A medical image segmentation method based on contrast manifold regularization and related device

Abstract

The application discloses a medical image segmentation method based on contrast manifold regularization and related devices. The method comprises: after importing medical image data and completing preprocessing, calculating the similarity between samples to construct a manifold regularization term; performing feature decomposition and enhancement on the data to obtain anatomical structure features, pathological semantic features and their enhanced versions; for each sample, the original-enhanced feature pair of the same sample is taken as a positive sample, different sample feature cross combinations are taken as negative samples, and a contrast loss term is calculated; the manifold regularization term and the contrast loss term are dynamically weighted and summed to form a dynamic contrast manifold regularization term; the image segmentation model is trained and the weight is obtained, and then precise segmentation is realized. The application significantly improves the image segmentation accuracy, robustness and generalization ability in the small sample scene by fusing the manifold structure and cross-sample contrast constraint, and reduces the dependence on a large amount of labeled data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical image processing, in particular to a medical image segmentation method based on contrast manifold regularization and related devices. BACKGROUND

[0002] In the field of medical image segmentation, accurate target region segmentation is crucial for disease diagnosis, treatment planning, and surgical navigation. However, existing segmentation techniques face many challenges. On the one hand, the medical image labeling process is time-consuming and labor-intensive, requiring the participation of professional doctors, resulting in a lack of labeled data. On the other hand, medical images have complex anatomical structures and variable pathological characteristics, which pose higher requirements for the generalization ability of segmentation models. Traditional supervised learning methods rely on a large amount of labeled data, making it difficult to meet actual needs. Semi-supervised learning methods can utilize unlabeled data, but their application in the medical image field still has limitations. In addition, existing segmentation models often ignore the manifold structure of medical images, failing to fully utilize the intrinsic geometric characteristics of the data, affecting the segmentation accuracy and generalization performance of the model. SUMMARY

[0003] To overcome the shortcomings of the prior art, the present application proposes a medical image segmentation method based on contrast manifold regularization and related devices, including but not limited to a medical image segmentation device based on contrast manifold regularization, an electronic device, a computer readable storage medium, and a computer program product.

[0004] The first aspect provides a medical image segmentation method based on contrast manifold regularization, comprising:

[0005] A. Importing medical image data and performing preprocessing to obtain preprocessed data;

[0006] B. Calculating the similarity between samples in the preprocessed data to obtain a manifold regularization term;

[0007] C. Feature decomposition and feature enhancement of the preprocessed data to obtain anatomical structure features, pathological semantic features, enhanced anatomical structure features, and enhanced pathological semantic features;

[0008] D. Constructing positive and negative sample pairs for each sample, taking the anatomical structure features and enhanced anatomical structure features, and the pathological semantic features and enhanced pathological semantic features of the same sample in the preprocessed data as positive sample pairs, and taking the cross combination of anatomical structure features and pathological semantic features of different samples as negative sample pairs; and calculating a contrast loss term based on the positive and negative sample pairs; and dynamically weighting and summing the manifold regularization term and the contrast loss term to obtain a dynamic contrast manifold regularization term;

[0009] E. training an image segmentation model using the preprocessed data based on the dynamic contrast manifold regularization term, to obtain training weights;

[0010] F. performing image segmentation using the image segmentation model based on the training weights, to obtain a segmentation result.

[0011] In combination with any embodiment of the present application, the similarity between samples in the preprocessed data is calculated, including:

[0012] The similarity matrix between the samples is calculated using a heat kernel function, wherein a kernel width parameter of the heat kernel function is automatically determined by a spectral analysis method;

[0013] A Laplacian matrix is constructed based on the similarity matrix, and a low-dimensional embedding representation of a manifold structure is obtained by eigen decomposition.

[0014] In combination with any embodiment of the present application, the positive and negative sample pairs are constructed, including:

[0015] For each sample, the distance between the sample and other samples in the feature space is calculated;

[0016] Based on the distance, M samples closest to the sample are selected to form a positive sample pair, and N samples farthest from the sample are selected to form a negative sample pair; wherein M and N are preset positive integers.

