Measurement system and measurement method of pathologic features of hypertensive retinopathy

Abstract

This disclosure provides a measurement system and method for the pathological features of hypertensive retinopathy. The measurement system includes: an acquisition module that acquires fundus images; a partitioning module that receives the fundus images and identifies and partitions the optic disc region; a segmentation module that uses an arteriovenous segmentation model to segment the fundus images into arteries and veins to obtain a three-valued image of the segmentation results. The arteriovenous vessel marking results include arterial and venous marking results formed by marking the boundaries of vessels with diameters larger than a preset vessel diameter in the training fundus images, and small vessel marking results formed by marking the course of vessels with diameters not larger than the preset vessel diameter. The measurement module measures the pathological features of hypertensive retinopathy in the fundus images based on the vessels larger than the preset vessel diameter in the partitioning and segmentation results. This disclosure enables efficient and objective measurement of the characteristics of hypertensive retinopathy.

Description

[0001] This application is a divisional application of the patent application filed on December 28, 2020, with application number 202011585139.9, entitled "Method and System for Measuring the Lesion Characteristics of Hypertensive Retinopathy". Technical Field

[0002] This application relates to the field of retinal disease measurement, specifically to a measurement system and method for measuring the pathological characteristics of hypertensive retinopathy. Background Technology

[0003] Currently, hundreds of millions of people suffer from hypertension, and what's even more worrying is that the age of onset is becoming increasingly younger. Hypertension easily leads to complications such as myocardial infarction, kidney failure, cerebral hemorrhage, cerebral infarction, and uremia, which can endanger life. Therefore, early identification and intervention of hypertension are of great significance.

[0004] In clinical practice, hypertension can be diagnosed by measuring blood pressure. However, blood pressure measurement generally cannot assess a patient's recent or long-term hypertension. Currently, observing changes in the blood vessels of the fundus can reveal pathological features of hypertensive retinopathy that reflect hypertension (such as arteriovenous crossing impressions, localized arteriolar stenosis, or generalized arteriolar stenosis), thus allowing assessment of the severity of hypertension. Generally, localized arteriolar stenosis indicates recent blood pressure elevation, while arteriovenous crossing impressions and generalized arteriolar stenosis indicate long-term, persistent blood pressure elevation. In such cases, other features (such as retinal hemorrhage or microaneurysms) can be combined to assist ophthalmologists in identifying hypertension. Existing methods typically utilize fundus photography to obtain fundus images, facilitating the observation and measurement of pathological features of hypertensive retinopathy. However, manually measuring these features is time-consuming and inherently subjective. Summary of the Invention

[0005] This disclosure is made in view of the above-mentioned situation, and its purpose is to provide a method and system for measuring the pathological characteristics of hypertensive retinopathy that can efficiently and objectively measure the pathological characteristics of hypertensive retinopathy.

[0006] To this end, a first aspect of the present disclosure provides a method for measuring the lesion characteristics of hypertensive retinopathy, including: obtaining fundus images; identifying the optic disc region of the fundus images and dividing the fundus images into at least three regions including a first region, a second region, and a third region based on the optic disc region; using an arteriovenous segmentation model based on deep learning trained with training fundus images and arteriovenous vessel marking results to segment arteries and veins in the fundus images to obtain an arteriovenous segmentation result, where the arteriovenous segmentation result includes an artery segmentation result and a vein segmentation result; and measuring the lesion characteristics of hypertensive retinopathy in the fundus images based on the three regions and the arteriovenous segmentation result, the lesion characteristics including at least one of arteriovenous crossing indentation characteristics, local arteriole stenosis characteristics, and general arteriole stenosis characteristics, the arteriovenous vessel marking results including artery marking results and vein marking results formed by marking the boundaries of vessels with a vessel diameter greater than a preset vessel diameter in the training fundus images and small vessel marking results formed by marking the directions of vessels not greater than the preset vessel diameter. When calculating the loss function, the weights of the regions corresponding to the small vessel marking results are adjusted based on the small vessel marking results. In this case, the fundus images are partitioned based on the optic disc region to obtain at least three regions, and an arteriovenous segmentation model based on deep learning is used to segment arteries and veins in the fundus images to obtain an arteriovenous segmentation result. The lesion characteristics of hypertensive retinopathy in the fundus images are automatically measured based on the three regions and the arteriovenous segmentation result. Thus, the lesion characteristics of hypertensive retinopathy can be measured efficiently and objectively.

[0007] In addition, in the measurement method according to the first aspect of the present disclosure, optionally, the first region is a region of a first circle formed with the center of the circumcircle of the optic disc region as the center and a first preset multiple of the diameter of the circumcircle as the diameter, the second region is a region between the edge of the first region and a second circle formed with the center as the center and a second preset multiple of the diameter of the circumcircle as the diameter, and the third region is a region between the edge of the second region and a third circle formed with the center as the center and a third preset multiple of the diameter of the circumcircle as the diameter, where v1 < v2 < v3, v1 represents the first preset multiple, v2 represents the second preset multiple, and v3 represents the third preset multiple. Thus, three regions can be obtained based on the optic disc region.

[0008] Furthermore, in the measurement method according to the first aspect of this disclosure, optionally, if the arteriovenous crossing impression feature is measured, the arteriovenous segmentation result of the fundus region other than the first region and the second region is refined to obtain a first vascular skeleton including a plurality of skeleton pixels as first measurement pixels, and the number of skeleton pixels within a preset range of each first measurement pixel is obtained and used as the number of first adjacent points. The pixels in the arteriovenous segmentation result corresponding to the first measurement pixels whose number of first adjacent points is greater than a first preset number are taken as the arteriovenous crossing position. The arteriovenous crossing impression feature is measured based on the ratio of the proximal and distal vessel diameters on each side of the arteriovenous crossing position along the extension direction of the arteriovenous segmentation result. Thus, the arteriovenous crossing impression feature can be measured based on the arteriovenous crossing position.

[0009] Furthermore, in the measurement method according to the first aspect of this disclosure, optionally, if the local stenosis features of the small artery are to be measured, the artery segmentation result is refined to obtain a second vascular skeleton including multiple skeleton pixels as second measurement pixels. The number of skeleton pixels within a preset range of each second measurement pixel is obtained as the number of second neighboring points. Second measurement pixels with a number of second neighboring points greater than the second preset number are deleted to obtain multiple vascular segments. The local stenosis features of the small artery are measured based on the ratio of the minimum vascular diameter to the maximum vascular diameter of each vascular segment. Thus, the local stenosis features of the small artery can be measured.

[0010] Furthermore, in the measurement method according to the first aspect of this disclosure, optionally, if the general stenosis characteristics of the small artery are to be measured, the arterial and venous segments in the third region of the arteriovenous segmentation result are obtained, and the arteriovenous diameter ratio is obtained based on the arterial segment, the venous segment, and a preset formula, wherein the preset formula is a modified Knudtson formula, to measure the general stenosis characteristics of the small artery. Thus, the general stenosis characteristics of the small artery can be measured based on the arterial and venous segments in the third region.

[0011] Furthermore, in the measurement method according to the first aspect of this disclosure, optionally, when measuring the arteriovenous crossing impression features, the arterial segmentation result is dilated so that the dilated arterial segmentation result intersects with the venous segmentation result to determine the arteriovenous crossing position. This allows for a more accurate acquisition of the arteriovenous crossing position.

[0012] Furthermore, in the measurement method according to the first aspect of this disclosure, optionally, if the vein segmentation result in the arteriovenous segmentation result is discontinuous at the arteriovenous crossing position, then the proximal end on each side is the skeleton pixel point on the first vascular skeleton of the vein segmentation result that is closest to the arteriovenous crossing position; if the vein segmentation result in the arteriovenous segmentation result is continuous at the arteriovenous crossing position, then the proximal end on each side is the arteriovenous crossing position; and the distal end on each side is the skeleton pixel point on the first vascular skeleton of the vein segmentation result that is at a distance of a first preset distance from the arteriovenous crossing position. Thus, the proximal and distal ends of both sides of the vein segmentation result can be determined based on the arteriovenous crossing position.

[0013] Furthermore, in the measurement method involved in the first aspect of this disclosure, optionally, the first preset distance is 2 to 4 times the maximum blood vessel diameter, and the first preset quantity is 3. Thus, the first preset distance and the first preset quantity can be obtained.

[0014] Furthermore, in the measurement method involved in the first aspect of this disclosure, optionally, the second preset quantity is 2, v1 is 1, v2 is 2, v3 is 3, and the preset blood vessel diameter is 50 μm. Thus, the second preset quantity, the first preset multiple, the second preset multiple, the third preset multiple, and the preset blood vessel diameter can be obtained.

