Method, device and medium for performing blood vessel segmentation on medical images

Abstract

The present disclosure relates to a method, device, and medium for performing blood vessel segmentation on a medical image. The method may include obtaining a medical image containing several regions for blood vessel segmentation, wherein the blood vessels in each region have different structural properties. A processor may be used to divide the medical image into sub-medical images of each region according to the region. Based on the sub-medical images of each region, the processor may be used to determine the corresponding blood vessel segmentation results using the corresponding blood vessel segmentation model of each region. The processor may be used to fuse the blood vessel segmentation results of the sub-medical images of each region to obtain the blood vessel segmentation result of the medical image. The method and device differentiate and adapt the blood vessel segmentation model for each region to separately segment the blood vessels in each region, so that the entire blood vessel segmentation process is fast, effective, and highly accurate.

Description

[0001] This application is a divisional application of the Chinese invention patent application with application number 202111487795.X, application date December 8, 2021, and invention name “A method, device and medium for blood vessel segmentation in medical images”. Technical Field

[0002] The present invention relates to the technical field of medical image processing, and in particular to a method and device for performing blood vessel segmentation on a medical image. Background Art

[0003] With the advancement of medical imaging equipment, doctors can use medical images for disease screening and diagnosis, surgical planning, and prognosis assessment. For example, computed tomography angiography (CTA) or magnetic resonance angiography (MRA) can be used to image blood vessels, vascular variations, and vascular lesions throughout the body. Medical vascular image analysis plays a vital role in clinical practice, and vascular image segmentation is a crucial component of this process. Early vascular segmentation methods primarily performed feature selection and image analysis on angiographic images to obtain segmented results, but these methods were susceptible to interference from surrounding vascular structures, such as bone, plaque, and stents. Current vascular segmentation methods primarily utilize deep neural networks to directly segment angiographic images to obtain segmented results. While these deep neural network-based vascular segmentation methods are effective, they generally employ a single model. Due to the complexity of vascular images, their segmentation results can be problematic. Specifically, there may be an under-segmentation problem, where blood vessels with larger or smaller radii are easily missed and broken; there may be an over-segmentation problem, where non-vascular areas that are similar to blood vessels in the image are segmented, resulting in adhesion noise. Summary of the Invention

[0004] The present disclosure is intended to solve the above technical problems. The present disclosure is intended to provide a computer-implemented method and apparatus that can use a learning network to quickly and accurately segment blood vessels with different structural attributes in medical images of several parts that may contain blood vessels with complex and variable structural attributes (such as but not limited to morphology, size, etc.) (such as but not limited to a large range of blood vessel sizes, variable blood vessel morphology). The method and apparatus divide the medical image into sub-medical images of various parts according to the differences in the structural attributes of the blood vessels, apply an adapted learning network to each sub-medical image of each part to achieve the segmentation of each blood vessel, and fuse the blood vessel segmentation results of each part to obtain the blood vessel segmentation result of the entire medical image, so that the entire blood vessel segmentation process is fast and effective. Furthermore, the method and apparatus can target the differences in morphology, curvature, radius size of blood vessels at special locations (such as the intersection or boundary of blood vessels) or the focus areas preset by the doctor, and achieve accurate segmentation of blood vessels.

[0005] In one embodiment, the present disclosure relates to a method for performing vascular segmentation on a medical image. The method may include obtaining a medical image containing multiple regions for vascular segmentation, wherein the vascular structures of each region differ. The method may also include, using a processor, dividing the medical image into sub-medical images corresponding to each region. The method may also include, based on the sub-medical images of each region, determining, using the processor, a corresponding vascular segmentation model for each region. The method may also include, using the processor, fusing the vascular segmentation results of the sub-medical images of each region to obtain a vascular segmentation result for the medical image.

[0006] In another embodiment, the present disclosure relates to a device for performing blood vessel segmentation on a medical image. The device may include an interface and a processor. The interface may be configured to receive a medical image containing several parts for which blood vessel segmentation is to be performed, wherein the blood vessels in each part have different structural properties. The processor may be configured to obtain (for example, via the interface) a medical image containing several parts for which blood vessel segmentation is to be performed, wherein the blood vessels in each part have different structural properties. The processor may also be configured to divide the medical image into sub-medical images of each part according to the part. The processor may also be configured to determine a corresponding blood vessel segmentation result based on the sub-medical images of each part using a corresponding blood vessel segmentation model for each part. In addition, the processor may be configured to fuse the blood vessel segmentation results of the sub-medical images of each part to obtain a blood vessel segmentation result for the medical image.

[0007] In another embodiment, the present disclosure relates to a non-transitory computer-readable medium having instructions stored thereon, wherein the instructions, when executed by a processor, implement a method for performing vascular segmentation on a medical image. The method may include obtaining a medical image comprising multiple regions for which vascular segmentation is to be performed, wherein the vascular structures of the respective regions differ. The method may also include dividing the medical image into sub-medical images of the respective regions according to the regions. The method may also include determining corresponding vascular segmentation results based on the sub-medical images of the respective regions using corresponding vascular segmentation models for the respective regions. The method may also include fusing the vascular segmentation results of the sub-medical images of the respective regions to obtain a vascular segmentation result for the medical image.

[0008] The embodiments of the present disclosure have the following advantages over the prior art:

[0009] 1. Compared with traditional blood vessel segmentation methods, the present invention uses a deep neural network to realize automatic blood vessel segmentation in medical images. This method has 1) good blood vessel segmentation effect and is suitable for data from different patients and different manufacturers; 2) fast blood vessel segmentation speed.

[0010] 2. Compared with current vascular segmentation methods, this method also uses a deep neural network to automatically segment blood vessels in medical images. However, this method does not use a single model to complete the automatic segmentation of blood vessels. Instead, it takes into account the large differences in vascular morphology in different parts of the medical image. First, a learning network is used to divide the medical image into several partial regions. Then, several corresponding learning networks are used to separately segment these partial regions. Finally, the vascular segmentation results of the several partial regions are fused. This method 1) has good vascular segmentation effect and is suitable for medical images such as head and neck CTA, which have large differences in vascular morphology.

[0011] 2) Solve the problem of poor blood vessel segmentation effect of using a single model based on a learning network, avoid blood vessel rupture and adhesion noise, and thus prevent these situations from affecting doctors' diagnosis of cardiovascular diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In the drawings, which are not necessarily drawn to scale, the same reference numerals may describe similar parts in different views. The same numbers with letter suffixes or different letter suffixes may represent different instances of similar parts. The accompanying drawings generally illustrate various embodiments by way of example rather than by way of limitation, and are used together with the description and claims to illustrate the disclosed embodiments. Where appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present method, apparatus, or a non-transitory computer-readable medium having stored thereon instructions for implementing the method.

