A blood vessel image recognition method and device, electronic equipment and storage medium

Abstract

The application provides a blood vessel image recognition method and device, electronic equipment and storage medium; the method comprises the following steps: obtaining a to-be-recognized image of a to-be-recognized object; performing blood vessel recognition on the to-be-recognized image to obtain a blood vessel segmentation graph corresponding to the to-be-recognized image; performing arteriovenous recognition on the to-be-recognized image according to the blood vessel segmentation graph to obtain an arteriovenous segmentation graph corresponding to the to-be-recognized image; and performing post-processing on the arteriovenous segmentation graph according to the blood vessel segmentation graph to obtain a blood vessel three-dimensional reconstruction result of the to-be-recognized object. In this way, the blood vessel segmentation graph obtained in the previous stage can be effectively utilized in the process of arteriovenous recognition, the arteriovenous recognition is performed on the to-be-recognized image based on the blood vessel segmentation graph, and the accuracy of arteriovenous recognition is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to medical imaging technology, and in particular to a blood vessel image recognition method and device, an electronic device, and a storage medium. BACKGROUND

[0002] In a clinical operation, it is often necessary to evaluate the patient's vascular anatomy to determine the operation type and entry, especially the differentiation of arteriovenous vessels. The conventional method is to use CT images combined with some medical imaging tools to mark the arteriovenous vessels layer by layer, which requires a large amount of manpower to complete the identification of arteriovenous vessels. The conventional deep learning arteriovenous vessel segmentation method often cannot accurately distinguish arteriovenous vessels, and is accompanied by problems such as false identification, and the accuracy of arteriovenous vessel identification is low. SUMMARY

[0003] The present application provides a blood vessel image recognition method and device, an electronic device, and a storage medium.

[0004] According to a first aspect of the present application, a blood vessel image recognition method is provided, the method comprising: acquiring a to-be-recognized image of a to-be-recognized object; performing blood vessel recognition on the to-be-recognized image to obtain a blood vessel segmentation map corresponding to the to-be-recognized image; performing arteriovenous recognition on the to-be-recognized image according to the blood vessel segmentation map to obtain an arteriovenous segmentation map corresponding to the to-be-recognized image; and performing post-processing on the arteriovenous segmentation map according to the blood vessel segmentation map to obtain a blood vessel three-dimensional reconstruction result of the to-be-recognized object.

[0005] According to an embodiment of the present application, the acquiring of the to-be-recognized image of the to-be-recognized object comprises: acquiring an original image of the to-be-recognized object; and performing preprocessing on the original image to obtain the to-be-recognized image, the preprocessing comprising normalization processing, ROI region segmentation processing, image cropping processing, and data enhancement processing.

[0006] According to an embodiment of the present application, the performing of blood vessel recognition on the to-be-recognized image to obtain a blood vessel segmentation map corresponding to the to-be-recognized image comprises: performing downsampling processing and upsampling processing on the to-be-recognized image to obtain a first feature map, the downsampling processing comprising convolution processing and attention feature extraction on the to-be-recognized image; determining blood vessel pixel points and non-blood vessel pixel points in the first feature map; and determining the blood vessel segmentation map corresponding to the to-be-recognized image according to the blood vessel pixel points and the non-blood vessel pixel points.

[0007] According to an embodiment of the present application, the vein and artery recognition on the to-be-recognized image according to the blood vessel segmentation map comprises: performing feature extraction on the to-be-recognized image to obtain an initial feature map; performing feature fusion on the blood vessel segmentation map and the initial feature map to obtain a target feature map; and performing vein and artery pixel point prediction on the target feature map to obtain the vein and artery segmentation map corresponding to the to-be-recognized image.

[0008] According to an embodiment of the present application, the feature extraction on the to-be-recognized image to obtain an initial feature map comprises: performing convolutional down-sampling processing on the to-be-recognized image for a set number of times to obtain a down-sampling result, the down-sampling result comprising a second feature map output by each convolutional down-sampling processing; performing convolutional up-sampling processing on the second feature map output by the last convolutional down-sampling processing according to the second feature map output by each convolutional down-sampling processing; and obtaining the output of the current convolutional up-sampling processing to obtain an initial feature map when the number of times of the convolutional up-sampling processing reaches a first set number of times.

[0009] According to an embodiment of the present application, the feature fusion on the blood vessel segmentation map and the initial feature map to obtain a target feature map comprises: obtaining the feature dimension of the initial feature map; performing convolutional processing and pooling processing on the blood vessel segmentation map according to the feature dimension to obtain a third feature map; and performing feature fusion on the initial feature map and the third feature map to obtain a target feature map.

[0010] According to an embodiment of the present application, the vein and artery pixel point prediction on the target feature map to obtain the vein and artery segmentation map corresponding to the to-be-recognized image comprises: determining a second set number of times according to the set number of times and the first set number of times; performing convolutional up-sampling processing on the target feature map for a second set number of times according to the second feature map output by each convolutional down-sampling processing and the target feature map to obtain a final feature map; determining target artery pixel points, target vein pixel points and target non-blood vessel pixel points in the final feature map; determining the vein and artery segmentation map according to the target artery pixel points, the target vein pixel points and the target non-blood vessel pixel points; and the sum of the first set number of times and the second set number of times has a determined correlation with the set number of times.

[0011] According to an embodiment of the present application, the post-processing comprises vein and artery hole filling, maximum connected domain processing and surface rendering reconstruction.

[0012] According to a second aspect of the present application, a blood vessel image recognition device is provided, which comprises: an acquisition module configured to acquire a to-be-recognized image of a to-be-recognized object; a recognition module configured to perform blood vessel recognition on the to-be-recognized image to obtain a blood vessel segmentation map corresponding to the to-be-recognized image; a determination module configured to determine an artery and vein segmentation map corresponding to the to-be-recognized image according to the blood vessel segmentation map and the to-be-recognized image; and a post-processing module configured to perform post-processing on the artery and vein segmentation map according to the blood vessel segmentation map to obtain a blood vessel three-dimensional reconstruction result of the to-be-recognized object.

