A lymphocyte classification method and device for Sjögren's syndrome

Abstract

The present application discloses a lymphocyte classification method and device for Sjögren's syndrome, the method comprising: extracting a cell feature list and a tissue feature list in a pathological image matrix using different models; calculating the inter-cell distance matrix and the cell-tissue distance matrix based on the cell feature list and the tissue feature list; constructing a cell graph model between cells according to the cell feature list, the inter-cell distance matrix and the preset cell feature information; constructing a tissue graph model of cells and tissues according to the tissue feature list, the cell-tissue distance matrix and the preset cell feature information; performing graph embedding evolution analysis on the cell graph model and the tissue graph model respectively by the LightGCN algorithm to obtain the corresponding cell and tissue embedding evolution results; performing cell classification based on the cell and tissue embedding evolution results to obtain lymphocyte information. The present application can solve the technical problem that the prior art relies on manual judgment, resulting in poor efficiency and accuracy.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method and device for classifying lymphocytes of Sjögren's syndrome. Background Art

[0002] Sjogren's syndrome (SS) is one of the most common autoimmune diseases. It is characterized by chronic inflammation and progressive destruction of the salivary glands and lacrimal glands. Patients with lesions that only affect the lacrimal glands and salivary glands are called primary Sjogren's syndrome (pSS); secondary Sjogren's syndrome (sSS) is accompanied by autoimmune diseases such as rheumatoid arthritis and systemic lupus erythematosus. SS can seriously affect the quality of life and basic daily functions such as eating, speaking and sleeping. If SS can be detected, diagnosed and treated early, the patient's prognosis is good. Otherwise, it will cause damage to organs such as the lungs, kidneys, stomach, nerves, muscles, joints and limbs.

[0003] The diagnosis of SS is based on three aspects: salivary gland histopathological changes, clinical manifestations, and immunological changes. The histopathological evaluation of the salivary glands plays an irreplaceable role in the diagnosis of SS. Since SS patients often involve both large and small salivary glands, and their histopathology is basically the same, minor salivary gland biopsy (MSGB) is often used clinically.

[0004] The current technology for judging lymphocyte infiltration foci and counting lymphocytes in Sjögren's syndrome is to have pathologists make judgments and roughly count them one by one under a microscope. This method has the problems of being time-consuming, overly subjective, low in efficiency and poor in accuracy, and is difficult to meet the actual clinical analysis needs. Summary of the invention

[0005] The present application provides a method and device for classifying lymphocytes for Sjögren's syndrome, which are used to solve the technical problem that the prior art relies too much on manual judgment, resulting in poor efficiency and accuracy.

[0006] In view of this, the first aspect of the present application provides a lymphocyte classification method for Sjögren's syndrome, comprising:

[0007] A Stardist model is used to extract a cell feature list from a target pathological image matrix, wherein the cell feature list includes a plurality of cell features;

[0008] Extracting a tissue feature list in the target pathological image matrix by using a Segment Anything segmentation model, wherein the tissue feature list includes a plurality of tissue features;

[0009] Calculate the inter-cell distance matrix and the cell-tissue distance matrix based on the cell feature list and the tissue feature list respectively;

[0010] Constructing association edges between cells according to the cell feature list, the inter-cell distance matrix and preset cell feature information, and generating a cell graph model;

[0011] Constructing an association relationship between cells and tissue structures according to the tissue feature list, the cell-tissue distance matrix and the preset cell feature information, and generating a tissue map model;

[0012] Performing graph embedding evolution analysis on the cell graph model and the tissue graph model respectively by using the LightGCN algorithm to obtain corresponding cell embedding evolution results and tissue embedding evolution results;

[0013] Cell classification is performed based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information, wherein the lymphocyte information includes a lymphocyte list and a lymphocyte distance matrix.

[0014] Preferably, the Stardist model is used to extract a cell feature list from the target pathological image matrix, wherein the cell feature list includes a plurality of cell features, and also includes:

[0015] Obtaining the acquisition path of the target pathological image;

[0016] The target pathological image is extracted based on the acquisition path, and a target pathological image matrix is ​​generated.

