A noise-annotated blood vessel image segmentation method and system

Abstract

The present invention discloses a noise-annotated vascular image segmentation method, the method comprising: obtaining a retinal vascular data set; inputting the noise labels and corresponding vascular images in the retinal vascular data set into a pre-trained noise label completion unit, performing label completion processing through feature extraction operation and feature fusion to obtain a completed label; inputting the completed label and the corresponding vascular image into a dual-stream feature extraction network to perform feature extraction operation and feature fusion to obtain a final image segmentation result. The present invention achieves high-precision retinal vascular image segmentation.

Description

Technical Field

[0001] The present invention relates to the technical field of vascular image processing, and in particular to a noise-annotated vascular image segmentation method and system. Background Art

[0002] The current segmentation method for noisy datasets first uses confidence to identify the noise label, and then uses pixel reweighting, multi-network collaborative training, deformable space transformation to adjust the noise pixel-level label and other methods to reduce the impact of noise samples on the model. Since the semantic information contained in noise pixels is rich, if the noise pixels are directly discarded or the weight of the noise pixels is reduced, the information in the noise pixels cannot be fully utilized, and for datasets that are difficult to label and have a small number of labels, it will reduce the information that the model can learn. The current correction method for noise labels works well for images with blocky segmentation targets, but because the morphology of blood vessel images is very complex and has a certain regularity, the thickness of blood vessel branches varies, and the traditional noise label processing method has a significantly reduced effect on the processing of blood vessel labels, and the segmentation effect of thin blood vessels in the segmentation results is poor.

[0003] The Chinese invention with application number 202211126578.2: A retinal vascular segmentation method based on the fusion of UNet and Transformer, the technical solution adopted by which includes the following steps: Step 1, preprocessing the image to be trained to obtain a preprocessed image; Step 2, inputting the preprocessed image into a retinal vascular segmentation model based on the fusion of UNet and Transformer to obtain a weight file, the model includes an encoder, a decoder and a fusion attention mechanism, the encoder includes a multi-stream cascaded convolutional layer, multiple pooling layers and multiple residual modules, each convolutional layer uses a residual module, and the pooling layer Set between two convolutional layer units; the decoder includes a cascade of multiple improved residual modules based on extended convolution, multiple upsampling modules and deconvolution layers, and the upsampling module is set between two adjacent improved residual units based on extended convolution; the fusion attention mechanism uses the output of the pooling layer in the encoder and the output of the adjacent stream pooling layer as low-level feature input and high-level feature input respectively, and the high-level feature input of the fusion attention mechanism in the third layer is the feature map formed by the corresponding encoder pooling layer through the Transformer module; step 3, load the weight file, input the tested fundus image into the model to obtain the retinal vascular segmentation result. This scheme does not improve the accuracy of image segmentation by optimizing the quality of the data set. Summary of the invention

[0004] In order to solve the problem of insufficient accuracy of vascular segmentation in the prior art of noise-annotated vascular image segmentation due to complex vascular images, difficulty in manual labeling, and missing or wrong vascular labels, the present invention provides a noise-annotated vascular image segmentation method and system. The technical solution adopted by the present invention is:

[0005] A first aspect of the present invention provides a noise-annotated blood vessel image segmentation method, the method comprising:

[0006] Acquire retinal vascular dataset;

[0007] The noise labels and corresponding vascular images in the retinal vascular dataset are input into the pre-trained noise label completion unit, and the label completion process is performed through feature extraction and feature fusion to obtain the completed labels;

[0008] The completed labels and the corresponding blood vessel images are input into a dual-stream feature extraction network for feature extraction and feature fusion to obtain the final image segmentation result.

[0009] As a preferred solution, the training method of the noise label completion unit includes:

[0010] Performing data enhancement processing on the vascular image data and corresponding labels in the retinal vascular dataset;

[0011] A preset number of clean labels are selected; and a noise label completion unit is trained using the clean labels and the corresponding vascular images.

