A medical image intelligent segmentation network model and method
By introducing multi-scale convolutional modules and spatial transformation network modules into the U-Net model, the segmentation problem of organs of uneven size in medical images is solved, achieving accurate segmentation of large and small organs and improving the performance of the segmentation network.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SUPERACCURACY SCIENCE & TECHNOLOGY CO LTD
- Filing Date
- 2023-09-22
- Publication Date
- 2026-06-12
AI Technical Summary
Existing medical image segmentation network models struggle to accurately segment both large and small organs simultaneously when dealing with organs of uneven size, especially small organs, and traditional methods and deep learning models have failed to effectively optimize the segmentation of imbalanced organs.
We employ a multi-scale convolutional module and a spatial transformation network module based on the U-Net model. By gradually reducing the number of convolutional kernels and gradually increasing the number of kernels in multiple layers, we extract organ features at different scales. We then fuse feature information through the spatial transformation network module to improve the feature extraction efficiency and segmentation accuracy of small organs.
It improves the efficiency and accuracy of feature extraction and segmentation for both large and small organs, and can better handle multiple organs in complex anatomical structures such as the head and neck, enhancing the segmentation effect of small organs and the ability to identify difficult-to-learn regions.
Smart Images
Figure CN117274591B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing technology, specifically to a segmentation network model and method for medical images with uneven organ size. Background Technology
[0002] In the field of medical image segmentation (using CT images as an example), to control radiation dose and area during radiotherapy and reduce damage to normal tissues and organs, it is necessary to accurately delineate organs at risk. The accuracy and speed of manual delineation by radiologists heavily rely on expert experience, and traditional segmentation algorithms struggle to simultaneously achieve accurate segmentation of multiple organs of varying sizes in CT images, such as those in the head and neck region. The main challenge lies in the severe imbalance between large and small organs (e.g., the smallest organ occupies only 0.003% of the total 3D volume, while the parotid gland is more than 250 times larger than the lens). Current state-of-the-art segmentation neural networks perform poorly on small organs. Furthermore, limitations of CT technology and the complex anatomical structure and low-contrast organ contours of head and neck organs further complicate the process. All these factors combined make it difficult to develop a method that accurately segments both small and large organs simultaneously.
[0003] Over the past decade, numerous methods have been proposed to address the challenging problem of organ segmentation in medical images. Early methods included atlas-based approaches, dynamic contouring, and graph cutting. Atlas-based methods were often used when only a small number of labeled images were available; however, these methods rely on image registration techniques and can lead to incorrect organ segmentation if the organ is obscured by a tumor. Recently, convolutional neural networks have made revolutionary progress in many tasks due to their powerful feature representation capabilities. Current deep learning segmentation models primarily focus on improving upon the U-Net architecture, but these network models are not optimized for imbalanced organ segmentation. These networks typically perform better on large organs, while the segmentation of small organs is often sacrificed.
[0004] In view of this, this invention patent is hereby proposed. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a segmentation network model and method for addressing organ size imbalances in medical images. Specifically, the following technical solution is adopted:
[0006] A segmentation network model for medical images with uneven organ size includes a backbone segmentation network model based on the U-Net model. The backbone segmentation network model uses multi-scale convolutional layers with multi-scale convolutional modules in the encoder of the U-Net model, except for the first layer, and uses multi-scale convolutional layers with multi-scale convolutional modules in the decoder of the U-Net model, except for the last layer.
[0007] The multi-scale convolution module includes multiple convolutional layers. The number of convolutional kernels gradually decreases from the first layer to the last layer, while the number of convolutional kernels gradually increases, to extract organ features at different scales.