[0017] Preferably, the Stardist model is used to extract a cell feature list from the target pathological image matrix, and the cell feature list includes multiple cell features, including:

[0018] The Stardist model is used to extract cells from the target pathological image matrix to obtain a mask list, wherein each cell in the mask list corresponds to a mask matrix;

[0019] The center coordinates and multiple cell features of each cell in the image are obtained according to the mask list to generate a cell feature list, wherein the cell features include grayscale mean, grayscale variance, area, perimeter, major axis length, minor axis length, eccentricity and compactness of the cell.

[0020] Preferably, the calculating of the inter-cell distance matrix and the cell-tissue distance matrix based on the cell feature list and the tissue feature list, respectively, comprises:

[0021] Calculate the central Euclidean distance between any two cells based on the cell information in the cell feature list, and subtract the central Euclidean distance from the sum of the radii of the two cells to obtain the inter-cell distance, thereby generating an inter-cell distance matrix;

[0022] The relative Euclidean distance between cells and tissues is calculated according to the cell feature list and the tissue feature list to generate a cell-tissue distance matrix.

[0023] Preferably, the performing cell classification based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information includes:

[0024] Extracting node feature vectors of nodes corresponding to the same cell from the cell embedding evolution result and the tissue embedding evolution result respectively;

[0025] Concatenate the node feature vector with a preset single-cell feature vector to obtain a graph fusion feature vector;

[0026] A preset classifier is used to perform binary classification of cells based on the input graph fusion feature vector to obtain lymphocyte information.

[0027] Preferably, the cell classification is performed based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information, and then further comprises:

[0028] Constructing a lymphocyte cluster model according to the lymphocyte information, wherein each cell in the lymphocyte cluster model serves as a node;

[0029] Detecting connected domains in the lymphocyte cluster model by using a breadth-based search algorithm to obtain a connected domain set;

[0030] The number of nodes in each of the connected domains is counted, and if the number of nodes is greater than a pathological threshold, the corresponding target pathological image is determined to be positive.

[0031] The second aspect of the present application provides a lymphocyte classification device for Sjögren's syndrome, comprising:

[0032] A cell feature extraction unit, used for extracting a cell feature list from a target pathological image matrix using a Stardist model, wherein the cell feature list includes a plurality of cell features;

[0033] A tissue feature extraction unit, used for extracting a tissue feature list from the target pathological image matrix by using a Segment Anything segmentation model, wherein the tissue feature list includes a plurality of tissue features;

[0034] A distance matrix calculation unit, used to calculate the inter-cell distance matrix and the cell-tissue distance matrix based on the cell feature list and the tissue feature list respectively;

[0035] A cell graph model construction unit, used to construct association edges between cells according to the cell feature list, the inter-cell distance matrix and preset cell feature information, and generate a cell graph model;

[0036] A tissue map model construction unit, used to construct an association relationship between cells and tissue structures according to the tissue feature list, the cell-tissue distance matrix and the preset cell feature information, and generate a tissue map model;

[0037] An embedding evolution analysis unit, used to perform graph embedding evolution analysis on the cell graph model and the tissue graph model respectively by using the LightGCN algorithm to obtain corresponding cell embedding evolution results and tissue embedding evolution results;

[0038] A lymphocyte classification unit is used to classify cells based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information, wherein the lymphocyte information includes a lymphocyte list and a lymphocyte distance matrix.

[0039] Preferably, the cell feature extraction unit is specifically used for:

[0040] The Stardist model is used to extract cells from the target pathological image matrix to obtain a mask list, wherein each cell in the mask list corresponds to a mask matrix;

[0041] The center coordinates and multiple cell features of each cell in the image are obtained according to the mask list to generate a cell feature list, wherein the cell features include grayscale mean, grayscale variance, area, perimeter, major axis length, minor axis length, eccentricity and compactness of the cell.