[0015] In addition, in the measurement method according to the first aspect of this disclosure, optionally, measuring the blood vessel diameter includes: enhancing the resolution of the arteriovenous segmentation result by a preset multiple to generate an enhanced arteriovenous segmentation result; extracting the vascular skeleton from the enhanced arteriovenous segmentation result and fitting the vascular skeleton to obtain a continuous vascular skeleton and a diameter measurement direction of a third measurement pixel, wherein the third measurement pixel is a plurality of pixels on the continuous vascular skeleton, and the diameter measurement direction is perpendicular to the tangent of the continuous vascular skeleton at the third measurement pixel; generating a blood vessel contour corresponding to the third measurement pixel using an interpolation algorithm based on the enhanced arteriovenous segmentation result, the third measurement pixel, the diameter measurement direction of the third measurement pixel, and a preset precision; calculating the blood vessel diameter l corresponding to the third measurement pixel based on the number of blood vessel pixels in the blood vessel contour corresponding to the third measurement pixel, the preset multiple, and the preset precision, wherein the blood vessel diameter l corresponding to the third measurement pixel satisfies: l = n × s / e, where n is the number of blood vessel pixels in the blood vessel contour corresponding to the third measurement pixel, s is the preset precision, and e is the preset multiple. In this scenario, the resolution of blood vessels in fundus images can be increased, and more pixels can be used to measure vessel diameter. This enables automated super-resolution measurement of vessel diameter and improves the accuracy of vessel diameter measurements.

[0016] Furthermore, in the measurement method according to the first aspect of this disclosure, the weights may optionally be adjusted to zero. In this case, the contribution of small vessel marking results that only indicate direction and their surrounding areas to the loss function can be excluded. Thus, the influence of small vessel marking results with low accuracy for arteriovenous segmentation on the arteriovenous segmentation model can be avoided.

[0017] The second aspect of this disclosure provides a method for measuring the pathological features of hypertensive retinopathy, comprising: an acquisition module for acquiring a fundus image; a partitioning module for receiving the fundus image, identifying the optic disc region of the fundus image, and dividing the fundus image into at least three regions, including a first region, a second region, and a third region, based on the optic disc region; a segmentation module for performing arterial and venous segmentation on the fundus image using a deep learning-based arterial and venous segmentation model trained on the fundus image and arterial and venous vascular markers to obtain arterial and venous segmentation results, wherein the arterial and venous segmentation results include arterial segmentation results and venous segmentation results; and a measurement module. The method measures the pathological features of hypertensive retinopathy in the fundus image based on the three regions and the arteriovenous segmentation results. The pathological features include at least one of arteriovenous crossing impression features, localized arteriovenous stenosis features, and generalized arteriovenous stenosis features. The arteriovenous vessel marking results include arterial and venous marking results formed by marking the boundaries of vessels with diameters larger than a preset vessel diameter in the training fundus image, and small vessel marking results formed by marking the course of vessels with diameters not larger than a preset vessel diameter. When calculating the loss function, the weights of the regions corresponding to the small vessel marking results are adjusted based on the small vessel marking results. In this case, the fundus image is divided into three regions based on the optic disc region, and the arteriovenous segmentation model based on deep learning is used to segment the fundus image into arteries and veins to obtain arteriovenous segmentation results. The pathological features of hypertensive retinopathy in the fundus image are automatically measured based on the three regions and the arteriovenous segmentation results. Therefore, the pathological features of hypertensive retinopathy can be measured efficiently and objectively.

[0018] According to this disclosure, a method and system for measuring the pathological characteristics of hypertensive retinopathy are provided, which can efficiently and objectively measure the pathological characteristics of hypertensive retinopathy. Attached Figure Description

[0019] This disclosure will now be explained in further detail by way of example only with reference to the accompanying drawings, in which:

[0020] Figure 1 This is a schematic diagram illustrating an application scenario of a method for measuring the pathological characteristics of hypertensive retinopathy as described in this disclosure example.

[0021] Figure 2 This is a flowchart illustrating a method for measuring the pathological characteristics of hypertensive retinopathy as described in this disclosure example.

[0022] Figure 3 This is a schematic diagram illustrating a fundus image relevant to an example of this disclosure.

[0023] Figure 4 This is a schematic diagram illustrating the disc area involved in the example of this disclosure.

[0024] Figure 5 This is a schematic diagram illustrating three regions of a fundus image involved in the example of this disclosure.

[0025] Figure 6 This is a flowchart illustrating the training method of the deep learning-based arteriovenous segmentation model involved in the examples of this disclosure.

[0026] Figure 7(a) is a schematic diagram illustrating the results of arteriovenous vascular marking as described in the example of this disclosure.

[0027] Figure 7(b) is a partial schematic diagram showing the results of arteriovenous vascular marking as described in the example of this disclosure.

[0028] Figure 8 This is a schematic diagram illustrating the results of arteriovenous segmentation as described in the examples of this disclosure.

[0029] Figure 9 This is a flowchart illustrating the measurement of arteriovenous crossing impression features as described in this disclosure example.

[0030] Figure 10 This is a flowchart illustrating the measurement of localized stenosis characteristics of small arteries as described in this disclosure example.

[0031] Figure 11 This is a flowchart illustrating the measurement of blood vessel diameter as described in this disclosure example.

[0032] Figure 12 This is a schematic diagram illustrating the pipe diameter measurement direction as described in the example of this disclosure.

[0033] Figure 13 This is a flowchart illustrating the generation of blood vessel contours as described in this disclosure.

[0034] Figure 14 This is a schematic diagram illustrating an image of a straightened blood vessel as described in the examples of this disclosure.

[0035] Figure 15 This is a block diagram illustrating a measurement system for the pathological characteristics of hypertensive retinopathy as described in this disclosure example.

[0036] Explanation of main labels:

[0037] 100…Application scenario, 110…Operator, 120…Terminal, 130…Acquisition device, 140…Human eye, 150…Server, A…Blood vessel, B…Visual disc region, C1…First region, C2…Second region, C3…Third region, D1…First circle, D2…Second circle, D3…Third circle, E1…Artery marking result, E2…Venus marking result, E3…Small vessel marking result, F1…Artery segmentation result, F2…Venus segmentation result, G1…Third measurement pixel, L1…Continuous vascular skeleton, L2…Tangent, L3…Diameter measurement direction, A'…Straightening blood vessel, 200…Measurement system, 210…Acquisition module, 220…Segmentation module, 230…Measurement module. Detailed Implementation

[0038] The preferred embodiments of this disclosure are described in detail below with reference to the accompanying drawings. In the following description, the same reference numerals are used for the same components, and repeated descriptions are omitted. Furthermore, the drawings are merely schematic diagrams, and the proportions of the components or the shapes of the components may differ from actual figures. It should be noted that the terms "comprising" and "having," and any variations thereof, in this disclosure, do not necessarily limit the process, method, system, product, or apparatus to the explicitly listed steps or units, but may include or have other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses. All methods described in this disclosure may be performed in any suitable order unless otherwise indicated herein or clearly contradicted by the context.

[0039] Figure 1 This is a schematic diagram illustrating an application scenario of the method for measuring the pathological characteristics of hypertensive retinopathy according to the examples of this disclosure. In some examples, the method for measuring the pathological characteristics of hypertensive retinopathy according to this disclosure (sometimes simply referred to as the measurement method) can be applied to, for example... Figure 1 In application scenario 100, operator 110 can control acquisition device 130 connected to terminal 120 to acquire fundus images of human eye 140. After acquisition device 130 completes fundus image acquisition, terminal 120 can submit the fundus images to server 150 via computer network. Server 150 executes computer program instructions to implement a measurement method that obtains lesion features (described later, sometimes simply referred to as lesion features) of hypertensive retinopathy in the fundus images and returns them to terminal 120. In some examples, terminal 120 can display whether lesion features are present. In other examples, lesion features can be stored as intermediate results in the memory of terminal 120 or server 150.

[0040] In some examples, operator 110 may be a physician with expertise in measuring lesion features in fundus images. In other examples, operator 110 may be a general practitioner familiar with how to automatically measure lesion features via terminal 120. Terminal 120 may include, but is not limited to, a laptop, tablet, or desktop computer. In some examples, terminal 120 may be a dedicated device for measuring lesion features, including a processor, memory, display screen, and acquisition device 130. Acquisition device 130 may include, but is not limited to, a camera. For example, a color fundus camera may be used. In some examples, acquisition device 130 may be connected to terminal 120 via a serial port or integrated into terminal 120.

