Pathological image analysis method and tumor computer-aided diagnosis method

Abstract

The embodiment of the invention provides a pathological image analysis method and a tumor computer-aided diagnosis method, and the method comprises the steps: obtaining pathological image information corresponding to a target pathological image which comprises at least one cell entity, the pathological image information comprises cell attribute information corresponding to each cell entity; acquiring density characteristic information corresponding to each cell entity based on the cell attribute information corresponding to each cell entity; generating pathological image feature information corresponding to the target pathological image according to the density feature information corresponding to each cell entity; and generating an image analysis result corresponding to the target pathological image based on the pathological image feature information. The pathological image information of the cell attribute information corresponding to the plurality of cell entities corresponds to the pathological image, so that the spatial characteristics of the cells in the pathological image can be comprehensively captured, and the limitation of a traditional method on clinical importance is made up.

Description

Technical Field

[0001] The embodiments of the present specification relate to the field of computer technology, and in particular to a pathological image analysis method. Background Art

[0002] With the rapid development of artificial intelligence (AI) technology, the field of pathological image analysis has benefited greatly. AI technology has significantly improved the accuracy and efficiency of diagnosis. Through methods such as deep learning, it can learn and identify complex patterns from a large amount of pathological image data, helping doctors to quickly process large-scale slices, locate lesions, and detect abnormalities that are difficult to detect with the naked eye.

[0003] Currently, pathological image analysis mainly uses traditional image perception frameworks to generate general image features by pre-training pathological basic models on large-scale datasets to support downstream tasks. However, this method ignores the spatial distribution of cells, which is closely related to molecular profiles, tumor progression, and prognostic biomarkers, limiting its comprehensiveness in clinical importance. Therefore, in order to solve the above problems, a pathological image analysis method is needed. Summary of the invention

[0004] In view of this, the embodiments of this specification provide a pathological image analysis method, a computer-aided diagnosis method for tumors, a pathological image analysis method applied to cloud devices, a computer-aided diagnosis method for tumors applied to clients, and a computer-aided diagnosis method for tumors applied to cloud devices. This specification also involves a computer-aided diagnosis system for tumors, a pathological image analysis device, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

[0005] According to a first aspect of an embodiment of this specification, a pathological image analysis method is provided, comprising:

[0006] Acquiring pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity;

[0007] Based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained;

[0008] Generating pathological image characteristic information corresponding to the target pathological image according to the density characteristic information corresponding to each cell entity;

[0009] An image analysis result corresponding to the target pathological image is generated based on the pathological image feature information.

[0010] According to a second aspect of the embodiments of this specification, a computer-aided diagnosis method for a tumor is provided, comprising:

[0011] Acquiring pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity;

[0012] Based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained;

[0013] Generating pathological image characteristic information corresponding to the target pathological image according to the density characteristic information corresponding to each cell entity;

[0014] An image recognition result corresponding to the target pathological image is generated based on the pathological image feature information.

[0015] According to a third aspect of the embodiments of this specification, a computer-aided diagnosis method for tumors is provided, which is applied to a client and includes:

[0016] Acquiring pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity;

[0017] Based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained;

[0018] According to the density feature information corresponding to each cell entity, pathological image feature information corresponding to the target pathological image is generated; and based on the pathological image feature information, an image recognition result corresponding to the target pathological image is generated.

[0019] According to a fourth aspect of the embodiments of this specification, a computer-aided diagnosis method for tumors is provided, which is applied to a cloud device, and includes:

[0020] A tumor diagnosis request sent by a receiving end-side device, wherein the tumor diagnosis request carries a target pathological image;

[0021] Acquiring pathological image information corresponding to the target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity;

[0022] Based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained;

[0023] Generating pathological image characteristic information corresponding to the target pathological image according to the density characteristic information corresponding to each cell entity;

[0024] An image recognition result corresponding to the target pathological image is generated based on the pathological image feature information, and the image recognition result is sent to the terminal side device.

[0025] According to a fifth aspect of an embodiment of this specification, a computing device is provided, including:

[0026] Memory and processor;

[0027] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above-mentioned pathological image analysis method and computer-aided diagnosis method for tumors are implemented.

[0028] According to the sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, which, when executed by a processor, implement the steps of the above-mentioned pathological image analysis method and computer-aided diagnosis method for tumors.

[0029] According to the seventh aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned pathological image analysis method and computer-aided diagnosis method for tumors.

[0030] By applying the solution of the embodiments of the present specification, pathological image information of the target pathological image is obtained, including cell attribute information of each cell entity, and density feature information of each cell entity is extracted based on this information, and pathological image feature information corresponding to the target pathological image is further generated, and an image analysis result of the target pathological image is generated. This method can effectively support a comprehensive analysis of the target pathological image. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a flow chart of a pathological image analysis method provided by an embodiment of this specification;

[0032] Figure 2 is a schematic diagram of a pathological image analysis process provided by an embodiment of this specification;

[0033] Figure 3 is a flow chart of a computer-aided diagnosis method for tumors provided by one embodiment of this specification;

[0034] Figure 4 is a flow chart of a pathological image analysis method applied to a cloud device provided by an embodiment of this specification;

[0035] Figure 5is a flowchart of a computer-aided diagnosis method for tumors applied to a client provided by an embodiment of this specification;

[0036] Figure 6 is a flow chart of a computer-aided diagnosis method for tumors applied to a cloud device provided by one embodiment of this specification;

[0037] Figure 7 is an architecture diagram of a computer-aided diagnosis system for tumors provided by one embodiment of this specification;

[0038] Figure 8 is a processing flow chart of a computer-assisted prognostic analysis method provided in one embodiment of the present specification;

[0039] Fig. 9 is a schematic diagram of the structure of a pathological image analysis device provided by an embodiment of this specification;

[0040] Fig.10 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0041] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0042] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0043] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0044] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0045] First, the terms involved in one or more embodiments of this specification are explained.

[0046] Principal Component Analysis (PCA): is a statistical method used to transform a set of possibly correlated variables into a set of linearly uncorrelated variables, called principal components, through orthogonal transformation. This method can be used for dimensionality reduction and extracting key information from data, and is often used for data preprocessing and feature extraction.

[0047] Farthest Point Sampling (FPS): It is a sampling algorithm in point cloud processing. The core idea is to select a set of points from the point cloud so that the distance between these points is as large as possible, thereby reducing the amount of data while retaining the structural characteristics of the point cloud.

[0048] Multilayer Perceptron (MLP): is a feedforward neural network consisting of at least three layers of nodes: input layer, hidden layer, and output layer. Except for the input node, each node is a neuron with a nonlinear activation function. MLP learns the pattern of input data by adjusting the weights and biases of neurons between layers, and is suitable for a variety of tasks such as classification and regression.

[0049] In this specification, a pathological image analysis method, a computer-aided diagnosis method for tumors, a pathological image analysis method applied to a cloud device, a computer-aided diagnosis method for tumors applied to a client, and a computer-aided diagnosis method for tumors applied to a cloud device are provided. This specification also involves a computer-aided diagnosis system for tumors, a pathological image analysis device, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0050] The use of artificial intelligence in pathological image analysis can significantly improve the accuracy and efficiency of diagnosis. Through technologies such as deep learning, artificial intelligence can automatically learn and identify complex patterns and features from large amounts of pathological image data. It can not only process and analyze thousands of slices in a short period of time, helping doctors quickly locate the lesion area, but also assist in detecting tiny abnormalities that are difficult to detect with the naked eye, reducing misdiagnosis and missed diagnosis due to fatigue or lack of experience; it helps with digital storage and transmission, making remote consultations more convenient and efficient, and promoting the optimal allocation of medical resources; in addition, artificial intelligence-based systems can also achieve effective tracking of disease progression, provide a basis for personalized treatment plans, and accelerate the discovery of new knowledge and the advancement of clinical practice.

[0051] However, existing methods usually analyze pathological images through traditional image perception frameworks, where image representation is the cornerstone of downstream tasks. Therefore, many pathological base models have been proposed, which are pre-trained on large-scale datasets for universal representation. However, unlike natural images, the analysis of the spatial distribution of cells within pathological images has been proven to be clinically important and related to molecular profiles, tumor progression, prognostic biomarkers, etc. Over-reliance on pathological base models, coupled with the neglect of the spatial distribution of cells, leads to high computational cost and poor performance.

[0052] See also Figure 1 , Figure 1 A flow chart of a pathological image analysis method provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0053] Step 102: Acquire pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity.

[0054] In practical applications, the target pathological image is the pathological image that needs to be analyzed, which usually contains a complete tissue or slice image and is used to identify disease characteristics or diagnostic indicators; pathological image information refers to data related to the target pathological image, covering cell distribution, morphology and other structured information that characterizes tissue pathological characteristics; cell entities are the basic components in pathological images, representing specific cells in tissues, such as tumor cells or immune cells; cell attribute information is the core feature of cell entities, including their category, location, morphological characteristics and other parameters used to describe cell status or function.

[0055] The target pathology image can be understood as the basic data source used to extract features and generate diagnostic results in pathology analysis. The target pathology image is usually a high-resolution digital slice image containing a variety of cell and tissue structures, such as a tissue slice from breast cancer or lung cancer. In practical applications, the target pathology image may be a local lesion area or a complete tissue slice image. For example, a target pathology image may include multiple tumor cell clusters, inflammatory areas, and normal tissue areas for further analysis of cancer staging or treatment plans.

[0056] Pathological image information can be understood as a multi-dimensional data set extracted from the target pathological image, which is used to describe the characteristics and spatial distribution of each component in the image. This information includes cell location information, category information (such as cancer cells, immune cells, etc.), density characteristics, and tissue region segmentation information. For example, the pathological image information of a breast cancer pathological image may include the number and distribution pattern of tumor cells, the proportion of immune cells in tumor infiltration, and the structural characteristics of the boundary between tumor and normal tissue.

[0057] Cell entities can be understood as each independent cell unit in the target pathological image, which is the basic building block of tissues and lesions. Cell entities can be represented as single cell objects in two-dimensional or three-dimensional images, usually associated with specific biological or pathological functions. For example, tumor cell entities may have a large nuclear area and high staining intensity, while immune cell entities may appear as small nuclei, high density distribution and located in the peripheral area of ​​the tumor. In pathological analysis, identifying these cell entities is a key step in extracting higher-level features (such as cell density or neighborhood relationships).

[0058] Cell attribute information can be understood as a set of specific parameters used to describe the characteristics of cell entities, including but not limited to the cell category, location, morphological characteristics (such as area, roundness, nuclear-cytoplasmic ratio), spatial relationship, and related molecular marker expression levels. For example, the attribute information of a tumor cell may include its category as "cancer cell", location as "(x, y) coordinates", area as "50μm 2 ”, the roundness is “0.8”, and is accompanied by specific markings.

[0059] Acquiring pathological image information corresponding to pathological images provides basic support for further analysis of the spatial distribution patterns of cells, histopathological characteristics and clinical indicators.

[0060] Furthermore, obtaining pathological image information corresponding to the target pathological image includes:

[0061] Acquire target pathological images;

[0062] Inputting the target pathological image into a cell annotation model to obtain cell position information and cell category information corresponding to each cell entity generated by the cell annotation model;

[0063] Based on the cell position information and cell category information corresponding to each cell entity, the cell attribute information corresponding to each cell entity is generated.

[0064] In practical applications, the cell annotation model is a model tool used to automatically identify and annotate cell entities in target pathological images. It masters the key features of cells by learning a large number of annotated samples. The cell location information is the geometric information that describes the specific spatial position of each cell entity in the pathological image, usually presented in the form of coordinates. The cell category information is the classification result of the biological properties of the cell entity, such as tumor cells, immune cells, etc.

[0065] The cell annotation model can be understood as an automated annotation tool built through training data using deep learning or other machine learning technologies, and is a model specifically used to extract cell-related information from target pathology images. For example, a cell annotation model for breast cancer slice images may be trained to identify and annotate cancer cells, immune cells, and other background cells. By inputting the target pathology image, the cell annotation model can automatically generate annotation results containing cell location information and cell category information, thereby greatly reducing the workload of manual annotation and improving efficiency and consistency.

[0066] Cell location information can be understood as the spatial coordinate data of each cell entity in the pathological image, usually expressed in two or three dimensions. For example, in a tissue slice, the cell location information can be recorded as the center point coordinates (x, y) of each cell. These location data not only provide a basis for the individual description of cells, but also lay an important foundation for the subsequent analysis of cell spatial distribution characteristics (such as density characteristics or neighborhood relationships). For example, by analyzing the cell location information in cancer slices, the spatial differences between the core area of ​​the tumor and the surrounding tissue can be identified.

