WO2021208739A1

WO2021208739A1 - Method and apparatus for evaluating blood vessel in fundus color image, and computer device and medium

Info

Publication number: WO2021208739A1
Application number: PCT/CN2021/084542
Authority: WO
Inventors: 柳杨; 吕彬; 吕传峰
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-11-25
Filing date: 2021-03-31
Publication date: 2021-10-21
Also published as: CN112465772B; CN112465772A

Abstract

A method and apparatus for evaluating a blood vessel in a fundus color image, and a computer device and a medium. The method comprises: using a pre-trained segmentation model to extract a retinal blood vessel image of a topological structure from a fundus color image; extracting a blood vessel center line from the retinal blood vessel image, and setting an error band of the blood vessel center line by taking a preset standard blood vessel center line as a reference (S120); according to the error band, calculating an error of the blood vessel center line relative to the standard blood vessel center line, so as to obtain a blood vessel topological structure error of the retinal blood vessel image (S130); and according to the blood vessel topological structure error, evaluating the segmentation of the fundus color image by a segmentation model, and generating an evaluation result (S140), so that the blood vessel topological structure error is used for evaluating the integrity and accuracy of a blood vessel topological structure, and the evaluation effect is better.

Description

眼底彩照图像血管评估方法、装置、计算机设备和介质Method, device, computer equipment and medium for evaluating blood vessel of fundus color photograph image

本申请要求于2020年11月25日提交中国专利局、申请号为202011344724.X，发明名称为“眼底彩照图像血管评估方法、装置、计算机设备和介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。This application claims the priority of the Chinese patent application filed with the Chinese Patent Office on November 25, 2020, the application number is 202011344724.X, and the invention title is "Fundus Color Image Blood Vessel Evaluation Method, Apparatus, Computer Equipment and Medium", all of which The content is incorporated in this application by reference.

技术领域Technical field

本申请涉及数字医疗的疾病风险评估技术领域，具体而言，本申请涉及一种眼底彩照图像血管评估方法、装置、计算机设备和介质。This application relates to the technical field of disease risk assessment of digital medical treatment. Specifically, this application relates to a method, device, computer equipment, and medium for evaluating blood vessels in color fundus images.

背景技术Background technique

眼底彩照是一种无创、非接触式成像技术，能直接观察到视网膜血管、视盘、黄斑等眼底组织结构，被广泛用于临床视网膜眼底病变筛查和诊断。其中，视网膜血管是人体唯一能直接观察到的血管***。视网膜血管形态上的变化(如血管密度、管径大小、弯曲度、分支角度等)是糖尿病、高血压、肾病等很多心血管慢病早期诊断及跟踪随访的重要依据。Fundus color photography is a non-invasive, non-contact imaging technology that can directly observe retinal blood vessels, optic discs, macula and other fundus tissue structures, and is widely used in clinical screening and diagnosis of retinal fundus diseases. Among them, the retinal blood vessel is the only blood vessel system that can be directly observed in the human body. The morphological changes of retinal blood vessels (such as blood vessel density, tube diameter, curvature, branch angle, etc.) are important basis for early diagnosis and follow-up of many chronic cardiovascular diseases such as diabetes, hypertension, and nephropathy.

目前关于视网膜血管分割或动静脉分类算法方面的研究很多，但发明人意识到它们均采用Dice、Accuracy等指标对血管分割准确率进行评估，然而这类指标无法体现血管分割结果在拓扑结构上的完整性，评估效果较差。At present, there are many researches on retinal blood vessel segmentation or arteriovenous classification algorithms, but the inventor realizes that they all use indicators such as Dice and Accuracy to evaluate the accuracy of blood vessel segmentation. However, such indicators cannot reflect the topological structure of blood vessel segmentation results. Completeness, poor evaluation effect.

技术问题technical problem

目前关于视网膜血管分割或动静脉分类算法方面的研究很多，但它们均采用Dice、Accuracy等指标对血管分割准确率进行评估，然而这类指标无法体现血管分割结果在拓扑结构上的完整性，评估效果较差。At present, there are many researches on retinal blood vessel segmentation or arteriovenous classification algorithms, but they all use indicators such as Dice and Accuracy to evaluate the accuracy of blood vessel segmentation. However, such indicators cannot reflect the completeness of the topological structure of the blood vessel segmentation results. The effect is poor.

技术解决方案Technical solutions

为克服以上技术问题，特别是现有技术中对血管分割准确率进行评估时，无法体现血管分割结果在拓扑结构上的完整性的缺陷，特提出以下技术方案：In order to overcome the above technical problems, especially when evaluating the accuracy of blood vessel segmentation in the prior art, the defect that the integrity of the blood vessel segmentation result in the topological structure cannot be reflected, the following technical solutions are proposed:

第一方面，本申请提供一种眼底彩照图像血管评估方法，其包括以下步骤：In the first aspect, this application provides a method for evaluating blood vessels in fundus color images, which includes the following steps:

获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；Acquiring a color fundus photograph image, and extracting a retinal blood vessel image from the color fundus photograph image by using a pre-trained segmentation model, wherein the retinal blood vessel image is a topological structure image;

提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，其中，所述误差带为距离预设的标准血管中心线预设距离的图像区域；Extract the blood vessel center line in the retinal blood vessel image, and set the error band of the blood vessel center line based on the preset standard blood vessel center line, wherein the error band is a preset distance from the preset standard blood vessel center line Image area;

根据所述误差带计算所述各血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；Calculating the error of each blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the blood vessel topology error of the retinal blood vessel image;

根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。The segmentation model is evaluated to segment the fundus color photograph image according to the error of the blood vessel topology structure, and an evaluation result is generated.

第二方面，本申请还提供一种眼底彩照图像血管评估装置，其包括：In a second aspect, the present application also provides a blood vessel evaluation device for color fundus images, which includes:

提取模块，用于获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；An extraction module for acquiring a color fundus photo image, and extracting a retinal blood vessel image from the color fundus photo image using a pre-trained segmentation model, wherein the retinal blood vessel image is a topological structure image;

设置模块，用于提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，其中，所述误差带为距离预设的标准血管中心线预设距离的图像区域；The setting module is used to extract the blood vessel center line in the retinal blood vessel image, and set the error band of the blood vessel center line based on a preset standard blood vessel center line, wherein the error band is a standard blood vessel with a preset distance The image area with a preset distance from the center line;

计算模块，用于根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；A calculation module, configured to calculate the error of the blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the blood vessel topology structure error of the retinal blood vessel image;

生成模块，用于根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。The generating module is configured to evaluate the segmentation model to segment the fundus color photograph image according to the error of the blood vessel topology structure, and generate an evaluation result.

第三方面，本申请还提供一种计算机设备，其包括：In the third aspect, this application also provides a computer device, which includes:

一个或多个处理器；One or more processors;

存储器；Memory

一个或多个计算机程序，其中所述一个或多个计算机程序被存储在所述存储器中并被配置为由所述一个或多个处理器执行，所述一个或多个计算机程序配置用于执行第一方面中任意实施例中所述的眼底彩照图像血管评估方法；其中，所述眼底彩照图像血管评估方法包括：获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，其中，所述误差带为距离预设的标准血管中心线预设距离的图像区域；根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。One or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and the one or more computer programs are configured to execute The method for evaluating blood vessels in a color fundus photograph image according to any embodiment of the first aspect; wherein the method for evaluating blood vessels in a color fundus photograph image includes: obtaining a color fundus photograph image, and extracting from the fundus color photograph image by using a pre-trained segmentation model A retinal blood vessel image, wherein the retinal blood vessel image is a topological structure image; the blood vessel center line in the retinal blood vessel image is extracted, and the error band of the blood vessel center line is set based on a preset standard blood vessel center line, wherein, The error band is an image area at a preset distance from a preset standard blood vessel center line; according to the error band, the error of the blood vessel center line compared to the standard blood vessel center line is calculated to obtain the image area of the retinal blood vessel image Vascular topology structure error; according to the vascular topology structure error, the segmentation model is evaluated to segment the fundus color photograph image, and an evaluation result is generated.

第四方面，本申请还提供一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，该计算机程序被处理器执行时实现第一方面中任意实施例中所述的眼底彩照图像血管评估方法；其中，所述眼底彩照图像血管评估方法包括：获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，其中，所述误差带为距离预设的标准血管中心线预设距离的图像区域；根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, the fundus described in any of the embodiments in the first aspect A method for evaluating blood vessels in a color photograph image; wherein the method for evaluating blood vessels in a color fundus photograph image includes: obtaining a color fundus photograph image, and extracting a retinal blood vessel image from the fundus color photograph image using a pre-trained segmentation model, wherein the retinal blood vessel image is Topological structure image; extracting the blood vessel center line in the retinal blood vessel image, setting the error band of the blood vessel center line based on the preset standard blood vessel center line, wherein the error band is the distance from the preset standard blood vessel center Calculate the error of the blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the blood vessel topology error of the retinal blood vessel image; according to the blood vessel topology error The segmentation model is segmented to evaluate the fundus color photograph image, and an evaluation result is generated.

有益效果Beneficial effect

通过利用预先训练好的分割模型从眼底彩照图像中提取出拓扑结构的视网膜血管图像，并提取视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，根据误差带计算所述各血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差，根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。由于视网膜血管图像为拓扑结构图像，因此可直接地从视网膜血管图像中确定血管中心线，利于后续计算血管拓扑结构误差，以及用于分析各血管中心线的关系。此外，本申请的血管拓扑结构误差用于评估血管拓扑结构的完整性和准确性，以对分割模型分割所述眼底彩照图像进行更加全面地评估，评估效果更好。By using a pre-trained segmentation model to extract the topological structure of the retinal blood vessel image from the fundus color photo image, and extract the blood vessel centerline in the retinal blood vessel image, set the centerline of the blood vessel centerline based on the preset standard blood vessel centerline Error band, calculating the error of each blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the blood vessel topology structure error of the retinal blood vessel image, and segmenting the segmentation model according to the blood vessel topology structure error The color fundus photo image is evaluated and the evaluation result is generated. Since the retinal blood vessel image is a topological structure image, the blood vessel centerline can be directly determined from the retinal blood vessel image, which is beneficial to the subsequent calculation of the error of the blood vessel topology structure, and is used to analyze the relationship between the centerline of each blood vessel. In addition, the blood vessel topology structure error of the present application is used to evaluate the completeness and accuracy of the blood vessel topology structure, so as to perform a more comprehensive evaluation of the fundus color photograph image segmented by the segmentation model, and the evaluation effect is better.