[0041] In some examples, the fundus of the human eye 140 refers to the tissue in the posterior part of the eyeball, which may include the endothelial membrane, retina, macula, and blood vessels (retinal arteries and veins). In some examples, lesion characteristics can be identified by monitoring changes in the blood vessels of the fundus of the human eye 140. In some examples, the server 150 may include one or more processors and one or more memories. The processor may include a central processing unit, a graphics processing unit, and any other electronic components capable of processing data and executing computer program instructions. The memory may be used to store computer program instructions. In some examples, a measurement method can be implemented by executing the computer program instructions in the memory. In some examples, the server 150 may also be a cloud server.

[0042] Figure 2 This is a flowchart illustrating a method for measuring the pathological characteristics of hypertensive retinopathy as described in this disclosure example. Figure 3 This is a schematic diagram illustrating fundus images relevant to examples of this disclosure. In some examples, such as Figure 2 As shown, the measurement method may include acquiring a fundus image (step S110), dividing the fundus image into regions to obtain at least three regions (step S120), performing arteriovenous segmentation on the fundus image to obtain arteriovenous segmentation results (step S130), and measuring the lesion features in the fundus image based on the three regions and the arteriovenous segmentation results (step S140). In this case, the lesion features of hypertensive retinopathy in the fundus image can be automatically measured. Therefore, the lesion features of hypertensive retinopathy can be measured efficiently and objectively.

[0043] In some examples, a fundus image may be acquired in step S110. In some examples, the fundus image may be a color fundus image. A color fundus image can clearly present fundus information such as the inner membrane of the eyeball, retina, macula, and blood vessels (retinal arteries and veins). In other examples, the fundus image may be a grayscale image. In some examples, the fundus image may be a fundus image acquired by the acquisition device 130. For example, as a fundus image... Figure 3 An image of the fundus taken by a fundus camera is shown, wherein the fundus image may include a blood vessel A. Blood vessel A may include arteries and veins (not shown).

[0044] In some examples, the fundus image may be preprocessed in step S110. Generally, because fundus images may have different image formats and sizes, preprocessing can convert them into images of a fixed standard format. A fixed standard format can mean that the images have the same format and consistent size. For example, in some examples, the preprocessed fundus images may have a uniform width of 512 or 1024 pixels.

[0045] Figure 4 This is a schematic diagram illustrating the optic disc region B involved in the examples of this disclosure. In some examples, in step S120, the fundus image may be partitioned to obtain at least three regions. In some examples, the optic disc region B of the fundus image (see...) Figure 4 The optic disc region B in the fundus image can be segmented using image segmentation algorithms to locate it. In some examples, deep learning image segmentation models (such as the U-Net model) can be used to segment the fundus image to locate the optic disc region B. For example, as an example of the optic disc region B... Figure 4 A schematic diagram of the view disk region B located using the U-Net model is shown. However, the examples disclosed herein are not limited to this; in other examples, the image segmentation algorithm may also be active contouring, Grabcut (graph cut), or thresholding, etc.

[0046] Figure 5 This is a schematic diagram illustrating three regions of a fundus image involved in the example of this disclosure.

[0047] In some examples, the fundus image can be divided into at least three regions based on optic disc region B. For example... Figure 5 As shown, in some examples, the three regions may include a first region C1, a second region C2, and a third region C3. Thus, it is possible to obtain the three regions based on the visual disc region B.

[0048] In some examples, the first region C1 can be the region of a first circle D1 formed with the center of the circumcircle of the optic disc region B as the center and the first predetermined multiple of the diameter of the circumcircle as the diameter. In some examples, the first predetermined multiple can be represented by v1. In some examples, v1 can be 1. In some examples, the second region C2 can be the region between the edge of the first region C1 and a second circle D2 formed with the center of the circumcircle as the center and the second predetermined multiple of the diameter of the circumcircle as the diameter. In some examples, the second predetermined multiple can be represented by v2. In some examples, v2 can be 2. In some examples, the third region C3 can be the region between the edge of the second region C2 and a third circle D3 formed with the center of the circumcircle as the center and the third predetermined multiple of the diameter of the circumcircle as the diameter. In some examples, the third predetermined multiple can be represented by v3. In some examples, v3 can be 3. In some examples, it can be made that v1 < v2 < v3. In other examples, according to actual needs, the fundus image can be partitioned in other ways. For example, the first region C1 and the second region C2 can be combined into one region and the third region C3 can be used as another region.

[0049] In some examples, in step S130, the fundus image can be subjected to arteriovenous segmentation to obtain an arteriovenous segmentation result. Thus, the arterial and venous regions can be identified, thereby removing the influence of other structures other than arteries and veins.

[0050] In some examples, the arteries and veins in the fundus image can be segmented to directly obtain an arterial segmentation result and a venous segmentation result (i.e., an arteriovenous segmentation result). In some examples, a training fundus image and its arteriovenous blood vessel labeling result can be used as a training set to train a deep learning-based arteriovenous segmentation model to perform arteriovenous segmentation on the fundus image in step S110. The training method of the deep learning-based arteriovenous segmentation model will be described in detail below with reference to the accompanying drawings. Figure 6 FIG. is a flowchart showing a training method of a deep learning-based arteriovenous segmentation model according to an example of the present disclosure.

[0051] As Figure 6 shown, in some examples, the training method of the arteriovenous segmentation model can include obtaining a training fundus image and an arteriovenous blood vessel labeling result (step S121), preprocessing the training fundus image and the arteriovenous blood vessel labeling result to obtain a preprocessed fundus image and a preprocessed arteriovenous blood vessel labeling result (step S122), and training the arteriovenous segmentation model based on the preprocessed fundus image and the preprocessed arteriovenous blood vessel labeling result to obtain an optimal model (step S123). In this case, the deep learning-based arteriovenous segmentation model can automatically learn arterial and venous features and output an arteriovenous segmentation result.

[0052] Figure 7(a) is a schematic diagram illustrating the arteriovenous vessel marking results according to the present disclosure example. Figure 7(b) is a partial schematic diagram illustrating the arteriovenous vessel marking results according to the present disclosure example. In some examples, in step S121, the training fundus image can be a fundus image obtained by taking a photograph of the fundus. In some examples, the training fundus image can be annotated to obtain the arteriovenous vessel marking results of the training fundus image. In some examples, the arteriovenous vessel marking results can include artery marking results and vein marking results (not shown).

[0053] In some examples, the arteriovenous vessel labeling results can be formed by labeling the blood vessels in the training fundus image. In some examples, the arteriovenous vessel labeling results can be formed by labeling the boundaries of the blood vessels in the training fundus image. In some examples, the arteriovenous vessel labeling results can be formed by labeling the boundaries of all blood vessels in the training fundus image.

[0054] In some examples, arteriovenous vessel labeling results can be formed by labeling the boundaries of vessels in portions of the training fundus image. Generally, due to the similarity of vessel features, labeling only the boundaries of vessels larger than a preset vessel diameter can lead to confusion between non-arteriovenous regions (i.e., background) and vessels smaller than the preset vessel diameter (referred to as small vessels). Because of the high similarity between vessels smaller and larger than the preset vessel diameter, labeling only vessels larger than the preset diameter severely affects the accuracy of arteriovenous segmentation. Therefore, it is necessary to label small vessels, and subsequently, the contribution of this region to the loss function can be controlled by adjusting the weights to improve the accuracy of arteriovenous segmentation. In some examples, the preset vessel diameter can be 50 μm.

[0055] Specifically, the boundaries of blood vessels with diameters larger than a preset diameter in the training fundus images can be marked to form artery and vein marking results, and the direction of small blood vessels can be marked to form small blood vessel marking results. That is, the arteriovenous vessel marking results can include artery marking result E1 (larger than the preset diameter), vein marking result E2 (larger than the preset diameter), and small blood vessel marking result E3 (see Figure 7). When calculating the loss function, the weights of the regions corresponding to the small blood vessel marking results can be adjusted based on the small blood vessel marking results. In this case, compared to accurately marking the boundaries of small blood vessels, marking only the direction of small blood vessels can reduce the marking difficulty and workload, thereby reducing the marking cost. When training the arteriovenous segmentation model, adjusting the weights of the regions corresponding to the small blood vessel marking results can control the contribution of small blood vessel marking results (marking only the direction) to the loss function, thereby improving the accuracy of arteriovenous segmentation.

