A method for extracting tumor pathological microenvironment features

Abstract

The present invention relates to a method for extracting features of a tumor pathological microenvironment, comprising the following steps: S1, inputting the coordinates of all cell types on a tumor microenvironment pathological image; wherein the coordinates are pixel-level two-dimensional coordinates; S2, generating a list of each cell type and a master table containing all cells; S3, numbering and connecting the points in the master table, and generating an edge table; S4, performing module partitioning using a community discovery algorithm; S5, identifying the cell types; and S6, counting the number of connections and spatial distances between each two cell types based on the edge table. The present invention achieves accurate identification of different cell types, precise statistics of the number of connections between two cell types, and rapid calculation of the spatial relationship between the two cell types, providing powerful assistance for studying the connections between various cell types in the tumor microenvironment.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image feature extraction, and in particular to a method for extracting tumor pathological microenvironment features. Background Art

[0002] With the development of computer and machine learning technologies, the application of image databases has become increasingly widespread. How to effectively establish, manage, and fully utilize image database resources to achieve intelligent and efficient image recognition and retrieval has been a hot topic for researchers both domestically and internationally. These databases have found widespread application in scientific research, defense and military, industrial production, aerospace, biomedicine, traffic monitoring, and other fields. Faced with the large volume, multimodality, complex imaging, and highly targeted nature of biomedical image data, feature extraction technology continues to face new opportunities and challenges.

[0003] Currently, many tumor microenvironment feature extraction methods extract visual features of the image, such as cell color, grayscale, shape, texture, and edge features. These features reflect the characteristics of the cells themselves, which have no correlation with the numerous other cells in the tumor microenvironment and cannot reflect the connection between various cells.

[0004] Therefore, it is necessary to find a method to extract features of the tumor pathological microenvironment, such as the number of connections and spatial relationships between cells, to investigate whether there are connections between various cell types in the tumor microenvironment. These features can provide reference for disease diagnosis, provide a basis for prognosis prediction, and serve as research data for discovering new targeted therapies. Summary of the Invention

[0005] To solve the technical problems existing in the prior art, the present invention provides a method for extracting features of a tumor pathological microenvironment. By inputting the coordinates of all types of cells on a tumor microenvironment pathological image, a list of each type of cell and a general table containing all cells are generated. The points in the general table are numbered and connected, and an edge table is generated. A community discovery algorithm is used for module division and cell type identification. The number of connections and spatial distances between each two types of cells are counted based on the edge table, thereby achieving accurate identification of different types of cells, precise statistics of the number of connections between two types of cells, and rapid calculation of the spatial relationship between the two types of cells.

[0006] The present invention adopts the following technical solution to achieve: a method for extracting tumor pathological microenvironment characteristics, comprising the following steps:

[0007] S1. Input the coordinates of all cell types on the tumor microenvironment pathology image; the coordinates are pixel-level two-dimensional coordinates;

[0008] S2. Generate a list of each type of cell and a master table containing all cells; store the coordinates of each type of cell in its own empty list, create a new empty table as the master table, and append the lists containing the coordinates of each type of cell to the end of the master table in a fixed order;

[0009] S3. Number and connect the points in the total table and generate an edge table; number the cells in the total table starting from 0 in the form of natural numbers, connect all the cells in the total table using a coordinate connection algorithm to generate an edge table;

[0010] S4. Use the community discovery algorithm to perform module division. First, use the LPA label propagation algorithm to iterate the cell connection graph k times to obtain the community rough division result. Then use the Louvain algorithm to perform community detection on the community rough division result. During this process, the quality function is used multiple times to evaluate the partition quality of the community detection result to find the community with the highest partition quality.

[0011] S5. Identify the cell type; calculate the length of the list storing the coordinates of each type of cell to obtain the number of cells of each type; merge the lists of each type of cell in a specific order and number the cells as natural numbers to obtain the number interval of the coordinates of each type of cell in the total list; and determine the type of the cell by finding the number interval to which the cell number belongs;

[0012] S6. Count the number of connections and spatial distances between each two types of cells according to the edge table; establish a loop, traverse each edge in each community edge table, determine the number interval where the numbers of the two endpoints of each edge are located to determine which two types of cells the edge is connected to, add 1 to the counter that counts the number of connections between the two types of cells, find the coordinates of the two endpoints in the total table by number, calculate the Euclidean distance between the two endpoints of each edge and accumulate them, end the loop, and take the quotient of the total distance between each two types of cells and the corresponding number of connections to obtain the corresponding average distance.