[0012] As a preferred solution, the data enhancement processing includes random flipping processing and brightness change processing; wherein the brightness change processing method includes:

[0013] A random function is used to generate a random number within a preset numerical range, and a preset value is used as a dividing line. If the value is greater than the preset value, the value is dimmed, and if the value is less than the preset value, the value is brightened. The vascular image data and the corresponding labels in the retinal vascular data set are transformed in brightness. The corresponding expression is:

[0014] S(o)=f(S(i))

[0015] Among them, S(o) is the processed image, S(i) is the original image, and f(·) is the brightness change function.

[0016] As a preferred solution, the noise labels and corresponding vascular images in the retinal vascular dataset are input into a pre-trained noise label completion unit, and label completion processing is performed through feature extraction operation and feature fusion. The method for obtaining the completed label includes:

[0017] Performing four feature extraction operations on the noise label to obtain a first label feature map, a second label feature map, a third label feature map, and a fourth label feature map;

[0018] Performing a feature extraction operation on the blood vessel image corresponding to the noise label to obtain a first image feature map;

[0019] Performing a feature fusion operation on the first label feature map and the first image feature map to obtain a first fused feature map; performing a feature extraction operation on the first fused feature map to obtain first fused feature information;

[0020] Performing a feature fusion operation on the first fused feature information and the second label feature map to obtain a second fused feature map; performing a feature extraction operation on the second fused feature map to obtain second fused feature information;

[0021] Performing a feature fusion operation on the second fused feature information and the third label feature map to obtain a third fused feature map; performing a feature extraction operation on the third fused feature map to obtain third fused feature information;

[0022] Performing a feature fusion operation on the third fused feature information and the fourth label feature map to obtain a fourth fused feature map; performing a feature extraction operation on the fourth fused feature map to obtain fourth fused feature information;

[0023] A decoding operation is performed on the fourth fusion feature information to obtain a completion label.

[0024] As a preferred solution, the method of inputting the completed label and the corresponding vascular image into a dual-stream feature extraction network for feature extraction and feature fusion to obtain the final segmentation result includes:

[0025] In the label feature extraction stream of the dual-stream feature extraction network, the completed label is subjected to five feature extraction operations to obtain a first completed label feature map, a second completed label feature map, a third completed label feature map, a fourth completed label feature map, and a fifth completed label feature map; wherein the fifth completed label feature map is subjected to a decoding operation to obtain a feature prediction result; finally, the completed label is used as the supervision information of the label feature extraction stream, and the label feature extraction stream is updated using the dice loss;

[0026] In the image segmentation stream of the dual-stream feature extraction network, a feature extraction operation is performed on the blood vessel image to obtain a first blood vessel image feature map;

[0027] Performing a feature fusion operation on the first complement label feature map and the first blood vessel image feature map to obtain a first complement fusion feature map; performing a feature extraction operation on the first complement fusion feature map to obtain first complement fusion feature information;

[0028] Performing a feature fusion operation on the first complement fusion feature information and the second complement label feature map to obtain a second complement fusion feature map; performing a feature extraction operation on the second complement fusion feature map to obtain second complement fusion feature information;

[0029] Performing a feature fusion operation on the second complement fusion feature information and the third complement label feature map to obtain a third complement fusion feature map; performing a feature extraction operation on the third complement fusion feature map to obtain third complement fusion feature information;

[0030] Performing a feature fusion operation on the third complement fusion feature information and the fourth complement label feature map to obtain a fourth complement fusion feature map; performing a feature extraction operation on the fourth complement fusion feature map to obtain fourth complement fusion feature information;

[0031] Performing a feature fusion operation on the fourth complement fusion feature information and the fifth complement label feature map to obtain a fifth complement fusion feature map;

[0032] The fifth complement fusion feature information is decoded to obtain an image segmentation result; finally, the complement label is used as the supervision information of the image segmentation flow, and the image segmentation flow is updated using the dice loss.

