Medical image segmentation method, device, equipment and storage medium

Abstract

The application provides a medical image segmentation method, device and equipment and a storage medium, and relates to the technical field of medical image processing. The method comprises the following steps: acquiring a medical image; the medical image comprises a target lesion region to be segmented. An anatomical structure segmentation is performed on the medical image, and structure space prior information is output. The structure space prior information is used to constrain the spatial position and anatomical morphology of the target lesion region. According to the structure space prior information, the target lesion region is segmented and processed, and a segmentation result is generated. The method is used to improve the segmentation accuracy of the medical image.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and more specifically, to a method, apparatus, device, and storage medium for segmenting medical images. Background Technology

[0002] Currently, medical image segmentation is a crucial step in computer-aided diagnosis and treatment planning. In pelvic fracture segmentation, accurate identification of the fracture region is essential for preoperative planning, intraoperative navigation, and postoperative assessment. However, the complex anatomy of the pelvis, diverse fracture morphologies, and the presence of noise and artifacts in CT images pose challenges to the precise segmentation of fracture boundaries.

[0003] In existing technologies, a two-stage cascaded segmentation strategy is widely used: the first stage segments the overall anatomical structure, and the second stage performs fine segmentation of the lesion area.

[0004] However, in existing technologies and methods, the structural segmentation results in the first stage are mostly used only for ROI trimming or post-processing reference, and are not directly injected into the second stage segmentation process as explicit structural space priors. This results in the two stages of learning being relatively independent, making it difficult to fully utilize the guiding role of anatomical structures in lesion area localization and limiting segmentation accuracy. Summary of the Invention

[0005] The purpose of this application is to provide a method, apparatus, device, and storage medium for segmenting medical images, which solves the above-mentioned problems existing in the prior art and can greatly improve the segmentation accuracy of medical images.

[0006] Firstly, a method for segmenting medical images is provided, which may include: Acquire medical images; the medical images include the target lesion region to be segmented; The medical image is segmented into anatomical structures, and structural spatial prior information is output; the structural spatial prior information is used to constrain the spatial location and anatomical morphology of the target lesion region; Based on the prior information of the structural space, the target lesion region is segmented to generate a segmentation result.

[0007] Secondly, a medical image segmentation apparatus is provided, which may include: The acquisition module is used to acquire medical images; the medical images include target lesion regions to be segmented. The prior module is used to segment the medical image into anatomical structures and output structural space prior information; the structural space prior information is used to constrain the spatial location and anatomical morphology of the target lesion region. The segmentation module is used to segment the target lesion region based on the prior information of the structural space and generate segmentation results.

[0008] Thirdly, an electronic device is provided, which includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements any of the steps described in the first aspect above.

[0009] Fourthly, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when executed by a processor, the computer program implements the steps of any of the methods described in the first aspect above.

[0010] This application provides a method, apparatus, device, and storage medium for segmenting medical images, acquiring medical images including target lesion regions to be segmented. Anatomical structure segmentation is performed on the medical image, outputting prior structural spatial information. This prior information constrains the spatial location and anatomical morphology of the target lesion region. Based on the prior structural spatial information, the target lesion region is segmented to generate a segmentation result. In this scheme, the target lesion region is segmented based on the output prior structural spatial information to generate a segmentation result. Therefore, the architecture that combines prior structural spatial information with further segmentation processing is a serial architecture, allowing the anatomical structure segmentation output to serve as an explicit spatial prior, flowing into the lesion region segmentation stage. This achieves effective constraints on the spatial location and anatomical morphology of the lesion region, enabling stable expression of the overall bone structure, maintaining the overall pelvic anatomy consistency, improving the accuracy of fracture region localization, and significantly improving segmentation accuracy and robustness. This application is particularly suitable for the accurate segmentation of complex lesion regions such as pelvic fractures, and can greatly improve the segmentation accuracy of medical images. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a medical image segmentation method provided in an embodiment of this application; Figure 2 A schematic diagram of the structure of a medical image segmentation device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by those skilled in the art. The words "first," "second," and similar terms used in this application do not indicate any order, quantity, or importance, but are only used to distinguish different components. The words "comprising" or "including," etc., mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but do not exclude other elements or objects. The words "connected," "coupled," or "connected," etc., are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0014] For ease of understanding, the terms used in the embodiments of this application are explained below: Multi-class label map: This usually refers to a map where each voxel (or pixel) is assigned a discrete integer value to represent the anatomical category to which the voxel belongs.