[0013] Figure 1 A method for segmenting blood vessels in a medical image according to an embodiment of the present disclosure is shown.

[0014] Figure 2 The process of performing blood vessel segmentation on a medical image according to an embodiment of the present disclosure is shown.

[0015] FIG3( a ) shows blood vessels after segmenting a head sub-medical image according to an embodiment of the present disclosure.

[0016] FIG3( b ) shows blood vessels after segmenting the neck sub-medical image according to an embodiment of the present disclosure.

[0017] FIG3( c ) shows blood vessels after segmenting the chest sub-medical image according to an embodiment of the present disclosure.

[0018] Figure 4 The detailed process of the method for segmenting blood vessels in a medical image according to an embodiment of the present disclosure is shown.

[0019] Figure 5 A method for segmenting blood vessels in a head sub-medical image according to an embodiment of the present disclosure is shown.

[0020] Figure 6 A method for segmenting blood vessels in a neck sub-medical image according to an embodiment of the present disclosure is shown.

[0021] Figure 7 A method for segmenting blood vessels in a chest sub-medical image according to an embodiment of the present disclosure is shown.

[0022] Figure 8 A block diagram illustrating an exemplary apparatus for segmenting blood vessels in a medical image according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0023] The present invention will be further described below with reference to the accompanying drawings and examples.

[0024] Figure 1 The method for segmenting blood vessels in a medical image according to an embodiment of the present disclosure is shown. Figure 1 As shown in , a method for performing blood vessel segmentation on a medical image begins with step S1, where a medical image containing multiple regions to be segmented is obtained, where the blood vessels in each region have different structural properties. In some embodiments, the structural properties of the blood vessels include at least one of morphology and size.

[0025] In step S2, a processor may be used to divide the medical image into sub-medical images of each part according to the parts. In some embodiments, the medical image is a 3D medical image, such as, but not limited to, a head and neck CTA image, a lung CT image, and the like. For a 3D medical image, dividing the medical image into sub-medical images of each part according to the parts specifically includes: based on the 3D medical image, using a slice classification model to identify key slices in the 3D medical image that are the junctions of adjacent parts; and using the identified key slices to achieve the division of the sub-medical images according to the parts. In some embodiments, the slice classification model can be implemented using a two-dimensional learning network, trained using training samples having classification information of slices corresponding to the parts.

[0026] The medical image is an image of one or more scanned parts of a patient, and the medical image generally contains vascular tissue. The medical image can be obtained by a computer device from a post-processing workstation or a picture archiving and communication system (PACS). In this embodiment, the computer device can obtain medical images uploaded to the PACS system by radiologists / imaging technicians in real time, or it can obtain all medical images within this time period from the PACS system at fixed time intervals. In this embodiment, the computer device can also obtain the medical images to be segmented from the hospital information management system (HIS), clinical information management system (CIS), radiology information management system (RIS), electronic medical record system (EMR) and related medical image cloud storage platforms.

[0027] In step S3, the processor can determine corresponding vascular segmentation results based on the sub-medical images of each region using the corresponding vascular segmentation model. In some embodiments, each vascular segmentation model is trained using training samples containing classification information for the corresponding vascular regions. The classification information can be multivariate; accordingly, the classification information for the vascular regions can include multiple labels to meet the physician's needs for more detailed vascular classification.

[0028] In step S4, the processor may be used to fuse the blood vessel segmentation results of the sub-medical images of various parts to obtain the blood vessel segmentation result of the medical image.

[0029] The present disclosure is described below using a head and neck CTA medical image as an example. However, it should be noted that the present disclosure is applicable to various medical images containing multiple parts. For example, the medical image may include at least one of a head and neck CTA image, a head and neck MRA image, and a lung CT image.

[0030] Figure 2 FIG. 4 shows a process of performing blood vessel segmentation on a medical image according to an embodiment of the present disclosure. Figure 2 As shown in FIG, a slice classification model 202 is first used to divide a head and neck CTA three-dimensional medical image 212 into three sub-regions: a head region 203, a neck region 204, and a chest region 205. The structural properties of the blood vessels in each region differ significantly. For example, the arteries in the head region 203 have a vessel radius of size S1, the arteries in the neck region 204 have a vessel radius of size S2, and the arteries in the chest region 205 have a vessel radius of size S3. For example, the vessel curvature of the arteries in the head region 203 is significantly higher than the vessel curvature of the arteries in the neck region 204 and the chest region 205. For another example, the vessel density of the arteries in the head region 203 is significantly higher than the vessel density of the arteries in the neck region 204 and the chest region 205.

[0031] The head vessel segmentation model 206 can be used to segment the blood vessels in the head region 203, the neck vessel segmentation model 207 can be used to segment the blood vessels in the neck region 204, and the chest vessel segmentation model 208 can be used to segment the blood vessels in the chest region 205. Finally, the blood vessel segmentation results of the head region 203, the neck region 204, and the chest region 205 are fused to obtain the final blood vessel segmentation result.

[0032] In some embodiments, before the processor performs various processing on the medical image or the sub-medical image, the medical image or the sub-medical image may be normalized. For example, the normalization may include resampling the medical image or the sub-medical image to obtain a medical image or sub-medical image with corresponding voxel spacing, clipping the grayscale values ​​of the resampled medical image or sub-medical image using a set window width and window level, and performing normalization.

[0033] In some embodiments, determining a vascular segmentation result for each sub-medical image at each site using the processor and the corresponding vascular segmentation model for each site may specifically include: performing a coarse segmentation on the sub-medical image at each site to obtain a first image block of a target region containing the blood vessel to be segmented. Coarse segmentation can be achieved using various methods, such as coarse segmentation of the blood vessels or coarse segmentation of characteristic structures (bone or tissue) in the target region to reduce the workload of subsequent segmentation. Subsequently, a first vascular segmentation result can be determined based on the first image block of the target region using the first vascular segmentation model, serving as the vascular segmentation result for the sub-medical image at each site.

[0034] Furthermore, the method for segmenting blood vessels in a medical image may further include, for at least one of the various parts: performing centerline extraction on the determined first blood vessel segmentation result, and intercepting a second image block at the end of the extracted centerline. For example, Figures 3(a), 3(b), and 3(c) respectively illustrate blood vessels after segmentation of a head sub-image, a neck sub-image, and a chest sub-image according to an embodiment of the present disclosure; 301 and 304 in Figure 3(a), 305 and 308 in Figure 3(b), and 309 and 310 in Figure 3(c) are the ends of the blood vessel centerlines. Accordingly, second image blocks may be intercepted at the ends 301 and 304 of the blood vessel centerline shown in Figure 3(a), the ends 305 and 308 of the blood vessel centerline shown in Figure 3(b), and the ends 309 and 310 of the blood vessel centerline shown in Figure 3(c). Note that the term "end" herein may include areas surrounding the endpoints, and the intercepted second image block may include image information of the blood vessel junction areas of the head, neck, and chest.