[0017] In combination with any embodiment of the present application, the similarity between samples in the preprocessed data is calculated, including:

[0018] The distance between the samples in the feature space is calculated;

[0019] Based on the distance, a similarity matrix is calculated using a heat kernel function;

[0020] A Laplacian matrix is constructed based on the similarity matrix;

[0021] Eigen decomposition is performed on the Laplacian matrix to obtain a low-dimensional embedding representation of a manifold structure, to obtain the manifold regularization term.

[0022] In combination with any embodiment of the present application, the feature decomposition and feature enhancement of the preprocessed data are performed, including:

[0023] Feature decoupling is performed on each sample to obtain anatomical structure features and pathological semantic features ;

[0024] Affine transformation is applied to the anatomical structure features to obtain enhanced features ;

[0025] Noise is injected into the pathological semantic features to obtain enhanced features The noise includes Gaussian noise or random occlusion with an occlusion ratio of 10%.

[0026] In combination with any of the embodiments of the present application, the positive and negative sample pairs are constructed for each sample, including:

[0027] The positive sample pair is constructed: the anatomical structure features of the same sample and the enhanced anatomical structure features form an anatomical structure positive sample pair, and the pathological semantic features of the same sample and the enhanced pathological semantic features form a pathological semantic positive sample pair.

[0028] The negative sample pair is constructed: the anatomical structure features and the pathological semantic features of different samples are cross combined into a negative sample pair.

[0029] The contrast loss function of the positive sample pair and the negative sample pair is calculated to obtain a contrast loss term.

[0030] In combination with any of the embodiments of the present application, the manifold regularization term and the contrast loss term are dynamically weighted and summed to obtain a dynamic contrast manifold regularization term, including:

[0031] A dynamic weight is established, and the formula is:

[0032] ,

[0033] Wherein w0 is an initial weight, k is a decay coefficient, and t is a training iteration round; Based on the dynamic weight, the manifold regularization term

[0034] and the contrast loss term are weighted and summed to obtain a dynamic contrast manifold regularization term, and the formula is:

[0035] .

[0036] In combination with any of the embodiments of the present application, based on the dynamic contrast manifold regularization term, the image segmentation model is trained using the preprocessed data, including:

[0037] The training parameters and the training stop condition are set, and the network weight of the image segmentation model is initialized;

[0038] In the training process, the loss value of the dynamic contrast manifold regularization term is calculated forwardly, and the network weight is updated;

[0039] When the training stop condition is met, the iteration is stopped, and the training weight is obtained.

[0040] In a second aspect, a medical image segmentation device based on contrast manifold regularization is provided, including:

[0041] ​The data input unit is configured to import medical image data and pre-process the medical image data to obtain pre-processed data.

[0042] The contrast learning unit is configured to perform feature decomposition and feature enhancement on the pre-processed data to obtain anatomical structure features, pathological semantic features, enhanced anatomical structure features, and enhanced pathological semantic features. The contrast learning unit is also configured to construct a positive sample and a negative sample for each sample, and calculate a contrast loss term based on the positive sample and the negative sample.

[0043] The manifold regularization unit is configured to calculate a similarity between samples in the pre-processed data to obtain a manifold regularization term.

[0044] The model training unit is configured to dynamically weight and sum the manifold regularization term and the contrast loss term to obtain a dynamic contrast manifold regularization term. The model training unit is also configured to train an image segmentation model using the pre-processed data based on the dynamic contrast manifold regularization term to obtain a training weight.

[0045] The image segmentation unit is configured to perform image segmentation using the image segmentation model based on the training weight to obtain a segmentation result.

[0046] In a third aspect, an electronic device is provided, comprising a processor, a communication module, a sensor, a user interface, and a storage unit, wherein the storage unit is configured to store computer program code, the program code comprising computer instructions. When the processor executes the instructions, the electronic device will perform the method described in the second aspect above and any of its implementation forms.

