A semi-supervised medical image segmentation method and system based on dual image filtering

Abstract

The present invention provides a dual filtering semi-supervised medical image segmentation method and system based on tree representation, comprising: converting the original medical image sample to obtain a filtered image that retains the key topological area after dual filtering in the tree representation space; inputting the labeled filtered image into two preset image segmentation neural networks with different initializations for independent training, inputting the unlabeled dual filtering image into two preset image segmentation neural networks with different initializations to generate pseudo label values ​​corresponding to the respective networks, using the pseudo labels of the two preset networks to supervise the training of the other preset network to obtain a semi-supervised medical image segmentation model; inputting the medical image to be segmented into the semi-supervised medical image segmentation model, and outputting the image segmentation result. The present invention can retain the key topological area of ​​the image without adding new object edges by adopting the filtering method based on tree representation, and has achieved significant improvement in semi-supervised medical image segmentation.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a semi-supervised medical image segmentation method and system based on dual image filtering. Background Art

[0002] In the field of medical image processing, a large number of advanced medical imaging technologies such as ultrasound detection, computed tomography (CT) and magnetic resonance imaging (MRI) are used to assist in the diagnosis of diseases involving organs such as the heart, liver, and kidneys. These non-invasive detection methods enable doctors to gain an in-depth understanding of the specific conditions of the patient's organs without causing any harm to the patient, and have therefore been widely adopted.

[0003] With the continuous progress of deep learning technology, important breakthroughs have been made in the field of medical image segmentation. At present, it has become a reality to use deep learning technology to assist doctors in diagnosis. However, the main challenge of deep learning in medical image analysis is that it requires a large number of samples labeled by professional doctors for training, and these labeling processes are time-consuming and expensive. Therefore, a semi-supervised medical image segmentation method that combines a large number of unlabeled images with a small number of labeled images has become a solution with broad potential and prospects. Organs and lesions in medical images generally have specific shapes. Existing semi-supervised segmentation technologies treat the medical image segmentation problem as an independent pixel-level classification task without considering these structural information, resulting in anatomically unreasonable segmentation results. Summary of the invention

[0004] The present invention provides a semi-supervised medical image segmentation method and system based on dual image filtering, which are used to solve the defects of the semi-supervised segmentation technology used in the prior art for medical image segmentation.

[0005] In a first aspect, the present invention provides a semi-supervised medical image segmentation method based on dual image filtering, comprising:

[0006] Obtaining raw medical image samples;

[0007] The original medical image samples are subjected to random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion to obtain a filtered image that retains the key topological area after tree representation space dual filtering;

[0008] Input the labeled filtered images into two preset image segmentation neural networks with different initializations for independent training, input the unlabeled dual filtered images into two preset image segmentation neural networks with different initializations to generate pseudo-label values ​​corresponding to the respective networks, and use the pseudo-labels of the two preset image segmentation neural networks to supervise the training of another preset network to obtain a semi-supervised medical image segmentation model;

[0009] The medical image to be segmented is input into the semi-supervised medical image segmentation model, and the image segmentation result is output.

[0010] According to a semi-supervised medical image segmentation method based on dual image filtering provided by the present invention, the original medical image sample is subjected to random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion to obtain a filtered image retaining key topological areas after tree representation space dual filtering, including:

[0011] The original medical image samples are randomly transformed by using monotonically increasing gray value mapping, and then the transformed images are converted into maximum trees and minimum trees based on inclusion relations;

[0012] Delete nodes in the maximum tree and the minimum tree that are smaller than a preset area threshold, and obtain a new maximum tree and a new minimum tree;

[0013] Deleting tree nodes whose corresponding parent nodes have only a single child node in the new maximum tree and the new minimum tree to obtain a simplified maximum tree and a simplified minimum tree;

[0014] Reconstructing the simplified maximum tree into a pixel representation image to obtain a filtered image with grayscale value enhancement and topological structure preservation;

[0015] The simplified minimum tree is reconstructed into a pixel representation image to obtain a filtered image with reduced grayscale value and preserved topological structure.