[0008] As an optional embodiment of the present invention, the multi-scale convolution module includes a first-level convolutional layer, a second-level convolutional layer, ..., an i-th-level convolutional layer; the first-level convolutional layer has M1 N1*N1 convolutional kernels, the second-level convolutional layer has M2 N1*N2 convolutional kernels, ..., the i-th-level convolutional layer has Mi Ni*Ni convolutional kernels, where M1 > M2 > ... > Mi, N1 < N2 < ... < Ni;
[0009] The first-level convolutional layer has M1 N1*N1 convolutional kernels connected in series, the second-level convolutional layer has M2 N2*N2 convolutional kernels connected in series, and so on, and the i-th level convolutional layer has Mi Ni*Ni convolutional kernels connected in series. The output of the first N1*N1 convolutional kernel of the first-level convolutional layer is input to the first convolutional kernel of the second to the i-th level, respectively. The last convolutional kernel of each level is connected in series. The output feature map of the intermediate convolutional kernel of the previous level is input to the first or intermediate convolutional kernel of the next level. The output feature maps of the last convolutional kernel of each level are fused to obtain the organ feature extraction map output by the multi-scale convolutional module.
[0010] As an optional embodiment of the present invention, the multi-scale convolution module includes a first-level convolutional layer, a second-level convolutional layer, and a third-level convolutional layer. The first-level convolutional layer has four 1*1 convolutional kernels for convolution operations, the second-level convolutional layer has two 3*3 convolutional kernels for convolution operations, and the third-level convolutional layer has one 5*5 convolutional kernel for convolution operations.
[0011] As an optional embodiment of the present invention, each multi-scale convolutional layer in the encoder of the backbone segmentation network model includes at least two multi-scale convolutional modules connected in series. The organ feature extraction map output by the last multi-scale convolutional module in the previous level multi-scale convolutional layer is respectively input to each multi-scale convolutional module in the next level multi-scale convolutional layer.
[0012] As an optional embodiment of the present invention, a segmentation network model for organs with size imbalance in medical images includes: Stag1, Stage2, ..., Stage(n), Stage(n+1), ..., Stage(2n-1), wherein Stag1, Stage2, ..., Stage(n-1) constitute an encoder, wherein Stag1 is the input layer, and Stage2, ..., Stage(n-1) are multi-scale convolutional layers employing multi-scale convolutional modules; Stage(n+1) ... Stage(2n-1) constitutes the decoder, Stage(2n-1) is the output layer, and Stage(n+1), ..., Stage(2n-2) are multi-scale convolutional layers employing at least two multi-scale convolutional modules; Stage(n) employs one multi-scale convolutional module, the organ feature extraction map output by Stage(n-1) is input to the multi-scale convolutional module of Stage(n), and the organ feature extraction map output by the multi-scale convolutional module of Stage(n) is input to Stage(n+1).
[0013] As an optional embodiment of the present invention, a segmentation network model for uneven organ size in medical images includes a spatial transformation network module, wherein the spatial transformation network module is serially connected between the multi-scale convolutional layers of the encoder and the multi-scale convolutional layers of the decoder at the same level.
[0014] This invention also provides a segmentation method for organs of uneven size in medical images, comprising:
[0015] The medical images are input into a segmentation network model designed for organ size imbalances in medical images;
[0016] The segmentation network model described above extracts feature information of organs of different sizes from medical images by using multi-scale convolutional layers of multi-scale convolutional modules, and outputs feature segmentation images of organs of different sizes from medical images.
[0017] As an optional embodiment of the present invention, a segmentation method for organs with uneven size in medical images includes:
[0018] The segmentation network model described above fuses the multi-scale features output from the multi-scale convolutional layers in the encoder through a spatial transformation network module. The fused features are then input into the multi-scale convolutional layers in the corresponding decoder to obtain extracted feature maps and location information for smaller organs and difficult-to-learn regions.
[0019] The present invention also provides a computer-readable storage medium storing a computer-executable program, characterized in that, when the computer-executable program is executed, it implements the aforementioned method for segmenting organs of uneven size in medical images.
[0020] The present invention also provides an electronic device, including a processor and a memory, wherein the memory is used to store a computer-executable program, and when the computer program is executed by the processor, the processor executes the aforementioned segmentation method for organs of different sizes in medical images.
[0021] This invention provides a segmentation network model for medical images with uneven organ size. For the first time, multiple serial STN modules are designed for use in medical image segmentation algorithms to extract small organs and difficult-to-identify regions at multiple levels.
[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0023] This invention proposes a new multi-scale convolutional module to replace the convolutional module in the U-Net model for a segmentation network model targeting organs of different sizes in medical images. This module can fuse features from multi-scale convolutions at different resolutions, making fuller use of feature information at different accuracies to extract richer multi-scale fusion features for both large and small organs, thus making organ segmentation more accurate.
