Blood vessel segmentation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses a blood vessel segmentation method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining a target medical image to be subjected to blood vessel segmentation and a trained target segmentation model used for blood vessel segmentation; based on the target medical image and the target segmentation model, determining a blood vessel segmentation result of the target medical image; wherein the target segmentation model comprises a coarse segmentation network, a feature aggregation network and a fine segmentation network, and the feature aggregation network is used for obtaining a target feature image based on the target medical image and a coarse feature image output by the coarse segmentation network, and inputting the target feature image into the fine segmentation network. According to the technical scheme of the embodiment of the invention, the accuracy of blood vessel segmentation is improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the technical field of image processing, and in particular, to a method, apparatus, electronic device, and storage medium for blood vessel segmentation. Background Art

[0002] In medicine, accurately segmenting blood vessels is of crucial significance for assisting doctors in diagnosing many cardiovascular diseases such as aneurysms, aortic dissections, and arteriosclerosis.

[0003] However, the accuracy of the currently adopted blood vessel segmentation method is not high and urgently needs to be solved. Summary of the Invention

[0004] Embodiments of the present invention provide a method, apparatus, electronic device, and storage medium for blood vessel segmentation, which improve the accuracy of blood vessel segmentation.

[0005] According to one aspect of the present invention, a method for blood vessel segmentation is provided, which may include:

[0006] Obtaining a target medical image to be segmented for blood vessels, and a trained target segmentation model for blood vessel segmentation;

[0007] Based on the target medical image and the target segmentation model, determining the blood vessel segmentation result of the target medical image;

[0008] Wherein, the target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network. The feature aggregation network is used to obtain a target feature image based on the target medical image and the coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network.

[0009] According to another aspect of the present invention, a device for blood vessel segmentation is provided, which may include:

[0010] A target segmentation model acquisition module, configured to obtain a target medical image to be segmented for blood vessels, and a trained target segmentation model for blood vessel segmentation;

[0011] A blood vessel segmentation result determination module, configured to determine the blood vessel segmentation result of the target medical image based on the target medical image and the target segmentation model;

[0012] Wherein, the target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network. The feature aggregation network is used to obtain a target feature image based on the target medical image and the coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network.

[0013] According to another aspect of the present invention, an electronic device is provided, which may include:

[0014] At least one processor; and

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

[0016] the memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, the at least one processor is caused to implement the vascular segmentation method provided in any embodiment of the present invention when executed.

[0017] According to another aspect of the present invention, there is provided a computer-readable storage medium having stored thereon computer instructions for causing a processor to implement the vascular segmentation method provided in any embodiment of the present invention when executed.

[0018] In the technical solution of the embodiment of the present invention, a target medical image to be subjected to vascular segmentation and a trained target segmentation model for vascular segmentation are obtained; based on the target medical image and the target segmentation model, a vascular segmentation result of the target medical image is determined; wherein the target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network, and the feature aggregation network is configured to obtain a target feature image based on the target medical image and a coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network. In the above technical solution, by using the feature aggregation network that can perform feature aggregation on the target medical image and the coarse feature image, the obtained vascular segmentation result can fully consider global information, avoid false positives or omissions in the vascular segmentation result, and thus improve the accuracy of vascular segmentation.

[0019] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0021] Figure 1 is a flowchart of a vascular segmentation method provided according to an embodiment of the present invention;

[0022] Figure 2 is a flowchart of another vascular segmentation method provided according to an embodiment of the present invention;

[0023] Figure 3It is a flowchart of yet another blood vessel segmentation method provided according to an embodiment of the present invention;

[0024] Figure 4 It is a flowchart of still another blood vessel segmentation method provided according to an embodiment of the present invention;

[0025] Figure 5 It is a structural diagram of an original segmentation model in still another blood vessel segmentation method provided according to an embodiment of the present invention;

[0026] Figure 6 It is a structural block diagram of a blood vessel segmentation device provided according to an embodiment of the present invention;

[0027] Figure 7 It is a schematic structural diagram of an electronic device for implementing the blood vessel segmentation method of the embodiment of the present invention. Specific embodiments

[0028] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order different from those illustrated or described here. The same is true for cases such as "target" and "original", which will not be elaborated here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0030] Figure 1 It is a flowchart of a blood vessel segmentation method provided in an embodiment of the present invention. This embodiment is applicable to the situation of blood vessel segmentation. This method can be executed by the blood vessel segmentation device provided in the embodiment of the present invention. The device can be implemented in software and / or hardware, and the device can be integrated on an electronic device, which can be various user terminals or servers.

[0031] See Figure 1, the method of the embodiment of the present invention specifically includes the following steps:

[0032] S110. Obtain a target medical image to be subjected to vascular segmentation and a trained target segmentation model for vascular segmentation. The target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network. The feature aggregation network is configured to obtain a target feature image based on the target medical image and the coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network.

[0033] Among them, the target medical image is a medical image to be subjected to vascular segmentation; the target medical image can be a multi-dimensional medical image such as two-dimensional, three-dimensional, or four-dimensional. For example, the target medical image can be a three-dimensional medical image; the target medical image can be a medical image obtained by Computed Tomography (CT), a medical image obtained by Computed Tomography Angiography (CTA) using computer tomography technology and a specific contrast agent to display the image inside the blood vessels, or a medical image obtained by Magnetic Resonance (MR); in the embodiment of the present invention, the source and type of the target medical image are not specifically limited. The coarse feature image is the feature image output by the coarse segmentation network for the target medical image. The target feature image is the feature image obtained by the feature aggregation network after feature aggregation of the target medical image and the coarse feature image.

[0034] It can be understood that the target segmentation model is a segmentation network for vascular segmentation. Essentially, the segmentation network classifies each pixel point in the image and distinguishes the part of interest and the background area.

[0035] In the embodiment of the present invention, the target medical image can be a medical image obtained by imaging an organ, tissue, or region such as the heart, brain, chest, abdomen, or head. Correspondingly, the blood vessels to be subjected to vascular segmentation can be blood vessels in organs, tissues, or regions such as the heart, brain, chest, abdomen, or head. For example, it can be the aorta in the heart. In the embodiment of the present invention, the object of imaging of the target medical image and the types of blood vessels to be subjected to vascular segmentation are not specifically limited.