[0013] According to a third aspect of the present application, an electronic device is provided, which comprises:

[0014] at least one processor; and

[0015] a memory connected to the at least one processor in communication; wherein

[0016] the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the present application.

[0017] According to a fourth aspect of the present application, a non-transitory computer readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to perform the method described in the present application.

[0018] The method of the embodiments of the present application acquires a to-be-recognized image of a to-be-recognized object; performs blood vessel recognition on the to-be-recognized image to obtain a blood vessel segmentation map corresponding to the to-be-recognized image; performs artery and vein recognition on the to-be-recognized image according to the blood vessel segmentation map to obtain an artery and vein segmentation map corresponding to the to-be-recognized image; and performs post-processing on the artery and vein segmentation map to obtain a blood vessel three-dimensional reconstruction result of the to-be-recognized object. By performing blood vessel recognition on the to-be-recognized image to obtain a blood vessel segmentation map, then performing artery and vein segmentation on the to-be-recognized image with reference to the blood vessel segmentation map, and then performing post-processing on the artery and vein segmentation map to obtain a three-dimensional reconstruction result, the blood vessel segmentation map obtained in the previous stage can be effectively utilized in the process of artery and vein recognition, and the accuracy of artery and vein recognition is improved.

[0019] It should be understood that the teachings of the present application do not require all the beneficial effects described above to be achieved, but specific technical solutions can achieve specific technical effects, and other embodiments of the present application can also achieve beneficial effects not mentioned above. BRIEF DESCRIPTION OF DRAWINGS

[0020] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description read in conjunction with the accompanying drawings, in which:

[0021] In the drawings, identical or corresponding components are denoted by identical or corresponding reference numerals.

[0022] Figure 1 An implementation flowchart of a blood vessel image recognition method provided by an embodiment of the present application is shown;

[0023] Figure 2 An implementation flowchart of a blood vessel image recognition method provided by an embodiment of the present application is shown;

[0024] Figure 3 An implementation flowchart of an artery and vein recognition operation of a blood vessel image recognition method provided by an embodiment of the present application is shown;

[0025] Figure 4 An implementation flowchart of a specific application example of a blood vessel image recognition method provided by an embodiment of the present application is shown;

[0026] Figure 5 An example diagram of a three-dimensional reconstruction result obtained based on a blood vessel image recognition method provided by an embodiment of the present application is shown;

[0027] Figure 6 An implementation flowchart of a blood vessel image recognition method provided by an embodiment of the present application is shown;

[0028] Figure 7 An implementation flowchart of a blood vessel image recognition method provided by an embodiment of the present application is shown; DETAILED DESCRIPTION

[0029] In order to make the purposes, features and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0030] Figure 1 An implementation flowchart of a blood vessel image recognition method provided by an embodiment of the present application is shown;

[0031] REFERENCE Figure 1The embodiment of the present application provides a blood vessel image recognition method, which comprises the following steps: in operation 101, obtaining an image to be recognized of an object to be recognized; in operation 102, performing blood vessel recognition on the image to be recognized to obtain a blood vessel segmentation map corresponding to the image to be recognized; in operation 103, performing arteriovenous recognition on the image to be recognized according to the blood vessel segmentation map to obtain an arteriovenous segmentation map corresponding to the image to be recognized; and in operation 104, performing post-processing on the arteriovenous segmentation map according to the blood vessel segmentation map to obtain a blood vessel three-dimensional reconstruction result of the object to be recognized.

[0032] In operation 101, an image to be recognized of an object to be recognized is obtained.

[0033] In an embodiment of the present application, the image to be recognized of the object to be recognized is obtained by the following operations: obtaining an original image of the object to be recognized, and performing preprocessing on the original image to obtain the image to be recognized. The preprocessing includes normalization processing, ROI region segmentation processing and image cutting processing.

[0034] Specifically, the original image of the object to be recognized can be a CT image of the object to be recognized. The CT image can be a CT image of different organs containing arteriovenous vessels, such as a lung CT image, a kidney CT image, a liver CT image and the like.

[0035] Further, in order to eliminate the redundant information in the CT image, restore the useful real information, enhance the detectability of the relevant information and simplify the data to the maximum, thereby increasing the reliability of the CT image recognition, it is generally necessary to perform preprocessing on the CT image. The preprocessing generally includes normalization processing, ROI region segmentation processing, image cutting processing and data enhancement processing.

[0036] The normalization processing can eliminate the difference of the CT image caused by the different CT parameters of different manufacturers. The standard of the normalization processing is different for the CT images of different organs. Different normalization window width and window level can be set for each CT image according to different parts corresponding to the CT image. For example, for a lung CT image, the window level WL is set to WL=-600 and the window width WW is set to 1200; for a liver CT image, the window level is set to WL=100 and the window width WW is set to 200; and for a kidney CT image, the window level is set to WL=0 and the window width WW is set to 200. The CT images of other parts except the above parts can be individually set by referring to the window width and window level actually used in the clinic.

[0037] The ROI region segmentation processing can remove the useless information recorded in the CT image, and roughly obtain the region where the blood vessels are located, so as to more accurately identify the blood vessels. The ROI region segmentation standards are different for CT images of different parts. For example, when identifying blood vessels in a lung CT image, the ROI region segmentation operation includes: obtaining the region where the lung parenchyma is located from the region below the threshold value -500 of the lung CT image, and then using a morphological closing operation image processing method to obtain a complete lung region, which contains all the blood vessels to be identified, that is, the ROI region. For parts other than the lung, a simple positioning method can be designed according to the actual organ where the blood vessels are located to obtain the ROI region where the blood vessels are located.

[0038] The image cropping processing can make the CT image better recognized. Generally, the size of the CT image is

D, H, W

[0039] The data enhancement processing can obtain diverse information for the CT image. The data enhancement processing method can be random flipping, salt and pepper noise, and contrast enhancement.