[0033] The second aspect of the present invention provides a noise-annotated blood vessel image segmentation system, comprising a data set acquisition module, a noise label completion module, a dual-stream feature extraction module, and a plurality of feature fusion modules;

[0034] The data set acquisition module is used to acquire a retinal blood vessel data set;

[0035] The noise label completion module is used to input the noise labels and corresponding vascular images in the retinal vascular data set into the pre-trained noise label completion unit, perform label completion processing through feature extraction operation and feature fusion, and obtain the completed label;

[0036] The dual-stream feature extraction module is used to input the completed label and the corresponding blood vessel image into the dual-stream feature extraction network to perform feature extraction operation and feature fusion to obtain the final image segmentation result.

[0037] As a preferred solution, the noise labels and corresponding vascular images in the retinal vascular dataset are input into a pre-trained noise label completion unit, and label completion processing is performed through feature extraction operation and feature fusion. The method for obtaining the completed label includes:

[0038] Performing four feature extraction operations on the noise label to obtain a first label feature map, a second label feature map, a third label feature map, and a fourth label feature map;

[0039] Performing a feature extraction operation on the blood vessel image corresponding to the noise label to obtain a first image feature map;

[0040] Performing a feature fusion operation on the first label feature map and the first image feature map through a feature fusion module to obtain a first fused feature map; performing a feature extraction operation on the first fused feature map to obtain first fused feature information;

[0041] Performing a feature fusion operation on the first fused feature information and the second label feature map through a feature fusion module to obtain a second fused feature map; performing a feature extraction operation on the second fused feature map to obtain second fused feature information;

[0042] The second fused feature information and the third label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a third fused feature map; a feature extraction operation is performed on the third fused feature map to obtain third fused feature information;

[0043] Performing a feature fusion operation on the third fused feature information and the fourth label feature map through a feature fusion module to obtain a fourth fused feature map; performing a feature extraction operation on the fourth fused feature map to obtain fourth fused feature information;

[0044] A decoding operation is performed on the fourth fusion feature information to obtain a completion label.

[0045] As a preferred solution, the method of inputting the completed label and the corresponding vascular image into a pre-trained dual-stream feature extraction network for feature extraction to obtain the final segmentation result includes:

[0046] In the label feature extraction stream of the dual-stream feature extraction network, the complementary label is subjected to five feature extraction operations to obtain a first complementary label feature map, a second complementary label feature map, a third complementary label feature map, a fourth complementary label feature map, and a fifth complementary label feature map; wherein the fifth complementary label feature map is subjected to a decoding operation to obtain a feature prediction result; and then the complementary label is used as the supervision information of the label feature extraction stream;

[0047] In the image segmentation stream of the dual-stream feature extraction network, a feature extraction operation is performed on the blood vessel image to obtain a first blood vessel image feature map;

[0048] Performing a feature fusion operation on the first complement label feature map and the first blood vessel image feature map through a feature fusion module to obtain a first complement fusion feature map; performing a feature extraction operation on the first complement fusion feature map to obtain first complement fusion feature information;

[0049] The first complementary fusion feature information and the second complementary label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a second complementary fusion feature map; a feature extraction operation is performed on the second complementary fusion feature map to obtain second complementary fusion feature information;

[0050] The second complement fusion feature information and the third complement label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a third complement fusion feature map; a feature extraction operation is performed on the third complement fusion feature map to obtain third complement fusion feature information;

[0051] The third complement fusion feature information and the fourth complement label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a fourth complement fusion feature map; a feature extraction operation is performed on the fourth complement fusion feature map to obtain fourth complement fusion feature information;

[0052] The fourth complement fusion feature information and the fifth complement label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a fifth complement fusion feature map;

[0053] The fifth complementary fusion feature information is decoded to obtain an image segmentation result.

[0054] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the aforementioned noise-annotated vascular image segmentation method are implemented.

