A blood vessel segmentation method, device, equipment and storage medium

Abstract

The present application provides a blood vessel segmentation method, apparatus, device and storage medium, the blood vessel segmentation method comprising: inputting an original medical image into a first blood vessel segmentation model to obtain a first blood vessel segmentation prediction result of a target blood vessel, and obtaining a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result; cutting the original medical image into a plurality of local image blocks, and cutting out a global region of the target blood vessel from the original medical image according to the position of each local image block in the original medical image and the blood vessel trunk positioning result; inputting the global region into the first blood vessel segmentation model to obtain a global intermediate feature map generated in an intermediate stage; inputting a plurality of local image blocks into a second blood vessel segmentation model, and introducing the global intermediate feature map as reference information for blood vessel segmentation prediction into the second blood vessel segmentation model in a data interactive manner to obtain a second blood vessel segmentation prediction result of the target blood vessel. In this way, the present application can effectively improve the accuracy of the blood vessel segmentation result.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a blood vessel segmentation method, device, equipment and storage medium. Background Art

[0002] In recent years, scholars have used deep learning technology to diagnose and analyze diseases. The application of deep learning in the medical field to improve disease diagnosis and auxiliary treatment has become a trend, and it has great research significance in medical image segmentation.

[0003] However, the accuracy of the current vascular segmentation algorithm still needs to be improved. Relying solely on a single vascular segmentation model can only obtain relatively rough vascular segmentation results. For example, in cases of acute lung infections such as COVID-19, when performing vascular segmentation on target blood vessels such as pulmonary blood vessels and liver blood vessels that contain both arteries and veins, since the appearance of arteries and veins is very similar, a single vascular segmentation model is difficult to effectively distinguish between arteries and veins, resulting in low accuracy of the vascular segmentation results output by the model. Summary of the invention

[0004] In view of this, the purpose of the present application is to provide a blood vessel segmentation method, apparatus, device and storage medium, which, by determining the blood vessel segmentation results in stages, introduces the global information between each local area and the blood vessel trunk into the second-stage model in a manner that conforms to the blood vessel topology to predict the blood vessel segmentation of each local area, thereby enabling the second-stage blood vessel segmentation model to obtain a feature representation with a stronger ability to distinguish arteriovenous categories, effectively improving the accuracy of the blood vessel segmentation results and reducing the segmentation errors of the blood vessel segmentation model due to arteriovenous confusion.

[0005] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings.

[0006] In a first aspect, an embodiment of the present application provides a blood vessel segmentation method, the blood vessel segmentation method comprising:

[0007] Inputting an original medical image containing a target blood vessel into a first blood vessel segmentation model, outputting a first blood vessel segmentation prediction result for the target blood vessel, and obtaining a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result; wherein the first blood vessel segmentation prediction result includes a first artery segmentation prediction result and a first vein segmentation prediction result of the target blood vessel;

[0008] The original medical image is cut into a plurality of local image blocks, and according to the position of each local image block in the original medical image and the vascular trunk positioning result, a global region corresponding to the target blood vessel is cut out from the original medical image; wherein the global region is used to characterize an image region including the vascular trunk of the target blood vessel and each local image block;

[0009] Inputting the global region into the first blood vessel segmentation model, and outputting a global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the global region through the first blood vessel segmentation model;

[0010] The multiple local image blocks are input into a second blood vessel segmentation model. In the process of the second blood vessel segmentation model performing blood vessel segmentation prediction on each local image block, the global intermediate feature map is introduced into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interactive manner, and a second blood vessel segmentation prediction result for the target blood vessel is output; wherein the second blood vessel segmentation prediction result includes a second artery segmentation prediction result and a second vein segmentation prediction result of the target blood vessel.

[0011] In an optional implementation, acquiring the main vessel trunk positioning result of the target vessel from the first vessel segmentation prediction result includes:

[0012] Extracting the arterial blood vessel midline of the target blood vessel from the maximum connected domain of the first artery segmentation prediction result;

[0013] Calculating the shortest distance between the arterial vessel midline and each voxel unit located on the surface of the arterial vessel in the first arterial segmentation prediction result, and taking the region where the arterial vessel is located where the calculation result is greater than or equal to the first preset distance threshold as the arterial vessel trunk positioning result in the vascular trunk positioning result;

[0014] Extracting the venous blood vessel centerline of the target blood vessel from the maximum connected domain of the first vein segmentation prediction result;

[0015] Calculate the shortest distance between the centerline of the vein and each voxel unit located on the surface of the vein in the first vein segmentation prediction result, and use the area where the vein is located where the calculation result is greater than or equal to the second preset distance threshold as the vein trunk positioning result in the blood vessel trunk positioning result.

[0016] In an optional implementation, outputting the global intermediate feature map generated in the intermediate stage of performing blood vessel segmentation prediction for the global region through the first blood vessel segmentation model includes:

[0017] In the intermediate stage of the first blood vessel segmentation model performing blood vessel segmentation prediction on the input global region, the first blood vessel segmentation model is used to output one or more feature representation results after intermediate processing on the global region; wherein each of the feature representation results is used to represent a global intermediate feature map.

[0018] In an optional implementation, in the process of the second blood vessel segmentation model performing blood vessel segmentation prediction for each local image block, introducing the global intermediate feature map as reference information for blood vessel segmentation prediction into the second blood vessel segmentation model in a data interactive manner, including:

[0019] For each of the input local image blocks, in an intermediate stage when the second blood vessel segmentation model performs blood vessel segmentation prediction on the local image block, obtaining a local intermediate feature map after the second blood vessel segmentation model performs intermediate processing on the local image block;

[0020] For each first voxel unit included in the local intermediate feature map, calculate the similarity between the first voxel unit and each second voxel unit included in the global intermediate feature map, and obtain a similarity calculation result corresponding to the first voxel unit;

[0021] According to the similarity calculation result and the target voxel unit included in the global intermediate feature map, an optimal path between a target second voxel unit and the target voxel unit is determined from the global intermediate feature map; wherein the target second voxel unit is used to represent a second voxel unit corresponding to the same physical position as the first voxel unit on the target blood vessel; and the target voxel unit is used to represent a voxel unit corresponding to the blood vessel trunk of the target blood vessel on the global intermediate feature map;

[0022] According to the similarity between each third voxel unit and the target second voxel unit, and the feature distance between each third voxel unit and the target second voxel unit, a fusion feature corresponding to the target second voxel unit is calculated by weighted summation; wherein the third voxel unit is used to characterize the voxel unit covered by the optimal path on the global intermediate feature map;

[0023] The fusion feature is spliced ​​with the local feature of the first voxel unit in the local intermediate feature map, and based on the feature splicing result corresponding to each of the first voxel units, the feature representation result corresponding to the local image block in the intermediate stage of blood vessel segmentation prediction is output.

[0024] In an optional implementation, determining an optimal path between a target second voxel unit and the target voxel unit from the global intermediate feature map according to the similarity calculation result and the target voxel unit included in the global intermediate feature map includes:

[0025] Determine, from the global intermediate feature map, a plurality of candidate paths between the target second voxel unit and the target voxel unit according to position information of the target voxel unit in the global intermediate feature map;

[0026] According to the similarity calculation result, a candidate path with the shortest distance and the largest sum of similarities between each second voxel unit covered on the path and the first voxel unit is selected from the multiple candidate paths as the optimal path.

[0027] In an optional implementation, the first blood vessel segmentation model is trained by the following method:

[0028] Acquire a plurality of medical images containing target blood vessels as first training sample data; wherein each of the medical images contains a blood vessel labeling result of the target blood vessel; the blood vessel labeling result includes: an arterial blood vessel labeling result and a venous blood vessel labeling result of the target blood vessel;

[0029] Inputting the first training sample data into a first blood vessel segmentation model, and outputting a third blood vessel segmentation prediction result for each of the medical images; wherein the third blood vessel segmentation prediction result includes an arterial blood vessel prediction result and a venous blood vessel prediction result of the target blood vessel;

[0030] Based on the segmentation loss between the third blood vessel segmentation prediction result of each of the medical images and the blood vessel labeling result corresponding to each of the medical images, the first blood vessel segmentation model is trained until the first blood vessel segmentation model reaches convergence.

