on the continuity of the boundary pixels with their neighboring pixels in terms of gradient direction and intensity.

[0028] A preset matching threshold is set. If the consistency score of a certain boundary segment is greater than the threshold, the corresponding pixel area is confirmed to be a real blood vessel.

[0029] Generate a binary image containing only all confirmed real blood vessel parts, where blood vessel pixels are assigned a value of 1 and background pixels are assigned a value of 0; this is the blood vessel image data.

[0030] S3. Based on vascular image data, identify vascular paths and obtain multiple vascular nodes in the vascular path to transform the two-dimensional vascular image into a one-dimensional path representation that is easy to perform topological structure analysis. The specific steps are as follows: The vascular region in the vascular image data is refined through an iterative pixel stripping process. Specifically, the pixel stripping process involves removing pixels layer by layer from the edge of the vascular region while maintaining the connectivity of the vascular network, until all blood vessels are refined to a center line of a single pixel width, which is the vascular path. Along each blood vessel path, a series of discrete points are obtained according to the preset sampling rules. These discrete points are the multiple blood vessel nodes in the blood vessel path, and these blood vessel nodes together constitute a digital description of the geometric morphology of the blood vessel.

[0031] The preferred preset sampling rule is to set a sampling point at fixed pixel intervals, or to set a sampling point at a location where the path curvature changes beyond a specific threshold.

[0032] Furthermore, it should be noted that the iterative pixel stripping process is preferably an image thinning algorithm, also known as a skeletonization algorithm. This algorithm removes pixels layer by layer, starting from the edge of the target region, i.e. the edge of the blood vessel, until the target is reduced to the center line of a single pixel width, while ensuring that its topological connectivity is not destroyed.

[0033] S4. Divide multiple vascular nodes into a set of primary control vascular nodes and a set of secondary control vascular nodes to classify them according to the importance of each node in the vascular network topology, so as to provide a basis for subsequent structure identification. The specific processing logic for the partitioning is as follows: For each vascular node, calculate the number of its neighboring nodes within a preset proximity radius, and define this number as the local connectivity of the node. The master vessel point is a key node with high structural complexity in the vascular network. It usually corresponds to the intersection or bifurcation point of the vessel. Therefore, all vascular nodes with local connectivity greater than the preset connectivity threshold are included in the master vessel point set.

[0034] Secondary control vessel points are defined as ordinary nodes that constitute the linear course of a vessel. All vessel nodes with local connectivity equal to or less than a preset connectivity threshold are included in the secondary control vessel point set.

[0035] By dividing the nodes as described above, we can effectively distinguish between the few nodes that represent structural changes and the majority nodes that represent the path itself.

[0036] The local connectivity greater than the preset connectivity threshold can preferably be greater than 2, indicating that the point is connected to at least three path segments.

[0037] Furthermore, local connectivity refers to the number of directly connected neighboring nodes within a predetermined radius for any given blood vessel node. This value is used to quantify the complexity of a node in its local topology.

[0038] S5. Based on the set of primary control vessel points and the set of secondary control vessel points, determine the branching and bending structures in the vessel path; Specifically, by analyzing the spatial arrangement and association of two types of nodes, the key vascular morphological features of blood vessels are identified. This process is divided into two parts: determining the branch structure and determining the bending structure.

[0039] The steps for determining the branch structure are as follows: Taking any one of the master control vessels in the master control vessel set as the center, identify all the secondary control vessels that are directly connected to it. All of the above secondary control vessels are considered as potential branch points. For any two potential branch points belonging to different vascular pathways, calculate the angle between the two line segments connecting the two potential branch points with the main controlling vascular point as the vertex. This angle is defined as the branch divergence. Determine whether the branch divergence meets the preset branch determination condition, that is, whether the divergence is greater than the preset branch angle threshold; If the branch divergence is greater than the branch angle threshold, it indicates that there is a significant vascular bifurcation, and a branch structure is determined. The main control vessel point is marked as a branch point. If the branch divergence is less than or equal to the branch angle threshold, the two potential branch points are considered to belong to the same continuous path and do not constitute a valid branch. They are maintained in the set of secondary control vessel points.

[0040] The steps for determining the bent structure are as follows: In the vascular pathway, identify the path segment consisting of consecutive secondary control vascular points; Calculate the node density of each such path segment, that is, the number of secondary control vessel points contained within a unit path length. Due to the sampling rules, the curved parts of the vessels usually generate higher node density. Define this node density as the curvature indicator value of the path. Before determining whether the indicator value is greater than the preset bending threshold, the indicator value is corrected according to the preset bending offset value to compensate for the systematic deviation introduced by different image resolutions or vessel thicknesses, and generate the corrected curvature indicator value. If the correction curvature indicator value is greater than the preset bending threshold, then the cumulative angle change formed by the connection between adjacent nodes on the path segment is further calculated. If the cumulative angle change is also greater than the preset angle change threshold, then the path segment is finally determined to be a bending structure, and all secondary control vessel points in the structure are marked as bending points.