[0013] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0014] 1. The present invention inputs the coordinates of all types of cells on the tumor microenvironment pathology image, generates a list of each type of cell and a general table containing all cells, numbers and connects the points in the general table, and generates an edge table. It then uses a community discovery algorithm to perform module division and identify cell types. The edge table then counts the number of connections and spatial distances between each two types of cells, achieving accurate identification of different types of cells, precise statistics of the number of connections between two types of cells, and rapid calculation of the spatial relationship between two types of cells, providing powerful assistance for studying the connections between various types of cells in the tumor microenvironment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a flow chart of the method of the present invention;

[0016] Figure 2 This is a DAPI-stained image of a tumor pathological microenvironment section;

[0017] Figure 3 (a) is a CD4 staining image of a tumor pathological microenvironment section;

[0018] Figure 3 (b) is a CD20 staining image of tumor pathological microenvironment sections;

[0019] Figure 3 (c) is a CD38 staining image of tumor pathological microenvironment sections;

[0020] Figure 3 (d) is a CD66B staining image of tumor pathological microenvironment sections;

[0021] Figure 3 (e) is a FOXP3 staining image of tumor pathological microenvironment sections;

[0022] Figure 4 It is a schematic diagram of the triangulated network output by the triangulation algorithm;

[0023] Figure 5 This is a schematic diagram of community discovery results;

[0024] Figure 6 This is a schematic diagram of the results of differentiating different types of cells;

[0025] Figure 7 It is a local schematic diagram of the feature extraction results of each community. DETAILED DESCRIPTION

[0026] The present invention will be described in further detail below with reference to the embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0027] Example

[0028] like Figure 1 As shown, this embodiment provides a method for extracting tumor pathological microenvironment features, including the following steps:

[0029] S1. Input the coordinates of all cell types on the tumor microenvironment pathology image; the coordinates are pixel-level two-dimensional coordinates;

[0030] S2. Generate a list of each type of cell and a master table containing all cells; store the coordinates of each type of cell in its own empty list, create a new empty table as the master table, and append the lists containing the coordinates of each type of cell to the end of the master table in a fixed order;

[0031] S3. Number and connect the points in the total table and generate an edge table; number the cells in the total table starting from 0 in the form of natural numbers, connect all the cells in the total table using a coordinate connection algorithm to generate an edge table;

[0032] S4. Use the community discovery algorithm to perform module division. First, use the LPA label propagation algorithm to iterate the cell connection graph k times to obtain the community rough division result. Then use the Louvain algorithm to perform community detection on the community rough division result. During this process, the quality function is used multiple times to evaluate the partition quality of the community detection result to find the community with the highest partition quality.

[0033] S5. Identify the cell type; calculate the length of the list storing the coordinates of each type of cell to obtain the number of cells of each type; merge the lists of each type of cell in a specific order and number the cells as natural numbers to obtain the number interval of the coordinates of each type of cell in the total list; and determine the type of the cell by finding the number interval to which the cell number belongs;

[0034] S6. Count the number of connections and spatial distances between each two types of cells according to the edge table; establish a loop, traverse each edge in each community edge table, determine the number interval where the numbers of the two endpoints of each edge are located to determine which two types of cells the edge is connected to, add 1 to the counter that counts the number of connections between the two types of cells, find the coordinates of the two endpoints in the total table by number, calculate the Euclidean distance between the two endpoints of each edge and accumulate them, end the loop, and take the quotient of the total distance between each two types of cells and the corresponding number of connections to obtain the corresponding average distance.

[0035] like Figure 2 As shown in this example, the two-dimensional coordinates of all cells on the DAPI staining map of the tumor pathological microenvironment slice are used as the research object, and Figure 2 Take as an example for implementation.

[0036] like Figure 3 (a) Figure 3 (b) Figure 3 (c) Figure 3 (d) Figure 3 As shown in (e), in this embodiment, CD4, CD20, CD38, CD66B, and FOXP3 staining images of tumor pathological microenvironment slices are used to assist in finding target cell areas and denoising.

