Vascular medical image segmentation method, device and apparatus

Abstract

The present application discloses a method, device and equipment for segmenting vascular medical images. The method comprises: obtaining a medical three-dimensional image of a blood vessel; inputting the medical three-dimensional image of the blood vessel into a pre-trained vascular image segmentation model, segmenting the three-dimensional image, and obtaining a segmented three-dimensional predicted image of the blood vessel; the vascular image segmentation model comprises an input layer, a first Transformer neural network unit block, a second Transformer neural network unit block, a third Transformer neural network unit block, a fifth deconvolution unit block, a first convolution unit block, a first deconvolution unit block, a second convolution unit block, a second deconvolution unit block and an output layer connected in sequence. The vascular image segmentation model of the present invention includes a Transformer neural network unit block; it has the ability to take into account the previous and next frame images, maintain the continuity of image features to a great extent, learn global information, and avoid the loss of important information, and has a better segmentation effect for the three-dimensional stereoscopic image of the blood vessel.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and device for segmenting vascular medical images. Background Art

[0002] Computed tomography angiography (CTA) is a widely used, minimally invasive and cost-effective imaging technology, which is commonly used for routine clinical diagnosis of head and neck blood vessels, especially cerebrovascular diseases. For CT images of blood vessels, vascular image segmentation has always been a hot topic in medical image analysis, which is crucial for diagnosis, treatment planning and execution, and evaluation of clinical outcomes in different fields such as intracardiac and intracranial blood vessels. However, in the existing technology, the accuracy of vascular image segmentation is relatively low. Because blood vessels are a complex three-dimensional shape, the existing technology uses three-dimensional images of blood vessels for segmentation, and the continuity is relatively poor. Summary of the invention

[0003] The main purpose of this application is to provide a vascular medical image segmentation method, device and equipment to solve the above problems.

[0004] In order to achieve the above object, according to one aspect of the present application, a vascular medical image segmentation method is provided, comprising:

[0005] Obtain medical 3D images of blood vessels;

[0006] Inputting the medical three-dimensional image of the blood vessel into a pre-trained blood vessel image segmentation model, segmenting the three-dimensional image, and obtaining a segmented three-dimensional predicted image of the blood vessel;

[0007] The vascular image segmentation model includes an input layer, a first Transformer neural network unit block, a second Transformer neural network unit block, a third Transformer neural network unit block, a fifth deconvolution unit block, a first convolution unit block, a first deconvolution unit block, a second convolution unit block, a second deconvolution unit block and an output layer, which are connected in sequence.

[0008] In one embodiment, the model further includes a third deconvolution unit and a fourth deconvolution unit;

[0009] The third deconvolution unit is connected to the output end of the first Transformer neural network unit block and the input end of the second convolution unit block;

[0010] A fourth deconvolution unit; the fourth deconvolution unit is connected to the output end of the first Transformer neural network unit block and the input end of the first convolution unit block.

[0011] In one implementation, the size of the image input to the input layer is (D, W, H);

[0012] Wherein, D is the width of the image; W is the length of the image; H is the thickness of the image; the number of the image is 1;

[0013] The size of the image output by the output layer is (D, W, H); the number is 1;

[0014] The model also includes a linear projection and a slice embedding module for slice processing of the image; after the slice processing, the number of pictures is 768; the size of a single picture is (D / 16, H / 16, W / 16);

[0015] The image size output by the third deconvolution unit is (D / 2, H / 2, W / 2); the number is 128;

[0016] The convolution kernel of the third deconvolution unit has a size of 2×2×2 and a number of 3;

[0017] The image size output by the fourth deconvolution unit is (D / 4, H / 4, W / 4); the number is 256;

[0018] The convolution kernel of the fourth deconvolution unit has a size of 2×2×2 and a number of 2;

[0019] The image size output by the fifth deconvolution unit is (D / 4, H / 4, W / 4); the number is 256;

[0020] The size of the convolution kernel of the fifth deconvolution unit is 2×2×2; the number is 1.

