Medical image segmentation method, medical image processing device and readable storage medium
By converting medical images from the first development stage to the second development stage, and using pre-trained segmentation models and style migration networks, the error problem in vascular segmentation is solved, and comprehensive and accurate segmentation of medical images is achieved and diagnostics is improved. and the accuracy of treatment.
Patent Information
- Application Number
- CN202310648274.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-01
- Publication Date
- 2024-12-03
AI Technical Summary
The prior art has errors in vascular segmentation in medical images, making it difficult to achieve comprehensive and accurate vascular segmentation, especially the time and positional errors between arteries and veins are difficult to avoid.
By converting the image from the first development stage to the image from the second development stage and performing vascular segmentation using a pre-trained segmentation model, combined with a combination of a style migration network and a segmentation model, the target vascular segmentation results, including venous and arterial vascular vascularization, are obtained by converting the image from the first development stage into images from the second development stage and using a pre-trained segmentation model. .
A comprehensive and accurate vascular segmentation of medical images is achieved, registration errors are reduced, and the accuracy of disease diagnosis and treatment is improved.
Smart Images

Figure CN119068008A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of medical image processing, and in particular, relates to a medical image segmentation method, a medical image processing device and a readable storage medium. Background Art
[0002] In current medical practice, vascular segmentation technology plays a vital role in image analysis and disease diagnosis. It is not only often used as a pre-processing for medical image analysis, but also can be used by doctors to diagnose and formulate treatment plans for patients. Therefore, it is of great significance for the early detection and treatment of vascular diseases.
[0003] Blood vessel segmentation is also one of the hot and difficult problems in the field of medical image processing. For example, in the existing technology, arteries and veins are usually segmented separately. However, due to the separate segmentation, there is a time error between the two, and the patient may have slight movements at different times. Therefore, there is still an error between the two and the actual position of the human body, so alignment is still needed.
[0004] It is difficult to avoid errors during the registration process. Therefore, how to comprehensively and accurately segment blood vessels in medical images is a problem that needs to be solved. Summary of the invention
[0005] The embodiments of the present application provide a medical image segmentation method, a medical image processing device, and a readable storage medium, which can solve the function of performing comprehensive and accurate blood vessel segmentation on medical images.
[0006] In a first aspect, an embodiment of the present application provides a medical image segmentation method, comprising:
[0007] converting an image in a first development phase into an image in a second development phase, the first development phase including one of an arterial phase, a venous phase, and a delayed phase, and the second development phase including another of the arterial phase, the venous phase, and the delayed phase;
[0008] Using a pre-trained first segmentation model belonging to the first development stage, performing blood vessel segmentation on the image of the first development stage to obtain a first blood vessel segmentation result;
[0009] Using a pre-trained second segmentation model belonging to the second development stage, performing blood vessel segmentation on the image of the second development stage to obtain a second blood vessel segmentation result;
[0010] A target blood vessel segmentation result is obtained according to the first blood vessel segmentation result and the second blood vessel segmentation result, wherein the target blood vessel segmentation result includes a venous blood vessel and an arterial blood vessel.
[0011] Optionally, converting the image in the first development stage into the image in the second development stage includes:
[0012] The image at the first development stage is converted into the image at the second development stage using a pre-trained style transfer network, wherein the style transfer network includes at least one of an adversarial neural network, a diffusion model, and a variational autoencoder.
[0013] Optionally, the method further includes:
[0014] Training is performed based on the images of the first development stage and the images of the second development stage of the training object to obtain the pre-trained style transfer network.
[0015] Optionally, the method further includes:
[0016] The training is performed based on the images of the first development stage of the training object and their annotation information to obtain a pre-trained first segmentation model, and the training is performed based on the images of the second development stage of the training object and their annotation information to obtain a pre-trained second segmentation model.
[0017] Optionally, the method further includes:
[0018] Acquire an image of a first development stage of a training object and its annotation information as a first sample;
[0019] Using a pre-trained style transfer network to convert the non-first development stage image of the training object into the first development stage image, and using the converted first development stage image and its annotation information as the third sample;
[0020] Perform training according to the first sample and the third sample to obtain the pre-trained first segmentation model;
[0021] and / or,
[0022] Acquire an image of the second development stage of the training object and its annotation information as a second sample;
[0023] Using a pre-trained style transfer network to convert the non-second development stage image of the training object into the second development stage image, and using the converted second development stage image and its annotation information as a fourth sample;
[0024] Training is performed according to the second sample and the fourth sample to obtain the pre-trained second segmentation model.
[0025] Optionally, the first segmentation result includes one of a single-type blood vessel segmentation result and a multi-type blood vessel segmentation result, and the second segmentation result includes the other of the single-type blood vessel segmentation result and the multi-type blood vessel segmentation result, the single-type blood vessel includes an arterial blood vessel or a venous blood vessel, and the multi-type blood vessel includes an arteriovenous blood vessel, and obtaining the target segmentation result according to the first blood vessel segmentation result and the second blood vessel segmentation result includes:
[0026] Matching different blood vessels from the first blood vessel segmentation result and the second blood vessel segmentation result as an intermediate blood vessel segmentation result, wherein the intermediate blood vessel segmentation result includes a single-type blood vessel segmentation result;
[0027] The target segmentation result is obtained by merging the single-type blood vessel segmentation result in the first blood vessel segmentation result and the second blood vessel segmentation result with the intermediate blood vessel segmentation result.
[0028] Optionally, the method further includes:
[0029] converting the image of the first development stage into an image of a third development stage, the third development stage being different from the first development stage and the second development stage;
[0030] Using a pre-trained third segmentation model belonging to the third development stage, performing blood vessel segmentation on the image of the third development stage to obtain a third blood vessel segmentation result;
[0031] A target blood vessel segmentation result is obtained according to the first blood vessel segmentation result, the second blood vessel segmentation result and the third blood vessel segmentation result.
[0032] Optionally, the first segmentation model includes a first rough segmentation model and a first fine segmentation model, and using the first segmentation model to perform blood vessel segmentation on the image in the first development stage to obtain a first blood vessel segmentation result includes:
[0033] Segmenting the image in the first development stage using a first rough segmentation model to obtain a rough segmentation result, wherein the rough segmentation result includes at least one initial blood vessel endpoint;
[0034] Extracting a sub-image to be segmented according to the blood vessel at the at least one initial blood vessel endpoint;
[0035] Segment the sub-image using a first fine segmentation model to obtain a fine segmentation result, wherein the fine segmentation result includes at least one new blood vessel endpoint;
[0036] updating the at least one initial vessel endpoint using the at least one new vessel endpoint;
[0037] Return to the step of extracting the sub-image to be segmented according to the blood vessel at the initial blood vessel endpoint until a preset end condition is met, and obtain a first blood vessel segmentation result according to the coarse segmentation result and each fine segmentation result.
[0038] Optionally, extracting the sub-image to be segmented according to the blood vessel at the blood vessel endpoint includes:
[0039] Determining the size of the sub-image according to the blood vessel size at the blood vessel endpoint;
[0040] Extracting the sub-image based on the size of the sub-image, the sub-image comprising a two-dimensional slice or a three-dimensional image block;
[0041] and / or
[0042] Determining the direction of the sub-image according to the direction of the blood vessel at the blood vessel endpoint;
[0043] Based on the orientation of the sub-image, the sub-image is extracted.
[0044] In a second aspect, an embodiment of the present application provides a medical image segmentation method, the method comprising:
[0045] training according to the images of the first development stage of the training object and their annotated information to obtain a pre-trained first segmentation model, and training according to the images of the second development stage of the training object and their annotated information to obtain a pre-trained second segmentation model, wherein the first development stage includes one of the arterial phase, the venous phase, and the delayed phase, and the second development stage includes another of the arterial phase, the venous phase, and the delayed phase, the first segmentation model is used to perform blood vessel segmentation on the images belonging to the first development stage, and the second segmentation model is used to perform blood vessel segmentation on the images belonging to the second development stage;
[0046] The training is performed based on the image of the first development stage of the training object and its annotation information, and the pre-trained first segmentation model is obtained, including:
[0047] Acquire an image of a first development stage of a training object and its annotation information as a first sample;
[0048] Using a pre-trained style transfer network to convert the non-first development stage image of the training object into the first development stage image, and using the converted first development stage image and its annotation information as the third sample;
[0049] Perform training according to the first sample and the third sample to obtain the pre-trained first segmentation model;
[0050] The pre-trained second segmentation model is obtained by training according to the image of the second development stage of the training object and its annotation information, including:
[0051] Acquire an image of the second development stage of the training object and its annotation information as a second sample;
[0052] Using a pre-trained style transfer network to convert the non-second development stage image of the training object into the second development stage image, and using the converted second development stage image and its annotation information as a fourth sample;
[0053] Training is performed according to the second sample and the fourth sample to obtain the pre-trained second segmentation model.