[0042] Preferably, the distance matrix calculation unit is specifically used for:

[0043] Calculate the central Euclidean distance between any two cells based on the cell information in the cell feature list, and subtract the central Euclidean distance from the sum of the radii of the two cells to obtain the inter-cell distance, thereby generating an inter-cell distance matrix;

[0044] The relative Euclidean distance between cells and tissues is calculated according to the cell feature list and the tissue feature list to generate a cell-tissue distance matrix.

[0045] Preferably, the lymphocyte classification unit is specifically used for:

[0046] Extracting node feature vectors of nodes corresponding to the same cell from the cell embedding evolution result and the tissue embedding evolution result respectively;

[0047] Concatenate the node feature vector with a preset single-cell feature vector to obtain a graph fusion feature vector;

[0048] A preset classifier is used to perform binary classification of cells based on the input graph fusion feature vector to obtain lymphocyte information.

[0049] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0050] In the present application, a lymphocyte classification method for Sjögren's syndrome is provided, including: extracting a cell feature list from a target pathological image matrix using a Stardist model, the cell feature list including multiple cell features; extracting a tissue feature list from a target pathological image matrix using a Segment Anything segmentation model, the tissue feature list including multiple tissue features; calculating an inter-cell distance matrix and a cell-tissue distance matrix based on the cell feature list and the tissue feature list, respectively; constructing association edges between cells according to the cell feature list, the inter-cell distance matrix and preset cell feature information, and generating a cell graph model; constructing an association relationship between cells and tissue structures according to the tissue feature list, the cell-tissue distance matrix and preset cell feature information, and generating a tissue graph model; performing graph embedding evolution analysis on the cell graph model and the tissue graph model respectively using the LightGCN algorithm to obtain corresponding cell embedding evolution results and tissue embedding evolution results; performing cell classification based on the cell embedding evolution results and the tissue embedding evolution results to obtain lymphocyte information, the lymphocyte information including a lymphocyte list and a lymphocyte distance matrix.

[0051] The lymphocyte classification method for Sjögren's syndrome provided in this application analyzes the cells and tissue structures in the target pathological image matrix respectively by feature extraction, and fuses the graph structure to ensure that in the process of cell classification, the features used for classification include both the characteristics of the cell itself and the characteristics of the cells and tissues surrounding the cell, thereby improving the classification accuracy; and using the Segment Anything segmentation model as a segmentation method, it is possible to more accurately extract tissue structure information that is beneficial to cell classification from pathological images, providing more reliable feature support for cell classification. Therefore, this application can solve the technical problem that the prior art relies too much on manual judgment, resulting in poor efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic diagram of a flow chart of a lymphocyte classification method for Sjögren's syndrome provided in an embodiment of the present application;

[0053] Figure 2 A schematic diagram of the structure of a lymphocyte classification device for Sjögren's syndrome provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0055] For easier understanding, see Figure 1 , an embodiment of a lymphocyte classification method for Sjögren's syndrome provided in the present application comprises:

[0056] Step 101: Use the Stardist model to extract a cell feature list from a target pathological image matrix, where the cell feature list includes multiple cell features.

[0057] Furthermore, step 101 includes:

[0058] The Stardist model is used to extract cells from the target pathological image matrix to obtain a mask list, in which each cell corresponds to a mask matrix;

[0059] The center coordinates and multiple cell features of each cell in the image are obtained according to the mask list to generate a cell feature list. The cell features include the grayscale mean, grayscale variance, area, perimeter, major axis length, minor axis length, eccentricity and compactness of the cell.

[0060] It should be noted that the Stardist model can use the cell extraction model in its database to extract cells from the target pathology image matrix to obtain the cell mask. Each cell corresponds to a mask matrix, and all cells as a whole form a mask list, that is, there are multiple mask matrices in the mask list. In addition, the target pathology image matrix is ​​a digital expression of the target pathology image, and the target pathology image can be directly read as a matrix form for easy analysis and calculation.

