Cell labeling model training method and tumor computer-aided diagnosis method

Abstract

The embodiment of the invention provides a cell labeling model training method and a tumor computer-aided diagnosis method, and the method comprises the steps: obtaining a target pathological image, and generating at least one pathological image slice which comprises at least one cell entity based on the target pathological image; inputting each pathological image slice into a cell labeling model, obtaining cell category information and cell position information corresponding to each cell entity in each pathological image slice generated by the cell labeling model, and performing cell labeling according to the cell category information and the cell position information corresponding to each cell entity in each pathological image slice, determining an initial slice category corresponding to each pathological image slice; inputting each pathological image slice into a slice verification model to obtain a target slice category corresponding to each pathological image slice generated by the slice verification model; and training a cell labeling model according to the initial slice category and the target slice category corresponding to each pathological image slice. The cost is reduced, the efficiency is improved, and new technical requirements are met.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and in particular to a cell labeling model training method. Background Art

[0002] With the continuous development of medical imaging technology, by accurately marking specific types of cells or structures (such as tumor cells), doctors can more accurately identify the lesion area and improve the efficiency and accuracy of diagnosis. Detailed and accurate cell annotation data can also be used to explore the mechanism of disease occurrence and development and discover new biomarkers.

[0003] Currently, the manual cell annotation method can annotate every region or cell of interest and complete the formulation of annotation standards. However, this method is costly and requires a lot of time and manpower; it is inefficient, the manual annotation speed is slow, and it is difficult to meet the rapidly growing scientific research needs; it has poor scalability, and the old annotation rules may not be able to adapt to the requirements of new technologies and new discoveries, and updating the database is time-consuming and labor-intensive. Therefore, in order to solve the above shortcomings, a cell annotation model training method is needed to obtain a cell annotation model to achieve automatic and accurate cell annotation. Summary of the invention

[0004] In view of this, the embodiments of this specification provide a cell annotation model training method, a cell annotation method, a computer-aided diagnosis method for tumors, a computer-aided diagnosis method for tumors applied to a client, a computer-aided diagnosis method for tumors applied to a cloud device, and a cell annotation method applied to a cloud device. This specification also involves a computer-aided diagnosis system for tumors, a cell annotation model training 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 cell labeling model training method is provided, comprising:

[0006] Acquire a target pathological image, and generate at least one pathological image slice based on the target pathological image, wherein the pathological image slice includes at least one cell entity;

[0007] Input each pathological image slice into a cell annotation model, obtain cell category information and cell position information corresponding to each cell entity in each pathological image slice generated by the cell annotation model, and determine 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;

[0008] Inputting each pathological image slice into a slice verification model to obtain a target slice category corresponding to each pathological image slice generated by the slice verification model;

[0009] The cell annotation model is trained according to the initial slice category and the target slice category corresponding to each pathological image slice.

[0010] According to a second aspect of an embodiment of this specification, a cell labeling method is provided, comprising:

[0011] Acquiring a pathological image to be processed, wherein the pathological image to be processed includes at least one cell entity;

[0012] Inputting the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method;

[0013] According to the cell category information and cell position information corresponding to each cell entity, a cell annotation result corresponding to the pathological image to be processed is generated.

[0014] According to a third aspect of the embodiments of this specification, a computer-aided diagnosis method for tumors is provided, comprising:

[0015] Acquiring a pathological image to be processed, wherein the pathological image to be processed includes at least one cell entity;

[0016] Inputting the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method;

[0017] According to the cell category information and cell position information corresponding to each cell entity, an image detection result corresponding to the pathological image to be processed is generated.

[0018] 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 client and includes:

[0019] Acquiring a pathological image to be processed, wherein the pathological image to be processed includes at least one cell entity;

[0020] Inputting the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method;

[0021] According to the cell category information and cell position information corresponding to each cell entity, an image detection result corresponding to the pathological image to be processed is generated.