[0067] Cell category information can be understood as the annotation data generated after classifying each cell entity, reflecting the category of the cell in terms of biological function or morphology. For example, in lung cancer pathology images, cell category information may include categories such as cancer cells, inflammatory cells, and vascular endothelial cells. This information is essential for analyzing the distribution patterns of different cell types in tissues and the interactions between them. For example, in the evaluation of immunotherapy, the spatial proximity and category ratio between immune cells and cancer cells can become important indicators, and these analyses are all based on accurate cell category information.

[0068] It should be noted that the cell annotation model can be obtained by obtaining a target pathological image and generating at least one pathological image slice based on the target pathological image, wherein the pathological image slice includes at least one cell entity; inputting each pathological image slice into the cell annotation model, obtaining the cell category information and cell position information corresponding to each cell entity in each pathological image slice generated by the cell annotation model, and determining the initial slice category corresponding to each pathological image slice according to the cell category information and cell position information corresponding to each cell entity in each pathological image slice; inputting each pathological image slice into a slice verification model, obtaining the target slice category corresponding to each pathological image slice generated by the slice verification model; training the above-mentioned cell annotation model according to the initial slice category and target slice category corresponding to each pathological image slice, and then further training the cell annotation model according to the data related to the target project. It can also be to obtain a pre-trained cell annotation model, and then further train the cell annotation model according to the data related to the target project, etc., and this specification does not impose any restrictions on this.

[0069] The cell location information and cell category information generated by the cell annotation model not only supports the further calculation of cell attribute information, but also provides an efficient and reliable data basis for subsequent pathological image feature extraction and image analysis, promoting the automation and refinement of pathological analysis.

[0070] Step 104: Obtain density feature information corresponding to each cell entity based on the cell attribute information corresponding to each cell entity.

[0071] In practical applications, density feature information is the key data for describing the spatial distribution characteristics of cells in the target pathological image, and is used to quantify and characterize the distribution pattern and proximity relationship of each cell entity within a specific spatial range.

[0072] For example, density feature information can be understood as a set of features extracted by statistically analyzing the spatial distribution of each cell entity in the pathological image. These features not only include the distribution status of individual cells in their spatial neighborhood, but also reflect the spatial structure and interaction relationship between cells in the entire tissue. For example, in a pathological image of breast cancer, the density feature information of the core area of ​​the tumor may indicate that the cells in this area are highly aggregated, while the density feature information of the surrounding tissue may show a lower cell distribution density. This spatial feature can reveal the boundary between the lesion area and the surrounding normal tissue, providing an important basis for analyzing the spread and invasion characteristics of the disease.

[0073] Through density feature information, we can fully understand the spatial distribution pattern of cells in the target pathological image. This feature has practical significance in many scenarios, such as tumor staging, lesion area identification, and cell microenvironment modeling. The extraction of density feature information lays a data foundation for subsequent pathological image analysis, and plays an important role in disease diagnosis, treatment effect evaluation, and the formulation of personalized medical plans.

[0074] Furthermore, density feature information corresponding to each cell entity is obtained based on the cell attribute information corresponding to each cell entity, including:

[0075] Parsing the cell attribute information corresponding to each cell entity, and obtaining the cell location information and cell category information corresponding to the cell attribute information corresponding to each cell entity;

[0076] Based on the cell position information and cell category information corresponding to each cell entity, density feature information corresponding to each cell entity is generated.

[0077] In practical applications, density feature information corresponding to each cell entity is generated based on the cell position information and cell category information corresponding to each cell entity. This can be understood as extracting features that characterize the degree of cell aggregation and neighborhood relationships by analyzing the cell spatial distribution pattern and category composition. The specific method can be to generate a density calculation radius threshold based on the cell position information corresponding to each cell entity, and determine at least one density calculation radius based on the density calculation radius threshold, so as to generate density feature information by counting the number of cells and category proportion within each density calculation radius; it can also be done by constructing a cell adjacency graph, treating the cell entity as a node in the graph, and generating density features with spatial distance as weight, and obtaining features by calculating the cell category distribution of each node and its neighborhood; it can also be done by dividing a fixed grid area, counting the number and category distribution of cells in each grid, and aggregating the grid data into the neighborhood of the target cell to calculate density feature information, etc. This manual does not impose any restrictions on this.

[0078] By generating density feature information corresponding to each cell entity through the cell position information and cell category information corresponding to each cell, the distribution pattern of cells in space and their relationship with surrounding cells can be effectively described, providing accurate characterization of tissue structure and pathological characteristics. This method can not only reveal the degree of aggregation and category distribution of cells in a local area, but also reflect the macroscopic spatial structure of cells in the overall tissue, laying the foundation for downstream pathological analysis tasks, such as cancer staging, tumor invasiveness assessment, and immune cell infiltration research, thereby greatly improving the accuracy and clinical relevance of the analysis.

[0079] Furthermore, based on the cell position information and cell category information corresponding to each cell entity, density feature information corresponding to each cell entity is generated, including:

[0080] generating a density calculation radius threshold according to the cell position information corresponding to each cell entity, and determining at least one density calculation radius based on the density calculation radius threshold;

[0081] Based on the density calculation radius and the cell position information and cell category information corresponding to each cell entity, density feature information corresponding to each cell entity is generated.

[0082] In practical applications, the density calculation radius threshold is the initial distance parameter used to determine the range of the cell neighborhood, and its purpose is to limit and define the spatial scale of the analysis; the density calculation radius is a multi-level spatial distance further subdivided within the density calculation radius threshold, which is used to calculate the cell density characteristics layer by layer.

[0083] The density calculation radius threshold can be understood as a reference value used to define the global scale of cell analysis in pathological images, which is automatically generated by combining cell location information and tissue characteristics. Its significance lies in providing a possible range limit value for subsequent cell neighborhood division. For example, in a tumor slice, the density calculation radius threshold can be determined based on the overall size of the tumor area and the cell density, thereby improving the efficiency of the calculation while ensuring that the analysis can cover the tumor core and its surrounding areas.

[0084] The density calculation radius can be understood as a multi-level spatial distance further divided on the basis of the density calculation radius threshold, which is used to capture the distribution pattern of cells in different spatial ranges. Its significance lies in that by refining the spatial scale, the aggregation relationship between cells and their interaction with the microenvironment can be described layer by layer. For example, in the analysis of breast cancer, multiple density calculation radii (such as 50μm, 100μm, 200μm) can be set to count the number and distribution characteristics of each cell within these ranges, thereby identifying the dense distribution in the core area of ​​the tumor and the sparse distribution in the peripheral area.

[0085] It should be noted that generating a density calculation radius threshold based on the cell position information corresponding to each cell entity can be understood as using the geometric distribution characteristics of cells in space to determine the global or local scale range of the analysis, so as to subsequently calculate the cell density characteristics. The specific method can be achieved by calculating the average shortest distance d between all cell entities in the data set. mean , and use the scaling factor λ r Dynamic adjustment, set the larger neighborhood radius r max =λ r d mean, to adaptively reflect the cell density; it is also possible to analyze the spatial distribution variance of cells in the pathological image, identify the difference between areas with higher spatial aggregation and areas with lower spatial aggregation, and thus generate a local density calculation radius with different weights; it is also possible to divide the pathological image into several cells based on the spatial grid division method, count the distribution of the number of cells in each cell, and calculate the global or local larger aggregation range as the density calculation radius threshold based on the statistical results, and this manual does not impose any restrictions on this.

[0086] It should be noted that the density feature information corresponding to each cell entity is generated based on each density calculation radius and the cell position information and cell category information corresponding to each cell entity. It can be understood that by comprehensively analyzing the spatial position and category distribution of the cells, density-related statistical features are extracted within different radius ranges to reflect the degree of spatial aggregation of cells and their relationship with other cell types. The specific method can be to take the target cell as the center, and count the number and category proportion of reference cells within different density calculation radii to form a density feature reflecting the neighborhood characteristics; it can also be based on the weight distribution of cell categories, and calculate the weighted average density of categories within each density calculation radius to quantify the contribution of different cell types to spatial characteristics; it can also construct a spatial relationship diagram, regard cell entities as nodes, calculate local graph features based on the distance between nodes and category information, and aggregate them into density features of target cells, etc. This manual does not impose any restrictions on this.

[0087] The density calculation method that uses multiple density calculation radii for stratification can accurately reflect the biological characteristics of tumors, thereby providing strong support for cancer staging and prognosis analysis.

[0088] Furthermore, based on the density calculation radius and the cell position information and cell category information corresponding to each cell entity, density feature information corresponding to each cell entity is generated, including:

[0089] Determine a target cell entity and at least one background cell entity, wherein the target cell entity is any one of the cell entities, and the background cell entity is any one of the cell entities except the target cell entity;

[0090] Based on each density calculation radius, the target cell position information corresponding to the target cell entity and the background cell position information corresponding to each background cell entity, determining a reference cell entity set of the target cell entity for each density calculation radius in each background cell entity, wherein the reference cell entity set includes at least one reference cell entity;

[0091] According to the cell category information corresponding to each reference cell entity in each reference cell entity set, the density feature information corresponding to the target cell entity is generated.

[0092] In practical applications, the background cell entity refers to all cell entities in the target pathological image except the target cell entity, and is mainly used to provide reference information of the environment in which the target cell is located; the reference cell entity set is a set of cell entities that have a certain spatial correlation with the target cell and are screened from the background cell entities based on the density calculation radius, and is mainly used to calculate the density feature information of the target cell.

[0093] Background cell entities can be understood as cell entities other than the target cell entity in the pathology slice image. These cell entities provide control information for analyzing the characteristics of the target cell. For example, in a breast cancer pathology image slice, when analyzing the neighborhood distribution characteristics of a tumor cell, for each cell entity, all cell entities other than the cell entity are considered as background cell entities. By analyzing the distribution pattern and category information of these background cells, the role and status of the target cell in the entire microenvironment can be more comprehensively reflected.

[0094] The reference cell entity set can be understood as a subset of cells that have a direct spatial association with the target cell, which is screened out from the background cell entity by setting the density calculation radius. The significance of this set is that it can focus on reflecting the neighborhood characteristics of the target cell within a specific spatial range. For example, in lung cancer pathology images, different density calculation radii (such as 50μm, 100μm) are set with a certain cell as the center, and each cell within the corresponding radius is screened out as a reference cell entity set, and the distribution characteristics of these sets are calculated to generate the density feature information of the target cell. This analysis method effectively narrows the scope of analysis and improves the accuracy of describing the spatial relationship of the target cells.

[0095] It should be noted that the density feature information corresponding to the target cell entity is generated according to the cell category information corresponding to each reference cell entity in each reference cell entity set. It can be understood that by analyzing the distribution characteristics of different categories of cells in the neighborhood of the target cell, statistical information and feature expressions reflecting the microenvironment of the target cell are extracted to further support pathological analysis. The specific method can be to generate the category distribution characteristics of the target cell neighborhood by counting the proportion of the number of different categories of cells in the reference cell entity set, thereby quantifying the contribution of different categories of cells; it is also possible to calculate the category density gradient based on the category information of the reference cells, and analyze the spatial aggregation degree of different categories of cells in the neighborhood of the target cell, such as the distribution changes of cancer cells and immune cells in the neighborhood; it is also possible to construct an interaction matrix between cell categories, evaluate the correlation strength between the target cell and its neighborhood reference cells, and use it as part of the high-dimensional density feature information, etc. This manual does not impose any restrictions on this.

[0096] By using multiple density calculation radii to obtain multiple reference cell entity sets for generating density feature information from the remaining cells in the slice, the distribution patterns of cells at different spatial scales and their neighborhood relationships with target cells can be comprehensively considered when subsequently generating density feature information, thereby more comprehensively reflecting the spatial characteristics and biological significance of target cells in their microenvironment. This method can not only reveal the relationship between cells in local high-density aggregation areas, but also capture the distribution trends of different cell types in a larger range, providing rich data support for analyzing complex tissue structures and disease-related spatial features, thereby improving the accuracy and clinical relevance of pathological image analysis.

[0097] Further, based on each density calculation radius, the target cell position information corresponding to the target cell entity and the background cell position information corresponding to each background cell entity, a reference cell entity set of the target cell entity for each density calculation radius is determined in each background cell entity, including:

[0098] Based on the target cell position information and the background cell position information, obtaining the distance information between the target cell entity and each background cell entity;

[0099] Obtaining a first density calculation radius, wherein the first density calculation radius is any one of the density calculation radii;

[0100] At least one reference cell entity is determined in each background cell entity, wherein the reference cell entity is a background cell entity whose distance information to the target cell entity is less than or equal to the first density calculation radius.

[0101] In practical applications, the distance information between the target cell entity and each background cell entity is the geometric distance data between the target cell and the surrounding background cells in space, which is used to quantify the spatial characteristics of the neighborhood relationship; the reference cell entity is a set of background cells that have spatial correlation with the target cell within the set density calculation radius, which is used to analyze the neighborhood characteristics of the target cell.