[0056] In some examples, the orientation of small blood vessels can be a low-precision vessel boundary, which can be used to estimate the region corresponding to the small vessels. In some examples, the orientation of small blood vessels can be any curve following the orientation of the small vessels. In some examples, when calculating the loss function, the weight of the region corresponding to the small vessel labeling result can be adjusted to 0. In this case, the contribution of small vessel labeling results that only label the orientation to the loss function can be excluded. Thus, the influence of low-precision small vessel labeling results on the arteriovenous segmentation model can be avoided. It should be noted that although training is performed on vessels larger than the preset vessel diameter, some small vessels can still be accurately segmented due to the similarity of vessel features. In some examples, the region corresponding to the small vessel labeling result can be the fundus region and its swollen area corresponding to the small vessel labeling result.

[0057] In some examples, experienced physicians can use labeling tools to label training fundus images to generate arteriovenous (AVM) labeling results. As an example of AVM labeling results, Figure 7(a) shows the AVM labeling results generated after labeling the training fundus image. These AVM labeling results can include artery labeling result E1, vein labeling result E2, and small vessel labeling result E3. To illustrate the AVM labeling results more clearly, Figure 7(b) shows a localized AVM labeling result. In some examples, the AVM labeling results may also include a background area (not shown).

[0058] In some examples, step S122 may preprocess the training fundus image to generate a preprocessed fundus image. In some examples, preprocessing the training fundus image may include cropping, noise reduction, and grayscale conversion. This can highlight blood vessels in the training fundus image. In some examples, the arteriovenous vessel marking results may be preprocessed to generate preprocessed arteriovenous vessel marking results. In some examples, the preprocessing of the arteriovenous vessel marking results may be the same as the preprocessing of the training fundus image.

[0059] In some examples, in step S123, the preprocessed arteriovenous vessel marking results can be used as ground truth to calculate the loss function. The arteriovenous segmentation model is then continuously optimized using this loss function until its value (i.e., the loss) converges to the optimal value, thus obtaining the optimal model. In some examples, the optimal model can output the probability that each pixel in the preprocessed fundus image belongs to an artery, vein, or background. This allows for the identification of arterial and vein regions that match the preprocessed arteriovenous vessel marking results, thereby directly obtaining the arterial and vein segmentation results (i.e., the arteriovenous segmentation results). Therefore, the arteriovenous segmentation results can be directly obtained using the arteriovenous segmentation model. In some examples, the arteriovenous segmentation model can be a U-Net model.

[0060] Figure 8 This is a schematic diagram illustrating the arteriovenous segmentation result involved in the example of this disclosure. In some examples, the fundus image obtained in step S110 can be input into the optimal model obtained through the above training method to perform arteriovenous segmentation on the fundus image, thereby generating an arteriovenous segmentation result. As an example of the arteriovenous segmentation result, for example... Figure 8 The diagram illustrates the arterial and vein segmentation results generated from a fundus image. These results can include artery segmentation result F1 and vein segmentation result F2. In some examples, the segmentation results can be three-valued images. In some examples, the three-valued image can include three grayscale values, representing the artery, vein, and background, respectively.

[0061] However, the examples disclosed herein are not limited to this. In other examples, the vascular segmentation results can be obtained by first performing vascular segmentation on the fundus image (i.e., without distinguishing between arteries and veins), and then the arteries and veins can be classified based on the vascular segmentation results to obtain artery segmentation result F1 and vein segmentation result F2.

[0062] In some examples, in step S140 of the measurement method, the lesion features of hypertensive retinopathy in the fundus image can be measured based on the three regions and the arteriovenous segmentation results. In some examples, the lesion features may include at least one of arteriovenous crossing impression features, localized arteriovenous stenosis features, and generalized arteriovenous stenosis features. For example, the lesion features may be one of arteriovenous crossing impression features, localized arteriovenous stenosis features, or generalized arteriovenous stenosis features. For example, the lesion features may be arteriovenous crossing impression features and localized arteriovenous stenosis features, arteriovenous crossing impression features and generalized arteriovenous stenosis features, etc. In some examples, the lesion features may be arteriovenous crossing impression features, localized arteriovenous stenosis features, and generalized arteriovenous stenosis features.

[0063] As described above, although training is performed on vessels larger than a preset vessel diameter, small vessels can still be segmented due to the similarity of vessel features. In some examples, vessels larger than a preset vessel diameter in the arteriovenous segmentation results can be selected for lesion feature measurement. However, the examples disclosed herein are not limited to this; in other examples, selection can be omitted, and the vessels can be directly used for lesion feature measurement. This further improves the accuracy of lesion feature measurement.

[0064] Figure 9 This is a flowchart illustrating the measurement of arteriovenous crossing impression features as described in this disclosure. Figure 9 As shown, in some examples, measuring the arteriovenous crossing impression features may include refining the vessels in the arteriovenous segmentation results to obtain a first vascular skeleton (step S1411), obtaining the arteriovenous crossing position (step S1412), obtaining the ratio of the proximal and distal vessel diameters on both sides of the arteriovenous crossing position (step S1413), and measuring the arteriovenous crossing impression features based on the ratio (step S1414).

[0065] In some examples, in step S1411, the blood vessels in the arteriovenous segmentation result can be refined to obtain a first vascular skeleton. Generally, the nerve fibers in the first region C1 and the second region C2 are dense, making the vessels at the arteriovenous crossing points appear narrow, which is not within the scope of measuring the arteriovenous crossing impression features. In this case, the arteriovenous segmentation result of the fundus region other than the first region C1 and the second region C2 can be refined to obtain the first vascular skeleton. In some examples, the first vascular skeleton may include multiple skeleton pixels that serve as the first measurement pixel. The fundus region may be a region within the eyeball contour. In some examples, morphological thinning algorithms can be used to refine the arteriovenous segmentation result. In some examples, morphological thinning algorithms may include, but are not limited to, the Hilditch thinning algorithm, the Pavlidis thinning algorithm, or the Rosenfeld thinning algorithm.

[0066] In some examples, the arteriovenous crossing location can be obtained in step S1412. In some examples, the number of skeleton pixels within a preset range of each first measured pixel of the first vascular skeleton can be obtained and used as the number of first neighboring points. In some examples, first measured pixels with a number of first neighboring points greater than a first preset number can be obtained and used as crossing points. In some examples, the pixel corresponding to the crossing point in the arteriovenous segmentation result can be used as the arteriovenous crossing location. In some examples, the preset range can be the eight neighborhoods surrounding the first measured pixel. In some examples, the first preset number can be 3.

[0067] Because the vein segmentation result F2 may be discontinuous at the arteriovenous crossing point, typically the arterial segmentation result F1 and the vein segmentation result F2 do not intersect at this point, making some arteriovenous crossing points unobtainable directly through steps S1411 and S1412. In some examples, the arterial segmentation result F1 can be expanded to intersect with the vein segmentation result F2 to determine the arteriovenous crossing point. Specifically, the intersecting expanded arterial segmentation result F1 and vein segmentation result F2 can be refined to obtain the vascular skeleton, and the arteriovenous crossing point can be determined based on this vascular skeleton. Refer to the description of obtaining the arteriovenous crossing point based on the first vascular skeleton in steps S1411 and S1412 for details. This allows for more accurate arteriovenous crossing points to be obtained.

[0068] In some examples, the maximum blood vessel diameter can be an estimated maximum blood vessel diameter. In some examples, the estimated maximum blood vessel diameter can be twice the maximum distance between a skeleton pixel on the first blood vessel skeleton and its nearest non-vascular pixel. This distance can be a Euclidean distance. Specifically, the Euclidean distance between each skeleton pixel and its nearest non-vascular pixel is calculated, and twice the maximum Euclidean distance is selected as the estimated maximum blood vessel diameter. However, the examples in this disclosure are not limited to this; in other examples, other methods can be used to estimate the maximum blood vessel diameter.

[0069] In some examples, in step S1413, the ratio of the vessel diameters of the vein segmentation result F2 at the proximal and distal ends on both sides of the arteriovenous crossing position can be calculated. In some examples, the proximal and distal ends on both sides of the arteriovenous crossing position can be the proximal and distal ends of each side of the arteriovenous segmentation result along the extension direction of the vein segmentation result F2 and located on both sides of the arteriovenous crossing position.