[0055] A fourth aspect of the present invention provides a computer device, comprising a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor, wherein the computer program implements the steps of the aforementioned noise-annotated vascular image segmentation method when executed by the processor.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The present invention obtains a completed label by inputting the noise label and the corresponding vascular image in the retinal vascular data set into a noise label completion unit for label completion processing, thereby alleviating the label annotation pressure and improving the quality of the noise data set; then, a dual-stream feature extraction network is used to extract features from the completed label and the corresponding vascular image to obtain an image segmentation result, thereby achieving high-precision vascular image segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A flow chart of a noise-annotated blood vessel image segmentation method provided by an embodiment of the present invention;

[0059] Figure 2 A block diagram of a noise-annotated blood vessel image segmentation system provided by an embodiment of the present invention;

[0060] Figure 3 A schematic diagram of a noise label completion module provided in an embodiment of the present invention;

[0061] Figure 4 This is a diagram of the blood vessel image segmentation result provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The accompanying drawings are only used for illustrative purposes and are not to be construed as limiting the present invention;

[0063] It should be clear that the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the embodiments of the present application.

[0064] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0065] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0066] In addition, in the description of this application, unless otherwise specified, "multiple" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0067] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0068] Example 1

[0069] Please refer to Figure 1 This embodiment provides a noise-annotated blood vessel image segmentation method, the method comprising:

[0070] S1: Acquire retinal vascular dataset;

[0071] Specifically, the retinal vascular dataset is a DRIVE retinal vascular dataset;

[0072] S2: Input the noise labels and corresponding vascular images in the retinal vascular dataset into the pre-trained noise label completion unit, perform label completion processing through feature extraction operation and feature fusion, and obtain the completed label;

[0073] In a specific embodiment, the training method of the noise label completion unit includes:

[0074] Performing data enhancement processing on the vascular image data and corresponding labels in the retinal vascular dataset;

[0075] A preset number of clean labels are selected; and a noise label completion unit is trained using the clean labels and the corresponding vascular images.

[0076] In a specific embodiment, the method for processing brightness changes includes:

[0077] A random function is used to generate a random number within a preset numerical range, and a preset value is used as a dividing line. If the value is greater than the preset value, the value is dimmed, and if the value is less than the preset value, the value is brightened. The vascular image data and the corresponding labels in the retinal vascular data set are transformed in brightness. The corresponding expression is:

[0078] S(o)=f(S(i))

[0079] Among them, S(o) is the processed image, S(i) is the original image, and f(·) is the brightness change function.

[0080] In a specific embodiment, the noise labels and corresponding vascular images in the retinal vascular dataset are input into a pre-trained noise label completion unit, and label completion processing is performed through feature extraction operation and feature fusion to obtain the completed label. The method includes:

[0081] Performing four feature extraction operations on the noise label to obtain a first label feature map, a second label feature map, a third label feature map, and a fourth label feature map;

[0082] Performing a feature extraction operation on the blood vessel image corresponding to the noise label to obtain a first image feature map;

[0083] Performing a feature fusion operation on the first label feature map and the first image feature map to obtain a first fused feature map; performing a feature extraction operation on the first fused feature map to obtain first fused feature information;

[0084] Performing a feature fusion operation on the first fused feature information and the second label feature map to obtain a second fused feature map; performing a feature extraction operation on the second fused feature map to obtain second fused feature information;

[0085] Performing a feature fusion operation on the second fused feature information and the third label feature map to obtain a third fused feature map; performing a feature extraction operation on the third fused feature map to obtain third fused feature information;

[0086] Performing a feature fusion operation on the third fused feature information and the fourth label feature map to obtain a fourth fused feature map; performing a feature extraction operation on the fourth fused feature map to obtain fourth fused feature information;

[0087] A decoding operation is performed on the fourth fusion feature information to obtain a completion label.