[0022] According to a fifth aspect of the embodiments of this specification, a computer-aided diagnosis method for tumors is provided, which is applied to a cloud device, including:

[0023] A tumor diagnosis request sent by a receiving end-side device, wherein the tumor diagnosis request includes a pathological image to be processed, and the pathological image to be processed includes at least one cell entity;

[0024] Inputting the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method;

[0025] According to the cell category information and cell position information corresponding to each cell entity, an image detection result corresponding to the pathological image to be processed is generated, and the image detection result is sent to the terminal side device.

[0026] According to a sixth aspect of an embodiment of this specification, a computing device is provided, including:

[0027] Memory and processor;

[0028] 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 cell labeling model training method, cell labeling method and computer-aided diagnosis method of tumors are implemented.

[0029] According to the seventh 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 cell labeling model training method, cell labeling method and computer-aided diagnosis method of tumors.

[0030] According to an eighth 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 cell labeling model training method, cell labeling method and computer-aided diagnosis method for tumors.

[0031] By applying the solution of the embodiments of this specification, through the collaborative training of the cell annotation model and the slice verification model, iterative optimization of the annotation process is achieved, which can quickly and accurately complete the classification and positioning of large-scale cells, significantly improve the annotation efficiency, meet the rapidly growing scientific research needs, and then realize automated technology to process the target pathological images without a lot of manual intervention, significantly reducing the time and labor cost of annotation. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of a cell labeling model training method provided by one embodiment of this specification;

[0033] Figure 2 is a schematic diagram of a cell labeling model training process provided by an embodiment of this specification;

[0034] Figure 3 is a flow chart of a cell labeling method provided by one embodiment of this specification;

[0035] Figure 4a is a schematic diagram of a target pathological image provided by an embodiment of this specification;

[0036] Figure 4b is a schematic diagram of a cell labeling result provided in one embodiment of this specification;

[0037] Figure 5 is a flow chart of a computer-aided diagnosis method for tumors provided by one embodiment of this specification;

[0038] Figure 6 is a flowchart of a computer-aided diagnosis method for tumors applied to a client provided by an embodiment of this specification;

[0039] Figure 7 is a flow chart of a computer-aided diagnosis method for tumors applied to a cloud device provided by one embodiment of this specification;

[0040] Figure 8 is a flow chart of a cell labeling method applied to a cloud device provided by an embodiment of this specification;

[0041] Fig. 9 is an architecture diagram of a computer-aided diagnosis system for tumors provided by one embodiment of this specification;

[0042] Fig.10 is a processing flow chart of a computer-assisted prognostic analysis method provided in one embodiment of the present specification;

[0043] Fig.11It is a structural schematic diagram of a cell labeling model training device provided by an embodiment of this specification;

[0044] Fig.12 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0045] 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.

[0046] 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.

[0047] 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, this 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".

[0048] 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.

[0049] In one or more embodiments of this specification, a large model refers to a deep learning model with large-scale model parameters, which usually contains hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than 10 trillion model parameters. A large model can also be called a foundation model / foundation model. The large model is pre-trained with large-scale unlabeled corpus to produce a pre-trained model with more than 100 million parameters. This model can adapt to a wide range of downstream tasks, and the model has good generalization ability, such as a large-scale language model (LLM), a multi-modal pre-training model, etc.

[0050] When the big model is used in practice, only a small number of samples are needed to fine-tune the pre-trained model and it can be applied to different tasks. The big model can be widely used in natural language processing (NLP), computer vision and other fields. Specifically, it can be applied to computer vision tasks such as visual question answering (VQA), image caption (IC), image generation, as well as natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. The main application scenarios of the big model include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0051] In this specification, a cell annotation model training method, a cell annotation method, a computer-aided diagnosis method for tumors, a computer-aided diagnosis method for tumors applied to a client, a computer-aided diagnosis method for tumors applied to a cloud device, and a cell annotation method applied to a cloud device are provided. This specification also involves a computer-aided diagnosis system for tumors, a cell annotation model training 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.

[0052] By accurately marking specific types of cells or structures (such as cancer cells), doctors can more accurately identify diseased areas and improve diagnostic efficiency and accuracy. Detailed and accurate cell annotation data can also be used to explore the mechanisms of disease occurrence and development, discover new biomarkers, etc. However, manual cell annotation requires professional knowledge background, and the process is very time-consuming; at the same time, the scarcity of professional talents leads to high labor costs. This limitation makes it not feasible to achieve large-scale annotation based on manual labor. In order to overcome the above challenges, it is particularly important to develop more efficient and cost-effective cell annotation methods.

