An Image Segmentation Method and Device Based on U-Shaped Siamese Network

Abstract

The present invention relates to an image segmentation method and apparatus based on a U-shaped twin network. The method includes: obtaining a target image to be segmented, where the target image is a chronic stroke imaging image; constructing and training a segmentation model based on the morphological structure of brain tissue and machine learning, and using the trained segmentation model to segment the brain tissue region of the target image; using the trained deep twin network to segment the lesions of the target image after the brain tissue region is segmented. The present invention processes the lesion image in stages through a deep twin network model and machine learning, and uses the difference between normal images and lesion images in the morphology of brain tissue as features to train the recognition model, thereby improving the robustness and accuracy of the recognition model.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical imaging and deep learning, and particularly relates to an image segmentation method and device based on a U-shaped twin network. Background Art

[0002] Chronic stroke usually causes sequelae such as hemiplegia, hemianopia, aphasia, etc. Clinically, doctors usually customize and modify treatment plans only by observing the patient's physical condition based on experience, and cannot provide quantitative diagnostic basis for patients. This situation will reduce the patient's cooperation and result in poor treatment effects. Computer-aided diagnosis methods based on imaging can greatly reduce the labor required for manual annotation and provide quantitative diagnostic basis for clinicians.

[0003] The Ischemic Stroke Lesion Segmentation (ISLES) challenge project provides multi-modal MRI stroke cases, and many algorithms with good accuracy have been proposed by researchers in this project. For example, the Choi team, the champion of this project in 2017, used a residual U-shaped network and a spatial-based pyramid pooling technique to achieve stroke lesion segmentation on a small dataset. Dolz et al. proposed a U-net model based on dense multi-path multi-modal to improve the multi-modal fusion technique and thus improve the algorithm segmentation accuracy; Qi et al. proposed an x-net network structure combined with a feature similarity model (FSM) to capture long-range dependence information in the network; Yang et al. designed a CLCI-net model to obtain cross-information between network layers and context information of images. Disadvantages of the prior art:

[0004] (1) Existing research mainly emphasizes obtaining high-precision lesion segmentation results, which requires providing multi-modal images including perfusion imaging (PWI), functional magnetic resonance imaging (fMRI), etc., and is not suitable for the actual needs of multiple follow-up diagnoses in the treatment of chronic stroke.

[0005] (2) The algorithms only consider the imaging features of the cases and do not take into account the imaging specificity and clinical knowledge of the diseases, resulting in low segmentation accuracy of these algorithms at present and unable to be applied clinically. Summary of the Invention

[0006] To improve the accuracy of chronic stroke lesion image segmentation and reduce the problem of requirements for target images, in the first aspect of the present invention, an image segmentation method based on a U-shaped twin network is provided, including: obtaining a target image to be segmented, where the target image is a chronic stroke imaging image; constructing and training a segmentation model based on the morphological structure of the brain tissue and machine learning, and using the trained segmentation model to perform brain tissue region segmentation on the target image; using the trained deep twin network to perform lesion segmentation on the target image after brain tissue region segmentation.

[0007] In some embodiments of the present invention, the trained segmentation model is trained through the following steps: obtaining multiple brain tissue images, using the unlabeled brain tissue images and the labeled atlas among them as samples and labels respectively to construct a data set; wherein the labeled atlas includes at least one unlabeled brain tissue image and its corresponding label image of the labeled or segmented brain tissue region; in each round of training, registering the sample image with the corresponding atlas, and calculating the confidence level transmitted by the label image according to the similarity between each atlas image and the target image, the morphological characteristics of the target tissue, and the spatial characteristics of the target tissue, and extracting the morphological feature vector of the target tissue according to it; calculating the norm of the morphological feature vector of the target tissue; traversing all the sample images in the data set until the norm of the morphological feature vector of the target tissue reaches a preset condition, to obtain the trained segmentation model.

[0008] Further, calculating the confidence level transmitted by the label image according to the similarity between each atlas image and the target image, the morphological characteristics of the target tissue, and the spatial characteristics of the target tissue includes: calculating the distance similarity and gray level similarity between each voxel in the target image and each atlas image, and calculating the image similarity between the target image and each atlas image according to them; based on the image similarity between the target image and each atlas image, calculating the confidence matrix between the target image and all atlas images.