[0024] For multi-scale convolutional modules, the number of convolutional kernels gradually decreases from the first layer to the last layer, while the number of convolutional kernels gradually increases. This can preserve more feature information of very small organs, improve the feature extraction efficiency of small and medium-sized organ networks, preserve feature information of medium and large organs, and more accurately segment organs of different sizes by combining features of different scales. Attached image description:
[0025] Figure 1 An embodiment of the present invention provides an overall structural diagram of a segmentation network model for medical images with uneven organ size.
[0026] Figure 2 A schematic diagram of the structure of a multi-scale convolution module in a segmentation network model for medical images with uneven organ size, according to an embodiment of the present invention;
[0027] Figure 3 This invention provides a schematic diagram of the spatial transformation network module in a segmentation network model for medical images with uneven organ size. Detailed Implementation
[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0029] Therefore, the following detailed description of embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely illustrates some embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0030] It should be noted that, unless otherwise specified, the embodiments and features and technical solutions in the present invention can be combined with each other.
[0031] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0032] In the description of this invention, it should be noted that the terms "upper," "lower," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. These terms are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0033] See Figure 1 As shown, this embodiment provides a segmentation network model for medical images with uneven organ size, including a backbone segmentation network model based on the U-Net model. The backbone segmentation network model uses multi-scale convolutional layers with multi-scale convolutional modules in the encoder of the U-Net model, except for the first layer, and uses multi-scale convolutional layers with multi-scale convolutional modules in the decoder of the U-Net model, except for the last layer.
[0034] The multi-scale convolution module includes multiple convolutional layers. The number of convolutional kernels gradually decreases from the first layer to the last layer, while the number of convolutional kernels gradually increases, to extract organ features at different scales.
[0035] This embodiment proposes a new multi-scale convolutional module to replace the convolutional module in the U-Net model for a segmentation network model targeting organs of different sizes in medical images. This module can fuse features from multi-scale convolutions at different resolutions, making fuller use of feature information at different accuracies to extract richer multi-scale fusion features for both large and small organs, thus making organ segmentation more accurate.
[0036] For multi-scale convolutional modules, the number of convolutional kernels gradually decreases from the first layer to the last layer, while the number of convolutional kernels gradually increases. This can preserve more feature information of very small organs, improve the feature extraction efficiency of small and medium-sized organ networks, preserve feature information of medium and large organs, and more accurately segment organs of different sizes by combining features of different scales.
[0037] See Figure 2 As shown in the figure, this embodiment provides a segmentation network model for organs with size imbalance in medical images. The multi-scale convolutional module includes a first-level convolutional layer, a second-level convolutional layer, ..., an i-th-level convolutional layer. The first-level convolutional layer has M1 N1*N1 convolutional kernels, the second-level convolutional layer has M2 N1*N2 convolutional kernels, ..., the i-th-level convolutional layer has Mi Ni*Ni convolutional kernels, where M1 > M2 > ... > Mi, and N1 < N2 < ... < Ni.
[0038] In this embodiment, the M1 N1*N1 convolutional kernels of the first-level convolutional layer are sequentially connected in series, the M2 N2*N2 convolutional kernels of the second-level convolutional layer are sequentially connected in series, and so on, and the Mi Ni*Ni convolutional kernels of the i-th-level convolutional layer are sequentially connected in series. The output of the first N1*N1 convolutional kernel of the first-level convolutional layer is input to the first convolutional kernel of the second to i-th layers respectively. The last convolutional kernels of each level of convolutional layer are sequentially connected in series. The output feature map of the intermediate convolutional kernel of the previous level of convolutional layer is input to the first or intermediate convolutional kernel of the next level of convolutional layer. The output feature maps of the last convolutional kernels of each level of convolutional layer are fused to obtain the organ feature extraction map output by the multi-scale convolutional module.