[0053] See also Figure 1 , Figure 1 A flowchart of a cell labeling model training method provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0054] Step 102: Acquire a target pathological image, and generate at least one pathological image slice based on the target pathological image, wherein the pathological image slice includes at least one cell entity.

[0055] In practical applications, the target pathological image is a complete pathological slice image containing multiple cells, which is used to provide comprehensive diagnostic information; the pathological image slice is a smaller image block extracted from the target pathological image, which is convenient for detailed analysis of local areas; the cell entity is the specific cell unit in the pathological image, and its category and location information are identified through the image annotation model to assist disease detection and classification.

[0056] A target pathology image can be understood as a pathology image that covers the entire picture of a tissue or organ, and is the basic data source for pathology analysis. This type of image is usually obtained from a glass slide sample using panoramic scanning technology and has ultra-high resolution. For example, a target pathology image may show a complete tissue section of a cancer patient, which includes both normal tissue areas and cancerous areas. By locating and segmenting the area of ​​interest on the target pathology image, accurate input data can be provided for subsequent analysis.

[0057] Pathology image slices can be understood as smaller image blocks cut out from the target pathology image for detailed analysis of local tissue or cell areas. For example, a target pathology image may be segmented into thousands of pathology image slices, each covering a few hundred pixels. The size and number of slices can be flexibly adjusted according to the research objectives or analysis requirements. This approach can significantly reduce computing costs while improving the efficiency of analyzing key areas.

[0058] Cell entities can be understood as specific cell units visible in pathological image slices, and are the main target of cell classification and detection models. For example, in cancer research, cell entities may include cancer cells, immune cells, or normal cells. Through pre-trained cell detection models, these cell entities can be accurately identified and classified into different cell types, while annotating their precise locations in the slices. These annotation results not only provide key evidence for disease diagnosis, but can also be used to explore disease progression and treatment effects.

[0059] It should be noted that generating at least one pathological image slice based on the target pathological image can be understood as a key step in extracting local areas from the panoramic pathological image for fine-grained analysis, and its significance lies in segmenting the target area for efficient implementation of subsequent cell annotation and classification analysis. Specifically, the target pathological image can be cut into multiple continuous image blocks of a preset size by a fixed-size sliding window method to cover the entire image area; it can also be done by an automatic segmentation method based on region of interest (ROI) detection to extract only areas containing key lesions or specific structures to reduce the processing burden of irrelevant areas; it can also be combined with a deep learning model, such as using a pre-trained semantic segmentation network to dynamically identify and extract slices with important pathological features in the image, etc. This manual does not impose any restrictions on this.

[0060] By acquiring multiple pathological image slices corresponding to the target pathological image, the data used for model training can be effectively expanded and the problem of increased training costs caused by excessive data volume during model training can be reduced.

[0061] Step 104: Input each pathological image slice into the cell annotation model, obtain 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 determine the initial slice category corresponding to each pathological image slice based on the cell category information and cell position information corresponding to each cell entity in each pathological image slice.

[0062] In practical applications, the cell annotation model is a machine learning model that can automatically detect and classify cells in pathological images, and is responsible for generating cell category information and location information; the cell category information is the classification label of each cell entity output by the cell annotation model, such as cancer cells or normal cells; the cell location information is the specific spatial coordinates of the cell entity in the pathological image, which is convenient for subsequent analysis and positioning; the initial slice category is the preliminary classification label of the pathological image slice determined according to the cell annotation model, which is used to guide further training and optimization of the model.

[0063] The cell annotation model can be understood as a deep learning model that is specifically used for the detection and classification of cells in pathological images. By inputting pathological image slices, the model can automatically identify the cell entities in the image and output the category information of each cell according to predefined classification rules. For example, in tumor research, the cell annotation model can classify cells as cancer cells, immune cells, or normal cells, thereby providing researchers with accurate cell distribution and category information and reducing manual annotation costs.