[0016] According to a semi-supervised medical image segmentation method based on dual image filtering provided by the present invention, a labeled filtered image is input into two preset image segmentation neural networks with different initializations for independent training, and an unlabeled dual filtered image is input into two preset image segmentation neural networks with different initializations to generate pseudo label values ​​corresponding to the respective networks, and the pseudo labels of the two preset image segmentation neural networks are used to supervise the training of another preset network to obtain a semi-supervised medical image segmentation model, including:

[0017] Inputting the labeled grayscale value-enhanced filtered image with topology-preserving structure and the grayscale value-reduced filtered image with topology-preserving structure into two preset image segmentation neural networks for independent training, respectively, wherein the two preset image segmentation neural networks are neural network structures with the same structure and different initializations;

[0018] The unlabeled filtered image with grayscale value increased and topology-retained and the filtered image with grayscale value reduced and topology-retained are respectively subjected to the two preset image segmentation neural networks to obtain the pseudo label value;

[0019] The semi-supervised medical image segmentation model is obtained by using the pseudo labels of two preset image segmentation neural networks to supervise another preset image segmentation neural network for training.

[0020] According to a semi-supervised medical image segmentation method based on dual image filtering provided by the present invention, a filtered image with grayscale value enhancement and topology retention and a filtered image with grayscale value reduction and topology retention are respectively input into two preset image segmentation neural networks for independent training, including:

[0021] The preset image segmentation neural network is trained by using the labeled filtered image and adopting the cross entropy loss function and the Dice loss function.

[0022] According to a semi-supervised medical image segmentation method based on dual image filtering provided by the present invention, the unlabeled grayscale value-enhanced topology-preserving filtered image and the grayscale value-reduced topology-preserving filtered image are respectively obtained by two preset image segmentation neural networks to obtain the pseudo label value, including:

[0023] The network output probability value is processed using the maximum independent variable point set Argmax function to obtain the pseudo label value.

[0024] According to a semi-supervised medical image segmentation method based on dual image filtering provided by the present invention, the semi-supervised medical image segmentation model is obtained by using the pseudo labels of two preset image segmentation neural networks to supervise another preset image segmentation neural network for training, including:

[0025] The pseudo-label values ​​of each of the preset image segmentation neural networks are used to supervise the output probability value of another image segmentation neural network, and the cross entropy loss function is used to train the unlabeled filtered image data to output the semi-supervised medical image segmentation model.

[0026] In a second aspect, the present invention further provides a semi-supervised medical image segmentation system based on dual image filtering, comprising:

[0027] An acquisition module, used for acquiring original medical image samples;

[0028] An image filtering module performs random transformation, tree representation conversion, tree representation dual filtering, and pixel representation conversion on the original medical image samples to obtain a filtered image that retains the key topological area after tree representation space dual filtering;

[0029] The training module inputs the labeled filtered images into two preset image segmentation neural networks with different initializations for independent training, and inputs the unlabeled dual filtered images into two preset image segmentation neural networks with different initializations to generate pseudo-label values ​​corresponding to the respective networks, and uses the pseudo-labels of the two preset image segmentation neural networks to supervise the training of another preset network to obtain a semi-supervised medical image segmentation model;

[0030] The segmentation module is used to input the medical image to be segmented into the semi-supervised medical image segmentation model and output the image segmentation result.

[0031] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the semi-supervised medical image segmentation method based on dual image filtering as described above is implemented.

[0032] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the semi-supervised medical image segmentation methods based on dual image filtering as described above.

[0033] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned semi-supervised medical image segmentation methods based on dual image filtering.

[0034] The semi-supervised medical image segmentation method and system based on dual image filtering provided by the present invention can better utilize the structural priors of medical image organs and lesions by adopting a tree-representation-based filtering method, can retain the key topological areas of the image, and does not add new object edges, thereby achieving significant improvement in semi-supervised medical image segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 It is a flow chart of a semi-supervised medical image segmentation method based on dual image filtering provided by the present invention;

[0037] Figure 2 It is a logic diagram of the structure-preserving dual image filtering method in the semi-supervised medical image segmentation task provided by the present invention;

[0038] Figure 3 It is a structural schematic diagram of a semi-supervised medical image segmentation system based on dual image filtering provided by the present invention;

[0039] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Figure 1 is a flow chart of a semi-supervised medical image segmentation method based on dual image filtering provided by an embodiment of the present invention, such as Figure 1 As shown, including:

[0042] Step 100: Obtaining an original medical image sample;

[0043] Step 200: performing random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion on the original medical image sample to obtain a filtered image retaining the key topological area after tree representation space dual filtering;

[0044] Step 300: inputting the labeled filtered image into two preset image segmentation neural networks with different initializations for independent training, inputting the unlabeled dual filtered image into two preset image segmentation neural networks with different initializations to generate pseudo label values ​​corresponding to the respective networks, and using the pseudo labels of the two preset image segmentation neural networks to supervise the training of another preset network to obtain a semi-supervised medical image segmentation model;

[0045] Step 400: input the medical image to be segmented into the semi-supervised medical image segmentation model, and output the image segmentation result.