[0037] like Figure 4 As shown, the coordinate connection algorithm in step S3 is a triangulation (Delaunay) algorithm, including the Lawson algorithm and the Bowyer-Watson algorithm. In this embodiment, the Bowyer-Watson algorithm is used for triangulation, which mainly includes the following steps:

[0038] S31. Construct a super triangle containing all scattered points and put them into the triangle linked list;

[0039] S32. Insert the scattered points in the point set in sequence, find the triangle whose circumcircle contains the insertion point in the triangle list as the influencing triangle of the point, delete the common edges of the influencing triangles, connect the insertion point with all the vertices of the influencing triangles, and thus complete the insertion of a point in the Delaunay triangle list;

[0040] S33, optimizing the newly formed local triangle according to the optimization criterion;

[0041] S34, putting the formed triangle into the Delaunay triangle list;

[0042] S35. Execute step S32 in a loop until all scattered points are inserted.

[0043] In this embodiment, the circumscribed circle of each triangle is found by calculating the center of the circumscribed circle. The formula for calculating the center of the circumscribed circle is as follows:

[0044]

[0045]

[0046] Among them, (a1,b1), (a2,b2), (a3,b3) are the coordinates of the three vertices of the triangle, and (a,b) are the coordinates of the center of the circumscribed circle.

[0047] In this embodiment, the edge table in step S3 is a list consisting of each edge after connecting the coordinates of all cells in the total table, and each edge is represented by the numbers of its two endpoints.

[0048] In this embodiment, the LPA label propagation algorithm in step S4 determines the characteristic properties of a point in the graph by using the neighboring nodes around the point. Closely connected points are divided into the same module to achieve the effect of "birds of a feather flock together". The algorithm includes the following steps:

[0049] Assign a unique label to all nodes;

[0050] Traverse each node and change its label to the label that appears most frequently among all its neighbors;

[0051] Iterate k times to roughly divide the community.

[0052] In this embodiment, the Louvain algorithm includes the following steps:

[0053] In the first stage, each node is considered as a separate community, and the nodes are moved to another community that maximizes the value of the quality function until all nodes cannot be moved;

[0054] The second phase treats the community in the first phase as a supernode and reconstructs the network;

[0055] These two stages are iterated until the quality function no longer improves.

[0056] like Figure 5 As shown in the figure, by combining the LPA label propagation algorithm and the Louvain algorithm, community discovery of cell networks is achieved. Each community has its own characteristics, and the cells within the community are closely connected, and there are differences in the cell functions between communities.

[0057] In this embodiment, the expression of the quality function in step S4 is as follows:

[0058]

[0059] Among them, n i is the number of nodes in community i, D(r i ||r) represents the density r of community i i The KL divergence of the density r of the entire graph is calculated as follows:

[0060]

[0061] By calculating the mass change ΔX of an edge moving from community a to community b (v) To control the end of the algorithm, the expression is as follows:

[0062]

[0063] in, a is the number of nodes in community a, n b is the number of nodes in community b, and These are the density changes of community a and community b respectively.

[0064] like Figure 6 As shown, the cell type is identified by finding the number interval to which the cell number belongs, and different colors are used to represent different cell types.

[0065] In this embodiment, in step S6, the Euclidean distance of each edge in the connection graph is calculated, and the average of the Euclidean distances of all cell types is taken as the spatial relationship between the two cells. The two-dimensional Euclidean distance expression is as follows:

[0066]

[0067] Where ρ is the Euclidean distance between the point (x1, y1) and the point (x2, y2), and |X| is the Euclidean distance from the point (x2, y2) to the origin.

[0068] In this embodiment, the spatial distance between each two types of cells is calculated according to the following formula:

[0069]

[0070] Where k is the number of connections between each two cell types, (x m ,y m ) and (x n ,y n ) are the coordinates of the two endpoints of any edge.

[0071] like Figure 7 As shown, in this embodiment, a loop is set up to traverse each edge in each community of cell connections, identifying which two cell types the edge connects by using the endpoint numbers, and simultaneously counting the number of connections and calculating the distance. When the entire image is traversed, the extracted features of each community are saved and output in an Excel spreadsheet.