[0047] In a fourth aspect, another electronic device is provided, comprising a processor, a wireless communication module, a touch screen, a speaker, and a storage unit, wherein the storage unit is configured to store computer program code, the program code comprising computer instructions. When the processor executes the instructions, the electronic device will perform the method described in the second aspect above and any of its implementation forms.

[0048] In a fifth aspect, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, the program comprising program instructions. When the instructions are executed by a processor, the processor will perform the method described in the second aspect above and any of its implementation forms.

[0049] In a sixth aspect, a computer program product is provided, the computer program product comprising a computer program or instructions. When the computer program or instructions are run on a computer, the computer will perform the method described in the second aspect above and any of its implementation forms.

[0050] It should be understood that the above general description and the following detailed description are only used as examples and explanations, and do not limit the present application in any way.

[0051] In the present application, compared with the prior art, the application first introduces a "contrast-manifold" double-constraint dynamic weighting framework in the medical image segmentation task, realizing high-precision segmentation in the small sample scene. Traditional methods either rely on a large amount of labeled data or only use a single regularization term, and it is difficult to balance structure preservation and semantic distinction. The present application improves through the following innovative process: 1. Calculate the similarity between samples based on the manifold hypothesis, forming a manifold regularization term for structure preservation; 2. Enhance the anatomical and pathological features of the same image respectively, and construct a contrast loss by taking the "original-enhanced" pair as positive samples and the cross-sample cross-pair as negative samples, and strengthen semantic discrimination; 3. The manifold regularization term and the contrast loss term are adaptively weighted according to the training stage to form a dynamic contrast manifold regularization term; 4. Train the segmentation model end-to-end with the regularization term as the optimization goal, which significantly improves the boundary accuracy and abnormal area detection rate, while reducing the dependence on the amount of labeled data. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background art, the drawings needed to be used in the embodiments of the present application or the background art will be described below.

[0053] The drawings herein are incorporated into the specification and form part of the specification, which show embodiments consistent with the present application, and together with the specification, serve to illustrate the technical solutions of the present application.

[0054] Figure 1 A flowchart of a medical image segmentation method based on contrast manifold regularization provided by an embodiment of the present application;

[0055] Figure 2 A structural schematic diagram of a medical image segmentation device based on contrast manifold regularization provided by an embodiment of the present application;

[0056] Figure 3 A hardware architecture schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to let the professionals in the technical field more comprehensively understand the technical solutions of the present application, the technical solutions of the present application will be explained in detail and clearly by the drawings. It should be particularly pointed out that the described embodiments are only part of the examples of the present application, and do not represent all. Based on these embodiments, those skilled in the art can directly deduce all other possible embodiments without creative thinking, and these are also included in the protection scope of the present application.

[0058] In the specification, claims and relevant drawings of the present application, the terms "first", "second" and the like are used only to distinguish different elements, and do not imply any specific order. Meanwhile, the use of "include" and "have" and their variants means non-exclusive inclusion. This means that if a process, method, system, product or device includes a series of steps or components, it does not limit the process, method, system, product or device to the listed steps or components, but can also include other steps or components not listed, or other steps or units inherent to it.

[0059] The "embodiments" mentioned in this paper refer to any instance combined with specific features, structures or properties, which may belong to at least one embodiment of the present application. The "embodiments" mentioned in this paper do not necessarily refer to the same specific case, nor do they represent mutually independent or exclusive alternatives. Those skilled in the art should understand that the embodiments described herein can be used with other embodiments. It should be clear that in this application, "at least one" includes one or more instances, "multiple" means two or more instances, and "at least two" means two or more instances.

[0060] It should be understood that the method embodiments of the present application can also be implemented by a processor executing computer program code. The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.