[0054] Optionally, the method comprises:
[0055] extracting a first sub-image based on an image of a first development stage of the training object;
[0056] Performing training according to the first sub-image and its annotation information to obtain the pre-trained first segmentation model;
[0057] and / or,
[0058] extracting a second sub-image based on the image of the training object at the second development stage;
[0059] Training is performed according to the second sub-image and its annotation information to obtain the pre-trained second segmentation model.
[0060] In a third aspect, an embodiment of the present application provides a medical image segmentation method, the method comprising:
[0061] Performing blood vessel segmentation on the medical image using a coarse segmentation model to obtain a coarse segmentation result, wherein the coarse segmentation result includes at least one initial blood vessel endpoint;
[0062] Determining the size of the sub-image to be segmented according to the blood vessel size at the at least one blood vessel endpoint;
[0063] Extracting the sub-image based on the size of the sub-image, the sub-image comprising a two-dimensional slice or a three-dimensional image block;
[0064] Segment the sub-image using a fine segmentation model to obtain a fine segmentation result, wherein the fine segmentation result includes at least one new blood vessel endpoint, and update the initial blood vessel endpoint with the new blood vessel endpoint;
[0065] Return to the step of determining the size of the sub-image to be segmented according to the blood vessel size at the at least one blood vessel endpoint until a preset end condition is met, and obtain a blood vessel segmentation result according to the coarse segmentation result and each fine segmentation result.
[0066] In a fourth aspect, an embodiment of the present application provides a medical image processing device, including:
[0067] A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, part or all of the steps of the medical image segmentation method of any one of the first aspects are implemented.
[0068] In the fifth aspect, an embodiment of the present application provides a surgical robot, comprising: a force-bearing mechanism, a motion execution mechanism, a force detection device arranged on the force-bearing mechanism, a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements part or all of the steps of the medical image segmentation method of any one of the first aspects mentioned above.
[0069] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, which may store a program, and when the program is executed by a processor, it implements part or all of the steps of the medical image segmentation method of any one of the above-mentioned first aspects.
[0070] In a seventh aspect, an embodiment of the present application provides a computer program product, which, when executed on a surgical robot, enables the surgical robot to perform part or all of the steps of the medical image segmentation method of any one of the first aspects described above.
[0071] It can be understood that the beneficial effects of the second to seventh aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.
[0072] Compared with the prior art, the embodiments of the present application have the following beneficial effects: by converting an image of a first development stage into an image of a second development stage, the first development stage includes one of an arterial stage, a venous stage and a delayed stage, and the second development stage includes another of the arterial stage, the venous stage and the delayed stage; using a pre-trained first segmentation model belonging to the first development stage to perform vascular segmentation on the image of the first development stage to obtain a first vascular segmentation result; using a pre-trained second segmentation model belonging to the second development stage to perform vascular segmentation on the image of the second development stage to obtain a second vascular segmentation result; according to the first vascular segmentation result and the second vascular segmentation result, a target vascular segmentation result is obtained, the target segmentation result includes venous vessels and arterial vessels, and the target segmentation result includes venous vessels and arterial vessels, thereby achieving comprehensive and accurate vascular segmentation of medical images. BRIEF DESCRIPTION OF THE DRAWINGS
[0073] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0074] Figure 1 is a flowchart of a medical image segmentation method provided by an embodiment of the present application;
[0075] Figure 2is a flowchart of a medical image segmentation method provided by another embodiment of the present application;
[0076] Figure 3 is a flowchart of a medical image segmentation method provided by another embodiment of the present application;
[0077] Figure 4 It is a schematic diagram of the overall structure of a medical image processing device according to an embodiment of the present application. DETAILED DESCRIPTION
[0078] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.
[0079] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.
[0080] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.
[0081] like Figure 1 As shown, a schematic diagram of a flow chart of a medical image segmentation method provided by an embodiment of the present application includes:
[0082] Step S10: converting the image of the first development stage into the image of the second development stage, the first development stage including one of the arterial phase, the venous phase and the delayed phase, and the second development stage including another of the arterial phase, the venous phase and the delayed phase.
[0083] When a patient undergoes a CT scan, contrast agents are injected according to the needs of the examination. After the contrast agents are injected into the blood vessels, the contrast agents flow in the blood vessels, and the blood vessel conditions are observed through the metabolism and development of the contrast agents in the blood vessels. According to the development stage, different periods of CT are divided into: plain scan period, arterial period, venous period and delayed period. For example, when the contrast agent flows to the artery, the artery is developed, which corresponds to the arterial period. When the contrast agent flows to the vein, both the artery and the vein are developed, which corresponds to the venous period; when the contrast agent flows out of the artery, it corresponds to the delayed period. In the embodiment of the present application, the medical images obtained by various imaging techniques in different development stages are referred to as images of a certain development stage, and can also be referred to as images belonging to a certain development stage. For example, the medical images obtained by various imaging techniques in the first development stage are referred to as images of the first development stage, and the medical images obtained by various imaging techniques in the second development stage are referred to as images of the second development stage. Medical images include blood vessels. At the same time, in the embodiment of the present application, the images of a certain development stage obtained after conversion are also referred to as images of the development stage. For example, the images of the second development stage obtained by conversion are also referred to as images of the second development stage. It can be understood that the specific duration of each development stage varies according to different organs. Existing imaging technologies include enhanced CT, ultrasound imaging, etc.
[0084] Step S20: using a pre-trained first segmentation model belonging to the first development stage to perform blood vessel segmentation on the image of the first development stage to obtain a first blood vessel segmentation result.
[0085] The first blood vessel segmentation result includes blood vessels corresponding to the first development stage. For example, if the first development stage is a blood vessel in the arterial phase, the first blood vessel segmentation result includes at least arterial blood vessels, if the first development stage is a blood vessel in the venous phase, the first blood vessel segmentation result includes at least arteriovenous blood vessels, and if the first development stage is a blood vessel in the delayed phase, the first blood vessel segmentation result includes at least venous blood vessels. Arteriovenous blood vessels include both arterial and venous blood vessels. The image of the first development stage may be at least one two-dimensional or three-dimensional image.
[0086] Step S30: using the pre-trained second segmentation model belonging to the first development stage to perform blood vessel segmentation on the image of the second development stage to obtain a second blood vessel segmentation result.
[0087] The second blood vessel segmentation result includes blood vessels corresponding to the second development stage. For example, if the second development stage is a blood vessel in the arterial phase, the second blood vessel segmentation result includes at least arterial blood vessels, if the second development stage is a blood vessel in the venous phase, the second blood vessel segmentation result includes at least arteriovenous blood vessels, and if the second development stage is a blood vessel in the delayed phase, the second blood vessel segmentation result includes at least venous blood vessels. The image of the second development stage may be at least one two-dimensional or three-dimensional image.
[0088] Step S40: Obtain a target blood vessel segmentation result based on the first blood vessel segmentation result and the second blood vessel segmentation result, wherein the target blood vessel segmentation result includes a venous blood vessel and an arterial blood vessel. Since the image of the second development stage is obtained by converting the image of the first development stage, the image of the first development stage and the image of the second development stage correspond to the same time and space coordinate system, and further obtaining the target blood vessel segmentation result based on the first blood vessel segmentation result and the second blood vessel segmentation result obtained by segmenting the image of the first development stage and the image of the second development stage can avoid tedious registration and errors caused by registration, thereby comprehensively and accurately realizing blood vessel segmentation, which is beneficial to the diagnosis and treatment of diseases.
[0089] In one embodiment, step S40 in the medical image segmentation method of the present application obtains a target blood vessel segmentation result according to the first blood vessel segmentation result and the second blood vessel segmentation result, including:
[0090] According to the different blood vessels in the first blood vessel segmentation result and the second blood vessel segmentation result, that is, matching the different blood vessels from the first blood vessel segmentation result and the second blood vessel segmentation result as an intermediate blood vessel segmentation result, wherein the intermediate blood vessel segmentation result includes a single type of blood vessel segmentation result;
[0091] Merging a single-type blood vessel segmentation result in the first blood vessel segmentation result and the second blood vessel segmentation result with the intermediate blood vessel segmentation result to obtain the target segmentation result;
[0092] or,
[0093] The first blood vessel segmentation result and the second blood vessel segmentation result are combined to obtain a target blood vessel segmentation result. Optionally, the combination of the first development stage and the second development stage includes 6 types:
[0094] Arterial phase + venous phase / delayed phase;
[0095] Venous phase + arterial phase / delayed phase;
[0096] Delayed phase + venous phase / arterial phase;
[0097] In the embodiment of the present application, the segmentation results of the segmentation model for the image in the arterial phase include a single type of blood vessels, the segmentation results of the segmentation model for the image in the venous phase include multiple types of blood vessels, and the segmentation results of the segmentation model for the image in the delayed phase include a single type of blood vessels.