[0102] The distance information between the target cell entity and each background cell entity can be understood as key data for evaluating the relative position relationship between the target cell and its surrounding background cells in space. Its significance lies in that through this distance information, it is possible to determine which background cells have a direct impact on the spatial environment of the target cell. For example, in tumor slice analysis, the distance information between the target cancer cells and the surrounding immune cells or normal cells can reveal the infiltration characteristics of tumor cells and the intensity of immune response. This distance information provides basic support for the subsequent generation of reference cell entity sets and density features.

[0103] The reference cell entity can be understood as a collection of cells selected based on the distance between the target cell and the background cell and the density calculation radius. These cells are directly involved in the analysis of the neighborhood characteristics of the target cell. Its significance lies in that by limiting the spatial range (for example, 50μm or 100μm), it is possible to focus on the neighborhood cells that have a greater impact on the spatial environment of the target cell, thereby improving the accuracy and pertinence of the feature analysis. For example, in breast cancer pathology images, the reference cell entities include various cell entities such as cancer cells, vascular endothelial cells, inflammatory cells, and immune cells within the density calculation radius.

[0104] By determining that the cell entities within each density calculation radius are multiple reference cell entity sets, and then analyzing the types and quantities of the cell entities in each reference cell entity set, density feature information of the target cancer cells can be generated to support cancer staging and treatment decisions.

[0105] Further, density feature information corresponding to the target cell entity is generated according to the cell category information corresponding to each reference cell entity in each reference cell entity set, including:

[0106] acquiring at least one cell type information based on each cell type information;

[0107] Generate at least one density feature sub-information corresponding to the target cell entity according to each reference cell entity set and each cell type information;

[0108] The density characteristic information corresponding to the target cell entity is generated according to each density characteristic sub-information.

[0109] In practical applications, cell type information is key data that describes the biological or morphological classification of different cells in the target pathological image, and is usually used to distinguish between tumor cells, inflammatory cells and other categories; density feature sub-information is a statistical feature generated by calculating the distribution characteristics of different types of cells within a radius based on the different densities of the target cell entity, and is mainly used for the refined characterization of the cell spatial environment.

[0110] Exemplarily, cell type information can be understood as the category label assigned to each cell entity in the pathological image, such as cancer cells, immune cells, vascular endothelial cells, etc. This information is of great significance in analyzing the spatial distribution patterns and biological functions between cells. For example, in a tumor slice, clear cell type information can reveal the invasion range of cancer cells and their interaction with immune cells, thereby providing basic data support for tumor microenvironment research. Density feature sub-information can be understood as the features generated by statistics and analysis of the distribution characteristics of each cell type in each reference cell entity set, which is used to refine the spatial relationship description of the target cells.

[0111] It should be noted that generating at least one density feature sub-information corresponding to the target cell entity based on each reference cell entity set and each cell type information can be understood as extracting statistical information or features of a specific dimension by analyzing the number and category distribution characteristics of reference cells in the neighborhood of the target cell, which is used to reflect the refined relationship of the cell microenvironment. The specific method can be to generate the category density gradient characteristics of the target cell by comparing the changes in the number of cells of a specific category within different radius ranges in the reference cell set sequence to reveal the spatial distribution dynamics of the cells of this category in the neighborhood; it is also possible to count the proportion of the number of cells of each category in each reference cell set, and generate each density sub-feature by combining the weighting factor to generate category normalization to quantify the relative importance of different categories of cells in the neighborhood of the target cell; it is also possible to construct a category interaction matrix of the reference cell set, and generate interaction density sub-features by evaluating the relationship strength between the target cell and the cells of different categories in the reference cell set, and this specification does not impose any restrictions on this.

[0112] It should be noted that the density feature information corresponding to the target cell entity generated according to each density feature sub-information can be understood as constructing a comprehensive statistical feature reflecting the relationship between the target cell and its neighboring cells by comprehensively analyzing the density feature sub-information of each dimension, providing a basis for subsequent analysis. The specific method can be to perform weighted averaging on different density feature sub-information, and generate feature information in the form of vectors after splicing multiple radius ranges; it can also be performed by performing dimensionality reduction analysis on each density feature sub-information, such as PCA (Principal Component Analysis), to extract key features to reduce redundant data and improve computational efficiency; it can also construct a multi-level feature matrix, and combine density feature sub-information of different categories and ranges into a multi-dimensional feature vector for input into a deep learning model, etc. This manual does not impose any restrictions on this.

[0113] By using multiple reference cell entity sets and multiple cell type information to generate multiple density feature sub-information, we can fully reflect the neighborhood distribution characteristics of target cells at different spatial scales and cell types, and provide support for the refined description of the cell microenvironment. This method can reveal the dynamic relationship between target cells and specific cell types in the neighborhood, such as density changes and spatial aggregation, while taking into account local and global distribution characteristics, providing a richer and more accurate data basis for subsequent pathological image analysis tasks, and helping to improve the clinical relevance and application value of the analysis.

[0114] Further, at least one density feature sub-information corresponding to the target cell entity is generated according to each reference cell entity set and each cell type information, including:

[0115] Determine first cell type information, a first reference cell entity set, and a second reference cell entity set, wherein the first cell type information is any one of the cell type information, the first reference cell entity set is any one of the reference entity sets, the second reference cell entity set is a reference cell entity set that is a previous reference cell entity set in a reference cell set sequence, and the reference cell set sequence is a sequence generated by sorting the reference entity sets in ascending order according to the density calculation radius corresponding to each reference entity set;

[0116] Acquire target first type cell quantity information corresponding to the first reference cell entity set, and initial first type cell quantity information corresponding to the second reference cell entity set;

[0117] Generate first type cell quantity change information corresponding to the first reference cell entity set according to the target first type cell quantity information and the initial first type cell quantity information, wherein the first type cell quantity change information represents a change in the quantity of first type cell entities in the first reference cell entity set compared with the quantity of first type cell entities in the second reference cell entity set, and the first type cell entity is a cell entity whose cell type information is the first type cell information;

[0118] According to the first cell type quantity change information, density feature sub-information of the first reference cell entity set for the first cell type information is obtained.

[0119] In practical applications, the reference cell set sequence is an ordered structure formed by sorting multiple reference cell entity sets in ascending order of radius based on the rule of increasing radius calculated by density, which is used to analyze the characteristics of cells at different spatial scales; the target first type cell quantity information is the statistical data of the target type cells in a specific reference cell entity set; the initial first type cell quantity information is the statistical data of the target type cells in the previous layer of reference cell entity set in the current reference cell set sequence; the first type cell quantity change information is the quantity change value of the target type cells between two adjacent layers of reference cell sets, which is used to describe the spatial distribution dynamics.

[0120] The reference cell collection sequence can be understood as a cell distribution hierarchy formed by expanding the neighborhood of the target cell layer by layer according to the radius, which is used to systematically study the multi-scale environmental characteristics of the target cell. For example, in tumor microenvironment analysis, the reference cell collection sequence can help identify the cell diffusion pattern and the degree of aggregation of immune cells in the tumor edge area, providing support for cancer staging.

[0121] The target first type of cell quantity information can be understood as the number statistics of a specific cell type in the target reference cell entity set. This type of information can intuitively reflect the distribution characteristics of different cell types in the target neighborhood. The initial first type of cell quantity information can be understood as the number statistics of target type cells in the previous layer of the current set in the reference cell set sequence.

[0122] The first type of cell quantity change information can be understood as the change in the quantity of the target cell type between two adjacent reference cell sets, which characterizes the expansion or contraction trend of the target cell type in the neighborhood. For example, in the study of inflammatory response, the recruitment effect of immune cells near the tumor can be evaluated through quantity change information, providing a basis for the diagnosis of the inflammatory area.

[0123] It should be noted that, according to the information on the change in the number of cells of the first type, obtaining the density feature sub-information of the first reference cell entity set for the first cell type information can be understood as extracting statistical features reflecting the local and global distribution states by combining the dynamic change characteristics of the number in the neighborhood of the target cell, so as to comprehensively describe the neighborhood environment of the target cell. The specific method can be to generate local density feature sub-information by calculating the ratio of the change in the number of local reference first type cells to the number in the current reference set, which is used to evaluate the density change of specific cell types in the neighborhood of the target cell, and to generate global density feature sub-information by normalizing the number and change information of the global reference first type cells, which is used to quantify the relative relationship between the target cell and the global cell distribution; the cumulative change trend analysis method can also be used to fit the quantity change data of the multi-level reference set to generate change trend feature sub-information, which is used to describe the dynamic distribution pattern of specific types of cells in the neighborhood of the target cell, and this manual does not impose any restrictions on this.

[0124] By obtaining the quantitative change information of the target cell neighborhood and extracting the corresponding density sub-feature information, the spatial distribution pattern of the target cell in its microenvironment can be accurately described, thereby helping to distinguish cell characteristics that vary significantly even within the same category. This method can reveal the relationship between the target cell and the specific cell types around it, such as whether the cancer cell is surrounded by a large number of immune cells or the extent to which it aggregates with other cancer cells, thereby providing more detailed and meaningful support for diagnosis, disease staging, and the formulation of potential treatment strategies.

[0125] Further, according to the information on the change in the number of cells of the first type, obtaining density feature sub-information of the first reference cell entity set for the first cell type information includes:

[0126] Acquire local reference first type cell quantity information and global reference first type cell quantity information, wherein the local reference first type cell quantity information represents the quantity information of the first type cell entity in the cell entity whose distance from the target cell entity is less than or equal to the density calculation radius threshold, and the global reference first type cell quantity information is the quantity information of each cell entity that is sorted from largest to smallest in the local reference first type cell quantity information;

[0127] Generate local density feature sub-information of the first reference cell entity set for the first cell type information according to the local reference first cell type number information and the first cell type number change information;

[0128] Generate global density feature sub-information of the first reference cell entity set for the first cell type information according to the global reference first cell type number information and the first cell type number change information;

[0129] Accordingly, density feature information corresponding to the target cell entity is generated according to each density feature sub-information, including:

[0130] Generate local density feature information corresponding to the target cell entity according to each local density feature sub-information, and generate global density feature information corresponding to the target cell entity according to each global density feature sub-information;

[0131] Density feature information corresponding to the target cell entity is generated based on the local density feature information and the global density feature information.

[0132] In practical applications, the local reference first type cell quantity information is a description of the number of first type cells in cell entities whose distance to the target cell entity is less than or equal to the density calculation radius threshold; the global reference first type cell quantity information is the larger value sorted by quantity among the local reference first type cell quantity information corresponding to all cell entities; the local density feature sub-information is feature information characterizing the local density distribution of the first reference cell entity set calculated based on the local reference first type cell quantity information and the first type cell quantity change information; the global density feature sub-information is feature information characterizing the global density distribution of the first reference cell entity set generated based on the global reference first type cell quantity information and the first type cell quantity change information; the local density feature information is a comprehensive representation of each local density feature sub-information; and the global density feature information is a comprehensive representation of each global density feature sub-information.

[0133] The local reference first type of cell quantity information can be understood as statistical data used to describe the closeness of the target cell to similar cells in its neighborhood, reflecting the distribution characteristics of adjacent cells in the target cell microenvironment. For example, in a tumor area, the local reference first type of cell quantity information can help identify which tumor cells are densely surrounded by immune cells, which is of great significance for analyzing the strength of the local immune response. The global reference first type of cell quantity information can be understood as comparing the local reference first type of cell quantity information of all target cells and selecting a larger value to measure the distribution characteristics of the cell category on a global scale. For example, when evaluating a pathological section, the global reference first type of cell quantity information can help determine the higher density of tumor cells in the entire section, thereby providing data support for grading and staging.

[0134] The local density feature sub-information can be understood as characterizing the detailed distribution of the target cell category in a specific area. By analyzing the changes in the local reference first type of cell quantity information, more detailed spatial distribution characteristics can be captured. For example, calculating the local density feature sub-information in neighborhoods of different radii can reveal the microstructural characteristics of tumor cell aggregation and provide a basis for further cancer staging. The global density feature sub-information can be understood as the calculation based on the global reference first type of cell quantity information within the entire slice, which is used to reflect the overall distribution trend of the target cell category. For example, analyzing the global density feature sub-information of tumor cells in the entire slice can help identify whether the tumor is unevenly distributed or diffuse, which is crucial for evaluating the aggressiveness of the tumor.

[0135] Local density feature information can be understood as a combination of local density feature sub-information to form an overall description of the distribution density of target cells in their neighborhood. For example, by integrating local density feature information at multiple radii, the distribution characteristics of target cells at different spatial scales can be comprehensively analyzed, providing support for multi-level modeling of the cell microenvironment. Global density feature information can be understood as a combination of global density feature sub-information to form the density distribution characteristics of the target cell category within the entire slice. For example, integrating global density feature information can reveal the macroscopic pattern of the distribution of a certain type of cell in the entire slice, providing key support for overall analysis at the slice level.