[0061] According to the mask list, the center coordinates of each cell in the image can be counted, and then the corresponding grayscale mean, grayscale variance, area, perimeter, major axis length, minor axis length, eccentricity and compactness can be calculated according to the cell mask; except for compactness, other parameters can be calculated based on existing technology and will not be described here. The calculation method of compactness is:

[0062]

[0063] Among them, C u is the tightness, P u , S u It is understood that, in addition to the above information, the cell features can also calculate other feature information based on the mask list, which is only an example and is not limited to this.

[0064] Furthermore, step 101, before that, also includes:

[0065] Obtaining the acquisition path of the target pathological image;

[0066] The target pathological image is extracted based on the acquisition path, and a target pathological image matrix is ​​generated.

[0067] It can be understood that the acquisition path here refers to the storage path of the target pathological image. The target pathological image can be extracted based on the acquisition path and directly read into a matrix form to obtain the expression of the target pathological image matrix, providing guarantee for subsequent calculation and analysis.

[0068] Step 102: extract a tissue feature list from the target pathological image matrix through the Segment Anything segmentation model, where the tissue feature list includes multiple tissue features.

[0069] The Segment Anything segmentation model can segment any object in the image without the need for labeled data; this eliminates the need for additional labeled data related to the tissue structure for training the segmentation model. This embodiment uses the Segment Anything segmentation model to extract tissue structure information from the target pathology image matrix, and can also obtain a mask matrix of the tissue structure, which constitutes a mask list as a whole; moreover, the mask matrix has the same size as the target pathology image matrix, 1 represents the tissue area, and 0 represents the background area.

[0070] The center coordinates of each tissue structure in the image and multiple tissue structure features are obtained according to the tissue structure mask list to generate a tissue feature list. The tissue structure features include the grayscale mean, grayscale variance, area, perimeter, major axis length, minor axis length, eccentricity and compactness of the tissue.

[0071] Step 103: Calculate the inter-cell distance matrix and the cell-tissue distance matrix based on the cell feature list and the tissue feature list respectively.

[0072] Further, in step 103, the central Euclidean distance between any two cells is calculated based on the cell information in the cell feature list, and the central Euclidean distance is subtracted from the sum of the radii of the two cells to obtain the inter-cell distance, thereby generating an inter-cell distance matrix;

[0073] The relative Euclidean distance between cells and tissues is calculated based on the cell feature list and the tissue feature list to generate a cell-tissue distance matrix.

[0074] It should be noted that the cell-to-cell distance matrix refers to the distance matrix formed by the relative distance between two cells, while the cell-tissue distance matrix refers to the distance matrix formed by the relative distance between cells and tissues.

[0075] The inter-cell distance in this embodiment is not a direct distance calculation, but first calculates the center Euclidean distance between two cells, and then subtracts it from the radius of the two cells to obtain the inter-cell distance. The specific calculation process is expressed as:

[0076]

[0077] Among them, S u , S v are the areas of cells u and v, respectively. The cell radius is calculated by taking the square root of the ratio of the area to π. c (u,v) is the Euclidean distance between two cells, d e (u,v) is the distance between cells. By solving the distance between all cells, we can get the distance matrix between cells, which is m×m in size, where m is the number of cells.

[0078] The cell-tissue distance is based on the center coordinates and cell radius of the cell, the center coordinates and radius of the tissue structure to calculate the relative distance between each cell and other tissues. The calculation method is the same as the calculation process of the distance between cells, so it will not be repeated here. The size of the cell-tissue distance matrix is ​​m×n, where n is the number of tissue structures.

[0079] Step 104: construct the association edges between cells according to the cell feature list, the cell distance matrix and the preset cell feature information, and generate a cell graph model.

[0080] The construction process of the cell graph model is essentially the process of constructing the associated edges between cells. Each cell can find a maximum of k1 cells and k2 tissue structures within a preset distance range. According to the inter-cell distance matrix and the preset cell feature information, the cell and the found cells and tissues can be constructed to construct the corresponding graph structure, that is, the cells are taken as nodes, and the associations between cells are expressed as edges. Specifically, an associated edge can be established between the nodes corresponding to two cells whose distance is less than a certain threshold. It can be understood that the preset cell feature information can be set according to actual needs and is not limited here. Moreover, the cell feature information of a cell node can form a single cell feature vector, where the number of features is the dimension of the vector. For example, if there are d features, the size of the vector is d-dimensional.