[0064] Cell category information can be understood as the classification label assigned by the cell annotation model to each cell entity. Each label represents the type of the cell entity, such as cancer cells, inflammatory cells, or other specific cell types. With this information, researchers can quickly understand the cellular composition of pathological images. For example, in cancer pathology analysis, cell category information can clearly show the distribution of cancer cells, helping doctors quickly locate tumor areas and develop treatment plans.

[0065] Cell location information can be understood as the precise coordinates of each cell entity in the pathological image. This location information is automatically generated by the model and represented in two-dimensional or three-dimensional form. For example, in a pathological image slice, the location information of a cancer cell may be represented as the pixel coordinates (x, y) on the slice. This location information not only helps to visualize the cells, but also serves as the basic data for subsequent analysis (such as cell spacing measurement or spatial distribution research).

[0066] The initial slice category can be understood as the overall slice classification label obtained by aggregating the category information and location information of each cell entity in the slice. For example, a pathology image slice may be labeled as "high-density cancer cell area" or "mainly composed of normal cells" to guide the model to further verify and optimize the slice. The initial slice category provides an important basis for subsequent slice classification and model fine-tuning, and helps to quickly screen out key slice samples for in-depth analysis.

[0067] It should be noted that, according to the cell category information and cell position information corresponding to each cell entity in each pathological image slice, determining the initial slice category corresponding to each pathological image slice can be understood as comprehensively evaluating the overall characteristics of the slice through the output information of the cell annotation model, providing a reference basis for subsequent verification and optimization, and thus improving the accuracy and availability of pathological image slice classification. The specific method can be to extract the main category information and classify the slice based on the preset threshold by counting the proportion of each cell category in the slice; the cell distribution characteristics generated based on the spatial density and position relationship of certain categories of cells, and then constructing the feature vector based on the above cell distribution characteristics and inputting it into the classification model for label prediction; the overall distribution characteristics of the cell category and position information in the slice can also be analyzed through a clustering algorithm to generate the initial category label of the slice, etc. This manual does not impose any restrictions on this.

[0068] By using the location information and category information of each cell entity generated by the cell annotation model to obtain the initial slice category corresponding to the cell slice, a more accurate input basis can be provided for subsequent model optimization and data analysis, thereby reducing manual intervention and improving processing efficiency and accuracy. This method can quickly identify the main cell features in the slice, and use this as a basis for sample screening and preliminary assignment of classification labels, which helps to reduce computational complexity while ensuring the consistency of slice classification, laying a solid foundation for further verification and optimization of the annotation model. Accurate initial classification labels can also accelerate the analysis and processing of large-scale pathological images, providing efficient support for scientific research and clinical applications.

[0069] Considering the need to locate and classify cell entities in slices, the cell annotation model includes a cell location module and a cell classification module;

[0070] Furthermore, each pathological image slice is input into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity in each pathological image slice generated by the cell annotation model, including:

[0071] Determine a first pathological image slice, wherein the first pathological image slice is any one of the pathological image slices;

[0072] Inputting the first pathological image slice into the cell positioning module to obtain cell position information corresponding to each cell entity in the first pathological image slice generated by the cell positioning module;

[0073] The first pathological image slice and each cell position information are processed according to the cell classification module to obtain cell category information corresponding to each cell entity.

[0074] In practical applications, the cell localization module is a component used to accurately mark the spatial position of each cell entity in pathological image slices, providing basic information for subsequent cell classification; the cell classification module is a component that determines the category of located cells, and outputs the specific category information of each cell entity through a classification algorithm to support pathological analysis and diagnosis.

[0075] The cell localization module can be understood as a model component that focuses on detecting the location of cells in pathological images. Its function is to accurately locate the spatial coordinates of each cell entity by analyzing the pixel distribution of pathological image slices. For example, in a panoramic pathological image, the cell localization module can mark the center point or boundary area of ​​a cancer cell colony and provide information on the geometric distribution of cells in the slice. This positioning capability can not only significantly reduce the workload of manual annotation, but also provide high-quality location information for subsequent cell classification and feature extraction.

