A medical image segmentation method, system and storage medium

Abstract

The application discloses a medical image segmentation method and system and a storage medium, and belongs to the field of medical image processing; the method comprises the following steps: acquiring a medical image as an input image to be segmented; extracting a foreground from the input image to be segmented to obtain a foreground class segmentation image to be processed; calculating the pose features of a target in the foreground class segmentation image to be processed to obtain the pose features of a series of targets to be segmented; recognizing and adjusting the pose of the input image to be segmented, and extracting the target; mapping the extracted target back to the pose of the corresponding target in the input image to be segmented to obtain a series of target segmentation results; mapping the series of target segmentation results to the input image to be segmented to obtain a target segmentation image; and integrating the target segmentation image to finally obtain a prediction image; the application can segment images of any data dimension and modality, and solves the problems of small target proportion and unclear contour through a two-stage segmentation method.

Description

Technical Field

[0001] This invention relates to the technical field of medical image processing, and in particular to a medical image segmentation method, system, and storage medium. Background Technology

[0002] Image segmentation is a crucial task in computer vision, aiming to assign pixels in an image to different semantic categories. The accuracy of image segmentation is critical in many scenarios, especially when precise differentiation and localization of different objects within an image are required. In medical image analysis, the accuracy of image segmentation is closely related to the accurate identification and localization of lesions, organs, and tissues. High-precision image segmentation can assist doctors in diagnostic work, surgical planning, and treatment monitoring.

[0003] In the field of medical image processing, the main existing methods include thresholding and region growing.

[0004] Thresholding segmentation primarily relies on image grayscale values ​​for segmentation. This characteristic makes it difficult to achieve effective segmentation with a single grayscale threshold when dealing with multimodal images, such as medical images that fuse optical and infrared information. Furthermore, it is typically only suitable for simple two-dimensional images; its segmentation effect is poor for three-dimensional and higher-dimensional images, limiting its application to specific images with significant grayscale differences. In terms of segmentation methods, it only sets a threshold to separate the target from the background. For images where the target occupies a small portion of the foreground and its outline is unclear, this single approach is prone to missegmentation and cannot meet the requirements for high-precision segmentation. Simultaneously, thresholding segmentation does not consider factors such as the target's pose characteristics. When the target's shape is complex, it cannot adjust the segmentation strategy according to the actual situation, resulting in poor segmentation results. Moreover, because it only relies on grayscale thresholding for segmentation, the foreground and target are not processed and fused separately during the segmentation process, making it difficult to ensure that the segmented target outline completely matches the real outline. This problem is even more pronounced in complex images.

[0005] Region growing methods have poor adaptability to different image modalities, and it is difficult to find universal growth criteria in medical images with different imaging principles. When processing high-dimensional images, the complexity of growth calculations makes accurate segmentation difficult, greatly limiting its application scenarios. This method is based on seed points and similarity criteria. For images where the target occupies a small proportion in the foreground and the outline is unclear, if the seed point is not properly selected or the growth criteria are not suitable, it is difficult to effectively distinguish between the foreground and the target, resulting in poor segmentation. Region growing methods mainly focus on the similarity of local pixels and do not involve the analysis of the overall pose features of the target. When the target shape is complex, it is difficult to flexibly adjust the growth process according to the target pose, making it difficult to accurately extract the target. During the growth process, the results of foreground segmentation and target segmentation are not fused and optimized, which can easily lead to incomplete or inaccurate target outlines, resulting in deviations from the true outline.

[0006] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a medical image segmentation method, system, and storage medium, which can segment images of any data dimension and modality. It solves the problems of small target proportion and unclear contours through a two-stage segmentation method, adjusts the image by calculating the target pose features to cope with complex morphological changes, and fuses the two-stage results to make the target contour more closely match the real contour.