[0021] In one embodiment, training a blood vessel image segmentation model includes:

[0022] Obtain medical 3D images of blood vessels;

[0023] Performing data enhancement processing on the medical three-dimensional image to obtain training samples for a segmentation model;

[0024] The training samples of the segmentation model are input into the segmentation model, the segmentation model is trained, and the training is stopped when the loss function of the segmentation model is less than a predetermined threshold.

[0025] In one embodiment, performing data enhancement processing on the medical three-dimensional image includes:

[0026] flipping the medical three-dimensional image horizontally; or,

[0027] The medical three-dimensional image is randomly rotated, and the rotation angle is less than a predetermined angle threshold; or,

[0028] The medical three-dimensional image is moved horizontally or vertically, and the moving distance is less than a predetermined number of pixel distances.

[0029] In a second aspect, the present application also proposes a vascular image segmentation model training method, comprising:

[0030] Obtain medical 3D images of blood vessels;

[0031] Performing data enhancement processing on the medical three-dimensional image to obtain training samples for a segmentation model;

[0032] The training samples of the segmentation model are input into the segmentation model, the segmentation model is trained, and the training is stopped when the loss function of the segmentation model is less than a predetermined threshold.

[0033] In one embodiment, performing data enhancement processing on the medical three-dimensional image includes:

[0034] flipping the medical three-dimensional image horizontally; or,

[0035] The medical three-dimensional image is randomly rotated, and the rotation angle is less than a predetermined angle threshold; or,

[0036] The medical three-dimensional image is moved horizontally or vertically, and the moving distance is less than a predetermined number of pixel distances.

[0037] In a third aspect, the present application further proposes a vascular medical image segmentation device, comprising:

[0038] An acquisition module, used for acquiring a medical three-dimensional image of a blood vessel;

[0039] A processing module, used for inputting the medical three-dimensional image of the blood vessel into a pre-trained blood vessel image segmentation model, segmenting the three-dimensional image, and obtaining a segmented three-dimensional image of the blood vessel;

[0040] The vascular image segmentation model includes an input layer, a first Transformer neural network unit block, a second Transformer neural network unit block, a third Transformer neural network unit block, a fifth deconvolution unit block, a first convolution unit block, a first deconvolution unit block, a second convolution unit block, a second deconvolution unit block and an output layer, which are connected in sequence.

[0041] In a fourth aspect, the present application proposes a vascular image segmentation model training device, comprising:

[0042] An acquisition module, used for acquiring a medical three-dimensional image of a blood vessel;

[0043] An enhancement module, used for performing data enhancement processing on the medical three-dimensional image to obtain training samples for a segmentation model;

[0044] The processing module is used to input the training samples of the segmentation model into the segmentation model, train the segmentation model, and stop the training when the loss function of the segmentation model is less than a predetermined threshold.

[0045] In a fifth aspect, the present application proposes a vascular medical image segmentation device, comprising: at least one processor and at least one memory; the memory is used to store one or more program instructions; the processor is used to run one or more program instructions to execute any of the above methods to achieve the above purpose.

[0046] According to a sixth aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium contains one or more program instructions, and the one or more program instructions are used to execute any one of the steps described above.

[0047] In the embodiment of the present application, the above-mentioned technical solution of the present invention includes a Transformer neural network unit block in the vascular image segmentation model; the characteristics of the Transformer neural network unit block are fully utilized. The Transformer neural network unit can take into account the previous and next frame images and pay attention to the temporal connection. Not completely relying on the convolutional neural network as an encoder can maintain the consistency of image features to a great extent, learn global information, and avoid the loss of important information. Therefore, it will have a better segmentation effect for the 3D stereo image of the blood vessel. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings constituting a part of this application are used to provide a further understanding of this application, so that other features, purposes and advantages of this application become more obvious. The schematic embodiment drawings and their descriptions of this application are used to explain this application and do not constitute an improper limitation on this application. In the drawings:

[0049] Figure 1 is a flow chart of a vascular medical image segmentation method according to an embodiment of the present application;

[0050] Figure 2 is a three-dimensional predicted image of a blood vessel according to an embodiment of the present application;

[0051] Figure 3 is a schematic diagram of the structure of a Transformer neural network unit according to an embodiment of the present application;

[0052] Figure 4 is a schematic diagram of the structure of a blood vessel image segmentation model according to an embodiment of the present application;

[0053] Figure 5 is an original CTA image of a blood vessel according to an embodiment of the present application;

[0054] Figure 6 is a blood vessel annotated image according to an embodiment of the present application;

[0055] Figure 7 is a 3D image of blood vessels annotated according to an embodiment of the present application;

[0056] Figure 8 is a structural schematic diagram of another blood vessel image segmentation model according to an embodiment of the present application;

[0057] Fig. 9 is a structural schematic diagram of a vascular medical image segmentation device according to an embodiment of the present application;

[0058] Fig.10 is a structural schematic diagram of a blood vessel image segmentation model training device according to an embodiment of the present application;

[0059] Fig.11 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0061] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0062] In addition, the terms "installed", "set", "provided with", "connected", "connected", and "socketed" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection, or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, elements, or components. For those of ordinary skill in the art, the specific meanings of the above terms in this utility model can be understood according to specific circumstances.

[0063] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0064] Current cerebrovascular segmentation techniques use classic rule-based algorithms, such as centerline tracking, active contour models, or region growing. However, such algorithms cannot distinguish between vascular systems and artifacts and are easily affected by other tissues. For example, the CT values ​​of blood vessels at the edge of the skull are highly similar to those of the skull, resulting in adhesion of extravascular tissues, making it impossible to accurately segment and extract the state of arterial vessels.

[0065] Deep learning-based segmentation methods have attracted more and more people's interest due to their self-learning and generalization capabilities for large amounts of data, especially the widely accepted U-Net architecture, which is a CNN-based segmentation network and is considered to be an effective alternative to traditional cerebrovascular segmentation methods. However, because the structure of blood vessels is tortuous and has many branches, a major technical challenge of this method is how to accurately segment uninterrupted blood vessels.

[0066] Based on this, this application proposes a vascular medical image segmentation method, see the attached Figure 1 A flow chart of a vascular medical image segmentation method is shown, the method comprising:

[0067] Step S102, obtaining a medical three-dimensional image of a blood vessel;

[0068] Specifically, a three-dimensional CTA image of a blood vessel can be obtained;

[0069] Step S104, inputting the medical three-dimensional image of the blood vessel into a pre-trained blood vessel image segmentation model, segmenting the three-dimensional image, and obtaining a segmented three-dimensional predicted image of the blood vessel;

[0070] For example, see the attached Figure 2 The 3D predicted image of blood vessels is shown in FIG. The 3D predicted image of blood vessels is clear, complete and coherent, which is beneficial for doctors to conduct further medical analysis.

[0071] The vascular image segmentation model includes an input layer, a first Transformer neural network unit block, a second Transformer neural network unit block, a third Transformer neural network unit block, a fifth deconvolution unit block, a first convolution unit block, a first deconvolution unit block, a second convolution unit block, a second deconvolution unit block and an output layer, which are connected in sequence.

[0072] Among them, see Appendix Figure 3 The schematic diagram of the structure of a Transformer neural network unit is shown in the figure; the neural network unit includes an input layer, a first LayerNormal layer, an MSA layer, a second LayerNormal layer, an MLP layer, and an output layer. Each Transformer neural network unit contains two sublayers, the first layer combines a normalized layer and a multi-head attention mechanism, and the second sublayer combines a normalized layer and a multi-layer perceptron.