[0046] The embodiment of the present invention targets the acquired original medical image samples, performs dual image filtering processing for structure preservation in the semi-supervised medical image segmentation task, including random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion, to obtain a filtered image that retains the key topological area after the tree representation space dual filtering, and inputs the labeled filtered image into two preset image segmentation neural networks with different initializations for independent training. The unlabeled dual filtered image is input into two preset image segmentation neural networks with different initializations to generate pseudo-label values ​​corresponding to the respective networks, and the pseudo-labels of the two preset networks are used to supervise the other preset network for training, to obtain a semi-supervised medical image segmentation model. In practical applications, the medical image to be segmented is input into the trained semi-supervised medical image segmentation model to obtain the image segmentation result.

[0047] Specifically, if Figure 2 As shown, the dual structure-preserving image filtering method (DSPIF) for semi-supervised medical image segmentation tasks proposed in the embodiment of the present invention has the following four stages:

[0048] In the first stage, the original image is randomly transformed (Monotone change) using monotonically increasing gray value mapping, and then the transformed image is converted into a maximum tree (Max-tree) and a minimum tree (Min-tree) based on a tree structure representation;

[0049] In the second stage, the nodes whose areas are smaller than a given threshold in the Max-tree and Min-tree are deleted to obtain new Max-trees and Min-trees, and then the nodes whose parent nodes have only a single child node in the new Max-tree and Min-tree are deleted to obtain simplified Max-trees and Min-trees, namely Filtered Max-trees and Filtered Min-trees.

[0050] In the third stage, the simplified Max-tree and Min-tree are reconstructed into images represented by pixels. The image reconstructed by the simplified Max-tree is called the grayscale value-enhanced topology-preserving filtered image (Upper Structure Preserving Image Filtering, USPIF), and the image reconstructed by the simplified Min-tree is called the grayscale value-reduced topology-preserving filtered image (Lower Structure Preserving Image Filtering, LSPIF);

[0051] In the fourth stage, USPIF and LSPIF are randomly used as two network inputs with the same structure but different initializations. The pseudo labels generated by the two networks are used to supervise the unlabeled data, and the labeled data are supervised by manual annotations to train the network, thus obtaining a semi-supervised medical image segmentation model.

[0052] It should be noted that, in the embodiment of the present invention, the generation process of Max-tree and Min-tree is invariant to the monotonically increasing gray value mapping, and deleting a node whose parent node has only a single child node in the Max-tree and Min-tree space will not change the topological structure of the space, so this method is a structure-preserving dual image filtering method.

[0053] The present invention designs a filtering method that maintains key topological structures in a tree representation space by utilizing prior structural information of medical organs and lesions, thereby effectively improving the semi-supervised segmentation effect of medical images and significantly improving the generalization ability of the network.

[0054] Based on the above embodiment, the original medical image sample is subjected to random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion to obtain a filtered image retaining the key topological area after tree representation space dual filtering, including:

[0055] The original medical image samples are randomly transformed by using monotonically increasing gray value mapping, and then the transformed images are converted into maximum trees and minimum trees based on inclusion relations;

[0056] Delete nodes in the maximum tree and the minimum tree that are smaller than a preset area threshold, and obtain a new maximum tree and a new minimum tree;

[0057] Deleting tree nodes whose corresponding parent nodes have only a single child node in the new maximum tree and the new minimum tree to obtain a simplified maximum tree and a simplified minimum tree;

[0058] Reconstructing the simplified maximum tree into a pixel representation image to obtain a filtered image with grayscale value enhancement and topological structure preservation;

[0059] The simplified minimum tree is reconstructed into a pixel representation image to obtain a filtered image with reduced grayscale value and preserved topological structure.