[0008] To achieve the above objectives, the present invention provides a medical image segmentation method, comprising the following steps:

[0009] S1: Acquire medical images as input images to be segmented;

[0010] S2: Extract the foreground from the input image to be segmented to obtain the foreground category segmentation image to be processed;

[0011] S3: Calculate the pose features of the targets in the foreground category segmentation image to be processed, and obtain a series of pose features of the targets to be segmented;

[0012] S4: Based on the pose features of a series of targets to be segmented, the pose of the input image to be segmented is identified and adjusted, and the targets are extracted from the pose-adjusted input image; the extracted targets are mapped back to the pose of the corresponding targets in the input image to be segmented, and a series of target segmentation results are obtained.

[0013] S5: Map the series of target segmentation results onto the input image to be segmented to obtain the target segmentation image;

[0014] S6: The foreground category segmentation image is used to segment the target image, and the final predicted image is obtained.

[0015] In one embodiment of the present invention, step S2 is performed in a pre-trained foreground category segmentation model; the foreground category segmentation model is based on Unet and includes an encoder and a decoder. The encoder is used to extract the semantic features of the input image to be segmented and convert the input data into a low-dimensional representation; the decoder is used to parse the semantic features of the input image into a foreground category segmentation image to be processed and convert the low-dimensional representation back to the original data space.

[0016] The encoder's structure consists of several alternating convolutional modules and downsampling modules.

[0017] The decoder is composed of several convolutional modules and upsampling modules that alternate, and it fuses features at different resolutions in the encoder with features at the corresponding resolution level in the decoder through skip connections.

[0018] In one embodiment of the present invention, the specific steps for calculating the pose features of the target in the foreground category segmentation image to be processed in step S3 are as follows:

[0019] S301: Calculate the center point P of the specified target A in the foreground category segmentation image to be processed. A And calculate the center point P of the nearest target. A-1 and P A+1 Calculate the direction vector of target A. :

[0020] ;

[0021] S302: Direction vector through target A Calculate the normalized pose of target A. rotation matrix ; where, rotation matrix The solution is obtained using the following formula:

[0022] ;

[0023] in, Let A be the direction vector of the normalized attitude of target A;

[0024] S303: From target A to standardized attitude rotation matrix Calculate the standardized attitude Rotation matrix to target A :

[0025] ;

[0026] S304: Select the center point of target A Target A to standardized posture rotation matrix and standardized posture Rotation matrix to target A As the pose feature of target A.

[0027] In one embodiment of the present invention, the specific process of recognizing and adjusting the pose of the input image to be segmented in step S4 is as follows:

[0028] S401: Standardized attitude towards target A Calculate the range of the target standardized image ;

[0029] S402: Range of the target-normalized image Standardized posture Rotation matrix to target A and the center point P of target A A Calculate the segmentation range of target A in the input image. :

[0030] ;

[0031] S403: Use the segmentation range of the target A in the input image. and target A to standardized posture rotation matrix The input image to be segmented is processed to obtain a pose adjustment for target A and convert it into a standardized pose. The input image below :

[0032] .

[0033] Wherein, Input represents the input image to be segmented.

[0034] In one embodiment of the present invention, the specific process of mapping the extracted target back to the pose of the corresponding target in the input image to be segmented in step S4 is as follows:

[0035] Target segmentation results extracted by pose-normalized segmentation network Through standardized posture Rotation matrix to target A and the center point of target A The target A extracted by the pose-normalized segmentation network is mapped back to the location of target A in the input image to be segmented, and then mapped back to the location of the target in the input image to be segmented. :

[0036]

[0037] In one embodiment of the present invention, step S4 is performed in a pre-trained pose normalization segmentation model; the pose normalization segmentation model includes an encoder and a decoder; the encoder is used to extract semantic features of the pose-adjusted input image and convert the input data into a low-dimensional representation; the decoder is used to parse the semantic features of the pose-adjusted input image into targets in the pose-adjusted input image and convert the low-dimensional representation back to the original data space.

[0038] The encoder's structure consists of several alternating convolutional modules and downsampling modules;

[0039] The decoder is composed of several convolutional modules and upsampling modules that alternate, and it fuses features at different resolutions in the encoder with features at the corresponding resolution level in the decoder through skip connections.

[0040] In one embodiment of the present invention, in step S4, if there are multiple targets in the foreground category segmentation image to be processed, the segmentation results of all targets are merged to obtain a target segmentation image.