[0039] Specifically, as a specific implementation method of this embodiment, see [link to relevant documentation]. Figure 2 As shown in the figure, this embodiment is a segmentation network model for organs with uneven size in medical images. The multi-scale convolution module includes a first-level convolutional layer, a second-level convolutional layer, and a third-level convolutional layer. The first-level convolutional layer has four 1*1 convolutional kernels for convolution operations, the second-level convolutional layer has two 3*3 convolutional kernels for convolution operations, and the third-level convolutional layer has one 5*5 convolutional kernel for convolution operations.
[0040] For the multi-scale convolution module, the first convolutional layer uses four 1*1 convolutional kernels to perform convolution operations, which can retain more feature information of very small organs. The second convolutional layer uses two 3*3 convolutional kernels, which can improve the feature extraction efficiency of small and medium-sized organ networks. Finally, a 5*5 convolutional kernel is used to retain the feature information of medium and large organs. By combining features of different scales, organs of different sizes can be segmented more accurately.
[0041] See Figure 1 As shown in this embodiment, a segmentation network model for medical images with uneven organ size is described. The encoder of the backbone segmentation network model includes at least two multi-scale convolutional modules connected in series in each multi-scale convolutional layer. The organ feature extraction map output by the last multi-scale convolutional module in the previous level multi-scale convolutional layer is respectively input into each multi-scale convolutional module in the next level multi-scale convolutional layer.
[0042] As an optional implementation of this embodiment, a segmentation network model for medical image organ size imbalance includes: Stag1, Stage2, ..., Stage(n), Stage(n+1), ..., Stage(2n-1), wherein Stag1, Stage2, ..., Stage(n-1) constitute an encoder, wherein Stag1 is the input layer, and Stage2, ..., Stage(n-1) are multi-scale convolutional layers employing multi-scale convolutional modules; Stage(n+1) ... Stage(n-1) constitutes the decoder, where Stage(n-1) is the output layer, and Stage(n+1), ..., Stage(2n-2) are multi-scale convolutional layers employing at least two multi-scale convolutional modules; Stage(n) employs one multi-scale convolutional module, the organ feature extraction map output by Stage(n-1) is input to the multi-scale convolutional module of Stage(n), and the organ feature extraction map output by the multi-scale convolutional module of Stage(n) is input to Stage(n+1).
[0043] Specifically, see Figure 2As shown, this embodiment of a segmentation network model for uneven organ size in medical images includes: Stag1, Stage2, Stage3, Stage4, Stage5, Stage6, and Stage7. Stag1, Stage2, and Stage3 constitute an encoder, where Stag1 is the input layer, and Stage2 and Stage3 are multi-scale convolutional layers using multi-scale convolutional modules. Stage5, Stage6, and Stage7 constitute a decoder, where Stage7 is the output layer, and Stage5 and Stage6 are multi-scale convolutional layers using at least two multi-scale convolutional modules. Stage4 uses one multi-scale convolutional module. The organ feature extraction map output from Stage3 is input to the multi-scale convolutional module of Stage4, and the organ feature extraction map output from the multi-scale convolutional module of Stage4 is input to Stage5.
[0044] See Figure 1 As shown in the figure, the segmentation network model for the size imbalance of organs in medical images described in this embodiment includes a spatial transformation network module (STN module), which is serially connected between the multi-scale convolutional layer of the encoder and the multi-scale convolutional layer of the decoder at the same level.
[0045] The STN module consists of three parts. First, the feature map is processed by a local network to output a transformation matrix. Then, the parameters of the transformation matrix are continuously corrected using a loss function to obtain the desired affine transformation matrix. After obtaining the output feature map, the most important step is to obtain the pixel value at each location in the output feature map. Next, a parameterized network is used for sampling to obtain the position of the input feature map corresponding to the coordinates of the coordinates of the output feature map. Finally, a sampler (differential image sampling) is used to calculate the corresponding grayscale value using the desired interpolation method. This process yields feature maps and location information for smaller organs and difficult-to-learn regions. By padding the local features generated by the STN module with zeros, the output and input feature dimensions are kept consistent, making feature fusion easier. Figure 1 Three serial STN modules are used to locate small organ regions of three different scales.
[0046] For the decoder, we input the multi-scale features obtained in Stage 2 into the STN module, and then input the fused features into the corresponding decoder module. Similarly, Stage 3 performs the same operation, maximizing the learning of small organ information and difficult regions, ultimately improving the detection rate of small organs and the performance indicators of difficult-to-learn regions.