[0040] After the CT image is preprocessed, a to-be-identified image is obtained.

[0041] In operation 102, the to-be-identified image is subjected to blood vessel identification, and a blood vessel segmentation image corresponding to the to-be-identified image is obtained.

[0042] Specifically, a commonly used target segmentation model can be used to identify the blood vessels of the to-be-identified image after simple training, and a blood vessel segmentation image is obtained. The to-be-identified image can also be subjected to blood vessel identification by using a commonly used image segmentation method, such as threshold segmentation, region segmentation or edge segmentation, to obtain a blood vessel segmentation image. The blood vessel segmentation image includes a blood vessel region and a non-blood vessel region.

[0043] In operation 103, according to the blood vessel segmentation image, the to-be-identified image is subjected to arteriovenous identification, and an arteriovenous segmentation image corresponding to the to-be-identified image is obtained.

[0044] Specifically, when the to-be-identified image is subjected to arteriovenous segmentation, the blood vessel region shown in the blood vessel segmentation image can be referred to, and the arteriovenous identification region is focused on the blood vessel region of the to-be-identified image, so as to improve the accuracy of arteriovenous identification.

[0045] In an embodiment of the present application, the arteriovenous recognition of the blood vessel region can also be directly based on the blood vessel segmentation map, to obtain an arteriovenous segmentation map of the to-be-recognized image, which includes an arterial region, a venous region and a background region. The arteriovenous recognition of the blood vessel region can be performed by a commonly used target segmentation method or by pre-training an identification model for arteriovenous recognition.

[0046] Operation 104: Post-processing the arteriovenous segmentation map to obtain a blood vessel three-dimensional reconstruction result of the to-be-recognized object.

[0047] Specifically, after obtaining the blood vessel segmentation map and the arteriovenous segmentation map, in order to complete the final three-dimensional reconstruction of the to-be-recognized object, a certain post-processing needs to be performed on the arteriovenous segmentation map.

[0048] In an embodiment of the present application, the post-processing includes arteriovenous hole filling, maximum connected domain processing and surface rendering reconstruction.

[0049] Specifically, the arteriovenous segmentation map is prone to pixel loss at the pixel level, thereby causing arteriovenous holes and leading to incomplete reconstruction morphology. Therefore, in combination with the blood vessel segmentation map, for the missing of the blood vessel segmentation map in the arteriovenous segmentation map, the missing arteriovenous label is given according to the distance of the missing distance arteriovenous distance to complete the filling of the missing pixels. The distance calculation method used is the Euclidean distance.

[0050] Further, the arterial blood vessel or the venous blood vessel usually has a complete single connected domain, and the pixels that are free in the edge of the arterial blood vessel or the venous blood vessel often exist in the arteriovenous segmentation result. In order to filter this part of noise, the maximum connected domain preserving processing needs to be performed on the arteriovenous segmentation map.

[0051] After the venous hole filling and the maximum connected domain processing of the arteriovenous segmentation result, the arteries and veins on the arteriovenous segmentation map are converted into 3D models to obtain the final three-dimensional reconstruction result. The classic surface reconstruction algorithm (Marching Cube) can be used to convert the arteriovenous segmentation map into a 3D model, thereby obtaining the final three-dimensional reconstruction result, which includes an arterial 3D model and a venous 3D model of the to-be-recognized object.

[0052] Therefore, in the embodiments of the present application, the blood vessel recognition is performed first, and then the arteriovenous recognition is performed based on the blood vessel segmentation map obtained by the blood vessel recognition. This avoids the problem of low arteriovenous recognition accuracy caused by the similarity or high adhesion degree of the arteriovenous blood vessels, and improves the accuracy of arteriovenous segmentation.

[0053] Figure 2 A flowchart of a blood vessel image recognition operation provided by an embodiment of the present application is shown.

[0054] Reference Figure 2 In an embodiment of the present application, operation 102, the blood vessel recognition is performed on the to-be-recognized image to obtain a blood vessel segmentation map corresponding to the to-be-recognized image, comprising: operation 201, downsampling processing and upsampling processing are performed on the to-be-recognized image to obtain a first feature map, the downsampling processing comprising convolution processing and attention feature extraction on the to-be-recognized image; operation 202, blood vessel pixel points and non-blood vessel pixel points in the first feature map are determined; operation 203, the blood vessel segmentation map corresponding to the to-be-recognized image is determined according to the blood vessel pixel points and the non-blood vessel pixel points.

[0055] In operation 201, downsampling processing and upsampling processing are performed on the to-be-recognized image to obtain a first feature map, the downsampling processing comprising convolution processing and attention feature extraction on the to-be-recognized image.

[0056] Specifically, in order to better recognize the blood vessels of the to-be-recognized image, feature extraction can be performed on the to-be-recognized image through upsampling and downsampling to obtain a first feature map for the to-be-recognized image. In the process of feature extraction on the to-be-recognized image, first, downsampling processing is performed, and then upsampling processing is performed.

[0057] In an embodiment of the present application, in the process of downsampling processing on the to-be-recognized image, attention feature extraction is also added to capture more features.

[0058] In operation 202, blood vessel pixel points and non-blood vessel pixel points in the first feature map are determined.

[0059] Specifically, for a plurality of pixel points in the first feature map, the probabilities of the plurality of pixel points belonging to a blood vessel region and a non-blood vessel region are predicted to obtain blood vessel pixel points and non-blood vessel pixel points.

[0060] In operation 203, the blood vessel segmentation map corresponding to the to-be-recognized image is determined according to the blood vessel pixel points and the non-blood vessel pixel points.

[0061] Specifically, the blood vessel pixel points and the non-blood vessel pixel points can be displayed in different forms respectively to obtain a blood vessel segmentation map, which includes a blood vessel region and a non-blood vessel region.

[0062] In an embodiment of the present application, the blood vessel segmentation map can be a blood vessel binary image, and the blood vessel pixel points and the non-blood vessel pixel points correspond to different gray values respectively.

