Medical image segmentation method, system and device based on large model labeling and distributed learning, and storage medium

Abstract

The invention discloses a medical image segmentation method, system and device based on large model labeling and distributed learning and a storage medium, and belongs to the technical field of medical image segmentation. The method comprises the following steps: acquiring a target image picture, and preprocessing the target image picture to obtain a target image picture set; inputting the target image picture set into a pre-trained Auto-MedSAM model to generate a pseudo tag set of the target image; inputting the target image picture set and the pseudo label set into a pre-trained image segmentation model to obtain a segmented image of the target image; wherein the image segmentation model comprises a quality evaluation module, a feature extraction module, a random deviation module, an annotator deviation module and a fusion module. According to the method, through the Auto-MedSAM model, a pseudo label with excellent quality can be obtained through a small number of annotation images, the selection range of a data set is expanded, and the manual annotation cost is reduced; the quality evaluation module solves the problems that the quality of pseudo labels is not uniform, and the low-quality pseudo labels affect the segmentation result.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image segmentation, and specifically relates to a medical image segmentation method, system, device and storage medium based on large model annotation and distributed learning. Background Art

[0002] Medical image segmentation, as a core step in medical image analysis, aims to accurately extract specific areas or structures in the image, thereby helping doctors to more accurately identify lesions, assess the condition and develop treatment plans. Therefore, medical image segmentation will directly affect the doctor's diagnosis, and its accuracy is particularly important.

[0003] In the existing technology, most medical images are annotated manually, which requires a lot of human resources. At the same time, due to manual annotation, the annotated data will have differences in annotation quality, image resolution and image type, which affects the accuracy of segmentation. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a medical image segmentation method, system, device and storage medium based on large model annotation and distributed learning, which can obtain high-quality pseudo labels with a small amount of annotated images and improve the accuracy of image segmentation.

[0005] The present invention provides the following technical solutions:

[0006] In a first aspect, a medical image segmentation method based on large model annotation and distributed learning is provided, comprising: obtaining a target image picture, and preprocessing the target image picture to obtain a target image picture set;

[0007] Inputting the target image picture set into a pre-trained Auto-MedSAM model to generate a pseudo label set of the target image;

[0008] Inputting the target image picture set and the pseudo label set into a pre-trained image segmentation model to obtain a segmented image of the target image;

[0009] The image segmentation model includes a quality assessment module, a feature extraction module, a random deviation module, an annotator deviation module and a fusion module;

[0010] The quality assessment module is used to calculate the similarity index of each pseudo label in the pseudo label set, so as to select pseudo labels whose similarity index is greater than a set value, and obtain a high-quality pseudo label set;

[0011] The feature extraction module is used to extract features in the target image through an encoder and a decoder to obtain an image feature map;

[0012] The random deviation module is used to separate the random deviation of each pseudo label in the high-quality pseudo label set based on the image feature map to obtain a first uncertainty picture;

[0013] The plurality of annotator bias modules are used to respectively separate the annotator biases in the first uncertainty image to obtain second uncertainty images with respective high-quality pseudo labels;

[0014] The fusion module is used to fuse the second uncertainty images of each high-quality pseudo-label to obtain a segmented image of the target image.

[0015] As a preferred technical solution of the present invention, the target image picture is preprocessed to obtain a target image picture set, including: rotating the target image picture by a random angle and scaling it by a random multiple to obtain a number of extended pictures; and using the target image picture and the several extended pictures as the target image picture set.

[0016] As a preferred technical solution of the present invention, the Auto-MedSAM model includes a coarse label module and a MedSAM module; the coarse label module is pre-trained through a labeled image set, and the coarse label module is used to generate corresponding coarse labels for each image in the target image set; the MedSAM module is used to generate pseudo labels for each image based on each coarse label.

[0017] As a preferred technical solution of the present invention, the quality assessment module is used to calculate the similarity index of each pseudo label in the pseudo label set to screen pseudo labels with a similarity index greater than a set value to obtain a high-quality pseudo label set, including:

[0018] Calculate the similarity index of each pseudo label, expressed as:

[0019] ;

[0020] in, represents the similarity index of the i-th pseudo label, represents the pseudo label corresponding to the i-th image, represents the i-th picture;

[0021] Will Pseudo labels with values ​​greater than the set value are regarded as high-quality pseudo labels.