[0060] Specifically, for the processing of the original medical image sample, the embodiment of the present invention first uses a monotonically increasing gray value mapping to randomly transform the image, and then converts the transformed image into a tree-structured representation Max-tree and Min-tree. The monotonically increasing gray value mapping can be any monotonically increasing mapping function suitable for image processing, and the method of converting the image into Max-tree and Min-tree can be arbitrary.

[0061] Then, the nodes in the Max-tree and Min-tree whose areas are smaller than the given threshold are deleted to obtain new Max-tree and Min-tree, and then the nodes whose parent nodes have only a single child node in the new Max-tree and Min-tree are deleted to obtain simplified Max-tree and Min-tree. The given threshold here is also flexibly set according to image features and processing requirements.

[0062] Furthermore, the simplified Max-tree and Min-tree are reconstructed into an image represented by pixel points. The image reconstructed by the simplified Max-tree is called USPIF, and the image reconstructed by the simplified Min-tree is called LSPIF. Here, the reconstruction algorithm for reconstructing the simplified Max-tree and Min-tree into an image represented by pixel points can be arbitrary.

[0063] Based on the above embodiment, the labeled filtered image is input into two preset image segmentation neural networks with different initializations for independent training, and the unlabeled dual filtered image is respectively input into two preset image segmentation neural networks with different initializations to generate pseudo label values ​​corresponding to the respective networks, and the pseudo labels of the two preset image segmentation neural networks are used to supervise the training of another preset network to obtain a semi-supervised medical image segmentation model, including:

[0064] Inputting the labeled grayscale value-enhanced filtered image with topology-preserving structure and the grayscale value-reduced filtered image with topology-preserving structure into two preset image segmentation neural networks for independent training, respectively, wherein the two preset image segmentation neural networks are neural network structures with the same structure and different initializations;

[0065] The unlabeled filtered image with grayscale value increased and topology-retained and the filtered image with grayscale value reduced and topology-retained are respectively subjected to the two preset image segmentation neural networks to obtain the pseudo label value;

[0066] The semi-supervised medical image segmentation model is obtained by using the pseudo labels of two preset image segmentation neural networks to supervise another preset image segmentation neural network for training.

[0067] Specifically, the embodiment of the present invention independently trains two networks by randomly taking the labeled USPIF and LSPIF as inputs of networks with the same structure but different initializations, and the loss functions used for training include cross entropy segmentation loss and Dice loss segmentation loss.

[0068] For the unlabeled USPIF and LSPIF, the pseudo labels generated by the two networks are used to supervise the prediction results of the other network, and the pseudo labels used for training are obtained by performing Argmax on the network segmentation results; the neural network model is trained according to the unlabeled data, and the loss function used for training includes the cross entropy segmentation loss between the obtained pseudo labels and the network prediction results.

[0069] It should be noted that the above network input uses two networks with the same structure but different initializations, which can be any neural network structure suitable for image segmentation.

[0070] The embodiment of the present invention uses any one of the two networks to segment the input medical image to obtain the lesion area or the organ area. The entire segmentation process does not increase any time overhead and computing cost, and can be better used to assist clinicians in diagnosis.

[0071] Experiments show that the embodiment of the present invention adopts commonly used medical evaluation indicators: Dice Coefficient (Dice) and Jaccard Index (JI) as evaluation criteria, and the effect is verified on three commonly used semi-supervised medical image segmentation data sets. The experimental results show that the segmentation method of the embodiment of the present invention can significantly improve the segmentation effect of the neural network for medical images and has good generalization performance.

[0072] The semi-supervised medical image segmentation system based on dual image filtering provided by the present invention is described below. The semi-supervised medical image segmentation system based on dual image filtering described below and the semi-supervised medical image segmentation method based on dual image filtering described above can be referenced to each other.

[0073] Figure 3 is a schematic diagram of the structure of a semi-supervised medical image segmentation system based on dual image filtering provided by an embodiment of the present invention, such as Figure 3 As shown, it includes: an acquisition module 31, an image filtering module 32, a training module 33 and a segmentation module 34, wherein:

[0074] The acquisition module 31 is used to acquire original medical image samples; the image filtering module 32 is used to perform random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion on the original medical image samples to obtain a filtered image that retains the key topological area after tree representation space dual filtering; the training module 33 is used to input the labeled filtered image into two preset image segmentation neural networks with different initializations for independent training, and input the unlabeled dual filtered image into two preset image segmentation neural networks with different initializations to generate pseudo-label values ​​corresponding to the respective networks, and use the pseudo-labels of the two preset image segmentation neural networks to supervise the training of another preset network to obtain a semi-supervised medical image segmentation model; the segmentation module 34 is used to input the medical image to be segmented into the semi-supervised medical image segmentation model and output the image segmentation result.