[0035] A second blood vessel segmentation result at the end portion can be determined using a second blood vessel segmentation model based on the second image block at the end portion. Compared to the first blood vessel segmentation model, the second blood vessel segmentation model focuses more on a smaller field of view, which facilitates accurate segmentation of blood vessels at the end portion with varying structural properties, and avoids problems such as blood vessel breakage and adhesion noise.

[0036] Finally, the first and second vessel segmentation results can be fused to obtain a vessel segmentation result for the at least one sub-medical image. Benefiting from the coarse segmentation—segmentation of the target region using the first vessel segmentation model and precise segmentation of edge vessels in image blocks at the ends using the second vessel segmentation model—this not only achieves more accurate segmentation results for complex vessels across multiple locations, avoiding vessel breakage and adhesion noise, but also reduces workload and achieves faster computational speed.

[0037] In some embodiments, the at least one location from which the centerline is to be extracted and the image block at the end of the extracted centerline is to be intercepted can be directly set to each of the multiple locations, or to a location of interest preset by a doctor among the multiple locations. The doctor can use the preset location of interest to express feedback on more precise requirements for the vessel segmentation results and greater segmentation difficulty for the location of interest, thereby specifically performing the small-field-of-view vessel fine segmentation process for the location of interest, namely centerline extraction, end image block interception, and intercepted image block, to respond to the doctor's feedback and meet the doctor's requirements.

[0038] Furthermore, the method for segmenting blood vessels in medical images may further include selecting, from the plurality of regions, a region where blood vessel segmentation results are likely to break at the intersection of adjacent regions as the at least one region. This can significantly reduce or even prevent blood vessel breakage at the intersection of adjacent regions.

[0039] Furthermore, the method for segmenting blood vessels in medical images further includes selecting, from the plurality of regions, a region where the segmented blood vessels at the end of an adjacent region meet at least one of the following conditions as the at least one region. By applying the process of centerline extraction, end image block clipping, and small-field-of-view vessel segmentation to only those regions meeting these conditions, the workload can be reduced and the segmentation speed increased, significantly reducing or even preventing blood vessel breakage at the intersection of adjacent regions.

[0040] These determination conditions may include the diameter of the blood vessel at the end where it intersects with an adjacent portion being less than a first predetermined threshold. For example, in FIG3(a) end 302, FIG3(b) end 306, and FIG3(b) end 307, blood vessels with diameters less than the first predetermined threshold are more likely to be broken during segmentation.

[0041] These determination conditions may include that the curvature of the blood vessel at the end where it intersects with the adjacent portion is greater than a second predetermined threshold. For example, in the end 303 of FIG3(a), a blood vessel with a curvature greater than the second predetermined threshold is more likely to be broken during segmentation.

[0042] These criteria may include the contrast between a blood vessel at an end that intersects with an adjacent part and the surrounding area being less than a third predetermined threshold. For example, in end 302 in Figure 3(a) and end 307 in Figure 3(b), blood vessels with a contrast ratio less than the third predetermined threshold relative to the surrounding area are more likely to experience breakage and / or adhesion noise during segmentation.

[0043] In this disclosure, different body parts such as the head, neck and chest are mainly used as examples of several parts, but it should be understood that the several parts can not only be different body parts, but also different organs, or at least one of different organ parts in the same organ.

[0044] The following combination Figure 4 , the detailed process of the method for segmenting blood vessels in medical images is exemplified. Figure 4 As shown, in step S401, a medical image to be segmented is obtained.

[0045] In this embodiment, after the computer device acquires the medical image to be segmented, it can also preprocess the medical image, including converting the medical image format from Digital Imaging and Communications in Medicine (DICOM) format to NIFTI format, resampling the medical image to a medical image with a first voxel spacing, setting the window width and window level to clip the grayscale value range of the medical image, and normalizing the image to a value between [0, 1] to obtain a standardized image. Resampling refers to resampling. Given an original image with an original size of (512, 512, 256) and an original voxel spacing of (0.6, 0.6, 0.4), and a target voxel spacing of (1.0, 1.0, 1.0), the target size can be calculated to be (307, 307, 102), specifically calculated as 512×0.6 / 1.0=307 and 256×0.4 / 1.0=102. To transform from a size of (512,512,256) to (307,307,102), the grayscale of each voxel needs to be re-determined and obtained through resampling. Voxel spacing is a comprehensive reflection of the pixel spacing in each dimension. Voxel spacing is used for three-dimensional images, while pixel spacing is used for two-dimensional images. Specifically, it means the distance from the center of a voxel to the center of another adjacent voxel, measured in millimeters. For example, a voxel spacing of (0.6,0.6,0.4) means that the size of one voxel represents an actual or physical size of 0.6×0.6×0.4 cubic millimeters.

[0046] In step S402, the medical image may be input into a preset slice classification model to obtain a plurality of sub-medical images of the medical image; wherein the slice classification model is obtained by training based on training samples with information of the part of interest.

[0047] Specifically, the computer device may input the preprocessed medical image into a preset slice classification model. The slice classification model extracts and classifies the medical image features to obtain a site determination result. The site determination result may be the result of dividing the preprocessed medical image into several sub-images. In this embodiment, the site determination result may be the result of dividing the preprocessed medical image into three sub-images, representing a head sub-image, a neck sub-image, and a chest sub-image, respectively. Obviously, determining the three sub-images requires two key slices.

[0048] Therefore, the above-mentioned slice classification model is obtained by training based on training samples with slice classification information of the part of interest. The training method of the slice classification model may include marking two key slices in the training sample image based on experienced radiologists, collecting slices of the head, neck, and chest based on the marked key slice information, and using this as training samples and gold standard classification information corresponding to the training samples. The training sample is then input into the slice classification model to obtain the slice classification result of the training sample, and the loss between the slice classification result and the gold standard classification information is calculated. The network parameters of the slice classification model are adjusted according to the loss. When the loss is less than or equal to the preset threshold or convergence is reached, it indicates that the slice classification model training has converged. In this embodiment, the cross entropy loss function or other types of loss functions can be used when calculating the loss; when adjusting the network parameters, a stochastic gradient descent SGD optimizer or other types of optimizers can be used.

[0049] In this embodiment, since the slice classification of medical images is performed, the slice classification model can be a two-dimensional model. The two-dimensional model can be a convolutional neural network model, a recurrent neural network model or other deep learning model, which is not specifically limited here. For the obtained part determination results of the medical image, please refer to Figure 2 On the left, 1 medical image is divided into 3 sub-medical images.

[0050] In step S403, the plurality of sub-medical images may be input into corresponding vascular segmentation models to obtain vascular segmentation results of the plurality of sub-medical images; wherein the plurality of vascular segmentation models are obtained by training based on training samples with classification information of the vascular of interest.