[0031] In an optional implementation, the second blood vessel segmentation model is trained by the following method:

[0032] For each of the medical images, the medical image is cut into a plurality of sample image blocks, and according to the position of each of the sample image blocks in the medical image and the positioning result of the vascular trunk of the target blood vessel in the medical image, an initial global region corresponding to the target blood vessel is cut out from the medical image; wherein the initial global region is used to characterize an image region including the vascular trunk of the target blood vessel and each of the sample image blocks;

[0033] Inputting the initial global region into a trained first blood vessel segmentation model, and outputting an initial global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the initial global region through the trained first blood vessel segmentation model;

[0034] The plurality of sample image blocks are input into a second blood vessel segmentation model. During the process of the second blood vessel segmentation model performing blood vessel segmentation prediction for each sample image block, the initial global intermediate feature map is introduced into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interactive manner, and a fourth blood vessel segmentation prediction result for the medical image is output; wherein the fourth blood vessel segmentation prediction result includes an arterial blood vessel prediction result and a venous blood vessel prediction result of the target blood vessel;

[0035] Based on the segmentation loss between the fourth blood vessel segmentation prediction result of each of the medical images and the blood vessel labeling result corresponding to each of the medical images, the second blood vessel segmentation model is trained until the second blood vessel segmentation model reaches convergence.

[0036] In a second aspect, an embodiment of the present application provides a blood vessel segmentation device, the blood vessel segmentation device comprising:

[0037] a coarse segmentation module, configured to input an original medical image containing a target blood vessel into a first blood vessel segmentation model, output a first blood vessel segmentation prediction result for the target blood vessel, and obtain a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result; wherein the first blood vessel segmentation prediction result includes a first artery segmentation prediction result and a first vein segmentation prediction result of the target blood vessel;

[0038] A cropping module, used for cropping the original medical image into a plurality of local image blocks, and cropping a global region corresponding to the target blood vessel from the original medical image according to the position of each local image block in the original medical image and the vascular trunk positioning result; wherein the global region is used to represent an image region including the vascular trunk of the target blood vessel and each local image block;

[0039] an intermediate output module, configured to input the global region into the first blood vessel segmentation model, and output a global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the global region through the first blood vessel segmentation model;

[0040] a data interaction module, configured to input the plurality of local image blocks into a second blood vessel segmentation model; and in a process in which the second blood vessel segmentation model performs blood vessel segmentation prediction on each local image block, introduce the global intermediate feature map into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interaction manner, and output a second blood vessel segmentation prediction result for the target blood vessel; wherein the second blood vessel segmentation prediction result includes a second artery segmentation prediction result and a second vein segmentation prediction result of the target blood vessel.

[0041] In a third aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned blood vessel segmentation method when executing the computer program.

[0042] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned blood vessel segmentation method are executed.

[0043] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0044] The embodiments of the present application provide a blood vessel segmentation method, apparatus, device and storage medium, which input an original medical image containing a target blood vessel into a first blood vessel segmentation model, output a first blood vessel segmentation prediction result for the target blood vessel, and obtain a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result; cut the original medical image into multiple local image blocks, and cut out a global region corresponding to the target blood vessel from the original medical image according to the position of each local image block in the original medical image and the blood vessel trunk positioning result; input the global region into the first blood vessel segmentation model, and output a global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the global region through the first blood vessel segmentation model; input multiple local image blocks into a second blood vessel segmentation model, and in the process of the second blood vessel segmentation model performing blood vessel segmentation prediction on each local image block, introduce the global intermediate feature map into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interactive manner, and output a second blood vessel segmentation prediction result for the target blood vessel.

[0045] By determining the vascular segmentation results in this staged manner, the present application can introduce the global information between each local area and the vascular trunk into the second-stage model's vascular segmentation prediction process for each local area in a manner that conforms to the vascular topology, thereby enabling the second-stage vascular segmentation model to obtain a feature representation with a stronger ability to distinguish between arteriovenous categories, effectively improving the accuracy of the vascular segmentation results and reducing segmentation errors caused by arteriovenous confusion in the vascular segmentation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A schematic diagram showing a flow chart of a blood vessel segmentation method provided in an embodiment of the present application;

[0048] Figure 2 A schematic diagram showing a flow chart of a method for obtaining a vascular trunk positioning result provided in an embodiment of the present application;

[0049] Figure 3 A schematic diagram of the model structure of a first blood vessel segmentation model provided in an embodiment of the present application is shown;

[0050] Figure 4a A schematic diagram of a process for introducing a global intermediate feature map provided in an embodiment of the present application is shown;

[0051] Figure 4b A schematic diagram of data interaction between a first blood vessel segmentation model and a second blood vessel segmentation model provided in an embodiment of the present application is shown;

[0052] Figure 4c An interactive schematic diagram of determining an optimal path provided by an embodiment of the present application is shown;

[0053] Figure 5 A schematic flow chart of a method for training a first blood vessel segmentation model provided in an embodiment of the present application is shown;

[0054] Figure 6 A schematic flow chart of a method for training a second blood vessel segmentation model provided in an embodiment of the present application is shown;

[0055] Figure 7 A schematic diagram of the structure of a blood vessel segmentation device provided in an embodiment of the present application is shown;

[0056] Figure 8 It is shown that an embodiment of the present application provides a computer device 800. DETAILED DESCRIPTION

[0057] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.

[0058] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0059] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0060] At present, the accuracy of the vascular segmentation algorithm needs to be improved. Relying solely on a single vascular segmentation model can only obtain relatively rough vascular segmentation results. At this time, when performing vascular segmentation on target blood vessels such as pulmonary blood vessels and liver blood vessels that contain both arteries and veins, since the appearance of arteries and veins is very similar, a single vascular segmentation model is difficult to effectively distinguish between arteries and veins, resulting in low accuracy of the vascular segmentation results output by the model.

[0061] Based on this, the embodiments of the present application provide a blood vessel segmentation method, apparatus, device and storage medium, which input an original medical image containing a target blood vessel into a first blood vessel segmentation model, output a first blood vessel segmentation prediction result for the target blood vessel, and obtain a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result; cut the original medical image into multiple local image blocks, and according to the position of each local image block in the original medical image and the blood vessel trunk positioning result, cut out a global region corresponding to the target blood vessel from the original medical image; input the global region into the first blood vessel segmentation model, and output a global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the global region through the first blood vessel segmentation model; input multiple local image blocks into a second blood vessel segmentation model, and in the process of the second blood vessel segmentation model performing blood vessel segmentation prediction on each local image block, introduce the global intermediate feature map into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interactive manner, and output a second blood vessel segmentation prediction result for the target blood vessel.

[0062] By determining the vascular segmentation results in this staged manner, the present application can introduce the global information between each local area and the vascular trunk into the second-stage model's vascular segmentation prediction process for each local area in a manner that conforms to the vascular topology, thereby enabling the second-stage vascular segmentation model to obtain a feature representation with a stronger ability to distinguish between arteriovenous categories, effectively improving the accuracy of the vascular segmentation results and reducing segmentation errors caused by arteriovenous confusion in the vascular segmentation model.

[0063] It should be noted that the blood vessel segmentation method provided in the embodiment of the present application is applicable to a blood vessel segmentation device, and the blood vessel segmentation device can be integrated into a computer device.

[0064] Specifically, the above-mentioned computer device can be a terminal device, such as a mobile phone, a tablet computer, a laptop computer, a desktop computer, etc.; the above-mentioned computer device can also be a server, which can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms, but is not limited to these.

[0065] To facilitate understanding of the embodiments of the present application, a blood vessel segmentation method, apparatus, device and storage medium provided in the embodiments of the present application are described in detail below.

[0066] Reference Figure 1 As shown, Figure 1 A schematic flow chart of a blood vessel segmentation method provided in an embodiment of the present application is shown, and the blood vessel segmentation method includes steps S101-S104; specifically:

[0067] S101, inputting an original medical image containing a target blood vessel into a first blood vessel segmentation model, outputting a first blood vessel segmentation prediction result for the target blood vessel, and obtaining a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result.