[0036] In an embodiment of the present invention, a target segmentation model capable of performing vascular segmentation can be pre-trained; the target segmentation model includes a coarse segmentation network capable of coarsely segmenting blood vessels in the target medical image to locate the blood vessels, a feature aggregation network capable of aggregating features in the target medical image, and a fine segmentation network capable of finely segmenting blood vessels in the target medical image to further improve the morphological outline of the blood vessels on the basis of the coarse segmentation; the coarse segmentation network and the fine segmentation network can adopt networks capable of performing vascular segmentation, and the structures of the coarse segmentation network and the fine segmentation network can be the same or different. For example, the coarse segmentation network and the fine segmentation network can adopt the Unet network with the same structure. Therefore, the target segmentation model can be named as the CF-Unet (Coarse-to-Fine Unet) network; the feature aggregation network can be used to obtain a target feature image based on the target medical image and the coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network.

[0037] In an embodiment of the present invention, the feature aggregation network may, for example, include a multiplication module for multiplying the labeled medical image and the coarse feature image to obtain a target feature image.

[0038] It should be noted that the coarse feature image output by the coarse segmentation network is not necessarily the coarse segmentation result output by the coarse segmentation network, but the last feature information output by the coarse segmentation network, that is, the coarse feature image can be the output result of the last module in the coarse segmentation network that can output feature information; the module for outputting the coarse segmentation result of the coarse segmentation network can only participate in the training process of the target segmentation model.

[0039] Exemplarily, since the aorta is the largest artery in the human body, starting from the left ventricle of the heart and supplying oxygen and nutrients to almost all organs and tissues of the body, accurate segmentation of the aorta is of crucial significance for diagnosing many cardiovascular diseases, such as aortic aneurysm, aortic dissection, aortic sclerosis, etc. However, the density difference between the aorta and other tissues or organs is small, resulting in it being difficult to obtain satisfactory results by direct threshold segmentation. Moreover, the branched blood vessels of the aorta are numerous and widely distributed, and some small branched blood vessels are easily overlooked or mis-sected. Also, due to the influence of physiological factors such as breathing and heartbeat, the medical images obtained by image acquisition for the aorta may have noise or motion artifacts. While CTA images can provide very clear blood vessel images, especially for the display of the aorta and its branches. Therefore, a CTA image to be subjected to aortic blood vessel segmentation can be obtained as the target medical image, and a pre-trained target segmentation model for performing vascular segmentation.

[0040] S120. Determine the vascular segmentation result of the target medical image based on the target medical image and the target segmentation model.

[0041] In the embodiments of the present invention, the target medical image can be directly input into the target segmentation model, and according to the output result of the target segmentation model, the blood vessel segmentation result of the target medical image can be obtained; alternatively, the target medical image can be processed, and the processed result is input into the target segmentation model, and according to the output result of the target segmentation model, the blood vessel segmentation result of the target medical image can be obtained.

[0042] Exemplarily, the size of the target medical image is 2N*2N*2N. The target medical image can be resampled to a size of N*N*N, and the resampled target medical image is input into the coarse segmentation network in the target segmentation model to obtain a coarse feature image; the target medical image with a size of 2N*2N*2N that has not been resampled can be cropped to obtain the middle region block of the target medical image with a size of N*N*N; the middle region block and the coarse feature image are input into the feature aggregation network; the feature aggregation network can include a feature cropping module, a resampling module, and a multiplication module. The coarse feature image can be input into the cropping module, and the cropping module can crop the coarse feature image to obtain the central region block of the coarse feature image with a size of (N / 2)*(N / 2)*(N / 2); the central region block is input into the resampling module to resample to obtain a central region block with a size of N*N*N; the central region block with a size of N*N*N and the middle region block are input into the multiplication module, and the multiplication module can perform a multiplication operation on the central region block with a size of N*N*N and the middle region block to obtain a target feature image; the target feature image is input into the fine segmentation network to obtain the blood vessel segmentation result of the target medical image.

[0043] In the embodiments of the present invention, before determining the blood vessel segmentation result of the target medical image based on the target medical image and the target segmentation model, the target medical image can also be subjected to normalization processing such as maximum-minimum normalization through the formula and the target medical image is updated according to the processing result of the normalization processing, thereby enhancing the generalization ability of the target segmentation model, so that the target segmentation model can function properly in different medical devices. Here, x refers to the target medical image.

[0044] In the embodiments of the present invention, a target segmentation model including a coarse segmentation network, a feature aggregation network, and a fine segmentation network is used for blood vessel segmentation, which can solve the problems existing in blood vessel segmentation methods including techniques such as threshold processing, edge detection, and morphological operations. When segmenting the aorta in CTA images, the performance is limited by image noise, artifacts, and density differences between the aorta and other tissues. It can also solve the problem that methods based on region growing and level sets require manual selection of seed points or initialization of contours, which are complex and time-consuming operations. It can also solve the problem that the convolutional neural network structure based on the encoder-decoder structure of Unet, which is widely used in medical image segmentation, combines a classification loss function to classify each pixel point affected to obtain the final segmentation result. For high-resolution CTA images, due to hardware limitations, it is often difficult to consider global information and resolution simultaneously, resulting in false positives or missing in the segmentation result. In particular, it can solve the problem that in the aorta blood vessel segmentation task, when the entire CTA image is downsampled to a specified size, some small branch blood vessels in the image cannot be segmented. In the case of using a general slider strategy, a region of a specified size is cropped around the blood vessel, and in this way, the model lacks global information and is prone to false positives and blood vessel missing. The technical solution of the embodiments of the present invention can obtain global information through the coarse segmentation network with a coarse resolution, then obtain detailed information through the fine segmentation network with a fine resolution, and merge the two, thereby solving the problem of limited resolution and receptive field existing in blood vessel segmentation, making the target segmentation model more robust in segmentation, and thus achieving both global information and resolution while ensuring the resolution, thereby improving the accuracy of blood vessel segmentation.

[0045] In the embodiments of the present invention, a segmentation threshold can be preset, and the blood vessel segmentation result is binarized according to the segmentation threshold to obtain a binarized segmentation result. The scattered false positives in the binarized segmentation result are removed using connected components, and the blood vessel segmentation result is updated according to the removal result.