[0063] In an embodiment of the present application, the above operations 201-203 can be implemented based on a pre-trained blood vessel recognition model.

[0064] Specifically, first, a plurality of CT images are acquired, the plurality of CT images are labeled, the non-vascular region, the arterial region and the venous region in each CT image are labeled, the arterial region and the venous region are defined as the blood vessel region in the training process, the basic model is trained, and a blood vessel recognition model is obtained. The plurality of CT images acquired can be CT images preprocessed based on original CT images. The preprocessing of the original CT images can refer to the preprocessing operation in operation 101, which will not be described here. The network of the basic model can be implemented by V-Net, or can basically use V-Net, and the intermediate convolution module of V-Net is replaced by an attention feature extraction module.

[0065] Further, the network for training the basic model is a supervised training method, and the supervision label is a single-channel binary image, wherein 0 represents the background (i.e. the non-vascular region), and 1 represents the blood vessel (i.e. the blood vessel region, which includes the venous region and the arterial region). The training of the basic model includes training and testing. In the case where the recall rate of the blood vessel reaches a set condition, it is determined that the model training is completed, and a blood vessel recognition model is obtained. The set condition can be configured according to actual conditions. For example, the set condition can be configured as a model recall rate of 95% or more.

[0066] After the blood vessel recognition model is trained, the image to be recognized is input into the blood vessel recognition model, and operations 201-203 are implemented to obtain a blood vessel segmentation map.

[0067] For example, the network structure of V-Net includes an input module, a compression module, a decompression module and an output module. The compression module includes four-stage convolution downsampling, and the decompression module includes four-stage convolution upsampling. The network of the blood vessel recognition model of the embodiment of the application is obtained by adding an attention feature extraction module to the compression module of V-Net. Therefore, compared with V-Net, the blood vessel recognition model further performs attention feature extraction in the compression module at a set stage. The stage at which the attention feature extraction is performed can be set according to actual conditions, and can be performed after each stage of convolution downsampling or after each stage of convolution upsampling.

[0068] The image to be recognized is input into the blood vessel recognition model, the image to be recognized is subjected to upsampling processing and attention feature extraction in the compression module, and then the image to be recognized is subjected to downsampling processing in the decompression stage. After the downsampling is completed, a feature map for classification is obtained, the output module predicts the blood vessel pixel points and the non-vascular pixel points in the feature map, and a blood vessel segmentation map including the blood vessel region and the non-vascular region is obtained.

[0069] It should be noted that the present application only takes the above network structure as an implementable manner of the blood vessel recognition model of the embodiments of the present application, and the implementation of operations 201-203 is described, and other network structures can also be used to implement operations 201-203, which are not limited in the present application.

[0070] Therefore, the embodiments of the present application perform blood vessel recognition on the to-be-recognized image by training the blood vessel segmentation model in advance, which avoids the problem of large subjectivity and easy mislabeling of manual labeling compared with the conventional manual labeling method, and improves the accuracy of blood vessel recognition.

[0071] Figure 3 An implementation flowchart of the arteriovenous recognition operation of the blood vessel image recognition method provided by the embodiments of the present application is shown.

[0072] Reference Figure 3 In an embodiment of the present application, the operation 103 described above performs arteriovenous recognition on the to-be-recognized image according to the blood vessel segmentation map to obtain the arteriovenous segmentation map corresponding to the to-be-recognized image, which includes: operation 301, performing feature extraction on the to-be-recognized image to obtain an initial feature map; operation 302, performing feature fusion on the blood vessel segmentation map and the initial feature map to obtain a target feature map; and operation 303, performing arteriovenous recognition on the target feature map to obtain the arteriovenous segmentation map corresponding to the to-be-recognized image.

[0073] In operation 301, feature extraction is performed on the to-be-recognized image to obtain an initial feature map.

[0074] Specifically, in the process of arteriovenous recognition of the to-be-recognized image, direct recognition of the arteriovenous in the to-be-recognized image may not be able to accurately distinguish the arteriovenous, and therefore, in the process of arteriovenous recognition of the to-be-recognized image, the arteriovenous recognition can be made easier and more accurate by adding prior knowledge.

[0075] Further, after obtaining the blood vessel segmentation map, the blood vessel segmentation map includes the blood vessel region and the non-blood vessel region, and the blood vessel segmentation map can be introduced as prior knowledge when the to-be-recognized image is arteriovenous recognized, so that the arteriovenous recognition is focused on the blood vessel region of the to-be-recognized image.

[0076] In this embodiment of the present application, the prior knowledge is introduced in the form of feature combination, and the arteriovenous recognition is performed by fusing the features of the blood vessel segmentation map and the to-be-recognized image and predicting the arteriovenous region of the fused features. In the process of arteriovenous recognition, the blood vessel segmentation image is introduced as local feature, that is, the global feature of the to-be-recognized image is considered, and the local feature of the blood vessel segmentation image is also considered.

[0077] Specifically, the features of the to-be-recognized image are first extracted to obtain an initial feature map.

[0078] In an embodiment of the present application, the initial feature extraction map is obtained by performing convolution down-sampling processing on the to-be-identified image for a set number of times to obtain a second feature map output by each convolution down-sampling processing, and performing first set number of times of convolution up-sampling processing on the second feature map output by the last convolution down-sampling to obtain an initial feature extraction result.

[0079] Specifically, in order to extract the initial feature map of the to-be-identified image, the to-be-identified image can be subjected to set number of times of convolution up-sampling processing and first set number of times of convolution up-sampling processing. The conventional feature extraction process can be set number of times of convolution down-sampling processing and set number of times of convolution up-sampling processing, each convolution down-sampling processing outputs a second feature map obtained by each convolution down-sampling, and each convolution up-sampling processing uses the second feature map output by the corresponding convolution down-sampling processing. The embodiment of the present application introduces the blood vessel segmentation map in the conventional feature extraction process, and performs feature fusion on the blood vessel segmentation map and the feature map obtained in the intermediate process, and then performs feature extraction based on the feature map after feature fusion.

