Tumor pathology image segmentation method and system based on graph convolutional contrastive learning network

Abstract

The present invention discloses a tumor pathology image segmentation method and system based on a graph convolutional contrastive learning network, which relates to the field of image segmentation technology. The method includes: obtaining a tumor pathology image and a segmentation label corresponding to the tumor pathology image, generating a tumor pathology image training set based on the tumor pathology image and the segmentation label corresponding to the tumor pathology image; preprocessing the tumor pathology image training set, inputting the preprocessed tumor pathology image training set into a pre-established offset graph convolutional contrastive learning network model for training, and obtaining a trained offset graph convolutional contrastive learning network model; receiving a tumor pathology image to be segmented, inputting the tumor pathology image to be segmented into the trained offset graph convolutional contrastive learning network model, and outputting a tumor pathology image segmentation result to achieve accurate image segmentation.

Description

Technical Field

[0001] The present invention relates to the field of image segmentation technology, and in particular to a tumor pathology image segmentation method and system based on a graph convolutional contrast learning network. Background Art

[0002] By deeply integrating data-driven learning capabilities with the precision needs of the medical field, deep learning-based tumor pathology image segmentation methods not only overcome the efficiency bottlenecks and subjective biases of manual analysis, but also provide a technical foundation for the transition of pathology diagnosis from qualitative description to quantitative analysis. Tumor pathology images are complex and have ambiguous boundaries between classes. Tissue morphology is diverse and heterogeneous. The boundaries between different classes in tumor pathology images may exhibit irregular shapes (such as the jagged shape of tumor infiltration boundaries) or contain transitional regions (such as the gradual boundary between inflammatory cells and normal tissue), making it difficult to define adjacency relationships using simple geometric rules. For example, the boundary between poorly differentiated tumors and stroma may lack clear texture distinctions, making traditional methods prone to misclassifying adjacent regions as the same class. The impact of imaging noise and annotation errors. Factors such as uneven staining and varying tissue section thickness can introduce image noise, causing pixel features in adjacent regions to overlap (e.g., similar grayscale values ​​between tumor cells and normal cells). Furthermore, manually annotated boundaries may contain subjective errors, resulting in unclear "true values" of adjacency relationships in the training data, hindering model learning. In tumor pathology image segmentation, the adjacency relationship between different tissue types (e.g., tumor, stroma, etc.) is crucial for accurate segmentation, but existing methods often struggle to effectively capture this information. Common loss functions (such as cross-entropy and Dice Loss) focus primarily on the classification accuracy of individual pixels rather than the category consistency of adjacent pixels. While data augmentation operations such as random rotation and flipping can improve model generalization, they can disrupt the inherent spatial structure in tumor pathology images (e.g., the anatomical relationship of organs), making it difficult for the network to learn stable adjacency patterns. Summary of the Invention

[0003] In order to address the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a tumor pathology image segmentation method and system based on graph convolutional contrastive learning network.

[0004] In a first aspect, the purpose of the present invention can be achieved by the following technical solution: a tumor pathology image segmentation method based on a graph convolutional contrastive learning network, the method comprising the following steps:

[0005] Obtaining a tumor pathology image and a segmentation label corresponding to the tumor pathology image, and generating a tumor pathology image training set based on the tumor pathology image and the segmentation label corresponding to the tumor pathology image;

[0006] Preprocessing a tumor pathology image training set, inputting the preprocessed tumor pathology image training set into a pre-established offset graph convolutional contrastive learning network model for training, and obtaining a trained offset graph convolutional contrastive learning network model;

[0007] The tumor pathology image to be segmented is received, the tumor pathology image to be segmented is input into the trained offset graph convolutional contrast learning network model, and the tumor pathology image segmentation result is output.

[0008] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the tumor pathology image training set includes tumor pathology images and its corresponding segmentation label , Represents the number of tumor pathology images in the training set.

[0009] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: a process of preprocessing the tumor pathology image training set includes:

[0010] For each tumor pathology image , the color image transformed into other color spaces is represented as , The segmentation label is also .

[0011] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the pre-established offset graph convolution contrastive learning network model includes: a feature extractor , offset graph convolutional network and decoder .

[0012] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the feature extractor Using a U-Net based encoder, By the total Level coding module; when the tumor pathology image Send in Afterwards, Level Encoding Module Output features ,in, Indicates the Level Encoding Module Output features, Level Encoding Module The input features are ; When tumor pathology images Send in Afterwards, Level Encoding Module Output features ,in, Indicates the Level Encoding Module Output features, Level Encoding Module The input features are .

