Image-based arteriovenous vessel separation method and device, electronic equipment and medium

Abstract

The present application relates to artificial intelligence and digital medical treatment, and provides an arteriovenous vessel separation method and device based on images, an electronic device and a medium, which extracts a blood vessel skeleton in a CT image; divides the blood vessel into a plurality of blood vessel segments according to the blood vessel skeleton; constructs a blood vessel topology graph based on the plurality of blood vessel segments; extracts a plurality of features of each blood vessel segment; trains an arteriovenous vessel separation model based on the blood vessel topology graph and the plurality of features of each blood vessel segment; and inputs a CT image to be processed into the trained arteriovenous vessel separation model to separate arteriovenous vessels. The present application can improve the accuracy of arteriovenous vessel separation.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to an image-based method, apparatus, electronic device, and medium for separating arteries and veins. Background Technology

[0002] Fundus photography is a non-invasive imaging method for observing the state of microvessels in the human eye. It can not only visually reflect fundus lesions such as hemorrhage and hard exudates, but also observe morphological changes in anatomical structures such as the width and tortuosity of fundus vessels. With the development of medical imaging technology, a series of deep learning-based fundus arteriovenous segmentation techniques have emerged using fundus photography images.

[0003] Previous studies have often used fully convolutional networks (FCN), U-net, etc. to segment arteries and veins in fundus images, but their disadvantages are: (1) It is difficult to distinguish between arteries and veins, and it is easy to missegment the same blood vessel, resulting in inconsistent arteries and veins (a section of blood vessel is partly vein and partly artery). (2) It is easy to miss some thin blood vessels. (3) The generalization ability is weak for images from different devices. Summary of the Invention

[0004] In view of the above, it is necessary to propose an image-based method, device, electronic device and medium for arteriovenous vessel separation, which can improve the accuracy of image-based arteriovenous vessel separation.

[0005] A first aspect of the present invention provides an image-based method for separating arteries and veins, the method comprising:

[0006] Extracting the vascular skeleton from CT images;

[0007] The blood vessel is divided into multiple blood vessel segments according to the vascular skeleton;

[0008] A vascular topology diagram is constructed based on the multiple vascular segments;

[0009] Extract multiple features from each of the vascular segments;

[0010] An arteriovenous vessel separation model is obtained by training based on the aforementioned vascular topology diagram and multiple features of each vascular segment;

[0011] The CT image to be processed is input into the trained arteriovenous vessel separation model for arteriovenous vessel separation.

[0012] According to an optional embodiment of the present invention, the extraction of the vascular skeleton from the CT image includes:

[0013] The CT image is segmented using a preset blood vessel segmentation model to obtain a binarized blood vessel mask image;

[0014] The blood vessel mask image is refined to obtain the blood vessel skeleton.

[0015] According to an optional embodiment of the present invention, refining the vascular mask image to obtain the vascular skeleton includes:

[0016] The blood vessel mask image is filtered to obtain a filtered image;

[0017] The initial vascular skeleton is extracted from the filtered image using a thinning algorithm;

[0018] The initial vascular skeleton is fitted to obtain a continuous vascular skeleton;

[0019] The continuous vascular skeleton is obtained by performing single-pixel processing.

[0020] According to an optional embodiment of the present invention, the step of dividing the blood vessel into multiple blood vessel segments based on the vascular skeleton includes:

[0021] Obtain the vascular branch points in the vascular skeleton;

[0022] Using the vascular branch points as segmentation points, the vascular skeleton is segmented to obtain the multiple vascular segments; or

[0023] The vascular branch points are deleted to split the vascular skeleton into multiple branches, and each branch is identified as a vascular segment to obtain the multiple vascular segments.

[0024] According to an optional embodiment of the present invention, the extraction of multiple features from each of the vascular segments includes:

[0025] Obtain the tight bounding box for each of the aforementioned vascular segments;

[0026] Obtain the feature map output by the preset blood vessel segmentation model, and extract the first feature corresponding to the tight bounding box from the feature map;

[0027] The first feature is normalized to obtain the normalized feature;

[0028] For each of the blood vessel segments, multiple preset feature extraction models are used to calculate multiple second features based on the tight bounding boxes corresponding to the blood vessel segments in the feature map.

[0029] According to an optional embodiment of the present invention, the process of training the arteriovenous vessel separation model based on the vascular topology map and multiple features of each vascular segment includes:

[0030] The overall features are obtained based on the normalized features corresponding to each of the blood vessel segments and the plurality of second features;

[0031] The vascular topology map and multiple overall features are input into a preset neural network, and the predicted label of each vascular segment output by the preset neural network is obtained.

[0032] Calculate the first loss function value based on the preset label and the corresponding real label;

[0033] The second loss function value is calculated based on the gold standard image corresponding to the vascular mask image and the CT image;

[0034] The gradient descent algorithm is used to train the arteriovenous vessel separation model and the preset vessel segmentation model based on the first loss function value and the second loss function value, so as to obtain the trained arteriovenous vessel separation model and vessel segmentation model.

[0035] According to an optional embodiment of the present invention, after obtaining the arterial and venous vessels, the method further includes:

[0036] Obtain the largest and smallest arterial vessels;

[0037] The equivalent value of the central retinal artery diameter is calculated based on the diameter of the largest artery and the diameter of the smallest artery.

[0038] Obtain the largest and smallest veins;

[0039] The equivalent value of the central retinal vein diameter is calculated based on the diameter of the largest vein and the diameter of the smallest vein.

[0040] The quantitative values ​​of arterial and venous vessel diameters are calculated based on the equivalent values ​​of the central retinal artery diameter and the central retinal vein diameter.

[0041] A second aspect of the present invention provides an image-based arteriovenous vessel separation device, the device comprising:

[0042] The extraction module is used to extract the vascular skeleton from CT images;

[0043] A segmentation module is used to segment the blood vessel into multiple blood vessel segments based on the vascular skeleton;

[0044] A construction module is used to construct a vascular topology diagram based on the multiple vascular segments;

[0045] A calculation module is used to extract multiple features from each of the vascular segments;

[0046] The training module is used to train an arteriovenous vessel separation model based on the vascular topology diagram and multiple features of each vascular segment.

[0047] The separation module is used to input the CT image to be processed into the trained arteriovenous vessel separation model for arteriovenous vessel separation.

[0048] A third aspect of the present invention provides an electronic device comprising a processor and a memory, the processor being configured to implement the image-based arteriovenous vessel separation method when executing a computer program stored in the memory.

[0049] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the image-based arteriovenous vessel separation method.