[0046] The technical solution of the embodiments of the present invention obtains a target medical image to be segmented for blood vessels, and a trained target segmentation model for blood vessel segmentation; based on the target medical image and the target segmentation model, the blood vessel segmentation result of the target medical image is determined; wherein, the target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network, and the feature aggregation network is used to obtain a target feature image based on the target medical image and the coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network. The above technical solution can make the obtained blood vessel segmentation result fully consider global information through the feature aggregation network that can perform feature aggregation on the target medical image and the coarse feature image, avoid false positives or missing in the blood vessel segmentation result, and thus improve the accuracy of blood vessel segmentation.

[0047] Figure 2 It is a flowchart of another method for blood vessel segmentation provided in an embodiment of the present invention. This embodiment is optimized based on the above technical solutions. In this embodiment, optionally, based on a target medical image and a target segmentation model, a blood vessel segmentation result of the target medical image is determined, including: performing a block processing on the target medical image to obtain at least one pixel block; for each pixel block among the at least one pixel block, performing a block processing on the pixel block to obtain a first intermediate block; inputting the pixel block and the first intermediate block into the target segmentation model to obtain a pixel output result output by the target segmentation model for the pixel block; and obtaining a blood vessel segmentation result of the target medical image according to the pixel output results respectively corresponding to the at least one pixel block. Among them, the explanations of the same or corresponding terms as those in the above embodiments are not elaborated herein.

[0048] See Figure 2 , the method of this embodiment may specifically include the following steps:

[0049] S210. Obtain a target medical image to be segmented for blood vessels and a trained target segmentation model for blood vessel segmentation, where the target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network, and the feature aggregation network is used to obtain a target feature image based on the target medical image and a coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network.

[0050] S220. Perform a block processing on the target medical image to obtain at least one pixel block.

[0051] Among them, the pixel block is an image block obtained by performing a block processing on the target medical image.

[0052] It should be noted that there may be overlapping regions among the at least one pixel block. For example, the entire target medical image can be traversed along three axial positions of the target medical image with a step size of N to obtain at least one pixel block with a size of 2N*2N*2N until every pixel point in the target medical image is traversed, that is, the traversal step size is N*N*N. However, since the size of the at least one pixel block obtained by block processing is 2N*2N*2N, there are overlapping regions among the at least one pixel block. The above traversal of the entire target medical image until every pixel point in the target medical image is traversed can be that, for each pixel block among the at least one pixel block, a block processing is performed on the pixel block to obtain a first intermediate block, so that the at least one first intermediate block corresponding to the at least one pixel block can include every pixel point in the target medical image, so as to ensure that there is no data loss problem in the determined blood vessel segmentation result of the target medical image.

[0053] S230. For each of at least one pixel block, perform a block processing on the pixel block to obtain a first intermediate block.

[0054] Wherein, the first intermediate block is an image block of the intermediate region of the pixel block obtained by performing a block processing on the pixel block.

[0055] In an embodiment of the present invention, the side length of the first intermediate block can be half of the side length of the pixel block. For example, if the size of the pixel block is 2N * 2N * 2N, the size of the first intermediate block is N * N * N.

[0056] S240. Input the pixel block and the first intermediate block into a target segmentation model to obtain a pixel output result output by the target segmentation model for the pixel block.

[0057] Wherein, the pixel output result is the output result obtained by inputting the pixel block and the first intermediate block into the target segmentation model.

[0058] S250. Obtain a blood vessel segmentation result of the target medical image according to the pixel output results respectively corresponding to at least one pixel block.

[0059] In an embodiment of the present invention, the pixel output results respectively corresponding to at least one pixel block can be integrated according to the positions of the pixel block and / or the first intermediate block corresponding in the target medical image to obtain a blood vessel segmentation result of the target medical image. For example, for each of at least one pixel block, if the size of the first intermediate block is N * N * N and the size of the obtained pixel output result is also N * N * N, the relative position of the position corresponding to the first intermediate block in the region where the blood vessel segmentation result of the target medical image is placed can be determined according to the position of the first intermediate block in the target medical image, and the pixel output result is placed at this relative position, thereby obtaining a blood vessel segmentation result of the target medical image.

[0060] The technical solution of the embodiment of the present invention obtains at least one pixel block by performing a block processing on the target medical image; for each of at least one pixel block, perform a block processing on the pixel block to obtain a first intermediate block; input the pixel block and the first intermediate block into a target segmentation model to obtain a pixel output result output by the target segmentation model for the pixel block; obtain a blood vessel segmentation result of the target medical image according to the pixel output results respectively corresponding to at least one pixel block. The above technical solution can perform a block processing on the target medical image and input the pixel block and the first intermediate block into the target segmentation model, so that the blood vessel segmentation result corresponding to the first intermediate block region in the target medical image not only considers the local information of the first intermediate block region, but also can consider the information in a larger range of pixel blocks, thereby making the obtained blood vessel segmentation result further fully consider the global information, and further improving the accuracy of blood vessel segmentation.

[0061] An optional technical solution is to input a pixel block and a first intermediate block into a target segmentation model to obtain a pixel output result output by the target segmentation model for the pixel block, including: inputting the pixel block into a coarse segmentation network to obtain a coarse feature image; inputting the coarse feature image and the first intermediate block into a feature aggregation network to obtain a target feature image; inputting the target feature image into a fine segmentation network to obtain a first output result output by the fine segmentation network for the pixel block, and using the first output result as the pixel output result output by the target segmentation model for the pixel block.

[0062] Wherein, the first output result is the output result obtained by inputting the target feature image into the fine segmentation network and output by the fine segmentation network for the pixel block.

[0063] It should be noted that the coarse feature image in the embodiment of the present invention is the feature image output by the coarse segmentation network for the pixel block, and the target feature image is the feature image obtained by the feature aggregation network after aggregating the coarse feature image and the first intermediate block.

[0064] Exemplarily, the size of the pixel block is 2N*2N*2N. The pixel block can be resampled to a size of N*N*N, and the resampled pixel block is input into the coarse segmentation network in the target segmentation model to obtain a coarse feature image; the pixel block with a size of 2N*2N*2N without resampling can be cropped to obtain a first intermediate block with a size of N*N*N; the first intermediate block and the coarse feature image are input into the feature aggregation network; the feature aggregation network can include a feature cropping module, a resampling module, and a multiplication module. The coarse feature image can be input into the cropping module, and the cropping module can crop the coarse feature image to obtain a central region block of the coarse feature image with a size of (N / 2)*(N / 2)*(N / 2); the central region block is input into the resampling module to resample to obtain a central region block with a size of N*N*N; the central region block with a size of N*N*N and the first intermediate block are input into the multiplication module, and the multiplication module can perform a multiplication operation on the central region block with a size of N*N*N and the first intermediate block to obtain a target feature image; the target feature image is input into the fine segmentation network to obtain the blood vessel segmentation result of the target medical image.