[0061] Please refer to Figure 1 , Figure 1 A flowchart of a medical image segmentation method based on contrast manifold regularization is provided for the embodiments of the present application. The flowchart shows the complete process from image acquisition to segmentation result generation in detail.

[0062] 101, data import and preprocessing: import medical image dataset and perform preprocessing to obtain preprocessed data.

[0063] In this embodiment, the medical image dataset usually contains target regions (such as liver, brain, etc.) and possible non-target regions (such as background or other tissues).

[0064] In this embodiment, the modalities of the samples in the medical image dataset include but are not limited to CT, MRI, ultrasound, etc.

[0065] In this embodiment, the medical image data acquisition methods include but are not limited to direct import from medical imaging equipment, import from hospital database, etc.

[0066] In this embodiment, the preprocessing includes but is not limited to size unification, cropping, etc.

[0067] 102、obtaining a manifold regularization term: calculating the similarity between samples in the preprocessed data to obtain a manifold regularization term.

[0068] In this embodiment, the manifold regularization term is constructed using a reproducing kernel Hilbert space method.

[0069] In this embodiment, a heat kernel function is used to calculate the similarity matrix between samples, wherein the kernel width parameter of the heat kernel function is automatically determined by a spectral analysis method; a Laplacian matrix is constructed based on the similarity matrix, and a low-dimensional embedding representation of the manifold structure is obtained by eigen decomposition.

[0070] In another possible implementation, the manifold regularization term can be constructed using other manifold regularization methods.

[0071] 103、obtaining a contrastive loss term: constructing a positive and negative sample pair for each sample, and calculating a contrastive loss term based on the positive and negative sample pair.

[0072] In this embodiment, the preprocessed data is first subjected to eigen decomposition and feature enhancement to obtain anatomical structure features , pathological semantic features , enhanced anatomical structure features , and enhanced pathological semantic features ; then for each sample, the anatomical structure features and enhanced anatomical structure features of the sample itself, and the pathological semantic features and enhanced pathological semantic features are taken as the positive sample pair, and the cross combination of the anatomical structure features and pathological semantic features of the sample and other samples are taken as the negative sample pair; finally, the contrastive loss term is calculated based on the positive and negative sample pair.

[0073] In this embodiment, the calculation method of the contrastive loss term includes but is not limited to InfoNCE loss, Margin loss, and triplet loss.

[0074] In another possible implementation, the positive and negative sample pair can be constructed by clustering method, specifically, clustering analysis is performed on the preprocessed data to obtain a plurality of sample clusters, and samples in the same cluster are taken as the positive sample pair, and samples in different clusters are taken as the negative sample pair.

[0075] In another possible implementation, the construction of the positive and negative sample pairs can adopt a hard example mining strategy, specifically focusing on selecting those sample pairs that are difficult to distinguish; in the negative sample pairs, selecting samples with higher similarity, i.e. those negative samples that are easy to be misjudged by the model as positive samples; in the positive sample pairs, selecting samples with lower similarity, i.e. those positive samples that are difficult for the model to identify as the same class.

[0076] In another possible implementation, for a semi-supervised learning task, the construction of the positive and negative sample pairs can adopt a random cropping manner, specifically, performing a random cropping operation on the preprocessed data, taking two different random crops of the same medical image as a positive sample pair, and taking different medical images as negative sample pairs.

[0077] 104. Constructing an optimization objective function: dynamically weighting and summing the manifold regularization term and the contrastive loss term to obtain a dynamic contrastive manifold regularization term as an optimization objective function.

[0078] In this embodiment, the calculation formula of the dynamic contrastive manifold regularization term is:

[0079]

[0080] wherein, is the manifold regularization term, is the contrastive loss term, is a dynamic weight, and the formula of the dynamic weight is:

[0081]

[0082] wherein is an initial weight, k is a decay coefficient, and t is a training iteration round.