[0073] See attached Figure 4 The structural schematic diagram of the vascular image segmentation model of the present application is shown; the first Transformer neural network unit block, the second Transformer neural network unit block, and the third Transformer neural network unit block of the present application are Transformer×3 respectively; that is, they are composed of 3 Transformer units connected in series.

[0074] In the output layer, the convolution kernel is 1×1×1 and the activation function is Softmax.

[0075] In the above technical solution of the present invention, the blood vessel image segmentation model includes a Transformer neural network unit block; the characteristics of the Transformer neural network unit block are fully utilized. The Transformer neural network unit can take into account the previous and next frame images and pay attention to the temporal connection. It does not completely rely on the convolutional neural network as an encoder, which can maintain the consistency of image features to a great extent, learn global information, and avoid the loss of important information. Therefore, it will have a better segmentation effect for the 3D stereo image of the blood vessel.

[0076] In order to compensate for the image loss of the first convolution unit block and the second convolution unit block, in one embodiment, see the attached Figure 4 As shown, the model also includes a third deconvolution unit and a fourth deconvolution unit;

[0077] The third deconvolution unit is connected to the output end of the first Transformer neural network unit block and the input end of the second convolution unit block;

[0078] The third deconvolution unit outputs the image to the input end of the second convolution unit block;

[0079] A fourth deconvolution unit; the fourth deconvolution unit is connected to the output end of the first Transformer neural network unit block and the input end of the first convolution unit block.

[0080] The fourth deconvolution unit outputs the image to the input end of the first convolution unit block, and plays a role in image compensation.

[0081] In one implementation, the size of the image input to the input layer is (D, W, H);

[0082] Wherein, D is the width of the image; W is the length of the image; H is the thickness of the image; the number of the image is 1;

[0083] The size of the image output by the output layer is (D, W, H); the number is 1;

[0084] In one embodiment, it also includes a linear projection and slice embedding module for processing a three-dimensional image into a one-dimensional linear slice sequence, and the one-dimensional linear slice sequence is input into the transformer model.

[0085] In the one-dimensional linear slice sequence, the number of images is 768; the size of a single image is (D / 16, H / 16, W / 16);

[0086] The image size output by the third deconvolution unit is (D / 2, H / 2, W / 2); the number is 128;

[0087] The convolution kernel of the third deconvolution unit has a size of 2×2×2 and a number of 3;

[0088] The image size output by the fourth deconvolution unit is (D / 4, H / 4, W / 4); the number is 256;

[0089] The convolution kernel of the fourth deconvolution unit has a size of 2×2×2 and a number of 2;

[0090] The image size output by the fifth deconvolution unit is (D / 4, H / 4, W / 4); the number is 256;

[0091] The size of the convolution kernel of the fifth deconvolution unit is 2×2×2; the number is 1.

[0092] In one embodiment, when training the vascular image segmentation model, a medical three-dimensional image of the blood vessel is acquired;

[0093] Performing data enhancement processing on the medical three-dimensional image to obtain training samples for a segmentation model;

[0094] The training samples of the segmentation model are input into the segmentation model, the segmentation model is trained, and the training is stopped when the loss function of the segmentation model is less than a predetermined threshold.

[0095] In one embodiment, before performing the data enhancement process on the medical three-dimensional image, a CTA image and corresponding annotations are obtained.

[0096] For example, see the attached Figure 5 Original CTA images of the vessels shown;

[0097] The CTA sequences that meet the requirements are manually selected from all the original sequences of the case, and then the bone-free CTA sequences are obtained through image clipping technology, and the results are fed back to the expert group for data annotation.

[0098] For example, see the attached Figure 6 Annotated images of blood vessels are shown; see Appendix Figure 7 The 3D image with blood vessel annotation is shown.

[0099] Data annotation is divided into three steps. First, the boneless CTA sequence is imported into the Mimics software to obtain a preliminary vascular annotation model. Then the expert group adjusts and deletes the vascular model to ensure that a uniform standard fine-tuned vascular annotation model is obtained. Finally, it is handed over to experienced doctors for final review to obtain the final vascular annotation model. The obtained vascular annotations form a one-to-one corresponding usable data set with the previously selected CTA sequence.