[0065] In the embodiment of the present invention, by inputting the pixel block into the coarse segmentation network to obtain a coarse feature image; inputting the coarse feature image and the first intermediate block into the feature aggregation network to obtain a target feature image; inputting the target feature image into the fine segmentation network to obtain a first output result output by the fine segmentation network for the pixel block, and using the first output result as the pixel output result output by the target segmentation model for the pixel block. Through the above solution, feature aggregation is achieved by inputting the coarse feature image and the first intermediate block into the feature aggregation network, thereby further improving the accuracy of blood vessel segmentation.

[0066] Figure 3 It is a flowchart of another blood vessel segmentation method provided in an embodiment of the present invention. This embodiment is optimized based on the above technical solutions. In this embodiment, optionally, the target segmentation model is pre-trained through the following steps: obtaining a pre-built original segmentation model, sample medical images, and blood vessel segmentation labels of the sample medical images, and using the sample medical images and the blood vessel segmentation labels as a set of training samples; training the original segmentation model based on multiple sets of training samples to obtain the target segmentation model. Among them, the explanations of the same or corresponding terms as those in the above embodiments are not repeated here.

[0067] See Figure 3 , the method of this embodiment may specifically include the following steps:

[0068] S310. Obtain a pre-built original segmentation model, sample medical images, and blood vessel segmentation labels of the sample medical images, and use the sample medical images and the blood vessel segmentation labels as a set of training samples.

[0069] Among them, the sample medical images are pre-determined sample images including blood vessels for training to obtain the target segmentation model. The blood vessel segmentation labels are the labels of the blood vessel segmentation results obtained after segmenting the sample medical images; the staff can pre-annotate according to the sample medical images to obtain the blood vessel segmentation labels. The training samples can be understood as the sample data used to train the target segmentation model, and a set of training samples can include a sample medical image and its corresponding blood vessel segmentation label. The original segmentation model can be understood as a model that can be used for blood vessel segmentation and is to be trained.

[0070] In the embodiment of the present invention, before using the sample medical images and the blood vessel segmentation labels as a set of training samples, the sample medical images can also be subjected to normalization processing such as maximum-minimum normalization, and the sample medical images can be updated according to the processing results of the normalization processing.

[0071] S320. Train the original segmentation model based on multiple sets of training samples to obtain the target segmentation model.

[0072] In the embodiments of the present invention, loss calculation can be performed through multiple sets of training samples and loss functions such as the BCE (Binary Cross Entropy) loss function and / or the Dice (Dice coefficient) loss function, etc., to adjust the parameters in the original segmentation model and obtain the target segmentation model. In the embodiments of the present invention, there are no specific limitations on the manner of training the original segmentation model based on multiple sets of training samples to obtain the target segmentation model and the loss function used. It can be understood that a loss function is a function that maps the value of a random event or its related random variable to a non-negative real number to represent the "risk" or "loss" of the random event. In applications, the loss function is usually associated with the learning criterion and the optimization problem, that is, the model is solved and evaluated by minimizing the loss function.

[0073] In the embodiments of the present invention, the original segmentation model can be trained based on multiple sets of training samples to obtain the target segmentation model; or the original segmentation model can be trained based on some of the training samples in multiple sets of training samples to obtain the target segmentation model, and then the remaining training samples in multiple sets of training samples can be used to evaluate the obtained target segmentation model.

[0074] In the embodiments of the present invention, the model obtained by training the original segmentation model based on multiple sets of training samples can be used as the target segmentation model; or the original segmentation model can be trained based on multiple sets of training samples to obtain an intermediate segmentation model, and then the intermediate segmentation model can be adjusted to obtain the target segmentation model. For example, considering that the rough feature image output by the rough segmentation network is not necessarily the rough segmentation result output by the rough segmentation network, but the last feature information output by the rough segmentation network, the convolutional layer and other modules used to output the rough segmentation result of the rough segmentation network can only participate in the training process of the target segmentation model. After the training of the original segmentation model is completed to obtain the intermediate segmentation model, the convolutional layer and other modules used to output the rough segmentation result of the rough segmentation network are deleted to obtain the target segmentation model.

[0075] S330. Obtain a target medical image to be subjected to blood vessel segmentation, and a trained target segmentation model for blood vessel segmentation, where the target segmentation model includes a rough segmentation network, a feature aggregation network, and a fine segmentation network, and the feature aggregation network is used to obtain a target feature image based on the target medical image and the rough feature image output by the rough segmentation network, and input the target feature image into the fine segmentation network.

[0076] S340. Determine the blood vessel segmentation result of the target medical image based on the target medical image and the target segmentation model.

[0077] In the technical solution of the embodiment of the present invention, the target segmentation model is pre-trained through the following steps: obtaining a pre-built original segmentation model, a sample medical image, and a vascular segmentation label of the sample medical image, and using the sample medical image and the vascular segmentation label as a set of training samples; training the original segmentation model based on multiple sets of training samples to obtain the target segmentation model. In the above technical solution, the target segmentation model is trained by using training samples to obtain the target segmentation model, which can improve the accuracy of the target segmentation model for vascular segmentation.

[0078] Figure 4 It is a flowchart of another vascular segmentation method provided in the embodiment of the present invention. This embodiment is optimized based on the above technical solutions. In this embodiment, optionally, training the original segmentation model based on multiple sets of training samples to obtain the target segmentation model includes: for each set of training samples in multiple sets of training samples, determining a rough segmentation label according to the vascular segmentation label in the training samples; training the original segmentation model based on multiple sets of training samples and the rough segmentation labels respectively corresponding to the multiple sets of training samples to obtain the target segmentation model. Among them, the explanations of the same or corresponding terms as those in the above embodiments will not be repeated here.

[0079] See Figure 4 , the method of this embodiment may specifically include the following steps:

[0080] S410. Obtain a pre-built original segmentation model, a sample medical image, and a vascular segmentation label of the sample medical image, and use the sample medical image and the vascular segmentation label as a set of training samples.