[0070] In some examples, if the vein segmentation result F2 in the arteriovenous segmentation result is discontinuous at the arteriovenous intersection, then the proximal end on each side can be the skeleton pixel on the first vascular skeleton of the vein segmentation result F2 that is closest to the arteriovenous intersection. In some examples, if the vein segmentation result F2 in the arteriovenous segmentation result is continuous at the arteriovenous intersection, then the proximal end on each side can be the arteriovenous intersection. In some examples, the distal end on each side can be the skeleton pixel on the first vascular skeleton of the vein segmentation result F2 that is at a distance of a first preset distance from the arteriovenous intersection. Thus, the proximal and distal ends of both sides of the vein segmentation result can be determined based on the arteriovenous intersection. In some examples, the first preset distance can be 2 to 4 times the maximum vessel diameter. In some examples, the first preset distance can be 3 times the maximum vessel diameter.

[0071] In some examples, in step S1413, the arteriovenous segmentation results can be cropped with the arteriovenous crossing point as the center and a radius of 4 to 6 times the maximum vessel diameter. This allows for convenient display of the arterial segmentation result F1 and the vein segmentation result F2 near each arteriovenous crossing point.

[0072] In some examples, in step S1414, the arteriovenous crossing impression feature can be measured based on a ratio. In some examples, if the ratio on both sides is not greater than 1 / 2, it indicates the presence of an arteriovenous crossing impression feature. In some examples, if the ratio on only one side is not greater than 1 / 2, it indicates the presence of a suspected arteriovenous crossing impression feature. In some examples, if the ratio on both sides is greater than 1 / 2, it indicates the absence of an arteriovenous crossing impression feature.

[0073] Figure 10 This is a flowchart illustrating the measurement of localized stenosis characteristics of small arteries as described in this disclosure. As stated above, lesion characteristics may include localized stenosis characteristics of small arteries. Figure 10 As shown, in some examples, measuring the local stenosis features of small arteries may include refining the artery segmentation results to obtain a second vascular skeleton (step S1421), obtaining the number of second neighboring points (step S1422), deleting second measurement pixels based on the number of second neighboring points to obtain multiple vascular segments (step S1423), obtaining the ratio of the minimum vascular diameter to the maximum vascular diameter of each vascular segment (step S1424), and measuring the local stenosis features of small arteries based on the ratio (step S1425).

[0074] In some examples, in step S1421, the artery segmentation result F1 with a vessel diameter larger than a preset vessel diameter can be refined to obtain a second vessel skeleton. For a description of the refinement process, please refer to the relevant description in step S1411. In some examples, the second vessel skeleton may include multiple skeleton pixels serving as second measurement pixels. In some examples, the preset vessel diameter may be 50 μm.

[0075] In some examples, in step S1422, the number of skeleton pixels within a preset range of each second measurement pixel can be used as the number of second neighboring points. In some examples, the preset range can be the eight neighborhoods surrounding the second measurement pixel.

[0076] In some examples, in step S1423, second measurement pixels with a number greater than a second preset number can be deleted to obtain multiple blood vessel segments. In some examples, the second preset number can be 2.

[0077] In some examples, in step S1424, the vessel diameter of each vessel segment at each skeleton pixel can be calculated to obtain the minimum vessel diameter w of each vessel segment. min and the largest blood vessel diameter w max The ratio. In some examples, in step S1425, the characteristics of localized arterial stenosis can be measured based on the ratio. In some examples, if wmin / wmax ≥ 2 / 3, then no localized arterial stenosis characteristics are present; otherwise, localized arterial stenosis characteristics are present.

[0078] As described above, the lesion features may include generalized stenosis of small arteries. In some examples, when measuring generalized stenosis of small arteries, arterial and venous segments within the third region C3 of the arteriovenous segmentation results can be obtained. In some examples, the arteriovenous ratio (AVR) can be obtained based on the arterial and venous segments and a preset formula. In some examples, generalized stenosis of small arteries can be measured based on the AVR. Thus, generalized stenosis of small arteries can be measured based on the arterial and venous segments within the third region C3. In some examples, the AVR can be the ratio of the central retinal artery equivalent (CRAE) diameter to the central retinal vein equivalent (CRVE) diameter.

[0079] In some examples, the preset formula can be a modified version of Knudtson's formula. This allows for the measurement of generalized stenosis characteristics of small arteries based on Knudtson's modified formula. Specifically, the process of measuring generalized stenosis characteristics of small arteries will be described using Knudtson's modified formula as an example. Generally, the arterial and venous segments within region C3 are independent vascular segments. In some examples, if bifurcated vascular segments exist, please refer to the relevant description of obtaining vascular segments in the section on measuring localized stenosis characteristics of small arteries above.

[0080] In some examples, the central retinal artery equivalent diameter (CRAE) can be obtained iteratively from the vessel diameters of the arterial segments within the third region C3. During calculation, the arterial segment with the largest diameter can be paired with the one with the smallest diameter, the second largest diameter with the second smallest diameter, and so on. The calculation can be iteratively performed according to equation (1) to calculate the central retinal artery equivalent diameter (CRAE). Here, w1 represents the smaller vessel diameter, and w2 represents the larger vessel diameter. Multiple intermediate values ​​w are generated and used as vessel diameters, iterating in the same manner until only one intermediate value w remains. This intermediate value w is then used as the central retinal artery equivalent diameter (CRAE). In some examples, if the number of arterial segments is odd, the diameter of a single arterial segment can be carried over to the next iteration.

[0081] In some examples, the equivalent diameter of the central retinal vein (CRVE) can be calculated based on the venous segment and by iterating according to equation (2): Where x1 is the smaller vessel diameter, x2 is the larger vessel diameter, and x is an intermediate value. For details, please refer to the description related to calculating the central retinal artery equivalent diameter (CRAE).

[0082] The equivalent diameters of the central retinal artery (CRAE) and central retinal vein (CRVE) can be obtained through the above steps. The formula for the arteriovenous vessel diameter ratio (AVR) is as follows: AVR = CRAE / CRVE (3). In some examples, if AVR is less than 2 / 3, there is a generalized stenosis of small arteries; otherwise, there is no generalized stenosis of small arteries.

[0083] Figure 11 This is a flowchart illustrating the measurement of blood vessel diameter as described in this disclosure example.

[0084] In some examples, such as Figure 11 As shown, measuring the aforementioned blood vessel diameter may include enhancing resolution (step S210), acquiring a continuous vascular skeleton and diameter measurement direction (step S220), generating a vascular contour (step S230), and calculating the blood vessel diameter (step S240). In this case, the resolution of blood vessels in the fundus image can be increased, and more pixels can be used to measure the blood vessel diameter. Therefore, automatic super-resolution measurement of blood vessel diameter is possible, and the accuracy of blood vessel diameter measurement can be improved.

[0085] Generally, the width (i.e., the number of pixels) used to measure the diameter of blood vessels in fundus images is a positive integer, such as 1, 3, 6, or 8. However, super-resolution technology based on resolution enhancement and preset precision can perform sub-pixel level measurements of blood vessel diameters, allowing the measurement width to be a decimal, such as 1.23, 3.12, 5.63, or 7.56. This enables more accurate measurements of blood vessel diameters.

[0086] In some examples, in step S210, the arteriovenous segmentation result can be a binary image; as described above, the arteriovenous segmentation result can be a ternary image. In some examples, the arteriovenous segmentation result can be converted into a binary image based on whether a pixel is a blood vessel for use in blood vessel diameter measurement, where blood vessel pixels are white and non-blood vessel pixels are black. However, the examples in this disclosure are not limited to this; in other examples, the blood vessel diameter can also be measured directly using a ternary image. In some examples, the arteriovenous segmentation result can be enhanced in resolution by a preset factor to generate an enhanced arteriovenous segmentation result. For example, in some examples, an arteriovenous segmentation result with a resolution of 140×63 can be enhanced by 10 times to generate an enhanced arteriovenous segmentation result with a resolution of 1400×630.

[0087] In some examples, linear interpolation methods can be used to enhance the resolution of arteriovenous segmentation results. However, the examples disclosed herein are not limited to this; in other examples, deep learning-based image super-resolution methods can be used to enhance the resolution of arteriovenous segmentation results. Furthermore, in some examples, the preset multiplier can be an integer greater than 1. This increases the resolution of blood vessels in fundus images, thereby improving the accuracy of subsequent measurements of vessel diameter. In some examples, the preset multiplier can be between 5 and 15. For example, the preset multiplier can be 5, 10, or 15, etc.