[0015] The medical image segmentation method provided in this application embodiment can be applied to electronic devices, terminal devices, medical image segmentation devices or equipment, or other devices or equipment capable of executing this embodiment, and there are no limitations on this application. In this embodiment, the execution subject is described as an electronic device.

[0016] The terminal can be a user equipment (UE) such as a mobile phone, smartphone, laptop computer, digital broadcast receiver, personal digital assistant (PDA), or tablet computer (PAD), handheld device, in-vehicle device, wearable device, computing device, or other processing device connected to a wireless modem, mobile station (MS), or mobile terminal. This terminal has the ability to communicate with one or more core networks via a radio access network (RAN).

[0017] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0018] Figure 1 This is a flowchart illustrating a medical image segmentation method provided in an embodiment of this application. Figure 1 As shown, the method may include: Step S101: Acquire medical images; the medical images include the target lesion area to be segmented.

[0019] For example, an electronic device acquires medical images. The medical images may be raw pelvic CT images; the medical images include at least one target lesion region to be segmented. The target lesion region includes any one or more of the following: fracture region, tumor region, organ injury region; the fracture region includes any one or more of the following: pelvic fracture, acetabular fracture, sacral fracture. The target lesion region and fracture region are merely examples and are not intended to be limiting. Other lesions (such as tumors, organ injuries, etc.) may be included. Step S102: Perform anatomical structure segmentation on the medical image and output structural space prior information; the structural space prior information is used to constrain the spatial location and anatomical morphology of the target lesion area.

[0020] For example, the first-stage segmentation network is used to segment anatomical structures in medical images, outputting prior information about the structural space. The first-stage segmentation network is a pre-trained or jointly trained segmentation model whose task is to extract anatomical structure information from the input medical image. For instance, in a pelvic fracture segmentation task, the first-stage network segments the overall pelvic structure. The output of this network is the "prior information about the structural space," which can take the form of a binary mask, a probability map, or a multi-class label map.

[0021] For example, taking a raw pelvic CT image as an example, the raw pelvic CT image is input into a first-stage segmentation network. This network is a 3D U-Net architecture and outputs a binary mask of the overall pelvic anatomical structure. This mask is the prior information of the structural space. The mask includes voxels, and each voxel has a value of 0 or 1, indicating whether the voxel belongs to the pelvic anatomical structure.

[0022] Step S103: Based on the prior information of the structural space, the target lesion area is segmented to generate the segmentation result.

[0023] For example, the target lesion region is segmented based on prior structural spatial information and the second-stage segmentation network to generate a segmentation result. The second-stage segmentation network is responsible for the accurate segmentation of the target lesion region (such as a fracture region). This network receives not only the original medical image but also the prior structural spatial information output from the first stage and performs segmentation based on this information. The prior structural spatial information is used to constrain the spatial location and anatomical morphology of the target lesion region during the segmentation process.

[0024] For example, the original pelvic CT image and the structural space prior information output from the first stage are simultaneously input into the second-stage segmentation network. In the second-stage segmentation network, feature fusion is used to inject the structural space prior information into different stages of the network. Based on the injected structural space prior information, the second-stage segmentation network accurately segments the fracture region in the original pelvic CT image, outputting a segmentation result with anatomical constraints. The segmentation result can be in the form of a semantic segmentation mask, a probabilistic map, or an instance-level segmentation result. For example, if the target lesion region is a fracture region, the segmentation result is a fracture region segmentation mask.

[0025] Therefore, by constraining the anatomical structure, the lesion area segmented in the second stage conforms to the anatomical structure in spatial location and maintains anatomical consistency in morphology, thereby suppressing false detection, missed detection and morphological abnormalities.