[0081] S420. For each set of training samples in multiple sets of training samples, determine a rough segmentation label according to the vascular segmentation label in the training samples.

[0082] Among them, the rough segmentation label is a label of the vascular segmentation result obtained by performing rough vascular segmentation on the sample medical image; the segmentation accuracy of the rough segmentation label is lower than that of the vascular segmentation label.

[0083] It can be understood that the vascular segmentation label is the vascular segmentation result of a relatively accurate sample medical image. However, the rough segmentation network in the original segmentation model is only a network for rough vascular segmentation. That is, in the rough segmentation stage, only the overall structure of the blood vessels needs to be found, and there is no need to pay too much attention to whether the blood vessels are completely segmented. That is, only the position where the blood vessels are located needs to be focused on. If only the vascular segmentation label is used to train the rough segmentation network, the trained rough segmentation network will not meet the actual needs. Therefore, in the embodiments of the present invention, for each group of training samples in multiple groups of training samples, a rough segmentation label can be determined according to the vascular segmentation label in the training sample, so that the rough segmentation label can participate in the training of the original segmentation model in the subsequent process, so that the trained target segmentation model can better meet the actual needs.

[0084] In the embodiments of the present invention, a rough segmentation label is determined according to the vascular segmentation label in the training sample. For example, the segmented blood vessels in the vascular segmentation label can be smoothed, etc. In the embodiments of the present invention, the method for determining the rough segmentation label according to the vascular segmentation label in the training sample is not specifically limited.

[0085] S430. Train the original segmentation model based on multiple groups of training samples and the rough segmentation labels respectively corresponding to the multiple groups of training samples to obtain a target segmentation model.

[0086] In the embodiments of the present invention, the entire original segmentation model can be trained based on multiple groups of training samples and the rough segmentation labels respectively corresponding to the multiple groups of training samples to obtain a target segmentation model; the rough segmentation network in the original segmentation model can also be trained based on multiple groups of training samples and the rough segmentation labels respectively corresponding to the multiple groups of training samples, and the other parts of the original segmentation model except the rough segmentation network can be trained based on multiple groups of training samples to obtain a target segmentation model; the rough segmentation network in the original segmentation model can also be trained based on the rough segmentation labels respectively corresponding to multiple groups of training samples, and the other parts of the original segmentation model except the rough segmentation network can be trained based on multiple groups of training samples to obtain a target segmentation model. In the embodiments of the present invention, the method for training the original segmentation model based on multiple groups of training samples and the rough segmentation labels respectively corresponding to the multiple groups of training samples to obtain a target segmentation model is not specifically limited.

[0087] S440. Obtain a target medical image to be subjected to vascular segmentation, and a trained target segmentation model for vascular segmentation. The target segmentation model includes a rough segmentation network, a feature aggregation network, and a fine segmentation network. The feature aggregation network is used to obtain a target feature image based on the target medical image and the rough feature image output by the rough segmentation network, and input the target feature image into the fine segmentation network.

[0088] S450. Determine the vascular segmentation result of the target medical image based on the target medical image and the target segmentation model.

[0089] In the technical solution of the embodiment of the present invention, for each group of training samples in multiple groups of training samples, according to the vascular segmentation label in the training sample, a rough segmentation label is determined; based on multiple groups of training samples and the rough segmentation labels respectively corresponding to the multiple groups of training samples, the original segmentation model is trained to obtain the target segmentation model. Through the above technical solution, by making the rough segmentation label participate in the training of the original segmentation model, the trained target segmentation model can better meet the actual requirements.

[0090] An optional technical solution for training the original segmentation model based on multiple groups of training samples and the rough segmentation labels respectively corresponding to the multiple groups of training samples to obtain the target segmentation model includes: for each group of training samples in multiple groups of training samples, the sample medical image in the training sample is divided into blocks to obtain at least one sample block; for each sample block in the at least one sample block, the sample block is further divided into blocks to obtain a second intermediate block; the sample block is input into the rough segmentation network in the original segmentation model to obtain a first sample feature image and a rough segmentation result; the first sample feature image and the second intermediate block are input into the feature aggregation network in the original segmentation model to obtain a second sample feature image; the second sample feature image is input into the fine segmentation network to obtain a second output result output by the fine segmentation network for the sample block; based on the rough segmentation label, the vascular segmentation label in the training sample, the rough segmentation result, and the second output result, the original segmentation model is trained to obtain the target segmentation model.

[0091] Among them, the sample block is an image obtained by dividing the sample medical image into blocks. The second intermediate block is an image block in the middle area of the sample block obtained by further dividing the sample block; the side length of the second intermediate block can be half of the side length of the sample block. The first sample feature image is the feature image output by the rough segmentation network for the sample block, that is, the first sample feature image is the last feature information output after inputting the sample block into the rough segmentation network. The rough segmentation result is the segmentation result obtained by performing rough vascular segmentation on the sample block output by the rough segmentation network. The second sample feature image is the feature image obtained by the feature aggregation network aggregating the first sample feature image and the second intermediate block. The second output result is the output result output by the fine segmentation network for the sample block after inputting the second sample feature image into the fine segmentation network.

[0092] It should be noted that there may be overlapping regions between at least one sample block.

[0093] It should be noted that when using the target segmentation model, the target medical image is segmented into at least one pixel block by traversing the entire target medical image until each pixel point in the target medical image is traversed. However, when segmenting the sample medical image in the training sample to obtain at least one sample block, it is not necessary to traverse the entire sample medical image. It is only necessary to segment the sample medical image in the training sample to obtain at least one sample block including part or all of the sample medical image. For example, at least one sample block can be randomly segmented and cropped from the sample medical image.

[0094] Exemplarily, the sample medical image in the training sample can be segmented to randomly crop at least one sample block with a size of 2N*2N*2N; for each sample block in the at least one sample block, the sample block can be resampled to a size of N*N*N, and the resampled sample block is input into the coarse segmentation network in the original segmentation model to obtain a first sample feature image and a coarse segmentation result; the sample block with a size of 2N*2N*2N that has not been resampled can be cropped to obtain a second intermediate block with a size of N*N*N; the second intermediate block and the first sample feature image are input into the feature aggregation network; the feature aggregation network can include a feature cropping module, a resampling module, and a multiplication module. The first sample feature image can be input into the cropping module, and the cropping module can crop the first sample feature image to obtain a feature center block of the first sample feature image with a size of (N / 2)*(N / 2)*(N / 2); the feature center block is input into the resampling module to resample to obtain a feature center block with a size of N*N*N; the feature center block with a size of N*N*N and the second intermediate block are input into the multiplication module, and the multiplication module can perform a multiplication operation on the feature center block with a size of N*N*N and the second intermediate block to obtain a second sample feature image; the second sample feature image is input into the fine segmentation network to obtain a second output result output by the fine segmentation network for the sample block.