[0041] S6. Based on the branch point and bend point, or based on the matching score of the branch point, determine the vascular path as a main vessel or a branch vessel. This step integrates the structural information identified in the previous step to determine the attributes of blood vessels. This determination is based on the presence of bending structures and uses the following two methods: If a bend is identified in the vascular pathway, the distance along the vascular pathway between the branch point and the nearest bend point is calculated.

[0042] Since the main blood vessel is usually relatively straight, while the branch vessels often branch off from the main blood vessel at a certain angle and may be accompanied by bends, if the distance is less than the preset distance threshold, it indicates that the branch is close to the bend, which is consistent with the typical shape of the branch vessel. Therefore, the blood vessel path containing the branch point is identified as the branch vessel. Conversely, if the distance is greater than or equal to the threshold, it is determined to be the main blood vessel; If no bending structure is found in the vascular path, the matching score is calculated by comparing the local geometric features of the branch points with preset standard samples.

[0043] The aforementioned preset standard samples store a large set of typical parameters for confirmed branch points of main and tributary vessels, such as branch angles and the ratio of vessel diameters before and after branching. The specific process for calculating the matching score is as follows: Extract the geometric feature parameters around the current branch point to be determined; calculate the quantitative difference between this set of parameters and the standard feature sets of the main blood vessel and the branch blood vessel in the preset standard samples respectively; The matching score is generated in reverse based on the magnitude of the difference. The smaller the difference, the higher the score. The branch point to be judged is classified into the category with the higher score. That is, if its matching score with the standard feature set of the main blood vessel is higher, the blood vessel path is determined to be the main blood vessel, otherwise it is determined to be a branch blood vessel.

[0044] The matching score is a quantized score that represents the degree of similarity between the local geometric features of the branch point to be judged and the typical feature set of a certain category in the standard feature library. This score is generated inversely from the quantized difference score; a higher score indicates a higher similarity.

[0045] Quantitative difference is a numerical value used to measure the difference between the set of geometric feature parameters of the branch point to be determined and a certain standard feature set in the standard feature library. For example, it is obtained by calculating the weighted sum of squares of the differences of each parameter. Example

[0046] See Figure 2 This embodiment provides a classification system for retinal vascular pathways based on image recognition, including: The vascular path recognition module is configured to identify vascular paths including multiple vascular nodes from the input retinal image. In the specific execution process, the module preprocesses the original retinal image, that is, performs multi-level cropping and denoising processing to generate initial image data that is more conducive to subsequent processing.

[0047] The vascular path recognition module extracts the vascular morphology features of blood vessels based on the initial image data by performing a multi-stage boundary refinement process; and matches the extracted vascular morphology features with the initial image data to generate a clear and connected vascular image data, which accurately depicts the vascular network in the retina.

[0048] Based on the vascular image data, the vascular path recognition module identifies individual vascular paths through an iterative pixel stripping process, and discretizes each path into a sequence of multiple ordered vascular nodes for subsequent analysis.

[0049] The structural feature determination module is configured to receive the vascular path and its multiple vascular nodes output by the vascular path recognition module. It is mainly used to determine the structural features of the vascular path, specifically including at least one of branch points and bending structures.

[0050] To determine the branch point, the following steps are taken: This module divides multiple vascular nodes in a vascular path into a set of primary control vascular points and a set of secondary control vascular points based on the local connectivity of each vascular node. It extracts the adjacent nodes of each primary control vascular point in the set of primary control vascular points and uses these adjacent nodes as potential branch points. For each potential branch point, it calculates its offset according to preset branch formation conditions. When the offset meets the preset branch determination conditions, the module finally determines the potential branch point as a branch point.

[0051] To determine the bending structure, the following steps are used: This module focuses on the set of secondary control vessel points, statistically analyzes the spatial distribution of these points in the vessel path, and calculates a quantitative index, namely the induction force, based on this distribution. This induction force reflects the degree of curvature in a local area of ​​the path. When the calculated induction force is greater than a preset bending threshold, the module determines that the vessel path segment constitutes a bending structure and marks all vessel nodes within the bending structure as bending points.

[0052] The vascular classification module is configured to classify vascular pathways into main vessels or branch vessels based on preset classification rules using structural features, in response to the structural feature determination module's determination of structural features. The classification logic of this module is conditional, invoking different classification rules based on the presence or absence of bending structures.

[0053] When the structural feature determination module determines that a bending structure exists, the module starts the first set of preset classification rules and calculates the distance between the bending point in the bending structure and the branch point on the path. If the distance is less than the preset distance threshold, it means that the bend is adjacent to the branch, which is usually a typical feature of branch vessels. Therefore, the module determines the vessel path containing the branch point as a branch vessel. Conversely, if the distance is greater than or equal to a preset distance threshold, the module determines the blood vessel path as the main blood vessel.

[0054] When the structural feature determination module fails to identify the existence of a bent structure, the module activates the second set of preset classification rules to calculate the matching score between the branch point and the preset standard sample. The specific steps are as follows: The module obtains the location and orientation distribution of blood vessel segments around the branch point. Based on this location and orientation distribution, the module further calculates the branch proportion of the blood vessel path associated with the branch point. The calculated branch proportions are compared with preset standard samples to generate the final matching score. The module sorts all vascular pathways according to their matching scores. The blood vessel paths with higher scores in the ranking results are identified as main blood vessels, while the blood vessel paths with lower scores are identified as branch blood vessels.