[0100] The DICOM format CTA sequence and NII format annotation files are read from the dataset and converted into a three-dimensional image matrix. The dataset is divided into a training set and a validation set in a ratio of 9:1, and the three-dimensional images in the training set are subjected to the above data enhancement processing.

[0101] In one embodiment, when performing data enhancement processing on the medical three-dimensional image, the following steps are taken:

[0102] horizontally flipping the medical three-dimensional image;

[0103] Specifically, after horizontal flipping, an axisymmetric image is formed. Since the brain structure is roughly symmetrical, horizontal flipping of the image allows the model to learn the vascular structures on both sides of the left and right sides through one case.

[0104] Randomly rotating the medical three-dimensional image, wherein the rotation angle is less than a predetermined angle threshold;

[0105] The angle can be set flexibly. For example, the angle range of rotation around the z-axis is (-45, 45) degrees, and the angle threshold around the x-axis and / or y-axis must be less than 30 degrees.

[0106] The medical three-dimensional image is moved along the x, y, and z coordinate axes respectively, and the moving distance is less than a pixel distance threshold.

[0107] Exemplarily, the pixel distance threshold is 5 pixels, which can be set flexibly.

[0108] Randomly rotating images and moving images horizontally or vertically takes into account that different patients may obtain CTA images in different positions due to different body positions when undergoing CTA examinations.

[0109] In one implementation, a portion of the image may be randomly covered.

[0110] After the above data enhancement process, the data is preprocessed into 3D image blocks. The CTA images in the training set after 3D image enhancement are divided into 3D image blocks. 100 3D image blocks are extracted from each image for subsequent deep learning model training.

[0111] Specifically, two methods are used: random sampling and traversal sampling. In the random sampling, 50 three-dimensional image blocks of size 80×80×80 are randomly selected on the image. In the traversal sampling, the sliding window size is set to half the size of the image block to translate the sliding window and extract 50 three-dimensional image blocks of 80×80×80.

[0112] See attached Figure 8 The structural diagram of another vascular image segmentation model shown in FIG. divides the data-enhanced training set image into three-dimensional image blocks as the model input, with a size of 1×80×80×80. The model output is a segmentation prediction map of the same size, with two categories.

[0113] First, the input 3D image block is flattened into a one-dimensional sequence and projected into a K-dimensional embedding space using a linear layer, and the one-dimensional learnable position information is incorporated into it as the input of the Transformer encoder. The encoder consists of multiple stacked Transformer modules.

[0114] The present invention selects to construct a 3D segmentation model based on the characteristics of CTA images. Compared with the prior art, in which the deep learning model is used to process 2D images, the present invention constructs a 3D segmentation model and designs a 3D data enhancement method to ensure the integrity and relevance of the data to the greatest extent, and plays a role in improving the accuracy of brain vascular segmentation.

[0115] The present invention uses a variety of 3D data enhancement methods, combined with the characteristics of brain CTA images, such as horizontal flipping of images because the brain structure is symmetrical; random rotation of images and horizontal or vertical movement of images take into account the different body positions of different patients during CTA examinations. There are relatively few data enhancement methods applicable to 3D medical images, but their impact on model results is crucial, so choosing the above data enhancement methods can effectively improve model performance and enhance its robustness.

[0116] This paper proposes a technology for intracranial vessel segmentation based on deep learning algorithm and three-dimensional image enhancement. A neural network is designed to perform intracranial vessel segmentation on three-dimensional CTA images. The Transformer structure is used in conjunction with the three-dimensional image enhancement method to improve the accuracy of intracranial vessel segmentation and the robustness of the model.