附图说明Description of the drawings

图1是本申请中的一个实施例的眼底彩照图像血管评估方法的流程图；FIG. 1 is a flowchart of a method for evaluating blood vessels in a color fundus photograph image according to an embodiment of the present application;

图2为本申请中的标准血管中心线的误差带的位置示意图；Figure 2 is a schematic diagram of the position of the error band of the standard blood vessel centerline in this application;

图3是本申请中的从眼底彩照图像中提取视网膜血管图像的示意图；Fig. 3 is a schematic diagram of extracting retinal blood vessel images from fundus color photograph images in this application;

图4为本申请中的眼底彩照图像、视网膜血管图像、血管中心线的示意图；Fig. 4 is a schematic diagram of the fundus color photograph image, retinal blood vessel image, and blood vessel centerline in this application;

图5为本申请中的血管结构断点修复前后的示意图；Fig. 5 is a schematic diagram before and after repair of a vascular structure breakpoint in this application;

图6为本申请中的血管半径的测量示意图；Fig. 6 is a schematic diagram of measuring the radius of a blood vessel in this application;

图7为本申请中的计算血管曲率时各点选取的示意图；FIG. 7 is a schematic diagram of each point selected when calculating the curvature of a blood vessel in this application;

图8是本申请中的一个实施例的眼底彩照图像血管评估装置的模块图；FIG. 8 is a block diagram of a blood vessel evaluation device for color fundus images according to an embodiment of the present application;

图9为本申请中的一个实施例的计算机设备的结构示意图。FIG. 9 is a schematic structural diagram of a computer device according to an embodiment of the application.

本发明的最佳实施方式The best mode of the present invention

本申请提供了一种眼底彩照图像血管评估方法，可参考图1，图1是一个实施例的眼底彩照图像血管评估方法的流程图，该方法包括以下步骤：This application provides a method for evaluating blood vessels in a color fundus photograph. Refer to FIG. 1. FIG. 1 is a flowchart of a method for evaluating blood vessels in a color fundus photograph according to an embodiment. The method includes the following steps:

S110、获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像。S110. Acquire a fundus color photograph image, and extract a retinal blood vessel image from the fundus color photograph image using a pre-trained segmentation model, where the retinal blood vessel image is a topological structure image.

视网膜血管分割是眼科计算机辅助诊断和大规模疾病筛查***的基础，当眼器官发生视觉疾病的时候，视网膜血管的直径、颜色和弯曲程度等就会出现异常，辅助眼科医生作出诊断。对眼底彩照图像进行血管分割常用的方法有：基于血管跟踪的方法、基于匹配滤波的方法、基于形态学处理的方法、基于形变模型的方法和基于机器学习的方法等。本申请对眼底彩照图像运用深度学习方法，首先获取待分割的眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，并将视网膜血管图像转换成拓扑结构的图像，从而将眼底彩照图像中的血管结构单独分割出来，便于观察研究。其中，拓扑结构图像可把实体抽象成与其大小、形状无关的“点”，而把连接实体的线路抽象成“线”，进而以图的形式来表示这些点与线之间关系的方法，从而可直观地研究这些点、线之间的相连关系。Retinal blood vessel segmentation is the basis of ophthalmology computer-aided diagnosis and large-scale disease screening system. When visual diseases occur in the eye organs, the diameter, color and degree of curvature of the retinal blood vessels will be abnormal, which assists the ophthalmologist in making a diagnosis. Commonly used methods for blood vessel segmentation on color fundus images include: blood vessel tracking-based methods, matched filtering-based methods, morphological processing-based methods, deformation model-based methods, and machine learning-based methods. This application applies a deep learning method to the fundus color photograph image, firstly obtains the fundus color photograph image to be segmented, uses a pre-trained segmentation model to extract the retinal blood vessel image from the fundus color photograph image, and converts the retinal blood vessel image into a topological structure image , So that the blood vessel structure in the fundus color photo image is segmented separately, which is convenient for observation and research. Among them, the topological structure image can abstract entities into "points" that have nothing to do with their size and shape, and abstract the lines connecting entities into "lines", and then express the relationship between these points and lines in the form of graphs. The connection relationship between these points and lines can be studied intuitively.

此外，本申请在生成训练好的分割模型时，首先获取多张眼底彩照图像，对所述多张眼底彩照图像进行标注，以得到每张所述眼底彩照图像人工分割血管后的准确标签；基于所述每张眼底彩照图像和所述准确标签，生成样本集，将所述样本集作为训练集和测试集，将训练集输入预设的深度神经网络模型对其进行训练，再将测试集输入训练完成的深度神经网络模型进行测试，获取所述深度神经网络模型输出的测试结果，将所述测试结果与预设的比对标签进行匹配，当所述测试结果与所述比对标签匹配时，将所述深度神经网络模型作为训练好的分割模型。In addition, when generating the trained segmentation model, this application first obtains multiple fundus color photograph images, and annotates the multiple fundus color photograph images, so as to obtain accurate labels for each of the fundus color photograph images after artificially segmented blood vessels; Generate a sample set for each of the fundus color photograph image and the accurate label, use the sample set as a training set and a test set, input the training set into a preset deep neural network model for training, and then input the test set The trained deep neural network model is tested, the test result output by the deep neural network model is obtained, the test result is matched with a preset comparison label, and when the test result matches the comparison label , And use the deep neural network model as a trained segmentation model.

S120、提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，其中，所述误差带为距离预设的标准血管中心线预设距离的图像区域；由于尖端毛细血管边缘不清晰且比较细小，在整个血管结构中的占比比较小，常规采用Dice系数和Accuracy作为目标函数指导神经网络模型训练的方式仅对全局血管分割误差进行计算，忽视了毛细血管在血管拓扑结构中的重要性。因此本申请为了使得血管分割保留更完整的拓扑结构，分割模型的损失函数可采用全局血管分割误差和血管拓扑结构误差的形式。S120. Extract the blood vessel center line in the retinal blood vessel image, and set an error band of the blood vessel center line based on a preset standard blood vessel center line, where the error band is a preset distance from the standard blood vessel center line. Set the distance of the image area; because the edge of the tip capillary is not clear and relatively small, it accounts for a relatively small proportion in the entire vascular structure. Conventionally, the Dice coefficient and Accuracy are used as the objective function to guide the neural network model training. Only the global blood vessel segmentation error The calculation ignores the importance of capillaries in the vascular topology. Therefore, in this application, in order to make the blood vessel segmentation retain a more complete topology structure, the loss function of the segmentation model can adopt the form of global blood vessel segmentation error and blood vessel topology structure error.

具体的，结合图2所示，本申请获取分割后的视网膜血管图像后，根据分割后的视网膜血管图像确定血管的起始血管段的起始血管层和终止血管段的终止血管层，并将所述起始血管段的起始血管层的中心作为血管中心线的起点，将所述终止血管段的终止血管层的中心作为血管中心线的终点，根据所述起点和所述终点在所述分割后的视网膜血管图像中提取血管中心线，并以对应专家标记的标准血管中心线为基准设置所述血管中心线的矩形误差带，具体地，将距离预设的标准血管中心线预设距离的图像区域设置为所述血管中心线的误差带，例如，将包裹标准血管中心线矩形或圆形区域设置为误差带。Specifically, in conjunction with FIG. 2, after the application obtains the segmented retinal blood vessel image, according to the segmented retinal blood vessel image, the initial blood vessel layer of the initial blood vessel segment and the terminating blood vessel layer of the terminating blood vessel segment are determined, and the The center of the starting blood vessel layer of the starting blood vessel segment is taken as the starting point of the blood vessel centerline, and the center of the ending blood vessel layer of the ending blood vessel segment is taken as the end point of the blood vessel center line. The blood vessel center line is extracted from the segmented retinal blood vessel image, and the rectangular error band of the blood vessel center line is set based on the standard blood vessel center line marked by the expert. Specifically, the distance from the preset standard blood vessel center line is preset The image area of is set as the error band of the center line of the blood vessel, for example, a rectangular or circular area that wraps the center line of the standard blood vessel is set as the error band.

S130、根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；S130: Calculate the error of the blood vessel center line compared with the standard blood vessel center line according to the error band, to obtain the blood vessel topology structure error of the retinal blood vessel image;

进一步地，本步骤根据矩形误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像在拓扑结构的血管拓扑结构误差；例如，计算任一根血管中心线与该血管中心线对应的标准血管中心线的误差时，可将距离标准血管中心线预设距离或预设角度的区域作为误差带，若所述血管中心线全部或大部分落入误差带内，则表示该血管中心线为真实的血管结构。需要说明的是，误差带的设置区域越小，则计算得到的误差的精准度越高。其中，所述标准血管中心线为真实的血管结构，所述血管拓扑结构误差包括完整性误差及准确性误差，所述完整性误差用于评估眼底彩照图像中真实的血管结构未被分割模型提取的占比，所述准确性误差用于评估分割模型提取视网膜血管图像的血管结构为非真实的血管结构的占比。Further, this step calculates the error of the centerline of the blood vessel compared with the centerline of the standard blood vessel according to the rectangular error band, to obtain the topological structure error of the blood vessel in the topological structure of the retinal blood vessel image; for example, calculate the center of any blood vessel When there is an error between the line and the center line of the standard blood vessel corresponding to the center line of the blood vessel, the area with a preset distance or a preset angle from the center line of the standard blood vessel can be used as the error band. Inside, it means that the centerline of the blood vessel is a real blood vessel structure. It should be noted that the smaller the setting area of the error band, the higher the accuracy of the calculated error. Wherein, the standard blood vessel centerline is a real blood vessel structure, the blood vessel topology structure error includes an integrity error and an accuracy error, and the integrity error is used to evaluate that the true blood vessel structure in the fundus color photograph image is not extracted by the segmentation model The accuracy error is used to evaluate the proportion of unreal blood vessel structures extracted from the retinal blood vessel image by the segmentation model.