[0041] In one embodiment of the present invention, step S6 specifically includes the following steps:

[0042] S601: Take the complement of the foreground category segmentation image and the target segmentation image;

[0043] S602: Cluster all pixels in the complement set according to the category of the predicted target;

[0044] S603: Merge the results into the target segmentation image to obtain the predicted image.

[0045] The present invention also provides a medical image segmentation system for performing the above-described medical image segmentation method, the system comprising:

[0046] The image acquisition module is used to acquire medical images as input images to be segmented;

[0047] The foreground category segmentation module is used to extract the foreground from the input image to be segmented, and obtain the foreground category segmentation image to be processed;

[0048] The pose feature calculation module is used to calculate the pose features of targets in the foreground category segmentation image to be processed, and obtain a series of pose features of the targets to be segmented.

[0049] The pose normalization segmentation model is used to identify and adjust the pose of the input image to be segmented by using the pose features of a series of targets to be segmented, and to extract the targets from the pose-adjusted input image; the extracted targets are then mapped back to the pose of the corresponding targets in the input image to be segmented to obtain a series of target segmentation results.

[0050] The pose normalization segmentation module is used to map a series of target segmentation results onto the input image to be segmented, so as to obtain the target segmented image;

[0051] The image integration module is used to segment images according to the foreground category to be processed, integrate the target segmented images, and finally obtain the predicted image.

[0052] The present invention also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described medical image segmentation method.

[0053] Compared with the prior art, the medical image segmentation method, system, and storage medium according to the present invention have the following advantages and beneficial effects:

[0054] (1) The present invention can segment images of any data dimension and any modality.

[0055] (2) The present invention adopts a two-stage segmentation method, first segmenting the overall foreground and then segmenting the target in the foreground, which can solve the problem of poor image segmentation effect caused by factors such as the target having a small proportion in the foreground and unclear target outline.

[0056] (3) After foreground segmentation, the present invention calculates the pose features of the target in the foreground and adaptively adjusts the image to be segmented according to the pose features of the target, which can solve the problem of the target's complex shape changes and difficulty in segmentation, and improve the segmentation effect of specific targets.

[0057] (4) The present invention integrates the results of foreground category segmentation and target segmentation in two stages, so that the outline of the target can more completely match the real outline. Attached Figure Description

[0058] Figure 1 This is a flowchart of a medical image segmentation method according to an embodiment of the present invention;

[0059] Figure 2 This is a schematic diagram of a medical image segmentation system according to an embodiment of the present invention. Detailed Implementation

[0060] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0061] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0062] like Figures 1 to 2 As shown, a medical image segmentation method according to a preferred embodiment of the present invention includes the following steps:

[0063] S1: Acquire a medical image as the input image to be segmented; the medical image can be any modality of medical image, such as CT, CBCT, MRI, PET, etc.

[0064] S2: Extract the foreground from the input image to be segmented to obtain the foreground category segmentation image to be processed.

[0065] Step S2 primarily operates within a pre-trained foreground category segmentation model. This model is trained using any modality of medical image as training samples, with the aim of obtaining foreground category segmentation images that approximate standard foreground category segmentation images.

[0066] The foreground category segmentation model is based on Unet and includes an encoder and a decoder. The encoder is used to extract the semantic features of the input image to be segmented and convert the input data into a low-dimensional representation. The decoder is used to parse the semantic features of the input image into a foreground category segmentation image to be processed and convert the low-dimensional representation back into the original data space.

[0067] The encoder's structure consists of several alternating convolutional modules and downsampling modules.

[0068] The decoder is composed of several convolutional modules and upsampling modules, and features at different resolutions in the encoder are fused with features at the corresponding resolution level in the decoder through skip connections to compensate for the loss of detail caused by downsampling and upsampling.

[0069] S3: Calculate the pose features of the targets in the foreground category segmentation image to be processed, and obtain a series of pose features of the targets to be segmented.

[0070] In step S3, the specific steps for calculating the pose features of the target in the foreground category segmentation image to be processed are as follows:

[0071] S301: Calculate the center point of the specified target A in the foreground category segmentation image to be processed. And calculate the center point of the nearby target. and Through points and Calculate the direction vector of target A :

[0072] .