[0117] In a second aspect, the present application also proposes a vascular image segmentation model training method, comprising:

[0118] Obtain medical 3D images of blood vessels;

[0119] Performing data enhancement processing on the medical three-dimensional image to obtain training samples for a segmentation model;

[0120] The training samples of the segmentation model are input into the segmentation model, the segmentation model is trained, and the training is stopped when the loss function of the segmentation model is less than a predetermined threshold.

[0121] In one embodiment, performing data enhancement processing on the medical three-dimensional image includes:

[0122] flipping the medical three-dimensional image horizontally; or,

[0123] The medical three-dimensional image is randomly rotated, and the rotation angle is less than a predetermined angle threshold; or,

[0124] The medical three-dimensional image is moved horizontally or vertically, and the moving distance is less than a predetermined number of pixel distances.

[0125] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0126] In the third aspect, the present application also proposes a vascular medical image segmentation device, see the attached Fig. 9 A structural schematic diagram of a vascular medical image segmentation device is shown; the device comprises:

[0127] An acquisition module 91, used for acquiring a medical three-dimensional image of a blood vessel;

[0128] A processing module 92, configured to input the medical three-dimensional image of the blood vessel into a pre-trained blood vessel image segmentation model, segment the three-dimensional image, and obtain a segmented three-dimensional image of the blood vessel;

[0129] The vascular image segmentation model includes an input layer, a first Transformer neural network unit block, a second Transformer neural network unit block, a third Transformer neural network unit block, a fifth deconvolution unit block, a first convolution unit block, a first deconvolution unit block, a second convolution unit block, a second deconvolution unit block and an output layer, which are connected in sequence.

[0130] In one embodiment, it further includes an enhanced training module 93 for acquiring a medical three-dimensional image of a blood vessel;

[0131] Performing data enhancement processing on the medical three-dimensional image to obtain training samples for a segmentation model;

[0132] The training samples of the segmentation model are input into the segmentation model, the segmentation model is trained, and the training is stopped when the loss function of the segmentation model is less than a predetermined threshold.

[0133] In one embodiment, the enhanced training module 93 is further used to horizontally flip the medical three-dimensional image; or,

[0134] The medical three-dimensional image is randomly rotated, and the rotation angle is less than a predetermined angle threshold; or,

[0135] The medical three-dimensional image is moved horizontally or vertically, and the moving distance is less than a predetermined number of pixel distances.

[0136] In a fourth aspect, the present application also proposes a vascular image segmentation model training device, see the attached Fig.10 The structure diagram of a blood vessel image segmentation model training device shown in FIG. 1 includes:

[0137] An acquisition module 101 is used to acquire a medical three-dimensional image of a blood vessel;

[0138] An enhancement module 102, configured to perform data enhancement processing on the medical three-dimensional image to obtain training samples for a segmentation model;

[0139] The processing module 103 is used to input the training samples of the segmentation model into the segmentation model, train the segmentation model, and stop the training when the loss function of the segmentation model is less than a predetermined threshold.

[0140] In one implementation, the enhancement module 102 is further configured to horizontally flip the medical three-dimensional image; or,

[0141] The medical three-dimensional image is randomly rotated, and the rotation angle is less than a predetermined angle threshold; or,

[0142] The medical three-dimensional image is moved horizontally or vertically, and the moving distance is less than a predetermined number of pixel distances.

[0143] According to a fifth aspect of the present application, an electronic device is provided. Fig.11 A structural diagram of an electronic device is shown; it includes at least one processor 111 and at least one memory 112; the memory 112 is used to store one or more program instructions; the processor 111 is used to run one or more program instructions to execute any of the above methods.

[0144] In a sixth aspect, the present application further proposes a computer-readable storage medium, which contains one or more program instructions, and the one or more program instructions are used to execute any of the methods described above.

[0145] The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The processor reads the information in the storage medium and completes the steps of the above method in combination with its hardware.

[0146] The storage medium may be a memory, which may be, for example, a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memory.