[0050] The image-based arteriovenous vessel separation method, device, electronic device, and medium provided by this invention combine deep neural networks and graph neural networks (GCN) for joint training, which not only makes the classification of blood vessels more accurate, but also makes the classification between arteries and veins simpler and more precise. Classification based on the binary mask image of the entire blood vessel can avoid the problem of misclassification of the same blood vessel leading to inconsistency between arteries and veins, that is, a segment of blood vessel is partly vein and partly artery, ensuring the continuity of arterial and venous blood vessels. Secondly, it can avoid missing some thin blood vessels. Attached Figure Description

[0051] Figure 1 This is a flowchart of an image-based arteriovenous vessel separation method provided in Embodiment 1 of the present invention.

[0052] Figure 2 This is a structural diagram of the image-based arteriovenous vessel separation device provided in Embodiment 2 of the present invention.

[0053] Figure 3 This is a schematic diagram of the structure of the electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0054] To better understand the above-mentioned objects, features, and advantages of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing an embodiment in one alternative implementation and is not intended to be limiting of the invention.

[0056] The image-based arteriovenous vessel separation method provided in this embodiment of the invention is executed by an electronic device, and correspondingly, the image-based arteriovenous vessel separation device operates in the electronic device.

[0057] The embodiments of this invention can standardize data processing based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0058] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0059] Arteriovenous separation offers unique advantages in various medical scenarios and under diverse disease or pathological conditions. In disease research, characteristics such as vascular density, structure, and size can be used to assess disease, and pulmonary vascular diseases may specifically affect arteries or veins through different physiological mechanisms. Therefore, studying arteries and veins can reveal the complex anatomical relationship between lung lesions and the vascular system, which is beneficial for improving the accuracy of lung disease diagnosis. Arteriovenous separation helps in the early screening and tracking of lung diseases, allowing doctors to diagnose and monitor patients' pathological states. It also provides effective references for preoperative planning, intraoperative navigation, and postoperative evaluation in lung surgery, which has significant clinical implications. The fundus is the only part of the body where arteries, veins, and capillaries can be directly observed with the naked eye. Numerous long-term follow-up studies have shown that quantitative indicators of fundus arteriovenous vessels (central arteriovenous equivalent, arteriovenous fractal dimension, arteriovenous ratio, etc.) are significantly correlated with systemic chronic diseases (hypertension, diabetes, cardiovascular and cerebrovascular diseases, etc.). The segmentation and extraction technology of fundus arteriovenous vessels is the cornerstone for achieving automated quantification of fundus arteriovenous vessels. Therefore, an effective arteriovenous extraction method is an urgent need in the field of medical imaging.

[0060] Example 1

[0061] Figure 1This is a flowchart of an image-based arteriovenous vessel separation method provided in an embodiment of the present invention. The image-based arteriovenous vessel separation method specifically includes the following steps. Depending on different requirements, the order of the steps in this flowchart can be changed, and some steps can be omitted.

[0062] S11, extract the vascular skeleton from the CT image.

[0063] The CT images are either fundus images obtained by scanning the patient's fundus using computed tomography (CT) technology, or lung images obtained by scanning the patient's lungs using computed tomography (CT) technology. Eye diseases can be identified by separating retinal arteries and veins in fundus CT images, or lung diseases can be identified by separating pulmonary arteries and veins in lung CT images.

[0064] The CT images can be obtained by electronic devices from a digital medical database. This digital medical database can be a digital repository storing patient medical records in a hospital, or it can be a networked database of multiple hospitals; this invention does not impose any limitations.

[0065] Electronic devices can acquire CT images of multiple patients, with each patient corresponding to one or more CT images. An arteriovenous vessel separation model is trained based on these acquired CT images, and then used to perform vessel segmentation, i.e., arteriovenous separation.

[0066] In an optional implementation, extracting the vascular skeleton from the CT image includes:

[0067] The CT image is segmented using a preset blood vessel segmentation model to obtain a binarized blood vessel mask image;

[0068] The blood vessel mask image is refined to obtain the blood vessel skeleton.

[0069] The preset blood vessel segmentation model can be a model trained using deep learning networks, such as U-net or Fully Convolutional Networks (FCN). The process of creating the blood vessel segmentation model is existing technology and will not be described in detail here.

[0070] Electronic devices use a preset vascular segmentation model to segment CT images, which means separating blood vessels from the background in the CT image without distinguishing between arteries and veins. The grayscale value of pixels in the binarized vascular mask image is 0 or 255, i.e., a black and white image. In the binarized vascular mask image, white areas represent blood vessels, and black areas represent the surrounding background. This implementation method reduces the amount of data in the CT image by obtaining a binarized vascular mask image, thereby facilitating the extraction of the vascular skeleton, i.e., highlighting the contours of the blood vessels.

[0071] In an optional implementation, refining the blood vessel mask image to obtain the blood vessel skeleton includes:

[0072] The blood vessel mask image is filtered to obtain a filtered image;

[0073] The initial vascular skeleton is extracted from the filtered image using a thinning algorithm;

[0074] The initial vascular skeleton is fitted to obtain a continuous vascular skeleton;

[0075] The continuous vascular skeleton is obtained by performing single-pixel processing.

[0076] The vascular skeleton is a topological description of the geometric features of blood vessels. It reflects the connectivity, structural information, and orientation of the vessels. The vascular skeleton is often located at the center of the vessel and is also called the vascular centerline. The vascular skeleton can be understood as the central axis of the vessel.

[0077] The electronic device can perform median filtering on the vascular mask image, which removes possible bifurcations at the ends of the vascular skeleton in the vascular mask image. Median filtering is a non-linear smoothing technique that sets the gray value of each pixel to the median of the gray values ​​of all pixels within a certain neighborhood window of that pixel.

[0078] A thinning algorithm is used to extract the vascular skeleton from the filtered vascular mask image, thereby removing redundant boundary points and retaining important image nodes such as junctions, endpoints, and isolated points. Electronic devices can employ morphological thinning algorithms to thin the vascular mask image to extract the vascular skeleton, which is then called the initial vascular skeleton. 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. Morphological thinning algorithms are existing technologies and will not be described in detail here.

[0079] Since the initial vascular skeleton is composed of discrete pixels, the electronic device fits the initial vascular skeleton to obtain a continuous vascular skeleton. In some implementations, a least-squares cubic spline interpolation algorithm can be used to fit the initial vascular skeleton.

[0080] Furthermore, since morphological thinning algorithms cannot guarantee that the extracted vascular skeleton is a single pixel, and non-single-pixel skeletons are not conducive to measuring blood vessel diameter, the initial vascular skeleton extracted by the morphological thinning algorithm needs to be further processed into single pixels. In specific implementation, the width of the continuous vascular skeleton is thinned to one pixel width towards the center of the blood vessel to form a single-pixel vascular skeleton, while maintaining the basic topological structure of the blood vessel shape of the single-pixel vascular skeleton.

[0081] S12, the blood vessel is divided into multiple blood vessel segments according to the vascular skeleton.