[0051] Specifically, the computer device can pre-process the aforementioned sub-medical images and then input them into several preset vascular segmentation models to obtain vascular segmentation results for the sub-medical images. In this embodiment, a single medical image is divided into three sub-medical images, which are then input into the three preset vascular segmentation models to obtain three vascular segmentation results. The three vascular segmentation models are a head vascular segmentation model, a neck vascular segmentation model, and a chest vascular segmentation model. The results of the slice classification model are a head sub-medical image, a neck sub-medical image, and a chest sub-medical image.

[0052] Before inputting the head medical image into the preset head vascular segmentation model, the computer device preprocesses the sub-medical image. This includes resampling it to a medical image with a second voxel spacing, setting the window width and window level to crop the grayscale value range of the medical image, and normalizing the image to a range between -1 and 1 to obtain a standardized image. The head vascular segmentation model extracts and maps features from the sub-medical image to obtain a head vascular segmentation result. This vascular segmentation result can be used to distinguish the head vascular region image from the background image.

[0053] Therefore, the above-mentioned head blood vessel segmentation model is obtained by training based on training samples with blood vessel information of interest. The training method of the blood vessel segmentation model may include: based on experienced radiologists, marking the head blood vessels in the training sample images, and using this as the gold standard during training. The training sample images are then input into the head blood vessel segmentation model to obtain the head blood vessel segmentation results, and the loss between the head blood vessel segmentation results and the gold standard is calculated. The network parameters of the head segmentation model are adjusted based on the loss. When the loss is less than or equal to a preset threshold or convergence is reached, it indicates that the head blood vessel segmentation model training has converged. In this embodiment, the Dice loss function, the cross entropy loss function or other types of loss functions can be used when calculating the loss, and no specific limitation is made here; when adjusting the network parameters, a stochastic gradient descent SGD optimizer or other types of optimizers can be used.

[0054] Before the neck sub-image is input into the pre-set neck vessel segmentation model, the computer preprocesses the image. This includes resampling it to a medical image with a third voxel spacing, setting the window width and window level to crop the grayscale value range of the medical image, and normalizing the image to a value between -1 and 1 to produce a standardized image. The neck vessel segmentation model extracts and maps features from the medical image to produce a neck vessel segmentation result that distinguishes the neck vessel region from the background image.

[0055] Therefore, the above-mentioned neck vessel segmentation model is obtained by training based on training samples with information about the vessels of interest. The training method of the vessel segmentation model may include: based on experienced radiologists, marking the neck vessels in the training sample images, and using this as the gold standard during training. The training sample images are then input into the neck vessel segmentation model to obtain the neck vessel segmentation results, and the loss between the neck vessel segmentation results and the gold standard is calculated. The network parameters of the neck segmentation model are adjusted based on the loss. When the loss is less than or equal to a preset threshold or convergence is reached, it indicates that the neck vessel segmentation model training has converged. When calculating the loss, the Dice loss function, the cross entropy loss function or other types of loss functions can be used, and no specific limitations are made here; when adjusting the network parameters, the stochastic gradient descent SGD optimizer or other types of optimizers can be used.

[0056] Before the chest sub-image is input into the pre-set chest vessel segmentation model, the computer preprocesses the image. This includes resampling it to a medical image with a fourth voxel spacing, setting the window width and window level to crop the grayscale value range of the medical image, and normalizing the image to a value between -1 and 1 to produce a standardized image. The chest vessel segmentation model extracts and maps features from the medical image to produce a chest vessel segmentation result that distinguishes the chest vessel region from the background image.

[0057] Therefore, the chest vessel segmentation model is trained based on training samples containing information about the vessels of interest. Training of the vessel segmentation model can include labeling the chest vessels in training sample images by experienced radiologists, using this as a gold standard for training. The training sample images are then input into the chest vessel segmentation model to obtain chest vessel segmentation results, and the loss between the chest vessel segmentation results and the gold standard is calculated. The network parameters of the chest segmentation model are adjusted based on this loss. When the loss is less than or equal to a preset threshold or convergence is achieved, the chest vessel segmentation model training has converged. Loss calculations can use Dice loss functions, cross-entropy loss functions, or other types of loss functions, without specific limitations herein. Network parameter adjustments can use stochastic gradient descent (SGD) optimizers or other types of optimizers.

[0058] In this embodiment, since blood vessel segmentation is performed on multiple sub-medical images, the multiple blood vessel segmentation models can be three-dimensional models. The three-dimensional model can be a convolutional neural network model, a recurrent neural network model, or other deep learning models, such as 3D U-Net and 3D V-Net, which are not limited in this embodiment. The multiple blood vessel segmentation results of the multiple sub-medical images can be referred to. Figure 2 In the middle, one sub-medical image corresponds to one blood vessel segmentation result.

[0059] In this embodiment, considering the GPU memory size of the computer device, which may be insufficient when processing the entire sub-medical image, the computer device can sequentially capture image blocks of the same size (e.g., the image block of the size shown in screenshot 1803) from the upper left corner of the medical image. Each image block is input into the vascular segmentation model to obtain a vascular segmentation result for each image block. The vascular segmentation results for each image block are then concatenated in the order in which they were captured to form the vascular segmentation result for the entire sub-medical image.

[0060] In step S404 , the blood vessel segmentation results of the plurality of sub-medical images may be fused to obtain a blood vessel segmentation result of the medical image.

[0061] Specifically, the computer device uses a linear or nonlinear fusion method to fuse the blood vessel segmentation results of several sub-medical images to obtain the blood vessel segmentation result of the final medical image. Figure 2 Right side.

[0062] In the medical image vascular segmentation method provided in this embodiment, a computer device first inputs a medical image to be segmented into a preset slice classification model to obtain several sub-images of the medical image. The computer then inputs the sub-images to be segmented into a preset vascular segmentation model to obtain sub-vessel segmentation results for the sub-images. Finally, the sub-vessel segmentation results are fused to obtain a vascular segmentation result for the entire medical image. The slice classification model is trained using training samples containing information about regions of interest. The multiple vascular segmentation models are trained using training samples containing information about vessels of interest. The resulting slice classification model takes into account the characteristic that head and neck sub-images contain different vessel sizes from bottom to top, enabling a faster and more efficient vascular segmentation process. The resulting vascular segmentation models also take into account the similar morphology, trajectory, and radius of vessels in the same region from different patients. This allows the multiple vascular segmentation models to focus more on the vascular segmentation process at the corresponding region, enabling more accurate vascular segmentation.

[0063] Figure 5 A method for segmenting blood vessels in a head sub-medical image according to an embodiment of the present disclosure is shown. Figure 5 As shown, the method for segmenting blood vessels in a head sub-medical image includes

[0064] Step S501: Acquire a head sub-medical image to be segmented. Specifically, the head sub-medical image to be segmented is a cut-off sub-medical image output by a slice classification model, which generally contains vascular tissue, brain tissue, and skull tissue.