[0147] Among them, the non-volatile memory can be a read-only memory (ROM), a programmable ROM (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory.

[0148] The volatile memory may be a random access memory (RAM) which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDRSDRAM), enhanced synchronous DRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus DRAM (DRRAM).

[0149] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0150] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0151] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A vascular medical image segmentation method, characterized in that: include: Obtain medical 3D images of blood vessels; Inputting the medical three-dimensional image of the blood vessel into a pre-trained blood vessel image segmentation model, segmenting the three-dimensional image, and obtaining a segmented three-dimensional predicted image of the blood vessel; The vascular image segmentation model includes an input layer, a first Transformer neural network unit block, a second Transformer neural network unit block, a third Transformer neural network unit block, a fifth deconvolution unit block, a first convolution unit block, a first deconvolution unit block, a second convolution unit block, a second deconvolution unit block and an output layer, which are connected in sequence.

2. The vascular medical image segmentation method according to claim 1, characterized in that: The model also includes a third deconvolution unit and a fourth deconvolution unit; The third deconvolution unit is connected to the output end of the first Transformer neural network unit block and the input end of the second convolution unit block; A fourth deconvolution unit; the fourth deconvolution unit is connected to the output end of the first Transformer neural network unit block and the input end of the first convolution unit block.

3. The vascular medical image segmentation method according to claim 2, characterized in that: The size of the image input to the input layer is (D, W, H); Wherein, D is the width of the image; W is the length of the image; H is the thickness of the image; the number of the image is 1; The size of the image output by the output layer is (D, W, H); the number is 1; The model also includes a linear projection and a slice embedding module for slice processing of the image; after the slice processing, the number of pictures is 768; the size of a single picture is (D / 16, H / 16, W / 16); The image size output by the third deconvolution unit is (D / 2, H / 2, W / 2); the number is 128; The convolution kernel of the third deconvolution unit has a size of 2×2×2 and a number of 3; The image size output by the fourth deconvolution unit is (D / 4, H / 4, W / 4); the number is 256; The convolution kernel of the fourth deconvolution unit has a size of 2×2×2 and a number of 2; The image size output by the fifth deconvolution unit is (D / 4, H / 4, W / 4); the number is 256; The size of the convolution kernel of the fifth deconvolution unit is 2×2×2; the number is 1.

4. The vascular medical image segmentation method according to claim 1, characterized in that: Training of vascular image segmentation model, including: Obtain medical 3D images of blood vessels; Performing data enhancement processing on the medical three-dimensional image to obtain training samples for a segmentation model; The training samples of the segmentation model are input into the segmentation model, the segmentation model is trained, and the training is stopped when the loss function of the segmentation model is less than a predetermined threshold.

5. The vascular medical image segmentation method according to claim 4, characterized in that: Performing data enhancement processing on the medical three-dimensional image includes: flipping the medical three-dimensional image horizontally; or, The medical three-dimensional image is randomly rotated, and the rotation angle is less than a predetermined angle threshold; or, The medical three-dimensional image is moved horizontally or vertically, and the moving distance is less than a predetermined number of pixel distances.

6. A vascular medical image segmentation device, characterized in that: include: An acquisition module, used for acquiring a medical three-dimensional image of a blood vessel; A processing module, used for inputting the medical three-dimensional image of the blood vessel into a pre-trained blood vessel image segmentation model, segmenting the three-dimensional image, and obtaining a segmented three-dimensional image of the blood vessel; The vascular image segmentation model includes an input layer, a first Transformer neural network unit block, a second Transformer neural network unit block, a third Transformer neural network unit block, a fifth deconvolution unit block, a first convolution unit block, a first deconvolution unit block, a second convolution unit block, a second deconvolution unit block and an output layer, which are connected in sequence.

7. A vascular medical image segmentation device, characterized in that: include: at least one processor and at least one memory; The memory is used to store one or more program instructions; The processor is used to run one or more program instructions to execute the method according to any one of claims 1-5.

Images