[0068] Here, the original medical image is used to represent a three-dimensional image containing a target blood vessel. For example, it may be a CT (Computed Tomography) image containing a target blood vessel, or a CTA image (i.e., CT vascular imaging) containing a target blood vessel, etc.; the embodiment of the present application does not impose any limitation on the specific data type to which the above-mentioned original medical image belongs.

[0069] In an embodiment of the present application, the blood vessel segmentation method provided in the embodiment of the present application can effectively reduce the segmentation errors of the blood vessel segmentation model for arteriovenous confusion by determining the pulmonary blood vessel segmentation results in stages; therefore, in a preferred embodiment, the target blood vessel can be used to characterize a blood vessel that includes both an arterial blood vessel and a venous blood vessel (e.g., the target blood vessel can be a pulmonary blood vessel including a pulmonary artery and a pulmonary vein, or a liver blood vessel including a hepatic artery and a hepatic vein).

[0070] At this time, in the above-mentioned preferred embodiment, the first blood vessel segmentation prediction result output by the first blood vessel segmentation model includes both the first artery segmentation prediction result of the target blood vessel (i.e., the image region where the artery is located in the original medical image) and the first vein segmentation prediction result of the target blood vessel (i.e., the image region where the vein is located in the original medical image); for example, if the target blood vessel is a pulmonary blood vessel, the original medical image may be a three-dimensional image of the lungs, and the three-dimensional image of the lungs may be input into the first blood vessel segmentation model, so that the pulmonary artery segmentation prediction result and the pulmonary vein prediction result may be output.

[0071] It should be noted that, in other optional implementations, the target blood vessel may also be an ordinary blood vessel without artery / vein distinction. In this case, the first blood vessel segmentation prediction result output by the first blood vessel segmentation model is: the image region where the target blood vessel is located in the original medical image; whether the target blood vessel is divided into artery / vein does not affect the blood vessel segmentation problem solved by the embodiment of the present application for the target blood vessel. Based on this, the embodiment of the present application does not impose any limitation on the specific blood vessel type to which the above-mentioned target blood vessel belongs.

[0072] Specifically, when the target blood vessel is divided into artery / vein, the main blood vessel trunk positioning result obtained from the first blood vessel segmentation result includes: the positioning result of the arterial blood vessel trunk in the target blood vessel (that is, the image area belonging to the arterial blood vessel trunk in the first blood vessel segmentation result) and the positioning result of the venous blood vessel trunk (that is, the image area belonging to the venous blood vessel trunk in the first blood vessel segmentation result); for example, if the target blood vessel is a pulmonary blood vessel, the main blood vessel trunk positioning result obtained from the first blood vessel segmentation result includes: the pulmonary artery trunk positioning result and the pulmonary vein trunk positioning result.

[0073] Specifically, when the target blood vessel does not have artery / vein distinction, the blood vessel trunk positioning result obtained from the first blood vessel segmentation result includes: the blood vessel trunk positioning result of the target blood vessel (ie, the image area belonging to the target blood vessel trunk in the first blood vessel segmentation result).

[0074] S102, cutting the original medical image into a plurality of local image blocks, and cutting out a global region corresponding to the target blood vessel from the original medical image according to the position of each local image block in the original medical image and the blood vessel trunk positioning result.

[0075] Here, in the above-mentioned step S101, the embodiment of the present application only obtains a rough segmentation prediction result of the target blood vessel through the first blood vessel segmentation model in the first stage of blood vessel segmentation, and only in the second stage of blood vessel segmentation (i.e., step S104) does it need to use the second blood vessel segmentation model to obtain a more refined blood vessel segmentation result.

[0076] Specifically, in order to obtain more precise blood vessel segmentation results, unlike the first blood vessel segmentation model in the first stage of blood vessel segmentation, which directly uses the global image of the target blood vessel (i.e., the original medical image containing the target blood vessel) as model input data, in the second stage of blood vessel segmentation, the second blood vessel segmentation model uses a local image of the target blood vessel with higher resolution (i.e., the multiple local image blocks after cropping mentioned above) as model input data.

[0077] Based on this, precisely because the second blood vessel segmentation model uses the above-mentioned multiple cropped local image blocks as model input data, therefore, before the second blood vessel segmentation model performs blood vessel segmentation prediction, it is necessary to implement step S102 to crop the global region of the target blood vessel corresponding to the above-mentioned multiple local image blocks from the original medical image, so that through the implementation of subsequent steps S103-S104, when the second blood vessel segmentation model performs segmentation prediction for each input local image block, the relevant feature representation of the above-mentioned global region is introduced, so that the second blood vessel segmentation model can obtain a feature representation with a stronger ability to distinguish arteriovenous categories.

[0078] Here, the global region is used to characterize an image region that includes both the blood vessel trunk of the target blood vessel and each of the local image blocks, and the position information of each local image block in the original medical image is simultaneously recorded in the global region.

[0079] Specifically, the above-mentioned position information can be understood as the specific three-dimensional position coordinates corresponding to each local image block in the original medical image; it can also be understood as the physical position information actually corresponding to the target blood vessel area contained in each local image block on the target blood vessel (since the original medical image contains a complete target blood vessel, the actual physical position information can also play the role of locating each local image block). Based on this, the embodiment of the present application does not impose any limitation on the specific information form of the above-mentioned position information.

[0080] It should be noted that when cropping the original medical image, the original medical image can be cropped into multiple local image blocks of the same size; or the original medical image can be cropped into multiple local image blocks of indefinite sizes by random cropping; the embodiment of the present application does not impose any limitation on whether the size of each local image block obtained after cropping is the same.

[0081] S103, inputting the global region into the first blood vessel segmentation model, and outputting a global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the global region through the first blood vessel segmentation model.

[0082] It should be noted that the first vascular segmentation model can be a vascular segmentation model constructed based on a Unet segmentation network, or a vascular segmentation model constructed based on other types of segmentation networks such as a ResUNet network; the embodiment of the present application does not impose any limitation on the specific model structure of the first vascular segmentation model.

[0083] Here, in the segmentation prediction process between the model input and the model output, the first vascular segmentation model will perform multiple intermediate processing on the input global area. For example, when the first vascular segmentation model is a vascular segmentation model constructed based on the Unet segmentation network, the first vascular segmentation model will perform multiple downsampling processing and multiple upsampling processing on the input global area to obtain the final vascular segmentation prediction result for the global area.

[0084] Based on this, when executing S103, as an optional embodiment, in the intermediate stage of performing vascular segmentation prediction on the input global region by the first vascular segmentation model, one or more feature representation results after intermediate processing on the global region can be output through the first vascular segmentation model; wherein each feature representation result can be used to represent a global intermediate feature map. The embodiment of the present application does not limit whether to specifically introduce one global intermediate feature map or multiple global intermediate feature maps in the subsequent step S104.

[0085] S104, inputting the multiple local image blocks into a second blood vessel segmentation model, and in the process of the second blood vessel segmentation model performing blood vessel segmentation prediction on each local image block, introducing the global intermediate feature map as reference information for blood vessel segmentation prediction into the second blood vessel segmentation model in a data interactive manner, and outputting a second blood vessel segmentation prediction result for the target blood vessel.

[0086] Here, since the second blood vessel segmentation model is still a blood vessel segmentation prediction for the target blood vessel (it is only based on the introduction of the global intermediate feature map and the use of multiple local image blocks as the processing method of the model input data, it can obtain a more accurate blood vessel segmentation prediction result than the first blood vessel segmentation prediction result), therefore, when the target blood vessel is divided into arteries and veins, the second blood vessel segmentation prediction result includes: a second artery segmentation prediction result and a second vein segmentation prediction result of the target blood vessel.

[0087] It should be noted that, similar to the first blood vessel segmentation prediction result, if there is no distinction between arteries and veins in the target blood vessel, the second blood vessel segmentation prediction result output by the second blood vessel segmentation model is: the image region where the target blood vessel is located. Based on this, the embodiment of the present application does not impose any limitation on the specific result content represented by the above-mentioned second blood vessel segmentation prediction result.