[0082] The entire vascular skeleton is not conducive to the separation of arteries and veins. Therefore, after extracting the target vascular skeleton, the electronic device segments the target vascular skeleton to obtain multiple vascular segments. Each vascular segment is either a vein or an artery, which facilitates the separation of arteries and veins.

[0083] After extracting the vascular skeleton, the electronic device can use a point-capturing tool to determine the intersections of the vascular skeleton and segment the blood vessel into multiple vascular segments based on these intersections, thereby obtaining multiple vascular segment images. These intersections can be intersections between arteries, branching points of arteries, intersections between veins, branching points of veins, or intersections between arteries and veins; this application does not specifically limit the type of intersection.

[0084] In an optional implementation, dividing the blood vessel into multiple vascular segments based on the vascular skeleton includes:

[0085] Obtain the vascular branch points in the vascular skeleton;

[0086] Using the vascular branch points as segmentation points, the vascular skeleton is segmented to obtain the multiple vascular segments; or

[0087] The vascular branch points are deleted to split the vascular skeleton into multiple branches, and each branch is identified as a vascular segment to obtain the multiple vascular segments.

[0088] Pixels located at branch points share a common characteristic: they must have three neighboring pixels within their eight-neighborhood. Based on this characteristic, eight-neighborhood filtering can be used to detect vascular branch points in the vascular skeleton. The electronic device performs eight-neighborhood filtering on the image corresponding to the vascular skeleton, then calculates the number of eight neighbors for each pixel. Pixels with three eight-neighborhoods are identified as suspicious points, and their pixel values ​​are obtained. The determination of whether a suspicious point is a vascular branch point is based on the obtained pixel value. Specifically, if the obtained pixel value is 0, it indicates that the suspicious point is a background point, and therefore it is not a vascular branch point; if the obtained pixel value is 1, it indicates that the suspicious point is the center pixel, and therefore it is a vascular branch point.

[0089] S13, Construct a vascular topology diagram based on the multiple vascular segments.

[0090] In this embodiment, each blood vessel segment is treated as a node. If blood vessel segments are connected to each other, an undirected edge is established between the corresponding two nodes. If blood vessel segments are not connected to each other, no undirected edge is established between the corresponding two nodes. In this way, a blood vessel topology graph is constructed.

[0091] S14, extract multiple features of each of the vascular segments.

[0092] After obtaining multiple blood vessel segments, in order to determine whether each blood vessel segment is a venous blood vessel segment or an arterial blood vessel segment, the electronic device can extract multiple features of each blood vessel segment, and then train a graph neural network based on the multiple features of each blood vessel segment, and perform binary classification on each blood vessel segment through the graph neural network.

[0093] In an optional implementation, the extraction of multiple features from each of the vascular segments includes:

[0094] Obtain the tight bounding box for each of the aforementioned vascular segments;

[0095] Obtain the feature map output by the preset blood vessel segmentation model, and extract the first feature corresponding to the tight bounding box from the feature map;

[0096] The first feature is normalized to obtain the normalized feature;

[0097] For each of the blood vessel segments, multiple preset feature extraction models are used to calculate multiple second features based on the tight bounding boxes corresponding to the blood vessel segments in the feature map.

[0098] Here, the tight bounding box refers to the smallest bounding rectangle that can enclose each blood vessel segment in the blood vessel mask image.

[0099] The preset blood vessel segmentation model is a deep learning model for segmenting blood vessels in CT images. It can obtain the feature map output from the penultimate layer of the preset blood vessel segmentation model, and this feature map is the same size as the blood vessel mask image. The electronic device can obtain the first position coordinates of the tight bounding box corresponding to each blood vessel segment in the blood vessel mask image, for example, the first position coordinates of the four vertices of the tight bounding box. Then, it obtains the second position coordinates corresponding to the first position coordinates in the feature map. Multiple first features enclosed by the rectangles corresponding to the second position coordinates in the feature map are used as the feature matrix of the blood vessel segment corresponding to the tight bounding box.

[0100] To accelerate model convergence and improve efficiency during subsequent training, the electronic device normalizes the feature matrix of each blood vessel segment after obtaining it, resulting in normalized features for that segment. Specifically, this involves calculating the average of all eigenvalues ​​in the feature matrix, then calculating the difference between each eigenvalue and the average, summing the squares of these differences, and finally calculating the mean of all sums as the normalized feature.

[0101] Multiple preset feature extraction models are pre-defined computational models for extracting multiple features of blood vessel segments, such as a length feature computational model, a diameter feature computational model, and a grayscale feature computational model. Specifically, the length feature of a blood vessel segment is the first number of pixels corresponding to the segment in the binarized blood vessel mask image, and the diameter feature of a blood vessel segment is the ratio of the second number to the first number of pixels within the corresponding tight bounding box in the feature map.

[0102] S15, an arteriovenous vessel separation model is obtained by training based on the vascular topology diagram and multiple features of each vascular segment.

[0103] The electronic device initializes the network architecture of the arteriovenous vessel separation model. For example, a graph neural network can be used as the network architecture for the arteriovenous vessel separation model. The constructed vascular topology map and multiple features extracted from each vascular segment are simultaneously input into the graph neural network for iterative training and binary classification prediction. The graph neural network predicts the category of each node in the vascular topology map, i.e., whether it is an artery or a vein. By classifying the nodes, the arteriovenous classification of the vascular mask image is obtained, thereby outputting the arteriovenous separation result.

[0104] In an optional implementation, the process of training the arteriovenous vessel separation model based on the vascular topology map and multiple features of each vascular segment includes:

[0105] The overall features are obtained based on the normalized features corresponding to each of the blood vessel segments and the plurality of second features;

[0106] The vascular topology map and multiple overall features are input into a preset neural network, and the predicted label of each vascular segment output by the preset neural network is obtained.

[0107] Calculate the first loss function value based on the preset label and the corresponding real label;

[0108] The second loss function value is calculated based on the gold standard image corresponding to the vascular mask image and the CT image;

[0109] The gradient descent algorithm is used to train the arteriovenous vessel separation model and the preset vessel segmentation model based on the first loss function value and the second loss function value, so as to obtain the trained arteriovenous vessel separation model and vessel segmentation model.

[0110] The normalized features and multiple second features corresponding to each blood vessel segment are concatenated to obtain the overall features. The blood vessel topology map and multiple overall features are used as input to a preset neural network, which outputs a predicted label for each blood vessel segment. The predicted label can be an artery or a vein. The true label is annotated by experts based on professional medical knowledge. A first loss function value is calculated based on the preset label and the corresponding true label. The first loss function value reflects the closeness between the predicted label output by the arteriovenous vessel separation model and the true label. The larger the first loss function value, the more consistent the predicted label is with the true label; the smaller the first loss function value, the more different the predicted label is from the true label. The first loss function value L1 can be calculated using the cross-entropy loss function, L1 = -SUM(y*logy'), where y represents the true label and y' represents the predicted label.