[0075] Figure 4 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430 and a communication bus 440, wherein the processor 410, the communication interface 420 and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute a semi-supervised medical image segmentation method based on dual image filtering, the method comprising: obtaining an original medical image sample; performing random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion on the original medical image sample to obtain a filtered image retaining a key topological area after tree representation space dual filtering; inputting the labeled filtered image into two preset image segmentation neural networks with different initializations for independent training. Inputting the unlabeled dual filtered images into two preset image segmentation neural networks with different initializations to generate pseudo label values ​​corresponding to the respective networks, using the pseudo labels of the two preset image segmentation neural networks to supervise another preset network for training, and obtaining a semi-supervised medical image segmentation model; inputting the medical image to be segmented into the semi-supervised medical image segmentation model, and outputting the image segmentation result.

[0076] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0077] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the semi-supervised medical image segmentation method based on dual image filtering provided by the above methods, the method including: obtaining an original medical image sample; performing random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion on the original medical image sample to obtain a filtered image that retains the key topological area after tree representation space dual filtering; inputting the labeled filtered image into two preset image segmentation neural networks with different initializations for independent training, and inputting the unlabeled dual filtered image into two preset image segmentation neural networks with different initializations to generate pseudo-label values ​​corresponding to each network, and using the pseudo-labels of each of the two preset image segmentation neural networks to supervise another preset network for training to obtain a semi-supervised medical image segmentation model; inputting the medical image to be segmented into the semi-supervised medical image segmentation model, and outputting the image segmentation result.

[0078] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the semi-supervised medical image segmentation method based on dual image filtering provided by the above-mentioned methods, the method comprising: obtaining an original medical image sample; performing random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion on the original medical image sample to obtain a filtered image that retains the key topological area after tree representation space dual filtering; inputting the labeled filtered image into two preset image segmentation neural networks with different initializations for independent training, and inputting the unlabeled dual filtered image into two preset image segmentation neural networks with different initializations to generate pseudo-label values ​​corresponding to their respective networks, and using the pseudo-labels of each of the two preset image segmentation neural networks to supervise another preset network for training to obtain a semi-supervised medical image segmentation model; inputting the medical image to be segmented into the semi-supervised medical image segmentation model, and outputting the image segmentation result.

[0079] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0080] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A semi-supervised medical image segmentation method based on dual image filtering, characterized in that: include: Obtaining raw medical image samples; The original medical image samples are subjected to random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion to obtain a filtered image that retains the key topological area after tree representation space dual filtering; Input the labeled filtered images into two preset image segmentation neural networks with different initializations for independent training, input the unlabeled dual filtered images into two preset image segmentation neural networks with different initializations to generate pseudo-label values ​​corresponding to the respective networks, and use the pseudo-labels of the two preset image segmentation neural networks to supervise the training of another preset network to obtain a semi-supervised medical image segmentation model; Inputting the medical image to be segmented into the semi-supervised medical image segmentation model, and outputting the image segmentation result; The original medical image samples are subjected to random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion to obtain The filtered image retains key topological regions after dual filtering in the tree representation space, including: The original medical image samples are randomly transformed by using monotonically increasing gray value mapping, and then the transformed images are converted into maximum trees and minimum trees based on inclusion relations; Delete nodes in the maximum tree and the minimum tree that are smaller than a preset area threshold, and obtain a new maximum tree and a new minimum tree; Deleting the tree nodes whose corresponding parent nodes have only a single child node in the new maximum tree and the new minimum tree, to obtain a simplified maximum tree and a simplified minimum tree; Reconstructing the simplified maximum tree into a pixel representation image to obtain a filtered image with grayscale value enhancement and topological structure preservation; The simplified minimum tree is reconstructed into a pixel representation image to obtain a filtered image with reduced grayscale value and preserved topological structure.