[0088] In some examples, in step S220, after obtaining the enhanced arteriovenous segmentation result generated in step S210, the vascular skeleton (also called the vascular centerline) can be extracted from the enhanced arteriovenous segmentation result. The vascular skeleton can be the midline of the blood vessel. In some examples, the vascular skeleton is composed of discrete pixels. In some examples, in step S220, a morphological thinning algorithm can be used to thin the enhanced arteriovenous segmentation result to extract the vascular skeleton. That is, the width of the blood vessel in the enhanced arteriovenous segmentation result is thinned to one pixel width towards the center of the blood vessel to form the vascular skeleton, while maintaining the basic topological structure of the blood vessel shape in the enhanced arteriovenous segmentation result. In some examples, median filtering can be performed on the enhanced arteriovenous segmentation result before extracting the vascular skeleton. This can remove possible bifurcations at the ends of the vascular skeleton. In some examples, the vascular skeleton can be fitted to obtain a continuous vascular skeleton. In some examples, a least-squares cubic spline interpolation algorithm can be used to fit the vascular skeleton. This can obtain a continuous vascular skeleton and a fitting equation.

[0089] Figure 12 This is a schematic diagram illustrating the pipe diameter measurement direction as described in the example of this disclosure.

[0090] Additionally, in some examples, in step S220, the third measurement pixel can be multiple pixels on a continuous vascular skeleton. In some examples, the diameter measurement direction can be perpendicular to the tangent line of the continuous vascular skeleton at the third measurement pixel. In some examples, the tangent line of the third measurement pixel can be obtained using the first derivative of the above fitting equation. Figure 12 As shown, the tangent line to the continuous vascular skeleton L1 of blood vessel A at the third measurement pixel point G1 can be the tangent line L2. The straight line passing through the third measurement pixel point G1 and perpendicular to the tangent line L2 can be the diameter measurement direction L3.

[0091] In some examples, in step S230, the vessel contour corresponding to the third measurement pixel can be generated based on the enhanced arteriovenous segmentation result obtained in step S210, the third measurement pixel obtained in step S220, the diameter measurement direction of the third measurement pixel obtained in step S220, and a preset precision. In some examples, an interpolation algorithm can be used to interpolate the enhanced arteriovenous segmentation result based on the enhanced arteriovenous segmentation result, the third measurement pixel, the diameter measurement direction of the third measurement pixel, and the preset precision to generate the vessel contour corresponding to the third measurement pixel. Thus, interpolation can be performed on the enhanced vessel image based on a preset precision to obtain the vessel contour. In some examples, the interpolation algorithm can be a cubic spline interpolation algorithm. Therefore, a cubic spline interpolation algorithm can be used to interpolate the enhanced arteriovenous segmentation result. Additionally, in some examples, the preset precision can be a decimal greater than 0 and less than 1. This increases the resolution of vessels in the fundus image, thereby improving the accuracy of subsequent vessel diameter measurements. In some examples, the preset precision can be between 0.01 and 0.10. For example, the preset precision can be 0.01, 0.05, or 0.10, etc.

[0092] The following section, with reference to the accompanying drawings, details the process of generating the blood vessel contour. Figure 13 This is a flowchart illustrating the blood vessel contour generation involved in the examples of this disclosure. (See attached diagram.) Figure 13 As shown, in some examples, the process of generating the blood vessel contour in step S230 may include obtaining the width of the blood vessel contour (step S231), obtaining the interpolation sampling interval (step S232), generating interpolation points (step S233), performing interpolation on the arteriovenous segmentation results to determine the pixel values ​​of the blood vessel contour (step S234), and generating the blood vessel contour based on the pixel values ​​of the blood vessel contour (step S235).

[0093] In some examples, in step S231, the width of the vessel contour can be N times the maximum vessel diameter. Here, N can be an integer between 2 and 5. In some examples, the width of the vessel contour can be twice the maximum vessel diameter. In some examples, the maximum vessel diameter can be an estimated maximum vessel diameter; for details, please refer to the description of the maximum vessel diameter in step S1413. Therefore, the widest vessel in the enhanced arteriovenous segmentation result can be fully presented in the vessel contour.

[0094] In some examples, in step S232, the interpolation sampling interval can be obtained based on the width of the blood vessel contour and a preset precision. Specifically, assuming the width of the blood vessel contour can be represented by wi and the preset precision by s, the value of the interpolation sampling interval can be between -(wi-1) / 2 and (wi-1) / 2, increasing in steps with the preset precision s. For example, the value of the interpolation sampling interval can be -(wi-1) / 2, (wi-1) / 2+s, (wi-1) / 2+2×s, or (wi-1) / 2, etc.

[0095] In some examples, in step S233, interpolation points can be generated based on the interpolation sampling interval, the diameter measurement direction, and the continuous vascular skeleton obtained in step S232. Specifically, assuming the interpolation sampling interval is represented by inc, the diameter measurement direction can be represented by (dx, dy), and the continuous vascular skeleton can be represented by (x, y), then the interpolation point can be represented as (x + dx × inc, y + dy × inc). In this case, the generated interpolation points are distributed along the diameter measurement direction. This increases the number of pixels near the diameter measurement direction, thereby improving the accuracy of vascular diameter measurement.

[0096] In some examples, in step S234, an interpolation algorithm can be used to interpolate the arteriovenous segmentation results based on the interpolation points obtained in step S233 to determine the pixel values ​​of the vessel contour. In some examples, the interpolation algorithm can be a cubic spline interpolation algorithm. Therefore, a cubic spline interpolation algorithm can be used to interpolate the enhanced arteriovenous segmentation results.

[0097] Figure 14 This is a schematic diagram illustrating an image of a straightened blood vessel as described in the examples of this disclosure.

[0098] In some examples, in step S235, the blood vessel contour corresponding to each third measurement pixel can be output based on the pixel values ​​of the blood vessel contour. In some examples, the blood vessel contours corresponding to each third measurement pixel can be arranged side by side in a collinear manner according to the arrangement order of the third measurement pixels on the continuous blood vessel skeleton to form a straightened blood vessel image. As an example of a straightened blood vessel image, Figure 14An image of a straightened blood vessel is shown. Here, straightened blood vessel A' represents the straightened blood vessel. This allows for convenient acquisition of the blood vessel contour corresponding to each third measurement pixel.

[0099] In some examples, in step S240 of measuring the vessel diameter, the vessel diameter corresponding to the third measured pixel can be calculated based on the number of vessel pixels in the vessel contour corresponding to the third measured pixel, a preset multiple, and a preset precision. The vessel contour corresponding to the third measured pixel can be obtained in step S230. In some examples, the number n of vessel pixels in the vessel contour corresponding to the third measured pixel can be calculated using the following formula: n = card({p: p∈P, f(p)>T}), where p is the third measured pixel, f(p) is the pixel value corresponding to the third measured pixel p, T is a preset threshold parameter, P is the set of pixels in the vessel contour corresponding to the third measured pixel p, and card represents the cardinality of the set. Thus, the number of vessel pixels in the vessel contour corresponding to the third measured pixel can be calculated. In some examples, T can be 0.9. Therefore, it is possible to distinguish between vessel pixels and non-vessel pixels in the vessel contour corresponding to the third measured pixel. In some examples, the blood vessel diameter *l* corresponding to the third measured pixel can satisfy: *l* = *n* × *s* / *e*, where *n* is the number of blood vessel pixels in the blood vessel contour corresponding to the third measured pixel, *s* is the preset precision, and *e* is the preset multiplier. Therefore, the blood vessel diameter can be calculated.

[0100] As mentioned above, the preset multiplier can be an integer greater than 1. The preset precision can be a decimal greater than 0 and less than 1. In some examples, different preset multipliers e and preset precisions s can be used to measure the diameter of the same blood vessel segment to obtain a more accurate measurement result. In some examples, the measurement results of the same blood vessel segment are shown in Table 1, where the mean is the average blood vessel diameter and the standard deviation is the standard deviation of the blood vessel diameter.

[0101] Table 1. Partial results of blood vessel diameter measurement comparison

[0102] Measurement parameters mean Standard deviation Manual annotation 6.4094 0.421 e = 1, s = 1 5.9365 0.7319 e = 5, s = 0.5 6.4457 0.4737 e = 5, s = 0.1 6.435 0.4723 e = 5, s = 0.05 6.4341 0.4707 e = 5, s = 0.01 6.4345 0.471 e = 10, s = 0.5 6.5142 0.4799 e = 10, s = 0.1 6.5078 0.4791 e = 10, s = 0.05 6.5069 0.4787 e = 10, s = 0.01 6.507 0.4788 e = 15, s = 0.5 6.5341 0.4757 e = 15, s = 0.1 6.5293 0.4757 e = 15, s = 0.05 6.5288 0.4754 e = 15, s = 0.01 6.5289 0.4755

[0103] As shown in Table 1, the mean blood vessel diameter measured by this invention is 6.4094, with a standard deviation of 0.421. Without super-resolution technology (e=1, s=1), the mean blood vessel diameter measured by this invention is 5.9365, with a standard deviation of 0.7319, which differs significantly from the manually labeled result. However, when using super-resolution technology (e∈{5,10,15} and s∈{0.01,0.05,0.1,0.5}), both the mean and standard deviation of the diameter measurement are close to the manually labeled result. Specifically, when e=5 and s=0.05, the mean blood vessel diameter is 6.4341, with a standard deviation of 0.4707, comparable to the manually labeled blood vessel diameter. Therefore, the measurement results using super-resolution technology are closer to the manually labeled results, indicating that super-resolution technology effectively improves the accuracy of blood vessel diameter measurement.