[0083] In another possible implementation, the dynamic weight can also be determined by cross-validation, specifically: dividing the preprocessed data into a training set and a validation set, training the image segmentation model on the training set, evaluating the performance of the image segmentation model under different weights on the validation set, and selecting the weight with the best performance.

[0084] In another possible implementation, the dynamic weight can adopt other dynamic weighting strategies. For example, at the beginning of training, a larger can be used to enable the image segmentation model to quickly learn the distinguishing features between classes; as the training progresses, the is gradually reduced to guide the image segmentation model to learn more generalizable feature representations.

[0085] ​​In another possible implementation, the optimization objective function can be further constructed by weighting and summing the dynamic contrast manifold regularization term and other deep learning loss terms.

[0086] 105. Training an image segmentation model: training an image segmentation model based on the dynamic contrast manifold regularization term using the preprocessed data to obtain trained weights.

[0087] In this embodiment, the image segmentation model is constructed based on deep learning principles.

[0088] In this embodiment, the loss value of the student model is calculated using the optimization objective function, and the parameters of the student model are updated according to the loss value.

[0089] 106. Generating a segmentation result: performing image segmentation using the image segmentation model based on the trained weights to obtain a segmentation result.

[0090] In another possible implementation, the segmentation result is post-processed, including but not limited to connected region elimination, hole filling, etc.

[0091] Those skilled in the art can understand that in the above method of the specific embodiment, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0092] The above describes the method of the embodiments of the present application in detail. The device of the embodiments of the present application is provided below.

[0093] Please refer to Figure 2 , Figure 2 A structure diagram of a medical image segmentation device based on contrast manifold regularization provided by the embodiments of the present application is provided. The image segmentation device 1 includes a data input unit 11, a contrast learning unit 12, a manifold regularization unit 13, a model training unit 14, and an image segmentation unit 15, specifically:

[0094] The data input unit 11 is used to import medical image data and perform preprocessing to obtain preprocessed data.

[0095] The contrast learning unit 12 is used to perform feature decomposition and feature enhancement on the preprocessed data to obtain anatomical structure features, pathological semantic features, enhanced anatomical structure features, and enhanced pathological semantic features. It is also used to construct positive and negative sample pairs for each sample and calculate a contrast loss term based on the positive and negative sample pairs.

[0096] The manifold regularization unit 13 is used to calculate the similarity between samples in the preprocessed data to obtain a manifold regularization term.

[0097] The model training unit 14 is configured to sum the manifold regularization term and the contrast loss term dynamically to obtain a dynamic contrast manifold regularization term, and train an image segmentation model based on the dynamic contrast manifold regularization term using the preprocessed data to obtain a training weight.

[0098] The image segmentation unit 15 is configured to perform image segmentation based on the training weight using the image segmentation model to obtain a segmentation result.

[0099] In some embodiments, the apparatus provided by the embodiments of the present application has functions or includes modules that can be used to perform the methods described in the above method embodiments, and the specific implementation can refer to the description of the above method embodiments. For brevity, it will not be repeated here.

[0100] Please refer to Figure 3 , Figure 3 A hardware architecture diagram of an electronic device is shown. The electronic device 2 mainly includes a processor 21 and a memory 22. In addition, the device can also include an input device 23 and an output device 24. The processor 21, the memory 22, the input device 23 and the output device 24 are connected to each other through connection components. These connection components can be various interfaces, data lines or communication buses, etc. The embodiments of the present application do not make specific provisions for this.

[0101] The processor 21 can be one or more graphics processing units (GPUs). If the processor 21 is a GPU, the GPU can be single-core or multi-core. As an option, the processor 21 can also be a processor group composed of multiple GPUs connected to each other through one or more buses. In addition, the processor can also be other types of processors, and the embodiments of the present application do not make specific limitations.

[0102] The memory 22 is designed to save the instructions of the computer program and various program codes required for the implementation of the scheme of the present application. As an option, the memory can include but is not limited to random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or compact disc read-only memory (CD-ROM), which are used to store related instructions and data.