[0088] S3: inputting the completed label and the corresponding blood vessel image into a dual-stream feature extraction network for feature extraction and feature fusion to obtain a final image segmentation result;

[0089] In a specific embodiment, the method of inputting the completed label and the corresponding blood vessel image into a dual-stream feature extraction network for feature extraction and feature fusion to obtain the final segmentation result includes:

[0090] In the label feature extraction stream of the dual-stream feature extraction network, the completed label is subjected to five feature extraction operations to obtain a first completed label feature map, a second completed label feature map, a third completed label feature map, a fourth completed label feature map, and a fifth completed label feature map; wherein the fifth completed label feature map is subjected to a decoding operation to obtain a feature prediction result; finally, the completed label is used as the supervision information of the label feature extraction stream, and the label feature extraction stream is updated using the dice loss;

[0091] In the image segmentation stream of the dual-stream feature extraction network, a feature extraction operation is performed on the blood vessel image to obtain a first blood vessel image feature map;

[0092] Performing a feature fusion operation on the first complement label feature map and the first blood vessel image feature map to obtain a first complement fusion feature map; performing a feature extraction operation on the first complement fusion feature map to obtain first complement fusion feature information;

[0093] Performing a feature fusion operation on the first complement fusion feature information and the second complement label feature map to obtain a second complement fusion feature map; performing a feature extraction operation on the second complement fusion feature map to obtain second complement fusion feature information;

[0094] Performing a feature fusion operation on the second complement fusion feature information and the third complement label feature map to obtain a third complement fusion feature map; performing a feature extraction operation on the third complement fusion feature map to obtain third complement fusion feature information;

[0095] Performing a feature fusion operation on the third complement fusion feature information and the fourth complement label feature map to obtain a fourth complement fusion feature map; performing a feature extraction operation on the fourth complement fusion feature map to obtain fourth complement fusion feature information;

[0096] Performing a feature fusion operation on the fourth complement fusion feature information and the fifth complement label feature map to obtain a fifth complement fusion feature map;

[0097] The fifth complement fusion feature information is decoded to obtain an image segmentation result; finally, the complement label is used as the supervision information of the image segmentation flow, and the image segmentation flow is updated using the dice loss.

[0098] Example 2

[0099] Please refer to Figure 2 as well as Figure 3 ,This embodiment provides a noise labeled vascular image segmentation system, including a data set acquisition module, a noise label completion module, a dual-stream feature extraction module, and several feature fusion modules;

[0100] The data set acquisition module is used to acquire a retinal blood vessel data set;

[0101] The noise label completion module is used to input the noise labels and corresponding vascular images in the retinal vascular data set into the pre-trained noise label completion unit, perform label completion processing through feature extraction operation and feature fusion, and obtain the completed label;

[0102] The dual-stream feature extraction module is used to input the completed label and the corresponding blood vessel image into the dual-stream feature extraction network to perform feature extraction operation and feature fusion to obtain the final image segmentation result.

[0103] In a specific embodiment, the noise labels and corresponding vascular images in the retinal vascular dataset are input into a pre-trained noise label completion unit, and label completion processing is performed through feature extraction operation and feature fusion to obtain the completed label. The method includes:

[0104] Performing four feature extraction operations on the noise label to obtain a first label feature map, a second label feature map, a third label feature map, and a fourth label feature map;

[0105] Performing a feature extraction operation on the blood vessel image corresponding to the noise label to obtain a first image feature map;

[0106] Performing a feature fusion operation on the first label feature map and the first image feature map through a feature fusion module to obtain a first fused feature map; performing a feature extraction operation on the first fused feature map to obtain first fused feature information;

[0107] Performing a feature fusion operation on the first fused feature information and the second label feature map through a feature fusion module to obtain a second fused feature map; performing a feature extraction operation on the second fused feature map to obtain second fused feature information;

[0108] The second fused feature information and the third label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a third fused feature map; a feature extraction operation is performed on the third fused feature map to obtain third fused feature information;

[0109] Performing a feature fusion operation on the third fused feature information and the fourth label feature map through a feature fusion module to obtain a fourth fused feature map; performing a feature extraction operation on the fourth fused feature map to obtain fourth fused feature information;

[0110] A decoding operation is performed on the fourth fusion feature information to obtain a completion label.