所述标准血管中心线还可通过大数据分析确定。具体的，可预先收集大量的健康人眼的视网膜血管图像，从视网膜血管图像中分别提取出健康的血管中心线，得到多根血管中心线，并分析各根血管中心线的特征点，从所述多根血管中心线中筛选出具有特征点最多的一根血管中心线作为标准血管中心线。The standard blood vessel centerline can also be determined through big data analysis. Specifically, a large number of retinal blood vessel images of healthy human eyes can be collected in advance, and healthy blood vessel center lines can be extracted from the retinal blood vessel images to obtain multiple blood vessel center lines, and analyze the characteristic points of each blood vessel center line. Among the multiple blood vessel center lines, a blood vessel center line with the most characteristic points is selected as the standard blood vessel center line.

其中，所述血管中心线可以由一根或多根支线组成，所述血管拓扑结构误差包括完整性误差及准确性误差，所述完整性误差可用于评估眼底彩照图像中真实的血管结构未被分割模型提取的占比，用于评估血管拓扑结构的完整性，所述完整性误差越大，血管拓扑结构的完整性越低，所述血管拓扑结构的完整性可采用以下公式计算：Wherein, the blood vessel centerline may be composed of one or more branch lines, the blood vessel topology error includes integrity error and accuracy error, and the integrity error can be used to evaluate the true blood vessel structure in the fundus color photograph image. The proportion extracted by the segmentation model is used to evaluate the integrity of the blood vessel topology. The greater the integrity error, the lower the integrity of the blood vessel topology. The integrity of the blood vessel topology can be calculated using the following formula:

Topology_Completeness＝TP/(TP+FN)；Topology_Completeness=TP/(TP+FN);

其中，所述TP为被提取出的血管结构为真实的血管结构的数量，所述FN为真实的血管结构未被分割模型提取的数量。例如，假设眼底彩照图像中真实的血管结构有100根，被分割模型准确提取的真实的血管结构有80根，而其余20根都未被提取出来，则所述全局血管分割误差的比例为20％，完整性为80％。Wherein, the TP is the number of extracted blood vessel structures that are real blood vessel structures, and the FN is the number of real blood vessel structures that are not extracted by the segmentation model. For example, assuming that there are 100 real blood vessel structures in the fundus color photo image, 80 real blood vessel structures accurately extracted by the segmentation model, and the remaining 20 are not extracted, the ratio of the global blood vessel segmentation error is 20 %, the integrity is 80%.

所述准确性误差用于评估分割模型提取视网膜血管图像的血管结构为非真实的血管结构的占比，用于评估血管拓扑结构的准确性，所述准确性误差越大，所述血管拓扑结构的准确性越低。所述血管拓扑结构的准确性可采用以下公式计算：The accuracy error is used to evaluate the proportion of unreal blood vessel structures extracted from the retinal blood vessel image by the segmentation model, and is used to evaluate the accuracy of the blood vessel topology. The greater the accuracy error, the greater the blood vessel topology The lower the accuracy. The accuracy of the blood vessel topology can be calculated using the following formula:

Topology_Correctness＝TP/(TP+FP)；Topology_Correctness=TP/(TP+FP);

其中，所述TP为被提取出的血管结构为真实的血管结构的数量，所述FP为被提取出的血管结构为非真实的血管结构的数量。例如，假设从眼底彩照图像中共提取出100根血管结构，生成视网膜血管图像，被分割模型准确提取的真实的血管结构有90根，而其余10根都是非真实的血管结构，为眼底彩照图像其他的结构，则准确性误差的比例为10％，所述血管拓扑结构的准确性为90％。Wherein, the TP is the number of the extracted blood vessel structures that are real blood vessel structures, and the FP is the number of the extracted blood vessel structures that are not real blood vessel structures. For example, suppose that a total of 100 blood vessel structures are extracted from the fundus color photo image to generate a retinal blood vessel image. There are 90 real blood vessel structures accurately extracted by the segmentation model, and the remaining 10 are all unreal blood vessel structures, which are the fundus color photo images. The accuracy error ratio is 10%, and the accuracy of the blood vessel topology is 90%.

S140、根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。S140: Evaluate segmentation of the fundus color photograph image by the segmentation model according to the error of the blood vessel topology structure, and generate an evaluation result.

本步骤血管拓扑结构误差结合完整性误差和准确性误差进行综合评价，生成视网膜血管图像的评估结果。其中，所述血管拓扑结构误差可采用如下公式计算：In this step, the vascular topology structure error is combined with the integrity error and the accuracy error for comprehensive evaluation, and the evaluation result of the retinal blood vessel image is generated. Wherein, the error of the blood vessel topology structure can be calculated using the following formula:

Topology _Loss＝1-TP/(TP+FP+FN)； Topology _Loss = 1-TP/(TP+FP+FN);

其中，所述TP为被提取出的血管结构为真实的血管结构的数量，所述FN为真实的血管结构未被分割模型提取的数量，所述FP为被提取出的血管结构为非真实的血管结构的数量。当然，如图3所示，本申请的评估结果还可结合血管拓扑结构误差及全局血管分割误差，从而综合评估分割模型分割后的血管结构，用于后续精准指导血管分割模型的训练。所述全局血管分割误差可采用常规的Dice_loss计算方式，在此不再赘述。Wherein, the TP is the number of vascular structures that are extracted as real vascular structures, the FN is the number of real vascular structures that are not extracted by the segmentation model, and the FP is the number of extracted vascular structures that are not real. The number of vascular structures. Of course, as shown in FIG. 3, the evaluation result of the present application can also be combined with the vessel topology structure error and the global vessel segmentation error, thereby comprehensively evaluating the vessel structure after the segmentation model is segmented, and is used to accurately guide the subsequent training of the vessel segmentation model. The global blood vessel segmentation error can use a conventional Dice_loss calculation method, which will not be repeated here.

如图4所示，图4示出了眼底彩照图像各阶段处理后的结果，其中，图4中的(a)为眼底彩照图像，(b)为视网膜血管图像，(c)为血管中心线图像，从而经过本申请上述方式处理，得到清晰的血管中心线图像，以辅助眼科医生作出诊断。其中，所述评估结果可根据所述血管拓扑结构误差以图表的形式生成，以评估分割模型分割所述眼底彩照图像的分割效果。As shown in Figure 4, Figure 4 shows the results of the various stages of the fundus color photograph image processing. In Figure 4, (a) is the fundus color photograph image, (b) is the retinal blood vessel image, and (c) is the blood vessel centerline. The image is processed in the manner described above to obtain a clear image of the centerline of the blood vessel to assist the ophthalmologist in making a diagnosis. Wherein, the evaluation result may be generated in the form of a graph according to the error of the blood vessel topology structure to evaluate the segmentation effect of the segmentation model for segmenting the color fundus image.

本申请所提供的一种眼底彩照图像血管评估方法，通过利用预先训练好的分割模型从眼底彩照图像中提取出拓扑结构的视网膜血管图像，并提取视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，根据误差带计算所述各血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差，根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。由于视网膜血管图像为拓扑结构图像，因此可直接地从视网膜血管图像中确定血管中心线，利于后续计算血管拓扑结构误差，以及用于分析各血管中心线的关系。此外，本申请的血管拓扑结构误差用于评估血管拓扑结构的完整性和准确性，以对分割模型分割所述眼底彩照图像进行更加全面地评估，评估效果更好。在其中一个实施例中，提取所述视网膜血管图像中的血管中心线，可包括：The method for evaluating blood vessels in a color fundus photograph image provided by this application extracts a topological structure of the retinal blood vessel image from the fundus color photograph image by using a pre-trained segmentation model, and extracts the blood vessel center line in the retinal blood vessel image to preset Set the error band of the blood vessel center line based on the standard blood vessel center line of, and calculate the error of each blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the blood vessel topology error of the retinal blood vessel image Evaluate segmentation of the fundus color photograph image by the segmentation model according to the error of the blood vessel topology structure, and generate an evaluation result. Since the retinal blood vessel image is a topological structure image, the blood vessel centerline can be directly determined from the retinal blood vessel image, which is beneficial to the subsequent calculation of the error of the blood vessel topology structure, and is used to analyze the relationship between the centerline of each blood vessel. In addition, the blood vessel topology structure error of the present application is used to evaluate the completeness and accuracy of the blood vessel topology structure, so as to perform a more comprehensive evaluation of the fundus color photograph image segmented by the segmentation model, and the evaluation effect is better. In one of the embodiments, extracting the blood vessel centerline in the retinal blood vessel image may include:

对所述视网膜血管图像二值化处理，得到二值图像；Binarize the retinal blood vessel image to obtain a binary image;

利用形态学骨架提取算法对所述二值图像的血管结构进行骨架提取，得到血管中心线，所述血管中心线为血管结构各处内切圆的圆心连接线。A morphological skeleton extraction algorithm is used to perform skeleton extraction on the blood vessel structure of the binary image to obtain the blood vessel center line, which is the center connection line of the inscribed circles of the blood vessel structure.