[0065] In this embodiment, the head sub-medical image to be segmented can be obtained by dividing the head CTA angiography image according to parts.

[0066] In step S502, the skull of the head sub-medical image can be segmented using a preset threshold range of bone tissue grayscale, and the head sub-medical image containing the skull area can be cropped; the cropped head sub-medical image can be subjected to the first head vascular segmentation model to obtain an initial vascular segmentation result.

[0067] Specifically, before the medical image is input into the preset first head blood vessel segmentation model, the computer device preprocesses the medical image, including resampling it to a medical image with a fifth voxel spacing, setting the window width and window level to crop the grayscale value range of the medical image, and normalizing the image to a range between -1 and 1 to obtain a standardized image. The first head blood vessel segmentation model extracts and maps features from the medical image to obtain a first head blood vessel segmentation result. This blood vessel segmentation result can be the result of distinguishing the head blood vessel region image from the background image.

[0068] Specifically, through extensive data verification, a pair of bone tissue threshold ranges [a, b] are determined for optimal skull segmentation. Grayscale values ​​less than grayscale threshold a are set to 0, grayscale values ​​greater than grayscale threshold b are set to 0, and grayscale values ​​between a and b are set to 1, resulting in a skull segmentation result. The skull segmentation calculations yield a bounding box [x1, y1, z1, x2, y2, z2] surrounding the skull. This bounding box is then used to crop the head medical image. This disclosure does not specifically limit the skull segmentation method.

[0069] In step S503, the centerline of the initial blood vessel segmentation result is extracted, an image block is intercepted at the end of the extracted centerline, and the second head blood vessel segmentation model is used to obtain the head blood vessel segmentation result at the end.

[0070] Specifically, centerlines are extracted from the initial vessel segmentation results. In this embodiment, a 3D skeletonization method may be used, although this embodiment is not limited thereto. The skeletonization results are analyzed to determine the vessel endpoint locations. A medical image is captured at the vessel endpoint locations and input into a pre-set second head vessel segmentation model to obtain the vessel segmentation results at the endpoints.

[0071] In this embodiment, before the medical image block is input into the preset second head blood vessel segmentation model, the computer device preprocesses the medical image block, including resampling it to a medical image with a sixth voxel interval, setting the window width and window position to crop the grayscale value range of the medical image, and normalizing the maximum and minimum values ​​of the image to between [-1, 1] to obtain a standardized image.

[0072] In step S504 , the initial blood vessel segmentation result and the head blood vessel segmentation result at the end may be fused to obtain a blood vessel segmentation result of the head sub-medical image.

[0073] Therefore, the above-mentioned first head blood vessel segmentation model is obtained by training based on training samples with blood vessel information of interest. The training method of the blood vessel segmentation model may include: based on experienced radiologists, marking the head blood vessels in the training sample images, and using this as the gold standard during training. The training sample images are then input into the head blood vessel segmentation model to obtain the head blood vessel segmentation results, and the loss between the head blood vessel segmentation results and the gold standard is calculated. The network parameters of the head segmentation model are adjusted based on the loss. When the loss is less than or equal to a preset threshold or convergence is reached, it indicates that the head blood vessel segmentation model training has converged. In this embodiment, the Dice loss function, the cross entropy loss function or other types of loss functions can be used when calculating the loss, and no specific limitation is made here; when adjusting the network parameters, a stochastic gradient descent SGD optimizer or other types of optimizers can be used.

[0074] S505 : Fusing the initial blood vessel segmentation result and the head blood vessel segmentation result at the end to obtain a blood vessel segmentation result of the head medical sub-image.

[0075] Figure 6 A method for segmenting blood vessels in a neck sub-medical image according to an embodiment of the present disclosure is shown. Figure 6 As shown, the method for performing vascular segmentation on a neck sub-medical image includes step S601, obtaining a neck sub-medical image to be segmented. Specifically, the neck sub-medical image to be segmented is a truncated sub-medical image output by a slice classification model, and the medical image generally contains vascular tissue and bone tissue.

[0076] In the present disclosure, the neck sub-medical image to be segmented may be from a neck CTA angiography image. This embodiment does not limit the source of the neck sub-medical image.

[0077] In step S602 , the neck sub-medical image may be input into a first neck blood vessel segmentation model to obtain an initial blood vessel segmentation result.

[0078] Specifically, before the medical image is input into the preset first neck vessel segmentation model, the computer device preprocesses the medical image, including resampling it to a medical image with a seventh voxel interval, setting the window width and window level to crop the grayscale value range of the medical image, and normalizing the image to a value between -1 and 1 to obtain a standardized image. The first neck vessel segmentation model extracts and maps features from the medical image to obtain a first neck vessel segmentation result. This vessel segmentation result can be the result of distinguishing the neck vessel region image from the background image.

[0079] In this embodiment, the neck vascular region can also exist in the form of multiple labels. For example, the left common carotid artery can be defined as label 1, the right common carotid artery as label 2, the brachiocephalic trunk as label 3, the left internal carotid artery as label 4, the right internal carotid artery as label 5, and so on.

[0080] In step S603 , the neck sub-medical image may be cropped using the initial blood vessel segmentation result and input into a second neck blood vessel segmentation model to obtain a refined blood vessel segmentation result.

[0081] Specifically, before the medical image is input into the preset second neck vessel segmentation model, the computer device preprocesses the medical image, including resampling it to a medical image with an eighth voxel spacing, setting the window width and window level to crop the grayscale value range of the medical image, and normalizing the image to a value between -1 and 1 to obtain a standardized image. The second neck vessel segmentation model extracts and maps features from the medical image to obtain a second neck vessel segmentation result. This vessel segmentation result can be the result of distinguishing the neck vessel region image from the background image.

[0082] In this embodiment, the neck vascular region can also exist in the form of multiple labels. For example, the left common carotid artery can be defined as label 1, the right common carotid artery as label 2, the brachiocephalic trunk as label 3, the left internal carotid artery as label 4, the right internal carotid artery as label 5, and so on.

[0083] S604 , performing centerline extraction on the refined blood vessel segmentation result, intercepting an image block at the end of the extracted centerline, and inputting the intercepted image block into the third neck vessel segmentation model to obtain a neck vessel segmentation result at the end.

[0084] Specifically, centerline extraction is performed on the refined vessel segmentation results. In this embodiment, a 3D skeletonization method may be used, although this embodiment is not limited thereto. The skeletonization results are analyzed to determine the vessel endpoint locations. A medical image is captured at the vessel endpoint locations and input into the pre-defined third neck vessel segmentation model to obtain the vessel segmentation results at the endpoints.