[0103] The input device 23 is used to input data and / or signals, and the output device 24 is used to output data and / or signals. The input device 23 and the output device 24 can be independent devices or a whole device.

[0104] It should be recognized that in the embodiments of the present application, the memory 22 can not only save related instructions, but also save related data. The embodiments of the present application do not make specific provisions for the specific data content stored in the memory.

[0105] It should be noted that, Figure 3 Only a simplified design of an electronic device is shown. In actual use, the electronic device can also include other necessary components, such as different numbers of input / output devices, processors, memories, etc. All electronic devices capable of implementing the embodiments of the present application are within the protection scope of the present application.

[0106] Those skilled in the art should recognize that, according to the components and algorithm steps of various examples described in the embodiments disclosed herein, these functions can be realized by electronic hardware or in combination with computer software and electronic hardware. Whether these functions are performed by hardware or software depends on the specific application requirements and design constraints of the technical solutions. The skilled person can adopt different implementation methods according to the requirements of each specific application, but such implementation methods should not be regarded as beyond the protection scope of the present application.

[0107] Professionals should understand that, in order to facilitate description and simplify the description, the specific operation process of the above system, device and component can refer to the corresponding steps in the previous method embodiments, which will not be repeated here. At the same time, professionals should also understand that each embodiment in the present application has its own emphasis, and in order to facilitate description and simplify, the same or similar content may not be repeated in different embodiments. Therefore, if the part is not mentioned or not described in detail in an embodiment, it can be referred to the relevant description of other embodiments.

[0108] In several embodiments provided in the present application, it should be recognized that the disclosed system, device and method can also be realized by other ways. For example, the described device embodiments are only exemplary, and the division of the units is only logical functional division, and different division methods can exist in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be omitted, or some steps can not be performed. In addition, the connection between the units shown or discussed, whether direct or indirect, whether coupled or communicatively connected, can be realized through interfaces, devices or units in electrical, mechanical or other forms.

[0109] The units described as independent components can actually be physically separated or not; the parts presented as units can be physical entities or not, that is, they can be concentrated in one location or dispersed on multiple network nodes. According to the actual needs, some or all of these units can be selected to achieve the goal of the present embodiment.

[0110] Moreover, in each embodiment of the present application, each functional unit can be integrated into one processing unit, or each independent physical unit, or two or more units can be combined into one unit. In the foregoing embodiments, the related functions can be fully or partially implemented by software, hardware, firmware or any combination thereof. If software implementation is chosen, it can be implemented in whole or in part in the form of a computer program product. The computer program product contains one or more computer instructions. When the instructions are loaded and executed on a computer, they produce all or part of the processes or functions described in the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, DSL) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any computer-accessible available medium, or a data storage facility such as a server, data center, etc. integrated with one or more available media. These available media can include magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), semiconductor media (such as SSDs), etc. Those skilled in the art should understand that all or part of the processes of the above-mentioned embodiments can be completed by computer program instruction related hardware, and these programs can be stored in a computer-readable storage medium. When these programs are executed, they will contain the processes of the above-mentioned embodiments. The above-mentioned storage medium includes but is not limited to read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and various media that can store program codes.

Claims

1. A medical image segmentation method based on contrast manifold regularization, characterized in that, The method comprises the following steps: A. Importing medical image data and preprocessing to obtain preprocessed data; B. Calculating the distance of samples in the preprocessed data in the feature space, calculating the Laplacian matrix based on the distance, performing eigen decomposition on the Laplacian matrix, obtaining the low-dimensional embedding representation of the manifold structure, and obtaining the manifold regularization term; C. Eigen decomposition and feature enhancement of the preprocessed data to obtain anatomical structure features, pathological semantic features, enhanced anatomical structure features, and enhanced pathological semantic features; D. Constructing positive and negative sample pairs for each sample, using the anatomical structure features and enhanced anatomical structure features of the same sample, and the pathological semantic features and enhanced pathological semantic features as the positive sample pair, and using the cross combination of anatomical structure features and pathological semantic features of different samples as the negative sample pair; and calculating the contrast loss term based on the positive and negative sample pairs; E. Establishing dynamic weights The formula is: ， wherein is the initial weight, k is the decay coefficient, and t is the training iteration round. and based on the dynamic weight, the manifold regularization term is weighted summed with the contrast loss term to obtain a dynamic contrast manifold regularization term, and the formula is: ； F. Training an image segmentation model using the preprocessed data based on the dynamic contrast manifold regularization term to obtain training weights; G. Using the image segmentation model to perform image segmentation based on the training weights to obtain a segmentation result.