[0095] Exemplarily, refer to Figure 5 , the original segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network; the coarse segmentation network sequentially includes a first convolutional layer, a first downsampling layer, a second convolutional layer, a second downsampling layer, a third convolutional layer, a third downsampling layer, a fourth convolutional layer, a fourth downsampling layer, a fifth convolutional layer, a sixth convolutional layer, a first upsampling layer, a seventh convolutional layer, a second upsampling layer, an eighth convolutional layer, a third upsampling layer, a ninth convolutional layer, a fourth upsampling layer, and a tenth convolutional layer. The output of the fourth upsampling layer is the first sample feature image, and the output of the tenth convolutional layer is the coarse segmentation result. For the specific data flow in aspects such as cascading in the coarse segmentation network, refer to Figure 5As shown; it should be noted that there may also be a fifth downsampling layer, which may be located inside the original segmentation model, before the first convolutional layer, or may also be located outside the original segmentation model. The fifth downsampling layer is used to resample a sample block of size 2N*2N*2N to a size of N*N*N and input the resampled sample block into the first convolutional layer; the specific modules included in the feature aggregation network are not limited, as long as the feature aggregation network can crop and intercept the first sample feature image input into it to obtain a feature center block of size (N / 2)*(N / 2)*(N / 2), magnify and resample the feature center block to obtain a feature center block of size N*N*N, perform a multiplication operation on the feature center block of size N*N*N and the second intermediate block to obtain a second sample feature image, and input the second sample feature image into the fine segmentation network; the fine segmentation network may sequentially include an eleventh convolutional layer, a fifth downsampling layer, a twelfth convolutional layer, a sixth downsampling layer, a thirteenth convolutional layer, a seventh downsampling layer, a fourteenth convolutional layer, an eighth downsampling layer, a fifteenth convolutional layer, a sixteenth convolutional layer, a fifth upsampling layer, a seventeenth convolutional layer, a sixth upsampling layer, an eighteenth convolutional layer, a seventh upsampling layer, a nineteenth convolutional layer, an eighth upsampling layer, and a twentieth convolutional layer. The output of the twentieth convolutional layer is the rough segmentation result. For the specific data flow in aspects such as cascading in the fine segmentation network, refer to Figure 5 As shown; among them, the convolutional layers used above can all be convolutional layers with a convolutional kernel size of 3*3, and the upsampling layers used above can all be transposed convolutional layers with a convolutional kernel size of 2*2. Specifically, an original segmentation model, a sample medical image, and the vascular segmentation label of the sample medical image can be obtained in advance; according to the vascular segmentation label, a rough segmentation label can be determined; the sample medical image in the training sample is divided into blocks to randomly extract at least one sample block of size 2N*2N*2N; for each sample block in at least one sample block, the sample block is cropped to obtain a second intermediate block of size N*N*N; the at least one sample block of size 2N*2N*2N randomly extracted and the second intermediate block of size N*N*N are input into the original segmentation model to obtain a rough segmentation result and a second output result; based on the rough segmentation label, the vascular segmentation label, the rough segmentation result, and the second output result, the original segmentation model is trained to obtain a target segmentation model.

[0096] Exemplarily, since in the rough segmentation stage, only the overall architecture of the blood vessels needs to be found, without paying too much attention to whether the blood vessels are completely segmented, that is, only the position where the blood vessels are located needs to be focused on. Therefore, a loss function L based on the BCE loss function and divided into two parts, namely the normal loss of the overall blood vessels and the vascular framework loss, can be used. c = -(y×logp+(1 - y)×log(1 - p)+k×(y kernel× log(p kernel ) + (1 - y kernel ) * log(1 - p kernel ), based on the rough segmentation label, the vascular segmentation label in the training sample, and the rough segmentation result, train the rough segmentation network in the original segmentation model, and / or, use the loss function L f = BCE + Dice, that is, use the loss function fused by the BCE loss function and the Dice loss function, and train the fine segmentation network in the original segmentation model based on the vascular segmentation label and the second output result to obtain the target segmentation model, so that when training the rough segmentation network, it can pay more attention to the overall framework of blood vessels and not care about whether the vascular segmentation is appropriate. When training the fine segmentation network, the second sample feature image that fuses the original sample medical image and the deep feature information of the rough segmentation can be used. Based on the blood vessel framework, use the high-resolution information of the image for fine segmentation to easily segment blood vessels; that is, the loss function L = L c + αL f can be adopted. Based on the rough segmentation label, the vascular segmentation label in the training sample, the rough segmentation result, and the second output result, train the original segmentation model to obtain the target segmentation model. Among them, L c is the loss function used to train the rough segmentation network; L f is the loss function used to train the fine segmentation network; L is the overall loss function used to train the original segmentation model; y is the vascular segmentation label; y kernel is the rough segmentation label; p and p kernel can both represent the rough segmentation result; k is a hyperparameter weight greater than 1, which can be set according to requirements or experience; α is a hyperparameter greater than 1, and the setting of α can make the final focus of the model on the result of the fine segmentation network.

[0097] It should be noted that in the case of training the original segmentation model based on the rough segmentation label, the vascular segmentation label in the training sample, the rough segmentation result, and the second output result to obtain the target segmentation model, it is also necessary to perform block processing on the rough segmentation label and the vascular segmentation label respectively to obtain labels corresponding to the size and position of the rough segmentation result and the second output result, so as to facilitate the training of the original segmentation model. In the embodiments of the present invention, the method of training the original segmentation model based on the rough segmentation label, the vascular segmentation label in the training sample, the rough segmentation result, and the second output result is not specifically limited.

[0098] In the embodiments of the present invention, data augmentation such as random rotation, random translation, random brightness change, and / or local brightness transformation can also be performed on the sample medical image and / or at least one sample block to increase the generalization ability of the trained target segmentation model.