[0085] In this embodiment, before the medical image block is input into the preset third neck blood vessel segmentation model, the computer device preprocesses the medical image block, including resampling it to a medical image with a ninth voxel interval, setting the window width and window position to crop the grayscale value range of the medical image, and normalizing the maximum and minimum values ​​of the image to between [-1, 1] to obtain a standardized image.

[0086] In step S605 , the initial blood vessel segmentation result and the neck blood vessel segmentation result at the end may be fused to obtain a blood vessel segmentation result of the neck sub-medical image.

[0087] Figure 7 A method for segmenting blood vessels in a chest sub-medical image according to an embodiment of the present disclosure is shown. Figure 7 As shown, the method for performing blood vessel segmentation on a chest medical sub-image may include step 701, obtaining a chest medical sub-image to be segmented. Specifically, the chest medical sub-image to be segmented is a truncated sub-image output by a slice classification model, and the medical image generally contains blood vessel tissue, aortic arch tissue, bone tissue, lung parenchyma tissue, etc.

[0088] In this embodiment, the chest sub-medical image to be segmented may be from a chest CTA angiography image. The present disclosure does not limit the source of the chest sub-medical image.

[0089] In step 702 , the chest medical sub-image may be input into a first chest vessel segmentation model to obtain an initial vessel segmentation result.

[0090] Specifically, before the medical image is input into the preset first chest vessel segmentation model, the computer device preprocesses the medical image, including resampling it to a medical image with a tenth voxel interval, setting the window width and window level to crop the grayscale value range of the medical image, and normalizing the image to a range between -1 and 1 to obtain a standardized image. The first chest vessel segmentation model extracts and maps features from the medical image to obtain a first chest vessel segmentation result. This vessel segmentation result can be the result of distinguishing the chest vessel region image from the background image.

[0091] In this embodiment, the chest vascular region can also exist in the form of multiple labels, that is, other vascular tissues except the aortic arch tissue (generally the common carotid artery, subclavian artery, etc.) are defined as label 1, and the aortic arch tissue is defined as label 2.

[0092] In step 703 , the neck sub-medical image may be cropped using the initial blood vessel segmentation result and input into a second chest blood vessel segmentation model to obtain a blood vessel segmentation result of the chest sub-medical image.

[0093] Specifically, before the medical image is input into the preset second chest vessel segmentation model, the computer device preprocesses the medical image. This includes resampling the image to a medical image with an eleventh pixel interval, setting the window width and window level to crop the grayscale value range of the medical image, and normalizing the image to a value between -1 and 1 to obtain a standardized image. The second chest vessel segmentation model extracts and maps features from the medical image to obtain a second neck vessel segmentation result. This vessel segmentation result can be distinguished from the background image by distinguishing the neck vessel region image.

[0094] In this embodiment, the chest vascular region can also exist in the form of multiple labels, that is, other vascular tissues except the aortic arch tissue (generally the common carotid artery, subclavian artery, etc.) are defined as label 1, and the aortic arch tissue is defined as label 2.

[0095] Figure 8 FIG. 1 is a block diagram illustrating an exemplary apparatus for segmenting blood vessels in a medical image according to an embodiment of the present disclosure, as shown in FIG. Figure 8 As shown, a blood vessel segmentation device 800 includes an interface 807 and a processor 801. Interface 807 can be configured to receive a medical image containing single data or multiple regions for blood vessel segmentation, where the blood vessels in each region have different structural properties. Processor 801 can be configured to execute the method for blood vessel segmentation in a medical image according to various embodiments of the present disclosure.

[0096] Through this interface 807, the device for performing blood vessel segmentation on medical images can be connected to a network (not shown), such as, but not limited to, a local area network in a hospital or the Internet. However, the communication method implemented by interface 807 is not limited to the network and can include NFC, Bluetooth, WIFI, etc.; it can be a wired connection or a wireless connection. Taking the network as an example, interface 807 can connect the device for performing blood vessel segmentation on medical images to external devices such as an image acquisition device (not shown), a medical image database 808, and an image data storage device 809. The image acquisition device can be any type of imaging modality, such as, but not limited to, computed tomography (CT), digital subtraction angiography (DSA), magnetic resonance imaging (MRI), functional MRI, dynamic contrast-enhanced MRI, diffusion MRI, spiral CT, cone-beam computed tomography (CBCT), positron emission tomography (PET), single-photon emission computed tomography (SPECT), X-ray imaging, optical tomography, fluorescence imaging, ultrasound imaging, and radiotherapy portal imaging.

[0097] In some embodiments, the vessel segmentation device 800 can be a dedicated intelligent device or a general-purpose intelligent device. For example, the vessel segmentation device 800 can be a computer customized for image data acquisition and image data processing tasks, or a server located in the cloud. For example, the device 800 can be integrated into an image acquisition device.

[0098] The blood vessel segmentation apparatus 800 may include a processor 801 and a memory 804 , and may further include at least one of an input / output 802 and an image display 803 .

[0099] The processor 801 may be a processing device including one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), etc. More specifically, the processor 801 may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor that runs other instruction sets, or a processor that runs a combination of instruction sets. The processor 801 may also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), a system on a chip (SoC), etc. As will be understood by those skilled in the art, in some embodiments, the processor 801 may be a special-purpose processor rather than a general-purpose processor. The processor 801 may include one or more known processing devices, such as a Pentium™, Core™, Xeon™, or Itanium series microprocessor manufactured by Intel™, a Turion™, Athlon™, Sempron™, Opteron™, FX™, or Phenom™ series manufactured by AMD™, or various processors manufactured by Sun Microsystems. Processor 801 may also include a graphics processing unit, such as a graphics processing unit from GPU manufactured by Nvidia TM series, GMA manufactured by Intel TM, Iris TM series, or Radeon TM series manufactured by AMD TM. Processor 801 may also include an accelerated processing unit, such as the Desktop A-4 (6,6) series manufactured by AMD TM or the Xeon Phi TM series manufactured by Intel TM. The disclosed embodiments are not limited to any type of processor or processor circuit, and are otherwise configured to acquire a medical image comprising multiple regions for blood vessel segmentation, wherein the blood vessels in each region have different structural properties; divide the medical image into sub-medical images of each region according to the region; determine a corresponding blood vessel segmentation result based on the sub-medical images of each region using a corresponding blood vessel segmentation model for each region; fuse the blood vessel segmentation results of the sub-medical images of each region to obtain a blood vessel segmentation result for the medical image, or manipulate any other type of data consistent with the disclosed embodiments. In addition, the term "processor" or "image processor" may include more than one processor, such as a multi-core design or multiple processors, each having a multi-core design. The processor 801 may execute sequences of computer program instructions stored in the memory 804 to perform the various operations, processes, and methods disclosed herein.