[0136] It should be noted that the density feature information corresponding to the target cell entity generated based on the local density feature information and the global density feature information can be understood as forming a comprehensive feature description by combining the density distribution characteristics of the target cell in the local and global space. The specific method can be to combine the local density feature information and the global density feature information in a weighted average manner to generate density feature information, thereby balancing the local detailed description and the global distribution trend; it can also be achieved by constructing a specific nonlinear model, inputting the local and global feature information into the model to generate high-dimensional comprehensive feature information, and capturing the complex relationship between the two; it can also be achieved by hierarchical aggregation, integrating the local and global feature information layer by layer, to generate density feature information that can reflect the multi-level spatial distribution characteristics. This specification does not impose any restrictions on this.

[0137] In one embodiment provided in this specification, the calculation method of the local density feature sub-information is as shown in Formula 1:

[0138]

[0139] in, The first reference cell entity set r corresponding to the radius j is calculated for the density of the target cell entity i (j) For the local density feature sub-information of cell category t, The first reference cell entity set r corresponding to the radius j is calculated for the density of the target cell entity i (j) The target first type of cell quantity information for the first cell type information t, Calculate the second reference cell entity set r corresponding to radius j-1 for the density of target cell entity i (j-1) For the initial first type of cell quantity information of the first cell type information t, That is, the first reference cell entity set r (j) The corresponding information on the change in the number of cells of the first type, Calculate the radius threshold N for the density of target cell entity i d The corresponding local reference first type cell quantity information for the first cell type information t.

[0140] The calculation method of the global density feature sub-information is shown in Formula 2:

[0141]

[0142] in, The first reference cell entity set r corresponding to the radius j is calculated for the density of the target cell entity i (j) The global density feature sub-information for cell category t, The global reference first type of cell quantity information is the global reference first type of cell quantity information that is sorted from largest to smallest in the local reference first type of cell quantity information corresponding to each cell entity.

[0143] After the local density feature sub-information and global density feature sub-information corresponding to each reference cell entity set and each cell category information are calculated using the above formula, each local density feature sub-information is spliced ​​to generate the local density feature information of the target cell entity i for The local density feature information is spliced ​​to generate the global density feature information of the target cell entity i for Then the above and After fusion, the density feature information F corresponding to the target cell entity is generated cell .

[0144] By integrating feature information from local and global perspectives, not only can the microenvironment of the target cell entity be carefully portrayed, but its distribution characteristics in the overall tissue can also be fully reflected, thereby providing richer and more accurate data support for subsequent analysis.

[0145] Step 106: Generate pathological image feature information corresponding to the target pathological image according to the density feature information corresponding to each cell entity.

[0146] In practical applications, pathological image feature information is a comprehensive expression of the spatial distribution of cells and other characterization information extracted from the target pathological image. It aims to describe the overall tissue structure characteristics and cell interaction patterns of the pathological image to support clinical analysis and model prediction.

[0147] Pathological image feature information can be understood as high-dimensional representation data generated by integrating and calculating the density feature information, positional relationship and category distribution of each cell entity in the target pathological image. These features are usually presented in the form of vectors or tensors to reflect the overall characteristics of the pathological image. For example, in a tumor slice, pathological image feature information may include features such as high cell density in the core area of ​​the tumor, sparse distribution of peripheral inflammatory cells, and linear arrangement of cells in the local vascular area. These feature information not only provides a statistical description at the cellular level, but also includes the spatial hierarchy of the entire tissue.

[0148] Pathological image feature information can comprehensively characterize the spatial distribution patterns and tissue structural characteristics of cells in the target pathological image, providing data support for downstream tasks. For example, by inputting pathological image feature information into a deep learning model, tasks such as cancer staging, tumor invasion prediction, or patient survival assessment can be achieved. In a lung cancer case, the model may use pathological image feature information to analyze the spatial distribution pattern of tumor cells and the infiltration of immune cells to determine the malignancy of the cancer and the patient's response to treatment. In addition, this feature information can also be used to optimize tissue classification algorithms and improve the efficiency and accuracy of pathological image analysis.

[0149] Furthermore, generating pathological image feature information corresponding to the target pathological image according to the density feature information corresponding to each cell entity includes:

[0150] Parsing the cell attribute information corresponding to each cell entity to obtain the cell position information corresponding to each cell entity, and determining each cell entity as a pathological feature extraction cell entity;

[0151] Determine the density feature information corresponding to each cell entity as the cell feature information corresponding to each pathological feature extraction cell entity;

[0152] Extracting cell feature information and cell position information corresponding to the cell entity based on each pathological feature to obtain at least one approximate cell set;

[0153] Extract cell feature information and cell position information corresponding to the cell entity according to each pathological feature in each approximate cell set, and generate aggregated cell entity information corresponding to each approximate cell set;

[0154] When the number of aggregated cell entity information is greater than the image feature extraction threshold, determining each aggregated cell entity information as a pathology feature extraction cell entity, and returning to execute the step of acquiring at least one approximate cell set based on the cell feature information and cell position information corresponding to each pathology feature extraction cell entity;

[0155] When the number of aggregated cell entity information is less than or equal to the image feature extraction threshold, pathological image feature information corresponding to the target pathological image is generated based on each aggregated cell entity information.

[0156] In practical applications, pathological feature extraction cell entities are cell entities determined for extracting pathological image features; approximate cell sets are cell sets representing local areas obtained based on feature information and position information of cell entities extracted from pathological features; and aggregated cell entity information is comprehensive descriptive data generated by integrating the information of each cell entity in the approximate cell set.

[0157] Pathological feature extraction of cell entities can be understood as, based on the cell entities in the target pathological image, by parsing their location information and attribute information, screening out cells that play an important role in the generation of pathological features. By performing screening in cycles for multiple times, the data dimension can be effectively reduced, allowing subsequent analysis to focus on more critical cell features. For example, in a tumor pathological image, the cell entities extracted by pathological features may be cells in the tumor boundary area, which directly reflect the trend of tumor spread.

[0158] Approximate cell collections can be understood as cell collections that are defined by setting spatial neighborhoods or specific similarity criteria based on the characteristic information and spatial location information of cell entities extracted based on pathological features. These collections can represent local features in the cell cloud and provide a data basis for subsequent aggregation steps. For example, in microenvironment research, a set of approximate cell collections can be generated by the distribution of distances and cell types within the neighborhood to further quantify the local microenvironment.

[0159] Aggregate cell entity information can be understood as a numerical expression describing local or global features generated by integrating and comprehensively calculating the characteristic information of cell entities in a collection of approximate cells. This information includes both the local distribution of cell types and the spatial relationship between cells, which helps to extract higher-level pathological image features. For example, aggregate cell entity information can represent the density difference between tumor cells and immune cells in a certain area, thus providing a direct basis for prognosis judgment.

[0160] It should be noted that obtaining at least one approximate cell set based on the cell feature information and cell position information corresponding to the cell entity extracted based on each pathological feature can be understood as extracting the characteristics and position of the cell through comprehensive pathological features, and delineating the neighborhood or similarity area around it, so as to generate a representative cell set. Specifically, the FPS (Farthest Point Sampling) method can be used to generate a preset number of grouping anchor points, and the cells can be clustered based on the anchor points to obtain multiple approximate cell sets; the density clustering method can also be used to delineate the cell neighborhood according to the distance between cells and the similarity threshold to form several approximate cell sets; the Euclidean distance between cells can also be calculated based on the feature vector, and multiple sets of adjacent cells can be formed through threshold segmentation. This specification does not impose any restrictions on this.

[0161] It should be noted that the cell feature information and cell location information corresponding to the cell entity are extracted according to each pathological feature in each approximate cell set, and the aggregated cell entity information corresponding to each approximate cell set is generated. It can be understood that by summarizing the features and spatial distribution within the approximate cell set, aggregated description information that can comprehensively represent the set is generated. The specific method can be based on the position coordinates of all cell entities in the target approximate cell set, calculate the center point or center of mass position, generate aggregated cell position information, and extract its main features by weighted average or principal component analysis of all cell feature information in the target approximate cell set to generate aggregated cell feature information; it can also be combined with aggregated cell position information and aggregated cell feature information, and use aggregation rules (for example, based on weighted distance or feature similarity) to generate complete aggregated cell entity information. This specification does not impose any restrictions on this.

[0162] In one embodiment provided in the present specification, the method for obtaining at least one approximate cell set based on the cell feature information and cell position information corresponding to each pathological feature extraction cell entity is to use the FPS method to generate a preset number of grouping anchor cell entities, then calculate the space-semantic perception distance between each cell and each grouping anchor cell entity, and then use the space-semantic perception distance to perform clustering to generate an approximate cell entity set corresponding to each grouping anchor cell entity. The space-semantic perception distance between cells is calculated as shown in Formula 3:

[0163]

[0164] in, is the spatial-semantic perception distance between the cell entity extracted by the i-th pathological feature and the cell entity of the grouping anchor point; exp() is an exponential function, which performs exponential operation on the negative value of the distance. When the distance between the two cells is close, exp(-||c (i) -c ref ||) is close to 1, and the greater the distance, the closer the value is to 0, so that the influence of distance on similarity has the characteristics of smooth and rapid decay; ||c (i) -c ref || Extract the cell location information c of the cell entity for the i-th pathological feature (i) The cell position information c of the grouped anchor cell entity ref The Euclidean distance norm of ; Extract cell feature information of cell entity for the i-th pathological feature Corresponding to the grouped anchor cell entity f ref The inner product (dot product) of the cell feature information can be used to measure the similarity between the two in the feature space. The larger the value, the higher the similarity in the features. dimThe dimension size of the cell feature information is that dividing the dot product result by the number of dimensions helps to normalize the feature similarity and avoid the deviation in the feature similarity measurement caused by the different number of dimensions.

[0165] like Figure 2 As shown, Figure 2 A schematic diagram of a pathological image analysis process provided for an embodiment of the present specification, wherein firstly, the pathological image information and the attribute information of each cell entity are used to obtain the density feature information corresponding to each cell entity, the cell entity marked as black in the pathological image information is a cell entity of the first cell category, and its attribute information includes the position information (x, y) and the category information 0, the gray cell entity is a cell entity of the second cell category, and its attribute information includes the position information (x, y) and the category information 1, and the white cell entity is a cell entity of the third cell category, and its attribute information includes the position information (x, y) and the category information 2, then clustering is performed according to the cell attribute information and density feature information corresponding to each cell entity to obtain multiple approximate cell sets of the first layer, and then the aggregated cell entity information corresponding to each approximate cell set is extracted. Then, the aggregated cell entity information of each aggregated cell entity is clustered again to obtain multiple approximate cell sets of the second layer, and the aggregated cell entity information corresponding to each approximate cell set is extracted, and so on, until it is clustered into an approximate cell set, the aggregated cell entity information corresponding to the approximate cell set is extracted, and the pathological image feature information is determined according to the aggregated cell entity information.

[0166] By repeatedly fusing and screening the pathological feature extraction cell entities used to extract pathological image features, the data dimension can be effectively reduced, allowing subsequent analysis to focus on more critical features in the pathological image.

[0167] Furthermore, cell feature information and cell location information corresponding to the cell entity are extracted according to each pathological feature in each approximate cell set, and aggregated cell entity information corresponding to each approximate cell set is generated, including:

[0168] Determine a target approximate cell set, wherein the target approximate cell set is any one of the approximate cell sets;

[0169] Extracting cell position information corresponding to cell entities based on each pathological feature in the target approximate cell set, and generating aggregated cell position information corresponding to the target approximate cell set;

[0170] Extracting cell feature information corresponding to cell entities according to each pathological feature in the target approximate cell set, and generating aggregated cell feature information corresponding to the target approximate cell set;

[0171] According to the aggregated cell position information and the aggregated cell characteristic information, aggregated cell entity information corresponding to the target approximate cell set is obtained.

[0172] In practical applications, the aggregated cell position information is the comprehensive coordinates used to characterize the spatial distribution of all cells in the approximate cell collection; the aggregated cell feature information is the comprehensive information used to describe the overall characteristics of cells in the approximate cell collection, which may include cell type distribution, density characteristics, etc.

[0173] Aggregate cell position information can be understood as calculating the spatial position representing the entire collection based on the positions of all cells in the approximate cell collection. For example, the coordinate center point or centroid of all cells is calculated, which can reflect the geometric center of cell distribution; the coordinates can also be adjusted based on the cell density weight to generate a density-weighted centroid, thereby highlighting the contribution of dense areas to the overall spatial distribution; the spatial coverage of the collection can also be characterized by the boundary information of the aggregation range, such as the center point of a smaller bounding box. This information can provide the spatial characteristics of the collection for further analysis and help identify distribution patterns or spatial anomalies.