[0009] In some embodiments of the present invention, the deep siamese network is trained through the following steps: obtaining multiple brain tissue images, using the normal images and the lesion images among them as samples, and using the difference information between each normal image and its corresponding lesion image as a label to construct a training data set; using the training data set to train the deep siamese U-shaped network until its error is lower than the threshold and tends to be stable, to obtain the trained deep siamese U-shaped network.

[0010] Further, obtaining multiple brain tissue images, using the normal images and the lesion images among them as samples, and using the difference information between each normal image and its corresponding lesion image as a label to construct a training data set includes: determining the scanning rule of the sample image, and selecting one or more normal images from the brain tissue image database as standard templates according to it; matching each normal image with one or more lesion images to obtain the difference information between each normal image and its corresponding lesion image.

[0011] In the above embodiments, the target image after lesion segmentation is used to optimize the deep siamese network according to the disease evolution or artificial experience.

[0012] In a second aspect of the present invention, there is provided an image segmentation device based on a U-shaped twin network, comprising: an acquisition module for acquiring a target image to be segmented, where the target image is a chronic stroke imaging image; a first segmentation module for constructing and training a segmentation model based on the morphological structure of the brain tissue and machine learning, and using the trained segmentation model to segment the brain tissue region of the target image; and a second segmentation module for using the trained deep twin network to segment the lesions of the target image after the brain tissue region is segmented.

[0013] In a third aspect of the present invention, there is provided an electronic device, comprising: one or more processors; a storage device for storing one or more programs, which when executed by the one or more processors cause the one or more processors to implement the image segmentation method based on the U-shaped twin network provided in the first aspect of the present invention.

[0014] In a fourth aspect of the present invention, there is provided a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the image segmentation method based on the U-shaped twin network provided in the first aspect of the present invention.

[0015] The beneficial effects of the present invention are as follows:

[0016] The present invention processes the lesion image in stages through a deep twin network model and machine learning, and uses the difference between normal images and lesion images in the morphology of the brain tissue as features to train the recognition model, thereby improving the robustness and accuracy of the recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the basic process of the image segmentation method based on the U-shaped twin network in some embodiments of the present invention;

[0018] Figure 2 is a schematic diagram of the training or recognition process of the segmentation model in some embodiments of the present invention;

[0019] Figure 3 is a schematic diagram of the training or recognition process of the deep twin network model in some embodiments of the present invention;

[0020] Figure 4 is a schematic diagram of the structure of the image segmentation device based on the U-shaped twin network in some embodiments of the present invention;

[0021] Figure 5 is a schematic diagram of the structure of the electronic device in some embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The principles and features of the present invention will be described below in conjunction with the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0023] Referring Figure 1 , in the first aspect of the present invention, an image segmentation method based on a U-shaped Siamese network is provided, including: S100. Obtain a target image to be segmented, where the target image is a chronic stroke imaging image; S200. Construct and train a segmentation model based on the morphological structure of the brain tissue and machine learning, and use the trained segmentation model to segment the brain tissue region of the target image; S300. Use the trained deep Siamese network to segment the lesions of the target image after the brain tissue region is segmented.

[0024] Referring Figure 2 , in step S200 of some embodiments of the present invention, the trained segmentation model is trained through the following steps: S201. Obtain multiple brain tissue images, and use the unlabeled brain tissue images and the labeled atlases therein as samples and labels respectively to construct a data set; where the labeled atlases include at least one unlabeled brain tissue image and its corresponding label image of the labeled or segmented brain tissue region; specifically, prepare a set of labeled atlas sets for training the model for brain tissue region segmentation. Among them, each atlas is determined by a set of brain MR scan images and corresponding expert labels. Optionally, a public data set for brain tissue segmentation is used as the standard healthy atlas resource.

[0025] It should be understood that in step S100, actually, the atlases in the atlas data set are registered to the target image space by a non-rigid registration method to provide prior information of the label image; then, according to the similarity between the atlas image and the target image, the morphological features of the target tissue, and the spatial characteristics of the target tissue, the confidence degree transmitted by the label image is calculated, and the morphological feature vector of the target tissue is extracted according to this confidence degree.