[0088] In the embodiment of the present application, the second blood vessel segmentation model can also be a blood vessel segmentation model constructed based on other types of segmentation networks such as Unet segmentation network / ResUNet network; the embodiment of the present application also does not impose any limitation on the specific model structure of the second blood vessel segmentation model.

[0089] It should be noted that, compared with the first vascular segmentation model, the model structure of the second vascular segmentation model can be the same or similar to the first vascular segmentation model. At this time, since the first vascular segmentation model is used to perform coarse segmentation prediction on the target vessel, the second vascular segmentation model is used to perform fine segmentation prediction on the target vessel. Therefore, although the structures of the two models can be the same or similar, the model depth of the second vascular segmentation model and the specific model parameters included will be higher than the first vascular segmentation model; that is, in the embodiment of the present application, the first vascular segmentation model and the second vascular segmentation model belong to different vascular segmentation models.

[0090] Specifically, based on the aforementioned step S103, it can be known that in step S103, one global intermediate feature map or multiple global feature maps can be obtained; wherein, when one global intermediate feature map is obtained in step S103, the embodiment of the present application does not impose any limitation on the intermediate stage in the second blood vessel segmentation model at which the global intermediate feature map is introduced; and when multiple global intermediate feature maps are obtained in step S103, based on the characteristic that the model structure of the second blood vessel segmentation model can be similar to that of the first blood vessel segmentation model, as a preferred implementation, each global intermediate feature map can also be introduced at the same intermediate stage in the second blood vessel segmentation model according to its source in the first blood vessel segmentation model, so that the second blood vessel segmentation model can obtain a feature representation with a stronger ability to distinguish arteriovenous categories, thereby further improving the accuracy of the second blood vessel segmentation result output by the model.

[0091] The specific implementation process of each of the above steps in the embodiment of the present application is described in detail below:

[0092] Regarding the specific implementation process of the above step S101, in addition to directly using the original medical image as the model input data of the first blood vessel segmentation model, considering that the first blood vessel segmentation prediction result obtained in the first stage does not need to be very precise, therefore, in another optional implementation, the original medical image can also be down-sampled so that its overall size can be controlled within a certain range, and the down-sampled image data is input into the first blood vessel segmentation model, which can also achieve the effect of learning the global information of the target blood vessel.

[0093] For example, taking the original medical image as a lung medical image with an image size of 300×300×400, by downsampling the original medical image, a lung medical image with an image size of 128×128×128 can be obtained, and the small-sized lung medical image obtained after the downsampling process is input into the first blood vessel segmentation model, and the first blood vessel segmentation prediction result can also be obtained.

[0094] It should be noted that the specific image size shown in the above example is only used for specific exemplary description. The embodiment of the present application does not impose any limitation on the image size of the original medical image used in the actual operation process and the specific image size of the medical image obtained by downsampling processing.

[0095] Regarding the specific implementation process of the above step S101, it should be noted that the vascular trunk positioning result does not need to be obtained through the model. Based on this, when there is a distinction between arteries and veins in the target blood vessel, refer to Figure 2 As shown, Figure 2 A flow chart of a method for obtaining a vascular trunk positioning result provided in an embodiment of the present application is shown, and the method includes steps S201-S204; specifically:

[0096] S201, extracting the arterial blood vessel midline of the target blood vessel from the maximum connected domain of the first artery segmentation prediction result.

[0097] Specifically, the first artery segmentation prediction result output by the first blood vessel segmentation model may include an image connected domain belonging to the artery (which may be multiple image connected domains of different sizes), a background image that does not belong to the artery, and discrete image units that belong to the artery but cannot constitute a connected domain.

[0098] At this time, based on the fact that the arterial trunk is the thicker part of the artery and the arterial trunk is connected to each vascular branch of the artery, the arterial midline of the target vessel can be directly extracted from the maximum connected domain (equivalent to the area where the arterial trunk is concentrated) of the first arterial segmentation prediction result.

[0099] S202, calculating the shortest distance between the centerline of the artery and each voxel unit located on the surface of the artery in the first artery segmentation prediction result, and taking the area where the artery is located where the calculation result is greater than or equal to the first preset distance threshold as the arterial trunk positioning result in the vascular trunk positioning result.

[0100] Here, based on the fact that the original medical image is a three-dimensional image, the above-mentioned voxel unit is equivalent to the basic image unit in the three-dimensional image, and its concept can be analogous to the pixel point in the two-dimensional image; the embodiment of the present application does not impose any limitation on the specific volume value represented by the above-mentioned voxel unit.

[0101] Specifically, since the blood vessel trunk is a thicker blood vessel part in the artery, based on the first preset distance threshold, the area where the arterial blood vessel where the above calculation result is greater than or equal to the first preset distance threshold is located can be used as the arterial blood vessel trunk positioning result.

[0102] It should be noted that the specific value of the first preset distance threshold can be determined according to the type of blood vessel to which the target blood vessel for actual blood vessel segmentation belongs, and this embodiment of the present application does not impose any limitation on this.

[0103] S203: Extracting the venous blood vessel centerline of the target blood vessel from the maximum connected domain of the first vein segmentation prediction result.

[0104] Here, the specific implementation of step S203 is the same as that of the above step S201, and the repeated parts are not repeated here.

[0105] S204, calculating the shortest distance between the centerline of the venous vessel and each voxel unit located on the surface of the venous vessel in the first venous segmentation prediction result, and taking the area where the venous vessel is located where the calculation result is greater than or equal to the second preset distance threshold as the venous vessel trunk positioning result in the vascular trunk positioning result.

[0106] Here, the specific implementation of step S204 is the same as that of the above step S202, and the repeated parts are not repeated here.

[0107] It should be noted that the specific value of the second preset distance threshold can also be determined according to the type of blood vessel to which the target blood vessel for actual blood vessel segmentation belongs. The embodiment of the present application does not impose any limitation on the specific value of the second preset distance threshold and whether the second preset distance threshold is the same as the first preset distance threshold.

[0108] Regarding the specific implementation process of the above step S103, taking the first blood vessel segmentation model as an example, the blood vessel segmentation model constructed by the Unet segmentation network is shown in FIG. Figure 3 As shown, Figure 3 FIG. 4 shows a schematic diagram of a model structure of a first blood vessel segmentation model provided in an embodiment of the present application. Figure 3 As shown, within the first blood vessel segmentation model, the first blood vessel segmentation model performs multiple downsampling processes 301 and multiple upsampling processes 302 on the input global area. At this time, in the intermediate stage of the first blood vessel segmentation model performing blood vessel segmentation prediction on the input global area, a first feature representation result 310 can be output after each downsampling process 301, and a second feature representation result 320 can be output after each upsampling process 302; wherein each first feature representation result 310 and each second feature representation result 320 can be used to represent a global intermediate feature map.

[0109] For the specific implementation process of the above step S104, refer to Figure 4a As shown, Figure 4aA schematic flow chart of a method for introducing a global intermediate feature map provided in an embodiment of the present application is shown, wherein the method comprises steps S401-S405; specifically:

[0110] S401, for each of the input local image blocks, in an intermediate stage when the second blood vessel segmentation model performs blood vessel segmentation prediction on the local image block, obtain a local intermediate feature map after the second blood vessel segmentation model performs intermediate processing on the local image block.

[0111] Here, the model structure of the second blood vessel segmentation model is similar to that of the first blood vessel segmentation model. Figure 3 In the first vascular segmentation model structure shown, during the segmentation prediction process between the model input and the model output, the second vascular segmentation model will also perform multiple intermediate processing on each input local image block. For example, when the second vascular segmentation model is a vascular segmentation model built based on the Unet segmentation network, the second vascular segmentation model will perform multiple downsampling processing and multiple upsampling processing on each input local image block, so as to obtain a vascular segmentation prediction result for each local image block.