[0073] The criteria for specifying target A are as follows: In the foreground segmentation image, select targets with clearly defined adjacent targets (one in front and one behind), and calculate their direction vectors using the center points of these adjacent targets. This rule ensures that the calculation of pose features has spatial relevance and physical meaning, providing a reliable basis for subsequent pose adjustment and target segmentation.

[0074] S302: Direction vector through target A Calculate the normalized pose of target A. rotation matrix .

[0075] Wherein, rotation matrix The solution is obtained using the following formula:

[0076]

[0077] in, Let be the direction vector of the normalized attitude of target A.

[0078] Standardized pose is a predefined target pose with a standard reference value. Its purpose is to convert targets with different poses to a standardized pose, facilitating consistent processing in subsequent segmentation operations. When analyzing the poses of multiple targets or the same target at different time points, standardized pose provides a unified reference frame. This allows different poses to be quantified, compared, and processed under the same standard, simplifying subsequent analysis operations.

[0079] S303: From target A to standardized attitude rotation matrix Calculate the standardized attitude Rotation matrix to target A :

[0080] .

[0081] S304: Select the center point of target A Target A to standardized posture rotation matrix and standardized posture Rotation matrix to target A As the pose feature of target A.

[0082] If there are multiple targets in the foreground category segmentation image to be processed, different pose features are calculated for each of the multiple targets.

[0083] S4: Based on the pose features of a series of targets to be segmented, the pose of the input image to be segmented is identified and adjusted, and the target is extracted from the pose-adjusted input image; the extracted target is mapped back to the pose of the corresponding target in the input image to be segmented, and a series of target segmentation results are obtained.

[0084] Step S4 primarily operates within a pre-trained pose-normalized segmentation model. This model is trained using the pose features of any modality of medical image and the target as training samples, with the aim of obtaining target segmentation results that approximate standard target segmentation results.

[0085] In step S4, the specific process of recognizing and adjusting the pose of the input image to be segmented is as follows:

[0086] S401: Standardized attitude towards target A Calculate the range of the target standardized image ;

[0087] The specific range of the target-normalized image needs to be manually set or determined through training data statistics based on the target characteristics and application scenario, rather than being calculated in real time using mathematical formulas. Its core function is to serve as a benchmark for spatial transformation, mapping the segmentation range of the original image to the target center point through a rotation matrix.

[0088] S402: Range of the target-normalized image Standardized posture Rotation matrix to target A and the center point P of target A A Calculate the segmentation range of target A in the input image. :

[0089] .

[0090] S403: Use the segmentation range of the target A in the input image. and target A to standardized posture rotation matrix The input image to be segmented is processed to obtain a pose adjustment for target A and convert it into a standardized pose. The input image below :

[0091] .

[0092] in, This represents the input image to be segmented.

[0093] In step S4, step S2 mainly operates within the pre-trained pose normalization segmentation model. The pre-trained pose normalization segmentation model includes an encoder and a decoder; the encoder is used to extract semantic features from the pose-adjusted input image, converting the input data into a low-dimensional representation; the decoder is used to parse the semantic features of the pose-adjusted input image into targets in the pose-adjusted input image, converting the low-dimensional representation back into the original data space.

[0094] The encoder's structure consists of several alternating convolutional modules and downsampling modules;

[0095] The decoder is composed of several convolutional modules and upsampling modules that alternate. It uses skip connections to fuse features at different resolutions in the encoder with features at the corresponding resolution level in the decoder to compensate for the loss of detail caused by downsampling and upsampling.

[0096] Both the pose normalization segmentation model and the foreground category segmentation model are based on the Unet architecture, but they are independent models at different stages of two-stage segmentation.

[0097] In step S4, the specific process of mapping the extracted target back to the pose of the corresponding target in the input image to be segmented is as follows:

[0098] Target segmentation results extracted by pose-normalized segmentation network Through standardized posture Rotation matrix to target A and the center point of target A The target A extracted by the pose-normalized segmentation network is mapped back to the location of target A in the input image to be segmented, and then mapped back to the location of the target in the input image to be segmented. :

[0099] .