[0111] The gold standard image refers to a pre-segmented blood vessel image. For example, experts, based on their professional medical knowledge, delineate the locations of the blood vessels to be segmented from unprocessed raw CT images; that is, the gold standard image pre-labels the blood vessels. The second loss function value reflects the closeness between the blood vessel mask image output by the blood vessel separation model and the corresponding gold standard image. A larger second loss function value indicates that the blood vessel mask image is more different from the corresponding gold standard image, and a smaller second loss function value indicates that the blood vessel mask image is closer to the corresponding gold standard image. The second loss function value L2 can be the Euclidean distance or cosine angle between the blood vessel mask image and the corresponding gold standard image.

[0112] The total loss function value is obtained by summing the first and second loss function values. Based on this total loss function value, the vessel segmentation network and the arteriovenous vessel separation network are jointly iteratively trained until the total loss function value converges, at which point the iterative training stops. Iterative training is a model training method in deep learning used to optimize the model. The iterative training process in this step is as follows: In each training iteration, all training samples are sequentially read in and the current total loss function value is calculated. A stochastic gradient descent algorithm is used to determine the gradient descent direction, causing the total loss function value to gradually decrease and reach a stable state, thereby optimizing the parameters of the constructed network model. Convergence of the total loss function value means that the total loss function value approaches 0, for example, less than 0.1.

[0113] S16, input the CT image to be processed into the trained arteriovenous vessel separation model for arteriovenous vessel separation.

[0114] The CT images to be processed refer to fundus CT images, lung CT images, or other medical images that require arteriovenous separation.

[0115] The CT image to be processed is input into the trained arteriovenous vessel separation model, which then outputs classification labels for arterial and venous vessel segments.

[0116] Disease prediction based on quantitative indicators of arteries and veins requires precise segmentation. However, previous deep learning-based arteriovenous segmentation methods, mostly based on CNNs, have significant limitations, leading to insufficient classification accuracy. This invention combines deep neural networks with graph neural networks (GCNs) for joint training, resulting in more accurate vessel classification and simpler, more precise classification between arteries and veins. Classification based on a binary mask of the entire vessel image avoids the problem of misclassification of the same vessel leading to inconsistencies between arteries and veins; that is, it ensures the continuity of arterial and venous vessels by identifying parts of a segment as both veins and arteries. Furthermore, it avoids missing some fine vessels. This invention transforms the problem of classifying arteries and veins for each pixel into classifying them for each vessel segment. This overcomes the problem of pixel-level classification algorithms potentially resulting in a single vessel containing both arterial and venous pixels, and also avoids issues such as arteriovenous intersections and branching that easily lead to misclassification. By extracting the vascular skeleton (vessel centerline) and determining the branching points of the vascular skeleton, the accuracy of segmenting the vessel image into vessel segment images can be improved.

[0117] In an optional implementation, prior to extracting the vascular skeleton from the CT image, the method further includes:

[0118] Convert the CT image into an image of a preset size;

[0119] Normalize CT images of a preset size to obtain normalized CT images;

[0120] The normalized CT image is enhanced to obtain an enhanced CT image.

[0121] The preset size can be 512×512.

[0122] Normalizing a CT image of a preset size aims to obtain a CT image with uniform pixel color. In practice, the preset-size CT image can first be Gaussian blurred to obtain a Gaussian image, and then the Gaussian image can be superimposed on the CT image in reverse.

[0123] Enhancement processing of normalized CT images may include: horizontal flipping with a preset probability; transposing with a preset probability; random gamma transformation; randomly changing the saturation value of the normalized CT image with a preset probability; histogram equalization processing of the normalized CT image; and random brightness and contrast adjustment.

[0124] Through the above optional implementation methods, CT images of different formats and sizes can be converted into images of a preset size. At the same time, normalization processing of CT images of the preset size can obtain CT images with uniform pixel color. Enhancement processing of CT images with uniform pixel color can obtain CT images with satisfactory brightness, clarity, and saturation values.

[0125] It should be understood that extracting the vascular skeleton from CT images includes extracting the vascular skeleton from enhanced CT images.

[0126] In an optional implementation, after obtaining the arterial and venous vessels, the method further includes:

[0127] Obtain the largest and smallest arterial vessels;

[0128] The equivalent value of the central retinal artery diameter is calculated based on the diameter of the largest artery and the diameter of the smallest artery.

[0129] Obtain the largest and smallest veins;

[0130] The equivalent value of the central retinal vein diameter is calculated based on the diameter of the largest vein and the diameter of the smallest vein.

[0131] The quantitative values ​​of arterial and venous vessel diameters are calculated based on the equivalent values ​​of the central retinal artery diameter and the central retinal vein diameter.

[0132] The equivalent value of the central retinal artery diameter (CRAE) in this embodiment is calculated using the following formula:

[0133] CRAE = (Ai 2 +Aj 2 ) 1 / 2 ,

[0134] Where Ai is the diameter of the largest arterial vessel obtained through iteration, and Aj is the diameter of the smallest arterial vessel obtained through iteration.

[0135] The equivalent value of the central retinal vein diameter (CRVE) in this embodiment is calculated using the following formula:

[0136] CRVE = (Vi 2 +Vj 2 ) 1 / 2 ,

[0137] Where Vi is the diameter of the largest vein obtained through iteration, and Vj is the diameter of the smallest vein obtained through iteration.

[0138] In this embodiment, the quantitative value of arterial and venous vessel diameter can be the ratio of the equivalent value of the central retinal artery diameter to the equivalent value of the central retinal vein diameter.

[0139] In the above-mentioned optional implementation, by quantifying the diameters of arteries and veins, equivalent values ​​for the diameters of the central retinal artery and central retinal vein are obtained, thereby yielding quantified values ​​for the arterial and venous vessel diameters. Because the separation accuracy of arteries and veins is high, the accuracy of the obtained arterial and venous vessel diameters is also high, resulting in high accuracy of the quantified arterial and venous vessel diameter values.

[0140] Medical evidence shows significant differences in microvessels in fundus images between individuals with and without hypertension. For example, under repeated high-pressure stimulation, the retinal arterioles in hypertension may initially show slight narrowing and mild hardening. If blood pressure remains elevated for a long period, the retina will undergo further changes, with persistent arterial narrowing, significant retinal arteriosclerosis, and the appearance of a "silver line reaction," uneven narrowing of the arterial diameter, and arteriovenous crossing indentations. This demonstrates that hypertension can affect or reduce the diameter of arteries and veins. In other words, the risk of hypertension in a subject can be predicted based on the quantitative values ​​of arteriovenous vessel diameter.

[0141] Example 2

[0142] Figure 2 This is a structural diagram of the image-based arteriovenous vessel separation device provided in an embodiment of the present invention.