[0081] Step 105: construct the association relationship between cells and tissue structures according to the tissue feature list, the cell-tissue distance matrix and the preset cell feature information, and generate a tissue map model.

[0082] The essence of the tissue map model is to describe the relationship between cells and tissues. Based on the preset cell feature information, tissue structure feature information, tissue feature list and cell-tissue distance matrix, the tissue structure associated with each cell can be found, and then the tissue map model can be generated. At this time, the node can be a cell or a tissue structure. All cells can build an association relationship with the tissue structure based on the distance threshold to generate an tissue map model.

[0083] Step 106: Use the LightGCN algorithm to perform graph embedding evolution analysis on the cell graph model and the tissue graph model respectively to obtain corresponding cell embedding evolution results and tissue embedding evolution results.

[0084] The LightGCN algorithm is an end-to-end graph recommendation model. Applying it to the graph embedding evolution analysis process of this embodiment can show stronger generalization ability and more efficient graph analysis ability in feature analysis. After the cell graph model performs graph embedding evolution analysis, the cell embedding evolution result can be obtained; after the organization graph model performs graph embedding evolution analysis, the organization embedding evolution result can be obtained.

[0085] Step 107: Perform cell classification based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information, where the lymphocyte information includes a lymphocyte list and a lymphocyte distance matrix.

[0086] Further, step 107 includes:

[0087] Extract node feature vectors of nodes corresponding to the same cell from the cell embedding evolution results and tissue embedding evolution results respectively;

[0088] The node feature vector is concatenated with the preset single-cell feature vector to obtain the graph fusion feature vector;

[0089] The preset classifier is used to perform binary classification of cells based on the input graph fusion feature vector to obtain lymphocyte information.

[0090] For the same cell, the feature vector of the corresponding node, i.e., the node feature vector, can be extracted from the cell embedding evolution result and the tissue embedding evolution result. In addition, the preset single cell feature vector can be the cell feature information in the above graph model, or can be configured according to actual needs, which is not limited here.

[0091] It is understandable that the preset classifier is a binary classifier and is a pre-trained model that can be directly used in the cell classification task. In addition, the preset classifier of this embodiment includes a fully connected neural network with two hidden layers. The process of cell binary classification is to determine whether the cell belongs to a lymphocyte by the preset classifier, filter non-lymphocytes, and obtain a lymphocyte list and a lymphocyte distance matrix.

[0092] Furthermore, step 107 further includes:

[0093] A lymphocyte cluster model is constructed based on lymphocyte information, and each cell in the lymphocyte cluster model is regarded as a node;

[0094] The connected domains are detected in the lymphocyte cluster model by using the breadth search algorithm to obtain the connected domain set;

[0095] The number of nodes in each connected domain is counted. If the number of nodes is greater than the pathological threshold, the corresponding target pathological image is judged to be positive.

[0096] It should be noted that the lymphocyte information includes the inter-cell distance matrix, so a lymphocyte cluster graph model can be constructed based on the distance between cells. In the graph model, each lymphocyte is regarded as a node. If the distance between every two lymphocytes is less than a given threshold, a connection is established between the two cell nodes.

[0097] The breadth search algorithm is also called the breadth-first search algorithm. It is a basic search algorithm that does not consider the possible location of the result and thoroughly searches the entire graph until it is found. Therefore, this embodiment uses the breadth search algorithm to find all connected domains in the lymphocyte cluster graph model to obtain a connected domain set in the graph model. Moreover, each detected connected domain represents a lymphocyte cluster.