[0098] Therefore, when the arterial phase and the delayed phase are not combined, the first segmentation result includes one of a single-type blood vessel and a multi-type blood vessel segmentation result, and the second segmentation result includes the other of a single-type blood vessel and a multi-type blood vessel segmentation result, wherein the single-type blood vessel includes an arterial blood vessel or a venous blood vessel, and the multi-type blood vessel includes an arteriovenous blood vessel. The same blood vessels in the first blood vessel segmentation result and the second blood vessel segmentation result can be marked as the first type, and the blood vessels that are different from the first blood vessel segmentation result and the second blood vessel segmentation result can be marked as the second type. The blood vessels of the single-type blood vessel segmentation result in the first blood vessel segmentation result and the second blood vessel segmentation result are marked as the first type. The first type is one of the arterial blood vessel and the venous blood vessel, and the second type is the other of the arterial blood vessel and the venous blood vessel. Specifically, how to match the first blood vessel segmentation result and the second blood vessel segmentation result to obtain the same or different blood vessels can adopt the existing technology, for example, different blood vessels can be obtained by subtraction processing in image processing, and the embodiment of the present application does not impose any limitation on this.
[0099] If a single type of blood vessel needs to be used alone in an actual application scenario, the single type of blood vessel segmentation result in the first segmentation result and the second segmentation result, or the intermediate segmentation result can be used. If a whole multi-type blood vessel needs to be used in an actual application scenario, the single type of blood vessel segmentation result in the first segmentation result and the second segmentation result, and the intermediate segmentation result are combined to obtain the final multi-type blood vessel segmentation result. In the embodiment of the present application, the blood vessel required in the actual application scenario is referred to as a target segmentation result, so the target segmentation result may include at least one of the single type of blood vessel segmentation result in the first segmentation result and the second segmentation result, and the intermediate segmentation result.
[0100] For example, Figure 2 The flowchart of the medical image segmentation method shown in the figure shows that the first development stage is the arterial stage, and the second development stage is the venous stage. In the first segmentation result and the second segmentation result, the first segmentation result is the segmentation result of a single-type arterial blood vessel, and the second segmentation result includes both arterial blood vessels and venous blood vessels, so it belongs to a multi-type blood vessel segmentation result. The intermediate blood vessel segmentation result is the segmentation result of a single-type venous blood vessel. Therefore, according to the actual application scenario, the target segmentation result may include at least one of the first segmentation result and the intermediate blood vessel segmentation result, i.e., artery, vein, or a combination of artery and vein. Specifically, how to merge multiple segmentation results can adopt existing technology, for example, the merging can be performed by addition processing in image processing, and the embodiment of the present application does not impose any limitation on this.
[0101] When the arterial phase and the delayed phase are combined, the first segmentation result includes a single-type blood vessel segmentation result, and the second segmentation result also includes a single-type blood vessel segmentation result. Further, the first blood vessel segmentation result and the second blood vessel segmentation result are merged to obtain the target blood vessel segmentation result. Exemplarily, the first development stage is the arterial phase, and the second development stage is the delayed phase, then the first segmentation result includes at least arterial blood vessels, and the second segmentation result includes at least veins. If a single-type blood vessel needs to be used alone in an actual application scenario, the first segmentation result or the second segmentation result can be used. If a whole multi-type blood vessel needs to be used in an actual application scenario, the first segmentation result and the second segmentation result can be merged to obtain the final segmentation result. In the embodiment of the present application, the blood vessel required for the actual application scenario is referred to as the target segmentation result, so the target segmentation result may also include at least one of the first segmentation result and the second segmentation result.
[0102] In the embodiment of the present application, the blood vessel segmentation result obtained by using the segmentation model to segment the images in the same development stage is the basis of medical image analysis, image three-dimensional reconstruction and medical image visualization, for example, it can be applied to positioning and navigation in surgical operations.
[0103] In one embodiment, after performing image segmentation according to the image of the first development stage and the image of the second development stage to obtain a segmentation result, if the segmentation effect is still not ideal, for example, when the developer dosage is insufficient or the quality of the medical image obtained in the development stage is not high, the medical image segmentation method provided in the embodiment of the present application may further include:
[0104] Step S50: converting the image of the first development stage into an image of the third development stage, where the third development stage is different from the first development stage and the second development stage, wherein the image of the first development stage is converted into the image of the third development stage using a style transfer network.
[0105] Step S70: Use the segmentation model of the third development stage to perform vascular segmentation on the image of the third development stage to obtain a third vascular segmentation result, the third segmentation model includes one of the arterial phase, venous phase and delayed phase segmentation models, and the third segmentation model is different from the first segmentation model and the second segmentation model.
[0106] Step S90: Obtaining a target blood vessel segmentation result according to the first blood vessel segmentation result, the second blood vessel segmentation result and the third blood vessel segmentation result.
[0107] When the combination of the first development stage and the second development stage does not include the combination of the arterial phase and the delayed phase, that is, the combination of the first development stage and the second development stage includes 4 types:
[0108] Arterial phase + venous phase;
[0109] Venous phase + arterial phase / delayed phase;
[0110] Delayed phase + venous phase.
[0111] Correspondingly, the third development stage includes: a delayed phase or an arterial phase, and the corresponding third blood vessel segmentation result is a single-type blood vessel segmentation result, so it is combined with the single-type segmentation results in the first blood vessel segmentation result and the second blood vessel segmentation result obtained above, and the single-type blood vessel segmentation results of the same type in the intermediate segmentation result to obtain this type of single-type blood vessel segmentation result. Further, according to application needs, this type of single-type blood vessel segmentation result can also be combined with another type of single-type blood vessel segmentation result to obtain a final multi-type blood vessel segmentation result.
[0112] When the combination of the first development stage and the second development stage includes a combination of the arterial phase and the delayed phase, that is, the combination of the first development stage and the second development stage includes two types:
[0113] Arterial phase + delayed phase;
[0114] Delayed phase + arterial phase;
[0115] Correspondingly, the third development stage may include: a venous stage. If the corresponding third blood vessel segmentation result is a multi-type blood vessel segmentation result, then a single-type blood vessel may be obtained according to the first blood vessel segmentation result and any single-type segmentation result of the second blood vessel segmentation result, and then the single-type blood vessel segmentation results of the artery and the vein may be obtained respectively. Furthermore, according to application requirements, the single-type blood vessel segmentation results of the artery and the vein may be merged to obtain the final multi-type blood vessel segmentation result.
[0116] For example, taking the first development stage as the arterial phase, the second development stage as the venous phase, and the third development stage as the delayed phase as an example, the medical image segmentation method provided in the embodiment of the present application includes:
[0117] Converting an arterial phase image into a venous phase image;
[0118] Segment the arterial phase image using a first segmentation model to obtain an arterial vessel segmentation result;
[0119] The converted venous phase image is segmented using the second segmentation model to obtain an arteriovenous blood vessel segmentation result;
[0120] converting the arterial phase image into a delayed phase image;
[0121] The transformed delayed period image is segmented using the third segmentation model to obtain a partial venous blood vessel segmentation result;
[0122] According to the common blood vessels in the artery segmentation results and the artery and vein segmentation results, a partial vein segmentation result is obtained;
[0123] The two partial venous vessel segmentation results are merged to obtain the venous vessel segmentation result.
[0124] Further, according to the needs of the application scenario, the arterial segmentation results and the venous segmentation results are merged to obtain the final blood vessel segmentation result.
[0125] The embodiment of the present application obtains a third blood vessel segmentation result by adding segmentation of the image of the third development stage, and obtains a target segmentation result by combining the first blood vessel segmentation result, the second blood vessel segmentation result and the third blood vessel segmentation result, so that the blood vessel segmentation result required in the application scenario can be obtained more comprehensively and accurately.
[0126] In one embodiment, the first segmentation model in step S20 includes a first rough segmentation model and a first fine segmentation model, and using the first segmentation model to perform blood vessel segmentation on the image in the first development stage to obtain a first blood vessel segmentation result includes:
[0127] The image of the first development stage is segmented using the first rough segmentation model to obtain a rough segmentation result, wherein the rough segmentation result includes at least one initial blood vessel endpoint. Relevant information of the at least one initial blood vessel endpoint, such as a position, a blood vessel size passing through the blood vessel endpoint, and a blood vessel running direction, can be obtained and stored according to the rough segmentation result.
[0128] According to the blood vessel at the at least one initial blood vessel endpoint, a sub-image to be segmented is extracted; when multiple initial blood vessel endpoints are included, some or all of the initial blood vessel endpoints can be selected to extract the corresponding sub-images, and the extraction can be done in parallel or in turn, and the embodiment of the present application does not impose any restrictions on this. Optionally, the sub-image is extracted from the image in the first development stage.