[0099] In an embodiment of the present invention, for each group of training samples among multiple groups of training samples, the sample medical images in the training samples are segmented to obtain at least one sample block; for each sample block among the at least one sample block, the sample block is segmented to obtain a second intermediate block; the sample block is input into the rough segmentation network in the original segmentation model to obtain a first sample feature image and a rough segmentation result; the first sample feature image and the second intermediate block are input into the feature aggregation network in the original segmentation model to obtain a second sample feature image; the second sample feature image is input into the fine segmentation network to obtain a second output result output by the fine segmentation network for the sample block; based on the rough segmentation label, the blood vessel segmentation label in the training sample, the rough segmentation result, and the second output result, the original segmentation model is trained to obtain a target segmentation model. The above technical solution can enable the rough segmentation label, the blood vessel segmentation label in the training sample, the rough segmentation result, and the second output result to participate in the training of the original segmentation model, so that the trained target segmentation model can better meet the actual needs.

[0100] Another alternative technical solution for determining the rough segmentation label according to the blood vessel segmentation label in the training sample includes: determining the blood vessel skeleton of the blood vessel to be segmented in the target medical image according to the blood vessel segmentation label in the training sample; performing dilation processing on the blood vessel skeleton to obtain the rough segmentation label.

[0101] Wherein, the blood vessel skeleton is the skeleton of the blood vessel to be segmented in the target medical image.

[0102] Exemplarily, the blood vessel segmentation label can be used to extract the skeleton of the image by algorithms such as the Skeleton algorithm to obtain the blood vessel skeleton of the blood vessel to be segmented in the target medical image; the blood vessel skeleton is dilated by algorithms such as the dilation algorithm to obtain the rough segmentation label.

[0103] In an embodiment of the present invention, the blood vessel skeleton of the blood vessel to be segmented in the target medical image can be determined according to the blood vessel segmentation label in the training sample; the blood vessel skeleton is dilated to obtain the rough segmentation label. The above technical solution can make the obtained rough segmentation label more corresponding to the rough segmentation result obtained by the rough segmentation network by dilating the blood vessel skeleton obtained according to the blood vessel segmentation label, thereby improving the training effect of training the original segmentation model.

[0104] Figure 6 The structural block diagram of the blood vessel segmentation device provided by the embodiment of the present invention is shown. This device is used to execute the blood vessel segmentation method provided in any of the above embodiments. This device and the blood vessel segmentation methods in the above embodiments belong to the same inventive concept. For the details not described in detail in the embodiment of the blood vessel segmentation device, reference can be made to the embodiments of the above blood vessel segmentation methods. SeeFigure 6 Specifically, the device may include: a target segmentation model acquisition module 510 and a blood vessel segmentation result determination module 520.

[0105] The target segmentation model acquisition module 510 is configured to acquire a target medical image to be subjected to blood vessel segmentation, and a trained target segmentation model for blood vessel segmentation.

[0106] The blood vessel segmentation result determination module 520 is configured to determine a blood vessel segmentation result of the target medical image based on the target medical image and the target segmentation model.

[0107] The target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network. The feature aggregation network is configured to obtain a target feature image based on the target medical image and the coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network.

[0108] Optionally, the blood vessel segmentation result determination module 520 may include:

[0109] A pixel block obtaining sub-module, configured to perform block processing on the target medical image to obtain at least one pixel block.

[0110] A first intermediate block obtaining sub-module, configured to perform block processing on each pixel block among the at least one pixel block to obtain a first intermediate block.

[0111] A pixel output result obtaining sub-module, configured to input the pixel block and the first intermediate block into the target segmentation model to obtain a pixel output result output by the target segmentation model for the pixel block.

[0112] A blood vessel segmentation result obtaining sub-module, configured to obtain a blood vessel segmentation result of the target medical image according to the pixel output results respectively corresponding to the at least one pixel block.

[0113] Optionally, based on the above device, the pixel output result obtaining sub-module may include:

[0114] A coarse feature image obtaining unit, configured to input the pixel block into the coarse segmentation network to obtain a coarse feature image.

[0115] A target feature image obtaining unit, configured to input the coarse feature image and the first intermediate block into the feature aggregation network to obtain a target feature image.

[0116] A pixel output result as unit, configured to input the target feature image into the fine segmentation network to obtain a first output result output by the fine segmentation network for the pixel block, and use the first output result as the pixel output result output by the target segmentation model for the pixel block.

[0117] Optionally, the device may further include the following modules to pre-train a target segmentation model:

[0118] A training sample serving as a module for obtaining a pre-built original segmentation model, sample medical images, and vascular segmentation labels of the sample medical images, and using the sample medical images and vascular segmentation labels as a set of training samples;

[0119] A target segmentation model obtaining module for training the original segmentation model based on multiple sets of training samples to obtain the target segmentation model.

[0120] Optionally, based on the above device, the target segmentation model obtaining module may include:

[0121] A rough segmentation label determining sub-module for determining a rough segmentation label for each set of training samples in multiple sets of training samples according to the vascular segmentation labels in the training samples;

[0122] A target segmentation model obtaining sub-module for training the original segmentation model based on multiple sets of training samples and the rough segmentation labels respectively corresponding to the multiple sets of training samples to obtain the target segmentation model.

[0123] Optionally, based on the above device, the target segmentation model obtaining sub-module may include:

[0124] A sample block obtaining unit for performing a blocking process on the sample medical images in each set of training samples in multiple sets of training samples to obtain at least one sample block;

[0125] A second intermediate block obtaining unit for performing a blocking process on each sample block in at least one sample block to obtain a second intermediate block;

[0126] A rough segmentation result waiting unit for inputting the sample block into the rough segmentation network in the original segmentation model to obtain a first sample feature image and a rough segmentation result;

[0127] A second sample feature image obtaining unit for inputting the first sample feature image and the second intermediate block into the feature aggregation network in the original segmentation model to obtain a second sample feature image;

[0128] A second output result obtaining unit for inputting the second sample feature image into the fine segmentation network to obtain a second output result output by the fine segmentation network for the sample block;

[0129] A target segmentation model obtaining unit for training the original segmentation model based on the rough segmentation label, the vascular segmentation label in the training sample, the rough segmentation result, and the second output result to obtain the target segmentation model.