[0055] The preset standard samples store the branch proportion characteristics of typical main blood vessels and tributaries.

Claims

1. A classification method for retinal vascular pathways based on image recognition, characterized in that, Includes the following steps: Identify vascular pathways including multiple vascular nodes from retinal images; determine the structural features of the vascular pathways, including at least one of branch points and bends. In response to the determination of structural features, the vascular pathways are classified into main vessels or branch vessels based on the preset classification rules that utilize the structural features. The process of identifying a vascular path including multiple vascular nodes from a retinal image includes: preprocessing the retinal image to generate initial image data; extracting vascular morphological features based on the initial image data; matching the vascular morphological features with the initial image data to generate vascular image data; and identifying the vascular path and obtaining multiple vascular nodes in the vascular path based on the vascular image data.

2. The method for retinal arteriovenous vessel identification based on image recognition according to claim 1, characterized in that, The branching points for determining the vascular pathway include: Multiple vascular nodes are divided into a set of primary control vascular points and a set of secondary control vascular points; adjacent nodes of the primary control vascular points in the set of primary control vascular points are extracted as potential branch points; Based on the preset branch formation conditions, the offset of potential branch points is calculated; and when the offset meets the preset branch determination conditions, the potential branch point is determined as a branch point.

3. The retinal arteriovenous vessel identification method based on image recognition according to claim 1, characterized in that, The tortuous structures that determine the vascular pathway include: The distribution of secondary control vessel points in the secondary control vessel point set within the vessel path is statistically analyzed, and the corresponding attraction force is calculated. When the attraction force is greater than a preset bending threshold, the vessel path is determined to be a bending structure, and the vessel nodes within the bending structure are marked as bending points.

4. The retinal arteriovenous vessel identification method based on image recognition according to claim 1, characterized in that, The pre-defined classification rules based on structural features are as follows: When a bent structure is determined, calculate the distance between the bend point and the branch point; If the distance is less than a preset distance threshold, the blood vessel path containing the branch point is identified as a branch vessel; if the distance is greater than or equal to the preset distance threshold, the blood vessel path containing the branch point is identified as a main vessel.

5. The method for identifying retinal arteries and veins based on image recognition according to claim 1, characterized in that, The preset classification rules also include: When the existence of a bending structure is not determined, the matching score between the branch point and the preset standard sample is calculated; the matching scores are sorted, and the blood vessel path with the higher score in the sorting results is identified as the main blood vessel, and the blood vessel path with the lower score is identified as the branch blood vessel.

6. A method for identifying retinal arteries and veins based on image recognition according to claim 5, characterized in that, The calculation of the matching score between the branch point and the preset standard sample includes: Obtain the location and orientation distribution of branch points; based on the location and orientation distribution, calculate the branch proportion of the vascular path associated with the branch point; compare the branch proportion with the preset standard sample to generate a matching score.

7. A retinal arteriovenous vessel recognition system based on image recognition, characterized in that, Includes the following modules: A vascular path recognition module is used to identify vascular paths, including multiple vascular nodes, from retinal images; The structural feature determination module is used to determine the structural features of the vascular path based on multiple vascular nodes. The structural features include at least one of branch points and bending structures. In addition, a blood vessel classification module is used to classify blood vessel paths into main vessels or branch vessels based on the pre-defined classification rules using the structural features, in response to the structural feature determination module's determination of structural features.

8. A retinal arteriovenous vessel recognition system based on image recognition according to claim 7, characterized in that, The branching points for determining the vascular pathway include: Multiple vascular nodes are divided into a set of primary control vascular points and a set of secondary control vascular points; adjacent nodes of the primary control vascular points in the set of primary control vascular points are extracted as potential branch points; Based on the preset branch formation conditions, the offset of potential branch points is calculated; when the offset meets the preset branch determination conditions, the potential branch point is determined as a branch point.

9. A retinal arteriovenous vessel recognition system based on image recognition according to claim 8, characterized in that, The tortuous structures that determine the vascular pathway include: The distribution of secondary control vessel points in the secondary control vessel point set within the vessel path is statistically analyzed, and the corresponding attraction force is calculated. And when the introduced force is greater than the preset bending threshold, the blood vessel path is determined to be a bending structure, and the blood vessel nodes within the bending structure are marked as bending points.

10. A retinal arteriovenous vessel recognition system based on image recognition according to claim 7, characterized in that, The vascular classification module is specifically configured as follows: In response to the determination of the existence of a bend, the distance between the bend point and the branch point is calculated, and the vascular path is classified as a main vessel or a branch vessel based on the comparison result of the distance with a preset distance threshold. In response to the absence of a bending structure, the matching score between the branch point and the preset standard sample is calculated, and the vascular path is classified as a main vessel or a branch vessel based on the ranking of the matching scores.

Images