[0072] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A method for extracting tumor pathological microenvironment features, characterized in that: The following steps are involved: S1. Input the coordinates of all cell types on the tumor microenvironment pathology image; the coordinates are pixel-level two-dimensional coordinates; S2. Generate a list of each type of cell and a master table containing all cells; store the coordinates of each type of cell in its own empty list, create a new empty table as the master table, and append the lists containing the coordinates of each type of cell to the end of the master table in a fixed order; S3. Number and connect the points in the total table and generate an edge table; number the cells in the total table starting from 0 in the form of natural numbers, connect all the cells in the total table using a coordinate connection algorithm to generate an edge table; S4. Use the community discovery algorithm to perform module division. First, use the LPA label propagation algorithm to iterate the cell connection graph k times to obtain the community rough division result. Then use the Louvain algorithm to perform community detection on the community rough division result. During this process, the quality function is used multiple times to evaluate the partition quality of the community detection result to find the community with the highest partition quality. S5. Identify the cell type; calculate the length of the list storing the coordinates of each type of cell to obtain the number of cells of each type; merge the lists of each type of cell in a specific order and number the cells as natural numbers to obtain the number interval of the coordinates of each type of cell in the total list; and determine the type of the cell by finding the number interval to which the cell number belongs; S6. Count the number of connections and spatial distances between each two types of cells according to the edge table; establish a loop, traverse each edge in each community edge table, determine the number interval where the numbers of the two endpoints of each edge are located to determine which two types of cells the edge is connected to, add 1 to the counter that counts the number of connections between the two types of cells, find the coordinates of the two endpoints in the total table by number, calculate the Euclidean distance between the two endpoints of each edge and accumulate them, end the loop, and take the quotient of the total distance between each two types of cells and the corresponding number of connections to obtain the corresponding average distance.

2. The method for extracting tumor pathological microenvironment characteristics according to claim 1, characterized in that: The coordinate connection algorithm in step S3 is a triangulation Delaunay algorithm, including the Lawson algorithm and the Bowyer-Watson algorithm. Among them, the triangulation using the Bowyer-Watson algorithm mainly includes the following steps: S31. Construct a super triangle containing all scattered points and put them into the triangle linked list; S32. Insert the scattered points in the point set in sequence, find the triangle whose circumcircle contains the insertion point in the triangle list as the influencing triangle of the point, delete the common edges of the influencing triangles, connect the insertion point with all vertices of the influencing triangles, and complete the insertion of the point in the Delaunay triangle list; S33, optimizing the newly formed local triangle according to the optimization criterion; S34, putting the formed triangle into the Delaunay triangle list; S35. Execute step S32 in a loop until all scattered points are inserted.

3. The method for extracting tumor pathological microenvironment features according to claim 2, characterized in that: The circumcircle of each triangle is found by calculating the center of the circumcircle. The formula for calculating the center of the circumcircle is as follows: Among them, (a1,b1), (a2,b2), (a3,b3) are the coordinates of the three vertices of the triangle, and (a,b) are the coordinates of the center of the circumscribed circle.

4. The method for extracting tumor pathological microenvironment characteristics according to claim 1, characterized in that: The LPA label propagation algorithm in step S4 determines the characteristic properties of a point in the graph by the neighboring nodes around the point. Closely connected points are divided into the same module. The algorithm includes the following steps: Assign a unique label to all nodes; Traverse each node and change its label to the label that appears most frequently among all its neighbors; Iterate k times to roughly divide the community.

5. The method for extracting tumor pathological microenvironment features according to claim 1, characterized in that: The Louvain algorithm in step S4 includes the following steps: In the first stage, each node is considered as a separate community and all nodes are traversed by moving a node to another community that maximizes the value of the quality function; The second phase treats the community in the first phase as a supernode and reconstructs the network; These two stages are iterated until the quality function is maximized.

6. The method for extracting tumor pathological microenvironment characteristics according to claim 1, characterized in that: The expression of the quality function in step S4 is as follows: Among them, n i is the number of nodes in community i, D(r i ||r) represents the density r of community i i The KL divergence of the density r of the entire graph is calculated as follows: By calculating the mass change ΔX of an edge moving from community a to community b (v) To control the end of the algorithm, the expression is as follows: in, a is the number of nodes in community a, n b is the number of nodes in community b, and These are the density changes of community a and community b respectively.

7. The method for extracting tumor pathological microenvironment characteristics according to claim 1, characterized in that: In step S6, the Euclidean distance of each edge in the connection graph is calculated, and the average of the Euclidean distances of all cell types is taken as the spatial relationship between the two cells. The two-dimensional Euclidean distance expression is as follows: Where ρ is the Euclidean distance between the point (x1, y1) and the point (x2, y2), and |X| is the Euclidean distance from the point (x2, y2) to the origin.

8. The method for extracting tumor pathological microenvironment features according to claim 7, characterized in that: The spatial distance between each two types of cells is calculated according to the following formula: Where k is the number of connections between each two cell types, (x m ,y m ) and (x n ,y n ) are the coordinates of the two endpoints of any edge.

Images