[0026] Therefore, in step S202: In each round of training, register the sample image with the corresponding atlas, and calculate the confidence degree transmitted by the label image according to the similarity between each atlas image and the target image, the morphological features of the target tissue, and the spatial characteristics of the target tissue, and extract the morphological feature vector of the target tissue according to it; calculate the norm of the morphological feature vector of the target tissue;

[0027] Further, in step S202, calculating the confidence of the label image transfer according to the similarity between each atlas image and the target image, the morphological features of the target tissue, and the spatial characteristics of the target tissue includes: S2021. Calculating the distance similarity and gray-scale similarity between the target image and each voxel in each atlas image, and calculating the image similarity between the target image and each atlas image based on these similarities; S2022. Calculating the confidence matrix of the target image and all atlas images based on the image similarity between the target image and each atlas image.

[0028] In an embodiment of the segmentation model in step S202 above, the input is: the atlas image I registered to the target image i , the target image to be segmented: the voxel x ∈ Ω in the region of interest of the target tissue to be segmented; the output is: the morphological feature vector f(x) of the target tissue. The specific calculation steps of the morphological feature vector of the target tissue are as follows:

[0029] Step 1: Calculate the distance similarity d1 of voxel x according to the formula

[0030] Step 2: Calculate the gray-scale similarity d2 of voxel x according to the formula d2(x) = |I i (x) - T(x)| * g s , where g s is the Gaussian kernel;

[0031] Step 3: Calculate the similarity s between the voxel x ∈ Ω in the region of interest of the target tissue to be segmented in the i-th atlas and the target image according to the formula i , where ε is a very small parameter used to prevent the denominator of the formula from being 0;

[0032] Step 4: Calculate the confidence matrix c at voxel x according to the formula n and C(x) = [c1(x) c2(x) … c i (x)];

[0033] Step 5: Calculate the confidence probability matrix P at voxel x according to the formula c and P n (x) = [p1(x) p2(x) … p c (x)];

[0034] Step 6: Calculate the morphological feature vector of the target tissue according to the formula f(x) = ||p c (x) · L(x)||, where L(x) represents the label vector, and each label vector includes the category of the lesion image represented by the label and the logical truth value.

[0035] S203. Traverse all sample images in the dataset until the norm of the target tissue morphological feature vector reaches a preset condition, and obtain a trained segmentation model. Specifically, using the target tissue morphological feature vector, the image detail information and label information provided by the atlas dataset, a machine learning method is used to train a brain tissue region segmentation model. Among them, a random forest can be used as the machine learning classifier to make the model faithfully reflect the data characteristics and avoid overfitting.

[0036] It can be understood that step S200 mainly includes three main steps: image registration, generating a feature vector based on morphological confidence, training of the model, and anatomical group region segmentation. First, the spatial feature information obtained in the training stage is combined with the image information to be segmented to extract the feature vector. Then, the segmentation result of the brain tissue region can be obtained by using the trained brain tissue region segmentation model.

[0037] Reference Figure 3 , in step S300 of some embodiments of the present invention, the deep siamese network is trained through the following steps: S301. Obtain multiple brain tissue images, use the normal images and diseased images among them as samples, and use the difference information between each normal image and its corresponding diseased image as a label to construct a training dataset; S302. Use the training dataset to train the deep siamese U-shaped network until its error is lower than the threshold and tends to be stable, and obtain a trained deep siamese U-shaped network.

[0038] Further, in step S301, the obtaining multiple brain tissue images, using the normal images and diseased images among them as samples, and using the difference information between each normal image and its corresponding diseased image as a label to construct a training dataset includes: determining the scanning rules of the sample images, and selecting one or more normal images from the brain tissue image database as standard templates according to them; matching each normal image with one or more diseased images to obtain the difference information between each normal image and its corresponding diseased image.

[0039] In a specific embodiment based on the above steps, the following steps are included: First, select a standard template from the healthy case database according to the scanning rules of the training images (including information such as the model of the scanning instrument and the scanning slice thickness). And register this template onto the training image to provide it with standard reference information. Find the area where lesions may occur as the region of interest (ROI) to be processed according to the differences between the training images with lesions and the standard images. Send the template image (normal brain tissue image) and the training image (lesion image) into the deep twin U-shaped network simultaneously. Use the twin network module to obtain the differences between the template image and the training image, and this difference is the prompt information of the lesions. Take this difference as feature information and transmit it to the end of the network through the skip connections of the U-shaped network. Enable the deep twin U-shaped network to simultaneously obtain the similarity of the lesions, the spatial position of the target area, and the detailed information of the image. In the prediction stage of the deep twin network (i.e., the segmentation stage of the lesions): As in the training stage, first, a standard template needs to be selected from the healthy case database, and this template is registered onto the image to be segmented to provide it with reference information. Find the area where lesions may occur as the region of interest to be processed according to the differences between the image to be segmented and the standard image. Finally, send the template image and the image to be segmented into the deep twin U-shaped network simultaneously to obtain the final segmentation result.