[0112] Specifically, based on the aforementioned steps S103-S104, it can be known that in step S103, one global intermediate feature map can be obtained, or multiple global feature maps can be obtained; wherein, when one global intermediate feature map is obtained in step S103, then when executing step S401, for each local image block, the embodiment of the present application does not impose any limitation on which step of the intermediate processing operation performed by the second blood vessel segmentation model on the local image block to introduce the obtained global feature map; when multiple global intermediate feature maps are obtained in step S103, based on the characteristic that the model structure of the second blood vessel segmentation model can be similar to that of the first blood vessel segmentation model (the specific model structure is as follows Figure 3 As shown in the figure, as a preferred embodiment, each global intermediate feature map can be introduced at the same intermediate stage in the second vessel segmentation model according to its origin in the first vessel segmentation model, so that the second vessel segmentation model can obtain a feature representation with a stronger ability to distinguish arteriovenous categories.

[0113] For an exemplary description, see Figure 4b As shown, Figure 4bA schematic diagram of data interaction between a first blood vessel segmentation model and a second blood vessel segmentation model provided by an embodiment of the present application is shown. In the intermediate stage of the first blood vessel segmentation model performing blood vessel segmentation prediction on the input global region, a first feature representation result 310 can be output after each downsampling process, and a second feature representation result 320 can be output after each upsampling process; wherein each first feature representation result 310 and each second feature representation result 320 can be used to represent a global intermediate feature map; based on this, when data interaction is performed between the first blood vessel segmentation model and the second blood vessel segmentation model, the second blood vessel segmentation model can introduce the global intermediate feature map represented by the first feature representation result 310 from the first blood vessel segmentation model in the intermediate stage of performing downsampling process on each input local image block; the second blood vessel segmentation model can introduce the global intermediate feature map represented by the second feature representation result 320 from the first blood vessel segmentation model in the intermediate stage of performing upsampling process on each input local image block.

[0114] Specifically, the method of obtaining the local intermediate feature map in step S401 is the same as the method of obtaining the global intermediate feature map in the above step S103, that is, the first blood vessel segmentation model inputs the above global area, so the intermediate feature map for the global area (i.e., the global intermediate feature map) can be obtained in the intermediate processing stage of the model; the second blood vessel segmentation model inputs each local image block after cropping, so the intermediate feature map for each local image block (i.e., the local intermediate feature map) can be obtained in the intermediate processing stage of the model; the repeated parts are not repeated here.

[0115] S402, for each first voxel unit included in the local intermediate feature map, calculate the similarity between the first voxel unit and each second voxel unit included in the global intermediate feature map, and obtain the similarity calculation result corresponding to the first voxel unit.

[0116] Here, the definition of the voxel unit is the same as that in the above step S202. The first voxel unit is used to represent the voxel unit contained in the local intermediate feature map, and its concept can be compared to the pixel point in a two-dimensional image. The embodiment of the present application does not impose any limitation on the specific volume value represented by the above first voxel unit.

[0117] Here, the second voxel unit is used to represent the voxel unit included in the global intermediate feature map. The embodiment of the present application also does not impose any limitation on the specific volume value represented by the second voxel unit.

[0118] Specifically, when executing step S402, the similarity between the first voxel unit and the second voxel unit can be calculated by any similarity calculation method, for example, similarity calculation methods such as Euclidean distance and Mahalanobis distance can be used; the embodiment of the present application does not impose any limitation on the specific similarity calculation method used in step S402.

[0119] S403, determining an optimal path between a target second voxel unit and the target voxel unit from the global intermediate feature map according to the similarity calculation result and the target voxel unit included in the global intermediate feature map.

[0120] Here, the target second voxel unit is used to represent a second voxel unit corresponding to the same physical position as the first voxel unit on the target blood vessel; wherein, based on the aforementioned step S102, it can be known that based on the specific three-dimensional position coordinates corresponding to each local image block in the original medical image / the physical position information actually corresponding to the target blood vessel area contained in each local image block on the target blood vessel and other position information, the target second voxel unit corresponding to each first voxel unit in the current local image block in the global intermediate feature map can be determined from the global intermediate feature map.

[0121] Here, the target voxel unit is used to characterize the voxel unit corresponding to the vascular trunk of the target blood vessel on the global intermediate feature map; that is, based on the vascular trunk positioning result of the target blood vessel obtained in the above step S101, the target second voxel unit corresponding to the vascular trunk of the target blood vessel can be determined from the global intermediate feature map.

[0122] It should be noted that when the target blood vessel is divided into arteries and veins, the vascular trunk of the above-mentioned target blood vessel includes the arterial vascular trunk and the venous vascular trunk. At this time, the second voxel units corresponding to the arterial vascular trunk and the venous vascular trunk both belong to the above-mentioned target second voxel unit.

[0123] Specifically, when executing step S403, the optimal path may be determined in the manner shown in the following steps a1-a2:

[0124] Step a1: According to the position information of the target voxel unit in the global intermediate feature map, multiple candidate paths between the target second voxel unit and the target voxel unit are determined from the global intermediate feature map.

[0125] For an exemplary description, see Figure 4c As shown, Figure 4cAn interactive schematic diagram of determining an optimal path provided by an embodiment of the present application is shown, wherein a first voxel unit in a local intermediate feature map 410 is taken as an example, the target second voxel unit corresponding to the first voxel unit on the global intermediate feature map 420 is denoted as A, and the target voxel unit corresponding to the vascular trunk of the target blood vessel on the global intermediate feature map 420 is denoted as C, such as Figure 4c As shown, on the global intermediate feature map 420, there are multiple candidate paths between the target second voxel unit A and the target voxel unit C, that is, each reachable path between the target second voxel unit A and the target voxel unit C can be used as a candidate path for screening the optimal path.

[0126] Step a2: According to the similarity calculation result, a candidate path with the shortest distance and the largest sum of similarities between each second voxel unit covered on the path and the first voxel unit is selected from the multiple candidate paths as the optimal path.

[0127] Exemplary descriptions, such as Figure 4c As shown, taking the candidate path between the grayed-out target second voxel unit A and the target voxel unit C as an example, the candidate path is covered with multiple second voxel units (grayed-out voxel units) on the global intermediate feature map 420, and the number and size of these grayed-out voxel units are used to determine the length of the distance between the target second voxel unit A and the target voxel unit C under the candidate path; by calculating the similarity between each grayed-out voxel unit and the corresponding first voxel unit of the target second voxel unit A on the local intermediate feature map 410, the above-mentioned similarity and value corresponding to the candidate path can be obtained; thereby, based on the calculation result of each candidate path, the above-mentioned optimal path is screened out from multiple candidate paths.

[0128] It should be noted that, in step S403, in addition to the optimal path determination method shown in the above steps a1-a2 (i.e., taking the candidate path with the shortest distance and the largest similarity and value between each second voxel unit covered and the first voxel unit as the optimal path), the screening method of the "optimal path" can also be adjusted based on actual path screening requirements. For example, when the candidate path with the shortest distance is different from the candidate path with the largest similarity and value, one of the candidate paths can be selected as the optimal path, or the overlapping area between the two candidate paths can be selected as the optimal path, etc.; the optimal path determination method shown in the above steps a1-a2 is only used to characterize a preferred implementation method provided in the embodiment of the present application, and the embodiment of the present application does not impose any limitation on the specific screening method of the optimal path.

[0129] S404, according to the similarity between each third voxel unit and the target second voxel unit, and the feature distance between each third voxel unit and the target second voxel unit, calculate the fusion feature corresponding to the target second voxel unit by weighted summation.

[0130] Here, the third voxel unit is used to represent the voxel unit covered by the optimal path on the global intermediate feature map; for example, Figure 4c Taking the path between the target second voxel unit A and the target voxel unit C marked in gray as an example, if this path is the optimal path, then each gray-marked voxel unit covered by the optimal path on the global intermediate feature map 420 is the above-mentioned third voxel unit.

[0131] Specifically, taking the third voxel unit covered under the above optimal path as n voxel units: F1...Fn as an example, if the similarities between each third voxel unit and the target second voxel unit are: W1...Wn, and the characteristic distance between each third voxel unit and the target second voxel unit is: Q1...Qn; then through the weighted summation method, the fusion feature corresponding to the target second voxel unit can be obtained = Q1×W1+Q2×W2+…+Qn×Wn.