[0129] Segment the sub-image using a first fine segmentation model to obtain a fine segmentation result, wherein the fine segmentation result includes at least one new blood vessel endpoint;
[0130] The at least one new blood vessel endpoint is used to update the at least one initial blood vessel endpoint; when multiple new blood vessel endpoints are included, some or all of the new blood vessel endpoints may be selected to update the original initial blood vessel endpoint.
[0131] Return to execute the step to extract the sub-image to be segmented according to the blood vessel at the at least one initial blood vessel endpoint until the preset end condition is met, and obtain the first blood vessel segmentation result according to the rough segmentation result and each fine segmentation result. Specifically, how to obtain the first blood vessel segmentation result according to the rough segmentation result and each fine segmentation result can adopt the existing technology, such as merging by addition processing in image processing, and the embodiment of the present application does not impose any limitation on this. Optionally, the rough segmentation result and each fine segmentation result can also be weighted first, and then merged to obtain the first blood vessel segmentation result. The weighting coefficients can be the same or different.
[0132] Similarly, the second segmentation model in step S30 may also include a second rough segmentation model and a second fine segmentation model. The second segmentation model is used to perform blood vessel segmentation on the image in the second development stage to obtain a second blood vessel segmentation result, including:
[0133] Segmenting the image in the second development stage using a second rough segmentation model to obtain a rough segmentation result, wherein the rough segmentation result includes at least one initial blood vessel endpoint;
[0134] According to the blood vessel at the at least one initial blood vessel endpoint, the sub-image to be segmented is extracted; when multiple initial blood vessel endpoints are included, some or all of the initial blood vessel endpoints can be selected to extract the corresponding sub-images, and the extraction can be done in parallel or in turn, and there is no restriction on the process between them. The embodiment of the present application does not impose any restriction on this. Optionally, the sub-image is extracted from the image in the second development stage.
[0135] Segment the sub-image using a second fine segmentation model to obtain a fine segmentation result, wherein the fine segmentation result includes at least one new blood vessel endpoint;
[0136] The at least one new blood vessel endpoint is used to update the at least one initial blood vessel endpoint; when multiple new blood vessel endpoints are included, some or all of the new blood vessel endpoints may be selected to update the original initial blood vessel endpoint.
[0137] Return to execute the step to extract the sub-image to be segmented according to the blood vessel at the initial blood vessel endpoint until the preset end condition is met, and obtain the second blood vessel segmentation result according to the rough segmentation result and each fine segmentation result. Specifically, how to obtain the second blood vessel segmentation result according to the rough segmentation result and each fine segmentation result can adopt the existing technology, such as merging by addition processing in image processing, and the embodiment of the present application does not impose any limitation on this. Optionally, the rough segmentation result and each fine segmentation result can also be weighted first, and then merged to obtain the second blood vessel segmentation result. The weighting coefficients can be the same or different.
[0138] Optionally, extracting a sub-image to be segmented according to the blood vessel at the at least one initial blood vessel endpoint includes:
[0139] Determining the size of the sub-image to be segmented according to the size of the blood vessel at the blood vessel endpoint;
[0140] Extracting the sub-image based on the size of the sub-image, the sub-image comprising a two-dimensional slice or a three-dimensional image block;
[0141] and / or
[0142] Determining the direction of the sub-image according to the direction of the blood vessel at the blood vessel endpoint;
[0143] Based on the orientation of the sub-image, the sub-image is extracted.
[0144] Furthermore, the direction of the sub-image forms a preset angle with the direction of the blood vessel, and the size of the sub-image is associated with the blood vessel size at the blood vessel endpoint. For example, the preset angle is 90°, that is, the sub-image is extracted along a direction perpendicular to the blood vessel direction at the blood vessel endpoint; the preset angle is 0°, that is, the sub-image is extracted along a direction parallel to the blood vessel direction at the blood vessel endpoint. Optionally, for a certain blood vessel endpoint, sub-images can be extracted according to the directions of multiple different sub-images, and then each sub-image is segmented using a fine segmentation model, and then the different fine segmentation results and the new blood vessel endpoint in the fine segmentation results are merged. Among them, the blood vessel direction at the endpoint can be the normal direction of the blood vessel centerline at the endpoint.
[0145] In the embodiment of the present application, the size of the sub-image to be segmented at each endpoint is determined according to the size of the blood vessel at each endpoint. The blood vessel size may include the blood vessel diameter or the cross-sectional area.
[0146] Optionally, the size of the sub-image is proportional to the size of the blood vessel, for example, patch = k*R, where k is the proportionality coefficient, R is the blood vessel diameter, and patch is the size parameter of the sub-image. Usually, the sub-image can be selected as a cube or a sphere, and patch is the edge length or radius of the image block. Or S1 = M*S2, where M is the proportionality coefficient, S2 is the cross-sectional area of the blood vessel at the blood vessel endpoint, and S1 is the cross-sectional area of the sub-image. The proportionality coefficient can be flexibly adjusted according to actual needs, for example, selected as 3 or 5. Optionally, the size of the corresponding sub-image is determined according to the range to which the blood vessel size at the blood vessel endpoint belongs. Optionally, for a certain blood vessel endpoint, sub-images can be extracted according to the sizes of multiple different sub-images, and then each sub-image can be segmented using a fine segmentation model, and then the different fine segmentation results and the new blood vessel endpoint in the fine segmentation results are merged.
[0147] The range of blood vessel sizes and the corresponding sub-image sizes can be adjusted according to the segmentation requirements and the required computing resources. For example, when the segmentation accuracy is high and the computing resources are sufficient, different sub-image size options can be added, and the range of blood vessel sizes can be more refined.
[0148] The blood vessel endpoint belongs to the sub-image. Optionally, the blood vessel endpoint is located on the outer surface of the image block, or at a certain point inside, such as the center point or a certain point at a certain distance from the center point. It is best if the sub-image includes more unsegmented areas, so as to obtain more segmentation results more efficiently.
[0149] Among them, the preset end condition may include that the blood vessel size at the new blood vessel endpoint is less than a preset threshold, or the number of blood vessels in the segmentation result is less than a preset threshold or the blood vessel size is less than a preset threshold, wherein each threshold can be flexibly adjusted according to the situation of the segmented organ.
[0150] The embodiment of the present application obtains more blood vessel endpoints, and determines the size of the sub-image to be segmented and the corresponding fine segmentation model according to the blood vessel size at the blood vessel endpoint, thereby combining rough segmentation with multiple rounds of fine segmentation to achieve fine and accurate segmentation of blood vessels, while saving computing resources. Since the segmentation results for each blood vessel endpoint can be merged, there is no limit on the order of segmentation of each blood vessel endpoint.
[0151] In one embodiment, the step S10 of converting the image in the first development stage into the image in the second development stage includes:
[0152] Use a pre-trained style transfer network to convert images from the first development stage to images from the second development stage.
[0153] Among them, the style transfer network can also be called a conversion model. The conversion model can include at least one of an adversarial neural network, a diffusion model, and a variational autoencoder. The adversarial neural network (Generative Adversarial Networks), referred to as GAN, includes a generator and a discriminator. The more commonly used adversarial neural networks include CycleGAN, StyleGAN, BigGAN or StackGAN.
[0154] Furthermore, the medical image segmentation method provided in the embodiment of the present application also includes: pre-training a style transfer network.
[0155] A large number of training samples are required for training the style transfer network. The training samples of the style transfer network include images of the first development stage and images of non-first development stages of the training object. The pre-trained style transfer network is obtained by training based on the images of the first development stage and images of non-first development stages of the training object. The non-first development stage may be the second development stage or the third development stage, etc. For example, if the training samples of the style transfer network include images of the first development stage and images of the second development stage of the training object, then the training is performed based on the images of the first development stage and images of the second development stage of the training object to obtain a style transfer network that can convert images of the first development stage into images of the second development stage; if the training samples of the style transfer network include images of the first development stage and images of the third development stage of the training object, then the training is performed based on the images of the first development stage and images of the third development stage of the training object to obtain a style transfer network that can convert images of the first development stage into images of the third development stage.
[0156] Furthermore, in the embodiment of the present application, since the first development stage can select one of the arterial phase, the venous phase and the delayed phase, and the second development stage can select another of the arterial phase, the venous phase and the delayed phase, the style transfer network can include at least the following 6 types:
[0157] converting an arterial phase image into a venous phase image or a delayed phase image; or
[0158] converting the venous phase image into an arterial phase image or a delayed phase image; or
[0159] The delayed phase image is converted into a venous phase image and an arterial phase image.