[0076] The cell classification module can be understood as a model component that identifies and classifies cell types based on cell location information. Its core task is to determine the specific category of the cell based on the location information of each cell output by the positioning module and the local features of the pathological image. For example, the classification module can mark the located cells as cancer cells, inflammatory cells, or normal cells, and further distinguish their subtypes. This module can help pathologists quickly understand the cellular composition and distribution characteristics of diseased tissues, and provide a basis for accurate diagnosis and treatment decisions.

[0077] It should be noted that the cell classification module processes the first pathological image slice and the position information of each cell to obtain the cell category information corresponding to each cell entity. It can be understood as a process of extracting a local image of a cell and determining its type using a classification algorithm. Its significance lies in giving each cell entity a clear category label, thereby providing reliable data support for subsequent analysis and research. The specific method can be to extract the local cell image of each cell entity and input it into the cell classification module one by one for separate classification to obtain the type information of each cell; it can also be batch processed by multiple cell images, combined with a deep learning model to generate the category prediction results of all cells at one time to improve processing efficiency; it can also integrate contextual information, such as analyzing the distribution pattern and correlation characteristics of cells and neighboring cells, to optimize the accuracy and consistency of classification results, etc. This manual does not impose any restrictions on this.

[0078] After determining the location information of each cell through the cell positioning module, the corresponding category of each cell is obtained, which can coordinate the division of labor between positioning and classification tasks, and significantly improve the accuracy and efficiency of model processing. This method can not only accurately mark the specific location of cells in the image, but also provide clear input for subsequent classification steps, so that the classification module can focus on the determination of cell types, thereby reducing interference and error rates. In addition, through this step-by-step processing method, it can also better adapt to the diverse features in complex pathological images, help to quickly identify lesion areas and improve the reliability of analysis, and provide strong technical support for disease research and diagnosis.

[0079] Further, the cell classification module processes the first pathological image slice and each cell position information to obtain cell category information corresponding to each cell entity, including:

[0080] Determine the cell image corresponding to each cell entity based on the position information of each cell in the first pathological image slice;

[0081] Each cell image is input into the cell classification module to obtain the cell category information corresponding to each cell entity generated by the cell classification module.

[0082] In practical applications, a cell image is a local image containing a single or a small number of cells extracted from a pathological image slice based on cell position information, which is used for subsequent classification and feature analysis. A cell image can be understood as a high-resolution local image that specifically displays the detailed features of a single or a small number of cells in a pathological image. Its function is to provide a direct analysis basis for the cell classification module by accurately extracting the target cell area. For example, from a pathological image slice, a cell image may be extracted as a local enlarged area containing cancer cell nuclei, making it easier for the cell classification module to identify its type, state or characteristics.

[0083] By processing cell images separately, the focus and accuracy of the analysis can be improved, while reducing the interference of redundant information, providing technical support for efficient cell classification and annotation. In addition, cell images can also be used for subsequent visualization and result verification, such as showing the typical morphology of a specific cell type in research, helping researchers to understand pathological data more intuitively.

[0084] Further, according to the cell category information and cell position information corresponding to each cell entity in each pathological image slice, the initial slice category corresponding to each pathological image slice is determined, including:

[0085] Determine the cell distribution characteristic information 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;

[0086] Based on the cell distribution characteristic information corresponding to each pathological image slice, the initial slice category corresponding to each pathological image slice is determined.

[0087] In practical applications, cell distribution feature information is structured data that describes the spatial distribution and aggregation characteristics of cells in slices, extracted based on the position information and category information of cells in pathological images. It is used to evaluate the overall cell characteristics of slices and their association with pathological significance. Cell distribution feature information can be understood as an information expression that comprehensively analyzes the spatial relationship and category composition of cells in pathological images. Its role is to help researchers and algorithm models more comprehensively understand the distribution pattern of cells in tissues. For example, in cancer pathology, cell distribution feature information can include the ratio of cancer cells to normal cells, the degree of aggregation of cancer cells, and the spatial distance between different cell types. This information can not only reveal the microscopic mechanism of disease development, but also support diagnostic decisions and treatment plan design.