[0111] In a specific embodiment, the method of inputting the completed label and the corresponding blood vessel image into a pre-trained dual-stream feature extraction network for feature extraction to obtain the final segmentation result includes:

[0112] In the label feature extraction stream of the dual-stream feature extraction network, the complementary label is subjected to five feature extraction operations to obtain a first complementary label feature map, a second complementary label feature map, a third complementary label feature map, a fourth complementary label feature map, and a fifth complementary label feature map; wherein the fifth complementary label feature map is subjected to a decoding operation to obtain a feature prediction result; and then the complementary label is used as the supervision information of the label feature extraction stream;

[0113] In the image segmentation stream of the dual-stream feature extraction network, a feature extraction operation is performed on the blood vessel image to obtain a first blood vessel image feature map;

[0114] Performing a feature fusion operation on the first complement label feature map and the first blood vessel image feature map through a feature fusion module to obtain a first complement fusion feature map; performing a feature extraction operation on the first complement fusion feature map to obtain first complement fusion feature information;

[0115] The first complementary fusion feature information and the second complementary label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a second complementary fusion feature map; a feature extraction operation is performed on the second complementary fusion feature map to obtain second complementary fusion feature information;

[0116] The second complement fusion feature information and the third complement label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a third complement fusion feature map; a feature extraction operation is performed on the third complement fusion feature map to obtain third complement fusion feature information;

[0117] The third complement fusion feature information and the fourth complement label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a fourth complement fusion feature map; a feature extraction operation is performed on the fourth complement fusion feature map to obtain fourth complement fusion feature information;

[0118] The fourth complement fusion feature information and the fifth complement label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a fifth complement fusion feature map;

[0119] The fifth complementary fusion feature information is decoded to obtain an image segmentation result.

[0120] Example 3

[0121] This example verifies and analyzes the method, more specifically:

[0122] Please refer to Figure 4 The present invention performs image enhancement on the training images of the DRIVE retinal vascular dataset, uses 100 samples as the training set, adds noise to the training set, trains our model in datasets with different noise contents and different noise addition degrees, obtains the final segmentation result through the model, and uses the same dataset to train three existing segmentation methods (respectively: UNet, SGL, MTCL), and compares them with the other three segmentation methods. Our method achieves the best in most cases, and can withstand different levels of noise. The performance under different noise conditions is good, as shown in Table 1, where α represents the proportion of noise data in the dataset, and β represents the noise addition degree of the noise label.

[0123] Table 1

[0124]

[0125] The segmentation results obtained by the present invention and other methods under different noise conditions are as follows Figure 4 shown.