[0160] When the selection of each development stage is uncertain, in order to efficiently and comprehensively complete the above 6 conversions, the embodiment of the present application can pre-train the above 6 different functional networks. It can be understood that if the first development stage and the second development stage have clear specific selections and no changes, the pre-training of some functional networks can be omitted, and only the network corresponding to the conversion of the first development stage to the second development stage can be trained. For example, if the first development stage is selected as the arterial stage and the second development stage is selected as the venous stage, then only the style transfer network that converts the image of the arterial stage to the image of the venous stage can be pre-trained.
[0161] Optionally, for different style transfer networks, due to different technologies, the same trained style transfer network may contain multiple conversion functions. For example, when the style transfer network is CycleGAN, CycleGAN can perform at least two conversion functions. For example, if the arterial phase is selected for the first development stage and the venous phase is selected for the second development stage, then the training samples include images of the arterial phase and images of the venous phase. The trained CycleGAN can realize the two functions of converting images of the first development stage into images of the second development stage, and converting images of the second development stage into images of the first development stage.
[0162] Optionally, the training samples of the style transfer network include not only images of the first development stage and images of non-first development stages, but also annotation information, and the annotation information includes a feature description corresponding to the conversion of the image of the first development stage to the image of the non-first development stage. When the style transfer network is cycleGAN, the training samples include not only images of the first development stage and images of non-first development stages, but also annotation information, and the annotation information includes a feature description corresponding to the conversion of the image of the first development stage to the image of the non-first development stage, and a feature description corresponding to the conversion of the image of the non-first development stage to the image of the first development stage. The annotation can be performed manually or automatically, the annotation information can be an image, and the specific annotation technology can adopt the existing technology, and the present application does not impose any limitation on this.
[0163] Among them, manual annotation is to manually mark some medical images, such as marking in red, drawing circles, drawing dots, etc., to mark the segmentation target. Taking blood vessel segmentation as an example, in lung CT images, the location of blood vessels is manually marked. Blood vessels generally have hole features. Manual annotation methods include selecting a point in the blood vessel, drawing a circle / box or other patterns in the blood vessel, and framing the blood vessel with a broken line / curve.
[0164] Since the manually given annotations only indicate the location of the blood vessels and often cannot accurately give the range of each blood vessel, the medical image is segmented according to the manually labeled results to obtain a labeled image. The segmentation refers to the binary segmentation of the medical image, and the segmentation result can be represented by a binary image (i.e., the labeled image), which has the same size as the medical image, and the pixel value of each pixel is used to indicate whether the corresponding pixel in the medical image belongs to the segmentation target. If it does, the pixel is called the foreground, and if it does not, the pixel is called the background. In the embodiments of the present application, the foreground pixel value is 1 and the background pixel value is 0. In fact, in other embodiments, the values can also be opposite, for example, the foreground pixel value is 0 and the background pixel value is 1, and the present application does not limit this.
[0165] Furthermore, when displaying, since the pixel value of a grayscale image generally ranges from 0 to 255, rather than simply 0 and 1, in order to be able to intuitively see the annotated image in the image, the pixel points with a pixel value of 1 are generally reassigned to 255 when displayed. Among them, for the purpose of clear display, the white part is the effect of reassigning the pixel points whose actual pixel value is 1 to 255. It should be noted that the actual value of the pixel value of the annotated image in this method is still 0 or 1. In the present application, when displaying a binary image, 1 corresponds to white and 0 corresponds to black; in actual applications, it can also be reversed, that is, 1 corresponds to black and 0 corresponds to white.
[0166] When the development stage is different from the above division, the training conversion model can be combined with the adaptability of the specific development stage.
[0167] In the embodiment of the present application, the images of the first development stage and the images of the non-first development stage are trained to obtain a pre-trained style transfer network, so that the images of the first development stage can be converted into images of the non-first development stage by using the style transfer network, thereby realizing blood vessel segmentation in the same development stage.
[0168] In one embodiment, the medical image segmentation method provided in the embodiment of the present application further includes:
[0169] Pre-training the first segmentation model. The first segmentation model requires a large number of training samples for training. Pre-training the first segmentation model includes:
[0170] Acquire an image of a first development stage of a training object and its annotation information as a first sample;
[0171] Training is performed according to the first sample to obtain the pre-trained first segmentation model belonging to the first development stage.
[0172] Pre-train the second segmentation model. The second segmentation model requires a large number of training samples for training. Pre-training the second segmentation model includes:
[0173] Acquire an image of the second development stage of the training object and its annotation information as a second sample;
[0174] Training is performed according to the second sample to obtain the pre-trained second segmentation model belonging to the second development stage.
[0175] Pre-train the third segmentation model. The third segmentation model requires a large number of training samples for training. Pre-training the third segmentation model includes:
[0176] Acquire an image of the third development stage of the training object and its annotation information as a fifth sample;
[0177] Training is performed according to the fifth sample to obtain the pre-trained third segmentation model belonging to the first development stage.
[0178] Furthermore, when training the segmentation model, images of the current development stage obtained by converting images of non-current development stages using a style transfer network may also be included, for example:
[0179] Pre-train the first segmentation model, including:
[0180] Acquire an image of a first development stage of a training object and its annotation information as a first sample;
[0181] The pre-trained style transfer network is used to convert the non-first development stage image of the training object into the first development stage image, and the converted first development stage image and its annotation information are used as the third sample; wherein the non-first development stage image may include the second and / or third development stage image. The converted first development stage image can also be understood as an image of the first development stage style, which has the same style as the original real first development stage image, and therefore can also be regarded as an image belonging to the first development stage.
[0182] Training is performed based on the first sample and the third sample to obtain the pre-trained first segmentation model belonging to the first development stage.
[0183] The embodiment of the present application expands the training samples by using the first sample and the third sample as training samples for training, so that a first segmentation model with better segmentation effect can be obtained.
[0184] Optionally, a pre-trained second segmentation model includes:
[0185] Acquire an image of the second development stage of the training object and its annotation information as a second sample;
[0186] Using a pre-trained style transfer network to convert the non-second development stage image of the training object into the second development stage image, and using the converted second development stage image and its annotation information as a fourth sample;
[0187] Training is performed based on the second sample and the fourth sample to obtain the pre-trained second segmentation model belonging to the second development stage.
[0188] Optionally, a pre-trained third segmentation model includes:
[0189] Acquire an image of the second development stage of the training object and its annotation information as a fifth sample;
[0190] Using a pre-trained style transfer network to convert the non-second development stage image of the training object into the second development stage image, and using the converted second development stage image and its annotation information as the sixth sample;
[0191] Training is performed according to the fifth sample and the sixth sample to obtain the pre-trained second segmentation model belonging to the second development stage.
[0192] Furthermore, since the first development stage can select one of the arterial phase, the venous phase and the delayed phase, the first segmentation model can include at least the following three types:
[0193] Use the first segmentation model to perform blood vessel segmentation on the image of the arterial phase to obtain arterial vessels; or,
[0194] The first segmentation model is used to perform blood vessel segmentation on the image in the venous phase to obtain arteriovenous blood vessels; or the first segmentation model is used to perform blood vessel segmentation on the image in the delayed phase to obtain venous blood vessels.
[0195] Similarly, the second segmentation model and the third segmentation model also include three types. As in the training of the aforementioned style transfer network, in order to efficiently and comprehensively complete the above three possible segmentations, the embodiment of the present application can pre-train the above three different functional segmentation models. It can be understood that the pre-training of the segmentation models of some functions can also be omitted. When the development stage is different from the above division, the training segmentation model can be combined with the specific development stage adaptability.
[0196] Exemplarily, corresponding to the actual application scenario, the first segmentation model belongs to one of the above three different segmentation model situations. For example, if the first development stage is selected as the arterial phase, the function of the first segmentation model is to segment the image of the arterial phase to obtain arterial blood vessels. Therefore, the training samples of the first segmentation model required for this actual application scenario include images of the arterial phase.
[0197] Optionally, the training samples of the first segmentation model include not only the training samples but also the annotation information. Exemplarily, pre-training the first segmentation model includes:
[0198] The image of the second development stage of the training object is converted into the image of the first development stage, and the conversion can be completed by using a style transfer network;
[0199] Training is performed based on the converted image of the first development stage, the image of the first development stage of the training object, and the annotation information to obtain a first segmentation model.
[0200] The labeling can be performed manually or automatically, and the labeling information can be an image. The labeling information is used to describe information such as the location or type of the blood vessel in the segmentation result. The specific labeling process can adopt existing technologies, and this application does not impose any limitation on this.
[0201] Further, extracting a first sub-image based on the image of the first development stage of the training object;
[0202] Performing training according to the first sub-image and its annotation information to obtain the pre-trained first segmentation model;
[0203] and / or,
[0204] extracting a second sub-image based on the image of the training object at the second development stage;
[0205] Training is performed according to the second sub-image and its annotation information to obtain the pre-trained second segmentation model.