[0100] S5: Map the series of target segmentation results onto the input image to be segmented to obtain the target segmentation image. If there are multiple targets in the foreground category segmentation image to be processed, merge the segmentation results of all targets to obtain the target segmentation image.

[0101] S6: The foreground category segmentation image is used to segment the target image, and the target segmentation image is integrated to obtain the final prediction image.

[0102] Step S6 specifically includes the following steps:

[0103] S601: Take the complement set of the foreground category segmentation image and the target segmentation image.

[0104] Foreground category segmentation identifies the foreground region in the image, while target segmentation identifies the region containing a specific target. The complement operation identifies the portion of the foreground region that does not belong to the target region.

[0105] S602: Cluster all pixels in the complement set according to the category of the predicted target.

[0106] Considering the real-time nature and target characteristics of medical images, clustering algorithms can be selected as follows: K-means clustering, which divides pixels into K classes based on the Euclidean distance of feature vectors, where K can be preset as the number of target classes; Density Peak Clustering (DPC), which is suitable for scenarios with high target pixel density and determines the cluster center through local density and distance; and supervised clustering, which can discover parts of the foreground that may have been ignored in the initial target segmentation for various reasons but actually belong to the target, thus capturing target information more comprehensively.

[0107] S603: Merge the results into the target segmentation image to obtain the predicted image.

[0108] Specifically, pixels identified as relevant to the target after clustering are added to the corresponding regions of the target segmentation image. The purpose is to improve the target segmentation results by supplementing previously missed target-related parts, so that the final predicted image can more accurately reflect the true situation of the target in the image, thereby improving the completeness and accuracy of target segmentation.

[0109] like Figure 2 As shown, a medical image segmentation system according to a preferred embodiment of the present invention is mainly used to perform the above-described medical image segmentation method. The system includes:

[0110] Image acquisition module 1 is used to acquire medical images as input images to be segmented.

[0111] Foreground category segmentation module 2 is used to extract the foreground from the input image to be segmented, and obtain the foreground category segmentation image to be processed.

[0112] The pose feature calculation module 3 is used to calculate the pose features of the targets in the foreground category segmentation image to be processed, and obtain a series of pose features of the targets to be segmented.

[0113] The pose normalization segmentation model 4 is used to identify and adjust the pose of the input image to be segmented by using the pose features of a series of targets to be segmented, extract the targets from the pose-adjusted input image, map the extracted targets back to the pose of the corresponding targets in the input image to be segmented, and finally obtain the segmentation results of a series of targets.

[0114] The pose normalization segmentation module 5 is used to map a series of target segmentation results onto the input image to be segmented, so as to obtain the target segmented image.

[0115] Image integration module 6 is used to segment images by foreground category to be processed, integrate target segmented images, and finally obtain predicted images.

[0116] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0117] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0118] 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.

[0119] 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.