二值化处理是让视网膜血管图像的像素点矩阵中的每个像素点的灰度值为0(黑色)或者255(白色)，也就是让整个视网膜血管图像呈现只有黑和白的效果。在灰度化的图像中灰度值的范围为0～255，在二值化处理后的视网膜血管图像中的灰度值范围是0或者255，更加有利于对视网膜血管图像的血管结构进行判别。The binarization process is to make the gray value of each pixel in the pixel matrix of the retinal blood vessel image 0 (black) or 255 (white), which means that the entire retinal blood vessel image presents only black and white effects. The gray value range in the grayscale image is 0～255, and the gray value range in the retinal blood vessel image after binarization is 0 or 255, which is more conducive to the identification of the vascular structure of the retinal blood vessel image. .

本申请进一步对二值化的血管结构采用形态学骨架提取算法，提取血管中心线，形态学骨架提取算法是提取视网膜血管图像的中心像素轮廓，其以骨架中心为准，对骨架进行细化，形成血管中心线，以精准地从所述视网膜血管图像中提取出血管中心线。This application further uses a morphological skeleton extraction algorithm for the binarized blood vessel structure to extract the center line of the blood vessel. The morphological skeleton extraction algorithm extracts the central pixel contour of the retinal blood vessel image, which is based on the skeleton center to refine the skeleton. The blood vessel centerline is formed to accurately extract the blood vessel centerline from the retinal blood vessel image.

此外，本申请还提出了一种基于距离和偏差角度的血管中心线断点修正方法，并进一步实现了分形维数、管径大小、弯曲度等量化指标的自动测量，以综合血管的强度信息和形态信息，对血管的特征参数进行量化，能够更为准确、全面的量化视网膜血管图像的血管特征，以用于识别血管疾病。In addition, this application also proposes a method for correcting the broken point of the blood vessel center line based on distance and deviation angle, and further realizes the automatic measurement of quantitative indicators such as fractal dimension, tube diameter, and curvature to integrate the strength information of the blood vessel. Quantification of the characteristic parameters of the blood vessel with morphological information can more accurately and comprehensively quantify the blood vessel characteristics of the retinal blood vessel image, which can be used to identify vascular diseases.

在其中一个实施例中，提取所述视网膜血管图像中的血管中心线之后，还可包括：In one of the embodiments, after extracting the blood vessel center line in the retinal blood vessel image, the method may further include:

计算每两根血管中心线之间的最短距离及夹角；Calculate the shortest distance and included angle between the center lines of every two blood vessels;

若所述最短距离小于阈值且夹度小于预设角度，获取所述两根血管中心线之间距离最短的断点，将所述断点相连。If the shortest distance is less than the threshold and the clamping degree is less than the preset angle, the break point with the shortest distance between the center lines of the two blood vessels is obtained, and the break points are connected.

在本实施例中，如图5所示，目前基于深度学习的血管提取模型多为像素级分割网络，没有考虑血管连续性，因此提取的血管结构可能存在断点问题，因此，本申请对于距离比较近的血管段，通过距离和偏差角度判定进行断点重连，生成连续的血管中心线。具体的，本申请两两选取视网膜血管图像的血管中心线，计算每两根血管中心线之间的最短距离及夹角，若最短距离小于阈值T＝5且每两根血管中心线之间的夹角小于30度，则将所述两根血管中心线之间距离最短的断点相连，实现将属于同一根血管的断点进行连接，生成完整的血管中心线，并确保相连的两根血管中心线同属一根血管。In this embodiment, as shown in Figure 5, the current deep learning-based blood vessel extraction model is mostly a pixel-level segmentation network, and the continuity of the blood vessel is not considered. Therefore, the extracted blood vessel structure may have a breakpoint problem. For the relatively close blood vessel segment, the breakpoint is reconnected based on the distance and the deviation angle to generate a continuous blood vessel centerline. Specifically, this application selects the blood vessel center lines of the retinal blood vessel image two by one, and calculates the shortest distance and the included angle between the center lines of each two blood vessels. If the shortest distance is less than the threshold T=5 and the center line of each two blood vessels If the included angle is less than 30 degrees, the break points with the shortest distance between the center lines of the two blood vessels are connected to realize the connection of the break points belonging to the same blood vessel to generate a complete center line of the blood vessel and ensure the two connected blood vessels The center line belongs to the same blood vessel.

在其中一个实施例中，根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差的步骤，包括：In one of the embodiments, the step of calculating the error of the blood vessel center line compared to the standard blood vessel center line according to the error band includes:

获取眼底彩照图像中血管中心线的总数量；Obtain the total number of blood vessel centerlines in the fundus color photo image;

计算所述血管中心线中未落入所述误差带内的第一数量；Calculating the first number of the centerline of the blood vessel that does not fall within the error band;

根据所述第一数量及总数量计算所述误差。The error is calculated according to the first quantity and the total quantity.

本实施例先计算眼底彩照图像中所有血管中心线的总数量，并获取完全未落入所述误差带内的所述血管中心线中，计算所述血管中心线中完全未落入所述误差带内的数量，得到第一数量，根据第一数量及总数量计算血管中心线相较于所述标准血管中心线的误差，操作方式简单，且精度高。This embodiment first calculates the total number of the center lines of all blood vessels in the fundus color photograph image, and obtains the blood vessel center lines that do not fall within the error band at all, and calculates that the blood vessel center lines do not fall into the error at all The number in the band obtains the first number, and the error of the blood vessel center line compared with the standard blood vessel center line is calculated according to the first number and the total number. The operation mode is simple and the accuracy is high.

在其中一个实施例中，提取所述视网膜血管图像中的血管中心线之后，还可包括：In one of the embodiments, after extracting the blood vessel centerline in the retinal blood vessel image, the method may further include:

计算所述视网膜血管图像的面积，利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数；Calculating the area of the retinal blood vessel image, traversing the retinal blood vessel image using a box of a preset size, and calculating the number of box traversals according to the area of the box and the area of the retinal blood vessel image;

计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数；Calculating the cumulative number of times that the box covers the retinal blood vessel image during traversal and includes the centerline of the blood vessel;

更换盒子的大小，继续执行利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数，计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数的步骤，得到不同尺寸的盒子的遍历次数及累计次数；Change the size of the box, continue to traverse the retinal blood vessel image with the box of the preset size, and calculate the number of box traversals based on the area of the box and the area of the retinal blood vessel image, and calculate the area covered by the retinal blood vessel image when the box is traversed Including the step of accumulating the number of times of the blood vessel centerline to obtain the number of traversals and accumulating times of boxes of different sizes;

根据所述不同尺寸的盒子的遍历次数及累计次数计算血管结构的分形维数。The fractal dimension of the blood vessel structure is calculated according to the traversal times and the accumulated times of the boxes of different sizes.

具体的，假设所述视网膜血管图像的面积为S，可依次使用2 ⁿ大小的盒子遍历视网膜血管图像，若盒子覆盖的视网膜血管图像的区域中包含血管中心线，则计数一次，计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数。更换盒子的大小，采用不同尺寸的盒子重新遍历视网膜血管图像，得到不同尺寸的盒子在遍历时覆盖视网膜血管图像的区域中包括所述血管中心线的累计次数，根据所述不同尺寸的盒子的遍历次数及累计次数计算血管结构的分形维数。这种血管结构的分形维数的计算方式操作简单、计算简便；且由于通过不同尺寸的盒子多次遍历视网膜血管图像，使计算得到的分形维数误差较小，精准度更高，所得分形维数能完整、有效反映被血管形态的复杂性和数量。 Specifically, assuming that the area of the retinal blood vessel image is S, a 2 ^n- sized box can be used in turn to traverse the retinal blood vessel image. If the area of the retinal blood vessel image covered by the box contains the center line of the blood vessel, it is counted once, and the box is traversed. The area covering the retinal blood vessel image includes the cumulative number of times of the blood vessel centerline. Change the size of the box, re-traverse the retinal blood vessel image with a box of different size, and obtain a box of different size in the area covering the retinal blood vessel image during the traversal, including the cumulative number of the blood vessel centerline, according to the traversal of the box of different size The number and cumulative number of times calculate the fractal dimension of the vascular structure. This method of calculating the fractal dimension of the blood vessel structure is simple to operate and easy to calculate; and because the retinal blood vessel image is traversed multiple times through boxes of different sizes, the calculated fractal dimension error is smaller, the accuracy is higher, and the fractal dimension is higher. The number can completely and effectively reflect the complexity and quantity of the vascular shape.

其中，分形维数可采用如下计算公式进行计算：Among them, the fractal dimension can be calculated using the following calculation formula:

其中，ε＝2 ⁿ/S，所述S为视网膜血管图像的面积，2 ⁿ为盒子的面积，N(ε)为盒子在遍历时覆盖视网膜血管图像的区域中包括所述血管中心线的累计次数，若盒子覆盖的区域中包含血管中心线，则N(ε)加1。进一步地，采用不同尺寸的盒子进行遍历，得到一系列(logN(ε)，log(1/ε))配对的点，logN(ε)为横轴坐标，log(1/ε)为纵轴坐标，可采用最小二乘拟合直线，所得直线斜率即为血管的分形维数。所述最小二乘法，又称最小平方法。它通过最小化误差的平方和寻找数据的最佳函数匹配。主要作用是从一堆相关数据中求解数据的一般性规律。在图像处理方面多用于各种形状的拟合。最小二乘拟合直线，主要体现为找到一条直线，使得所有已知的点到这条直线的欧式距离的和最小(或者理解为点到直线的误差平方和最小)。 Where ε=2 ⁿ /S, the S is the area of the retinal blood vessel image, 2 ⁿ is the area of the box, and N(ε) is the accumulation of the centerline of the blood vessel in the area covered by the retinal blood vessel image during the traversal of the box The number of times, if the area covered by the box contains the centerline of the blood vessel, then N(ε) plus one. Furthermore, using boxes of different sizes to traverse to obtain a series of (logN(ε), log(1/ε)) paired points, logN(ε) is the horizontal axis coordinate, log(1/ε) is the vertical axis coordinate , The least squares can be used to fit a straight line, and the slope of the obtained straight line is the fractal dimension of the blood vessel. The least square method is also called the least square method. It finds the best function match of the data by minimizing the sum of squares of the error. The main function is to solve the general law of data from a bunch of related data. In image processing, it is mostly used for fitting various shapes. Least squares fitting a straight line is mainly embodied in finding a straight line so that the sum of the Euclidean distances from all known points to this straight line is the smallest (or understood as the smallest sum of squared errors from the point to the straight line).