[0022] As a preferred technical solution of the present invention, the random deviation module is used to separate the random deviation of each pseudo label in the high-quality pseudo label set based on the image feature map to obtain a first uncertainty picture, including:

[0023] Calculate the average μ and standard deviation characteristics of the pseudo labels in the high-quality pseudo label set ,

[0024] Based on the mean value μ, the mean entropy map is calculated, which is expressed as:

[0025] ;

[0026] in, The entropy graph representing the mean value μ;

[0027] Calibration via entropy plot of mean μ , and the calibrated standard deviation characteristics are obtained, expressed as:

[0028] ;

[0029] in, represents the calibrated standard deviation feature, α represents the parameter, S represents the total number of scales, represents the entropy graph at the s-th scale, represents weight;

[0030] The standard deviation characteristic after calibration Calculate the standard deviation ;

[0031] The first uncertainty picture satisfies N(μ, )distributed.

[0032] As a preferred technical solution of the present invention, the plurality of annotator deviation modules are used to respectively separate the annotator deviations in the first uncertainty image to obtain the second uncertainty image of each pseudo label, including:

[0033] Computing dynamic structural elements , and through the dynamic structural elements Simulate the expansion or erosion of the target image, expressed as:

[0034] ;

[0035] in, and β represent adjustable parameters, Indicates that the image feature map is at position Gradient information at ;

[0036] Separate the annotator bias in each high-quality pseudo-label first uncertainty image to obtain the separated image ;

[0037] Separation images with high-quality pseudo-labels Perform sigmoid operation on it to obtain The second uncertainty picture, where R represents the total number of high-quality pseudo labels, represents the r-th high-quality pseudo label.

[0038] As a preferred technical solution of the present invention, the fusion module is used to fuse the second uncertainty images of each high-quality pseudo-label to obtain a segmented image of the target image, including:

[0039] For any pixel (m, n) of the target image, any two high-quality pseudo-labeled second uncertainty images and The fusion process is expressed as:

[0040] = ;

[0041] in, express and The fusion result, K represents the conflict coefficient, K= ;

[0042] Will Fuse it with the next high-quality pseudo-labeled second uncertainty image until all high-quality pseudo-labeled second uncertainty images are fused to obtain the uncertainty of the pixel point. ;

[0043] The segmented image of the target image is expressed as:

[0044] ;

[0045] in, represents the segmented image of the target image, represents the rth high-quality pseudo label, and R represents the total number of high-quality pseudo labels.

[0046] In a second aspect, a medical image segmentation system based on large model annotation and distributed learning is provided, comprising: an image acquisition module, for acquiring a target image, and preprocessing the target image to obtain a target image set;

[0047] A pseudo label generation module is used to input the target image picture set into a pre-trained Auto-MedSAM model to generate a pseudo label set of the target image;

[0048] An output module, used for inputting the target image picture set and the pseudo label set into a pre-trained image segmentation model to obtain a segmented image of the target image;

[0049] The image segmentation model includes a quality assessment module, a feature extraction module, a random deviation module, an annotator deviation module and a fusion module;

[0050] The quality assessment module is used to calculate the similarity index of each pseudo label in the pseudo label set, so as to select pseudo labels whose similarity index is greater than a set value, and obtain a high-quality pseudo label set;

[0051] The feature extraction module is used to extract features in the target image through an encoder and a decoder to obtain an image feature map;

[0052] The random deviation module is used to separate the random deviation of each pseudo label in the high-quality pseudo label set based on the image feature map to obtain a first uncertainty picture;

[0053] The plurality of annotator bias modules are used to respectively separate the annotator biases in the first uncertainty image to obtain second uncertainty images with respective high-quality pseudo labels;

[0054] The fusion module is used to fuse the second uncertainty images of each pseudo-label to obtain a segmented image of the target image.