[0098] The number of nodes in each connected domain, that is, the number of lymphocytes, is counted. If there is a connected domain with a number of nodes not less than the pathological threshold, it means that the corresponding pathological image is Sjögren's syndrome, that is, pathologically positive. On the contrary, if no lymphocyte cluster with a number of nodes greater than 50 is found in all connected domains, it means that there is no lymphocyte cluster that meets the conditions in the pathological image, and the pathological image is negative. This method can automatically classify and identify the pathological state of lymphocytes in Sjögren's syndrome, provide reliable theoretical support for the diagnosis of Sjögren's syndrome, and greatly improve the diagnostic efficiency of the disease.

[0099] The lymphocyte classification method for Sjögren's syndrome provided in the embodiment of the present application analyzes the cells and tissue structures in the target pathological image matrix respectively by feature extraction, and fuses the graph structure to ensure that in the process of cell classification, the features used for classification include both the characteristics of the cell itself and the characteristics of the cells and tissues surrounding the cell, thereby improving the classification accuracy; and using the Segment Anything segmentation model as a segmentation method, it is possible to more accurately extract tissue structure information that is beneficial to cell classification from the pathological image, providing more reliable feature support for cell classification. Therefore, the embodiment of the present application can solve the technical problem that the prior art relies too much on manual judgment, resulting in poor efficiency and accuracy.

[0100] For easier understanding, see Figure 2 The present application provides an embodiment of a lymphocyte classification device for Sjögren's syndrome, comprising:

[0101] A cell feature extraction unit 201 is used to extract a cell feature list from a target pathological image matrix using a Stardist model, where the cell feature list includes a plurality of cell features;

[0102] A tissue feature extraction unit 202 is used to extract a tissue feature list from a target pathological image matrix by using a Segment Anything segmentation model, where the tissue feature list includes a plurality of tissue features;

[0103] A distance matrix calculation unit 203, used to calculate the inter-cell distance matrix and the cell-tissue distance matrix based on the cell feature list and the tissue feature list respectively;

[0104] A cell graph model construction unit 204 is used to construct association edges between cells according to the cell feature list, the inter-cell distance matrix and the preset cell feature information, and generate a cell graph model;

[0105] The tissue map model construction unit 205 is used to construct the association relationship between cells and tissue structures according to the tissue feature list, the cell-tissue distance matrix and the preset cell feature information, and generate a tissue map model;

[0106] The embedding evolution analysis unit 206 is used to perform graph embedding evolution analysis on the cell graph model and the tissue graph model respectively by using the LightGCN algorithm to obtain corresponding cell embedding evolution results and tissue embedding evolution results;

[0107] The lymphocyte classification unit 207 is used to classify cells based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information, which includes a lymphocyte list and a lymphocyte distance matrix.

[0108] Furthermore, the cell feature extraction unit 201 is specifically used for:

[0109] The Stardist model is used to extract cells from the target pathological image matrix to obtain a mask list, in which each cell corresponds to a mask matrix;

[0110] The center coordinates and multiple cell features of each cell in the image are obtained according to the mask list to generate a cell feature list. The cell features include the grayscale mean, grayscale variance, area, perimeter, major axis length, minor axis length, eccentricity and compactness of the cell.

[0111] Further, the distance matrix calculation unit 203 is specifically used for:

[0112] Based on the cell information in the cell feature list, the central Euclidean distance between any two cells is calculated, and the central Euclidean distance is subtracted from the sum of the radii of the two cells to obtain the inter-cell distance, thereby generating an inter-cell distance matrix;

[0113] The relative Euclidean distance between cells and tissues is calculated based on the cell feature list and the tissue feature list to generate a cell-tissue distance matrix.

[0114] Furthermore, the lymphocyte classification unit 207 is specifically used for:

[0115] Extract node feature vectors of nodes corresponding to the same cell from the cell embedding evolution results and tissue embedding evolution results respectively;

[0116] The node feature vector is concatenated with the preset single-cell feature vector to obtain the graph fusion feature vector;

[0117] The preset classifier is used to perform binary classification of cells based on the input graph fusion feature vector to obtain lymphocyte information.