[0026] The method provided in this application acquires a medical image, which includes a target lesion region to be segmented. Anatomical structure segmentation is performed on the medical image, outputting prior structural spatial information. This prior information constrains the spatial location and anatomical morphology of the target lesion region. Based on the prior structural spatial information, the target lesion region is segmented to generate a segmentation result. In this scheme, the target lesion region is segmented based on the output prior structural spatial information to generate a segmentation result. Therefore, the architecture that combines prior structural spatial information with further segmentation processing is a serial architecture. This allows the anatomical structure segmentation output to serve as an explicit spatial prior, flowing into the lesion region segmentation stage. This achieves effective constraints on the spatial location and anatomical morphology of the lesion region, enabling stable representation of the overall bone structure, maintaining the overall pelvic anatomy consistency, improving the accuracy of fracture region localization, and significantly improving segmentation accuracy and robustness. This application is particularly suitable for the accurate segmentation of complex lesion regions such as pelvic fractures, and can greatly improve the segmentation accuracy of medical images.

[0027] This application provides a method for segmenting medical images. This embodiment... Figure 1 Based on the embodiments, the method is described in detail below, and the method includes: Step S201: Acquire medical images; the medical images include the target lesion area to be segmented.

[0028] In one example, the target lesion area includes any one or more of the following: fracture area, tumor area, organ damage area; the fracture area includes any one or more of the following: pelvic fracture, acetabular fracture, sacral fracture.

[0029] For example, this step is described in step S101, and will not be repeated here.

[0030] Step S202: Perform anatomical structure segmentation on the medical image and output structural space prior information; the structural space prior information is used to constrain the spatial location and anatomical morphology of the target lesion area.

[0031] In one example, the prior information of the structural space is a multi-class label map of the anatomical structure. The multi-class label map is a three-dimensional integer label map, where the integer value of each voxel in the three-dimensional integer label map represents the category of the anatomical substructure to which the voxel belongs.

[0032] For example, the prior information of the structural space is a multi-class label map of anatomical structures. This multi-class label map is a three-dimensional integer label map, where the integer value of each voxel represents a different anatomical substructure such as the pelvis, femur, or sacrum. For instance, in the multi-class label map, an integer value of 0 represents the background, an integer value of 1 represents the pelvis, an integer value of 2 represents the femur, and an integer value of 3 represents the sacrum. Alternatively, the prior information of the structural space can be a binary mask of the anatomical structures, where a foreground voxel value of 1 indicates that it belongs to an anatomical structure, and a background voxel value of 0 indicates that it does not belong to an anatomical structure.

[0033] Optionally, the medical image is a three-dimensional CT image with a voxel spacing between 0.5 mm and 2 mm and an image size of 512 × 512 × N, where N is the number of scan slices. Alternatively, the medical image is an MRI image, using any of the following sequences: T1-weighted, T2-weighted, or STIR.

[0034] Step S203: Based on the prior information of the structural space, the target lesion area is segmented to generate a segmentation result.

[0035] In one example, step S203 includes several implementations: The first implementation of step S203 is as follows: extract the intermediate layer feature map of the medical image; concatenate the structural space prior information with the intermediate layer feature map along the channel dimension to generate a concatenated feature map; input the concatenated feature map into the subsequent convolutional layer in the segmentation process for feature learning; based on the learned features, generate the segmentation result of the target lesion region; the spatial location and anatomical morphology of the target lesion region in the segmentation result are both constrained by the structural space prior information.

[0036] The second implementation of step S203 is as follows: Extract the intermediate layer feature map of the medical image; process the prior information of the structural space into an attention weight map; the value of each position in the attention weight map represents the degree of response of that position to the anatomical structure; multiply the attention weight map and the intermediate layer feature map element by element to generate a weighted feature map; wherein, the feature response of the position corresponding to the anatomical structure is greater than the original feature response, and the feature response of the position corresponding to the non-anatomical structure is less than the original feature response; perform subsequent feature learning based on the weighted feature map to generate the segmentation result of the target lesion area; the spatial position and anatomical shape of the target lesion area in the segmentation result are both constrained by the prior information of the structural space.