[0132] It should be noted that, when executing step S404, the similarity between the third voxel unit and the target second voxel unit can be calculated by any similarity calculation method, for example, similarity calculation methods such as Euclidean distance and Mahalanobis distance can be used; the embodiment of the present application also does not impose any limitation on the specific similarity calculation method used in step S404.

[0133] S405, performing feature splicing on the fusion feature and the local feature of the first voxel unit in the local intermediate feature map, and based on the feature splicing result corresponding to each of the first voxel units, outputting the feature representation result corresponding to the local image block in the intermediate stage of blood vessel segmentation prediction.

[0134] Here, the feature splicing is equivalent to splicing the fused feature with the local feature of the first voxel unit in the local intermediate feature map end to end, and outputting the feature splicing result obtained after the end-to-end splicing as the feature representation result corresponding to the local image block in the intermediate stage of vascular segmentation prediction, so that the second vascular segmentation model obtains a feature representation with a stronger ability to distinguish arteriovenous categories on the basis of the original local features (that is, the feature splicing result of the fused feature and the local feature), which effectively improves the accuracy of the vascular segmentation result output by the second vascular segmentation model and reduces the segmentation error of the second vascular segmentation model due to arteriovenous confusion.

[0135] For the model application process shown in the above steps S101-S104, during the model training process, refer to Figure 5 As shown, Figure 5 A flow chart of a method for training a first blood vessel segmentation model provided in an embodiment of the present application is shown. The method includes steps S501-S503; specifically:

[0136] S501, acquiring a plurality of medical images containing target blood vessels as first training sample data.

[0137] Here, each of the medical images contains a blood vessel marking result of the target blood vessel; the blood vessel marking result includes: an arterial blood vessel marking result and a venous blood vessel marking result of the target blood vessel.

[0138] It should be noted that the above-mentioned medical images are also used to represent three-dimensional images containing target blood vessels. For example, they can be CT (Computed Tomography) images containing target blood vessels, or CTA images (i.e., CT vascular imaging) containing target blood vessels, etc.; the embodiments of the present application do not impose any limitations on the specific data types to which the above-mentioned medical images belong.

[0139] S502: Input the first training sample data into a first blood vessel segmentation model, and output a third blood vessel segmentation prediction result for each of the medical images.

[0140] Here, the third blood vessel segmentation prediction result includes an arterial blood vessel prediction result and a venous blood vessel prediction result of the target blood vessel.

[0141] Specifically, the specific implementation of outputting the third blood vessel segmentation prediction result in the above step S502 is the same as the implementation of outputting the first blood vessel segmentation prediction result in the above step S101, and the repeated parts are not repeated here.

[0142] S503 , training the first blood vessel segmentation model based on the segmentation loss between the third blood vessel segmentation prediction result of each of the medical images and the blood vessel labeling result corresponding to each of the medical images until the first blood vessel segmentation model reaches convergence.

[0143] Specifically, when calculating the above-mentioned segmentation loss, any loss function available for image segmentation can be used for calculation, such as a cross entropy loss function, a focal loss loss function, etc. The embodiment of the present application does not impose any limitation on the specific calculation method of the above-mentioned segmentation loss.

[0144] With respect to the model application process shown in the above steps S101-S104, on the basis of obtaining the first blood vessel segmentation model trained above, in the model training process, refer to Figure 6 As shown, Figure 6 A flow chart of a method for training a second blood vessel segmentation model provided in an embodiment of the present application is shown. The method includes steps S601-S604; specifically:

[0145] S601, for each of the medical images, cutting the medical image into a plurality of sample image blocks, and cutting out an initial global region corresponding to the target blood vessel from the medical image according to a position of each of the sample image blocks in the medical image and a positioning result of the vascular trunk of the target blood vessel in the medical image.

[0146] Here, the initial global region is used to characterize an image region that includes both the blood vessel trunk of the target blood vessel and each of the sample image blocks.

[0147] Specifically, the cropping method of the sample image block and the cropping method of the initial global area in step S601 are the same as the cropping method of the local image block and the cropping method of the global area in step S102 above, and the repeated parts are not repeated here.

[0148] S602, inputting the initial global region into a trained first blood vessel segmentation model, and outputting an initial global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the initial global region through the trained first blood vessel segmentation model.

[0149] Here, the specific model training method of the first blood vessel segmentation model can refer to the model training steps shown in the aforementioned steps S501-S503, and the repeated parts will not be repeated here.

[0150] Specifically, the method of obtaining the initial global intermediate feature map in step S602 is the same as the method of obtaining the global intermediate feature map in the above step S103, and the repeated parts are not repeated here.

[0151] S603, inputting the multiple sample image blocks into the second vascular segmentation model, and during the process of the second vascular segmentation model performing vascular segmentation prediction on each sample image block, introducing the initial global intermediate feature map into the second vascular segmentation model as reference information for vascular segmentation prediction in a data interactive manner, and outputting a fourth vascular segmentation prediction result for the medical image.

[0152] Here, the fourth blood vessel segmentation prediction result includes an arterial blood vessel prediction result and a venous blood vessel prediction result of the target blood vessel.

[0153] Specifically, the method of obtaining the fourth blood vessel segmentation prediction result in step S603 is the same as the method of obtaining the second blood vessel segmentation prediction result in step S104, and the repeated parts are not repeated here.

[0154] S604: Training the second blood vessel segmentation model based on the segmentation loss between the fourth blood vessel segmentation prediction result of each of the medical images and the blood vessel labeling result corresponding to each of the medical images until the second blood vessel segmentation model reaches convergence.

[0155] Specifically, when calculating the above-mentioned segmentation loss in step S604, any loss function available for image segmentation can also be used for calculation, such as a cross entropy loss function, a focal loss loss function, etc. The embodiment of the present application does not impose any limitation on the specific calculation method of the above-mentioned segmentation loss.

[0156] The above-mentioned blood vessel segmentation method provided by the embodiment of the present application inputs an original medical image containing a target blood vessel into a first blood vessel segmentation model, outputs a first blood vessel segmentation prediction result for the target blood vessel, and obtains a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result; cuts the original medical image into multiple local image blocks, and cuts out a global region corresponding to the target blood vessel from the original medical image according to the position of each local image block in the original medical image and the blood vessel trunk positioning result; inputs the global region into the first blood vessel segmentation model, and outputs a global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the global region through the first blood vessel segmentation model; inputs multiple local image blocks into a second blood vessel segmentation model, and in the process of the second blood vessel segmentation model performing blood vessel segmentation prediction on each local image block, introduces the global intermediate feature map as reference information for blood vessel segmentation prediction into the second blood vessel segmentation model in a data interactive manner, and outputs a second blood vessel segmentation prediction result for the target blood vessel.

[0157] By determining the vascular segmentation results in this staged manner, the present application can introduce the global information between each local area and the vascular trunk into the second-stage model's vascular segmentation prediction process for each local area in a manner that conforms to the vascular topology, thereby enabling the second-stage vascular segmentation model to obtain a feature representation with a stronger ability to distinguish between arteriovenous categories, effectively improving the accuracy of the vascular segmentation results and reducing segmentation errors caused by arteriovenous confusion in the vascular segmentation model.

[0158] Based on the same inventive concept, the present application also provides a blood vessel segmentation device corresponding to the above-mentioned blood vessel segmentation method. Since the principle of solving the problem by the blood vessel segmentation device in the embodiment of the present application is similar to the above-mentioned blood vessel segmentation method in the embodiment of the present application, the implementation of the blood vessel segmentation device can refer to the implementation of the above-mentioned blood vessel segmentation method, and the repeated parts will not be repeated.