2. The semi-supervised medical image segmentation method based on dual image filtering according to claim 1, characterized in that: The labeled dual filter images are input into two preset image segmentation neural networks with different initializations for independent training, and the unlabeled dual filter images are respectively input into two preset image segmentation neural networks with different initializations to generate pseudo label values ​​corresponding to the respective networks. The pseudo labels of the two preset image segmentation neural networks are used to supervise the training of another preset network, and a semi-supervised medical image segmentation model is obtained, including: Inputting the labeled grayscale value-enhanced filtered image with topology-preserving structure and the grayscale value-reduced filtered image with topology-preserving structure into two preset image segmentation neural networks for independent training, respectively, wherein the two preset image segmentation neural networks are neural network structures with the same structure and different initializations; The unlabeled filtered image with grayscale value increased and topology-retained and the filtered image with grayscale value reduced and topology-retained are respectively subjected to the two preset image segmentation neural networks to obtain the pseudo label value; The semi-supervised medical image segmentation model is obtained by using the pseudo labels of two preset image segmentation neural networks to supervise another preset image segmentation neural network for training.

3. The semi-supervised medical image segmentation method based on dual image filtering according to claim 2, characterized in that: The labeled grayscale value-enhanced filtered image with topology-preserving structure and the grayscale value-reduced filtered image with topology-preserving structure are respectively input into two preset image segmentation neural networks for independent training, including: The preset image segmentation neural network is trained by using the labeled filtered image and adopting the cross entropy loss function and the Dice loss function.

4. The semi-supervised medical image segmentation method based on dual image filtering according to claim 2, characterized in that: The method of obtaining the pseudo label value by using the two preset image segmentation neural networks respectively for the unlabeled grayscale value-enhanced topological structure-preserving filtered image and the grayscale value-reduced topological structure-preserving filtered image, including: The network output probability value is processed using the maximum independent variable point set Argmax function to obtain the pseudo label value.

5. The semi-supervised medical image segmentation method based on dual image filtering according to claim 4, characterized in that: The semi-supervised medical image segmentation model is obtained by using the pseudo labels of two preset image segmentation neural networks to supervise another preset image segmentation neural network for training, including: The pseudo-label values ​​of each of the preset image segmentation neural networks are used to supervise the output probability value of another image segmentation neural network, and the cross entropy loss function is used to train the unlabeled filtered image to output the semi-supervised medical image segmentation model.

6. A semi-supervised medical image segmentation system based on dual image filtering, characterized in that: include: An acquisition module, used for acquiring original medical image samples; An image filtering module, used for performing random transformation, tree representation conversion, tree representation dual filtering and pixel representation conversion on the original medical image sample to obtain a filtered image retaining key topological areas after tree representation space dual filtering; A training module is used to input the labeled filtered images into two preset image segmentation neural networks with different initializations for independent training, and input the unlabeled dual filtered images into two preset image segmentation neural networks with different initializations to generate pseudo-label values ​​corresponding to the respective networks, and use the pseudo-labels of the two preset image segmentation neural networks to supervise the training of another preset network to obtain a semi-supervised medical image segmentation model; A segmentation module, used for inputting the medical image to be segmented into the semi-supervised medical image segmentation model and outputting an image segmentation result; The image filtering module is specifically used for: The original medical image samples are randomly transformed by using monotonically increasing gray value mapping, and then the transformed images are converted into maximum trees and minimum trees based on inclusion relations; Delete nodes in the maximum tree and the minimum tree that are smaller than a preset area threshold, and obtain a new maximum tree and a new minimum tree; Deleting the tree nodes whose corresponding parent nodes have only a single child node in the new maximum tree and the new minimum tree, to obtain a simplified maximum tree and a simplified minimum tree; Reconstructing the simplified maximum tree into a pixel representation image to obtain a filtered image with grayscale value enhancement and topological structure preservation; The simplified minimum tree is reconstructed into a pixel representation image to obtain a filtered image with reduced grayscale value and preserved topological structure.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the semi-supervised medical image segmentation method based on dual image filtering as described in any one of claims 1 to 5 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the semi-supervised medical image segmentation method based on dual image filtering as described in any one of claims 1 to 5 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the semi-supervised medical image segmentation method based on dual image filtering as described in any one of claims 1 to 5 is implemented.

Images