[0088] The cell distribution characteristic information obtained through the cell category information and cell location information corresponding to each cell entity can assist doctors in judging the aggressiveness of the tumor, or assist doctors in predicting the trend of pathological changes in a specific area, thereby providing more targeted clinical advice.

[0089] Step 106: input each pathological image slice into a slice verification model, and obtain a target slice category corresponding to each pathological image slice generated by the slice verification model.

[0090] In practical applications, the slice verification model is a deep learning model used to verify and optimize the preliminary classification results of pathological image slices. It improves the annotation quality by evaluating and correcting the slice categories. The target slice category is the classification label output by the slice verification model, which is used to represent pathological image slices to guide model training and sample screening.

[0091] The slice verification model can be understood as an auxiliary model specially designed to evaluate and correct the initial classification results of pathological image slices. It combines the prediction results of multiple pre-trained basic models and uses a voting mechanism or credibility judgment method to screen out classification labels with higher accuracy. For example, a pathological image slice is labeled as a "high-density cancer cell area" in the initial classification, but the slice verification model may reclassify it as an "inflammatory cell area" by analyzing the cell distribution characteristics and comprehensive judgment across models, thereby improving the consistency and reliability of the classification results.

[0092] The target slice category can be understood as the category label generated by the slice verification model for each pathological image slice, which represents the classification result generated by the verification model. The target slice category can accurately describe the main cell types or tissue characteristics in the slice. For example, a pathological image slice is initially labeled as a "normal cell area", but after analysis by the slice verification model, it is confirmed that it contains a small number of cancer cells. The target slice category may be adjusted to "potential cancer cell area". This result can not only provide higher quality samples for the training of subsequent cell annotation models, but also can be used to screen key pathological feature areas and accelerate the disease diagnosis and research process.

[0093] In order to further improve the credibility of the target slice category output by the verification model, the slice verification model includes at least one slice verification module;

[0094] Furthermore, each pathological image slice is input into a slice verification model to obtain a target slice category corresponding to each pathological image slice generated by the slice verification model, including:

[0095] Determine a second pathological image slice, wherein the second pathological image slice is any one of the pathological image slices;

[0096] Inputting the second pathological image slice into each slice verification module to obtain the reference slice category generated by each slice verification module;

[0097] A target slice category corresponding to the second pathological image slice is determined according to each reference slice category.

[0098] In practical applications, the slice verification module is an independent functional module used to verify and optimize the preliminary classification results of pathological image slices. The credibility of the target slice category is improved through the collaborative evaluation of multiple modules. The reference slice category is the intermediate classification result generated by the slice verification module according to the characteristics of the input slice, which serves as an important basis for further determining the target slice category.

[0099] The slice verification module can be understood as a system component that uses a variety of evaluation strategies to verify pathological image slices. Its role is to verify and optimize the classification labels of slices through the independent evaluation results of multiple models. For example, for a pathological image slice, the slice verification model may include three different slice verification modules, one of which is based on cell type distribution, another based on tissue structure characteristics, and another based on molecular marker signals. By integrating the output results of each module, the accuracy and consistency of the classification label are ensured. The introduction of the slice verification module not only improves the credibility of the target slice category, but also significantly reduces the impact of classification errors on subsequent analysis.

[0100] The reference slice category can be understood as the intermediate classification result generated by each slice verification module, which is used to provide preliminary classification information about the characteristics of the pathological image slice. For example, a pathological image slice may be independently labeled as "normal tissue", "inflammatory area" or "cancer cell dense area" by multiple modules during the verification process. These reference slice categories further determine the target category of the slice through voting or weighting mechanism.

[0101] It should be noted that inputting the second pathological image slice into each slice verification module and obtaining the reference slice category generated by each slice verification module can be understood as classifying and evaluating the pathological image slices through a variety of verification methods, and its significance lies in combining multi-angle analysis to improve the credibility and scientificity of the classification results. The specific method can be to splice the second pathological image slice with the pathological image slice prompt word template and input it into the slice verification module, and use the pre-trained slice verification module to generate a reference slice category; it can also be directly input into the deep learning verification module, and use feature extraction and classification networks to independently evaluate the slice features; it can also use the cell category statistics and spatial distribution characteristics of the second pathological image slice as input, and combine the rule classification model to generate a reference category, so as to supplement the judgment based on image features, etc. This manual does not impose any restrictions on this.