[0100] The processor 801 can be communicatively coupled to the memory 804 and configured to execute computer-executable instructions stored therein. The memory 804 may include read-only memory (ROM), flash memory, random access memory (RAM), dynamic random access memory (DRAM) such as synchronous DRAM (SDRAM) or Rambus DRAM, static memory (e.g., flash memory, static random access memory), etc., on which computer-executable instructions are stored in any format. In some embodiments, the memory 804 may store computer-executable instructions for one or more image processing programs 805. The computer program instructions can be accessed by the processor 801, read from the ROM or any other suitable memory location, and loaded into the RAM for execution by the processor 801. For example, the memory 804 may store one or more software applications. The software applications stored in the memory 804 may include, for example, an operating system (not shown) for a conventional computer system and an operating system for a soft control device.

[0101] Furthermore, the memory 804 may store an entire software application or only a portion of the software application (e.g., the image processing program 805) that is executable by the processor 801. Furthermore, the memory 804 may store multiple software modules for implementing the various steps of the method for segmenting blood vessels in medical images consistent with the present disclosure or the process for training a slice classification model or a corresponding blood vessel segmentation model for each region.

[0102] In addition, the memory 804 may store data generated / buffered during the execution of the computer program, for example, medical image data 806, including medical images transmitted from an image acquisition device, a medical image database 808, an image data storage device 809, etc. In some embodiments, the medical image data 806 may include a medical image containing multiple regions for which blood vessel segmentation is to be performed, and the image processing program 805 will perform region division, blood vessel segmentation for each region, and fusion of the blood vessel segmentation results.

[0103] In some embodiments, an image data storage device 809 may be provided to exchange image data with a medical image database 808, and the memory 804 may communicate with the medical image database 808 to obtain medical images containing multiple regions for which vessel segmentation is to be performed. For example, the image data storage device 809 may reside in another medical image acquisition device (e.g., a CT scan performed on the patient). The patient's medical images may be transmitted and stored in the medical image database 808, and the vessel segmentation device 800 may retrieve medical images of a specific patient from the medical image database 808 and perform vessel segmentation on the specific patient's medical images.

[0104] In some embodiments, the memory 804 may communicate with the medical image database 808 to transmit and save the extracted geometric features together with the obtained blood vessel segmentation result of the medical image into the medical image database 808 .

[0105] Furthermore, the parameters of the trained slice classification model and / or the corresponding vessel segmentation model for each region can be stored in the medical image database 808, allowing them to be accessed, obtained, and utilized by other vessel segmentation devices when needed. In this manner, when presented with a patient, the processor 801 can retrieve the trained slice classification model and / or the corresponding vessel segmentation model for each region of the patient population, and perform vessel segmentation based on the retrieved trained model.

[0106] In some embodiments, the slice classification model and / or the corresponding blood vessel segmentation model for each region (particularly a learning network) can be stored in the memory 804. Alternatively, the learning network can be stored in a remote device, a separate database (such as the medical image database 808), or a distributed device and can be used by the image processing program 805.

[0107] In addition to displaying medical images, the image display 803 can also display other information, such as the corresponding blood vessel segmentation results determined by using the corresponding blood vessel segmentation models for each region, and the blood vessel segmentation results of the medical image obtained by fusing the blood vessel segmentation results of the sub-medical images for each region. For example, the image display 803 can be an LCD, CRT, or LED display.

[0108] The input / output 802 can be configured to allow the vessel segmentation device 800 to receive and / or send data. The input / output 802 may include one or more digital and / or analog communication devices that allow the device to communicate with a user or other machines and devices. For example, the input / output 802 may include a keyboard and mouse that allow a user to provide input.

[0109] In some embodiments, the image display 803 may present a user interface so that the user can conveniently and intuitively modify (such as edit, move, modify, etc.) the generated anatomical labels using the input / output 802 together with the user interface.

[0110] The interface 807 may include a network adapter, a cable connector, a serial connector, a USB connector, a parallel connector, a high-speed data transmission adapter such as an optical fiber, USB 6.0, Lightning, a wireless network adapter such as a Wi-Fi adapter, a telecommunications (6G, 4G / LTE, etc.) adapter. The device may be connected to a network through the interface 807. The network may include a local area network (LAN), a wireless network, a cloud computing environment (e.g., software as a service, platform as a service, infrastructure as a service, etc.), a client-server, a wide area network (WAN), etc.

[0111] Various operations or functions are described herein that may be implemented as or defined as software code or instructions. Such content may be source code or differential code ("delta" or "patch" code) that can be directly executed ("object" or "executable" form). Software codes or instructions may be stored in a computer-readable storage medium and, when executed, may cause a machine to perform the functions or operations described, and include any mechanism for storing information in a form accessible to a machine (e.g., a computing device, an electronic system, etc.), such as recordable or non-recordable media (e.g., read-only memory (ROM), random access memory (RAM), magnetic disk storage media, optical storage media, flash memory devices, etc.).

[0112] The foregoing description has been presented for purposes of illustration. It is not exhaustive and is not intended to limit the invention to the precise forms or embodiments disclosed. Modifications and adaptations of the disclosed embodiments will be apparent from consideration of the specification and practice of the disclosed embodiments.

[0113] The exemplary methods described herein may be at least partially machine or computer-implemented. Some examples may include a computer-readable medium or machine-readable medium encoded with instructions operable to configure an electronic device to perform the methods described in the above examples. Implementations of such methods may include software code, such as microcode, assembly language code, higher-level language code, and the like. Various programs or program modules may be created using various software programming techniques. For example, program segments or program modules may be designed using Java, Python, C, C++, assembly language, or any known programming language. One or more such software components or modules may be integrated into a computer system and / or computer-readable medium. Such software code may include computer-readable instructions for performing the various methods. The software code may form part of a computer program product or computer program module. In addition, in one example, the software code may be tangibly stored on one or more volatile, non-transitory, or non-volatile tangible computer-readable media, such as during execution or at other times. Examples of these tangible computer-readable media may include, but are not limited to, hard disks, removable disks, removable optical disks (e.g., optical disks and digital video disks), magnetic tape cassettes, memory cards or sticks, random access memory (RAM), read-only memory (ROM), and the like.

[0114] Furthermore, although illustrative embodiments are described herein, the scope includes any and all embodiments having equivalent elements, modifications, omissions, combinations (e.g., combinations of schemes across various embodiments), adjustments, or variations based on the present disclosure. The elements of the claims are to be interpreted broadly based on the language used in the claims and are not limited to the examples described in this specification or during the continuation of this application, which examples are to be construed as non-exclusive. Furthermore, the steps of the disclosed methods may be modified in any manner, including by reordering steps or inserting or deleting steps. Therefore, it is intended that the description be considered merely as examples, with the true scope being indicated by the following claims and their full scope of equivalents.