[0174] Aggregating cell feature information can be understood as extracting a comprehensive description that reflects the overall characteristics of the collection by integrating the features of all cells in the approximate cell collection. For example, the simple average of cell features can be calculated to generate the overall feature information of the collection, which is suitable for analyzing the average state of the collection; the main change direction of the collection features can also be extracted through principal component analysis to emphasize the distribution differences of key features in the collection; or based on weighted aggregation methods, such as assigning different weights to different features, to highlight the impact of certain specific features on the collection.

[0175] By aggregating the individual cell entities in the approximate cell set, it can not only be used for feature comparison between sets, but also serve as input for subsequent classification or grouping, providing efficient feature expression for downstream tasks.

[0176] Further, extracting cell feature information corresponding to a cell entity according to each pathological feature in the target approximate cell set to generate aggregated cell feature information corresponding to the target approximate cell set includes:

[0177] Determine a first pathological feature extraction cell entity and at least one second pathological feature extraction cell entity in the target approximate cell set, wherein the first pathological feature extraction cell entity is any one of the pathological feature extraction cell entities in the target approximate cell set, and the second pathological feature extraction cell entity is a cell entity other than the first pathological feature extraction cell entity in the target approximate cell set;

[0178] Based on the cell feature information corresponding to the first pathological feature extraction cell entity and the cell feature information corresponding to each second pathological feature extraction cell entity, generate first verification attention feature information corresponding to the first pathological feature extraction cell entity and second verification attention feature information and second extended attention feature information corresponding to each second pathological feature extraction cell entity;

[0179] Generate extended cell feature information corresponding to the first pathological feature extraction cell entity according to the first verification attention feature information, the second verification attention feature information corresponding to each second pathological feature extraction cell entity, and the second extended attention feature information;

[0180] The extended cell feature information corresponding to the cell entity is extracted according to each pathological feature in the target approximate cell set, and the aggregated cell feature information corresponding to the target approximate cell set is generated.

[0181] In practical applications, the first verification attention feature information is the feature used to describe the interaction weight of the target pathological feature extraction cell entity with other cell entities in the attention mechanism; the second verification attention feature information is the corresponding feature generated by other pathological feature extraction cell entities in the attention mechanism of the target pathological feature extraction cell entity; the second extended attention feature information is the comprehensive feature further expanded and calculated based on the attention features of other cell entities, which is used to describe the overall interaction of the set; the extended cell feature information is the feature generated by combining the target cell entity and the interaction results of the remaining cells, which is used for high-level feature expression of cell entities.

[0182] The first verification of attention feature information can be understood as extracting the feature interactions between the cell entity and other cell entities by analyzing the target pathological features, and generating features that can quantify the importance of the target cell to other cells. For example, this feature can reflect its sensitivity to the surrounding environment by calculating the specific feature matching between the target cell and other cells; it can also calculate the weight based on the geometric distance and feature similarity to represent the spatial correlation between the target cell and other cells; it can also be generated by dynamically adjusting the weight to better adapt to cell interactions under different distributions.

[0183] The second verification of attention feature information can be understood as how other pathological features extract cell entities and affect the target cells, generating features that reflect the importance of these cells to the target cells. For example, features can be generated based on the feature differences between adjacent cells and target cells to quantify local cell differences; features can also be generated by focusing on the weight changes of certain specific types of cells to analyze the specific contributions of important cell categories; and the comprehensive impact of the environment on target cells can also be modeled through high-dimensional feature matching calculations.

[0184] The second extended attention feature information can be understood as combining the attention features of multiple cells to generate comprehensive features that represent higher-level interactions. For example, all relevant attention features can be fused through weighted summation to generate global interaction features; the attention features of specific types of cells can be normalized to highlight the influence of key cell categories; and more important feature interaction patterns can be extracted through feature dimensionality reduction technology to reduce redundant information.

[0185] Expanding cell feature information can be understood as generating high-dimensional features that characterize the role of cells in the current microenvironment by integrating the attention interaction information between the target cell and its surrounding environment. For example, the target cell features can be combined with the attention-weighted features of other cells to generate overall interaction features; spatiotemporal information can be introduced to generate temporal features that can describe the role of cells in a dynamic environment; and aggregate feature extraction methods can be used to generate comprehensive features containing multi-scale information.

[0186] By extracting the first pathological feature to extract the cell entity and the attention features of each second pathological feature to extract the cell entity through the attention mechanism, and realizing the fusion of cell entity features based on the above-mentioned attention features, the complex interaction between cells can be effectively captured, thereby more comprehensively characterizing the function and role of cells in a specific microenvironment. This method can enhance the expressive power of features, so that it is not only limited to the information of a single cell, but also includes the dynamic relationship between cells and the surrounding environment, thereby supporting more accurate pathological feature extraction and functional analysis, and providing a reliable data basis for subsequent pathological image feature generation and disease diagnosis.

[0187] Further, based on the cell feature information corresponding to the first pathological feature extraction cell entity and the cell feature information corresponding to each second pathological feature extraction cell entity, first verification attention feature information corresponding to the first pathological feature extraction cell entity and second verification attention feature information and second extended attention feature information corresponding to each second pathological feature extraction cell entity are generated, including:

[0188] Inputting the cell feature information corresponding to the first pathological feature extraction cell entity into the attention feature extraction network to obtain the first verification attention feature information corresponding to the first pathological feature extraction cell entity;

[0189] The cell feature information corresponding to each pathological feature extraction cell entity is input into the attention feature extraction network, and the second verification attention feature information and the second extended attention feature information corresponding to each pathological feature extraction cell entity are input into the attention feature extraction network.

[0190] In practical applications, the attention feature extraction network is a deep learning model that is used to extract key features from the feature information of cell entities through the attention mechanism. It can generate accurate verification attention features and extended attention features based on the input feature information to describe the correlation between cells and their role in a specific spatial environment.

[0191] Exemplarily, the attention feature extraction network can be understood as a neural network based on the attention mechanism, whose purpose is to extract information that is more critical to the target task by assigning different weights to the input features. For example, for the first pathological feature extraction cell entity, the network can generate verification attention feature information based on its input cell feature information to measure the correlation between the individual characteristics of the cell and the surrounding environment. At the same time, for the second pathological feature extraction cell entity, the network can generate extended attention feature information by comparing these cell features to reflect the interaction pattern between cells. Through these feature extractions, the network can capture the complex relationships between cells while also focusing on local information that is more closely related to the pathological features.

[0192] The feature data of cells is obtained through the attention mechanism, that is, more important parts are given higher weights in the input information, and irrelevant or noisy data is ignored. This mechanism enables the network to efficiently process large-scale cell feature data and accurately capture microscopic details in pathological feature extraction. For example, in cancer research, the network can be used to extract key features from the spatial distribution of tumor cells to help identify early cancer or predict patient prognosis. In summary, the attention feature extraction network provides a powerful tool for high-dimensional pathological data analysis and feature extraction, and has broad application potential.

[0193] Further, generating extended cell feature information corresponding to the first pathological feature extraction cell entity according to the first verification attention feature information, the second verification attention feature information corresponding to each second pathological feature extraction cell entity, and the second extended attention feature information, including:

[0194] Determine feature extension distance information corresponding to the second pathological feature extraction cell entity according to the cell position information corresponding to the first pathological feature extraction cell entity and the cell position information corresponding to each second pathological feature extraction cell entity, wherein the feature extension distance information is the distance information between the cell position information of each second pathological feature extraction cell entity and the cell position information of the first pathological feature extraction cell entity;

[0195] Generate feature extension weight information corresponding to each second pathological feature extraction cell entity according to the first verification attention feature information, the second verification attention feature information corresponding to each second pathological feature extraction cell entity, and the feature extension distance information;

[0196] Generate extended feature information corresponding to the first pathological feature extraction cell entity based on feature extension weight information, second extended attention feature information and feature extension distance information corresponding to each second pathological feature extraction cell entity;

[0197] The extended cell feature information corresponding to the first pathological feature extraction cell entity is generated according to the extended feature information.

[0198] In practical applications, the feature extension distance information is numerical information used to describe the spatial distance between the first pathological feature extraction cell entity and each second pathological feature extraction cell entity; the feature extension weight information is generated based on the first verification attention feature information, the second verification attention feature information and the feature extension distance information, and is weight information used to measure the degree of influence of the second pathological feature extraction cell entity on the feature extension process.

[0199] Feature extension distance information can be understood as a value calculated based on the spatial position information of cells, which aims to quantitatively characterize the physical distance between cells. For example, by measuring the Euclidean distance between the spatial coordinates of a cell entity extracted from a first pathological feature and the coordinates of a cell entity extracted from a second pathological feature, its feature extension distance information can be obtained. The importance of this information in feature extension lies in that it can provide a basis for the spatial relationship between cells, making the subsequent feature aggregation closer to the actual spatial distribution pattern.

[0200] The feature expansion weight information can be understood as a normalized weight generated based on multiple input features (such as the first verification attention feature information, the second verification attention feature information, and the feature expansion distance information), which is used to adjust the influence of different cells in the feature expansion process. For example, a cell entity with a closer distance can be given a larger weight, while a cell entity with a farther distance can be given a lower weight.

[0201] In one embodiment provided in this specification, the generation method of the extended cell feature information corresponding to the first pathological feature extraction cell entity is as shown in Formula 4:

[0202]

[0203] d (i，j) =c (i) -c (j) ,...Formula 4

[0204] Among them, d (i，j) is the feature extension distance information between the first pathological feature extraction cell entity i and the second pathological feature extraction cell entity j, c (i) Extract the cell location information of cell entity i for the first pathological feature, c (j)The cell location information of the cell entity j is extracted for the second pathological feature.

[0205] is the feature expansion weight vector information between the first pathological feature extraction cell entity i and the second pathological feature extraction cell entity j, M (j) is the calculation result of whether the space-semantic perception distance between the first pathological feature extraction cell entity i and the second pathological feature extraction cell entity j is greater than the threshold. If it is greater than the threshold, the value is determined to be 0; if it is less than the threshold, the value is determined to be 1. att () is an attention feature extraction network. In this embodiment, MLP (Multilayer Perceptron) is used as the attention feature extraction network. is a linear projection matrix for obtaining the first verification attention feature information (mapping the density feature information corresponding to the first pathological feature extraction cell entity i to the query space in the self-attention mechanism), is a linear projection matrix for obtaining the second verification attention feature information (mapping the density feature information corresponding to the second pathological feature extraction cell entity j to the key space in the self-attention mechanism), Extract the density feature information corresponding to cell entity i for the first pathological feature, The density feature information corresponding to the cell entity j is extracted for the second pathological feature.

[0206] Extract the extended cell feature information corresponding to cell entity i for the first pathological feature, The feature expansion weight vector information between the first pathological feature extraction cell entity i and each second pathological feature extraction cell entity j is The feature expansion weight information obtained after linear regression and normalization processing, represents a weighted summation operation on each second pathological feature extraction cell entity j in the reference cell entity set where the first pathological feature extraction cell entity i is located, It is a linear projection matrix used to obtain the second extended attention feature information (mapping the density feature information corresponding to the second pathological feature extracted cell entity j to the numerical space in the self-attention mechanism).

[0207] By calculating the feature extension weights of each second pathological feature extraction cell entity for the first pathological feature extraction cell entity, dynamic adjustment is achieved when the first pathological feature extraction cell entity is expanded, making the feature expansion process more accurate and effectively highlighting the cell information with a high correlation with the target pathological feature extraction cell entity, thereby improving the accuracy and applicability of the expanded cell feature information.

[0208] Step 108: Generate an image analysis result corresponding to the target pathological image based on the pathological image feature information.

[0209] In practical applications, image analysis results are high-level data expressions generated based on pathological image feature information, which are usually used to support disease diagnosis, treatment effect evaluation, and other clinical decisions. Image analysis results can be understood as medical conclusions or model outputs obtained through in-depth mining and analysis of pathological image feature information. These results can be quantitative indicators, such as cancer staging probability, survival prediction value, or qualitative judgments, such as lesion area classification or treatment response quality. For example, in breast cancer cases, image analysis results may include tumor invasiveness scores (high, medium, and low), predictions of whether there is distant metastasis, and the degree of immune cell infiltration. Through these results, doctors can more intuitively understand the patient's condition and develop appropriate treatment plans.

[0210] Image analysis results can significantly improve the efficiency and accuracy of clinical diagnosis and research. Taking lung cancer as an example, the results of pathological image analysis may reveal the distribution characteristics of immune cells in the tumor microenvironment, thereby helping to determine whether the patient is suitable for immunotherapy. In addition, these results can also be integrated into the hospital information system to generate standardized diagnostic reports to support doctors in making decisions quickly. Image analysis results have important application value in personalized medicine, treatment monitoring and prognosis evaluation.