[0159] Reference Figure 7 As shown, Figure 7A schematic diagram of the structure of a blood vessel segmentation device provided in an embodiment of the present application is shown, and the blood vessel segmentation device includes:

[0160] The coarse segmentation module 701 is used to input the original medical image containing the target blood vessel into the first blood vessel segmentation model, output a first blood vessel segmentation prediction result for the target blood vessel, and obtain a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result; wherein the first blood vessel segmentation prediction result includes a first artery segmentation prediction result and a first vein segmentation prediction result of the target blood vessel;

[0161] A cropping module 702 is used to crop the original medical image into a plurality of local image blocks, and crop a global region corresponding to the target blood vessel from the original medical image according to the position of each local image block in the original medical image and the vascular trunk positioning result; wherein the global region is used to represent an image region including the vascular trunk of the target blood vessel and each local image block at the same time;

[0162] An intermediate output module 703, configured to input the global region into the first blood vessel segmentation model, and output a global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the global region through the first blood vessel segmentation model;

[0163] The data interaction module 704 is used to input the multiple local image blocks into the second blood vessel segmentation model, and in the process of the second blood vessel segmentation model performing blood vessel segmentation prediction on each local image block, introduce the global intermediate feature map into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interaction manner, and output a second blood vessel segmentation prediction result for the target blood vessel; wherein the second blood vessel segmentation prediction result includes a second artery segmentation prediction result and a second vein segmentation prediction result of the target blood vessel.

[0164] In an optional implementation, when obtaining the vascular trunk positioning result of the target vessel from the first vessel segmentation prediction result, the coarse segmentation module 701 is configured to:

[0165] Extracting the arterial blood vessel midline of the target blood vessel from the maximum connected domain of the first artery segmentation prediction result;

[0166] Calculating the shortest distance between the arterial vessel midline and each voxel unit located on the surface of the arterial vessel in the first arterial segmentation prediction result, and taking the region where the arterial vessel is located where the calculation result is greater than or equal to the first preset distance threshold as the arterial vessel trunk positioning result in the vascular trunk positioning result;

[0167] Extracting the venous blood vessel centerline of the target blood vessel from the maximum connected domain of the first vein segmentation prediction result;

[0168] Calculate the shortest distance between the centerline of the vein and each voxel unit located on the surface of the vein in the first vein segmentation prediction result, and use the area where the vein is located where the calculation result is greater than or equal to the second preset distance threshold as the vein trunk positioning result in the blood vessel trunk positioning result.

[0169] In an optional implementation, when the first blood vessel segmentation model is used to output the global intermediate feature map generated in the intermediate stage of performing blood vessel segmentation prediction on the global region, the intermediate output module 703 is used to:

[0170] In the intermediate stage of the first blood vessel segmentation model performing blood vessel segmentation prediction on the input global region, the first blood vessel segmentation model is used to output one or more feature representation results after intermediate processing on the global region; wherein each of the feature representation results is used to represent a global intermediate feature map.

[0171] In an optional implementation, when the second blood vessel segmentation model performs blood vessel segmentation prediction for each local image block, the global intermediate feature map is introduced into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interactive manner, the data interactive module 704 is used to:

[0172] For each of the input local image blocks, in an intermediate stage when the second blood vessel segmentation model performs blood vessel segmentation prediction on the local image block, obtaining a local intermediate feature map after the second blood vessel segmentation model performs intermediate processing on the local image block;

[0173] For each first voxel unit included in the local intermediate feature map, calculate the similarity between the first voxel unit and each second voxel unit included in the global intermediate feature map, and obtain a similarity calculation result corresponding to the first voxel unit;

[0174] According to the similarity calculation result and the target voxel unit included in the global intermediate feature map, an optimal path between a target second voxel unit and the target voxel unit is determined from the global intermediate feature map; wherein the target second voxel unit is used to represent a second voxel unit corresponding to the same physical position as the first voxel unit on the target blood vessel; and the target voxel unit is used to represent a voxel unit corresponding to the blood vessel trunk of the target blood vessel on the global intermediate feature map;

[0175] According to the similarity between each third voxel unit and the target second voxel unit, and the feature distance between each third voxel unit and the target second voxel unit, a fusion feature corresponding to the target second voxel unit is calculated by weighted summation; wherein the third voxel unit is used to characterize the voxel unit covered by the optimal path on the global intermediate feature map;

[0176] The fusion feature is spliced ​​with the local feature of the first voxel unit in the local intermediate feature map, and based on the feature splicing result corresponding to each of the first voxel units, the feature representation result corresponding to the local image block in the intermediate stage of blood vessel segmentation prediction is output.

[0177] In an optional implementation, when determining the optimal path between the target second voxel unit and the target voxel unit from the global intermediate feature map according to the similarity calculation result and the target voxel unit included in the global intermediate feature map, the data interaction module 704 is used to:

[0178] Determine, from the global intermediate feature map, a plurality of candidate paths between the target second voxel unit and the target voxel unit according to position information of the target voxel unit in the global intermediate feature map;

[0179] According to the similarity calculation result, a candidate path with the shortest distance and the largest sum of similarities between each second voxel unit covered on the path and the first voxel unit is selected from the multiple candidate paths as the optimal path.

[0180] In an optional implementation, the coarse segmentation module 701 is used to train and obtain the first blood vessel segmentation model by the following method:

[0181] Acquire a plurality of medical images containing target blood vessels as first training sample data; wherein each of the medical images contains a blood vessel labeling result of the target blood vessel; the blood vessel labeling result includes: an arterial blood vessel labeling result and a venous blood vessel labeling result of the target blood vessel;

[0182] Inputting the first training sample data into a first blood vessel segmentation model, and outputting a third blood vessel segmentation prediction result for each of the medical images; wherein the third blood vessel segmentation prediction result includes an arterial blood vessel prediction result and a venous blood vessel prediction result of the target blood vessel;

[0183] Based on the segmentation loss between the third blood vessel segmentation prediction result of each of the medical images and the blood vessel labeling result corresponding to each of the medical images, the first blood vessel segmentation model is trained until the first blood vessel segmentation model reaches convergence.

[0184] In an optional implementation, the data interaction module 704 is used to train the second blood vessel segmentation model by the following method:

[0185] For each of the medical images, the medical image is cut into a plurality of sample image blocks, and according to the position of each of the sample image blocks in the medical image and the positioning result of the vascular trunk of the target blood vessel in the medical image, an initial global region corresponding to the target blood vessel is cut out from the medical image; wherein the initial global region is used to characterize an image region including the vascular trunk of the target blood vessel and each of the sample image blocks;

[0186] Inputting the initial global region into a trained first blood vessel segmentation model, and outputting an initial global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the initial global region through the trained first blood vessel segmentation model;

[0187] The plurality of sample image blocks are input into a second blood vessel segmentation model. During the process of the second blood vessel segmentation model performing blood vessel segmentation prediction for each sample image block, the initial global intermediate feature map is introduced into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interactive manner, and a fourth blood vessel segmentation prediction result for the medical image is output; wherein the fourth blood vessel segmentation prediction result includes an arterial blood vessel prediction result and a venous blood vessel prediction result of the target blood vessel;

[0188] Based on the segmentation loss between the fourth blood vessel segmentation prediction result of each of the medical images and the blood vessel labeling result corresponding to each of the medical images, the second blood vessel segmentation model is trained until the second blood vessel segmentation model reaches convergence.

[0189] Based on the same inventive concept, Figure 8 As shown, an embodiment of the present application provides a computer device 800 for executing the blood vessel segmentation method in the present application, and the device includes a memory 801, a processor 802, and a computer program stored in the memory 801 and executable on the processor 802, wherein the processor 802 implements the steps of the blood vessel segmentation method when executing the computer program.

[0190] Specifically, the memory 801 and the processor 802 may be general-purpose memories and processors, which are not specifically limited here. When the processor 802 runs the computer program stored in the memory 801, the blood vessel segmentation method described above can be executed.

[0191] Corresponding to the blood vessel segmentation method in the present application, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned blood vessel segmentation method are executed.

[0192] Specifically, the storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the storage medium is run, the above-mentioned blood vessel segmentation method can be executed.