[0104] In some examples, arteriovenous crossing impressions, localized arteriolar stenosis, and generalized arteriolar stenosis can reflect hypertension. For instance, localized arteriolar stenosis may indicate a recent increase in blood pressure. Conversely, arteriovenous crossing impressions and generalized arteriolar stenosis may indicate a long-term, persistent increase in blood pressure. The measurement methods based on this disclosure can assist ophthalmologists in identifying hypertension.

[0105] The measurement system 200 (sometimes simply referred to as measurement system 200) for measuring the pathological characteristics of hypertensive retinopathy disclosed herein will be described in detail below with reference to the accompanying drawings. The measurement system 200 disclosed herein is used to implement the measurement method described above. Figure 15 This is a block diagram illustrating a measurement system 200 for the pathological characteristics of hypertensive retinopathy as described in this disclosure example.

[0106] like Figure 15 As shown, in some examples, the measurement system 200 may include an acquisition module 210, a partitioning module 220, a segmentation module 230, and a measurement module 240. The acquisition module 210 can be used to acquire fundus images. The partitioning module 220 can be used to partition the fundus image to obtain at least three regions. The segmentation module 230 can be used to perform arteriovenous segmentation on the fundus image to obtain arteriovenous segmentation results. The measurement module 240 can measure the lesion features in the fundus image based on the three regions and the arteriovenous segmentation results. In this case, the lesion features of hypertensive retinopathy in the fundus image can be automatically measured. Therefore, the lesion features of hypertensive retinopathy can be measured efficiently and objectively.

[0107] In some examples, the acquisition module 210 can be used to acquire fundus images. The fundus images can include blood vessels, and the blood vessels can include arteries and veins. For specific descriptions, reference can be made to the relevant descriptions in step S110, which will not be elaborated here.

[0108] In some examples, in the partitioning module 220, the three regions can include a first region, a second region, and a third region. Thus, the three regions can be acquired based on the optic disc region. In some examples, the first region can be a region of a first circle formed with the center of the circumcircle of the optic disc region as the center and the diameter of the circumcircle multiplied by a first preset multiple as the diameter. In some examples, the first preset multiple can be represented by v1. In some examples, v1 can be 1. In some examples, the second region can be a region between the edge of the first region and a second circle formed with the center of the circumcircle as the center and the diameter of the circumcircle multiplied by a second preset multiple as the diameter. In some examples, the second preset multiple can be represented by v2. In some examples, v2 can be 2. In some examples, the third region can be a region between the edge of the second region and a third circle formed with the center of the circumcircle as the center and the diameter of the circumcircle multiplied by a third preset multiple as the diameter. In some examples, the third preset multiple can be represented by v3. In some examples, v3 can be 3. In some examples, it can be set that v1 < v2 < v3. For specific descriptions, reference can be made to the relevant descriptions in step S120, which will not be elaborated here.

[0109] In some examples, the segmentation module 230 can utilize a deep learning-based arteriovenous segmentation model to segment arteries and veins in fundus images to obtain arteriovenous segmentation results. In some examples, the arteriovenous segmentation model can be trained based on training fundus images and arteriovenous vessel labeling results. In some examples, the arteriovenous segmentation results can include both artery and vein segmentation results. In some examples, arteries and veins in the fundus image can be segmented to directly obtain arteriovenous segmentation results. In some examples, the arteriovenous vessel labeling results can include both artery and vein labeling results. In some examples, the arteriovenous vessel labeling results can be formed by labeling vessels in the training fundus image. In some examples, the arteriovenous vessel labeling results can be formed by labeling the boundaries of vessels with a diameter larger than a preset vessel diameter in the training fundus image to form artery and vein labeling results, and by labeling the course of small vessels to form small vessel labeling results. In some examples, when calculating the loss function, the weights of the regions corresponding to the small vessel labeling results can be adjusted based on the small vessel labeling results. In this scenario, marking the boundaries of larger blood vessels and the orientation of smaller blood vessels in the training fundus images effectively reduces annotation time, lowers annotation costs and difficulty, and controls the contribution of small vessel orientation-only marking results to the loss function by adjusting the weights of the regions corresponding to the small vessel marking results, thereby improving the accuracy of arteriovenous segmentation. In some examples, the preset blood vessel diameter can be 50 μm. In some examples, small blood vessels can be blood vessels no larger than the preset blood vessel diameter. In some examples, the weights of the regions corresponding to the small vessel marking results can be adjusted to 0 when calculating the loss function. In this case, the contribution of small vessel orientation-only marking results to the loss function can be excluded. Thus, the influence of small vessel marking results with low accuracy for arteriovenous segmentation on the arteriovenous segmentation model can be avoided. For a detailed description, please refer to the relevant description in step S130, which will not be repeated here.

[0110] In some examples, the measurement module 240 can measure the lesion features of hypertensive retinopathy in fundus images based on three regions and arteriovenous segmentation results. In some examples, the lesion features may include at least one of arteriovenous crossing impression features, localized arteriovenous stenosis features, and generalized arteriovenous stenosis features.

[0111] In some examples, the arteriovenous crossing impression features can be measured in the measurement module 240. In some examples, the arteriovenous segmentation results of the fundus region other than the first and second regions can be refined to obtain a first vascular skeleton. In some examples, the first vascular skeleton may include multiple skeleton pixels serving as first measurement pixels. In some examples, the number of skeleton pixels within a preset range for each first measurement pixel can be obtained and used as the number of first adjacent points. In some examples, pixels in the arteriovenous segmentation results corresponding to first measurement pixels with a number of first adjacent points greater than a first preset number can be used as arteriovenous crossing locations. In some examples, the arterial segmentation results can be dilated so that the dilated arterial segmentation results intersect with the vein segmentation results to determine the arteriovenous crossing locations. This allows for obtaining more accurate arteriovenous crossing locations. In some examples, the arteriovenous crossing impression features can be measured based on the ratio of the proximal and distal vessel diameters of the vein segmentation results on both sides of the arteriovenous crossing location. In some examples, the proximal and distal ends on both sides of the arteriovenous crossing location can be the proximal and distal ends of each side of the arteriovenous segmentation result along the extension direction of the vein segmentation result and located on both sides of the arteriovenous crossing location. In some examples, if the vein segmentation result in the arteriovenous segmentation result is discontinuous at the arteriovenous crossing location, the proximal end on each side can be the skeleton pixel point on the first vascular skeleton of the vein segmentation result that is closest to the arteriovenous crossing location. In some examples, if the vein segmentation result in the arteriovenous segmentation result is continuous at the arteriovenous crossing location, the proximal end on each side can be the arteriovenous crossing location. In some examples, the distal end on each side can be the skeleton pixel point on the first vascular skeleton of the vein segmentation result that is at a distance of a first preset distance from the arteriovenous crossing location. Thus, the proximal and distal ends on both sides of the vein segmentation result can be determined based on the arteriovenous crossing location. In some examples, the first preset distance can be 2 to 4 times the maximum vessel diameter. In some examples, the first preset distance can be 3 times the maximum vessel diameter. For a detailed description, please refer to the relevant description of measuring the arteriovenous crossing impression features in step S140, which will not be repeated here.

[0112] In some examples, the local stenosis features of small arteries can be measured in measurement module 240. In some examples, the artery segmentation results can be refined to obtain a second vascular skeleton. In some examples, the second vascular skeleton may include multiple skeleton pixels serving as second measurement pixels. In some examples, the number of skeleton pixels within a preset range for each second measurement pixel can be obtained as the number of second neighboring points. In some examples, second measurement pixels with a number of second neighboring points greater than a second preset number can be deleted to obtain multiple vascular segments. In some examples, the local stenosis features of small arteries can be measured based on the ratio of the minimum to the maximum vascular diameter of each vascular segment. For a detailed description, please refer to the relevant description of measuring the local stenosis features of small arteries in step S140, which will not be repeated here.