[0118] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0119] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0120] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0121] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for executing all or part of the steps of the method described in each embodiment of the present application through a computer device (which can be a personal computer, a server, or a network device, etc.). The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name in English: Read-Only Memory, English abbreviation: ROM), random access memory (full name in English: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store program codes.

[0122] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for classifying lymphocytes in Sjögren's syndrome, characterized in that: include: A Stardist model is used to extract a cell feature list from a target pathological image matrix, wherein the cell feature list includes a plurality of cell features; Extracting a tissue feature list in the target pathological image matrix by using a Segment Anything segmentation model, wherein the tissue feature list includes a plurality of tissue features; Calculate the inter-cell distance matrix and the cell-tissue distance matrix based on the cell feature list and the tissue feature list respectively; According to the cell feature list, the inter-cell distance matrix and the preset cell feature information, the associated edges between cells are constructed, and a cell graph model is generated. The specific process includes: Each cell finds the most cells and an organizational structure; construct the cell and the found cells and the corresponding graph structure of the organization according to the inter-cell distance matrix and the preset cell feature information; take the cell as a node, and express the association between cells as an edge; specifically, establish an association edge between the nodes corresponding to two cells whose distance is less than a preset threshold; the cell feature information of a cell node can form a single cell feature vector, where the number of features is the dimension of the vector; The association relationship between cells and tissue structures is constructed according to the tissue feature list, the cell-tissue distance matrix and the preset cell feature information, and a tissue map model is generated. The specific process includes: Based on the preset cell feature information, tissue structure feature information, tissue feature list and cell-tissue distance matrix, the tissue structure associated with each cell is found, and then the tissue map model is generated; the node includes a cell or a tissue structure; all cells are associated with the tissue structure based on the distance threshold to generate the tissue map model; Performing graph embedding evolution analysis on the cell graph model and the tissue graph model respectively by using the LightGCN algorithm to obtain corresponding cell embedding evolution results and tissue embedding evolution results; Cell classification is performed based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information, wherein the lymphocyte information includes a lymphocyte list and a lymphocyte distance matrix.

2. The method for classifying lymphocytes for Sjögren's syndrome according to claim 1, characterized in that: The Stardist model is used to extract a cell feature list from the target pathological image matrix, wherein the cell feature list includes multiple cell features, and also includes: Obtaining the acquisition path of the target pathological image; The target pathological image is extracted based on the acquisition path, and a target pathological image matrix is ​​generated.

3. The lymphocyte classification method for Sjögren's syndrome according to claim 1, characterized in that: The Stardist model is used to extract a cell feature list from the target pathological image matrix, and the cell feature list includes multiple cell features, including: The Stardist model is used to extract cells from the target pathological image matrix to obtain a mask list, wherein each cell in the mask list corresponds to a mask matrix; The center coordinates and multiple cell features of each cell in the image are obtained according to the mask list to generate a cell feature list, wherein the cell features include grayscale mean, grayscale variance, area, perimeter, major axis length, minor axis length, eccentricity and compactness of the cell.

4. The lymphocyte classification method for Sjögren's syndrome according to claim 1, characterized in that: The step of respectively calculating the inter-cell distance matrix and the cell-tissue distance matrix based on the cell feature list and the tissue feature list comprises: Calculate the central Euclidean distance between any two cells based on the cell information in the cell feature list, and subtract the central Euclidean distance from the sum of the radii of the two cells to obtain the inter-cell distance, thereby generating an inter-cell distance matrix; The relative Euclidean distance between cells and tissues is calculated according to the cell feature list and the tissue feature list to generate a cell-tissue distance matrix.

5. The lymphocyte classification method for Sjögren's syndrome according to claim 1, characterized in that: The cell classification based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information includes: Extracting node feature vectors of nodes corresponding to the same cell from the cell embedding evolution result and the tissue embedding evolution result respectively; Concatenate the node feature vector with a preset single-cell feature vector to obtain a graph fusion feature vector; A preset classifier is used to perform binary classification of cells based on the input graph fusion feature vector to obtain lymphocyte information.