[0206] In the embodiment of the present application, the first segmentation model is obtained by training with images from the first development stage, and the second segmentation model is obtained by training with images from the second development stage, so that the segmentation model can be used to segment and obtain blood vessel segmentation results in the same development stage.
[0207] In one embodiment, a medical image segmentation method is also provided, comprising:
[0208] The training is performed based on the images of the first development stage of the training object and their annotation information to obtain a pre-trained first segmentation model, and the training is performed based on the images of the second development stage of the training object and their annotation information to obtain a pre-trained second segmentation model, the first development stage includes one of the arterial phase, the venous phase and the delayed phase, the second development stage includes another of the arterial phase, the venous phase and the delayed phase, the first segmentation model is used to perform vascular segmentation on the images of the first development stage, and the second segmentation model is used to perform vascular segmentation on the images of the second development stage.
[0209] Optionally, the medical image segmentation method further includes:
[0210] Acquire an image of a first development stage of a training object and its annotation information as a first sample;
[0211] Using a pre-trained style transfer network to convert the non-first development stage image of the training object into the first development stage image, and using the converted first development stage image and its annotation information as the third sample;
[0212] Perform training according to the first sample and the third sample to obtain the pre-trained first segmentation model;
[0213] and / or,
[0214] Acquire an image of the second development stage of the training object and its annotation information as a second sample;
[0215] Using a pre-trained style transfer network to convert the non-second development stage image of the training object into the second development stage image, and using the converted second development stage image and its annotation information as a fourth sample;
[0216] Training is performed according to the second sample and the fourth sample to obtain the pre-trained second segmentation model.
[0217] Optionally, the medical image segmentation method further includes:
[0218] Acquire an image of the third development stage of the training object and its annotation information as a fifth sample;
[0219] Using a pre-trained style transfer network to convert the non-third development stage image of the training object into the third development stage image, and using the converted third development stage image and its annotation information as the sixth sample;
[0220] Training is performed according to the fifth sample and the sixth sample to obtain the pre-trained third segmentation model.
[0221] The specific training process of the first segmentation model, the second segmentation model, the third segmentation model and the style transfer network is the same as the above principles. For the sake of simplicity, they will not be described here one by one.
[0222] In the embodiment of the present application, a first segmentation model is obtained by training at least using images of the first development stage, and a second segmentation model is obtained by training using images of the second development stage, so that a blood vessel segmentation result can be obtained by using the segmentation model in the same development stage.
[0223] The medical image segmentation method provided in the embodiment of the present application can be applied to multiple organs of the human body, such as liver, lungs, eyes, heart, kidneys and bladder, etc. For example, taking kidneys and bladder as examples, the first development stage selects the arterial phase, and the second development stage selects the delayed phase. The medical image segmentation method includes:
[0224] converting the arterial phase image into a delayed phase image;
[0225] Using the first segmentation model to perform blood vessel segmentation on the image in the arterial phase to obtain a first blood vessel segmentation result, wherein the first blood vessel segmentation result at least includes an arterial blood vessel segmentation result;
[0226] Using the second segmentation model to perform blood vessel segmentation on the delayed period image to obtain a second blood vessel segmentation result, wherein the second blood vessel segmentation result at least includes a venous blood vessel segmentation result;
[0227] According to the application scenario, if a single type of blood vessel segmentation result is required, the arterial blood vessel segmentation result or the venous blood vessel segmentation result can be used alone. If multiple types of blood vessel segmentation results are required, the arterial blood vessel segmentation result and the venous blood vessel segmentation result can be combined. The first blood vessel segmentation result can reflect the kidney, and the second blood vessel segmentation result can reflect the bladder. The relative position relationship between the kidney and the bladder can be seen by combining the first blood vessel segmentation result and the second blood vessel segmentation result.
[0228] The embodiment of the present application achieves vascular segmentation of the kidney and bladder by using the first segmentation model to perform vascular segmentation on the image in the arterial phase and the second segmentation model to perform vascular segmentation on the image in the delayed phase, thereby facilitating the doctor's diagnosis and treatment.
[0229] In one embodiment, if Figure 3 FIG. 1 is a flow chart of a medical image segmentation method provided by another embodiment of the present application, comprising:
[0230] Step S10′: Use the rough segmentation model to perform blood vessel segmentation on the medical image to obtain a rough segmentation result, which includes at least one initial blood vessel endpoint. The rough segmentation model can be used to obtain thick blood vessels in the medical image, but the segmentation effect of some thin blood vessels is not ideal. The rough segmentation result obtained by the rough segmentation may include multiple blood vessel endpoints, which may not be the real blood vessel ends, so it is necessary to continue fine segmentation.
[0231] Step S20′: Determine the size of the sub-image to be segmented according to the vessel size at the at least one initial vessel endpoint. The rough segmentation result may include multiple vessel endpoints, and the sizes of the vessels at these different endpoints may be different. The embodiment of the present application determines the size of the sub-image to be segmented at each endpoint according to the vessel size at each endpoint. The vessel size may include the vessel diameter or cross-sectional area.
[0232] Optionally, the size of the sub-image is proportional to the size of the blood vessel, for example, patch=k*R, where k is a proportionality factor, R is a blood vessel diameter, and patch is a size parameter of the image block. Typically, the sub-image can be selected as a cube or a sphere, and patch is the edge length or radius of the image block.
[0233] Optionally, the size of the corresponding sub-image is determined according to the range to which the blood vessel size at the blood vessel endpoint belongs, for example
[0234] The vessel size belongs to (A, B), corresponding to the sub-image of the first size;
[0235] The vessel size belongs to (C, D), corresponding to the sub-image of the second size;
[0236] The vessel size belongs to (E, F), corresponding to the sub-image of the third size.
[0237] Exemplarily, the vessel size belongs to (1 mm, 2.9 mm), corresponding to a sub-image of 64*64*64 pixels;
[0238] The vessel size belongs to (3mm, 4.9mm), corresponding to a sub-image of 96*96*96 pixels;
[0239] The vessel size is (5mm, 5.9mm), corresponding to a sub-image of 128*128*128 pixels.
[0240] The range of blood vessel sizes and the corresponding sub-image sizes can be adjusted according to the segmentation requirements and the required computing resources. For example, when the segmentation accuracy is high and the computing resources are sufficient, different sub-image size options can be added, and the range of blood vessel sizes can be more refined.
[0241] Step S30′: extract the sub-image based on the size of the sub-image, wherein the sub-image includes a 2D image or a 3D image, wherein the 2D image can be called a slice and the 3D image can be called an image block. The vascular endpoint belongs to the sub-image, and optionally, the vascular endpoint is located on the outer surface of the image block, or at a certain point inside, such as the center point or a certain point at a certain distance from the center point. It is best if the sub-image includes more unsegmented areas, so as to obtain more segmentation results more efficiently.
[0242] Furthermore, extracting the sub-image based on the size of the sub-image may also include:
[0243] Obtain the direction of the blood vessel at the blood vessel endpoint;
[0244] Based on the size of the sub-image, a sub-image is extracted along a direction forming a preset angle with the direction of the blood vessel, and the size of the sub-image is associated with the blood vessel size at the blood vessel endpoint.
[0245] Optionally, the preset angle is 90°, i.e., the sub-image is extracted along a direction perpendicular to the direction of the blood vessel at the blood vessel endpoint; the preset angle is 0°, i.e., the sub-image is extracted along a direction parallel to the direction of the blood vessel at the blood vessel endpoint.
[0246] When multiple initial blood vessel endpoints are included, some or all of the initial blood vessel endpoints can be selected to extract their corresponding sub-images, and the extraction can be done in parallel or in turn. The embodiment of the present application does not impose any limitation on this.
[0247] Step S40′: Use the fine segmentation model to segment the sub-image to obtain a fine segmentation result, which includes at least one new vascular endpoint, and updates the at least one initial vascular endpoint with the new vascular endpoint. When multiple new vascular endpoints are included, some or all of the new vascular endpoints can be selected to update the original initial vascular endpoints. The fine segmentation model needs to be pre-trained and can include multiple fine segmentation models. The size of the fine segmentation model is associated with the size of the sub-image. Exemplarily, the size of the training sample of the fine segmentation model is the same as the size of the sub-image. The training sample size of the fine segmentation model is 64*64*64 pixels, and the size of the sub-image is also 64*64*64 pixels; or the size of the corresponding fine segmentation model is determined according to the range to which the sub-image size belongs, for example:
[0248] The sub-image belongs to (first size, second size), corresponding to the fine segmentation model of size G;
[0249] The sub-image belongs to (second size, third size), corresponding to the H-size fine segmentation model.