提取所述血管中心线上的任一点作为圆心，以所述圆心为中心设置一固定的ROI区域，在所述ROI区域内进行血管边界探测；Extracting any point on the center line of the blood vessel as the center of the circle, setting a fixed ROI area with the center of the circle as the center, and performing blood vessel boundary detection in the ROI area;

计算血管边界距离所述圆心的最短距离，得到血管半径。The shortest distance between the blood vessel boundary and the center of the circle is calculated to obtain the blood vessel radius.

在本实施例中，眼底彩照图像分析需要首先获取ROI区域，这样在后续的处理中能有效避免ROI区域外的像素的影响，降低运算的复杂度。本申请可提取血管中心线上的任一点作为圆心，以所述圆心为中心设置一固定的ROI区域，该ROI区域可以是矩形区域，在所述ROI区域内进行血管边界探测；计算提取的血管边界到血管中心线的最短距离作为血管半径。如图6所示，可在ROI区域内对1、2、3的位置进行血管边界探测，分别计算血管中心线到1、2、3血管边界距离，将提取的血管边界到血管中心线的最短距离作为血管半径，从图中可以看出，血管中心线到2的位置最近，因此将血管中心线到2的距离作为血管半径。In this embodiment, the analysis of the fundus color photograph image needs to first obtain the ROI area, so that the influence of pixels outside the ROI area can be effectively avoided in the subsequent processing, and the complexity of the calculation can be reduced. This application can extract any point on the center line of the blood vessel as the center of the circle, and set a fixed ROI area with the center of the circle as the center. The ROI area can be a rectangular area, and the blood vessel boundary detection is performed in the ROI area; calculate the extracted blood vessel The shortest distance from the boundary to the centerline of the blood vessel is taken as the blood vessel radius. As shown in Figure 6, blood vessel boundary detection can be performed on the positions of 1, 2, and 3 in the ROI area, and the distance between the blood vessel center line and the blood vessel boundary of 1, 2, and 3 can be calculated respectively, and the shortest distance between the extracted blood vessel boundary and the blood vessel center line The distance is taken as the radius of the blood vessel. It can be seen from the figure that the position of the blood vessel center line to 2 is the closest, so the distance from the blood vessel center line to 2 is taken as the blood vessel radius.

本申请通过以圆心为中心设置一固定的ROI区域，在ROI区域内进行血管边界探测，以缩小血管边界的探测范围，进而减少血管边界的探测次数，便于快速确定距离圆心最短的血管边界，提高血管半径的计算效率及准确性，以辅助医生根据血管半径进行疾病诊断，提高医疗效率。In this application, a fixed ROI area is set with the center of the circle as the center, and blood vessel boundary detection is performed in the ROI area to reduce the detection range of the blood vessel boundary, thereby reducing the detection times of the blood vessel boundary, and facilitating the rapid determination of the blood vessel boundary with the shortest distance from the center of the circle. The calculation efficiency and accuracy of the blood vessel radius can assist doctors in diagnosing diseases based on the blood vessel radius and improve medical efficiency.

测量所述血管中心线的弧长，计算所述血管中心线各点曲率的平方和；Measuring the arc length of the center line of the blood vessel, and calculating the sum of the squares of the curvature of each point of the center line of the blood vessel;

根据所述弧长及平方和计算血管结构的血管平均曲率。The average curvature of the blood vessel of the blood vessel structure is calculated based on the arc length and the square sum.

进一步地，在计算所述血管中心线各点曲率的平方和之前，还可包括：Further, before calculating the sum of the squares of the curvature of each point of the blood vessel centerline, it may further include:

从血管中心线上任意选取一点作为目标点，依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率；Any point selected from the center line of the blood vessel as the target point, the cosine of the included angle between the remaining points on the center line of the blood vessel and the target point is sequentially calculated, and the value with the largest included angle cosine is selected as the curvature of the target point;

从血管中心线上的其余点中选取一点作为目标点，重新执行依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率的步骤，直至得到所述血管中心线各点的曲率。Select a point from the remaining points on the blood vessel center line as the target point, re-calculate the cosine of the included angle between the remaining points on the blood vessel center line and the target point, and select the largest value of the included angle cosine as the curvature of the target point Steps until the curvature of each point of the center line of the blood vessel is obtained.

具体的，本申请从血管中心线上任意选取一点作为目标点，依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率；再从血管中心线上的其余点中选取一点作为目标点，重新执行依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率的步骤，直至完成血管中心线各点的曲率计算，得到所述血管中心线各点的曲率，并计算各点曲率的平方和，最后根据弧长及平方和计算血管结构的血管平均曲率。其中，计算血管中心线所有点的曲率时，可从血管中心线的起点至终点依次计算各点的曲率。Specifically, this application randomly selects a point from the blood vessel center line as the target point, sequentially calculates the angle cosine between the remaining points on the blood vessel center line and the target point, and selects the largest value of the angle cosine as the curvature of the target point; Then select a point from the remaining points on the center line of the blood vessel as the target point, re-calculate the cosine of the included angle between the remaining points on the center line of the blood vessel and the target point, and select the largest value of the included angle cosine as the curvature of the target point Until the curvature calculation of each point of the blood vessel centerline is completed, the curvature of each point of the blood vessel centerline is obtained, and the square sum of the curvature of each point is calculated, and finally the average curvature of the blood vessel of the blood vessel structure is calculated according to the arc length and the square sum. Among them, when calculating the curvature of all points of the blood vessel centerline, the curvature of each point can be calculated sequentially from the start point to the end point of the blood vessel centerline.

本申请通过遍历计算血管中心线上所有点的曲率，得到血管结构的血管平均曲率，以综合评估血管结构的弯曲度，使计算得到的血管结构弯曲度的精准度更高。In this application, the curvature of all points on the center line of the blood vessel is calculated by traversal to obtain the average curvature of the blood vessel structure to comprehensively evaluate the curvature of the blood vessel structure, so that the calculated curvature of the blood vessel structure has a higher accuracy.

可选地，血管平均曲率τ的计算公式为：Optionally, the calculation formula of the average curvature τ of the blood vessel is:

τ＝tsc(C)/s(C)；τ=tsc(C)/s(C);

其中，s(C)为血管中心线的弧长，tsc(C)为血管中心线上各点曲率的平方和。具体的，假设该血管中心线由n个点组成，则弧长的近似计算为：Among them, s(C) is the arc length of the blood vessel centerline, and tsc(C) is the sum of the squares of the curvature of each point on the blood vessel centerline. Specifically, assuming that the centerline of the blood vessel consists of n points, the approximate calculation of the arc length is:

其中，x _i、y _i分别为血管中心线某一点的横坐标和纵坐标；x _i+1、y _i+1为血管中心线另一点的横坐标和纵坐标。 Among them, x _i and y _i are the abscissa and ordinate of a certain point on the blood vessel center line respectively; x _i+1 and y _i+1 are the abscissa and ordinate of another point on the blood vessel center line.

如图7所示，计算各点曲率的平方和时，可在血管中心线上选取任一点K，采用K余弦曲率方式近似计算各点的曲率，依次计算曲线上前后各点与该K点形成的夹角余弦，取最大的夹角余弦值作为该点的曲率值。As shown in Figure 7, when calculating the sum of squares of the curvature of each point, you can select any point K on the center line of the blood vessel, use the K cosine curvature method to approximate the curvature of each point, and calculate the formation of each point on the curve and the K point in turn The angle cosine of, and the largest angle cosine value is taken as the curvature value of the point.

本申请实现了血管分形维数、管径大小、弯曲度等血管指标的自动量化计算，弥补了现阶段临床上使用的半自动测量方法鲁棒性较差、无法复现的不足，有助于提高临床诊断效率，为实现糖尿病、高血压、肾病等相关心血管慢病的早期诊断及跟踪随访的提供了重要的量化依据。This application realizes the automatic quantitative calculation of blood vessel indicators such as the fractal dimension of blood vessels, the size of the tube diameter, and the degree of curvature. The efficiency of clinical diagnosis provides an important quantitative basis for the early diagnosis and follow-up of chronic cardiovascular diseases such as diabetes, hypertension, and kidney disease.

基于与上述眼底彩照图像血管评估方法相同的申请构思，本申请实施例还提供了一种眼底彩照图像血管评估装置，如图8所示，包括：Based on the same application concept as the above-mentioned fundus color image blood vessel evaluation method, an embodiment of the present application also provides a fundus color image blood vessel evaluation device, as shown in FIG. 8, including:

提取模块310，用于获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；The extraction module 310 is configured to obtain a color fundus photo image, and extract a retinal blood vessel image from the color fundus photo image using a pre-trained segmentation model, where the retinal blood vessel image is a topological structure image;

设置模块320，用于提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带；The setting module 320 is configured to extract the blood vessel center line in the retinal blood vessel image, and set the error band of the blood vessel center line based on a preset standard blood vessel center line;

计算模块330，用于根据所述误差带计算所述各血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的全局血管分割误差及血管拓扑结构误差；生成模块340，用于根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。The calculation module 330 is configured to calculate the error of each blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the global blood vessel segmentation error and the blood vessel topology error of the retinal blood vessel image; generating module 340 , Used for evaluating the segmentation model to segment the fundus color photograph image according to the vascular topology structure error, and generating an evaluation result.