[0102] It should be noted that determining the target slice category corresponding to a pathological image slice according to each reference slice category can be understood as screening and confirming a more suitable classification label by integrating the evaluation results of multiple reference categories. Its significance lies in improving the credibility and consistency of the classification results and providing a reliable basis for subsequent analysis and model optimization. The specific method can be to use a voting mechanism to count all reference slice categories and select categories with higher frequency as target slice categories; or to use a weighted scoring method to calculate the comprehensive score of each reference category according to the weights of different verification modules and select categories with higher scores as target categories; or to set threshold rules, for example, when the number of reference results of a certain category exceeds a predetermined threshold, directly determine it as a target slice category, thereby quickly determining the target classification label of the slice, etc. This manual does not impose any restrictions on this.

[0103] Determining the target slice category through multiple slice verification modules ensures that the target classification results are fully combined with the judgment of multiple modules, improves the scientificity and reliability of the labels, and provides a solid foundation for the subsequent optimization and application of cell labeling models.

[0104] Further, determining the target slice category corresponding to the second pathological image slice according to each reference slice category includes:

[0105] determining a first slice category among the reference slice categories;

[0106] Acquire slice check module quantity information, and determine a check category determination threshold according to the slice check module quantity information;

[0107] When the number corresponding to each first slice category is greater than the verification category determination threshold, it is determined that the target slice category corresponding to the second pathological image slice is the first slice category.

[0108] In practical applications, the first slice category is a priority category selected through statistics or rules in the reference slice category, indicating a more likely classification label for the pathological image slice; the slice verification module quantity information is a key parameter indicating the number of modules involved in the verification task, which is used to adjust the classification strategy; the verification category determination threshold is a threshold set according to the slice verification module quantity information, which is used to determine the basis for determining the target slice category.

[0109] The first slice category can be understood as the priority classification result obtained by integrating various reference slice categories. Its role is to provide a preliminary and reliable label to guide the further determination of the target slice category. For example, among multiple reference slice categories, if the "cancer cell high density area" category appears more frequently, then this category is determined as the first slice category, and the subsequent decision-making process will give priority to this label. This setting can effectively reduce the computational complexity and provide a reliable basis for target classification.

[0110] The number of slice verification modules can be understood as a record of the number of verification modules involved in the pathological image slice verification task, and its role is to provide contextual support for classification decisions. For example, if there are 5 verification modules in the system, the number of slice verification modules is 5. The number of verification modules can also be used to dynamically adjust the classification strategy. For example, increasing the number of modules can increase the robustness of the classification and reduce the probability of misjudgment.

[0111] The verification category determination threshold can be understood as a classification decision threshold set according to the verification module quantity information, and its role is to define the determination condition of the target slice category to ensure that the acquired target slice category is a credible category. Usually, the verification category determination threshold is set to half the number of slice modules. For example, when the slice verification module quantity information is 5, the threshold is set to 3. When a slice category is supported by at least 3 modules in the verification result, the category will be determined as the target slice category.

[0112] It should be noted that when the number corresponding to each first slice category is less than or equal to the verification category determination threshold, it means that each slice verification module cannot determine the category of the slice, and further means that the target slice category corresponding to the second pathological image slice cannot be determined.

[0113] By determining the verification category determination threshold and determining the target slice category according to the verification category determination threshold, the accuracy and credibility of the classification can be flexibly controlled to adapt to application requirements in different scenarios, such as high-precision diagnosis or rapid initial screening tasks.

[0114] Step 108: training the cell annotation model according to the initial slice category and the target slice category corresponding to each pathological image slice.