[0040] In the above embodiment, the target image after lesion segmentation is used to optimize the deep twin network according to the disease evolution or artificial experience. Specifically, based on the characteristics that chronic stroke requires repeated and multiple observations and treatments, after using the deep twin U-shaped network for lesion segmentation, the segmentation results are applied to clinical cases and combined with expert opinions to optimize the model and conduct prognostic assessment of the disease development. It specifically includes three parts: First, summarize the influence of parameter changes in the experiment on the model results to improve the robustness of the deep twin U-shaped network. Deeply analyze the characteristics of this network to provide a theoretical basis for transplantation to the image segmentation of other related diseases. Second, hand over the lesion area obtained by the automatic segmentation algorithm to the doctor, and the imaging expert corrects the automatic segmentation results. The corrected results are fed back into the deep twin U-shaped network model to specifically optimize the model for chronic stroke cases. The improved deep twin network model to be designed is as Figure 3 shown. Third, combine the doctor's conclusions on clinical cases to find the relationship between the size, shape, gray scale, etc. of the imaging chronic stroke lesions and the actual clinic, and provide qualitative data for the doctor to assist in diagnosis and treatment.

[0041] Embodiment 2

[0042] Reference Figure 4, the second aspect of the present invention provides an image segmentation device 1 based on a U-shaped twin network, including: an acquisition module 11 for acquiring a target image to be segmented, where the target image is a chronic stroke imaging image; a first segmentation module 12 for constructing and training a segmentation model based on the morphological structure of brain tissue and machine learning, and using the trained segmentation model to segment the brain tissue region of the target image; a second segmentation module 13 for using the trained deep twin network to segment the lesions of the target image after brain tissue region segmentation.

[0043] Further, the first segmentation module 12 includes: an acquisition unit for acquiring multiple brain tissue images, using the unlabeled brain tissue images and the labeled atlases among them as samples and labels respectively to construct a data set; where the labeled atlases include at least one unlabeled brain tissue image and its corresponding label image of the labeled or segmented brain tissue region; a calculation unit for registering the sample image with the corresponding atlas in each round of training, and calculating the confidence transferred by the label image according to the similarity between each atlas image and the target image, the morphological characteristics of the target tissue, and the spatial characteristics of the target tissue, and extracting the morphological feature vector of the target tissue according to it; calculating the norm of the morphological feature vector of the target tissue; a traversal module for traversing all sample images in the data set until the norm of the morphological feature vector of the target tissue reaches a preset condition to obtain the trained segmentation model.

[0044] Embodiment 3

[0045] Reference Figure 5 , the third aspect of the present invention provides an electronic device, including: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, enabling the one or more processors to implement the image segmentation method based on the U-shaped twin network in the first aspect of the present invention.

[0046] The electronic device 500 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 502 or the program loaded from the storage device 508 into the random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. The input / output (I / O) interface 505 is also connected to the bus 504.

[0047] Typically, the following devices can be connected to the I / O interface 505: an input device 506 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 508 including, for example, a hard disk, etc.; and a communication device 509. The communication device 509 can allow the electronic device 500 to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 5 an electronic device 500 with various devices is shown, it should be understood that it is not required to implement or have all the shown devices. More or fewer devices can be alternatively implemented or had. Figure 5 Each block shown in

[0048] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product that includes a computer program carried on a computer-readable medium, and the computer program includes program code for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by a processing device 501, the above-mentioned functions defined in the methods of embodiments of the present disclosure are performed. It should be noted that the computer-readable medium described in embodiments of the present disclosure can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In embodiments of the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In embodiments of the present disclosure, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and the computer-readable signal medium can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0049] The above computer-readable medium can be included in the above electronic device; or it can exist separately and not be assembled into the electronic device. The above computer-readable medium carries one or more computer programs, and when the above one or more programs are executed by the electronic device, the electronic device is caused to:

[0050] Computer program code for performing the operations of the embodiments of the present disclosure may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, Python, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may execute entirely on the user's computer, partially on the user's computer, execute as a stand-alone software package, execute partially on the user's computer and partially on a remote computer, or execute entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0051] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should be noted that each block in the block diagram and / or flowchart, and the combinations of blocks in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system that performs the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.