[0013] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the offset graph convolutional network Using graph convolutional subnetwork , Depend on When the tumor pathology image Send in After that, for the feature extractor Output features ,in, Represents the number of channels, height and width of the feature respectively; the feature extractor Output Level Features Using adaptive pooling operation becomes Features with the same number of pixels ,in, Representation characteristics The number of channels. Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of , Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of ;

[0014] For the For the level graph convolution layer , , , , Indicates the Level Graph Convolutional Layer No. nodes, express No. The connection of the first level graph convolutional layer Node and The edge between the nodes, Level adjacency matrix , the adjacency matrix The calculation formula for each element is:

[0015]

[0016] in, Represents calculation The cosine similarity between two vectors; calculate the Degree Matrix The diagonal elements of , the rest of the elements are 0, where , is the identity matrix; compute the graph convolution subnetwork No. Level The graph convolution features of each node are as follows:

[0017]

[0018] in, represents the activation function, Indicates the Level The graph convolution features of nodes, represents a learnable parameter; when hour, Corresponding to The pixel features, Corresponding to The pixel features, Corresponding to The pixel features;

[0019] Using graph convolutional subnetwork , Depend on When the tumor pathology image Send in After that, for the feature extractor Output features ,in, Represents the number of channels, height and width of the feature respectively; the feature extractor Output Level Features Using adaptive pooling operation becomes Features with the same number of pixels ,in, Representation characteristics The number of channels; Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of , Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of ;

[0020] For the For the level graph convolution layer , , , , Indicates the Level Graph Convolutional Layer No. nodes, express No. The connection of the first level graph convolutional layer Node and The edge between the nodes; Level adjacency matrix , the adjacency matrix The calculation formula for each element is:

[0021]

[0022] in, Represents calculation The cosine similarity between two vectors is calculated Degree Matrix The diagonal elements of , the rest of the elements are 0, where , is the identity matrix; compute the graph convolution subnetwork No. Level The graph convolution features of each node are as follows:

[0023]

[0024] in, represents the activation function, Indicates the Level The graph convolution features of nodes, represents a learnable parameter; when hour, Corresponding to The pixel features, Corresponding to The pixel features, Corresponding to The pixel features;

[0025] Using graph convolutional subnetwork , Depend on Level graph convolution layer composition, for feature extractor Output image features 、 Sum the elements to get ,in, Represents the number of channels, height and width of the feature respectively;

[0026] For the For the level graph convolution layer , subgraph , , express The first level of the graph convolutional layer nodes, express The connection of the first level graph convolutional layer Nodes With the Nodes The edge between Level adjacency matrix , the adjacency matrix The calculation formula for each element is:

[0027]

[0028] in, Represents calculation Cosine similarity between two vectors, Representing a subgraph No. The characteristics of a node, Representing a subgraph No. Calculate the characteristics of the node. Degree Matrix The diagonal elements of , the rest of the elements are 0, where , is the identity matrix; graph convolution subnetwork No. Level The graph convolution feature of each node is calculated as follows:

[0029]

[0030] in, represents the activation function, Indicates the Level The graph convolution features of nodes, Indicates the Level The graph convolution features of nodes, 、 represents a learnable parameter.

[0031] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the decoder The processing process:

[0032] Will Plus Construct updated image features Feed into decoder , The structure uses the UNet decoder, which consists of Level decoding module composition; for tumor pathology images No. The input of the decoding module consists of upsampled 、 Composition, when hour, ; For tumor pathology images No. The input samples of the decoding module 、 Composition, when hour, , for tumor pathology images Decoder Output is segmentation map ; For tumor pathology images Decoder Output is segmentation map .

[0033] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the loss of the trained offset graph convolutional contrastive learning network model is as follows:

[0034] Through contrastive learning loss Maximize the similarity between the convolutional features of corresponding graphs in different color spaces:

[0035]

[0036] in, represents the exponential function;

[0037] Defining geometric consistency loss , constraining the geometric consistency of the adjacency relationship of the convolution graphs in different color spaces:

[0038]

[0039] in, Representation feature map 、 height, Representation feature map 、 width;

[0040] Calculate segmentation loss :

[0041]

[0042] in, Representing an image The number of pixels, Indicates the number of categories for segmentation, express No. Class pixels, Represents the segmentation map No. Class pixels, Represents the segmentation map No. Class pixels;

[0043] Total loss of offset graph convolutional contrastive learning network model for:

[0044] .