[0047] This embodiment presents a segmentation network model for medical images with uneven organ size. For the first time, multiple serial STN modules are designed for use in medical image segmentation algorithms to extract small organs and difficult-to-identify regions at multiple levels.
[0048] This embodiment also provides a segmentation method for organs of uneven size in medical images, including:
[0049] The medical image is input into a segmentation network model for medical images with uneven organ size, as described in any one of claims 1-6;
[0050] The segmentation network model described above extracts feature information of organs of different sizes from medical images by using multi-scale convolutional layers of multi-scale convolutional modules, and outputs feature segmentation images of organs of different sizes from medical images.
[0051] As an optional implementation of this embodiment, a segmentation method for organs with uneven size in medical images includes:
[0052] The segmentation network model described above fuses the multi-scale features output from the multi-scale convolutional layers in the encoder through a spatial transformation network module. The fused features are then input into the multi-scale convolutional layers in the corresponding decoder to obtain extracted feature maps and location information for smaller organs and difficult-to-learn regions.
[0053] This embodiment also provides a computer-readable storage medium storing a computer-executable program, which, when executed, implements the segmentation method for organs of uneven size in medical images as described above.
[0054] The computer-readable storage medium described in this embodiment may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable storage medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.
[0055] This embodiment also provides an electronic device, including a processor and a memory, wherein the memory is used to store a computer-executable program, and when the computer program is executed by the processor, the processor executes the aforementioned segmentation method for organs of uneven size in medical images.
[0056] The electronic device is manifested in the form of a general-purpose computing device. It may contain one or more processors that work collaboratively. This invention also does not preclude distributed processing, meaning that processors may be distributed across different physical devices. The electronic device of this invention is not limited to a single entity, but may also be the sum of multiple physical devices.
[0057] The memory stores a computer-executable program, typically machine-readable code. The computer-readable program can be executed by the processor to enable the electronic device to perform the method of the present invention, or at least some steps of the method.
[0058] The memory includes volatile memory, such as random access memory (RAM) and / or cache memory, and may also be non-volatile memory, such as read-only memory (ROM).
[0059] It should be understood that the electronic device of the present invention may also include elements or components not shown in the examples above. For example, some electronic devices also include display units such as a display screen, and some electronic devices also include human-computer interaction elements such as buttons and keyboards. Any electronic device capable of executing a computer-readable program in its memory to implement the method of the present invention or at least some steps of the method can be considered as an electronic device covered by the present invention.
[0060] From the above description of the embodiments, those skilled in the art will readily understand that the present invention can be implemented by hardware capable of executing specific computer programs, such as the system of the present invention, and the electronic processing unit, server, client, mobile phone, control unit, processor, etc. included in the system. The present invention can also be implemented by computer software that executes the methods of the present invention, for example, by control software executed by a microprocessor, electronic control unit, client, server, etc. However, it should be noted that the computer software executing the methods of the present invention is not limited to execution in one or a specific set of hardware entities; it can also be implemented in a distributed manner by unspecified hardware. For computer software, the software product can be stored in a computer-readable storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) or distributed across a network, as long as it enables electronic devices to execute the methods according to the present invention.
[0061] The above embodiments are only used to illustrate the present invention and are not intended to limit the technical solutions described herein. Although the present invention has been described in detail with reference to the above embodiments, the present invention is not limited to the specific embodiments described above. Therefore, any modifications or equivalent substitutions to the present invention, as well as all technical solutions and improvements that do not depart from the spirit and scope of the invention, are covered within the scope of the claims of the present invention.