[0115] The above description is intended to be illustrative and not restrictive. For example, the above examples (or one or more schemes thereof) can be used in combination with each other. Other embodiments may be used by those of ordinary skill in the art when viewing the above description. Moreover, in the above detailed description, various features can be combined together to simplify the present disclosure. This should not be interpreted as intending that disclosed features that are not claimed for protection are essential to any claim. Therefore, the following claims are hereby incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently a separate embodiment, and it is conceivable that these embodiments can be combined with each other in various combinations or permutations. The scope of the present invention should be determined with reference to the appended claims and the full scope of equivalents to which these claims are assigned.

Claims

1. A method for segmenting blood vessels in a medical image, characterized in that: include: Obtain a medical image containing several regions for vascular segmentation, where the blood vessels in each region have different structural properties; Using a processor, dividing the medical image into sub-medical images of each part according to the part; performing a coarse segmentation on the sub-medical image of each part to obtain a first image block of a target area containing a blood vessel to be segmented; determining a first blood vessel segmentation result based on the first image block of the target area using a first blood vessel segmentation model as the blood vessel segmentation result of the sub-medical image of each part; and For at least one of the parts: performing centerline extraction on the determined first blood vessel segmentation result, intercepting a second image block at an end of the extracted centerline; and determining a second blood vessel segmentation result at the end based on the second image block at the end using a second blood vessel segmentation model, wherein a part is selected as the at least one part from among the multiple parts, wherein the segmented blood vessel at the end intersecting with the adjacent part satisfies at least one of the following conditions: a diameter of the blood vessel at the end intersecting with the adjacent part is less than a first predetermined threshold; and a curvature of the blood vessel at the end intersecting with the adjacent part is greater than a second predetermined threshold. fusing the first blood vessel segmentation result and the second blood vessel segmentation result to obtain a blood vessel segmentation result of the sub-medical image of the at least one part; The processor is used to fuse the blood vessel segmentation results of the sub-medical images of various parts to obtain the blood vessel segmentation result of the medical image.

2. The method according to claim 1, characterized in that The structural properties of the blood vessels include at least one of morphology and size.

3. The method according to claim 1, characterized in that The medical image is a 3D medical image, and dividing the medical image into sub-medical images of each part according to the parts specifically includes: based on the 3D medical image, using a slice classification model, identifying key slices in the 3D medical image that are the junctions of adjacent parts; using the identified key slices to realize the division of the sub-medical images according to the parts.

4. The method according to claim 3, characterized in that The slice classification model is implemented using a two-dimensional learning network and is trained using training samples with classification information of slices at corresponding locations.

5. The method according to any one of claims 1 to 4, characterized in that Before the processor performs various processing on the medical image or the sub-medical image, the medical image or the sub-medical image is subjected to a standardization process.

6. The method according to claim 5, characterized in that The standardization processing includes resampling the medical image or the sub-medical image to obtain a medical image or sub-medical image with corresponding voxel spacing, clipping the grayscale value of the resampled medical image or sub-medical image with a set window width and window position, and normalizing the medical image.

7. The method according to any one of claims 1 to 4, characterized in that Each blood vessel segmentation model is trained using training samples having classification information of blood vessels at corresponding locations.

8. The method according to claim 7, characterized in that The classification information of the blood vessels at the corresponding part includes a plurality of labels.

9. The method according to claim 1, characterized in that The at least one part is each part of the several parts, or is a part of interest preset by a doctor among the several parts.

10. The method according to claim 1, characterized in that The method further includes selecting, from the plurality of parts, a part where a segmented blood vessel at the junction with an adjacent part is likely to be broken as the at least one part.

11. The method according to claim 1, characterized in that Also includes: A portion where the segmented blood vessel at the end intersecting the adjacent portion satisfies the following condition is selected from the plurality of portions as the at least one portion: the contrast between the blood vessel at the end intersecting the adjacent portion and the surrounding area is less than a third predetermined threshold.

12. The method according to any one of claims 1 to 4, characterized in that The multiple parts are at least one of different body parts, different organs, and different organ parts in the same organ.

13. The method according to any one of claims 1 to 4, characterized in that The medical image includes at least one of a head and neck CTA image, a head and neck MRA image, and a lung CT image.

14. The method according to claim 13, characterized in that In the case where the medical image is a head and neck CTA image or a head and neck MRA image, the various parts include the head, neck and chest, Based on the sub-medical images of each part, the corresponding vascular segmentation model of each part is used to determine the corresponding vascular segmentation results, specifically including: Segmenting the skull of the head sub-medical image using a preset threshold range of bone tissue grayscale, and cropping the head sub-medical image containing the skull region; obtaining an initial blood vessel segmentation result using the first head blood vessel segmentation model on the cropped head sub-medical image; Performing centerline extraction on the initial blood vessel segmentation result, intercepting an image block at the end of the extracted centerline, and using the second head blood vessel segmentation model to obtain the head blood vessel segmentation result at the end; The initial blood vessel segmentation result and the head blood vessel segmentation result at the end are fused to obtain a blood vessel segmentation result of the head sub-medical image.

15. The method according to claim 14, characterized in that Based on the sub-medical images of each part, the corresponding vascular segmentation model of each part is used to determine the corresponding vascular segmentation results, specifically including: Inputting the neck sub-medical image into the first neck vascular segmentation model to obtain an initial vascular segmentation result; cropping the neck sub-medical image using the initial vascular segmentation result and inputting it into the second neck vascular segmentation model to obtain a refined vascular segmentation result; Performing centerline extraction on the refined blood vessel segmentation result, intercepting an image block at the end of the extracted centerline, and inputting the intercepted image block into the third neck vessel segmentation model to obtain a neck vessel segmentation result at the end; The initial blood vessel segmentation result and the neck blood vessel segmentation result at the end are fused to obtain a blood vessel segmentation result of the neck sub-medical image.

16. The method according to claim 14, characterized in that Based on the sub-medical images of each part, the corresponding vascular segmentation model of each part is used to determine the corresponding vascular segmentation results, specifically including: The chest sub-medical image is input into the first chest vascular segmentation model to obtain an initial vascular segmentation result. The neck sub-medical image is cropped using the initial vascular segmentation result and input into the second chest vascular segmentation model to obtain a vascular segmentation result of the chest sub-medical image.

17. A device for segmenting blood vessels in medical images, characterized in that: include: The interface is configured to: receive a medical image containing a plurality of parts for which blood vessels are to be segmented, wherein the blood vessels in each part have different structural properties; as well as A processor configured to execute the method for segmenting blood vessels in a medical image according to any one of claims 1 to 16.

18. A non-transitory computer-readable medium having instructions stored thereon, the instructions, when executed by a processor, implementing the method for segmenting blood vessels in a medical image according to any one of claims 1 to 16.

Images