[0211] The scheme of the embodiments of the present specification is applied, by acquiring the pathological image information corresponding to the target pathological image, extracting the position information, category information and cell attribute information of each cell entity generated thereby, and calculating the density feature information of each cell entity based on this information, on the basis of retaining the spatial distribution pattern of the cells, the target cell entity is associated with the background cell entity with the density calculation radius as a constraint to form a reference cell entity set, thereby accurately characterizing the spatial distribution and category characteristics of the cells in the local area; further generating local density feature information and global density feature information of the target cell entity through local and global reference methods, realizing multi-scale feature extraction of the target pathological image; combining these cell entity features to generate the analysis results of the target pathological image, which can fully support the analysis and interpretation of pathological images from both spatial and semantic dimensions, and provide reliable data support for disease diagnosis, classification and prognosis evaluation in clinical applications.

[0212] Corresponding to the above method embodiment, this specification also provides a computer-aided diagnosis method embodiment of a tumor, see Figure 3 , Figure 3 A flowchart of a computer-aided diagnosis method for a tumor provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0213] Step 302: Acquire pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity.

[0214] Step 304: Obtain density feature information corresponding to each cell entity based on the cell attribute information corresponding to each cell entity.

[0215] Step 306: Generate pathological image feature information corresponding to the target pathological image according to the density feature information corresponding to each cell entity.

[0216] Step 308: Generate an image recognition result corresponding to the target pathological image based on the pathological image feature information.

[0217] In practical applications, image recognition results are preliminary diagnosis-related data outputs generated based on pathological image feature information in computer-aided diagnosis of tumors. They are mainly used to determine whether a tumor exists in a pathological image and its possible category.

[0218] The image recognition results can be understood as comprehensive results of tumor presence and category prediction generated by high-level feature extraction and pattern analysis of pathological images through tumor computer-aided diagnosis solutions. For example, in the auxiliary diagnosis of lung cancer, the image recognition results may output the judgment of "suspected tumor" or "no tumor found", and provide the possible tumor type (such as adenocarcinoma or squamous cell carcinoma). These results can serve as an important basis in the diagnosis process and provide a reference for doctors to further evaluate.

[0219] The above is a schematic scheme of a computer-aided diagnosis method for tumors in this embodiment. It should be noted that the technical scheme of the computer-aided diagnosis method for tumors and the technical scheme of the above-mentioned pathological image analysis method belong to the same concept, and the details of the technical scheme of the computer-aided diagnosis method for tumors that are not described in detail can be referred to the description of the technical scheme of the above-mentioned pathological image analysis method.

[0220] By applying the scheme of the embodiments of this specification, the pathological image information of the target pathological image is obtained, including the cell attribute information of each cell entity, and the density feature information of each cell entity is extracted based on this information, and the pathological image feature information of the target pathological image is further generated and the image recognition result is generated accordingly, so that the accurate analysis of the pathological image can be achieved. In particular, in the analysis process, by combining the cell-centered attribute information with the density feature, the distribution characteristics and tissue relationship of the cells in the target pathological image can be effectively revealed, thereby improving the accuracy and comprehensiveness of image recognition and providing more clinically valuable support for disease auxiliary diagnosis.

[0221] Corresponding to the above method embodiment, this specification also provides a pathological image analysis method embodiment applied to a cloud device, see Figure 4 , Figure 4 A flow chart of a pathological image analysis method applied to a cloud device provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0222] Step 402: receiving a pathological image analysis request sent by a terminal-side device, wherein the pathological image analysis request includes a target pathological image.

[0223] Step 404: Acquire pathological image information corresponding to the target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity.

[0224] Step 406: Obtain density feature information corresponding to each cell entity based on the cell attribute information corresponding to each cell entity.

[0225] Step 408: Generate pathological image feature information corresponding to the target pathological image according to the density feature information corresponding to each cell entity.

[0226] Step 410: Generate an image analysis result corresponding to the target pathological image based on the pathological image feature information, and send the image analysis result to the terminal side device.

[0227] The above is a schematic scheme of a pathological image analysis method applied to a cloud device in this embodiment. It should be noted that the technical scheme of the pathological image analysis method applied to the cloud device belongs to the same concept as the technical scheme of the pathological image analysis method described above. For details not described in detail in the technical scheme of the pathological image analysis method applied to the cloud device, please refer to the description of the technical scheme of the pathological image analysis method described above.

[0228] By applying the solution of the embodiments of this specification, by executing the pathological image analysis process on the cloud device, it is possible to utilize the target pathological image information sent by the end-side device, combined with the powerful computing power of the cloud, efficiently extract density features based on the cell attribute information of each cell entity, and generate pathological image feature information, thereby further generating image analysis results and returning them to the end-side device. This method not only optimizes the computing resources of the pathological image analysis process, but also effectively improves the accuracy and efficiency of the analysis, supports the efficient implementation of multi-end collaboration and remote pathological diagnosis, and meets the needs of complex pathological image analysis tasks for large-scale data processing capabilities.

[0229] Corresponding to the above method embodiment, this specification also provides a computer-aided diagnosis method embodiment of a tumor applied to a client, see Figure 5 , Figure 5 A flowchart of a computer-aided diagnosis method for tumors applied to a client provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0230] Step 502: Acquire pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity.

[0231] Step 504: based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained.

[0232] Step 506: Generate pathological image feature information corresponding to the target pathological image according to the density feature information corresponding to each cell entity.

[0233] Step 508: Generate an image recognition result corresponding to the target pathological image based on the pathological image feature information.

[0234] The above is a schematic scheme of a computer-aided diagnosis method for tumors applied to a client in this embodiment. It should be noted that the technical scheme of the computer-aided diagnosis method for tumors applied to the client and the technical scheme of the computer-aided diagnosis method for tumors described above belong to the same concept, and the details of the technical scheme of the computer-aided diagnosis method for tumors applied to the client that are not described in detail can all be referred to the description of the technical scheme of the computer-aided diagnosis method for tumors described above.

[0235] The solution of the embodiment of this specification is applied. By executing the process of extracting density features from cell attribute information on the client side, aggregating this information into pathological image features and forming image recognition results, it is not only conducive to efficient and accurate auxiliary diagnosis of tumor pathological images, but also better reveals the spatial distribution characteristics at the cellular level and its intrinsic relationship with the surrounding tissue structure. In this way, in-depth analysis results of tumor lesion characteristics can be quickly obtained in clinical scenarios without cumbersome external data transmission and waiting, which brings significant clinical value and application prospects for real-time decision-making and the formulation of personalized treatment plans.

[0236] Corresponding to the above method embodiment, this specification also provides a computer-aided diagnosis method embodiment of a tumor applied to a cloud device, see Figure 6 , Figure 6 A flowchart of a computer-aided diagnosis method for tumors applied to a cloud device according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0237] Step 602: receiving a tumor diagnosis request sent by a terminal device, wherein the tumor diagnosis request carries a target pathological image.

[0238] Step 604: Acquire pathological image information corresponding to the target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity.

[0239] Step 606: Obtain density feature information corresponding to each cell entity based on the cell attribute information corresponding to each cell entity.

[0240] Step 608: Generate pathological image feature information corresponding to the target pathological image according to the density feature information corresponding to each cell entity.

[0241] Step 610: Generate an image recognition result corresponding to the target pathological image based on the pathological image feature information, and send the image recognition result to the terminal side device.

[0242] The above is a schematic scheme of a computer-aided diagnosis method for tumors applied to a cloud device in this embodiment. It should be noted that the technical scheme of the computer-aided diagnosis method for tumors applied to a cloud device and the technical scheme of the computer-aided diagnosis method for tumors described above belong to the same concept, and the details of the technical scheme of the computer-aided diagnosis method for tumors applied to a cloud device that are not described in detail can all be referred to the description of the technical scheme of the computer-aided diagnosis method for tumors described above.

[0243] By applying the solution of the embodiments of this specification, the target pathological image is graded and analyzed in the cloud device, and the cell attribute information and density feature information of each cell entity are extracted and integrated, which can give full play to the flexibility and high performance of cloud computing resources and network connections. This feature extraction and result generation in the cloud helps to achieve rapid processing and remote diagnosis support for large-scale pathological image data, improves the efficiency and reliability of tumor diagnosis through accurate quantification and identification of tissue structure and cell distribution, and provides richer and more practical references for clinical decision-making and subsequent treatment strategy formulation.

[0244] See also Figure 7 , Figure 7 The architecture diagram of a computer-aided diagnosis system for tumors provided in one embodiment of the present specification is shown. The computer-aided diagnosis system for tumors may include a client 100 and a server 200;

[0245] The client 100 is used to send a tumor diagnosis request to the server 200, wherein the tumor diagnosis request carries a target pathology image, wherein the target pathology image includes at least one cell entity;

[0246] The server 200 is used to obtain pathological image information corresponding to the target pathological image, wherein the pathological image information includes cell attribute information corresponding to each cell entity; based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained; based on the density feature information corresponding to each cell entity, pathological image feature information corresponding to the target pathological image is generated; based on the pathological image feature information, an image recognition result corresponding to the target pathological image is generated; and the image recognition result is sent to the client 100;

[0247] The client 100 is also used to receive the image recognition result sent by the server 200.

[0248] By applying the solution of the embodiments of this specification, by providing a computer-aided diagnosis system architecture for tumors, efficient processing of target pathological images from request initiation, image information acquisition, feature extraction to result generation can be achieved. By sending a target pathological image diagnosis request, the client enables the server to extract the attribute information of the cell entity in the target pathological image, and generate density feature information based on these attribute information, thereby forming a complete pathological image feature. On this basis, the server generates an image recognition result of the target pathological image and returns it to the client, realizing rapid interaction of image analysis results. Through the collaborative architecture of the server and the client, the efficiency and accuracy of pathological image analysis are significantly improved, providing comprehensive support for computer-aided diagnosis of tumors.

[0249] The computer-aided diagnosis system for tumors may include multiple clients 100 and a server 200, wherein the client 100 may be referred to as a terminal-side device and the server 200 may be referred to as a cloud-side device. Multiple clients 100 may establish a communication connection through the server 200. In the computer-aided diagnosis scenario for tumors, the server 200 is used to provide computer-aided diagnosis services for tumors between multiple clients 100. Multiple clients 100 may serve as a sending end or a receiving end, respectively, and realize communication through the server 200.

[0250] The user can interact with the server 200 through the client 100 to receive data sent by other clients 100, or send data to other clients 100, etc. In the computer-aided diagnosis scenario of tumors, the user can publish a data stream to the server 200 through the client 100, and the server 200 generates an image recognition result based on the data stream and pushes the image recognition result to other clients that have established communication.

[0251] The client 100 and the server 200 are connected via a network. The network provides a medium for a communication link between the client 100 and the server 200. The network may include various connection types, such as wired or wireless communication links or optical fiber cables, etc. The data transmitted by the client 100 may need to be encoded, transcoded, compressed, etc. before being released to the server 200.

[0252] The client 100 can be a browser, an APP (Application), or a web application such as an H5 (HyperText Markup Language 5, Hypertext Markup Language Version 5) application, or a light application (also known as a mini-program, a lightweight application) or a cloud application, etc. The client 100 can be based on the software development kit (SDK, Software Development Kit) of the corresponding service provided by the server 200, such as based on the real-time communication (RTC, Real Time Communication) SDK development and acquisition. The client 100 can be deployed in an electronic device, and needs to rely on the device to run or some APPs in the device to run. For example, the electronic device can have a display screen and support information browsing, such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications can also be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0253] The server 200 may include servers that provide various services, such as servers that provide communication services to multiple clients, servers for background training that provide support for models used on clients, and servers that process data sent by clients. It should be noted that the server 200 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server of a distributed system, or a server combined with a blockchain. The server can also be a cloud server for basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content Delivery Network), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0254] It is worth noting that the computer-aided diagnosis method for tumors provided in the embodiments of this specification is generally executed by the server, but in other embodiments of this specification, the client may also have similar functions to the server, thereby executing the computer-aided diagnosis method for tumors provided in the embodiments of this specification. In other embodiments, the computer-aided diagnosis method for tumors provided in the embodiments of this specification may also be jointly executed by the client and the server.

[0255] The following combination Figure 8 , taking the application of the pathological image analysis method provided in this specification in computer-assisted prognosis prediction as an example, the pathological image analysis method is further described. Among them, Figure 8 A processing flow chart of a computer-assisted prognostic analysis method provided in one embodiment of the present specification is shown, which specifically includes the following steps.

[0256] Step 802: Acquire a pathological image that needs to be analyzed for prognosis, and acquire pathological image information corresponding to the pathological image, including attribute information corresponding to a plurality of cell entities.

[0257] Step 804: based on the category information and position information corresponding to each cell entity and the set density calculation radius threshold, the global density information and local density information corresponding to each cell entity are obtained.

[0258] Step 806: Generate density feature information of each cell entity for pathological image information according to the global density information and local density information corresponding to each cell entity.