[0143] In some embodiments, the image-based arteriovenous vessel separation device 20 may include multiple functional modules composed of computer program segments. The computer programs for each program segment in the image-based arteriovenous vessel separation device 20 may be stored in the memory of an electronic device and executed by at least one processor to perform (see details). Figure 1 (Description) Functional image-based arteriovenous vessel separation.

[0144] In this embodiment, the image-based arteriovenous vessel separation device 20 can be divided into multiple functional modules according to its function. These functional modules may include: an extraction module 201, a segmentation module 202, a construction module 203, a calculation module 204, a training module 205, a separation module 206, an enhancement module 207, and a quantization module 208. The module referred to in this invention is a series of computer program segments that can be executed by at least one processor and perform a fixed function, stored in memory. In this embodiment, the functions of each module will be detailed in subsequent embodiments.

[0145] The extraction module 201 is used to extract the vascular skeleton in CT images.

[0146] The CT images are either fundus images obtained by scanning the patient's fundus using computed tomography (CT) technology, or lung images obtained by scanning the patient's lungs using computed tomography (CT) technology. Eye diseases can be identified by separating retinal arteries and veins in fundus CT images, or lung diseases can be identified by separating pulmonary arteries and veins in lung CT images.

[0147] The CT images can be obtained by electronic devices from a digital medical database. This digital medical database can be a digital repository storing patient medical records in a hospital, or it can be a networked database of multiple hospitals; this invention does not impose any limitations.

[0148] Electronic devices can acquire CT images of multiple patients, with each patient corresponding to one or more CT images. An arteriovenous vessel separation model is trained based on these acquired CT images, and then used to perform vessel segmentation, i.e., arteriovenous separation.

[0149] In an optional implementation, extracting the vascular skeleton from the CT image includes:

[0150] The CT image is segmented using a preset blood vessel segmentation model to obtain a binarized blood vessel mask image;

[0151] The blood vessel mask image is refined to obtain the blood vessel skeleton.

[0152] The preset blood vessel segmentation model can be a model trained using deep learning networks, such as U-net or Fully Convolutional Networks (FCN). The process of creating the blood vessel segmentation model is existing technology and will not be described in detail here.

[0153] Electronic devices use a preset vascular segmentation model to segment CT images, which means separating blood vessels from the background in the CT image without distinguishing between arteries and veins. The grayscale value of pixels in the binarized vascular mask image is 0 or 255, i.e., a black and white image. In the binarized vascular mask image, white areas represent blood vessels, and black areas represent the surrounding background. This implementation method reduces the amount of data in the CT image by obtaining a binarized vascular mask image, thereby facilitating the extraction of the vascular skeleton, i.e., highlighting the contours of the blood vessels.

[0154] In an optional implementation, refining the blood vessel mask image to obtain the blood vessel skeleton includes:

[0155] The blood vessel mask image is filtered to obtain a filtered image;

[0156] The initial vascular skeleton is extracted from the filtered image using a thinning algorithm;

[0157] The initial vascular skeleton is fitted to obtain a continuous vascular skeleton;

[0158] The continuous vascular skeleton is obtained by performing single-pixel processing.

[0159] The vascular skeleton is a topological description of the geometric features of blood vessels. It reflects the connectivity, structural information, and orientation of the vessels. The vascular skeleton is often located at the center of the vessel and is also called the vascular centerline. The vascular skeleton can be understood as the central axis of the vessel.

[0160] The electronic device can perform median filtering on the vascular mask image, which removes possible bifurcations at the ends of the vascular skeleton in the vascular mask image. Median filtering is a non-linear smoothing technique that sets the gray value of each pixel to the median of the gray values ​​of all pixels within a certain neighborhood window of that pixel.

[0161] A thinning algorithm is used to extract the vascular skeleton from the filtered vascular mask image, thereby removing redundant boundary points and retaining important image nodes such as junctions, endpoints, and isolated points. Electronic devices can employ morphological thinning algorithms to thin the vascular mask image to extract the vascular skeleton, which is then called the initial vascular skeleton. 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. Morphological thinning algorithms are existing technologies and will not be described in detail here.

[0162] Since the initial vascular skeleton is composed of discrete pixels, the electronic device fits the initial vascular skeleton to obtain a continuous vascular skeleton. In some implementations, a least-squares cubic spline interpolation algorithm can be used to fit the initial vascular skeleton.

[0163] Furthermore, since morphological thinning algorithms cannot guarantee that the extracted vascular skeleton is a single pixel, and non-single-pixel skeletons are not conducive to measuring blood vessel diameter, the initial vascular skeleton extracted by the morphological thinning algorithm needs to be further processed into single pixels. In specific implementation, the width of the continuous vascular skeleton is thinned to one pixel width towards the center of the blood vessel to form a single-pixel vascular skeleton, while maintaining the basic topological structure of the blood vessel shape of the single-pixel vascular skeleton.

[0164] The segmentation module 202 is used to segment the blood vessel into multiple blood vessel segments according to the vascular skeleton.

[0165] The entire vascular skeleton is not conducive to the separation of arteries and veins. Therefore, after extracting the target vascular skeleton, the electronic device segments the target vascular skeleton to obtain multiple vascular segments. Each vascular segment is either a vein or an artery, which facilitates the separation of arteries and veins.

[0166] After extracting the vascular skeleton, the electronic device can use a point-capturing tool to determine the intersections of the vascular skeleton and segment the blood vessel into multiple vascular segments based on these intersections, thereby obtaining multiple vascular segment images. These intersections can be intersections between arteries, branching points of arteries, intersections between veins, branching points of veins, or intersections between arteries and veins; this application does not specifically limit the type of intersection.

[0167] In an optional implementation, dividing the blood vessel into multiple vascular segments based on the vascular skeleton includes:

[0168] Obtain the vascular branch points in the vascular skeleton;

[0169] Using the vascular branch points as segmentation points, the vascular skeleton is segmented to obtain the multiple vascular segments; or

[0170] The vascular branch points are deleted to split the vascular skeleton into multiple branches, and each branch is identified as a vascular segment to obtain the multiple vascular segments.

[0171] Pixels located at branch points share a common characteristic: they must have three neighboring pixels within their eight-neighborhood. Based on this characteristic, eight-neighborhood filtering can be used to detect vascular branch points in the vascular skeleton. The electronic device performs eight-neighborhood filtering on the image corresponding to the vascular skeleton, then calculates the number of eight neighbors for each pixel. Pixels with three eight-neighborhoods are identified as suspicious points, and their pixel values ​​are obtained. The determination of whether a suspicious point is a vascular branch point is based on the obtained pixel value. Specifically, if the obtained pixel value is 0, it indicates that the suspicious point is a background point, and therefore it is not a vascular branch point; if the obtained pixel value is 1, it indicates that the suspicious point is the center pixel, and therefore it is a vascular branch point.

[0172] The construction module 203 constructs a vascular topology diagram based on the multiple vascular segments.