[0045] In a second aspect, in order to achieve the above-mentioned purpose, the present invention discloses a tumor pathology image segmentation system based on a graph convolutional contrastive learning network, comprising:

[0046] An image acquisition module is used to acquire tumor pathology images and segmentation labels corresponding to the tumor pathology images, and generate a tumor pathology image training set based on the tumor pathology images and the segmentation labels corresponding to the tumor pathology images;

[0047] A model training module is used to preprocess the tumor pathology image training set, input the preprocessed tumor pathology image training set into a pre-established offset graph convolutional contrast learning network model for training, and obtain a trained offset graph convolutional contrast learning network model;

[0048] The image segmentation module is used to receive the tumor pathology image to be segmented, input the tumor pathology image to be segmented into the trained offset graph convolution contrast learning network model, and output the tumor pathology image segmentation result.

[0049] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, the tumor pathology image segmentation method based on the graph convolutional contrast learning network as described above is adopted.

[0050] Beneficial effects of the present invention:

[0051] The present invention transforms each tumor pathology image into another color space and feeds it into the feature extractor. This is to achieve data amplification and improve the generalization of the segmentation model. The features of the two different color space images are then fed into the offset graph convolution contrast learning network for fusion, reducing the impact of image noise on segmentation caused by factors such as uneven staining.

[0052] The offset graph convolution contrast learning network of the present invention includes a subnetwork constructed for the original image By using different levels of image features, we can dynamically adjust the adjacency relationship of different layers of the graph convolutional network, and capture the personalized morphology of different patients in the original color space. Sub-networks constructed for other color images By using different levels of image features, we can dynamically adjust the adjacency relationship of different layers of the graph convolutional network, and capture the personalized morphology of different patients in other color spaces. ,Through different levels of color space image offset features, the ,adjacency relationships of different layers of the graph convolutional ,network can be dynamically adjusted, and the changes in color space image offset ,features of different patients can be captured.

[0053] The contrastive learning loss of the present invention maximizes the similarity between the convolution features of corresponding graphs in different color spaces, and the geometric consistency loss is used to constrain the geometric consistency of the convolution adjacency relationship of corresponding graphs in different color spaces, thereby improving the model learning adjacent areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0055] Figure 1 It is a schematic flow chart of the method of the present invention;

[0056] Figure 2 Schematic diagram of the offset graph convolution contrastive learning network of the present invention;

[0057] Figure 3 2. It is a schematic diagram comparing the image segmentation method of the present invention and the segmentation results of the existing method;

[0058] Figure 4 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making any creative efforts shall fall within the scope of protection of the present invention.

[0060] Example 1:

[0061] like Figure 1 As shown, a tumor pathology image segmentation method based on a graph convolutional contrast learning network includes the following steps:

[0062] S101: Obtain a tumor pathology image and a segmentation label corresponding to the tumor pathology image, and generate a tumor pathology image training set based on the tumor pathology image and the segmentation label corresponding to the tumor pathology image;

[0063] Among them, tumor pathology images It is a color image represented by the red, green and blue color space. The color image transformed into other color spaces (for example, hue, saturation and brightness space) is represented as ,Will and After inputting the offset graph convolution contrast learning network, the corresponding segmentation image is obtained and .

[0064] The tumor pathology image training set includes tumor pathology images and its corresponding segmentation label , Represents the number of tumor pathology images in the training set.

[0065] The process of preprocessing the tumor pathology image training set includes:

[0066] For each tumor pathology image , the color image transformed into other color spaces (for example, hue saturation brightness space) is represented as , The segmentation label is also .

[0067] S102: preprocessing the tumor pathology image training set, inputting the preprocessed tumor pathology image training set into a pre-established offset graph convolutional contrastive learning network model for training, thereby obtaining a trained offset graph convolutional contrastive learning network model;

[0068] like Figure 2 As shown, the pre-established offset graph convolution contrast learning network model includes: feature extractor , offset graph convolutional network and decoder ;

[0069] Among them, the feature extractor , the structure can use the UNet encoder, composed of When the tumor pathology image Send in Afterwards, Level encoding module output features ; When tumor pathology images Send in Afterwards, Level encoding module output features .