[0037] In one example, the attention weight map and the intermediate layer feature map have the same spatial size and number of channels, and element-wise multiplication is the multiplication of corresponding elements at each position.

[0038] In one example, the segmentation result includes any of the following: a 3D semantic segmentation mask, a probabilistic graph, or an instance-level segmentation mask.

[0039] For example, the implementation methods of feature fusion include channel dimension splicing, attention weighting, or gating mechanisms, etc., without limitation; the gating mechanism is to dynamically filter the prior information of the structural space through learnable gating units, and then inject the filtered information into the second-stage segmentation network.

[0040] In the first implementation, intermediate layer feature maps of the medical image are extracted. These intermediate layer feature maps refer to the feature maps output by one or more intermediate layers during the second-stage segmentation network's processing of the medical image. These feature maps already contain certain semantic information of the image. The structural space prior information is concatenated with the intermediate layer feature maps along the channel dimension to generate a concatenated feature map. The structural space prior information (such as a binary mask) itself can be considered as a single-channel or multi-channel feature map. Connecting this feature map with the intermediate layer feature maps along the channel dimension forms a new feature map with an increased number of channels (i.e., the concatenated feature map). The concatenated feature map is then input into subsequent convolutional layers in the segmentation process for feature learning. Since the concatenated feature map contains both the original image features and the structural space prior information, subsequent convolutional layers can learn both types of information simultaneously, guiding the feature extraction process with structural priors. Finally, after learning by subsequent layers, the final segmentation result is output. The segmentation result is constrained by the structural space prior information, reflected in aspects such as the target lesion region being located inside the anatomical structure and the boundary aligning with the anatomical structure.

[0041] Therefore, by using a feature fusion method that splices together channels, prior information of the structural space is used as an additional feature channel. Subsequent convolutional layers can learn both image features and prior features simultaneously, making the segmentation results more accurate at the anatomical location.

[0042] In the second implementation, firstly, intermediate layer feature maps of the medical image are extracted. The structural spatial prior information is processed into an attention weight map. The structural spatial prior information (such as an anatomical structure mask) is equivalent to the attention map, where anatomical structure regions have high weights and background regions have low weights; alternatively, the structural spatial prior information can be learned by the network to transform into a more suitable weight distribution. The attention weight map and the intermediate layer feature map have the same spatial dimensions, and the value at each position in the attention weight map represents the intensity at which that position should be focused in subsequent processing. Then, the attention weight map and the intermediate layer feature map are multiplied element-wise to generate a weighted feature map. The feature response at the anatomical structure location is greater than the original feature response, while the feature response at the non-anatomical structure location is less than the original feature response. Specifically, through element-wise multiplication, the feature values ​​at the anatomical structure location are preserved or enhanced (multiplied by a weight close to 1), while the feature values ​​at the non-anatomical structure location are suppressed (multiplied by a weight close to 0). This operation makes the network pay more attention to the anatomical structure region in subsequent processing. Finally, based on the weighted feature map, subsequent feature learning is performed to generate the segmentation result of the target lesion area. The spatial location and anatomical shape of the target lesion area in the segmentation result are both constrained by the prior information of the structural space. Therefore, the subsequent convolutional layer further extracts features on the basis of the weighted feature map, so that the final segmentation result focuses on the interior of the anatomical structure or near the boundary.

[0043] Optionally, the attention weight map and the intermediate layer feature map have the same spatial size and number of channels, and element-wise multiplication is performed by multiplying corresponding elements in the two maps. Therefore, using the prior information of the structural space as the attention weight map and performing element-wise multiplication with the intermediate layer feature map of the second-stage segmentation network can enhance the response of the anatomical structural regions and suppress interference from unstructured regions.

[0044] Therefore, by using attention-weighted feature fusion, prior information is used as weights to modulate feature responses, enhancing features in anatomical regions and suppressing features in non-anatomical regions, making the segmentation boundary more closely resemble the real anatomical structure.