[0173] In this embodiment, each blood vessel segment is treated as a node. If blood vessel segments are connected to each other, an undirected edge is established between the corresponding two nodes. If blood vessel segments are not connected to each other, no undirected edge is established between the corresponding two nodes. In this way, a blood vessel topology graph is constructed.

[0174] The calculation module 204 is used to extract multiple features of each of the blood vessel segments.

[0175] After obtaining multiple blood vessel segments, in order to determine whether each blood vessel segment is a venous blood vessel segment or an arterial blood vessel segment, the electronic device can extract multiple features of each blood vessel segment, and then train a graph neural network based on the multiple features of each blood vessel segment, and perform binary classification on each blood vessel segment through the graph neural network.

[0176] In an optional implementation, the extraction of multiple features from each of the vascular segments includes:

[0177] Obtain the tight bounding box for each of the aforementioned vascular segments;

[0178] Obtain the feature map output by the preset blood vessel segmentation model, and extract the first feature corresponding to the tight bounding box from the feature map;

[0179] The first feature is normalized to obtain the normalized feature;

[0180] For each of the blood vessel segments, multiple preset feature extraction models are used to calculate multiple second features based on the tight bounding boxes corresponding to the blood vessel segments in the feature map.

[0181] Here, the tight bounding box refers to the smallest bounding rectangle that can enclose each blood vessel segment in the blood vessel mask image.

[0182] The preset blood vessel segmentation model is a deep learning model for segmenting blood vessels in CT images. It can obtain the feature map output from the penultimate layer of the preset blood vessel segmentation model, and this feature map is the same size as the blood vessel mask image. The electronic device can obtain the first position coordinates of the tight bounding box corresponding to each blood vessel segment in the blood vessel mask image, for example, the first position coordinates of the four vertices of the tight bounding box. Then, it obtains the second position coordinates corresponding to the first position coordinates in the feature map. Multiple first features enclosed by the rectangles corresponding to the second position coordinates in the feature map are used as the feature matrix of the blood vessel segment corresponding to the tight bounding box.

[0183] To accelerate model convergence and improve efficiency during subsequent training, the electronic device normalizes the feature matrix of each blood vessel segment after obtaining it, resulting in normalized features for that segment. Specifically, this involves calculating the average of all eigenvalues ​​in the feature matrix, then calculating the difference between each eigenvalue and the average, summing the squares of these differences, and finally calculating the mean of all sums as the normalized feature.

[0184] Multiple preset feature extraction models are pre-defined computational models for extracting multiple features of blood vessel segments, such as a length feature computational model, a diameter feature computational model, and a grayscale feature computational model. Specifically, the length feature of a blood vessel segment is the first number of pixels corresponding to the segment in the binarized blood vessel mask image, and the diameter feature of a blood vessel segment is the ratio of the second number to the first number of pixels within the corresponding tight bounding box in the feature map.

[0185] The training module 205 is used to train an arteriovenous vessel separation model based on the vascular topology diagram and multiple features of each vascular segment.

[0186] The electronic device initializes the network architecture of the arteriovenous vessel separation model. For example, a graph neural network can be used as the network architecture for the arteriovenous vessel separation model. The constructed vascular topology map and multiple features extracted from each vascular segment are simultaneously input into the graph neural network for iterative training and binary classification prediction. The graph neural network predicts the category of each node in the vascular topology map, i.e., whether it is an artery or a vein. By classifying the nodes, the arteriovenous classification of the vascular mask image is obtained, thereby outputting the arteriovenous separation result.

[0187] In an optional implementation, the process of training the arteriovenous vessel separation model based on the vascular topology map and multiple features of each vascular segment includes:

[0188] The overall features are obtained based on the normalized features corresponding to each of the blood vessel segments and the plurality of second features;

[0189] The vascular topology map and multiple overall features are input into a preset neural network, and the predicted label of each vascular segment output by the preset neural network is obtained.

[0190] Calculate the first loss function value based on the preset label and the corresponding real label;

[0191] The second loss function value is calculated based on the gold standard image corresponding to the vascular mask image and the CT image;

[0192] The gradient descent algorithm is used to train the arteriovenous vessel separation model and the preset vessel segmentation model based on the first loss function value and the second loss function value, so as to obtain the trained arteriovenous vessel separation model and vessel segmentation model.

[0193] The normalized features and multiple second features corresponding to each blood vessel segment are concatenated to obtain the overall features. The blood vessel topology map and multiple overall features are used as input to a preset neural network, which outputs a predicted label for each blood vessel segment. The predicted label can be an artery or a vein. The true label is annotated by experts based on professional medical knowledge. A first loss function value is calculated based on the preset label and the corresponding true label. The first loss function value reflects the closeness between the predicted label output by the arteriovenous vessel separation model and the true label. The larger the first loss function value, the more consistent the predicted label is with the true label; the smaller the first loss function value, the more different the predicted label is from the true label. The first loss function value L1 can be calculated using the cross-entropy loss function, L1 = -SUM(y*logy'), where y represents the true label and y' represents the predicted label.

[0194] The gold standard image refers to a pre-segmented blood vessel image. For example, experts, based on their professional medical knowledge, delineate the locations of the blood vessels to be segmented from unprocessed raw CT images; that is, the gold standard image pre-labels the blood vessels. The second loss function value reflects the closeness between the blood vessel mask image output by the blood vessel separation model and the corresponding gold standard image. A larger second loss function value indicates that the blood vessel mask image is more different from the corresponding gold standard image, and a smaller second loss function value indicates that the blood vessel mask image is closer to the corresponding gold standard image. The second loss function value L2 can be the Euclidean distance or cosine angle between the blood vessel mask image and the corresponding gold standard image.

[0195] The total loss function value is obtained by summing the first and second loss function values. Based on this total loss function value, the vessel segmentation network and the arteriovenous vessel separation network are jointly iteratively trained until the total loss function value converges, at which point the iterative training stops. Iterative training is a model training method in deep learning used to optimize the model. The iterative training process in this step is as follows: In each training iteration, all training samples are sequentially read in and the current total loss function value is calculated. A stochastic gradient descent algorithm is used to determine the gradient descent direction, causing the total loss function value to gradually decrease and reach a stable state, thereby optimizing the parameters of the constructed network model. Convergence of the total loss function value means that the total loss function value approaches 0, for example, less than 0.1.

[0196] The separation module 206 is used to input the CT image to be processed into the trained arteriovenous vessel separation model for arteriovenous vessel separation.

[0197] The CT images to be processed refer to fundus CT images, lung CT images, or other medical images that require arteriovenous separation.

[0198] The CT image to be processed is input into the trained arteriovenous vessel separation model, which then outputs classification labels for arterial and venous vessel segments.