[0080] Further, the first set number can be determined according to the set number and actual conditions. For example, the set number can be configured as 4 times, and in order to better extract the fused features after feature fusion, the first set number can be configured as 2 times or 3 times.

[0081] In operation 302, the blood vessel segmentation map and the initial feature map are subjected to feature fusion to obtain a target feature map.

[0082] Specifically, after obtaining the initial feature map, the blood vessel segmentation map and the initial feature map are subjected to feature fusion to obtain a target feature map.

[0083] In an embodiment of the present application, the target feature map is obtained by obtaining the feature dimension of the initial feature map, performing convolution processing and pooling processing on the blood vessel segmentation map according to the feature dimension to obtain a third feature map, and performing feature fusion on the initial feature map and the third feature map to obtain the target feature map.

[0084] Specifically, the dimension of the feature map needs to be considered in the feature fusion process to ensure that the dimensions of the two feature maps to be fused are the same. Therefore, after obtaining the initial feature map, the dimension of the initial feature map needs to be obtained.

[0085] Further, the dimension of the feature map can be changed by convolution processing and pooling processing. For the blood vessel segmentation map, convolution processing and pooling processing are adopted to make the dimension of the blood vessel segmentation map the same as the dimension of the initial feature map, and a third feature map is obtained.

[0086] The third feature map is fused with the initial feature map to obtain a target feature map. The feature fusion can be feature addition, for example, by means of add (element-wise add, tensor addition).

[0087] In operation 303, vein pixel points are predicted based on the target feature map to obtain a vein segmentation map corresponding to the to-be-identified image.

[0088] Specifically, after the vein segmentation map and the initial feature map are fused, vein pixel points are predicted based on the obtained target feature map.

[0089] In an embodiment of the present application, the following operations are used to predict vein pixel points based on the target feature map: a second set number of times is determined according to the set number of times and the first set number of times, the target feature map is subjected to convolution up-sampling processing for the second set number of times according to the second feature map output by each convolution down-sampling processing and the target feature map, a final feature map is obtained, target artery pixel points, target vein pixel points and target non-vein pixel points in the final feature map are determined, and a vein segmentation map is determined according to the target artery pixel points, the target vein pixel points and the target non-vein pixel points. The sum of the first set number of times and the second set number of times has a determined correlation with the set number of times.

[0090] Specifically, after the convolution down-sampling processing for the set number of times is performed, equivalent convolution up-sampling processing is required to restore the feature map after feature extraction to the same dimension as the input of the convolution down-sampling processing.

[0091] Further, in the process of extracting the initial feature map, the convolution down-sampling processing for the set number of times and the convolution up-sampling processing for the first set number of times are performed, and the convolution up-sampling processing for the second set number of times is also required, and the sum of the second set number of times and the first set number of times is the set number of times. Therefore, the convolution up-sampling processing for the second set number of times is performed on the target feature map obtained by feature fusion to obtain a final feature map.

[0092] Based on the final feature map, vein pixel points are predicted, including prediction of target artery pixel points, target vein pixel points and target non-vein pixel points in the final feature map, and finally a vein segmentation map is obtained based on the predicted target artery pixel points, target vein pixel points and target non-vein pixel points.

[0093] In an embodiment of the present application, the operations 301-303 described above can be implemented based on a pre-trained arteriovenous recognition model. The arteriovenous recognition model can be obtained by fine-tuning the blood vessel recognition model trained in operations 201-203 and introducing prior knowledge. After obtaining the arteriovenous recognition model, the to-be-recognized image and the blood vessel segmentation map obtained in operations 201-203 are input into the arteriovenous recognition model to obtain the arteriovenous segmentation map of the to-be-recognized object.

[0094] Specifically, the parameters of the feature extraction of the blood vessel recognition model can be initialized and fixed, the same training set as the blood vessel recognition model is used to fine-tune the latter half of the network of the blood vessel recognition model, so that the blood vessel recognition capability of the blood vessel recognition model is changed into arteriovenous recognition capability, thereby obtaining the arteriovenous recognition model. The network design of the arteriovenous recognition model is basically the same as that of the blood vessel recognition model, and the network is in a supervised training mode, and the supervision label is a single-channel multi-value image, 0 represents the background (i.e. non-blood vessel region), and 1 represents the artery and 2 represents the vein.

[0095] Further, in the process of fine-tuning the latter half of the network of the blood vessel recognition model, in order to improve the network's ability to recognize arteriovenous, the blood vessel result of the blood vessel recognition model is introduced as prior knowledge for arteriovenous recognition, so that the arteriovenous recognition model can focus on arteriovenous recognition in the blood vessel region.

[0096] In this embodiment of the present application, prior knowledge is introduced from the input, so that the model has two inputs in the process of use. One input is the to-be-recognized image, which is the initial input of the arteriovenous recognition model, and the other input is the blood vessel segmentation image for the to-be-recognized image, which is the intermediate input of the arteriovenous recognition model. The two inputs are fused in the middle stage of arteriovenous recognition. In this way, the arteriovenous recognition model uses the blood vessel segmentation image as a reference in the process of arteriovenous recognition and focuses on the blood vessel region for arteriovenous recognition.

[0097] The introduction of prior knowledge can be added after the second convolution upsampling process of the arteriovenous recognition model, and the output of the second convolution upsampling process can be fused, or it can be added after the third convolution upsampling, which is not limited in the present application.

[0098] It should be noted that the fine-tuning of the blood vessel recognition model and the introduction of prior knowledge can refer to the conventional model fine-tuning and model prior knowledge introduction scheme, which will not be described here.

[0099] After obtaining the arteriovenous recognition model, the to-be-recognized image and the blood vessel segmentation map obtained based on the blood vessel recognition model are input into the arteriovenous recognition model to obtain the arteriovenous segmentation map corresponding to the to-be-recognized image.