[0052] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. An image segmentation method based on a U-shaped twin network, characterized in that, Including: Obtain a target image to be segmented, where the target image is a chronic stroke imaging image; Construct and train a segmentation model based on the morphological structure of brain tissue and machine learning, and use the trained segmentation model to segment the brain tissue region of the target image; The trained segmentation model is trained through the following steps: Obtain multiple brain tissue images, and use the unlabeled brain tissue images and the labeled atlases among them as samples and labels respectively to construct a data set; Where the labeled atlas includes at least one unlabeled brain tissue image and its corresponding labeled or segmented brain tissue region label image; In each round of training, register the sample image with the corresponding atlas, and calculate the confidence transferred by the label image according to the similarity between each atlas image and the target image, the morphological characteristics of the target tissue, and the spatial characteristics of the target tissue: Calculate the distance similarity and gray similarity between each voxel in the target image and each atlas image, and calculate the image similarity between the target image and each atlas image based on this; Based on the image similarity between the target image and each atlas image, calculate the confidence matrix between the target image and all atlas images, and extract the morphological feature vector of the target tissue according to the confidence; Calculate the norm of the morphological feature vector of the target tissue; Traverse all sample images in the data set until the norm of the morphological feature vector of the target tissue reaches a preset condition to obtain a trained segmentation model; Use the trained deep twin network to perform lesion segmentation on the target image after brain tissue region segmentation: Obtain multiple brain tissue images, use the normal images and lesion images among them as samples, and use the difference information between each normal image and its corresponding lesion image as a label to construct a training data set; Use the training data set to train the deep twin U-shaped network until its error is lower than the threshold and tends to be stable to obtain a trained deep twin U-shaped network.

2. The image segmentation method based on the U-shaped twin network according to claim 1, wherein The step of obtaining multiple brain tissue images, using the normal images and lesion images among them as samples, and using the difference information between each normal image and its corresponding lesion image as a label to construct a training data set includes: Determine the scanning rules of the sample images, and select one or more normal images from the brain tissue image database as standard templates according to them; Match each normal image with one or more lesion images to obtain the difference information between each normal image and its corresponding lesion image.

3. The image segmentation method based on the U-shaped twin network according to any one of claims 1 to 2, characterized in that Also including: Optimize the deep twin network according to the disease evolution or artificial experience of the target image after lesion segmentation.

4. An image segmentation device based on a U-shaped twin network, characterized in that, Including: An acquisition module for obtaining a target image to be segmented, where the target image is a chronic stroke imaging image; The first segmentation module is used to construct and train a segmentation model based on the morphological structure of the brain tissue and machine learning, and use the trained segmentation model to segment the brain tissue region of the target image. The trained segmentation model is trained through the following steps: Obtain multiple brain tissue images, and use the unlabeled brain tissue images and the labeled atlases among them as samples and labels respectively to construct a dataset. The labeled atlas includes at least one unlabeled brain tissue image and its corresponding label image of the labeled or segmented brain tissue region. In each round of training, register the sample image with the corresponding atlas, and calculate the confidence transmitted by the label image according to the similarity between each atlas image and the target image, the morphological characteristics of the target tissue, and the spatial characteristics of the target tissue: Calculate the distance similarity and gray-scale similarity between each voxel in the target image and each atlas image, and calculate the image similarity between the target image and each atlas image based on this; Based on the image similarity between the target image and each atlas image, calculate the confidence matrix between the target image and all atlas images, and extract the morphological feature vector of the target tissue according to the confidence; Calculate the norm of the morphological feature vector of the target tissue. Traverse all sample images in the dataset until the norm of the morphological feature vector of the target tissue reaches a preset condition, and obtain the trained segmentation model. The second segmentation module is used to segment the lesions of the target image after the brain tissue region is segmented by using the trained deep twin network: Obtain multiple brain tissue images, and use the normal images and the lesion images among them as samples, and use the difference information between each normal image and its corresponding lesion image as a label to construct a training dataset; Use the training dataset to train the deep twin U-shaped network until its error is lower than the threshold and tends to be stable, and obtain the trained deep twin U-shaped network.

5. An electronic device, comprising: One or more processors; A storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the U-shaped twin network-based image segmentation method according to any one of claims 1 to 3.

6. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by the processor, it implements the U-shaped twin network-based image segmentation method according to any one of claims 1 to 3.

Images