[0120] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A medical image segmentation method, characterized in that, Includes the following steps: S1: Acquire medical images as input images to be segmented; S2: Extract the foreground from the input image to be segmented to obtain the foreground category segmentation image to be processed; S3: Calculate the pose features of the targets in the foreground category segmentation image to be processed, and obtain a series of pose features of the targets to be segmented; the specific steps of calculating the pose features of the targets in the foreground category segmentation image to be processed in step S3 are as follows: S301: Calculate the center point of the specified target A in the foreground category segmentation image to be processed. And calculate the center point of the nearby target. and Calculate the direction vector of target A. : ； S302: Direction vector through target A Calculate the normalized pose of target A. rotation matrix ; where, rotation matrix The solution is obtained using the following formula: ； in, Let A be the direction vector of the normalized attitude of target A; S303: From target A to standardized attitude rotation matrix Calculate the standardized attitude Rotation matrix to target A : ； S304: Select the center point of target A Target A to standardized posture rotation matrix and standardized posture Rotation matrix to target A As the pose feature of target A; S4: Based on the pose features of a series of targets to be segmented, the pose of the input image to be segmented is identified and adjusted, and the targets are extracted from the pose-adjusted input image; the extracted targets are mapped back to the pose of the corresponding targets in the input image to be segmented, and a series of target segmentation results are obtained. S5: Map the series of target segmentation results onto the input image to be segmented to obtain the target segmentation image; S6: Segment the image by the foreground category to be processed, integrate the target segmented image, and finally obtain the predicted image; In step S4, the specific process of recognizing and adjusting the pose of the input image to be segmented is as follows: S401: Standardized attitude towards target A Calculate the range of the target standardized image ; S402: Range of the target-normalized image Standardized posture Rotation matrix to target A and the center point P of target A A Calculate the segmentation range of target A in the input image. : ； S403: Use the segmentation range of the target A in the input image. and target A to standardized posture rotation matrix The input image to be segmented is processed to obtain a pose adjustment for target A and convert it into a standardized pose. The input image below : ； in, This represents the input image to be segmented; In step S4, the specific process of mapping the extracted target back to the pose of the corresponding target in the input image to be segmented is as follows: Target segmentation results extracted by pose-normalized segmentation network Through standardized posture Rotation matrix to target A and the center point of target A The target A extracted by the pose-normalized segmentation network is mapped back to the location of target A in the input image to be segmented, and then mapped back to the location of the target in the input image to be segmented. : 。 2. The medical image segmentation method as described in claim 1, characterized in that, Step S2 is performed in a pre-trained foreground category segmentation model. The foreground category segmentation model is based on Unet and includes an encoder and a decoder. The encoder is used to extract the semantic features of the input image to be segmented and convert the input data into a low-dimensional representation. The decoder is used to parse the semantic features of the input image into a foreground category segmentation image to be processed and convert the low-dimensional representation back to the original data space. The encoder's structure consists of several alternating convolutional modules and downsampling modules; The decoder is composed of several convolutional modules and upsampling modules that alternate, and it fuses features at different resolutions in the encoder with features at the corresponding resolution level in the decoder through skip connections.

3. The medical image segmentation method as described in claim 1, characterized in that, Step S4 is performed in a pre-trained pose normalization segmentation model; the pose normalization segmentation model includes an encoder and a decoder; the encoder is used to extract semantic features of the pose-adjusted input image and convert the input data into a low-dimensional representation; the decoder is used to parse the semantic features of the pose-adjusted input image into targets in the pose-adjusted input image and convert the low-dimensional representation back to the original data space. The encoder's structure consists of several alternating convolutional modules and downsampling modules; The decoder is composed of several convolutional modules and upsampling modules that alternate, and it fuses features at different resolutions in the encoder with features at the corresponding resolution level in the decoder through skip connections.

4. The medical image segmentation method as described in claim 1, characterized in that, In step S4, if there are multiple targets in the foreground category segmentation image to be processed, the segmentation results of all targets are merged to obtain the target segmentation image.

5. The medical image segmentation method as described in claim 1, characterized in that, Step S6 specifically includes the following steps: S601: Take the complement of the foreground category segmentation image and the target segmentation image; S602: Cluster all pixels in the complement set according to the category of the predicted target; S603: Merge the results into the target segmentation image to obtain the predicted image.

6. A medical image segmentation system, characterized in that, The system for performing the medical image segmentation method as described in any one of claims 1 to 5 includes: The image acquisition module is used to acquire medical images as input images to be segmented; The foreground category segmentation module is used to extract the foreground from the input image to be segmented, and obtain the foreground category segmentation image to be processed; The pose feature calculation module is used to calculate the pose features of targets in the foreground category segmentation image to be processed, and obtain a series of pose features of the targets to be segmented. The pose normalization segmentation model is used to identify and adjust the pose of the input image to be segmented by using the pose features of a series of targets to be segmented, and to extract the targets from the pose-adjusted input image; the extracted targets are then mapped back to the pose of the corresponding targets in the input image to be segmented to obtain a series of target segmentation results. The pose normalization segmentation module is used to map a series of target segmentation results onto the input image to be segmented, so as to obtain the target segmented image; The image integration module is used to segment images according to the foreground category to be processed, integrate the target segmented images, and finally obtain the predicted image.

7. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the medical image segmentation method as described in any one of claims 1 to 5.

Images