[0100] For example, the to-be-identified image is input into the arteriovenous recognition model, and the arteriovenous recognition model performs feature extraction on the to-be-identified image. In the process of feature extraction, the blood vessel segmentation image corresponding to the input to-be-identified image is received, the blood vessel segmentation image is fused with the feature map of the current feature extraction stage, the feature map after feature fusion is subjected to certain convolution upsampling processing, and arteriovenous pixel point prediction is performed to obtain an arteriovenous segmentation image.

[0101] Therefore, in the process of arteriovenous recognition, the blood vessel segmentation image is introduced as prior knowledge, so that the arteriovenous recognition process focuses on the blood vessel region for arteriovenous recognition, and the accuracy of arteriovenous recognition is improved.

[0102] Figure 4 An implementation flowchart of a specific application example of the blood vessel image recognition method provided by the embodiment of the application is shown.

[0103] To further understand the scheme, the following describes a specific application example of the embodiment of the application.

[0104] Reference Figure 4 In this specific application example of the application, the application scenario is three-dimensional reconstruction of the renal blood vessels of a patient, and this specific application example of the application includes the following steps.

[0105] S1, input an original CT image.

[0106] S2, perform image preprocessing on the original CT image to obtain a preprocessed image.

[0107] Specifically, the preprocessing includes window width and window level normalization, kidney region segmentation, image cropping, and data enhancement.

[0108] S3, input the preprocessed image into a one-stage blood vessel segmentation model to perform one-stage blood vessel recognition and obtain a blood vessel recognition result.

[0109] Specifically, the network of the one-stage blood vessel segmentation model basically adopts 3D V-Net, and the middle convolution module is replaced with an attention feature extraction module (Project&Excite) for implementation. It should be noted that the training process of the one-stage blood vessel recognition model can refer to the description of the blood vessel recognition model in the above Figure 2 The description of the blood vessel recognition model is not repeated here.

[0110] Further, after the preprocessed image is input into the one-stage blood vessel recognition model, the one-stage blood vessel recognition model determines the blood vessel pixel points and non-blood vessel pixel points in the preprocessed image to obtain the blood vessel recognition result.

[0111] S4, input the preprocessed image and the blood vessel recognition result into the two-stage arteriovenous segmentation model for two-stage arteriovenous recognition to obtain an arteriovenous recognition result.

[0112] Specifically, the two-stage arteriovenous recognition model is obtained by fine-tuning the one-stage blood vessel segmentation model and introducing a priori knowledge module. In this specific application example of the present application, the priori knowledge module is introduced in the second convolution upsampling stage of the two-stage blood vessel segmentation model.

[0113] Further, the preprocessed image is input into the two-stage arteriovenous segmentation model, and the feature dimension of the output of the second convolution upsampling processing is obtained in the second convolution upsampling processing stage. The blood vessel recognition result is reduced in dimension by convolution and pooling, so that the reduced blood vessel recognition result has the same feature dimension as the output of the third convolution upsampling processing. Then, the reduced blood vessel recognition result and the output of the second convolution upsampling processing are fused in feature to serve as the input of the third convolution upsampling processing. Finally, the arteriovenous recognition result is obtained along with the subsequent network layers of the two-stage arteriovenous recognition model.

[0114] S5, post-processing is performed on the arteriovenous recognition result to obtain a three-dimensional reconstruction result.

[0115] Specifically, the post-processing includes arteriovenous hole filling, maximum connected domain, and surface reconstruction rendering. The specific post-processing process can be referred to the description of operation 103 above, which will not be repeated here.

[0116] Further, after the post-processing of the arteriovenous recognition result, a three-dimensional reconstruction result is obtained, which includes 3D maps of the renal vein and the renal artery.

[0117] It should be noted that this specific application example of the present application only illustrates the three-dimensional reconstruction of the arteriovenous vessels of the kidney. The embodiments of the present application can perform three-dimensional reconstruction of the arteriovenous vessels of any other organ including arteriovenous vessels, for example, refer to Figure 5 , Figure 5 An example diagram of a three-dimensional reconstruction result obtained based on the blood vessel image recognition method provided by the embodiments of the present application is shown, Figure 5 a shows a three-dimensional reconstruction result of pulmonary blood vessels obtained based on the blood vessel image recognition method of the embodiments of the present application, Figure 5 b shows a three-dimensional reconstruction result of liver blood vessels obtained based on the blood vessel image recognition method of the embodiments of the present application.

[0118] Therefore, this specific application example of the embodiments of the present application improves the recognition accuracy of arteriovenous vessels by performing multi-stage recognition on the blood vessel part of the original CT image.

[0119] Figure 6A component structure schematic diagram of a blood vessel image recognition device provided by an embodiment of the present application is shown.

[0120] Reference Figure 6 Based on the blood vessel image recognition method, the present application provides a blood vessel image recognition device, which comprises: an acquisition module 601 configured to acquire a to-be-recognized image of a to-be-recognized object; a first recognition module 602 configured to perform blood vessel recognition on the to-be-recognized image to obtain a blood vessel segmentation map corresponding to the to-be-recognized image; a second recognition module 603 configured to perform artery and vein recognition on the to-be-recognized image according to the blood vessel segmentation map to obtain an artery and vein segmentation map corresponding to the to-be-recognized image; and a post-processing module 604 configured to perform post-processing on the artery and vein segmentation map according to the blood vessel segmentation map to obtain a blood vessel three-dimensional reconstruction result of the to-be-recognized object.

[0121] In an embodiment of the present application, the acquisition module 601 comprises: an acquisition sub-module configured to acquire an original image of a to-be-recognized object; and a preprocessing sub-module configured to perform preprocessing on the original image to obtain the to-be-recognized image, wherein the preprocessing comprises normalization processing, ROI region segmentation processing, image cropping processing, and data enhancement processing.