[0055] In a third aspect, a medical image segmentation device based on large model annotation and distributed learning is provided, comprising a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method described in the first aspect.

[0056] According to a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] The present invention provides a medical image segmentation method based on large model annotation and distributed learning. The Auto-MedSAM model is used to automatically generate pseudo labels by relying on a small number of annotated images, which greatly reduces labor costs and improves the quality of pseudo labels. The quality assessment module solves the problem of inconsistent pseudo label quality and low-quality pseudo labels affecting segmentation results. The annotator deviation module separates the different preferences of each annotator, and obtains the segmented image through the fusion module, thereby improving the accuracy of segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a schematic diagram of the structure of a medical image segmentation method according to an embodiment of the present invention;

[0060] Figure 2 is a flow chart of a medical image segmentation method according to an embodiment of the present invention;

[0061] Figure 3 is a schematic structural diagram of the Auto-MedSAM model in an embodiment of the present invention;

[0062] Figure 4 is a schematic diagram of the structure of a quality assessment module in an embodiment of the present invention;

[0063] Figure 5 is a schematic diagram of the structure of a feature extraction module in an embodiment of the present invention;

[0064] Figure 6 Schematic diagram of the structure of the random deviation module and the annotator deviation module in the embodiment of the present invention. DETAILED DESCRIPTION

[0065] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0066] Example 1

[0067] This embodiment provides a medical image segmentation method based on large model annotation and distributed learning. Figure 1 and Figure 2 As shown, the specific steps include:

[0068] Step 1: Obtain target image pictures, and preprocess the target image pictures to obtain a target image picture set.

[0069] In this embodiment, the target image picture is rotated by a random angle and scaled by a random multiple to obtain a number of extended pictures; the target image picture and the number of extended pictures are used as a target image picture set.

[0070] Step 2: Input the target image set into the pre-trained Auto-MedSAM model to generate a pseudo label set of the target image.

[0071] like Figure 3 As shown, the Auto-MedSAM model includes a coarse label module and a MedSAM module. The coarse label module is pre-trained with a labeled image set, and is used to generate a corresponding coarse label for each image in the target image set; the MedSAM module is used to generate a pseudo label for each image based on each coarse label. This step realizes the automatic generation of pseudo labels for unlabeled images, greatly reducing labor costs.

[0072] Step 3: Input the target image picture set and pseudo label set into a pre-trained image segmentation model to obtain a segmented image of the target image.

[0073] The image segmentation model includes a quality assessment module, a feature extraction module, a random deviation module, an annotator deviation module and a fusion module. Figure 1As shown, the qs module is the quality assessment module, the sbm module is the random deviation module, the nbm module is the annotator deviation module, and the EF module is the fusion module.

[0074] (1) The quality assessment module is used to calculate the similarity index of each pseudo-label in the pseudo-label set to screen out pseudo-labels whose similarity index is greater than a set value to obtain a high-quality pseudo-label set.

[0075] Specifically, after obtaining the pseudo-label set generated by Auto-MedSAM, the images in the pseudo-label set are subjected to corresponding inverse transformation to align them with the target image. After alignment, high-quality pseudo-labels are obtained through screening by the quality assessment module.

[0076] like Figure 4 As shown in the figure, a more accurate similarity index of pseudo labels is obtained by training the similarity index of real labels. The details are as follows:

[0077] Calculate the similarity index of each pseudo label, expressed as:

[0078] ;

[0079] in, represents the similarity index of the i-th pseudo label, represents the pseudo label corresponding to the i-th image, represents the i-th picture.

[0080] Calculate the similarity index of the true label, expressed as:

[0081] ;

[0082] in, represents the similarity index of the i-th true label, represents the pseudo label corresponding to the i-th image, Represents the true label corresponding to the i-th image.