[0045] Optionally, the segmentation result can include any of the following: a 3D semantic segmentation mask, a probabilistic map, or an instance-level segmentation mask. For a 3D semantic segmentation mask, each voxel is assigned a category label (e.g., 0 for background, 1 for fracture region), and this result can be directly used for visualization or as input for subsequent processing. For a probabilistic map, each voxel takes a value between 0 and 1, representing the probability that the voxel belongs to the target lesion region. The probabilistic map retains more uncertainty information and can be used in scenarios requiring confidence assessment. For an instance-level segmentation mask, it not only distinguishes between the target and the background but also between different instances (e.g., different fracture fragments). This output is particularly important for tasks requiring fragment-level analysis (e.g., fracture reduction planning).

[0046] Optionally, structural space prior information can be injected into at least one stage of the second-stage segmentation network, either the encoding stage, decoding stage, or skip connection stage, through feature fusion. For example, structural space prior information can be directly injected into the second-stage segmentation network, including specific implementations such as injection into the encoder, decoder, bottleneck layer, or skip connections, without limitation. Specifically, structural space prior information can be introduced after the first convolutional layer in the encoding stage to establish anatomical constraints early in feature extraction. Alternatively, structural space prior information can be introduced before upsampling in the last layer of the decoding stage to refine the segmentation boundary near the output layer. Or, structural space prior information can be introduced at multiple skip connections to form multi-scale anatomical constraints. Therefore, directly injecting structural space prior information into the intermediate layers of the network allows it to participate in the subsequent feature learning process, thereby substantially constraining the segmentation results.

[0047] Optionally, the first-stage segmentation network and the second-stage segmentation network can be trained jointly, with the total loss function being the weighted sum of the first-stage and second-stage segmentation losses. The weight coefficients are adaptively adjusted based on the validation set performance. Alternatively, the first-stage and second-stage segmentation networks can be trained in stages, i.e., the first-stage segmentation network is trained independently until convergence, and then the parameters of the first-stage segmentation network are fixed before training the second-stage segmentation network.

[0048] The method provided in this application acquires a medical image, which includes a target lesion region to be segmented. Anatomical structure segmentation is performed on the medical image, outputting prior structural spatial information. This prior information constrains the spatial location and anatomical morphology of the target lesion region. Based on the prior structural spatial information, the target lesion region is segmented to generate a segmentation result. In this scheme, the target lesion region is segmented based on the output prior structural spatial information to generate a segmentation result. Therefore, the architecture that combines prior structural spatial information with further segmentation processing is a serial architecture. This allows the anatomical structure segmentation output to serve as an explicit spatial prior, flowing into the lesion region segmentation stage. This achieves effective constraints on the spatial location and anatomical morphology of the lesion region, enabling stable representation of the overall bone structure, maintaining the overall pelvic anatomy consistency, improving the accuracy of fracture region localization, and significantly improving segmentation accuracy and robustness. This application is particularly suitable for the accurate segmentation of complex lesion regions such as pelvic fractures, and can greatly improve the segmentation accuracy of medical images.

[0049] Corresponding to the above method, embodiments of this application also provide a medical image segmentation device, such as... Figure 2 As shown, the device includes: Acquisition module 41 is used to acquire medical images; the medical images include target lesion regions to be segmented; Prior module 42 is used to segment the medical image into anatomical structures and output structural space prior information; the structural space prior information is used to constrain the spatial location and anatomical morphology of the target lesion area; The segmentation module 43 is used to segment the target lesion region based on the prior information of the structural space and generate a segmentation result.

[0050] The functions of each functional unit of the medical image segmentation device provided in the above embodiments of this application can be implemented through the above method steps. Therefore, the specific working process and beneficial effects of each unit in the medical image segmentation device provided in the embodiments of this application will not be repeated here.

[0051] This application also provides an electronic device, such as... Figure 3 As shown, it includes a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540.

[0052] Memory 530 is used to store computer programs; The processor 510 performs the above steps when executing the program stored in the memory 530.