[0250] Step S50′: returning to the above step to determine the size of the sub-image to be segmented according to the blood vessel size at the at least one initial blood vessel endpoint, until a preset end condition is met, and obtaining a first blood vessel segmentation result according to the coarse segmentation result and each fine segmentation result.
[0251] After segmentation using the refined segmentation model, more new blood vessels will be segmented, so new blood vessel endpoints will be generated accordingly, and the at least one initial blood vessel endpoint can be updated with the new blood vessel endpoint, that is, the new blood vessel endpoint is used as the basis for the next round of segmentation, and step S20' is returned to be executed. It can be understood that when there are multiple blood vessel endpoints, they can be processed in parallel or in turn, and this application does not impose any limitation on this.
[0252] Optionally, for a certain vascular endpoint, sub-images can be extracted according to the sizes of multiple different sub-images, and then each sub-image is segmented using a fine segmentation model, and then each different fine segmentation result and the new vascular endpoint in the fine segmentation result are merged. Optionally, for a certain vascular endpoint, sub-images can be extracted according to the directions of multiple different sub-images, and then each sub-image is segmented using a fine segmentation model, and then each different fine segmentation result and the new vascular endpoint in the fine segmentation result are merged. The embodiment of the present application uses multiple rounds of cycles to continuously segment using new endpoints in the fine segmentation results and merge each fine segmentation result, so that the vascular segmentation can be completed more comprehensively and accurately.
[0253] Among them, the preset end condition may include that the blood vessel size at the new blood vessel endpoint is less than a preset threshold, or the number of blood vessels in the segmentation result is less than a preset threshold or the blood vessel size is less than a preset threshold, wherein each threshold can be flexibly adjusted according to the situation of the segmented organ.
[0254] Wherein, obtaining a first blood vessel segmentation result according to the rough segmentation result and each fine segmentation result includes:
[0255] The rough segmentation result and each fine segmentation result may be weighted first, and then combined to obtain the first blood vessel segmentation result. The weighting coefficients may be the same or different.
[0256] The embodiment of the present application determines the size of the sub-image to be segmented and the corresponding fine segmentation model according to the blood vessel size at the blood vessel endpoint, thereby achieving fine and accurate segmentation of the blood vessels by combining coarse segmentation with multiple rounds of fine segmentation, while saving computing resources.
[0257] In one embodiment, the medical image segmentation method further includes pre-training a fine segmentation model, specifically including:
[0258] Extracting a sub-image according to the size of the blood vessel at the pixel point in the medical image, wherein the size of the sub-image is associated with the size of the blood vessel at the pixel point;
[0259] Performing training according to the sub-images and their annotation information to obtain a pre-trained fine segmentation model;
[0260] and / or
[0261] Determining the direction of the sub-image according to the direction of blood vessels at the pixel points in the medical image;
[0262] Based on the orientation of the sub-image, the sub-image is extracted.
[0263] Training is performed based on the sub-images and their annotation information to obtain a pre-trained fine segmentation model.
[0264] In order to achieve the targeted segmentation effect of the fine segmentation model, for example, suitable for segmenting sub-images of a certain size range, it is also necessary to select when obtaining training samples. For example, large-sized sub-images are extracted from thick blood vessel areas for training, and small-sized images are extracted from thin blood vessel areas for training. In this way, multiple fine segmentation models are obtained by training separately, and each fine segmentation model is suitable for segmenting sub-images of different size ranges and obtaining blood vessels of different thicknesses. The size of the sub-image is the same as the above principle, and will not be repeated here.
[0265] Exemplarily, the pre-trained fine segmentation model includes:
[0266] A plurality of sub-images are extracted according to the size of blood vessels at the pixel points in the image of the first development stage. The size of the sub-image is associated with the size of blood vessels at the pixel points. The pixel points may include a plurality of sub-images. For example, the size of the sub-image is 64*64*64 pixels.
[0267] Training is performed based on the multiple sub-images of 64*64*64 pixels to obtain a pre-trained fine segmentation model.
[0268] The image of the first development stage may include the image of the first development stage obtained according to the imaging technology, and may also include the image of the first development stage obtained by conversion.
[0269] Alternatively, the pre-trained fine segmentation model may also include:
[0270] Multiple sub-images are extracted according to the blood vessel size at the pixel point in the image of the second development stage. The size of the sub-image is associated with the blood vessel size at the pixel point. The pixel point may include multiple sub-images. For example, the size of the sub-image is 64*64*64 pixels.
[0271] Training is performed based on the multiple sub-images of 64*64*64 pixels to obtain a pre-trained fine segmentation model.
[0272] The image of the second development stage may include the image of the second development stage obtained according to the imaging technology, and may also include the image of the second development stage obtained by conversion.
[0273] Alternatively, pre-trained fine-segmentation models include:
[0274] Determining the direction of the sub-image according to the direction of the blood vessel at the pixel point in the image of the first development stage, wherein the pixel point may include a plurality of pixels;
[0275] The sub-image is extracted based on the direction of the sub-image. For example, the direction of the sub-image is at a 90° angle to the direction in which the blood vessel runs, that is, a direction perpendicular to the direction in which the blood vessel runs.
[0276] The training is performed based on a plurality of sub-images extracted in a direction perpendicular to the direction of the blood vessel, thereby obtaining a pre-trained fine segmentation model.
[0277] The image of the first development stage may include the image of the first development stage obtained according to the imaging technology, and may also include the image of the first development stage obtained by conversion.
[0278] Alternatively, pre-trained fine-segmentation models include:
[0279] The size of the sub-image is determined according to the size of the blood vessel at the pixel point in the image of the second development stage. For example, the sub-image is a two-dimensional slice, the size of the sub-image is 32*32 pixels, and the cross-sectional area of the sub-image is 3 times the cross-sectional area of the blood vessel at the pixel point.
[0280] And according to the direction of the blood vessel at the pixel point in the image of the second development stage, the direction of the sub-image is determined, for example, the direction of the sub-image is at a 90° angle to the direction of the blood vessel, that is, the direction of the sub-image is perpendicular to the direction of the blood vessel.
[0281] The sub-image is extracted based on a size of 32*32 pixels and in a direction perpendicular to the direction of the blood vessel.
[0282] The training is performed based on a plurality of sub-images extracted in a direction perpendicular to the direction of the blood vessel, thereby obtaining a pre-trained fine segmentation model.
[0283] The image of the second development stage may include the image of the second development stage obtained according to the imaging technology, and may also include the image of the second development stage obtained by conversion.
[0284] Furthermore, it also includes pre-training a rough segmentation model. Similarly, it can be trained according to the images of the first development stage, the images of the second development stage, and the images of the third development stage, respectively, to obtain a rough segmentation model suitable for each development stage. The rough segmentation model can adopt the network architecture in the prior art, and the embodiments of the present application do not impose any limitation on this.
[0285] Optionally, the process of obtaining the annotation information is the same as the above principle and will not be repeated here.
[0286] The embodiment of the present application obtains multiple fine segmentation models by extracting sub-images of associated sizes according to the size of the blood vessels, and training the sub-images of different sizes as training samples respectively, thereby facilitating the fine segmentation model to segment blood vessels of different thicknesses, thereby achieving comprehensive and accurate blood vessel segmentation.
[0287] Figure 4 4 is a schematic diagram of a medical image processing device 4 provided in an embodiment of the present invention, comprising: a processor 40, a memory 41, and a computer program 42 stored in the memory 41 and executable on the processor 40. When the processor 40 executes the computer program 42, the steps in the above-mentioned medical image segmentation method embodiment are implemented, for example, Figure 1 Steps S10 to S40 are shown.
[0288] Exemplarily, the computer program 42 may be divided into one or more modules / units, one or more modules / units are stored in the memory 41, and executed by the processor 40 to implement the present invention. One or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program 42 in the medical image processing device 4.
[0289] The medical image processing device 4 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The medical image processing device 4 may include, but is not limited to, a processor 40 and a memory 41. Those skilled in the art will appreciate that Figure 4 It is only an example of the medical image processing device 4 and does not constitute a limitation of the medical image processing device 4. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the medical image processing device 4 may also include input and output devices, network access devices, buses, etc.
[0290] The processor 40 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.
[0291] The memory 41 may be an internal storage unit of the medical image processing device 4, such as a hard disk or memory of the medical image processing device 4. The memory 41 may also be an external storage device of the medical image processing device 4, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the medical image processing device 4. Further, the memory 41 may also include both an internal storage unit and an external storage device of the medical image processing device 4. The memory 41 is used to store computer programs and other programs and data required by the medical image processing device 4. The memory 41 may also be used to temporarily store data that has been output or is to be output.
[0292] An embodiment of the present application provides a surgical robot, comprising: a main operating console, a slave operating device; and a medical image processing device, which is coupled to the main operating console and the slave operating device and is configured to execute the above-mentioned medical image segmentation method.