[0130] Optionally, based on the above device, the coarse segmentation label determination sub-module may include:

[0131] A vascular skeleton determination unit, configured to determine the vascular skeleton of the blood vessels to be segmented in the target medical image according to the vascular segmentation labels in the training samples;

[0132] A coarse segmentation label obtaining unit, configured to perform dilation processing on the vascular skeleton to obtain coarse segmentation labels.

[0133] The vascular segmentation device provided by the embodiments of the present invention obtains a target medical image to be subjected to vascular segmentation and a trained target segmentation model for performing vascular segmentation through a target segmentation model acquisition module; and determines the vascular segmentation result of the target medical image based on the target medical image and the target segmentation model through a vascular segmentation result determination module, wherein the target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network, and the feature aggregation network is configured to obtain a target feature image based on the target medical image and the coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network. The above device can enable the obtained vascular segmentation result to fully consider global information through the feature aggregation network that can perform feature aggregation on the target medical image and the coarse feature image, avoid false positives or omissions in the vascular segmentation result, and thus improve the accuracy of vascular segmentation.

[0134] The vascular segmentation device provided by the embodiments of the present invention can execute the vascular segmentation method provided by any embodiment of the present invention, and has corresponding function modules and beneficial effects for executing the method.

[0135] It should be noted that in the embodiments of the above vascular segmentation device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0136] Figure 7 FIG. shows a schematic structural diagram of an electronic device 10 that can be used to implement the embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, a personal digital processor, a cellular phone, a smart phone, a wearable device (such as a helmet, glasses, a watch, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present invention described herein and / or claimed.

[0137] AsFigure 7 As shown in Figure 7 , the electronic device 10 includes at least one processor 11 and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. The memory stores a computer program executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0138] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0139] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the blood vessel segmentation method.

[0140] In some embodiments, the blood vessel segmentation method can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the blood vessel segmentation method described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the blood vessel segmentation method by any other appropriate means (e.g., by means of firmware).

[0141] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which may be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0142] The computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The computer programs may be executed entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

[0143] In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium may be a machine-readable signal medium. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0144] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0145] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0146] The computing system can include a client and a server. The client and the server are generally far from each other and typically interact through a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system and solves the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0147] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is made herein.

[0148] The above specific embodiments do not constitute a limitation to the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for blood vessel segmentation, characterized in that, Including: Obtaining a target medical image to be subjected to blood vessel segmentation, and a trained target segmentation model for blood vessel segmentation; Determining a blood vessel segmentation result of the target medical image based on the target medical image and the target segmentation model; Wherein, the target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network, and the feature aggregation network is configured to obtain a target feature image based on the target medical image and a coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network.

2. The method according to claim 1, characterized in that, The determining the blood vessel segmentation result of the target medical image based on the target medical image and the target segmentation model includes: Performing a block processing on the target medical image to obtain at least one pixel block; For each pixel block in the at least one pixel block, performing a block processing on the pixel block to obtain a first intermediate block; Inputting the pixel block and the first intermediate block into the target segmentation model to obtain a pixel output result output by the target segmentation model for the pixel block; Obtaining a blood vessel segmentation result of the target medical image according to the pixel output results respectively corresponding to the at least one pixel block.

3. The method according to claim 2, characterized in that, The inputting the pixel block and the first intermediate block into the target segmentation model to obtain a pixel output result output by the target segmentation model for the pixel block includes: Inputting the pixel block into the coarse segmentation network to obtain a coarse feature image; Inputting the coarse feature image and the first intermediate block into the feature aggregation network to obtain a target feature image; Inputting the target feature image into the fine segmentation network to obtain a first output result output by the fine segmentation network for the pixel block, and using the first output result as the pixel output result output by the target segmentation model for the pixel block.

4. The method according to claim 1, wherein The target segmentation model is pre-trained through the following steps: Obtaining a pre-built original segmentation model, sample medical images, and blood vessel segmentation labels of the sample medical images, and using the sample medical images and the blood vessel segmentation labels as a group of training samples; Training the original segmentation model based on multiple groups of the training samples to obtain the target segmentation model.

5. The method according to claim 4, wherein The training the original segmentation model based on multiple groups of the training samples to obtain the target segmentation model includes: For each group of training samples in the multiple groups of training samples, determining a coarse segmentation label according to the blood vessel segmentation label in the training sample; Training the original segmentation model based on multiple groups of the training samples and the coarse segmentation labels respectively corresponding to the multiple groups of the training samples to obtain the target segmentation model.

6. The method according to claim 5, wherein The training the original segmentation model based on multiple groups of the training samples and the coarse segmentation labels respectively corresponding to the multiple groups of the training samples to obtain the target segmentation model includes: For each group of training samples in the multiple groups of training samples, performing a block processing on the sample medical image in the training sample to obtain at least one sample block; For each of the at least one sample block, the sample block is processed in blocks to obtain a second intermediate block; The sample block is input into the coarse segmentation network in the original segmentation model to obtain a first sample feature image and a coarse segmentation result; The first sample feature image and the second intermediate block are input into the feature aggregation network in the original segmentation model to obtain a second sample feature image; The second sample feature image is input into the fine segmentation network to obtain a second output result output by the fine segmentation network for the sample block; Based on the coarse segmentation label, the blood vessel segmentation label in the training sample, the coarse segmentation result, and the second output result, the original segmentation model is trained to obtain the target segmentation model.

7. The method according to claim 5, characterized in that The determining the coarse segmentation label according to the blood vessel segmentation label in the training sample includes: According to the blood vessel segmentation label in the training sample, determining the blood vessel skeleton of the blood vessels to be segmented in the target medical image; Performing dilation processing on the blood vessel skeleton to obtain a coarse segmentation label.

8. A blood vessel segmentation device, characterized in that, Including: A target segmentation model acquisition module, configured to acquire a target medical image to be subjected to blood vessel segmentation, and a trained target segmentation model for performing blood vessel segmentation; A blood vessel segmentation result determination module, configured to determine a blood vessel segmentation result of the target medical image based on the target medical image and the target segmentation model; Wherein, the target segmentation model includes a coarse segmentation network, a feature aggregation network, and a fine segmentation network, and the feature aggregation network is configured to obtain a target feature image based on the target medical image and the coarse feature image output by the coarse segmentation network, and input the target feature image into the fine segmentation network.

9. An electronic device, characterized in that, Including: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor executes the blood vessel segmentation method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and the computer instructions are used to cause the processor to implement the blood vessel segmentation method according to any one of claims 1-7 when executed.

Images