[0070] Among them, the offset graph convolutional network It is an undirected network consisting of three subgraph convolutional networks constitute.

[0071] Among them, the decoder , the structure can use the UNet decoder, which is composed of For tumor pathology images No. The input of the decoding module consists of upsampled 、 Composition; for tumor pathology images No. The input samples of the decoding module 、 Finally, for tumor pathology images Decoder Output is segmentation map ; For tumor pathology images Decoder Output is segmentation map .

[0072] Specifically, the feature extractor Using a U-Net based encoder, By the total Level coding module; when the tumor pathology image Send in Afterwards, Level Encoding Module Output features ,in, Indicates the Level Encoding Module Output features, Level Encoding Module The input features are ; When tumor pathology images Send in Afterwards, Level Encoding Module Output features ,in, Indicates the Level Encoding Module Output features, Level Encoding Module The input features are .

[0073] The offset graph convolutional network Using graph convolutional subnetwork , Depend on When the tumor pathology image Send in After that, for the feature extractor Output features ,in, Represents the number of channels, height and width of the feature respectively; the feature extractor Output Level Features Using adaptive pooling operation becomes Features with the same number of pixels ,in, Representation characteristics The number of channels. Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of , Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of ;

[0074] For the For the level graph convolution layer , , , , Indicates the Level Graph Convolutional Layer No. nodes, express No. The connection of the first level graph convolutional layer Node and The edge between the nodes, Level adjacency matrix , the adjacency matrix The calculation formula for each element is:

[0075]

[0076] in, Represents calculation The cosine similarity between two vectors; calculate the Degree Matrix The diagonal elements of , the rest of the elements are 0, where , is the identity matrix; compute the graph convolution subnetwork No. Level The graph convolution features of each node are as follows:

[0077]

[0078] in, represents the activation function, Indicates the Level The graph convolution features of nodes, represents a learnable parameter; when hour, Corresponding to The pixel features, Corresponding to The pixel features, Corresponding to The pixel features;

[0079] Using graph convolutional subnetwork , Depend on When the tumor pathology image Send in After that, for the feature extractor Output features ,in, Represents the number of channels, height and width of the feature respectively; the feature extractor Output Level Features Using adaptive pooling operation becomes Features with the same number of pixels ,in, Representation characteristics The number of channels; Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of , Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of ;

[0080] For the For the level graph convolution layer , , , , Indicates the Level Graph Convolutional Layer No. nodes, express No. The connection of the first level graph convolutional layer Node and The edge between the nodes; Level adjacency matrix , the adjacency matrix The calculation formula for each element is:

[0081]

[0082] in, Represents calculation The cosine similarity between two vectors is calculated Degree Matrix The diagonal elements of , the rest of the elements are 0, where , is the identity matrix; compute the graph convolution subnetwork No. Level The graph convolution features of each node are as follows:

[0083]

[0084] in, represents the activation function, Indicates the Level The graph convolution features of nodes, represents a learnable parameter; when hour, Corresponding to The pixel features, Corresponding to The pixel features, Corresponding to The pixel features;

[0085] Using graph convolutional subnetwork , Depend on Level graph convolution layer composition, for feature extractor Output image features 、 Sum the elements to get ,in, Represents the number of channels, height and width of the feature respectively;

[0086] For the For the level graph convolution layer , subgraph , , express The first level of the graph convolutional layer nodes, express The connection of the first level graph convolutional layer Nodes With the Nodes The edge between Level adjacency matrix , the adjacency matrix The calculation formula for each element is:

[0087]

[0088] in, Represents calculation Cosine similarity between two vectors, Representing a subgraph No. The characteristics of a node, Representing a subgraph No. Calculate the characteristics of the node. Degree Matrix The diagonal elements of , the rest of the elements are 0, where , is the identity matrix; graph convolution subnetwork No. Level The graph convolution feature of each node is calculated as follows:

[0089]

[0090] in, represents the activation function, Indicates the Level The graph convolution features of nodes, Indicates the Level The graph convolution features of nodes, 、 represents a learnable parameter.

[0091] The decoder The processing process:

[0092] Will Plus Construct updated image features Feed into decoder , The structure uses the UNet decoder, which consists of Level decoding module composition; for tumor pathology images No. The input of the decoding module consists of upsampled 、 Composition, when hour, ; For tumor pathology images No. The input samples of the decoding module 、 Composition, when hour, , for tumor pathology images Decoder Output is segmentation map ; For tumor pathology images Decoder Output is segmentation map .