进一步地，所述提取模块310进一步被配置为：Further, the extraction module 310 is further configured to:

利用形态学骨架提取算法对所述二值图像的血管结构进行骨架提取，得到各血管中心线，所述血管中心线为血管结构各处内切圆的圆心连接线。A morphological skeleton extraction algorithm is used to perform skeleton extraction on the blood vessel structure of the binary image to obtain the center line of each blood vessel, and the blood vessel center line is the center connection line of the inscribed circle of the blood vessel structure.

在其中一个实施例中，本申请的眼底彩照图像血管评估装置还可包括：In one of the embodiments, the fundus color photograph image blood vessel assessment device of the present application may further include:

第一计算模块，用于计算每两根血管中心线之间的最短距离及夹角；The first calculation module is used to calculate the shortest distance and the included angle between the center lines of every two blood vessels;

连接模块，用于若所述最短距离小于阈值且夹度小于预设角度，获取所述两根血管中心线之间距离最短的断点，将所述断点相连。The connection module is configured to obtain a break point with the shortest distance between the center lines of the two blood vessels if the shortest distance is less than a threshold and the clamping angle is less than a preset angle, and connect the break points.

第一遍历模块，用于计算所述视网膜血管图像的面积，利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数；The first traversal module is configured to calculate the area of the retinal blood vessel image, traverse the retinal blood vessel image with a box of a preset size, and calculate the number of box traversals according to the area of the box and the area of the retinal blood vessel image;

累计模块，用于计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数；The accumulation module is used to calculate the cumulative number of times that the centerline of the blood vessel is included in the area covered by the retinal blood vessel image when the box is traversed;

第二遍历模块，用于更换盒子的大小，继续执行利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数，计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数的步骤，得到不同尺寸的盒子的遍历次数及累计次数；The second traversal module is used to change the size of the box, continue to traverse the retinal blood vessel image with the preset size box, and calculate the number of box traversals according to the area of the box and the area of the retinal blood vessel image, and calculate the coverage of the box during the traversal The step of including the cumulative number of times of the centerline of the blood vessel in the region of the retinal blood vessel image to obtain the number of traversal times and the cumulative number of times of boxes of different sizes;

第二计算模块，用于根据所述不同尺寸的盒子的遍历次数及累计次数计算血管结构的分形维数。The second calculation module is used to calculate the fractal dimension of the blood vessel structure according to the traversal times and the cumulative times of the boxes of different sizes.

边界探测模块，用于提取所述血管中心线上的任一点作为圆心，以所述圆心为中心设置一固定的ROI区域，在所述ROI区域内进行血管边界探测；The boundary detection module is configured to extract any point on the center line of the blood vessel as the center of the circle, set a fixed ROI area with the center of the circle as the center, and perform blood vessel boundary detection in the ROI area;

第三计算模块，用于计算血管边界距离所述圆心的最短距离，得到血管半径。The third calculation module is used to calculate the shortest distance between the blood vessel boundary and the center of the circle to obtain the blood vessel radius.

测量模块，用于测量所述血管中心线的弧长，计算所述血管中心线各点曲率的平方和；The measurement module is used to measure the arc length of the centerline of the blood vessel, and calculate the sum of the squares of the curvature of each point on the centerline of the blood vessel;

第四计算模块，用于根据所述弧长及平方和计算血管结构的血管平均曲率。The fourth calculation module is used to calculate the average curvature of the blood vessel of the blood vessel structure according to the arc length and the square sum.

在其中一个实施例中，所述测量模块还被配置为：In one of the embodiments, the measurement module is further configured to:

在其中一个实施例中，所述设置模块320还被配置为：In one of the embodiments, the setting module 320 is further configured to:

将距离预设的标准血管中心线预设距离的图像区域设置为所述血管中心线的误差带。The image area at the preset distance from the preset standard blood vessel center line is set as the error band of the blood vessel center line.

在其中一个实施例中，所述计算模块330还被配置为：In one of the embodiments, the calculation module 330 is further configured to:

请参考图9，图9为一个实施例中计算机设备的内部结构示意图。该计算机设备包括通过***总线连接的处理器410、存储介质420、存储器430和网络接口440。其中，该计算机设备的存储介质420存储有操作***、数据库和计算机可读指令，数据库中可存储有控件信息序列，该计算机可读指令被处理器410执行时，可使得处理器410实现一种眼底彩照图像血管评估方法，处理器410能实现上述所示实施例中的一种眼底彩照图像血管评估装置中的功能。其中，所述眼底彩照图像血管评估方法包括：获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带；根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。在此不再赘述。Please refer to FIG. 9, which is a schematic diagram of the internal structure of a computer device in an embodiment. The computer device includes a processor 410, a storage medium 420, a memory 430, and a network interface 440 connected through a system bus. Wherein, the storage medium 420 of the computer device stores an operating system, a database, and computer-readable instructions. The database may store control information sequences. When the computer-readable instructions are executed by the processor 410, the processor 410 can implement a In the method for evaluating blood vessels in a color fundus photograph, the processor 410 can implement the functions of a blood vessel evaluation device in a color fundus photograph in the above-mentioned embodiment. Wherein, the method for evaluating blood vessels in a color fundus photograph image includes: acquiring a color fundus photograph image, and extracting a retinal blood vessel image from the fundus color photograph image using a pre-trained segmentation model, wherein the retinal blood vessel image is a topological structure image; For the blood vessel center line in the retinal blood vessel image, an error band of the blood vessel center line is set based on a preset standard blood vessel center line; according to the error band, the blood vessel center line is calculated as compared with the standard blood vessel center line The error of the retinal blood vessel image is obtained, and the blood vessel topology structure error of the retinal blood vessel image is obtained; the segmentation model is evaluated according to the error of the blood vessel topology structure to segment the fundus color photograph image, and the evaluation result is generated. I won't repeat them here.

该计算机设备的处理器410用于提供计算和控制能力，支撑整个计算机设备的运行。该计算机设备的存储器430中可存储有计算机可读指令，该计算机可读指令被处理器410执行时，可使得处理器410执行一种眼底彩照图像血管评估方法。该计算机设备的网络接口440用于与终端连接通信。本领域技术人员可以理解，图9中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。The processor 410 of the computer device is used to provide calculation and control capabilities, and supports the operation of the entire computer device. The memory 430 of the computer device may store computer readable instructions, and when the computer readable instructions are executed by the processor 410, the processor 410 can make the processor 410 execute a fundus color image blood vessel assessment method. The network interface 440 of the computer device is used to connect and communicate with the terminal. Those skilled in the art can understand that the structure shown in FIG. 9 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.

若所述计算机设备为终端，为了便于说明，仅示出了与本申请实施例相关的部分，具体技术细节未揭示的，请参照本申请实施例方法部分。该终端可以为包括手机、平板电脑、PDA(Personal Digital Assistant，个人数字助理)、POS(Point of Sales，销售终端)、车载电脑等任意终端设备。If the computer device is a terminal, for ease of description, only parts related to the embodiment of the present application are shown. For specific technical details that are not disclosed, please refer to the method part of the embodiment of the present application. The terminal can be any terminal device including a mobile phone, a tablet computer, a PDA (Personal Digital Assistant), a POS (Point of Sales, sales terminal), a vehicle-mounted computer, etc.

在一个实施例中，本申请还提出了一种存储有计算机可读指令的存储介质，所述存储介质为易失性存储介质或非易失性存储介质，该计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行一种眼底彩照图像血管评估方法；其中，所述眼底彩照图像血管评估方法包括：获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带；根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。在此不再赘述。In one embodiment, the present application also proposes a storage medium storing computer-readable instructions. The storage medium is a volatile storage medium or a non-volatile storage medium. The computer-readable instructions are stored by one or more When the two processors are executed, one or more processors are allowed to execute a method for evaluating blood vessels in a color fundus photograph image; wherein the method for evaluating blood vessels in a color fundus photograph image includes: acquiring a color fundus photograph image, and using a pre-trained segmentation model from the The retinal blood vessel image is extracted from the fundus color photograph image, wherein the retinal blood vessel image is a topological structure image; the blood vessel centerline in the retinal blood vessel image is extracted, and the centerline of the blood vessel centerline is set based on the preset standard blood vessel centerline Error band; calculate the error of the blood vessel center line compared to the standard blood vessel center line according to the error band, to obtain the blood vessel topology error of the retinal blood vessel image; segment the segmentation model according to the blood vessel topology error The color fundus photo images are described for evaluation, and the evaluation results are generated. I won't repeat them here.

综合上述实施例可知，本申请最大的有益效果在于：Based on the foregoing embodiments, it can be seen that the greatest beneficial effect of the present application lies in:

本申请所提供的一种眼底彩照图像血管评估方法、装置、计算机设备和计算机可读存储介质，通过利用预先训练好的分割模型从眼底彩照图像中提取出拓扑结构的视网膜血管图像，并提取视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，根据误差带计算所述各血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差，根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。由于视网膜血管图像为拓扑结构图像，因此可直接地从视网膜血管图像中确定血管中心线，利于后续计算血管拓扑结构误差，以及用于分析各血管中心线的关系。此外，本申请的血管拓扑结构误差用于评估血管拓扑结构的完整性和准确性，以对分割模型分割所述眼底彩照图像进行更加全面地评估，评估效果更好。本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，该计算机程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等存储介质，或随机存储记忆体(Random Access Memory，RAM)等。The method, device, computer equipment, and computer-readable storage medium for evaluating blood vessels in fundus color photograph images provided by this application extract the topological structure of the retinal blood vessel image from the fundus color photograph image by using a pre-trained segmentation model, and extract the retina For the blood vessel center line in the blood vessel image, set the error band of the blood vessel center line based on the preset standard blood vessel center line, and calculate the error of each blood vessel center line compared with the standard blood vessel center line according to the error band, The blood vessel topology structure error of the retinal blood vessel image is obtained, and the segmentation model is evaluated based on the blood vessel topology structure error to segment the fundus color photograph image, and an evaluation result is generated. Since the retinal blood vessel image is a topological structure image, the blood vessel centerline can be directly determined from the retinal blood vessel image, which is beneficial to the subsequent calculation of the error of the blood vessel topology structure, and is used to analyze the relationship between the centerline of each blood vessel. In addition, the blood vessel topology structure error of the present application is used to evaluate the completeness and accuracy of the blood vessel topology structure, so as to perform a more comprehensive evaluation of the fundus color photograph image segmented by the segmentation model, and the evaluation effect is better. A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through a computer program. The computer program can be stored in a computer readable storage medium. When executed, it may include the procedures of the above-mentioned method embodiments. Among them, the aforementioned storage medium may be a storage medium such as a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.