[0122] In an embodiment of the present application, the first recognition module 602 comprises: a sampling sub-module configured to perform down-sampling processing and up-sampling processing on the to-be-recognized image to obtain a first feature map, wherein the down-sampling processing comprises convolution processing and attention feature extraction on the to-be-recognized image; a first determination sub-module configured to determine blood vessel pixel points and non-blood vessel pixel points in the first feature map; and a second determination sub-module configured to determine a blood vessel segmentation map corresponding to the to-be-recognized image according to the blood vessel pixel points and the non-blood vessel pixel points.

[0123] In an embodiment of the present application, the second recognition module 603 comprises: an extraction sub-module configured to perform feature extraction on the to-be-recognized image to obtain an initial feature map; a fusion sub-module configured to perform feature fusion on the blood vessel segmentation map and the initial feature map to obtain a target feature map; and a prediction sub-module configured to perform artery and vein pixel point prediction on the target feature map to obtain an artery and vein segmentation map corresponding to the to-be-recognized image.

[0124] In an embodiment of the present application, the extraction sub-module comprises: a first processing unit configured to perform convolution down-sampling processing on the to-be-recognized image for a set number of times to obtain a down-sampling result, wherein the down-sampling result comprises a second feature map output by each convolution down-sampling processing; a second processing unit configured to perform convolution up-sampling processing on the second feature map output by the last convolution down-sampling according to the second feature map output by each convolution down-sampling; and a first acquisition unit configured to acquire an output of a current convolution up-sampling processing to obtain the initial feature map when a number of times of the convolution up-sampling processing reaches a first set number of times.

[0125] In an embodiment of the present application, the fusion submodule comprises: a second acquisition unit, configured to acquire a feature dimension of the initial feature map; a third processing unit, configured to perform convolution processing and pooling processing on the blood vessel segmentation map according to the feature dimension, to obtain a third feature map; and a fusion unit, configured to perform feature fusion on the initial feature map and the third feature map, to obtain a target feature map.

[0126] In an embodiment of the present application, the prediction submodule comprises: a first determination unit, configured to determine a second set number of times according to the set number of times and a first set number of times; a fourth processing unit, configured to perform convolution up-sampling processing on the target feature map for the second set number of times according to the second feature map output by each convolution down-sampling processing and the target feature map, to obtain a final feature map; a second determination unit, configured to determine target arterial pixel points, target venous pixel points and target non-blood vessel pixel points in the final feature map; and a third determination unit, configured to determine the arterial-venous segmentation map according to the target arterial pixel points, the target venous pixel points and the target non-blood vessel pixel points; wherein the sum of the first set number of times and the second set number of times has a determined correlation with the set number of times.

[0127] It should be noted that the description of the device of the embodiments of the present application is similar to the description of the method embodiments described above, and has similar beneficial effects as the method embodiments, and thus will not be described again. For technical details not described in the blood vessel image recognition device provided by the embodiments of the present application, they can be understood according to the description of any one of the accompanying drawings. Figures 1 to 3

[0128] According to the embodiments of the present application, the present application further provides an electronic device and a non-transitory computer readable storage medium.

[0129] Figure 7 A schematic block diagram of an example electronic device 70 that can be used to implement embodiments of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application described and / or claimed in this document.

[0130] As Figure 7 ​As shown, the electronic device 70 includes a computing unit 701 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 702 or a computer program loaded into a random access memory (RAM) 703 from a storage unit 708. In the RAM 703, various programs and data required for the operation of the electronic device 70 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 707 is also connected to the bus 704.

[0131] A plurality of components in the electronic device 70 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, and the like; an output unit 707, such as various types of displays, a speaker, and the like; a storage unit 708, such as a magnetic disk, an optical disk, and the like; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 709 allows the electronic device 70 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0132] The computing unit 701 can be various general-purpose and / or special-purpose processing components having processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, and the like. The computing unit 701 performs various methods and processes described above, such as the blood vessel image recognition method. For example, in some embodiments, the blood vessel image recognition method can be implemented as a computer software program that is tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 70 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the blood vessel image recognition method described above can be performed. Alternatively, in other embodiments, the computing unit 701 can be configured to perform the blood vessel image recognition method by any other appropriate means, such as by means of firmware.

[0133] The various embodiments of the systems and techniques described above can be implemented in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a load programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0134] Program code for carrying out methods of the present application can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program code, when executed by the processor or controller, causes the functions / acts specified in the flowcharts and / or block diagrams to be implemented. The program code can execute entirely on a machine, partly on a machine, partly on a remote machine or entirely on a remote machine or server.

[0135] In the context of the present application, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0136] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0137] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0138] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server can arise by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0139] It should be understood that various forms of flow shown above can be used, re-ordered, added to, or deleted from without departing from the spirit of the present disclosure. For example, the steps recited in the present disclosure can be executed in parallel, in series, or in a different order, as long as the desired results of the technology disclosed in the present disclosure are achieved, and the present disclosure is not limited herein.

[0140] The above description is only specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present disclosure, which should be covered in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A blood vessel image recognition method characterized by comprising: The method comprises: acquiring a to-be-recognized image of a to-be-recognized object; performing blood vessel recognition on the to-be-recognized image to obtain a blood vessel segmentation map corresponding to the to-be-recognized image; performing arteriovenous recognition on the to-be-recognized image according to the blood vessel segmentation map to obtain an arteriovenous segmentation map corresponding to the to-be-recognized image; performing post-processing on the arteriovenous segmentation map according to the blood vessel segmentation map to obtain a blood vessel three-dimensional reconstruction result of the to-be-recognized object; wherein the blood vessel recognition on the to-be-recognized image to obtain the blood vessel segmentation map corresponding to the to-be-recognized image comprises: inputting the to-be-recognized image into a one-stage blood vessel recognition model; and performing processing on the to-be-recognized image by using the one-stage blood vessel recognition model to determine blood vessel pixel points and non-blood vessel pixel points in the to-be-recognized image, thereby obtaining the blood vessel segmentation map; the arteriovenous recognition on the to-be-recognized image according to the blood vessel segmentation map to obtain the arteriovenous segmentation map corresponding to the to-be-recognized image comprises: fine-tuning the one-stage blood vessel segmentation model and introducing a priori knowledge module at a second convolution upsampling stage of the one-stage blood vessel segmentation model to obtain a two-stage arteriovenous segmentation model; inputting the to-be-recognized image into the two-stage arteriovenous segmentation model, acquiring a feature dimension output by a second convolution upsampling processing stage, performing convolution and pooling dimension reduction processing on the blood vessel segmentation map according to the feature dimension; performing feature fusion on the blood vessel segmentation map after the dimension reduction and the output of the second convolution upsampling processing; taking the fusion result as input of a third convolution upsampling processing, and performing subsequent network layer processing on the two-stage arteriovenous recognition model to obtain the arteriovenous segmentation map.