[0093] S103: Receive the tumor pathology image to be segmented, input the tumor pathology image to be segmented into the trained offset graph convolutional contrast learning network model, and output the tumor pathology image segmentation result.

[0094] The loss of the trained offset graph convolutional contrastive learning network model is as follows:

[0095] Through contrastive learning loss Maximize the similarity between the convolutional features of corresponding graphs in different color spaces:

[0096]

[0097] in, Represents the exponential function.

[0098] At the same time, in order to capture the local geometric alignment information, the geometric consistency loss is defined , constraining the geometric consistency of the adjacency relationship of the convolution graphs in different color spaces:

[0099]

[0100] in, Representation feature map 、 height, Representation feature map 、 width.

[0101] Calculate segmentation loss :

[0102]

[0103] in, Representing an image The number of pixels, Indicates the number of categories for segmentation, express No. Class pixels, Represents the segmentation map No. Class pixels, Represents the segmentation map No. Class pixels;

[0104] Total loss of offset graph convolutional contrastive learning network model for:

[0105] .

[0106] Specifically, the present invention is further described below through examples. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and experiments.

[0107] First, the data used in this paper was collected from an international public dataset (Semantic Segmentation of Breast Cancer, BCSS). This dataset includes whole-slide pathology images from 151 different patients, all stained with hematoxylin and eosin (HE). These whole-slide tumor pathology images were annotated by an experienced pathologist and include all regions within the image for five component categories: Tumor, Stroma, Lymphocytic Infiltrate, Necrosis, and Others. 100 images served as the training set, and 51 images served as the test set. The whole-slide tumor pathology images were processed using the following steps: A 512×512 sliding window was used to crop the whole-slide tumor pathology images into non-overlapping partial slice images. The cropped images comprised 32,000 partial slice images in the training set, and 15,000 partial slice images in the test set.

[0108] This paper uses the PyTorch deep learning framework, Torch version 1.13.0. All training and validation processes were performed on an NVIDIA GeForce RTX 3090 graphics card equipped with 24GB of video memory. During training, the neural network used a mini-batch approach to read data, with a batch size of 2. The Stochastic Gradient Descent (SGD) optimizer was used, with an initial learning rate of 0.01, an optimizer momentum of 0.9, and an optimizer regularization coefficient of 0.0001. The Poly method was used as the learning rate adjustment strategy.

[0109] The experimental results of the present invention are as follows:

[0110] In order to quantitatively evaluate the performance of the proposed method, two different evaluation metrics were selected to evaluate the performance of the neural network in the multi-component segmentation task of breast cancer: Dice similarity coefficient (DSC) and intersection over union (IoU). The DSC calculation formula is:

[0111]

[0112] in, is the number of true positive segmented pixels, is the number of false positive segmented pixels, is the number of false negative segmented pixels. The calculation formula of the intersection-over-union ratio is:

[0113]

[0114] The present invention compares the segmentation results with existing methods: multi-resolution self-supervised method, domain increment method, image hybrid synthesis method, consistency regularization method, and learnable hint method, as shown in Table 1. Table 1 compares the segmentation results of the present invention with those of others.

[0115] Table 1 Comparison of experimental results between the method of the present invention and the existing method

[0116]

[0117]

[0118] As shown in Table 1, compared to other existing methods, the proposed method outperforms other existing methods in terms of segmentation accuracy, both in terms of DSC and IoU. Compared to the domain increment method with the best IoU among other methods, the proposed method improves IoU by 2.64, reaching an IoU of 73.84%. Compared to the multi-resolution self-supervised method with the best DSC among other methods, the proposed method improves DSC by 2.15, reaching a DSC of 81.64.

[0119] The image segmentation method of the present invention is compared with the segmentation results of the existing method. Figure 3 As shown, Figure 3 Columns (a)-(h) represent the tumor pathology image, the segmentation label, the segmentation results of the proposed method, the segmentation results of the multi-resolution self-supervised method, the segmentation results of the domain increment method, the segmentation results of the image hybrid synthesis method, the segmentation results of the consistency regularization method, and the segmentation results of the learnable hint method, respectively. Other methods exhibited numerous segmentation errors. For example, in columns (d)-(h) of the first row, other methods missed lymphocyte infiltration; in columns (d)-(g) of the second row, the boundary between tumor and stroma segmented by other methods did not align with the boundary labeled in column (b) of the second row; and in column (h) of the second row, other methods exhibited numerous misclassifications between tumor, stroma, necrosis, and other categories.