[0053] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0054] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0055] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0056] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0057] The implementation methods and beneficial effects of the various components of the electronic device in the above embodiments for solving the problem can be found in [reference needed]. Figure 1 The steps in the illustrated embodiments are used to implement the electronic device. Therefore, the specific working process and beneficial effects of the electronic device provided in this application will not be repeated here.

[0058] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform any of the medical image segmentation methods described in the above embodiments.

[0059] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the medical image segmentation methods described in the above embodiments.

[0060] Those skilled in the art will understand that the embodiments in this application can be provided as methods, systems, or computer program products. Therefore, the embodiments in this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments in this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] This application describes embodiments of methods, apparatus (systems), and computer program products according to embodiments of this application with reference to flowchart illustrations and / or block diagrams. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0063] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0064] Although preferred embodiments have been described in this application, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of this application.

[0065] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims in this application and their equivalents, then this application also intends to include these modifications and variations.

Claims

1. A method for segmenting medical images, characterized in that, The method includes: Acquire medical images; the medical images include the target lesion region to be segmented; The medical image is segmented into anatomical structures, and structural spatial prior information is output; the structural spatial prior information is used to constrain the spatial location and anatomical morphology of the target lesion region; Based on the prior information of the structural space, the target lesion region is segmented to generate a segmentation result.

2. The method as described in claim 1, characterized in that, The step of segmenting the target lesion region based on the prior information of the structural space and generating a segmentation result includes: Extract the intermediate layer feature map of the medical image; The prior information of the structural space is concatenated with the intermediate layer feature map along the channel dimension to generate a concatenated feature map; The concatenated feature map is then input into the subsequent convolutional layers in the segmentation process for feature learning. Based on the learned features, a segmentation result of the target lesion region is generated; the spatial location and anatomical morphology of the target lesion region in the segmentation result are both constrained by the prior information of the structural space.

3. The method as described in claim 1, characterized in that, The step of segmenting the target lesion region based on the prior information of the structural space and generating a segmentation result includes: Extract the intermediate layer feature map of the medical image; The prior information of the structural space is processed into an attention weight map; the value of each position in the attention weight map represents the degree of response of that position to the anatomical structure. The attention weight map is multiplied element-wise with the intermediate layer feature map to generate a weighted feature map; wherein the feature response at the location corresponding to the anatomical structure is greater than the original feature response, and the feature response at the location corresponding to the non-anatomical structure is less than the original feature response. Subsequent feature learning is performed based on the weighted feature map to generate segmentation results of the target lesion region; the spatial location and anatomical morphology of the target lesion region in the segmentation results are both constrained by the prior information of the structural space.

4. The method as described in claim 3, characterized in that, The attention weight map and the intermediate layer feature map have the same spatial size and number of channels, and the element-wise multiplication means multiplying the elements at corresponding positions respectively.

5. The method according to any one of claims 1-4, characterized in that, The segmentation result includes any of the following: 3D semantic segmentation mask, probabilistic graph, or instance-level segmentation mask.

6. The method according to any one of claims 1-4, characterized in that, The prior information of the structural space is a multi-class label map of the anatomical structure. The multi-class label map is a three-dimensional integer label map. The integer value of each voxel in the three-dimensional integer label map represents the category of the anatomical substructure to which the voxel belongs.

7. The method according to any one of claims 1-4, characterized in that, The target lesion area includes any one or more of the following: Fracture area, tumor area, organ damage area; The fracture area includes any one or more of the following: Pelvic fracture, acetabular fracture, sacral fracture.

8. A medical image segmentation device, characterized in that, The device includes: The acquisition module is used to acquire medical images; the medical images include target lesion regions to be segmented. The prior module is used to segment the medical image into anatomical structures and output structural space prior information; the structural space prior information is used to constrain the spatial location and anatomical morphology of the target lesion region. The segmentation module is used to segment the target lesion region based on the prior information of the structural space and generate segmentation results.

9. An electronic device, characterized in that, The electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images