[0199] Disease prediction based on quantitative indicators of arteries and veins requires precise segmentation. However, previous deep learning-based arteriovenous segmentation methods, mostly based on CNNs, have significant limitations, leading to insufficient classification accuracy. This invention combines deep neural networks with graph neural networks (GCNs) for joint training, resulting in more accurate vessel classification and simpler, more precise classification between arteries and veins. Classification based on a binary mask of the entire vessel image avoids the problem of misclassification of the same vessel leading to inconsistencies between arteries and veins; that is, it ensures the continuity of arterial and venous vessels by identifying parts of a segment as both veins and arteries. Furthermore, it avoids missing some fine vessels. This invention transforms the problem of classifying arteries and veins for each pixel into classifying them for each vessel segment. This overcomes the problem of pixel-level classification algorithms potentially resulting in a single vessel containing both arterial and venous pixels, and also avoids issues such as arteriovenous intersections and branching that easily lead to misclassification. By extracting the vascular skeleton (vessel centerline) and determining the branching points of the vascular skeleton, the accuracy of segmenting the vessel image into vessel segment images can be improved.

[0200] In an optional implementation, the enhancement module 207 is configured to convert the CT image into an image of a preset size before extracting the vascular skeleton from the CT image; normalize the CT image of the preset size to obtain a normalized CT image; and enhance the normalized CT image to obtain an enhanced CT image.

[0201] The preset size can be 512×512.

[0202] Normalizing a CT image of a preset size aims to obtain a CT image with uniform pixel color. In practice, the preset-size CT image can first be Gaussian blurred to obtain a Gaussian image, and then the Gaussian image can be superimposed on the CT image in reverse.

[0203] Enhancement processing of normalized CT images may include: horizontal flipping with a preset probability; transposing with a preset probability; random gamma transformation; randomly changing the saturation value of the normalized CT image with a preset probability; histogram equalization processing of the normalized CT image; and random brightness and contrast adjustment.

[0204] Through the above optional implementation methods, CT images of different formats and sizes can be converted into images of a preset size. At the same time, normalization processing of CT images of the preset size can obtain CT images with uniform pixel color. Enhancement processing of CT images with uniform pixel color can obtain CT images with satisfactory brightness, clarity, and saturation values.

[0205] It should be understood that extracting the vascular skeleton from CT images includes extracting the vascular skeleton from enhanced CT images.

[0206] The quantization module 208 is used to obtain the quantified value of the arterial and venous vessel diameter after obtaining the arterial and venous vessels.

[0207] In an optional implementation, obtaining the quantitative value of the arterial and venous vessel diameter includes:

[0208] Obtain the largest and smallest arterial vessels;

[0209] The equivalent value of the central retinal artery diameter is calculated based on the diameter of the largest artery and the diameter of the smallest artery.

[0210] Obtain the largest and smallest veins;

[0211] The equivalent value of the central retinal vein diameter is calculated based on the diameter of the largest vein and the diameter of the smallest vein.

[0212] The quantitative values ​​of arterial and venous vessel diameters are calculated based on the equivalent values ​​of the central retinal artery diameter and the central retinal vein diameter.

[0213] The equivalent value of the central retinal artery diameter (CRAE) in this embodiment is calculated using the following formula:

[0214] CRAE = (Ai 2 +Aj 2 ) 1 / 2 ,

[0215] Where Ai is the diameter of the largest arterial vessel obtained through iteration, and Aj is the diameter of the smallest arterial vessel obtained through iteration.

[0216] The equivalent value of the central retinal vein diameter (CRVE) in this embodiment is calculated using the following formula:

[0217] CRVE = (Vi 2 +Vj 2 ) 1 / 2,

[0218] Where Vi is the diameter of the largest vein obtained through iteration, and Vj is the diameter of the smallest vein obtained through iteration.

[0219] In this embodiment, the quantitative value of arterial and venous vessel diameter can be the ratio of the equivalent value of the central retinal artery diameter to the equivalent value of the central retinal vein diameter.

[0220] In the above-mentioned optional implementation, by quantifying the diameters of arteries and veins, equivalent values ​​for the diameters of the central retinal artery and central retinal vein are obtained, thereby yielding quantified values ​​for the arterial and venous vessel diameters. Because the separation accuracy of arteries and veins is high, the accuracy of the obtained arterial and venous vessel diameters is also high, resulting in high accuracy of the quantified arterial and venous vessel diameter values.

[0221] Medical evidence shows significant differences in microvessels in fundus images between individuals with and without hypertension. For example, under repeated high-pressure stimulation, the retinal arterioles in hypertension may initially show slight narrowing and mild hardening. If blood pressure remains elevated for a long period, the retina will undergo further changes, with persistent arterial narrowing, significant retinal arteriosclerosis, and the appearance of a "silver line reaction," uneven narrowing of the arterial diameter, and arteriovenous crossing indentations. This demonstrates that hypertension can affect or reduce the diameter of arteries and veins. In other words, the risk of hypertension in a subject can be predicted based on the quantitative values ​​of arteriovenous vessel diameter.

[0222] Example 3

[0223] This embodiment provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the steps described in the above-described image-based arteriovenous vessel separation method embodiment, for example... Figure 1 S11-S16 as shown:

[0224] S11, extract the vascular skeleton from the CT image;

[0225] S12, the blood vessel is divided into multiple blood vessel segments according to the vascular skeleton;

[0226] S13, Construct a vascular topology diagram based on the multiple vascular segments;

[0227] S14, extract multiple features of each of the vascular segments;

[0228] S15, an arteriovenous vessel separation model is obtained by training based on the vascular topology diagram and multiple features of each vascular segment;

[0229] S16, input the CT image to be processed into the trained arteriovenous vessel separation model for arteriovenous vessel separation.

[0230] Alternatively, when the computer program is executed by the processor, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 2 Modules 201-206 in the middle:

[0231] The extraction module 201 is used to extract the vascular skeleton in CT images;

[0232] The segmentation module 202 is used to segment the blood vessel into multiple blood vessel segments according to the vascular skeleton;

[0233] The construction module 203 is used to construct a vascular topology diagram based on the plurality of vascular segments;

[0234] The calculation module 204 is used to extract multiple features of each of the blood vessel segments;

[0235] The training module 205 is used to train an arteriovenous vessel separation model based on the vascular topology diagram and multiple features of each vascular segment.

[0236] The separation module 206 is used to input the CT image to be processed into the trained arteriovenous vessel separation model for arteriovenous vessel separation.

[0237] When executed by a processor, the computer program also implements the enhancement module 207 and quantization module 208 in the above-described device embodiment. See details for further information. Figure 2 And its related descriptions.

[0238] Example 4

[0239] See Figure 3 The diagram shown is a structural schematic of an electronic device provided in an embodiment of the present invention. In a preferred embodiment of the present invention, the electronic device 3 includes a memory 31, at least one processor 32, at least one communication bus 33, and a transceiver 34.