[0083] In order to and Alignment introduces a loss function L, expressed as:

[0084] ;

[0085] in, represents the scaling constant, Represents the mean square error loss, used to measure and The average of the squared differences between the two is used to evaluate the quality of the pseudo-labels, expressed as:

[0086] ;

[0087] Where N represents the total number of pseudo labels, Denotes the RankNet loss, expressed as:

[0088] ;

[0089] Among them, P represents the probability that pseudo label i is better than pseudo label j, P= , Indicates the parameters, represents the actual order of pseudo-label i and pseudo-label j, When , the true label of pseudo label i is larger than pseudo label j, When , the true label of pseudo label i is smaller than pseudo label j.

[0090] Based on the loss function L, the similarity index of all pseudo labels can be obtained and Pseudo labels with a value greater than 0.9 are regarded as high-quality pseudo labels.

[0091] (2) The feature extraction module is used to extract features from the target image through an encoder and a decoder to obtain an image feature map.

[0092] like Figure 5 As shown, five encoders and five decoders are provided, and the corresponding encoders and decoders are connected.

[0093] (3) The random deviation module is used to separate the random deviation of each pseudo-label in the high-quality pseudo-label set based on the image feature map to obtain a first uncertainty image.

[0094] like Figure 6 As shown, the average value of the pseudo labels in the high-quality pseudo label set is calculated, which is expressed as:

[0095] μ= ;

[0096] Among them, μ represents the average value of pseudo labels, Represents the image feature map, represents a function about the mean value, express Parameters.

[0097] The feature of calculating the standard deviation of pseudo labels in the high-quality pseudo label set is expressed as:

[0098] ;

[0099] in, The characteristic of the standard deviation, Represents the image feature map, represents a function of the standard deviation, express Parameters.

[0100] Random deviations always occur in edge regions where the entropy of some annotations is high, so calibration is done based on entropy. The entropy graph of the mean value μ is represented as:

[0101] ;

[0102] in, An entropy plot representing the mean μ.

[0103] Calibration via entropy plot of mean μ , and the calibrated features are obtained, expressed as:

[0104] ;

[0105] in, represents the calibrated features, α represents the parameter, S represents the total number of scales, represents the entropy graph at the s-th scale, Represents weight.

[0106] After calibration, the features Calculate the standard deviation, expressed as:

[0107] ;

[0108] in, represents the standard deviation, represents a 1×1 convolutional layer, Indicates connection, express Parameters.

[0109] The first uncertainty picture satisfies N(μ, )distributed.

[0110] (4) The plurality of annotator bias modules are used to respectively separate the annotator biases in the first uncertainty image to obtain second uncertainty images with high-quality pseudo labels.

[0111] Since the target image is rotated and scaled, some defined areas will be larger or smaller than other areas, so they are simulated by regional expansion or erosion. Figure 6 As shown, the dynamic structural element is calculated , and through the dynamic structural elements Simulate the expansion or erosion of the target image, expressed as:

[0112] ;

[0113] in, and β represent adjustable parameters, G(x,y) represents the image feature map at position The gradient information at .

[0114] Separate the annotator bias in the first uncertainty image, expressed as:

[0115] ;

[0116] in, represents the separated image of the rth pseudo label, represents the convolution operation, express Parameters, represents a function, express function.

[0117] Through Gaussian sampling, the separation of high-quality pseudo-label images Perform sigmoid operation on it to obtain The second uncertainty picture, where R represents the total number of high-quality pseudo labels, represents the r-th high-quality pseudo label.

[0118] (5) The fusion module is used to fuse the second uncertainty images of each high-quality pseudo-label to obtain a segmented image of the target image.

[0119] For any pixel (m, n) of the target image, any two high-quality pseudo-labeled second uncertainty images and The fusion process is expressed as:

[0120] = ;

[0121] in, express and The fusion result, K represents the conflict coefficient, K= .

[0122] Will Fuse it with the next high-quality pseudo-labeled second uncertainty image until all high-quality pseudo-labeled second uncertainty images are fused to obtain the uncertainty of the pixel point. .

[0123] The segmented image of the target image is expressed as:

[0124] ;

[0125] in, represents the segmented image of the target image, represents the rth high-quality pseudo label, and R represents the total number of high-quality pseudo labels.

[0126] like Figure 1 As shown, the rightmost picture is the segmented image.