[0126] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A noise-annotated vascular image segmentation method, characterized in that: The method comprises: Acquire retinal vascular dataset; The noise labels and corresponding vascular images in the retinal vascular dataset are input into the pre-trained noise label completion unit, and the label completion process is performed through feature extraction and feature fusion to obtain the completed labels; Inputting the completed label and the corresponding blood vessel image into a dual-stream feature extraction network for feature extraction and feature fusion to obtain a final image segmentation result; The training method of the noise label completion unit includes: Performing data enhancement processing on the vascular image data and corresponding labels in the retinal vascular dataset; Screening a preset number of clean labels; using the clean labels and corresponding vascular images to train a noise label completion unit; The noise labels and corresponding vascular images in the retinal vascular dataset are input into the pre-trained noise label completion unit, and the label completion process is performed through feature extraction operation and feature fusion. The method for obtaining the completed label includes: Performing four feature extraction operations on the noise label to obtain a first label feature map, a second label feature map, a third label feature map, and a fourth label feature map; Performing a feature extraction operation on the blood vessel image corresponding to the noise label to obtain a first image feature map; Performing a feature fusion operation on the first label feature map and the first image feature map to obtain a first fused feature map; performing a feature extraction operation on the first fused feature map to obtain first fused feature information; Performing a feature fusion operation on the first fused feature information and the second label feature map to obtain a second fused feature map; performing a feature extraction operation on the second fused feature map to obtain second fused feature information; Performing a feature fusion operation on the second fused feature information and the third label feature map to obtain a third fused feature map; performing a feature extraction operation on the third fused feature map to obtain third fused feature information; Performing a feature fusion operation on the third fused feature information and the fourth label feature map to obtain a fourth fused feature map; performing a feature extraction operation on the fourth fused feature map to obtain fourth fused feature information; Performing a decoding operation on the fourth fusion feature information to obtain a completion label; The method of inputting the completed label and the corresponding blood vessel image into a dual-stream feature extraction network for feature extraction and feature fusion to obtain the final segmentation result includes: In the label feature extraction stream of the dual-stream feature extraction network, the completed label is subjected to five feature extraction operations to obtain a first completed label feature map, a second completed label feature map, a third completed label feature map, a fourth completed label feature map, and a fifth completed label feature map; wherein the fifth completed label feature map is subjected to a decoding operation to obtain a feature prediction result; finally, the completed label is used as the supervision information of the label feature extraction stream, and the label feature extraction stream is updated using the dice loss; In the image segmentation stream of the dual-stream feature extraction network, a feature extraction operation is performed on the blood vessel image to obtain a first blood vessel image feature map; Performing a feature fusion operation on the first complement label feature map and the first blood vessel image feature map to obtain a first complement fusion feature map; performing a feature extraction operation on the first complement fusion feature map to obtain first complement fusion feature information; Performing a feature fusion operation on the first complement fusion feature information and the second complement label feature map to obtain a second complement fusion feature map; performing a feature extraction operation on the second complement fusion feature map to obtain second complement fusion feature information; Performing a feature fusion operation on the second complement fusion feature information and the third complement label feature map to obtain a third complement fusion feature map; performing a feature extraction operation on the third complement fusion feature map to obtain third complement fusion feature information; Performing a feature fusion operation on the third complement fusion feature information and the fourth complement label feature map to obtain a fourth complement fusion feature map; performing a feature extraction operation on the fourth complement fusion feature map to obtain fourth complement fusion feature information; Performing a feature fusion operation on the fourth complement fusion feature information and the fifth complement label feature map to obtain a fifth complement fusion feature map; The fifth complement fusion feature information is decoded to obtain an image segmentation result; finally, the complement label is used as the supervision information of the image segmentation flow, and the image segmentation flow is updated by using the dice loss; The noise labels and corresponding vascular images in the retinal vascular dataset are input into the pre-trained noise label completion unit, and the label completion process is performed through feature extraction operation and feature fusion. The method for obtaining the completed label includes: Performing four feature extraction operations on the noise label to obtain a first label feature map, a second label feature map, a third label feature map, and a fourth label feature map; Performing a feature extraction operation on the blood vessel image corresponding to the noise label to obtain a first image feature map; Performing a feature fusion operation on the first label feature map and the first image feature map through a feature fusion module to obtain a first fused feature map; performing a feature extraction operation on the first fused feature map to obtain first fused feature information; Performing a feature fusion operation on the first fused feature information and the second label feature map through a feature fusion module to obtain a second fused feature map; performing a feature extraction operation on the second fused feature map to obtain second fused feature information; The second fused feature information and the third label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a third fused feature map; a feature extraction operation is performed on the third fused feature map to obtain third fused feature information; Performing a feature fusion operation on the third fused feature information and the fourth label feature map through a feature fusion module to obtain a fourth fused feature map; performing a feature extraction operation on the fourth fused feature map to obtain fourth fused feature information; A decoding operation is performed on the fourth fusion feature information to obtain a completion label.

2. The noise-annotated vascular image segmentation method according to claim 1, characterized in that: The data enhancement processing includes random flipping processing and brightness change processing; wherein the brightness change processing method includes: A random function is used to generate a random number within a preset numerical range, and a preset value is used as a dividing line. If the value is greater than the preset value, the value is dimmed, and if the value is less than the preset value, the value is brightened. The vascular image data and the corresponding labels in the retinal vascular data set are transformed in brightness. The corresponding expression is: S(o)=f(S(i)) Among them, S(o) is the processed image, S(i) is the original image, and f(·) is the brightness change function.