Claims

一种眼底彩照图像血管评估方法，其中，包括以下步骤：A method for evaluating blood vessels in color fundus photographs, which includes the following steps:

获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；Acquiring a color fundus photograph image, and extracting a retinal blood vessel image from the color fundus photograph image by using a pre-trained segmentation model, wherein the retinal blood vessel image is a topological structure image;

提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，其中，所述误差带为距离预设的标准血管中心线预设距离的图像区域；Extract the blood vessel center line in the retinal blood vessel image, and set the error band of the blood vessel center line based on the preset standard blood vessel center line, wherein the error band is a preset distance from the preset standard blood vessel center line Image area;

根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；Calculating the error of the blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the blood vessel topology error of the retinal blood vessel image;

根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。The segmentation model is evaluated to segment the fundus color photograph image according to the error of the blood vessel topology structure, and an evaluation result is generated.
根据权利要求1所述的眼底彩照图像血管评估方法，其中，提取所述视网膜血管图像中的血管中心线，包括：The method for evaluating blood vessels in a color fundus photograph image according to claim 1, wherein extracting the blood vessel center line in the retinal blood vessel image comprises:

对所述视网膜血管图像二值化处理，得到二值图像；Binarize the retinal blood vessel image to obtain a binary image;

利用形态学骨架提取算法对所述二值图像的血管结构进行骨架提取，得到各血管中心线，所述血管中心线为血管结构各处内切圆的圆心连接线。A morphological skeleton extraction algorithm is used to perform skeleton extraction on the blood vessel structure of the binary image to obtain the center line of each blood vessel, and the blood vessel center line is the center connection line of the inscribed circle of the blood vessel structure.
根据权利要求1所述的眼底彩照图像血管评估方法，其中，提取所述视网膜血管图像中的血管中心线之后，还包括：The method for evaluating blood vessels in a color fundus photograph image according to claim 1, wherein after extracting the blood vessel center line in the retinal blood vessel image, the method further comprises:

计算每两根血管中心线之间的最短距离及夹角；Calculate the shortest distance and included angle between the center lines of every two blood vessels;

若所述最短距离小于阈值且夹度小于预设角度，获取所述两根血管中心线之间距离最短的断点，将所述断点相连。If the shortest distance is less than the threshold and the clamping degree is less than the preset angle, the break point with the shortest distance between the center lines of the two blood vessels is obtained, and the break points are connected.
根据权利要求1所述的眼底彩照图像血管评估方法，其中，提取所述视网膜血管图像中的血管中心线之后，还包括：The method for evaluating blood vessels in a color fundus photograph image according to claim 1, wherein after extracting the blood vessel center line in the retinal blood vessel image, the method further comprises:

计算所述视网膜血管图像的面积，利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数；Calculating the area of the retinal blood vessel image, traversing the retinal blood vessel image using a box of a preset size, and calculating the number of box traversals according to the area of the box and the area of the retinal blood vessel image;

计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数；Calculating the cumulative number of times that the box covers the retinal blood vessel image during traversal and includes the centerline of the blood vessel;

更换盒子的大小，继续执行利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数，计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数的步骤，得到不同尺寸的盒子的遍历次数及累计次数；Change the size of the box, continue to traverse the retinal blood vessel image with the box of the preset size, and calculate the number of box traversals based on the area of the box and the area of the retinal blood vessel image, and calculate the area covered by the retinal blood vessel image when the box is traversed Including the step of accumulating the number of times of the blood vessel centerline to obtain the number of traversals and accumulating times of boxes of different sizes;

根据所述不同尺寸的盒子的遍历次数及累计次数计算血管结构的分形维数。The fractal dimension of the blood vessel structure is calculated according to the traversal times and the accumulated times of the boxes of different sizes.
根据权利要求1所述的眼底彩照图像血管评估方法，其中，提取所述视网膜血管图像中的血管中心线之后，还包括：The method for evaluating blood vessels in a color fundus photograph image according to claim 1, wherein after extracting the blood vessel center line in the retinal blood vessel image, the method further comprises:

测量所述血管中心线的弧长，计算所述血管中心线各点曲率的平方和；Measuring the arc length of the center line of the blood vessel, and calculating the sum of the squares of the curvature of each point of the center line of the blood vessel;

根据所述弧长及平方和计算血管结构的血管平均曲率。The average curvature of the blood vessel of the blood vessel structure is calculated based on the arc length and the square sum.
根据权利要求5所述的眼底彩照图像血管评估方法，其中，计算所述血管中心线各点曲率的平方和之前，还包括：The method for evaluating blood vessels in a color fundus photograph image according to claim 5, wherein before calculating the sum of the squares of the curvature of each point on the center line of the blood vessel, the method further comprises:

从血管中心线上任意选取一点作为目标点，依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率；Any point selected from the center line of the blood vessel as the target point, the cosine of the included angle between the remaining points on the center line of the blood vessel and the target point is sequentially calculated, and the value with the largest included angle cosine is selected as the curvature of the target point;

从血管中心线上的其余点中选取一点作为目标点，重新执行依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率的步骤，直至得到所述血管中心线各点的曲率。Select a point from the remaining points on the blood vessel center line as the target point, re-calculate the cosine of the included angle between the remaining points on the blood vessel center line and the target point, and select the largest value of the included angle cosine as the curvature of the target point Steps until the curvature of each point of the center line of the blood vessel is obtained.
根据权利要求1所述的眼底彩照图像血管评估方法，其中，根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差的步骤，包括：The method for evaluating blood vessels in a color fundus photograph image according to claim 1, wherein the step of calculating the error of the blood vessel center line compared to the standard blood vessel center line according to the error band comprises:

获取眼底彩照图像中血管中心线的总数量；Obtain the total number of blood vessel centerlines in the fundus color photo image;

计算所述血管中心线中未落入所述误差带内的第一数量；Calculating the first number of the centerline of the blood vessel that does not fall within the error band;

根据所述第一数量及总数量计算所述误差。The error is calculated according to the first quantity and the total quantity.
一种眼底彩照图像血管评估装置，其中，包括：A device for evaluating blood vessels in color fundus images, which includes:

提取模块，用于获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；An extraction module for acquiring a color fundus photo image, and extracting a retinal blood vessel image from the color fundus photo image using a pre-trained segmentation model, wherein the retinal blood vessel image is a topological structure image;

设置模块，用于提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，其中，所述误差带为距离预设的标准血管中心线预设距离的图像区域；The setting module is used to extract the blood vessel center line in the retinal blood vessel image, and set the error band of the blood vessel center line based on a preset standard blood vessel center line, wherein the error band is a standard blood vessel with a preset distance The image area with a preset distance from the center line;

计算模块，用于根据所述误差带计算所述各血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；A calculation module, configured to calculate the error of each blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the blood vessel topology structure error of the retinal blood vessel image;

生成模块，用于根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。The generating module is configured to evaluate the segmentation model to segment the fundus color photograph image according to the error of the blood vessel topology structure, and generate an evaluation result.
一种计算机设备，其中，包括：A computer device, which includes:

一个或多个处理器；One or more processors;

存储器；Memory

一个或多个计算机程序，其中所述一个或多个计算机程序被存储在所述存储器中并被配置为由所述一个或多个处理器执行，所述一个或多个计算机程序配置用于执行一种眼底彩照图像血管评估方法；One or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and the one or more computer programs are configured to execute A method for evaluating blood vessels in fundus color photographs;

其中，所述眼底彩照图像血管评估方法包括：Wherein, the blood vessel evaluation method of the fundus color photograph image includes:

获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；Acquiring a color fundus photograph image, and extracting a retinal blood vessel image from the color fundus photograph image by using a pre-trained segmentation model, wherein the retinal blood vessel image is a topological structure image;

提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，其中，所述误差带为距离预设的标准血管中心线预设距离的图像区域；Extract the blood vessel center line in the retinal blood vessel image, and set the error band of the blood vessel center line based on the preset standard blood vessel center line, wherein the error band is a preset distance from the preset standard blood vessel center line Image area;

根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；Calculating the error of the blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the blood vessel topology error of the retinal blood vessel image;

根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。The segmentation model is evaluated to segment the fundus color photograph image according to the error of the blood vessel topology structure, and an evaluation result is generated.
根据权利要求9所述的计算机设备，其中，提取所述视网膜血管图像中的血管中心线，包括：The computer device according to claim 9, wherein extracting the blood vessel centerline in the retinal blood vessel image comprises:

对所述视网膜血管图像二值化处理，得到二值图像；Binarize the retinal blood vessel image to obtain a binary image;