[0259] Step 808: Based on the density feature information and position information corresponding to each cell entity, cluster each cell entity and extract feature information of the clustered cell set.

[0260] Step 810: Aggregate the feature information of each cell set and extract the feature information of the clustered cell set, and repeat this process until a single piece of pathological image feature information for the pathological image information is obtained.

[0261] Step 812: Input the pathological image feature information into a decoding model that generates a prognostic analysis based on the pathological image feature information to obtain a prognostic analysis result of the pathological image.

[0262] The scheme of the embodiments of the present specification is applied to obtain pathological image information that requires prognostic analysis, including attribute information corresponding to multiple cell entities, and obtain global density information and local density information of each cell entity based on category information, location information and density calculation radius threshold, thereby generating density feature information for the pathological image information, and then further aggregate and obtain pathological image feature information for the pathological image information through clustering and feature extraction of each cell entity, and generate prognostic analysis results in combination with the decoding model. This method can realize hierarchical analysis from the cellular level to the global pathological image features, significantly improve the accuracy and practicality of pathological images in prognosis prediction, and help support more detailed disease mechanism research and the formulation of personalized diagnosis and treatment plans.

[0263] Corresponding to the above method embodiment, this specification also provides a pathological image analysis device embodiment, Fig. 9 FIG. 2 shows a schematic diagram of the structure of a pathological image analysis device provided by an embodiment of the present specification. Fig. 9 As shown, the device comprises:

[0264] The information acquisition module 902 is configured to acquire pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity;

[0265] A feature acquisition module 904 is configured to acquire density feature information corresponding to each cell entity based on cell attribute information corresponding to each cell entity;

[0266] The feature generation module 906 is configured to generate pathological image feature information corresponding to the target pathological image according to the density feature information corresponding to each cell entity;

[0267] The analysis module 908 is configured to generate an image analysis result corresponding to the target pathological image based on the pathological image feature information.

[0268] Optionally, the information acquisition module 902 is further configured to:

[0269] Acquire target pathological images;

[0270] Inputting the target pathological image into a cell annotation model to obtain cell position information and cell category information corresponding to each cell entity generated by the cell annotation model;

[0271] Based on the cell position information and cell category information corresponding to each cell entity, the cell attribute information corresponding to each cell entity is generated.

[0272] Optionally, the feature acquisition module 904 is further configured to:

[0273] Parsing the cell attribute information corresponding to each cell entity, and obtaining the cell location information and cell category information corresponding to the cell attribute information corresponding to each cell entity;

[0274] Based on the cell position information and cell category information corresponding to each cell entity, density feature information corresponding to each cell entity is generated.

[0275] Optionally, the feature acquisition module 904 is further configured to:

[0276] generating a density calculation radius threshold according to the cell position information corresponding to each cell entity, and determining at least one density calculation radius based on the density calculation radius threshold;

[0277] Based on the density calculation radius and the cell position information and cell category information corresponding to each cell entity, density feature information corresponding to each cell entity is generated.

[0278] Optionally, the feature acquisition module 904 is further configured to:

[0279] Determine a target cell entity and at least one background cell entity, wherein the target cell entity is any one of the cell entities, and the background cell entity is any one of the cell entities except the target cell entity;

[0280] Based on each density calculation radius, the target cell position information corresponding to the target cell entity and the background cell position information corresponding to each background cell entity, determining a reference cell entity set of the target cell entity for each density calculation radius in each background cell entity, wherein the reference cell entity set includes at least one reference cell entity;

[0281] According to the cell category information corresponding to each reference cell entity in each reference cell entity set, the density feature information corresponding to the target cell entity is generated.

[0282] Optionally, the feature acquisition module 904 is further configured to:

[0283] Based on the target cell position information and the background cell position information, obtaining the distance information between the target cell entity and each background cell entity;

[0284] Obtaining a first density calculation radius, wherein the first density calculation radius is any one of the density calculation radii;

[0285] At least one reference cell entity is determined in each background cell entity, wherein the reference cell entity is a background cell entity whose distance information to the target cell entity is less than or equal to the first density calculation radius.

[0286] Optionally, the feature acquisition module 904 is further configured to:

[0287] acquiring at least one cell type information based on each cell type information;

[0288] Generate at least one density feature sub-information corresponding to the target cell entity according to each reference cell entity set and each cell type information;

[0289] The density characteristic information corresponding to the target cell entity is generated according to each density characteristic sub-information.

[0290] Optionally, the feature acquisition module 904 is further configured to:

[0291] Determine first cell type information, a first reference cell entity set, and a second reference cell entity set, wherein the first cell type information is any one of the cell type information, the first reference cell entity set is any one of the reference entity sets, the second reference cell entity set is a reference cell entity set that is a previous reference cell entity set in a reference cell set sequence, and the reference cell set sequence is a sequence generated by sorting the reference entity sets in ascending order according to the density calculation radius corresponding to each reference entity set;

[0292] Acquire target first type cell quantity information corresponding to the first reference cell entity set, and initial first type cell quantity information corresponding to the second reference cell entity set;

[0293] Generate first type cell quantity change information corresponding to the first reference cell entity set according to the target first type cell quantity information and the initial first type cell quantity information, wherein the first type cell quantity change information represents a change in the quantity of first type cell entities in the first reference cell entity set compared with the quantity of first type cell entities in the second reference cell entity set, and the first type cell entity is a cell entity whose cell type information is the first type cell information;

[0294] According to the first cell type quantity change information, density feature sub-information of the first reference cell entity set for the first cell type information is obtained.

[0295] Optionally, the feature acquisition module 904 is further configured to:

[0296] Acquire local reference first type cell quantity information and global reference first type cell quantity information, wherein the local reference first type cell quantity information represents the quantity information of the first type cell entity in the cell entity whose distance from the target cell entity is less than or equal to the density calculation radius threshold, and the global reference first type cell quantity information is the quantity information of each cell entity that is sorted from largest to smallest in the local reference first type cell quantity information;

[0297] Generate local density feature sub-information of the first reference cell entity set for the first cell type information according to the local reference first cell type number information and the first cell type number change information;

[0298] Generate global density feature sub-information of the first reference cell entity set for the first cell type information according to the global reference first cell type number information and the first cell type number change information;

[0299] Accordingly, density feature information corresponding to the target cell entity is generated according to each density feature sub-information, including:

[0300] Generate local density feature information corresponding to the target cell entity according to each local density feature sub-information, and generate global density feature information corresponding to the target cell entity according to each global density feature sub-information;

[0301] Density feature information corresponding to the target cell entity is generated based on the local density feature information and the global density feature information.

[0302] Optionally, the feature generation module 906 is further configured to:

[0303] Parsing the cell attribute information corresponding to each cell entity to obtain the cell position information corresponding to each cell entity, and determining each cell entity as a pathological feature extraction cell entity;

[0304] Determine the density feature information corresponding to each cell entity as the cell feature information corresponding to each pathological feature extraction cell entity;

[0305] Extracting cell feature information and cell position information corresponding to the cell entity based on each pathological feature to obtain at least one approximate cell set;

[0306] Extract cell feature information and cell position information corresponding to the cell entity according to each pathological feature in each approximate cell set, and generate aggregated cell entity information corresponding to each approximate cell set;

[0307] When the number of aggregated cell entity information is greater than the image feature extraction threshold, determining each aggregated cell entity information as a pathology feature extraction cell entity, and returning to execute the step of acquiring at least one approximate cell set based on the cell feature information and cell position information corresponding to each pathology feature extraction cell entity;

[0308] When the number of aggregated cell entity information is less than or equal to the image feature extraction threshold, pathological image feature information corresponding to the target pathological image is generated based on each aggregated cell entity information.

[0309] Optionally, the feature generation module 906 is further configured to:

[0310] Determine a target approximate cell set, wherein the target approximate cell set is any one of the approximate cell sets;

[0311] Extracting cell position information corresponding to cell entities based on each pathological feature in the target approximate cell set, and generating aggregated cell position information corresponding to the target approximate cell set;

[0312] Extracting cell feature information corresponding to cell entities according to each pathological feature in the target approximate cell set, and generating aggregated cell feature information corresponding to the target approximate cell set;

[0313] According to the aggregated cell position information and the aggregated cell characteristic information, aggregated cell entity information corresponding to the target approximate cell set is obtained.

[0314] Optionally, the feature generation module 906 is further configured to:

[0315] Determine a first pathological feature extraction cell entity and at least one second pathological feature extraction cell entity in the target approximate cell set, wherein the first pathological feature extraction cell entity is any one of the pathological feature extraction cell entities in the target approximate cell set, and the second pathological feature extraction cell entity is a cell entity other than the first pathological feature extraction cell entity in the target approximate cell set;

[0316] Based on the cell feature information corresponding to the first pathological feature extraction cell entity and the cell feature information corresponding to each second pathological feature extraction cell entity, generate first verification attention feature information corresponding to the first pathological feature extraction cell entity and second verification attention feature information and second extended attention feature information corresponding to each second pathological feature extraction cell entity;

[0317] Generate extended cell feature information corresponding to the first pathological feature extraction cell entity according to the first verification attention feature information, the second verification attention feature information corresponding to each second pathological feature extraction cell entity, and the second extended attention feature information;

[0318] The extended cell feature information corresponding to the cell entity is extracted according to each pathological feature in the target approximate cell set, and the aggregated cell feature information corresponding to the target approximate cell set is generated.

[0319] Optionally, the feature generation module 906 is further configured to:

[0320] Inputting the cell feature information corresponding to the first pathological feature extraction cell entity into the attention feature extraction network to obtain the first verification attention feature information corresponding to the first pathological feature extraction cell entity;

[0321] The cell feature information corresponding to each pathological feature extraction cell entity is input into the attention feature extraction network, and the second verification attention feature information and the second extended attention feature information corresponding to each pathological feature extraction cell entity are input into the attention feature extraction network.

[0322] Optionally, the feature generation module 906 is further configured to:

[0323] Determine feature extension distance information corresponding to the second pathological feature extraction cell entity according to the cell position information corresponding to the first pathological feature extraction cell entity and the cell position information corresponding to each second pathological feature extraction cell entity, wherein the feature extension distance information is the distance information between the cell position information of each second pathological feature extraction cell entity and the cell position information of the first pathological feature extraction cell entity;

[0324] Generate feature extension weight information corresponding to each second pathological feature extraction cell entity according to the first verification attention feature information, the second verification attention feature information corresponding to each second pathological feature extraction cell entity, and the feature extension distance information;

[0325] Generate extended feature information corresponding to the first pathological feature extraction cell entity based on feature extension weight information, second extended attention feature information and feature extension distance information corresponding to each second pathological feature extraction cell entity;

[0326] The extended cell feature information corresponding to the first pathological feature extraction cell entity is generated according to the extended feature information.

[0327] By applying the scheme of the embodiments of this specification, through the modular design of the pathological image analysis device, it is possible to gradually parse the attribute information of each cell entity in the image on the basis of acquiring the target pathological image, and accurately extract density feature information based on these attribute information, generate pathological image feature information that characterizes the spatial distribution of cells, and form detailed image analysis results for the target pathological image. In the structure of this device, the collaboration of each module realizes the complete process from image raw data to feature extraction and then to the generation of analysis results. In particular, after obtaining cell location information and category information, through the strategy of flexibly generating density features, it can adapt to the analysis needs of pathological images of different scales and complexities, and further improve the efficiency and accuracy of image analysis. At the same time, through modular optional design, it supports the refined processing of the spatial characteristics and category distribution of the target cell entity and its surrounding background cells, providing strong technical support for complex pathological tasks.

[0328] The above is a schematic scheme of a pathological image analysis device of this embodiment. It should be noted that the technical scheme of the pathological image analysis device and the technical scheme of the pathological image analysis method described above are of the same concept, and the details not described in detail in the technical scheme of the pathological image analysis device can be found in the description of the technical scheme of the pathological image analysis method described above.

[0329] Fig.10 The block diagram of a computing device 1000 according to an embodiment of the present specification is shown. The components of the computing device 1000 include but are not limited to a memory 1010 and a processor 1020. The processor 1020 is connected to the memory 1010 via a bus 1030, and the database 1050 is used to store data.

[0330] The computing device 1000 also includes an access device 1040 that enables the computing device 1000 to communicate via one or more networks 1060. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 1040 may include one or more of any type of network interface (e.g., a network interface card (NIC)) of wired or wireless, such as an IEEE 1002.11 wireless local area network (WLAN) wireless interface, a world-wide interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, and a near field communication (NFC).

[0331] In one embodiment of the present specification, the above components of the computing device 1000 and Fig.10 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Fig.10 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0332] The computing device 1000 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 1000 may also be a mobile or stationary server.

[0333] The processor 1020 is used to execute the following computer executable instructions, which, when executed by the processor, implement the steps of the above-mentioned pathological image analysis method and computer-aided diagnosis method for tumors.