2. The method of claim 1, wherein, The acquiring of the to-be-recognized image of the to-be-recognized object comprises: acquiring an original image of the to-be-recognized object; performing preprocessing on the original image to obtain the to-be-recognized image, wherein the preprocessing comprises normalization processing, ROI region segmentation processing, image cropping processing and data enhancement processing.

3. The method of claim 1, wherein, The blood vessel recognition on the to-be-recognized image to obtain the blood vessel segmentation map corresponding to the to-be-recognized image comprises: performing downsampling processing and upsampling processing on the to-be-recognized image to obtain a first feature map, wherein the downsampling processing comprises convolution processing and attention feature extraction on the to-be-recognized image; determining blood vessel pixel points and non-blood vessel pixel points in the first feature map; determining the blood vessel segmentation map corresponding to the to-be-recognized image according to the blood vessel pixel points and the non-blood vessel pixel points.

4. The method of claim 1, wherein, The arteriovenous recognition on the to-be-recognized image according to the blood vessel segmentation map to obtain the arteriovenous segmentation map corresponding to the to-be-recognized image comprises: performing feature extraction on the to-be-recognized image to obtain an initial feature map; performing feature fusion on the blood vessel segmentation map and the initial feature map to obtain a target feature map; performing arteriovenous pixel point prediction on the target feature map to obtain the arteriovenous segmentation map corresponding to the to-be-recognized image.

5. The method of claim 4, wherein, The feature extraction on the to-be-recognized image to obtain an initial feature extraction map comprises: performing convolution downsampling processing on the to-be-recognized image for a set number of times to obtain a downsampling result, wherein the downsampling result comprises a second feature map output by each convolution downsampling processing; performing convolution up-sampling processing on the second feature map output by the last convolution down-sampling; in a case where the number of times of the convolution up-sampling processing reaches a first preset number of times, obtaining an output of the current convolution up-sampling processing to obtain an initial feature map.

6. The method of claim 4, wherein, performing feature fusion on the blood vessel segmentation map and the initial feature map to obtain a target feature map, including: obtaining a feature dimension of the initial feature map; performing convolution processing and pooling processing on the blood vessel segmentation map according to the feature dimension to obtain a third feature map; performing feature fusion on the initial feature map and the third feature map to obtain a target feature map.

7. The method of claim 5, wherein, performing arteriovenous pixel point prediction on the target feature map to obtain an arteriovenous segmentation map corresponding to the to-be-recognized image, including: determining a second preset number of times according to the preset number of times and the first preset number of times; performing second preset number of times of convolution up-sampling processing on the target feature map according to the second feature map output by each convolution down-sampling processing and the target feature map to obtain a final feature map; determining a target arterial pixel point, a target venous pixel point, and a target non-blood vessel pixel point in the final feature map; determining the arteriovenous segmentation map according to the target arterial pixel point, the target venous pixel point, and the target non-blood vessel pixel point; wherein the sum of the first preset number of times and the second preset number of times has a determined correlation with the preset number of times.

8. The method of claim 1, wherein, The post-processing includes arteriovenous hole filling, maximum connected domain processing, and surface rendering reconstruction.

9. A blood vessel recognition apparatus characterized by comprising: The device includes: an obtaining module configured to obtain a to-be-recognized image of a to-be-recognized object; an identifying module configured to perform blood vessel identification on the to-be-recognized image to obtain a blood vessel segmentation map corresponding to the to-be-recognized image; a determining module configured to perform arteriovenous identification on the to-be-recognized image according to the blood vessel segmentation map to determine an arteriovenous segmentation map corresponding to the to-be-recognized image; a post-processing module configured to perform post-processing on the arteriovenous segmentation map according to the blood vessel segmentation map to obtain a blood vessel three-dimensional reconstruction result of the to-be-recognized object; wherein the identifying module performs blood vessel identification on the to-be-recognized image to obtain a blood vessel segmentation map corresponding to the to-be-recognized image, including: inputting the to-be-recognized image into a one-stage blood vessel identification model; and performing processing on the to-be-recognized image by using the one-stage blood vessel identification model to determine blood vessel pixel points and non-blood vessel pixel points in the to-be-recognized image, thereby obtaining the blood vessel segmentation map; The determining module performs arteriovenous recognition on the to-be-recognized image according to the blood vessel segmentation map, and obtains an arteriovenous segmentation map corresponding to the to-be-recognized image, including: fine-tuning a one-stage blood vessel segmentation model, and introducing a priori knowledge module at a second convolution upsampling stage of the one-stage blood vessel segmentation model to obtain a two-stage arteriovenous segmentation model; inputting the to-be-recognized image into the two-stage arteriovenous segmentation model, obtaining a feature dimension output by a second convolution upsampling processing stage, and performing convolution and pooling dimension reduction processing on the blood vessel segmentation map according to the feature dimension; performing feature fusion on the blood vessel segmentation map after the dimension reduction and the output of the second convolution upsampling processing; taking the fusion result as input of third convolution upsampling processing, and performing subsequent network layer processing on the two-stage arteriovenous recognition model to obtain the arteriovenous segmentation map.

10. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are configured to cause a computer to perform the method of any one of claims 1-8.

Images