[0120] In addition, in order to verify the effectiveness of each part of the present invention, an ablation experiment was conducted, with the baseline model being UNet; other color spaces, segmentation losses were added in turn. , Subgraph Convolutional Network , Subgraph Convolutional Network , Subgraph Convolutional Network , contrastive learning loss , geometric consistency loss The ablation experiment results are shown in Table 2:

[0121] Table 2 Ablation experiment results

[0122]

[0123] From the ablation experiment in Table 2 above, it can be seen that compared with UNet, both DSC and IoU are gradually improving, which shows that each part of the method proposed in this invention is helpful to improve the segmentation accuracy.

[0124] Example 2: In order to achieve the above purpose, Figure 4 As shown, based on the first embodiment, the present invention discloses a tumor pathology image segmentation system based on a graph convolutional contrastive learning network, including:

[0125] An image acquisition module 11 is configured to acquire tumor pathology images and segmentation labels corresponding to the tumor pathology images, and generate a tumor pathology image training set based on the tumor pathology images and the segmentation labels corresponding to the tumor pathology images;

[0126] The model training module 12 is used to preprocess the tumor pathology image training set, input the preprocessed tumor pathology image training set into a pre-established offset graph convolutional contrast learning network model for training, and obtain a trained offset graph convolutional contrast learning network model;

[0127] The image segmentation module 13 is used to receive the tumor pathology image to be segmented, input the tumor pathology image to be segmented into the trained offset graph convolution contrast learning network model, and output the tumor pathology image segmentation result.

[0128] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0129] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, performs the above-described method. The storage medium may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0130] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0131] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present disclosure. Various changes and improvements may be made to the present disclosure without departing from the spirit and scope of the present disclosure, and such changes and improvements shall fall within the scope of the present disclosure.

Claims

1. A tumor pathology image segmentation method based on graph convolutional contrastive learning network, characterized by: The method comprises the following steps: Obtaining a tumor pathology image and a segmentation label corresponding to the tumor pathology image, and generating a tumor pathology image training set based on the tumor pathology image and the segmentation label corresponding to the tumor pathology image; The tumor pathology image training set includes tumor pathology images and its corresponding segmentation label , , represents the number of tumor pathology images in the training set; The process of preprocessing the tumor pathology image training set includes: For each tumor pathology image , the color image transformed into other color spaces is represented as , The segmentation label is ; Preprocessing a tumor pathology image training set, inputting the preprocessed tumor pathology image training set into a pre-established offset graph convolutional contrastive learning network model for training, and obtaining a trained offset graph convolutional contrastive learning network model; The pre-established offset graph convolution contrastive learning network model includes: feature extractor , offset graph convolutional network and decoder ; The feature extractor Using a U-Net based encoder, By the total Level coding module; when the tumor pathology image Send in Afterwards, Level Encoding Module Output features ,in, , Indicates the Level Encoding Module Output features, Level Encoding Module The input features are ; When tumor pathology images Send in Afterwards, Level Encoding Module Output features ,in, Indicates the Level Encoding Module Output features, Level Encoding Module The input features are ; The offset graph convolutional network Using graph convolutional subnetwork , Depend on When the tumor pathology image Send in After that, for the feature extractor Output features ,in, Represents the number of channels, height and width of the feature respectively; the feature extractor Output Level Features Using adaptive pooling operation becomes Features with the same number of pixels ,in, Representation characteristics The number of channels; Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of , Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of ; For the For the level-level graph convolutional layer, , , , , Indicates the Level Graph Convolutional Layer No. nodes, express No. The connection of the first level graph convolutional layer Node and The edge between the nodes, Level adjacency matrix , the adjacency matrix The calculation formula for each element is: in, Represents calculation The cosine similarity between two vectors; calculate the Degree Matrix The diagonal elements of , the rest of the elements are 0, where , is the identity matrix; compute the graph convolution subnetwork No. Level The graph convolution features of each node are as follows: in, represents the activation function, Indicates the Level The graph convolution features of nodes, represents a learnable parameter; when hour, Corresponding to The pixel features, Corresponding to The pixel features, Corresponding to The pixel features; Using graph convolutional subnetwork , Depend on When the tumor pathology image Send in After that, for the feature extractor Output features ,in, Represents the number of channels, height and width of the feature respectively; the feature extractor Output Level Features Using adaptive pooling operation becomes Features with the same number of pixels ,in, Representation characteristics The number of channels; Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of , Middle pixel features After the double-layer perceptron, the number of channels becomes The eigenvector of ; For the For the level-level graph convolutional layer, , , , , Indicates the Level Graph Convolutional Layer No. nodes, express No. The connection of the first level graph convolutional layer Node and The edge between the nodes; Level adjacency matrix , the adjacency matrix The calculation formula for each element is: in, Represents calculation The cosine similarity between two vectors is calculated Degree Matrix The diagonal elements of , the rest of the elements are 0, where , is the identity matrix; compute the graph convolution subnetwork No. Level The graph convolution features of each node are as follows: in, represents the activation function, Indicates the Level The graph convolution features of nodes, represents a learnable parameter; when hour, Corresponding to The pixel features, Corresponding to The pixel features, Corresponding to The pixel features; Using graph convolutional subnetwork , Depend on Level graph convolution layer composition, for feature extractor Output image features 、 Sum the elements to get ,in, Represents the number of channels, height and width of the feature respectively; For the For the level-level graph convolutional layer, , subgraph , , express The first level of the graph convolutional layer nodes, express The connection of the first level graph convolutional layer Nodes With the Nodes The edge between Level adjacency matrix , the adjacency matrix The calculation formula for each element is: in, Represents calculation Cosine similarity between two vectors, Representing a subgraph No. The characteristics of a node, Representing a subgraph No. The characteristics of each node; Calculate the Degree Matrix The diagonal elements of , the rest of the elements are 0, where , is the identity matrix; graph convolution subnetwork No. Level The graph convolution feature of each node is calculated as follows: in, represents the activation function, Indicates the Level The graph convolution features of nodes, Indicates the Level The graph convolution features of nodes, 、 represents a learnable parameter; The tumor pathology image to be segmented is received, the tumor pathology image to be segmented is input into the trained offset graph convolutional contrast learning network model, and the tumor pathology image segmentation result is output.