3. A noise-annotated vascular image segmentation system, characterized in that: It includes a data set acquisition module, a noise label completion module, a dual-stream feature extraction module, and several feature fusion modules; The data set acquisition module is used to acquire a retinal blood vessel data set; The noise label completion module is used to input the noise labels and corresponding vascular images in the retinal vascular data set into the pre-trained noise label completion unit, perform label completion processing through feature extraction operation and feature fusion, and obtain the completed label; The dual-stream feature extraction module is used to input the completed label and the corresponding blood vessel image into the dual-stream feature extraction network to perform feature extraction and feature fusion to obtain the final image segmentation result; The noise labels and corresponding vascular images in the retinal vascular dataset are input into the pre-trained noise label completion unit, and the label completion process is performed through feature extraction operation and feature fusion. The method for obtaining the completed label includes: Performing four feature extraction operations on the noise label to obtain a first label feature map, a second label feature map, a third label feature map, and a fourth label feature map; Performing a feature extraction operation on the blood vessel image corresponding to the noise label to obtain a first image feature map; Performing a feature fusion operation on the first label feature map and the first image feature map through a feature fusion module to obtain a first fused feature map; performing a feature extraction operation on the first fused feature map to obtain first fused feature information; Performing a feature fusion operation on the first fused feature information and the second label feature map through a feature fusion module to obtain a second fused feature map; performing a feature extraction operation on the second fused feature map to obtain second fused feature information; The second fused feature information and the third label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a third fused feature map; a feature extraction operation is performed on the third fused feature map to obtain third fused feature information; Performing a feature fusion operation on the third fused feature information and the fourth label feature map through a feature fusion module to obtain a fourth fused feature map; performing a feature extraction operation on the fourth fused feature map to obtain fourth fused feature information; Performing a decoding operation on the fourth fusion feature information to obtain a completion label; The method of inputting the completed label and the corresponding blood vessel image into a pre-trained two-stream feature extraction network for feature extraction to obtain the final segmentation result includes: In the label feature extraction stream of the dual-stream feature extraction network, the complementary label is subjected to five feature extraction operations to obtain a first complementary label feature map, a second complementary label feature map, a third complementary label feature map, a fourth complementary label feature map, and a fifth complementary label feature map; wherein the fifth complementary label feature map is subjected to a decoding operation to obtain a feature prediction result; and then the complementary label is used as the supervision information of the label feature extraction stream; In the image segmentation stream of the dual-stream feature extraction network, a feature extraction operation is performed on the blood vessel image to obtain a first blood vessel image feature map; Performing a feature fusion operation on the first complement label feature map and the first blood vessel image feature map through a feature fusion module to obtain a first complement fusion feature map; performing a feature extraction operation on the first complement fusion feature map to obtain first complement fusion feature information; The first complementary fusion feature information and the second complementary label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a second complementary fusion feature map; a feature extraction operation is performed on the second complementary fusion feature map to obtain second complementary fusion feature information; The second complement fusion feature information and the third complement label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a third complement fusion feature map; a feature extraction operation is performed on the third complement fusion feature map to obtain third complement fusion feature information; The third complement fusion feature information and the fourth complement label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a fourth complement fusion feature map; a feature extraction operation is performed on the fourth complement fusion feature map to obtain fourth complement fusion feature information; The fourth complement fusion feature information and the fifth complement label feature map are subjected to a feature fusion operation through a feature fusion module to obtain a fifth complement fusion feature map; The fifth complementary fusion feature information is decoded to obtain an image segmentation result.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the noise-annotated blood vessel image segmentation method according to any one of claims 1 to 2 are implemented.

5. A computer device, characterized in that: The method comprises a storage medium, a processor and a computer program stored in the storage medium and executable by the processor, wherein when the computer program is executed by the processor, the steps of the noise-annotated blood vessel image segmentation method according to any one of claims 1 to 2 are implemented.

Images