[0193] In the embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0194] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0195] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0196] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0197] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0198] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A blood vessel segmentation method, characterized in that: The blood vessel segmentation method comprises: Inputting an original medical image containing a target blood vessel into a first blood vessel segmentation model, outputting a first blood vessel segmentation prediction result for the target blood vessel, and obtaining a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result; wherein the first blood vessel segmentation prediction result includes a first artery segmentation prediction result and a first vein segmentation prediction result of the target blood vessel; The original medical image is cut into a plurality of local image blocks, and according to the position of each local image block in the original medical image and the vascular trunk positioning result, a global region corresponding to the target blood vessel is cut out from the original medical image; wherein the global region is used to characterize an image region including the vascular trunk of the target blood vessel and each local image block; Inputting the global region into the first blood vessel segmentation model, and outputting a global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the global region through the first blood vessel segmentation model; The multiple local image blocks are input into a second blood vessel segmentation model. In the process of the second blood vessel segmentation model performing blood vessel segmentation prediction on each local image block, the global intermediate feature map is introduced into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interactive manner, and a second blood vessel segmentation prediction result for the target blood vessel is output; wherein the second blood vessel segmentation prediction result includes a second artery segmentation prediction result and a second vein segmentation prediction result of the target blood vessel.

2. The blood vessel segmentation method according to claim 1, characterized in that: The step of obtaining the main vessel trunk positioning result of the target vessel from the first vessel segmentation prediction result includes: Extracting the arterial blood vessel midline of the target blood vessel from the maximum connected domain of the first artery segmentation prediction result; Calculating the shortest distance between the arterial vessel midline and each voxel unit located on the surface of the arterial vessel in the first arterial segmentation prediction result, and taking the region where the arterial vessel is located where the calculation result is greater than or equal to the first preset distance threshold as the arterial vessel trunk positioning result in the vascular trunk positioning result; Extracting the venous blood vessel centerline of the target blood vessel from the maximum connected domain of the first vein segmentation prediction result; Calculate the shortest distance between the centerline of the vein and each voxel unit located on the surface of the vein in the first vein segmentation prediction result, and use the area where the vein is located where the calculation result is greater than or equal to the second preset distance threshold as the vein trunk positioning result in the blood vessel trunk positioning result.

3. The blood vessel segmentation method according to claim 1, characterized in that: The step of outputting a global intermediate feature map generated in an intermediate stage of performing blood vessel segmentation prediction on the global region through the first blood vessel segmentation model comprises: In the intermediate stage of the first blood vessel segmentation model performing blood vessel segmentation prediction on the input global region, the first blood vessel segmentation model is used to output one or more feature representation results after intermediate processing on the global region; wherein each of the feature representation results is used to represent a global intermediate feature map.

4. The blood vessel segmentation method according to claim 1, characterized in that: In the process of performing blood vessel segmentation prediction for each local image block by the second blood vessel segmentation model, introducing the global intermediate feature map as reference information for blood vessel segmentation prediction into the second blood vessel segmentation model in a data interactive manner, comprising: For each of the input local image blocks, in an intermediate stage when the second blood vessel segmentation model performs blood vessel segmentation prediction on the local image block, obtaining a local intermediate feature map after the second blood vessel segmentation model performs intermediate processing on the local image block; For each first voxel unit included in the local intermediate feature map, calculate the similarity between the first voxel unit and each second voxel unit included in the global intermediate feature map, and obtain a similarity calculation result corresponding to the first voxel unit; According to the similarity calculation result and the target voxel unit included in the global intermediate feature map, an optimal path between a target second voxel unit and the target voxel unit is determined from the global intermediate feature map; wherein the target second voxel unit is used to represent a second voxel unit corresponding to the same physical position as the first voxel unit on the target blood vessel; and the target voxel unit is used to represent a voxel unit corresponding to the blood vessel trunk of the target blood vessel on the global intermediate feature map; According to the similarity between each third voxel unit and the target second voxel unit, and the feature distance between each third voxel unit and the target second voxel unit, a fusion feature corresponding to the target second voxel unit is calculated by weighted summation; wherein the third voxel unit is used to characterize the voxel unit covered by the optimal path on the global intermediate feature map; The fusion feature is spliced ​​with the local feature of the first voxel unit in the local intermediate feature map, and based on the feature splicing result corresponding to each of the first voxel units, the feature representation result corresponding to the local image block in the intermediate stage of blood vessel segmentation prediction is output.

5. The blood vessel segmentation method according to claim 4, characterized in that: The step of determining an optimal path between a target second voxel unit and the target voxel unit from the global intermediate feature map according to the similarity calculation result and the target voxel unit included in the global intermediate feature map comprises: Determine, from the global intermediate feature map, a plurality of candidate paths between the target second voxel unit and the target voxel unit according to position information of the target voxel unit in the global intermediate feature map; According to the similarity calculation result, a candidate path with the shortest distance and the largest sum of similarities between each second voxel unit covered on the path and the first voxel unit is selected from the multiple candidate paths as the optimal path.

6. The blood vessel segmentation method according to claim 1, characterized in that: The first blood vessel segmentation model is trained by the following method: Acquire a plurality of medical images containing target blood vessels as first training sample data; wherein each of the medical images contains a blood vessel labeling result of the target blood vessel; the blood vessel labeling result includes: an arterial blood vessel labeling result and a venous blood vessel labeling result of the target blood vessel; Inputting the first training sample data into a first blood vessel segmentation model, and outputting a third blood vessel segmentation prediction result for each of the medical images; wherein the third blood vessel segmentation prediction result includes an arterial blood vessel prediction result and a venous blood vessel prediction result of the target blood vessel; Based on the segmentation loss between the third blood vessel segmentation prediction result of each of the medical images and the blood vessel labeling result corresponding to each of the medical images, the first blood vessel segmentation model is trained until the first blood vessel segmentation model reaches convergence.

7. The blood vessel segmentation method according to claim 6, characterized in that: The second blood vessel segmentation model is trained by the following method: For each of the medical images, the medical image is cut into a plurality of sample image blocks, and according to the position of each of the sample image blocks in the medical image and the positioning result of the vascular trunk of the target blood vessel in the medical image, an initial global region corresponding to the target blood vessel is cut out from the medical image; wherein the initial global region is used to characterize an image region including the vascular trunk of the target blood vessel and each of the sample image blocks; Inputting the initial global region into a trained first blood vessel segmentation model, and outputting an initial global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the initial global region through the trained first blood vessel segmentation model; The plurality of sample image blocks are input into a second blood vessel segmentation model. During the process of the second blood vessel segmentation model performing blood vessel segmentation prediction for each sample image block, the initial global intermediate feature map is introduced into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interactive manner, and a fourth blood vessel segmentation prediction result for the medical image is output; wherein the fourth blood vessel segmentation prediction result includes an arterial blood vessel prediction result and a venous blood vessel prediction result of the target blood vessel; Based on the segmentation loss between the fourth blood vessel segmentation prediction result of each of the medical images and the blood vessel labeling result corresponding to each of the medical images, the second blood vessel segmentation model is trained until the second blood vessel segmentation model reaches convergence.

8. A blood vessel segmentation device, characterized in that: The blood vessel segmentation device comprises: a coarse segmentation module, configured to input an original medical image containing a target blood vessel into a first blood vessel segmentation model, output a first blood vessel segmentation prediction result for the target blood vessel, and obtain a blood vessel trunk positioning result of the target blood vessel from the first blood vessel segmentation prediction result; wherein the first blood vessel segmentation prediction result includes a first artery segmentation prediction result and a first vein segmentation prediction result of the target blood vessel; A cropping module, used for cropping the original medical image into a plurality of local image blocks, and cropping a global region corresponding to the target blood vessel from the original medical image according to the position of each local image block in the original medical image and the vascular trunk positioning result; wherein the global region is used to represent an image region including the vascular trunk of the target blood vessel and each local image block; an intermediate output module, configured to input the global region into the first blood vessel segmentation model, and output a global intermediate feature map generated in an intermediate stage of blood vessel segmentation prediction for the global region through the first blood vessel segmentation model; a data interaction module, configured to input the plurality of local image blocks into a second blood vessel segmentation model; and in a process in which the second blood vessel segmentation model performs blood vessel segmentation prediction on each local image block, introduce the global intermediate feature map into the second blood vessel segmentation model as reference information for blood vessel segmentation prediction in a data interaction manner, and output a second blood vessel segmentation prediction result for the target blood vessel; wherein the second blood vessel segmentation prediction result includes a second artery segmentation prediction result and a second vein segmentation prediction result of the target blood vessel.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of the blood vessel segmentation method according to any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the blood vessel segmentation method according to any one of claims 1 to 7 are executed.

Images