6. The method for classifying lymphocytes for Sjögren's syndrome according to claim 1, characterized in that: The method further comprises: performing cell classification based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information, and then: Constructing a lymphocyte cluster model according to the lymphocyte information, wherein each cell in the lymphocyte cluster model serves as a node; Detecting connected domains in the lymphocyte cluster model by using a breadth-based search algorithm to obtain a connected domain set; The number of nodes in each of the connected domains is counted, and if the number of nodes is greater than a pathological threshold, the corresponding target pathological image is determined to be positive.

7. A lymphocyte classification device for Sjögren's syndrome, characterized in that: include: A cell feature extraction unit, used for extracting a cell feature list from a target pathological image matrix using a Stardist model, wherein the cell feature list includes a plurality of cell features; A tissue feature extraction unit, used for extracting a tissue feature list from the target pathological image matrix by using a Segment Anything segmentation model, wherein the tissue feature list includes a plurality of tissue features; A distance matrix calculation unit, used to calculate the inter-cell distance matrix and the cell-tissue distance matrix based on the cell feature list and the tissue feature list respectively; The cell graph model construction unit is used to construct the association edges between cells according to the cell feature list, the inter-cell distance matrix and the preset cell feature information, and generate a cell graph model. The specific process includes: Each cell finds the most cells and an organizational structure; construct the cell and the found cells and the corresponding graph structure of the organization according to the inter-cell distance matrix and the preset cell feature information; take the cell as a node, and express the association between cells as an edge; specifically, establish an association edge between the nodes corresponding to two cells whose distance is less than a preset threshold; the cell feature information of a cell node can form a single cell feature vector, where the number of features is the dimension of the vector; The tissue map model construction unit is used to construct the association relationship between cells and tissue structures according to the tissue feature list, the cell-tissue distance matrix and the preset cell feature information, and generate a tissue map model. The specific process includes: Based on the preset cell feature information, tissue structure feature information, tissue feature list and cell-tissue distance matrix, the tissue structure associated with each cell is found, and then the tissue map model is generated; the node includes a cell or a tissue structure; all cells are associated with the tissue structure based on the distance threshold to generate the tissue map model; An embedding evolution analysis unit, used to perform graph embedding evolution analysis on the cell graph model and the tissue graph model respectively by using the LightGCN algorithm to obtain corresponding cell embedding evolution results and tissue embedding evolution results; A lymphocyte classification unit is used to classify cells based on the cell embedding evolution result and the tissue embedding evolution result to obtain lymphocyte information, wherein the lymphocyte information includes a lymphocyte list and a lymphocyte distance matrix.

8. The lymphocyte classification device for Sjögren's syndrome according to claim 7, characterized in that: The cell feature extraction unit is specifically used for: The Stardist model is used to extract cells from the target pathological image matrix to obtain a mask list, wherein each cell in the mask list corresponds to a mask matrix; The center coordinates and multiple cell features of each cell in the image are obtained according to the mask list to generate a cell feature list, wherein the cell features include grayscale mean, grayscale variance, area, perimeter, major axis length, minor axis length, eccentricity and compactness of the cell.

9. The lymphocyte classification device for Sjögren's syndrome according to claim 7, characterized in that: The distance matrix calculation unit is specifically used for: Calculate the central Euclidean distance between any two cells based on the cell information in the cell feature list, and subtract the central Euclidean distance from the sum of the radii of the two cells to obtain the inter-cell distance, thereby generating an inter-cell distance matrix; The relative Euclidean distance between cells and tissues is calculated according to the cell feature list and the tissue feature list to generate a cell-tissue distance matrix.

10. The lymphocyte classification device for Sjögren's syndrome according to claim 7, characterized in that: The lymphocyte classification unit is specifically used for: Extracting node feature vectors of nodes corresponding to the same cell from the cell embedding evolution result and the tissue embedding evolution result respectively; Concatenate the node feature vector with a preset single-cell feature vector to obtain a graph fusion feature vector; A preset classifier is used to perform binary classification of cells based on the input graph fusion feature vector to obtain lymphocyte information.

Images