2. The tumor pathology image segmentation method based on graph convolutional contrastive learning network according to claim 1, characterized in that: The decoder The processing process: Will Plus Construct updated image features Feed into decoder , The structure uses the UNet decoder, which consists of Level decoding module composition; for tumor pathology images No. The input of the decoding module consists of upsampled 、 Composition, when hour, ; For tumor pathology images No. The input of the level decoding module consists of the sampled 、 Composition, when hour, , for tumor pathology images Decoder Output is segmentation map ; For tumor pathology images Decoder Output is segmentation map .

3. The tumor pathology image segmentation method based on graph convolutional contrastive learning network according to claim 2, characterized in that: The loss of the trained offset graph convolutional contrastive learning network model is as follows: Through contrastive learning loss Maximize the similarity between the convolutional features of corresponding graphs in different color spaces: in, represents the exponential function; Defining geometric consistency loss , constraining the geometric consistency of the adjacency relationship of the convolution graphs in different color spaces: in, Representation feature map 、 height, Representation feature map 、 Width; Calculate segmentation loss : in, Representing an image The number of pixels, Indicates the number of categories for segmentation, express No. Class pixels, Represents the segmentation map No. Class pixels, Represents the segmentation map No. Class pixels; Total loss of offset graph convolutional contrastive learning network model for: 。 4. A tumor pathology image segmentation system based on a graph convolutional contrastive learning network, which adopts the tumor pathology image segmentation method based on a graph convolutional contrastive learning network according to any one of claims 1 to 3, characterized in that: include: An image acquisition module is used to acquire tumor pathology images and segmentation labels corresponding to the tumor pathology images, and generate a tumor pathology image training set based on the tumor pathology images and the segmentation labels corresponding to the tumor pathology images; A model training module is used to preprocess the tumor pathology image training set, input the preprocessed tumor pathology image training set into a pre-established offset graph convolutional contrast learning network model for training, and obtain a trained offset graph convolutional contrast learning network model; The image segmentation module is used to receive the tumor pathology image to be segmented, input the tumor pathology image to be segmented into the trained offset graph convolution contrast learning network model, and output the tumor pathology image segmentation result.

5. A terminal device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on a processor. When the processor loads and executes the computer program, the tumor pathology image segmentation method based on a graph convolutional contrast learning network described in any one of claims 1 to 3 is adopted.

Images