Claims
1. A segmentation network model for medical images with uneven organ size, characterized in that, This includes a backbone segmentation network model based on the U-Net model. The backbone segmentation network model uses multi-scale convolutional layers with multi-scale convolutional modules in the encoder of the U-Net model, except for the first layer. It also uses multi-scale convolutional layers with multi-scale convolutional modules in the decoder of the U-Net model, except for the last layer. The multi-scale convolution module includes multiple levels of convolutional layers. The number of convolutional kernels in the multi-level convolutional layers gradually decreases from the first layer to the last layer, while the number of convolutional kernels gradually increases, to extract organ features at different scales. The multi-scale convolutional module includes a first-level convolutional layer, a second-level convolutional layer, ..., an i-th-level convolutional layer; the first-level convolutional layer has M1 N1*N1 convolutional kernels, the second-level convolutional layer has M2 N2*N2 convolutional kernels, ..., the i-th-level convolutional layer has Mi Ni*Ni convolutional kernels, where M1 > M2 > ... > Mi, N1 < N2 < ... < Ni; The first-level convolutional layer has M1 N1*N1 convolutional kernels connected in series, the second-level convolutional layer has M2 N2*N2 convolutional kernels connected in series, and so on, and the i-th level convolutional layer has Mi Ni*Ni convolutional kernels connected in series. The output of the first N1*N1 convolutional kernel of the first-level convolutional layer is input to the first convolutional kernel of the second to the i-th level, respectively. The last convolutional kernel of each level is connected in series. The output feature map of the intermediate convolutional kernel of the previous level is input to the first or intermediate convolutional kernel of the next level. The output feature maps of the last convolutional kernel of each level are fused to obtain the organ feature extraction map output by the multi-scale convolutional module.
2. The segmentation network model for uneven organ size in medical images according to claim 1, characterized in that, The multi-scale convolution module includes a first-level convolutional layer, a second-level convolutional layer, and a third-level convolutional layer. The first-level convolutional layer has four 1*1 convolutional kernels for convolution operations, the second-level convolutional layer has two 3*3 convolutional kernels for convolution operations, and the third-level convolutional layer has one 5*5 convolutional kernel for convolution operations.
3. The segmentation network model for uneven organ size in medical images according to claim 1, characterized in that, The encoder of the backbone segmentation network model includes at least two multi-scale convolutional modules connected in series in each multi-scale convolutional layer. The organ feature extraction map output by the last multi-scale convolutional module in the previous level multi-scale convolutional layer is respectively input to each multi-scale convolutional module in the next level multi-scale convolutional layer.
4. A segmentation network model for uneven organ size in medical images according to claim 3, characterized in that, include: Stages Stag1, Stage2, ..., Stage(n), Stage(n+1), ..., Stage(2n-1) constitute an encoder, where Stag1 is the input layer and Stage2, ..., Stage(n-1) are multi-scale convolutional layers using multi-scale convolutional modules. Stage(n+1), ..., Stage(2n-1) constitute a decoder, where Stage(2n-1) is the output layer and Stage(n+1), ..., Stage(2n-2) are multi-scale convolutional layers using at least two multi-scale convolutional modules. Stage(n) uses one multi-scale convolutional module. The organ feature extraction map output by Stage(n-1) is input to the multi-scale convolutional module of Stage(n), and the organ feature extraction map output by the multi-scale convolutional module of Stage(n) is input to Stage(n+1).
5. A segmentation network model for uneven organ size in medical images according to any one of claims 1-4, characterized in that, It includes a spatial transformation network module, which is serially connected between the multi-scale convolutional layers of the encoder and the decoder at the same level.
6. A segmentation method for organs of uneven size in medical images, characterized in that, include: The medical image is input into a segmentation network model for medical images with size-disproportionate organs, as described in any one of claims 1-5. The segmentation network model described above extracts feature information of organs of different sizes from medical images by using multi-scale convolutional layers of multi-scale convolutional modules, and outputs feature segmentation images of organs of different sizes from medical images.
7. A segmentation method for organs of uneven size in medical images according to claim 6, characterized in that, include: The segmentation network model described above fuses the multi-scale features output from the multi-scale convolutional layers in the encoder through a spatial transformation network module. The fused features are then input into the multi-scale convolutional layers in the corresponding decoder to obtain extracted feature maps and location information for smaller organs and difficult-to-learn regions.
8. A computer-readable storage medium storing a computer-executable program, characterized in that, When the computer-executable program is executed, it implements a segmentation method for organs of uneven size in medical images as described in claim 6 or 7.
9. An electronic device comprising a processor and a memory, characterized in that, The memory is used to store a computer-executable program, which, when executed by the processor, performs a segmentation method for organs of different sizes in medical images as described in claim 6 or 7.