2. The method of claim 1, wherein, The Laplacian matrix is calculated based on the distance using a heat kernel function.

3. The method of claim 1, wherein, The eigen decomposition and feature enhancement of the preprocessed data comprises: Decoupling features from each sample yields anatomical features with pathological semantic features ; applying an affine transformation to the anatomical structure feature, resulting in an enhanced anatomical structure feature ; injecting noise to the pathological semantic feature to obtain an enhanced pathological semantic feature , the noise including Gaussian noise or random occlusion with an occlusion ratio of 10%.

4. The method of claim 1, wherein, The positive and negative sample pairs for each sample are constructed as follows: Positive sample pair construction: the anatomical structure features and enhanced anatomical structure features of the same sample form an anatomical structure positive sample pair, and the pathological semantic features and enhanced pathological semantic features of the same sample form a pathological semantic positive sample pair; Negative sample pair construction: the cross combination of anatomical structure features and pathological semantic features of different samples forms a negative sample pair; The contrast loss function of the positive and negative sample pairs is calculated to obtain the contrast loss term.

5. The method of claim 1, wherein, The training of the image segmentation model using the preprocessed data based on the dynamic contrast manifold regularization term comprises: Setting training parameters and training stop conditions, initializing the network weights of the image segmentation model; Calculating the loss value of the dynamic contrast manifold regularization term in the forward direction during training, and updating the network weights; When the training stop condition is met, the iteration is stopped, and the training weights are obtained.

6. A medical image segmentation apparatus based on contrast manifold regularization, characterized by, The method comprises the following steps: A data input unit is configured to import medical image data and perform preprocessing to obtain preprocessed data; A contrast learning unit is configured to perform eigen decomposition and feature enhancement on the preprocessed data to obtain anatomical structure features, pathological semantic features, enhanced anatomical structure features, and enhanced pathological semantic features, and to construct positive and negative sample pairs for each sample and calculate a contrast loss term based on the positive and negative sample pairs; A manifold regularization unit is configured to calculate the distance of samples in the preprocessed data in the feature space, calculate the Laplacian matrix based on the distance, perform eigen decomposition on the Laplacian matrix, obtain the low-dimensional embedding representation of the manifold structure, and obtain the manifold regularization term; Model training unit: for establishing dynamic weights The formula is: ， wherein is the initial weight, k is the decay coefficient, and t is the training iteration round. and based on the dynamic weight, the manifold regularization term is weighted summed with the contrast loss term to obtain a dynamic contrast manifold regularization term, and the formula is: ; further based on the dynamic contrast manifold regularization term, training an image segmentation model using the preprocessed data to obtain trained weights; An image segmentation unit is configured to use the image segmentation model to perform image segmentation based on the training weights to obtain a segmentation result.

7. An electronic device, comprising: An electronic device comprising a processor and a storage unit for storing computer program code, the code comprising computer instructions which, when executed by the processor, cause the electronic device to perform the method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, A computer readable storage medium having stored therein a computer program, the program comprising program instructions which, when executed by a processor, cause the processor to perform the method of any one of claims 1 to 5.

9. A computer program product, characterised in that, A computer program product comprising a computer program or instructions which, when run on a computer, cause the computer to perform the method of any one of claims 1 to 5.

Images