[0127] Example 2

[0128] This embodiment provides a medical image segmentation system based on large model annotation and distributed learning, including:

[0129] The image acquisition module is used to acquire target image pictures and pre-process the target image pictures to obtain a target image picture set.

[0130] The pseudo label generation module is used to input the target image picture set into the pre-trained Auto-MedSAM model to generate a pseudo label set of the target image.

[0131] The output module is used to input the target image picture set and the pseudo label set into a pre-trained image segmentation model to obtain a segmented image of the target image.

[0132] The image segmentation model includes a quality assessment module, a feature extraction module, a random deviation module, an annotator deviation module and a fusion module.

[0133] The quality assessment module is used to calculate the similarity index of each pseudo-label in the pseudo-label set to screen pseudo-labels with a similarity index greater than a set value to obtain a high-quality pseudo-label set.

[0134] The feature extraction module is used to extract features in the target image through an encoder and a decoder to obtain an image feature map.

[0135] The random deviation module is used to separate the random deviation of each pseudo label in the high-quality pseudo label set based on the image feature map to obtain a first uncertainty picture.

[0136] The multiple annotator deviation modules are used to separate the annotator deviations in the first uncertainty image respectively to obtain the second uncertainty images with high-quality pseudo labels.

[0137] The fusion module is used to fuse the second uncertainty images of each pseudo-label to obtain a segmented image of the target image.

[0138] Example 3

[0139] This embodiment provides a medical image segmentation device based on large model annotation and distributed learning, including a processor and a storage medium. The storage medium is used to store instructions. The processor is used to operate according to the instructions to execute the steps of the medical image segmentation method based on large model annotation and distributed learning described in Example 1.

[0140] Example 4

[0141] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the medical image segmentation method based on large model annotation and distributed learning described in Embodiment 1 are implemented.

[0142] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0143] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0144] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0146] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A medical image segmentation method based on large model annotation and distributed learning, characterized in that: include: Obtaining target image pictures, and preprocessing the target image pictures to obtain a target image picture set; Inputting the target image picture set into a pre-trained Auto-MedSAM model to generate a pseudo label set of the target image; Inputting the target image picture set and the pseudo label set into a pre-trained image segmentation model to obtain a segmented image of the target image; The image segmentation model includes a quality assessment module, a feature extraction module, a random deviation module, an annotator deviation module and a fusion module; The quality assessment module is used to calculate the similarity index of each pseudo label in the pseudo label set, so as to select pseudo labels whose similarity index is greater than a set value, and obtain a high-quality pseudo label set; The feature extraction module is used to extract features in the target image through an encoder and a decoder to obtain an image feature map; The random deviation module is used to separate the random deviation of each pseudo label in the high-quality pseudo label set based on the image feature map to obtain a first uncertainty picture; The plurality of annotator bias modules are used to respectively separate the annotator biases in the first uncertainty image to obtain second uncertainty images with respective high-quality pseudo labels; The fusion module is used to fuse the second uncertainty images of each high-quality pseudo-label to obtain a segmented image of the target image.

2. The medical image segmentation method based on large model annotation and distributed learning according to claim 1, characterized in that: The preprocessing of the target image pictures to obtain a target image picture set includes: Rotate the target image at a random angle and scale it by a random multiple to obtain a number of extended images; The target image picture and several extended pictures are used as a target image picture set.

3. The medical image segmentation method based on large model annotation and distributed learning according to claim 1, characterized in that: The Auto-MedSAM model includes a rough label module and a MedSAM module; The rough label module is pre-trained with a labeled picture set, and the rough label module is used to generate a corresponding rough label for each picture in the target image picture set; The MedSAM module is used to generate pseudo labels for each image based on each rough label.

4. The medical image segmentation method based on large model annotation and distributed learning according to claim 1, characterized in that: The quality assessment module is used to calculate the similarity index of each pseudo label in the pseudo label set to screen pseudo labels with a similarity index greater than a set value to obtain a high-quality pseudo label set, including: Calculate the similarity index of each pseudo label, expressed as: ； in, represents the similarity index of the i-th pseudo label, represents the pseudo label corresponding to the i-th image, represents the i-th picture; Will Pseudo labels with values ​​greater than the set value are regarded as high-quality pseudo labels.