[0240] Those skilled in the art should understand that Figure 3 The structure of the electronic device shown does not constitute a limitation of the embodiments of the present invention. It can be a bus structure or a star structure. The electronic device 3 may also include more or fewer other hardware or software than shown, or different component arrangements.

[0241] In some embodiments, the electronic device 3 is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), digital processors, and embedded devices. The electronic device 3 may also include client devices, including, but not limited to, any electronic product capable of human-computer interaction with a client via a keyboard, mouse, remote control, touchpad, or voice control device, such as personal computers, tablets, smartphones, and digital cameras.

[0242] The electronic device 3 described herein is merely an example. Other existing or future electronic products that are adaptable to this invention should also be included within the scope of protection of this invention and are incorporated herein by reference.

[0243] In some embodiments, the memory 31 stores a computer program that, when executed by the at least one processor 32, implements all or part of the steps in the image-based arteriovenous vessel separation method as described. The memory 31 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0244] Furthermore, the computer-readable storage medium may primarily include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application required for a function, etc.; and the data storage area may store data created based on the use of blockchain nodes, etc.

[0245] The blockchain referred to in this invention is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0246] In some embodiments, the at least one processor 32 is the control unit of the electronic device 3, connecting various components of the electronic device 3 via various interfaces and lines. It executes programs or modules stored in the memory 31 and calls data stored in the memory 31 to perform various functions and process data. For example, when the at least one processor 32 executes a computer program stored in the memory, it implements all or part of the steps of the image-based arteriovenous vessel separation method described in this embodiment of the invention; or it implements all or part of the functions of the image-based arteriovenous vessel separation device. The at least one processor 32 may be composed of integrated circuits, such as a single-packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0247] In some embodiments, the at least one communication bus 33 is configured to enable communication between the memory 31 and the at least one processor 32, etc.

[0248] Although not shown, the electronic device 3 may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 32 via a power management device, thereby enabling functions such as charging, discharging, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 3 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0249] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) or processor to execute portions of the methods described in the various embodiments of the present invention.

[0250] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0251] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0252] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0253] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other elements, and the singular does not exclude the plural. Multiple elements or devices recited in the specification may also be implemented by a single element or device in software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any particular order.

[0254] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An image-based method for separating arteries and veins, characterized in that, The method includes: The CT image is segmented using a preset blood vessel segmentation model to obtain a binarized blood vessel mask image; the blood vessel mask image is then filtered to obtain a filtered image; an initial blood vessel skeleton is extracted from the filtered image using a thinning algorithm; the initial blood vessel skeleton is fitted to obtain a continuous blood vessel skeleton; and the continuous blood vessel skeleton is then pixelated to obtain the final blood vessel skeleton. The blood vessel is divided into multiple blood vessel segments according to the vascular skeleton; A vascular topology diagram is constructed based on the multiple vascular segments; Obtain the tight bounding box of each blood vessel segment; obtain the feature map output by the preset blood vessel segmentation model, and extract the first feature corresponding to the tight bounding box from the feature map; normalize the first feature to obtain normalized features; for each blood vessel segment, use multiple preset feature extraction models to calculate multiple second features based on the tight bounding box corresponding to the blood vessel segment in the feature map; The overall features are obtained based on the normalized features and multiple second features corresponding to each blood vessel segment; the blood vessel topology map and multiple overall features are input into a preset neural network, and the predicted label of each blood vessel segment output by the preset neural network is obtained; a first loss function value is calculated based on the predicted label and the corresponding real label; a second loss function value is calculated based on the gold standard image corresponding to the blood vessel mask image and the CT image; the arteriovenous vessel separation model and the preset blood vessel segmentation model are trained using a gradient descent algorithm based on the first loss function value and the second loss function value, to obtain the trained arteriovenous vessel separation model and blood vessel segmentation model; The CT image to be processed is input into the trained arteriovenous vessel separation model for arteriovenous vessel separation.

2. The image-based arteriovenous vessel separation method as described in claim 1, characterized in that, The step of dividing the blood vessel into multiple blood vessel segments according to the vascular skeleton includes: Obtain the vascular branch points in the vascular skeleton; Using the vascular branch points as segmentation points, the vascular skeleton is segmented to obtain the multiple vascular segments; or The vascular branch points are deleted to split the vascular skeleton into multiple branches, and each branch is identified as a vascular segment to obtain the multiple vascular segments.

3. The image-based arteriovenous vessel separation method as described in claim 1 or 2, characterized in that, After obtaining the arterial and venous vessels, the method further includes: Obtain the largest and smallest arterial vessels; The equivalent value of the central retinal artery diameter is calculated based on the diameter of the largest artery and the diameter of the smallest artery. Obtain the largest and smallest veins; The equivalent value of the central retinal vein diameter is calculated based on the diameter of the largest vein and the diameter of the smallest vein. The quantitative values ​​of arterial and venous vessel diameters are calculated based on the equivalent values ​​of the central retinal artery diameter and the central retinal vein diameter.

4. An image-based arteriovenous vessel separation device, characterized in that, The device includes: An extraction module is used to segment CT images using a preset blood vessel segmentation model to obtain a binarized blood vessel mask image; to perform filtering operations on the blood vessel mask image to obtain a filtered image; to extract an initial blood vessel skeleton from the filtered image using a thinning algorithm; to fit the initial blood vessel skeleton to obtain a continuous blood vessel skeleton; and to perform single-pixel processing on the continuous blood vessel skeleton to obtain the blood vessel skeleton. A segmentation module is used to segment the blood vessel into multiple blood vessel segments based on the vascular skeleton; A construction module is used to construct a vascular topology diagram based on the multiple vascular segments; The calculation module is used to obtain the tight bounding box of each blood vessel segment; obtain the feature map output by the preset blood vessel segmentation model, and extract the first feature corresponding to the tight bounding box from the feature map; normalize the first feature to obtain normalized features; and for each blood vessel segment, use multiple preset feature extraction models to calculate multiple second features based on the tight bounding box corresponding to the blood vessel segment in the feature map. The training module is used to obtain overall features based on the normalized features and multiple second features corresponding to each blood vessel segment; input the blood vessel topology map and multiple overall features into a preset neural network, and obtain the predicted label of each blood vessel segment output by the preset neural network; calculate a first loss function value based on the predicted label and the corresponding real label; calculate a second loss function value based on the gold standard image corresponding to the blood vessel mask image and the CT image; and use a gradient descent algorithm to train the arteriovenous vessel separation model and the preset blood vessel segmentation model based on the first loss function value and the second loss function value to obtain the trained arteriovenous vessel separation model and blood vessel segmentation model. The separation module is used to input the CT image to be processed into the trained arteriovenous vessel separation model for arteriovenous vessel separation.

5. An electronic device, characterized in that, The electronic device includes a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement the image-based arteriovenous vessel separation method as described in any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the image-based arteriovenous vessel separation method as described in any one of claims 1 to 3.

Images