利用形态学骨架提取算法对所述二值图像的血管结构进行骨架提取，得到各血管中心线，所述血管中心线为血管结构各处内切圆的圆心连接线。A morphological skeleton extraction algorithm is used to perform skeleton extraction on the blood vessel structure of the binary image to obtain the center line of each blood vessel, and the blood vessel center line is the center connection line of the inscribed circle of the blood vessel structure.
根据权利要求9所述的计算机设备，其中，提取所述视网膜血管图像中的血管中心线之后，还包括：The computer device according to claim 9, wherein after extracting the blood vessel center line in the retinal blood vessel image, the method further comprises:

计算每两根血管中心线之间的最短距离及夹角；Calculate the shortest distance and included angle between the center lines of every two blood vessels;

若所述最短距离小于阈值且夹度小于预设角度，获取所述两根血管中心线之间距离最短的断点，将所述断点相连。If the shortest distance is less than the threshold and the clamping degree is less than the preset angle, the break point with the shortest distance between the center lines of the two blood vessels is obtained, and the break points are connected.
根据权利要求9所述的计算机设备，其中，提取所述视网膜血管图像中的血管中心线之后，还包括：The computer device according to claim 9, wherein after extracting the blood vessel center line in the retinal blood vessel image, the method further comprises:

计算所述视网膜血管图像的面积，利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数；Calculating the area of the retinal blood vessel image, traversing the retinal blood vessel image using a box of a preset size, and calculating the number of box traversals according to the area of the box and the area of the retinal blood vessel image;

计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数；Calculating the cumulative number of times that the box covers the retinal blood vessel image during traversal and includes the centerline of the blood vessel;

更换盒子的大小，继续执行利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数，计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数的步骤，得到不同尺寸的盒子的遍历次数及累计次数；Change the size of the box, continue to traverse the retinal blood vessel image with the box of the preset size, and calculate the number of box traversals based on the area of the box and the area of the retinal blood vessel image, and calculate the area covered by the retinal blood vessel image when the box is traversed Including the step of accumulating the number of times of the blood vessel centerline to obtain the number of traversals and accumulating times of boxes of different sizes;

根据所述不同尺寸的盒子的遍历次数及累计次数计算血管结构的分形维数。The fractal dimension of the blood vessel structure is calculated according to the traversal times and the accumulated times of the boxes of different sizes.
根据权利要求9所述的计算机设备，其中，提取所述视网膜血管图像中的血管中心线之后，还包括：The computer device according to claim 9, wherein after extracting the blood vessel center line in the retinal blood vessel image, the method further comprises:

测量所述血管中心线的弧长，计算所述血管中心线各点曲率的平方和；Measuring the arc length of the center line of the blood vessel, and calculating the sum of the squares of the curvature of each point of the center line of the blood vessel;

根据所述弧长及平方和计算血管结构的血管平均曲率。The average curvature of the blood vessel of the blood vessel structure is calculated based on the arc length and the square sum.
根据权利要求13所述的计算机设备，其中，计算所述血管中心线各点曲率的平方和之前，还包括：The computer device according to claim 13, wherein before calculating the sum of the squares of the curvature of each point of the blood vessel centerline, the method further comprises:

从血管中心线上任意选取一点作为目标点，依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率；Any point selected from the center line of the blood vessel as the target point, the cosine of the included angle between the remaining points on the center line of the blood vessel and the target point is sequentially calculated, and the value with the largest included angle cosine is selected as the curvature of the target point;

从血管中心线上的其余点中选取一点作为目标点，重新执行依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率的步骤，直至得到所述血管中心线各点的曲率。Select a point from the remaining points on the blood vessel center line as the target point, re-calculate the cosine of the included angle between the remaining points on the blood vessel center line and the target point, and select the largest value of the included angle cosine as the curvature of the target point Steps until the curvature of each point of the center line of the blood vessel is obtained.
一种计算机可读存储介质，其中，所述计算机可读存储介质上存储有计算机程序，该计算机程序被处理器执行时实现一种眼底彩照图像血管评估方法；A computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, a method for evaluating blood vessels in a fundus color photograph image is realized;

其中，所述眼底彩照图像血管评估方法包括：Wherein, the blood vessel evaluation method of the fundus color photograph image includes:

获取眼底彩照图像，利用预先训练好的分割模型从所述眼底彩照图像中提取视网膜血管图像，其中，所述视网膜血管图像为拓扑结构图像；Acquiring a color fundus photograph image, and extracting a retinal blood vessel image from the color fundus photograph image by using a pre-trained segmentation model, wherein the retinal blood vessel image is a topological structure image;

提取所述视网膜血管图像中的血管中心线，以预设的标准血管中心线为基准设置所述血管中心线的误差带，其中，所述误差带为距离预设的标准血管中心线预设距离的图像区域；Extract the blood vessel center line in the retinal blood vessel image, and set the error band of the blood vessel center line based on the preset standard blood vessel center line, wherein the error band is a preset distance from the preset standard blood vessel center line Image area;

根据所述误差带计算所述血管中心线相较于所述标准血管中心线的误差，得到所述视网膜血管图像的血管拓扑结构误差；Calculating the error of the blood vessel center line compared to the standard blood vessel center line according to the error band to obtain the blood vessel topology error of the retinal blood vessel image;

根据所述血管拓扑结构误差对分割模型分割所述眼底彩照图像进行评估，生成评估结果。The segmentation model is evaluated to segment the fundus color photograph image according to the error of the blood vessel topology structure, and an evaluation result is generated.
根据权利要求15所述的计算机可读存储介质，其中，提取所述视网膜血管图像中的血管中心线，包括：15. The computer-readable storage medium of claim 15, wherein extracting the blood vessel centerline in the retinal blood vessel image comprises:

对所述视网膜血管图像二值化处理，得到二值图像；Binarize the retinal blood vessel image to obtain a binary image;

利用形态学骨架提取算法对所述二值图像的血管结构进行骨架提取，得到各血管中心线，所述血管中心线为血管结构各处内切圆的圆心连接线。A morphological skeleton extraction algorithm is used to perform skeleton extraction on the blood vessel structure of the binary image to obtain the center line of each blood vessel, and the blood vessel center line is the center connection line of the inscribed circle of the blood vessel structure.
根据权利要求15所述的计算机可读存储介质，其中，提取所述视网膜血管图像中的血管中心线之后，还包括：15. The computer-readable storage medium according to claim 15, wherein after extracting the blood vessel centerline in the retinal blood vessel image, the method further comprises:

计算每两根血管中心线之间的最短距离及夹角；Calculate the shortest distance and included angle between the center lines of every two blood vessels;

若所述最短距离小于阈值且夹度小于预设角度，获取所述两根血管中心线之间距离最短的断点，将所述断点相连。If the shortest distance is less than the threshold and the clamping degree is less than the preset angle, the break point with the shortest distance between the center lines of the two blood vessels is obtained, and the break points are connected.
根据权利要求15所述的计算机可读存储介质，其中，提取所述视网膜血管图像中的血管中心线之后，还包括：15. The computer-readable storage medium according to claim 15, wherein after extracting the blood vessel centerline in the retinal blood vessel image, the method further comprises:

计算所述视网膜血管图像的面积，利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数；Calculating the area of the retinal blood vessel image, traversing the retinal blood vessel image using a box of a preset size, and calculating the number of box traversals according to the area of the box and the area of the retinal blood vessel image;

计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数；Calculating the cumulative number of times that the box covers the retinal blood vessel image during traversal and includes the centerline of the blood vessel;

更换盒子的大小，继续执行利用预设大小的盒子遍历所述视网膜血管图像，并根据盒子的面积及视网膜血管图像的面积计算盒子遍历次数，计算盒子在遍历时覆盖所述视网膜血管图像的区域中包括所述血管中心线的累计次数的步骤，得到不同尺寸的盒子的遍历次数及累计次数；Change the size of the box, continue to traverse the retinal blood vessel image with the box of the preset size, and calculate the number of box traversals based on the area of the box and the area of the retinal blood vessel image, and calculate the area covered by the retinal blood vessel image when the box is traversed Including the step of accumulating the number of times of the blood vessel centerline to obtain the number of traversals and accumulating times of boxes of different sizes;

根据所述不同尺寸的盒子的遍历次数及累计次数计算血管结构的分形维数。The fractal dimension of the blood vessel structure is calculated according to the traversal times and the accumulated times of the boxes of different sizes.
根据权利要求15所述的计算机可读存储介质，其中，提取所述视网膜血管图像中的血管中心线之后，还包括：15. The computer-readable storage medium according to claim 15, wherein after extracting the blood vessel centerline in the retinal blood vessel image, the method further comprises:

测量所述血管中心线的弧长，计算所述血管中心线各点曲率的平方和；Measuring the arc length of the center line of the blood vessel, and calculating the sum of the squares of the curvature of each point of the center line of the blood vessel;

根据所述弧长及平方和计算血管结构的血管平均曲率。The average curvature of the blood vessel of the blood vessel structure is calculated based on the arc length and the square sum.
根据权利要求19所述的计算机可读存储介质，其中，计算所述血管中心线各点曲率的平方和之前，还包括：18. The computer-readable storage medium according to claim 19, wherein before calculating the sum of the squares of the curvature of each point on the centerline of the blood vessel, the method further comprises:

从血管中心线上任意选取一点作为目标点，依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率；Any point selected from the center line of the blood vessel as the target point, the cosine of the included angle between the remaining points on the center line of the blood vessel and the target point is sequentially calculated, and the value with the largest included angle cosine is selected as the curvature of the target point;

从血管中心线上的其余点中选取一点作为目标点，重新执行依次计算血管中心线上其余点与所述目标点的夹角余弦，选取夹角余弦最大的值作为所述目标点的曲率的步骤，直至得到所述血管中心线各点的曲率。Select a point from the remaining points on the blood vessel center line as the target point, re-calculate the cosine of the included angle between the remaining points on the blood vessel center line and the target point, and select the largest value of the included angle cosine as the curvature of the target point Steps until the curvature of each point of the center line of the blood vessel is obtained.