5. The medical image segmentation method based on large model annotation and distributed learning according to claim 1, characterized in that: The random deviation module is used to separate the random deviation of each pseudo label in the high-quality pseudo label set based on the image feature map to obtain a first uncertainty picture, including: Calculate the average μ and standard deviation characteristics of the pseudo labels in the high-quality pseudo label set , Based on the mean value μ, the mean entropy map is calculated, which is expressed as: ； in, The entropy graph representing the mean value μ; Calibrate the standard deviation feature through the entropy map of the mean μ , and the calibrated standard deviation characteristics are obtained, expressed as: ； in, represents the calibrated standard deviation feature, α represents the parameter, S represents the total number of scales, represents the entropy graph at the s-th scale, represents weight; The standard deviation characteristic after calibration Calculate the standard deviation ; The first uncertainty picture satisfies N(μ, )distributed.

6. The medical image segmentation method based on large model annotation and distributed learning according to claim 5, characterized in that: The plurality of annotator deviation modules are used to respectively separate the annotator deviations in the first uncertainty image to obtain second uncertainty images of the respective pseudo labels, including: Computing dynamic structural elements , and through the dynamic structural elements Simulate the expansion or erosion of the target image, expressed as: ； in, and β represent adjustable parameters, Indicates that the image feature map is at position Gradient information at ; Separate the annotator bias in each high-quality pseudo-label first uncertainty image to obtain the separated image ; Separation images with high-quality pseudo-labels Perform sigmoid operation on it to obtain The second uncertainty picture, where R represents the total number of high-quality pseudo labels, represents the r-th high-quality pseudo label.

7. The medical image segmentation method based on large model annotation and distributed learning according to claim 1, characterized in that: The fusion module is used to fuse the second uncertainty images of the high-quality pseudo labels to obtain a segmented image of the target image, including: For any pixel (m, n) of the target image, any two high-quality pseudo-labeled second uncertainty images and The fusion process is expressed as: = ； in, express and The fusion result, K represents the conflict coefficient, K= ; Will Fuse it with the next high-quality pseudo-labeled second uncertainty image until all high-quality pseudo-labeled second uncertainty images are fused to obtain the uncertainty of the pixel point. ; The segmented image of the target image is expressed as: ； in, represents the segmented image of the target image, represents the rth high-quality pseudo label, and R represents the total number of high-quality pseudo labels.

8. A medical image segmentation system based on large model annotation and distributed learning, characterized in that: include: The image acquisition module is used to acquire target image pictures and pre-process the target image pictures to obtain a target image picture set; A pseudo label generation module is used to input the target image picture set into a pre-trained Auto-MedSAM model to generate a pseudo label set of the target image; An output module, used for inputting the target image picture set and the pseudo label set into a pre-trained image segmentation model to obtain a segmented image of the target image; The image segmentation model includes a quality assessment module, a feature extraction module, a random deviation module, an annotator deviation module and a fusion module; The quality assessment module is used to calculate the similarity index of each pseudo label in the pseudo label set, so as to select pseudo labels whose similarity index is greater than a set value, and obtain a high-quality pseudo label set; The feature extraction module is used to extract features in the target image through an encoder and a decoder to obtain an image feature map; The random deviation module is used to separate the random deviation of each pseudo label in the high-quality pseudo label set based on the image feature map to obtain a first uncertainty picture; The plurality of annotator bias modules are used to respectively separate the annotator biases in the first uncertainty image to obtain second uncertainty images with respective high-quality pseudo labels; The fusion module is used to fuse the second uncertainty images of each pseudo-label to obtain a segmented image of the target image.

9. A medical image segmentation device based on large model annotation and distributed learning, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the medical image segmentation method based on large model annotation and distributed learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the medical image segmentation method based on large model annotation and distributed learning as described in any one of claims 1 to 7 are implemented.

Images