[0293] Exemplarily, the computer program can be divided into one or more modules / units, one or more modules / units are stored in a memory and executed by a processor to complete the present invention. One or more modules / units can be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the surgical robot.
[0294] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.
[0295] The memory may be an internal storage unit of the surgical robot, such as a hard disk or memory of the surgical robot. The memory may also be an external storage device of the surgical robot, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the surgical robot. Furthermore, the memory may include both an internal storage unit and an external storage device of the surgical robot. The memory is used to store computer programs and other programs and data required by the terminal device. The memory may also be used to temporarily store data that has been output or is to be output.
[0296] The embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.
[0297] An embodiment of the present application provides a computer program product. When the computer program product runs on a mobile terminal, the mobile terminal can implement the steps in the above-mentioned method embodiments when executing the computer program product.
[0298] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.
[0299] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0300] As used in the specification of this application and the appended claims, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting" depending on the context. Similarly, the phrase "if it is determined" or "if the described condition or event is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce the described condition or event is detected" or "in response to detecting the described condition or event" depending on the context.
[0301] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.
[0302] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.
[0303] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0304] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0305] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0306] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0307] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.
[0308] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.
Claims
1. A medical image segmentation method, characterized in that: The method comprises: converting an image in a first development phase into an image in a second development phase, the first development phase including one of an arterial phase, a venous phase, and a delayed phase, and the second development phase including another of the arterial phase, the venous phase, and the delayed phase; Using a pre-trained first segmentation model belonging to the first development stage, performing blood vessel segmentation on the image of the first development stage to obtain a first blood vessel segmentation result; Using a pre-trained second segmentation model belonging to the second development stage, performing blood vessel segmentation on the image of the second development stage to obtain a second blood vessel segmentation result; A target blood vessel segmentation result is obtained according to the first blood vessel segmentation result and the second blood vessel segmentation result, wherein the target blood vessel segmentation result includes a venous blood vessel and an arterial blood vessel.
2. The medical image segmentation method according to claim 1, characterized in that: The converting the image of the first development stage into the image of the second development stage comprises: The image at the first development stage is converted into the image at the second development stage using a pre-trained style transfer network, wherein the style transfer network includes at least one of an adversarial neural network, a diffusion model, and a variational autoencoder.
3. The medical image segmentation method according to claim 2, characterized in that: The method further comprises: Training is performed based on the images of the first development stage and the images of the second development stage of the training object to obtain the pre-trained style transfer network.
4. The medical image segmentation method according to claim 1, characterized in that: The method further comprises: The training is performed based on the images of the first development stage of the training object to obtain a pre-trained first segmentation model, and the training is performed based on the images of the second development stage of the training object to obtain a pre-trained second segmentation model.
5. The medical image segmentation method according to claim 1, characterized in that: The method further comprises: Acquire an image of a first development stage of a training object and its annotation information as a first sample; Using a pre-trained style transfer network to convert the non-first development stage image of the training object into the first development stage image, and using the converted first development stage image and its annotation information as the third sample; Perform training according to the first sample and the third sample to obtain the pre-trained first segmentation model; and / or, Acquire an image of the second development stage of the training object and its annotation information as a second sample; Using a pre-trained style transfer network to convert the non-second development stage image of the training object into the second development stage image, and using the converted second development stage image and its annotation information as a fourth sample; Training is performed according to the second sample and the fourth sample to obtain the pre-trained second segmentation model.
6. The medical image segmentation method according to claim 1, characterized in that: The first segmentation result includes one of a single-type blood vessel and a multi-type blood vessel segmentation result, and the second segmentation result includes the other of the single-type blood vessel and the multi-type blood vessel segmentation result, wherein the single-type blood vessel includes an arterial blood vessel or a venous blood vessel, and the multi-type blood vessel includes an arteriovenous blood vessel, and the target segmentation result obtained according to the first blood vessel segmentation result and the second blood vessel segmentation result includes: Matching different blood vessels from the first blood vessel segmentation result and the second blood vessel segmentation result as an intermediate blood vessel segmentation result, wherein the intermediate blood vessel segmentation result includes a single-type blood vessel segmentation result; The target segmentation result is obtained by merging the single-type blood vessel segmentation result in the first blood vessel segmentation result and the second blood vessel segmentation result with the intermediate blood vessel segmentation result.
7. The medical image segmentation method according to claim 1, characterized in that: The method further comprises: converting the image of the first development stage into an image of a third development stage, the third development stage being different from the first development stage and the second development stage; Using a pre-trained third segmentation model belonging to the third development stage, performing blood vessel segmentation on the image of the third development stage to obtain a third blood vessel segmentation result; A target blood vessel segmentation result is obtained according to the first blood vessel segmentation result, the second blood vessel segmentation result and the third blood vessel segmentation result.
8. The medical image segmentation method according to claim 1, characterized in that: The first segmentation model includes a first rough segmentation model and a first fine segmentation model. The first segmentation model is used to perform blood vessel segmentation on the image in the first development stage to obtain a first blood vessel segmentation result, including: Segmenting the image in the first development stage using a first rough segmentation model to obtain a rough segmentation result, wherein the rough segmentation result includes at least one initial blood vessel endpoint; Extracting a sub-image to be segmented according to the blood vessel at the at least one initial blood vessel endpoint; Segment the sub-image using a first fine segmentation model to obtain a fine segmentation result, wherein the fine segmentation result includes at least one new blood vessel endpoint; updating the at least one initial vessel endpoint using the at least one new vessel endpoint; Return to the step of extracting the sub-image to be segmented according to the blood vessel at the initial blood vessel endpoint until a preset end condition is met, and obtain a first blood vessel segmentation result according to the coarse segmentation result and each fine segmentation result.
9. The medical image segmentation method according to claim 8, characterized in that: The extracting the sub-image to be segmented according to the blood vessel at the blood vessel endpoint comprises: Determining the size of the sub-image according to the blood vessel size at the blood vessel endpoint; Extracting the sub-image based on the size of the sub-image, the sub-image comprising a two-dimensional slice or a three-dimensional image block; and / or Determining the direction of the sub-image according to the direction of the blood vessel at the blood vessel endpoint; Based on the orientation of the sub-image, the sub-image is extracted.
10. A medical image segmentation method, characterized in that: The method comprises: training according to the images of the first development stage of the training object and their annotated information to obtain a pre-trained first segmentation model, and training according to the images of the second development stage of the training object and their annotated information to obtain a pre-trained second segmentation model, wherein the first development stage includes one of the arterial phase, the venous phase, and the delayed phase, and the second development stage includes another of the arterial phase, the venous phase, and the delayed phase, the first segmentation model is used to perform blood vessel segmentation on the images belonging to the first development stage, and the second segmentation model is used to perform blood vessel segmentation on the images belonging to the second development stage; The training is performed according to the image of the first development stage of the training object and its annotation information to obtain a pre-trained first segmentation model, which includes: Acquire an image of a first development stage of a training object and its annotation information as a first sample; Using a pre-trained style transfer network to convert the non-first development stage image of the training object into the first development stage image, and using the converted first development stage image and its annotation information as the third sample; Perform training according to the first sample and the third sample to obtain the pre-trained first segmentation model; The training is performed according to the image of the second development stage of the training object and its annotation information to obtain a pre-trained second segmentation model, which includes: Acquire an image of the second development stage of the training object and its annotation information as a second sample; Using a pre-trained style transfer network to convert the non-second development stage image of the training object into the second development stage image, and using the converted second development stage image and its annotation information as a fourth sample; Training is performed according to the second sample and the fourth sample to obtain the pre-trained second segmentation model.
11. A medical image segmentation method, characterized in that: The method comprises: Performing blood vessel segmentation on the medical image using a coarse segmentation model to obtain a coarse segmentation result, wherein the coarse segmentation result includes at least one initial blood vessel endpoint; Determining the size of the sub-image to be segmented according to the blood vessel size at the at least one blood vessel endpoint; Extracting the sub-image based on the size of the sub-image, the sub-image comprising a two-dimensional slice or a three-dimensional image block; Segment the sub-image using a fine segmentation model to obtain a fine segmentation result, wherein the fine segmentation result includes at least one new blood vessel endpoint, and update the initial blood vessel endpoint with the new blood vessel endpoint; Return to the step of determining the size of the sub-image to be segmented according to the blood vessel size at the at least one blood vessel endpoint until a preset end condition is met, and obtain a blood vessel segmentation result according to the coarse segmentation result and each fine segmentation result.
12. A medical image processing device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the medical image segmentation method according to any one of claims 1 to 11 when executing the computer program.
13. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the medical image segmentation method according to any one of claims 1 to 11 is implemented.