[0334] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the above-mentioned pathological image analysis method and computer-aided diagnosis method of tumors belong to the same concept, and the details not described in detail in the technical scheme of the computing device can be referred to the description of the technical scheme of the above-mentioned pathological image analysis method and computer-aided diagnosis method of tumors.

[0335] An embodiment of the present specification also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the above-mentioned pathological image analysis method and computer-aided diagnosis method for tumors.

[0336] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the above-mentioned pathological image analysis method and computer-aided diagnosis method of tumors belong to the same concept, and the details not described in detail in the technical scheme of the storage medium can be referred to the description of the technical scheme of the above-mentioned pathological image analysis method and computer-aided diagnosis method of tumors.

[0337] An embodiment of the present specification also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned pathological image analysis method and the computer-aided diagnosis method of tumors.

[0338] The above is a schematic scheme of a computer program of this embodiment. It should be noted that the technical scheme of the computer program and the technical scheme of the above-mentioned pathological image analysis method and computer-aided diagnosis method of tumors belong to the same concept, and the details not described in detail in the technical scheme of the computer program can be referred to the description of the technical scheme of the above-mentioned pathological image analysis method and computer-aided diagnosis method of tumors.

[0339] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0340] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0341] It should be noted that, for the convenience of description, the aforementioned method embodiments are all described as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0342] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0343] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A pathological image analysis method, comprising: Acquiring pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity; Based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained; Generating pathological image characteristic information corresponding to the target pathological image according to the density characteristic information corresponding to each cell entity; An image analysis result corresponding to the target pathological image is generated based on the pathological image feature information.

2. The method according to claim 1, obtaining pathological image information corresponding to the target pathological image, comprising: Acquire target pathological images; Inputting the target pathological image into a cell annotation model to obtain cell position information and cell category information corresponding to each cell entity generated by the cell annotation model; Based on the cell position information and cell category information corresponding to each cell entity, the cell attribute information corresponding to each cell entity is generated.

3. The method according to claim 1, obtaining density feature information corresponding to each cell entity based on the cell attribute information corresponding to each cell entity, comprising: Parsing the cell attribute information corresponding to each cell entity, and obtaining the cell location information and cell category information corresponding to the cell attribute information corresponding to each cell entity; Based on the cell position information and cell category information corresponding to each cell entity, density feature information corresponding to each cell entity is generated.

4. The method according to claim 3, generating density feature information corresponding to each cell entity based on the cell position information and cell category information corresponding to each cell entity, comprising: generating a density calculation radius threshold according to the cell position information corresponding to each cell entity, and determining at least one density calculation radius based on the density calculation radius threshold; Based on the density calculation radius and the cell position information and cell category information corresponding to each cell entity, density feature information corresponding to each cell entity is generated.

5. The method according to claim 4, generating density feature information corresponding to each cell entity based on each density calculation radius and cell position information and cell category information corresponding to each cell entity, comprising: Determine a target cell entity and at least one background cell entity, wherein the target cell entity is any one of the cell entities, and the background cell entity is any one of the cell entities except the target cell entity; Based on each density calculation radius, the target cell position information corresponding to the target cell entity and the background cell position information corresponding to each background cell entity, determining a reference cell entity set of the target cell entity for each density calculation radius in each background cell entity, wherein the reference cell entity set includes at least one reference cell entity; Density feature information corresponding to the target cell entity is generated according to the cell category information corresponding to each reference cell entity in each reference cell entity set.

6. The method according to claim 5, based on each density calculation radius, the target cell position information corresponding to the target cell entity and the background cell position information corresponding to each background cell entity, determining a reference cell entity set of the target cell entity for each density calculation radius in each background cell entity, comprising: Based on the target cell position information and the background cell position information, obtaining the distance information between the target cell entity and each background cell entity; Obtaining a first density calculation radius, wherein the first density calculation radius is any one of the density calculation radii; At least one reference cell entity is determined in each background cell entity, wherein the reference cell entity is a background cell entity whose distance information to the target cell entity is less than or equal to the first density calculation radius.

7. The method according to claim 5, generating density feature information corresponding to the target cell entity according to the cell category information corresponding to each reference cell entity in each reference cell entity set, comprising: acquiring at least one cell type information based on each cell type information; Generate at least one density feature sub-information corresponding to the target cell entity according to each reference cell entity set and each cell type information; The density characteristic information corresponding to the target cell entity is generated according to each density characteristic sub-information.

8. The method of claim 7, generating at least one density feature sub-information corresponding to the target cell entity according to each reference cell entity set and each cell type information, comprising: Determine first cell type information, a first reference cell entity set, and a second reference cell entity set, wherein the first cell type information is any one of the cell type information, the first reference cell entity set is any one of the reference entity sets, the second reference cell entity set is a reference cell entity set that is a previous reference cell entity set in a reference cell set sequence, and the reference cell set sequence is a sequence generated by sorting the reference entity sets in ascending order according to the density calculation radius corresponding to each reference entity set; Acquire target first type cell quantity information corresponding to the first reference cell entity set, and initial first type cell quantity information corresponding to the second reference cell entity set; Generate first type cell quantity change information corresponding to the first reference cell entity set according to the target first type cell quantity information and the initial first type cell quantity information, wherein the first type cell quantity change information represents a change in the quantity of first type cell entities in the first reference cell entity set compared with the quantity of first type cell entities in the second reference cell entity set, and the first type cell entity is a cell entity whose cell type information is the first type cell information; According to the first cell type quantity change information, density feature sub-information of the first reference cell entity set for the first cell type information is obtained.

9. The method of claim 8, wherein obtaining density feature sub-information of the first reference cell entity set for the first cell type information according to the first cell type number change information comprises: Acquire local reference first type cell quantity information and global reference first type cell quantity information, wherein the local reference first type cell quantity information represents the quantity information of the first type cell entity in the cell entity whose distance from the target cell entity is less than or equal to the density calculation radius threshold, and the global reference first type cell quantity information is the quantity information of each cell entity that is sorted from largest to smallest in the local reference first type cell quantity information; Generate local density feature sub-information of the first reference cell entity set for the first cell type information according to the local reference first cell type number information and the first cell type number change information; Generate global density feature sub-information of the first reference cell entity set for the first cell type information according to the global reference first cell type number information and the first cell type number change information; Accordingly, density feature information corresponding to the target cell entity is generated according to each density feature sub-information, including: Generate local density feature information corresponding to the target cell entity according to each local density feature sub-information, and generate global density feature information corresponding to the target cell entity according to each global density feature sub-information; Density feature information corresponding to the target cell entity is generated based on the local density feature information and the global density feature information.

10. The method according to claim 1, generating pathological image feature information corresponding to the target pathological image according to density feature information corresponding to each cell entity, comprising: Parsing the cell attribute information corresponding to each cell entity to obtain the cell position information corresponding to each cell entity, and determining each cell entity as a pathological feature extraction cell entity; Determine the density feature information corresponding to each cell entity as the cell feature information corresponding to each pathological feature extraction cell entity; Extracting cell feature information and cell position information corresponding to the cell entity based on each pathological feature to obtain at least one approximate cell set; Extract cell feature information and cell position information corresponding to the cell entity according to each pathological feature in each approximate cell set, and generate aggregated cell entity information corresponding to each approximate cell set; When the number of aggregated cell entity information is greater than the image feature extraction threshold, determining each aggregated cell entity information as a pathology feature extraction cell entity, and returning to execute the step of acquiring at least one approximate cell set based on the cell feature information and cell position information corresponding to each pathology feature extraction cell entity; When the number of aggregated cell entity information is less than or equal to the image feature extraction threshold, pathological image feature information corresponding to the target pathological image is generated based on each aggregated cell entity information.

11. The method according to claim 10, extracting cell feature information and cell location information corresponding to the cell entity according to each pathological feature in each approximate cell set, and generating aggregated cell entity information corresponding to each approximate cell set, comprising: Determine a target approximate cell set, wherein the target approximate cell set is any one of the approximate cell sets; Extracting cell position information corresponding to cell entities based on each pathological feature in the target approximate cell set, and generating aggregated cell position information corresponding to the target approximate cell set; Extracting cell feature information corresponding to cell entities according to each pathological feature in the target approximate cell set, and generating aggregated cell feature information corresponding to the target approximate cell set; According to the aggregated cell position information and the aggregated cell characteristic information, aggregated cell entity information corresponding to the target approximate cell set is acquired.

12. The method according to claim 11, extracting cell feature information corresponding to cell entities according to each pathological feature in the target approximate cell set, and generating aggregated cell feature information corresponding to the target approximate cell set, comprises: Determine a first pathological feature extraction cell entity and at least one second pathological feature extraction cell entity in the target approximate cell set, wherein the first pathological feature extraction cell entity is any one of the pathological feature extraction cell entities in the target approximate cell set, and the second pathological feature extraction cell entity is a cell entity other than the first pathological feature extraction cell entity in the target approximate cell set; Based on the cell feature information corresponding to the first pathological feature extraction cell entity and the cell feature information corresponding to each second pathological feature extraction cell entity, generate first verification attention feature information corresponding to the first pathological feature extraction cell entity and second verification attention feature information and second extended attention feature information corresponding to each second pathological feature extraction cell entity; Generate extended cell feature information corresponding to the first pathological feature extraction cell entity according to the first verification attention feature information, the second verification attention feature information corresponding to each second pathological feature extraction cell entity, and the second extended attention feature information; The extended cell feature information corresponding to the cell entity is extracted according to each pathological feature in the target approximate cell set, and the aggregated cell feature information corresponding to the target approximate cell set is generated.

13. The method according to claim 12, generating first verification attention feature information corresponding to the first pathological feature extraction cell entity and second verification attention feature information and second extended attention feature information corresponding to each second pathological feature extraction cell entity based on the cell feature information corresponding to the first pathological feature extraction cell entity and the cell feature information corresponding to each second pathological feature extraction cell entity, comprising: Inputting the cell feature information corresponding to the first pathological feature extraction cell entity into the attention feature extraction network to obtain the first verification attention feature information corresponding to the first pathological feature extraction cell entity; The cell feature information corresponding to each pathological feature extraction cell entity is input into the attention feature extraction network, and the second verification attention feature information and the second extended attention feature information corresponding to each pathological feature extraction cell entity are input into the attention feature extraction network.

14. The method of claim 12, generating extended cell feature information corresponding to the first pathological feature extraction cell entity according to the first verification attention feature information, the second verification attention feature information corresponding to each second pathological feature extraction cell entity, and the second extended attention feature information, comprising: Determine feature extension distance information corresponding to the second pathological feature extraction cell entity according to the cell position information corresponding to the first pathological feature extraction cell entity and the cell position information corresponding to each second pathological feature extraction cell entity, The feature extension distance information is the distance information between the cell position information of each second pathological feature extraction cell entity and the cell position information of the first pathological feature extraction cell entity; Generate feature extension weight information corresponding to each second pathological feature extraction cell entity according to the first verification attention feature information, the second verification attention feature information corresponding to each second pathological feature extraction cell entity, and the feature extension distance information; Generate extended feature information corresponding to the first pathological feature extraction cell entity based on feature extension weight information, second extended attention feature information and feature extension distance information corresponding to each second pathological feature extraction cell entity; The extended cell feature information corresponding to the first pathological feature extraction cell entity is generated according to the extended feature information.

15. A computer-aided diagnosis method for tumors, comprising: Acquiring pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity; Based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained; Generating pathological image characteristic information corresponding to the target pathological image according to the density characteristic information corresponding to each cell entity; An image recognition result corresponding to the target pathological image is generated based on the pathological image feature information.

16. A computer-aided diagnosis method for tumors, applied to a client, comprising: Acquiring pathological image information corresponding to a target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity; Based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained; According to the density feature information corresponding to each cell entity, pathological image feature information corresponding to the target pathological image is generated; and based on the pathological image feature information, an image recognition result corresponding to the target pathological image is generated.

17. A computer-aided diagnosis method for tumors, applied to a cloud device, comprising: A tumor diagnosis request sent by a receiving end-side device, wherein the tumor diagnosis request carries a target pathological image; Acquiring pathological image information corresponding to the target pathological image, wherein the target pathological image includes at least one cell entity, and the pathological image information includes cell attribute information corresponding to each cell entity; Based on the cell attribute information corresponding to each cell entity, density feature information corresponding to each cell entity is obtained; Generating pathological image characteristic information corresponding to the target pathological image according to the density characteristic information corresponding to each cell entity; An image recognition result corresponding to the target pathological image is generated based on the pathological image feature information, and the image recognition result is sent to the terminal side device.

18. A computing device comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer program / instructions are executed by the processor, the steps of the method described in any one of claims 1 to 17 are implemented.

19. A computer-readable storage medium storing a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the steps of the method according to any one of claims 1 to 17.

20. A computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 17.

Images