[0113] In some examples, the generalized stenosis characteristics of small arteries can be measured in measurement module 240. In some examples, arterial and venous segments within the third region of the arteriovenous segmentation result can be obtained. In some examples, the arteriovenous diameter ratio can be obtained based on the arterial and venous segments and a preset formula to measure the generalized stenosis characteristics of small arteries. Thus, the generalized stenosis characteristics of small arteries can be measured based on the arterial and venous segments within the third region. In some examples, the arteriovenous diameter ratio can be the ratio of the equivalent diameter of the central retinal artery to the equivalent diameter of the central retinal vein. In some examples, the preset formula can be a modified version of Knudtson's formula. Thus, the generalized stenosis characteristics of small arteries can be measured based on Knudtson's modified formula. For a detailed description, please refer to the relevant description of measuring the generalized stenosis characteristics of small arteries in step S140, which will not be repeated here.

[0114] In some examples, the measurement system 200 may also include a diameter calculation module (not shown). The diameter calculation module can be used to measure the diameter of blood vessels. In some examples, the arteriovenous segmentation results can be enhanced by a preset resolution to generate enhanced arteriovenous segmentation results. In some examples, the vascular skeleton in the enhanced arteriovenous segmentation results can be extracted and fitted to obtain a continuous vascular skeleton and the diameter measurement direction of the third measurement pixel. In some examples, the third measurement pixel can be multiple pixels on the continuous vascular skeleton. In some examples, the diameter measurement direction can be perpendicular to the tangent of the continuous vascular skeleton at the third measurement pixel. In some examples, an interpolation algorithm can be used to generate the vascular contour corresponding to the third measurement pixel based on the enhanced arteriovenous segmentation results, the third measurement pixel, the diameter measurement direction of the third measurement pixel, and a preset precision. In some examples, the vascular diameter corresponding to the third measurement pixel can be calculated based on the number of vascular pixels in the vascular contour corresponding to the third measurement pixel, a preset resolution, and a preset precision. In this case, the resolution of blood vessels in the fundus image can be increased, and more pixels can be used to measure the vascular diameter. Therefore, automatic super-resolution measurement of blood vessel diameter can be performed, improving the accuracy of blood vessel diameter measurement. In some examples, the blood vessel diameter l corresponding to the third measurement pixel can satisfy: l = n × s / e, where n is the number of blood vessel pixels in the blood vessel contour corresponding to the third measurement pixel, s is the preset precision, and e is the preset multiplier. For a detailed description, please refer to the relevant descriptions in steps S210 to S240, which will not be repeated here. However, the examples disclosed herein are not limited to this; in other examples, other methods can be used to calculate the blood vessel diameter.

[0115] While the present disclosure has been specifically described above in conjunction with the accompanying drawings and examples, it is to be understood that the foregoing description does not limit the present disclosure in any way. Those skilled in the art can make modifications and variations to the present disclosure as needed without departing from its essential spirit and scope, and all such modifications and variations shall fall within the scope of the present disclosure.

Claims

1. A system for measuring the pathological characteristics of hypertensive retinopathy, characterized in that, It includes an acquisition module, a partitioning module, a segmentation module, and a measurement module; The acquisition module is used to acquire fundus images; The partitioning module is used to receive the fundus image, identify the optic disc region of the fundus image, and divide the fundus image into at least three regions based on the optic disc region; The segmentation module is used to segment the fundus image into arteries and veins using a deep learning-based arteriovenous segmentation model trained on the fundus image and the arteriovenous vessel labeling results, so as to directly obtain a three-valued image of the arteriovenous segmentation results. The arteriovenous vessel labeling results include arterial labeling results and vein labeling results formed by labeling the boundaries of blood vessels with a diameter greater than a preset blood vessel diameter in the fundus image, and small blood vessel labeling results formed by labeling the direction of blood vessels with a diameter not greater than the preset blood vessel diameter. The direction is a curve following the direction of the blood vessel and is used to estimate the region corresponding to the blood vessel with a diameter not greater than the preset blood vessel diameter. The three-valued image includes three gray values ​​representing arteries, veins and background respectively. When calculating the loss function, the weight of the region corresponding to the small blood vessel labeling results is adjusted to zero to exclude the contribution of the small blood vessel labeling results that only label the direction to the loss function. The measurement module is used to measure the lesion characteristics based on the three regions and the vessels larger than the preset vessel diameter in the arteriovenous segmentation results.

2. The measurement system according to claim 1, characterized in that: The pathological features include at least one of the following: arteriovenous crossing impression features, localized arteriolar stenosis features, and generalized arteriolar stenosis features.

3. The measurement system according to claim 2, characterized in that: The three regions include a first region, a second region, and a third region; If the arteriovenous crossing impression features are measured, the arteriovenous segmentation results of the fundus regions other than the first and second regions are refined to obtain a first vascular skeleton including multiple skeleton pixels as first measurement pixels. The number of skeleton pixels within a preset range of each first measurement pixel is obtained and used as the number of first adjacent points. The pixels in the arteriovenous segmentation results corresponding to the first measurement pixels with a number of first adjacent points greater than a first preset number are taken as arteriovenous crossing positions. The arteriovenous crossing impression features are measured based on the ratio of the proximal and distal vessel diameters on each side of the arteriovenous crossing position along the extension direction of the vein segmentation results in the arteriovenous segmentation results; and / or If the local stenosis characteristics of the small artery are measured, the arterial segmentation result of the arteriovenous segmentation result is refined to obtain a second vascular skeleton including multiple skeleton pixels as second measurement pixels. The number of skeleton pixels within a preset range of each second measurement pixel is obtained as the number of second neighboring points. Second measurement pixels with a number of second neighboring points greater than the second preset number are deleted to obtain multiple vascular segments. The local stenosis characteristics of the small artery are measured based on the ratio of the minimum vascular diameter to the maximum vascular diameter of each vascular segment; and / or If the general stenosis characteristics of the small arteries are measured, the arterial and venous segments in the third region of the arteriovenous segmentation result are obtained, and the arteriovenous diameter ratio is obtained based on the arterial segment, the venous segment and a preset formula to measure the general stenosis characteristics of the small arteries.

4. The measurement system according to claim 3, characterized in that: The arteriovenous vessel diameter ratio is the ratio of the equivalent diameter of the central retinal artery to the equivalent diameter of the central retinal vein.

5. The measurement system according to claim 4, characterized in that, The method for calculating the equivalent diameter of the central retinal artery is as follows: Substitute the arterial segment with the largest diameter and the arterial segment with the smallest diameter into the calculation formula. From which, For smaller blood vessel diameters, It has a relatively large blood vessel diameter.

6. The measurement system according to claim 4, characterized in that, The method for calculating the equivalent diameter of the central retinal vein is as follows: Substitute the venous segment with the largest diameter and the venous segment with the smallest diameter into the calculation formula. From which, For smaller blood vessel diameters, It has a relatively large blood vessel diameter.

7. The measurement system according to claim 3, characterized in that: The preset range of the first measurement pixel is eight neighborhoods around the first measurement pixel, and the preset range of the second measurement pixel is eight neighborhoods around the second measurement pixel.

8. The measurement system according to claim 3, characterized in that: The refined algorithms include the Hilditch algorithm, the Pavlidis algorithm, or the Rosenfeld algorithm.

9. The measurement system according to claim 1, characterized in that: The preset blood vessel diameter is 50 μm.

10. A method for fundus image segmentation in hypertensive retinopathy, characterized in that, include: Acquire fundus images; The optic disc region of the fundus image is identified, and the fundus image is divided into at least three regions based on the optic disc region; A deep learning-based arteriovenous segmentation model, trained on fundus images and arteriovenous vessel labeling results, is used to segment the fundus images into arteries and veins to directly obtain a ternary image of the arteriovenous segmentation results. The arteriovenous vessel labeling results include arterial and vein labeling results formed by labeling the boundaries of vessels with diameters larger than a preset diameter in the trained fundus images, and small vessel labeling results formed by labeling the direction of vessels with diameters not larger than the preset diameter. The direction is a curve following the vessel's direction and is used to estimate the region corresponding to vessels with diameters not larger than the preset diameter. The ternary image includes three grayscale values ​​representing arteries, veins, and the background, respectively. When calculating the loss function, the weight of the region corresponding to the small vessel labeling results is adjusted to zero to exclude the contribution of the small vessel labeling results (which only label the direction) to the loss function.

Images