[0115] In practical applications, training a cell annotation model according to the initial slice category and target slice category corresponding to each pathological image slice can be understood as optimizing the classification and detection performance of the cell annotation model by comparing the information of the initial category and the target category. Its significance lies in improving the prediction accuracy of the model for cell category and position information through iterative learning, thereby meeting higher application requirements. The specific method can be to screen pathological image slices whose initial category is consistent with the target category as sample data, and use these credible samples to supervise the learning of the cell annotation model; it can also introduce a reinforcement learning method, use the matching degree between the initial category and the target category as a reward signal, and dynamically adjust the model parameters to ensure category consistency; it can also adopt a self-supervised learning strategy, by constructing a contrast loss function, to strengthen the model's mapping ability between the initial category and the target category, further optimize the annotation results, etc. This manual does not impose any restrictions on this.

[0116] By training the cell annotation model with the initial slice categories generated by the cell annotation model and the target slice categories generated by the slice verification model, the verification results can be effectively used to optimize the model parameters and improve the accuracy and reliability of the prediction of cell categories and location information. This training method can dynamically adjust the model's ability to recognize different slice features, making it better adaptable to diverse pathological image samples, while reducing classification bias, and providing high-quality basic support for subsequent large-scale annotation tasks and precise analysis. This not only improves training efficiency, but also ensures the robustness and wide applicability of the model in practical applications.

[0117] Furthermore, the cell annotation model is trained according to the initial slice category and the target slice category corresponding to each pathological image slice, including:

[0118] Determining at least one sample pathology image slice in each pathology image slice according to the initial slice category and the target slice category corresponding to each pathology image slice;

[0119] The cell annotation model is trained based on the pathological image slices of each sample.

[0120] In practical applications, sample pathology image slices are high-confidence slices selected from all pathology image slices. They usually meet the condition that the initial slice category is consistent with the target slice category, and are used to train the cell annotation model and optimize its performance. Sample pathology image slices can be understood as a type of pathology image slice that has been screened and used for model training. Its role is to provide high-quality input data for the model to ensure the accuracy and efficiency of the training process. For example, in a set of pathology image slices, if the initial classification results of certain slices (such as "high-density cancer cell areas") are completely consistent with the target classification results, these slices can be identified as sample pathology image slices and used to further train the cell annotation model.

[0121] By using sample pathology image slices for training, the model can gradually optimize its ability to predict cell categories and location information, thereby improving its adaptability in real scenarios. In addition, sample pathology image slices can also be used to verify model performance, such as evaluating the model's prediction effect on unseen data through cross-validation methods to ensure that the model has good generalization capabilities.

[0122] Further, determining at least one sample pathology image slice in each pathology image slice according to the initial slice category and the target slice category corresponding to each pathology image slice includes:

[0123] Determine a third pathological image slice, wherein the third pathological image slice is any one of the pathological image slices;

[0124] In a case where the initial slice category corresponding to the third pathological image slice is the same as the target slice category, the third pathological image slice is determined to be a sample pathological image slice.

[0125] In practical applications, pathological image slices whose initial slice categories are the same as the target slice categories are determined as sample pathological image slices. This can be understood as screening out highly reliable pathological image slices as sample data to provide reliable input for subsequent cell labeling model training, thereby significantly improving the classification accuracy and robustness of the model.

[0126] refer to Figure 2 , Figure 2 This is a schematic diagram of a cell annotation model training process provided by an embodiment of the present specification. First, the target pathological image is divided into multiple pathological image slices, and then each pathological image slice is input into the cell annotation model and the slice verification model respectively, and the cell category corresponding to each pathological image slice determined by the slice verification model and the cell category determined based on the pathological annotation model are obtained respectively, and it is determined whether the two cell categories corresponding to each pathological image slice are the same, and the same pathological image slice is used as a sample pathological image, and then the cell annotation model is trained according to the sample pathological image.

[0127] By reducing the interference of low-quality data, the training process is ensured to be more targeted and efficient, while optimizing the performance of the model in real application scenarios, laying a solid foundation for large-scale pathological image analysis and cell labeling tasks. In addition, iterative optimization using sample pathological image slices can further improve the adaptability and generalization ability of the model, effectively meeting the analysis needs of different types of pathological images.

[0128] Furthermore, the cell annotation model is trained based on the pathological image slices of each sample, including:

[0129] Based on the initial slice category and the target slice category corresponding to each sample pathological image slice, obtaining the sample slice category corresponding to each sample pathological image slice;

[0130] Inputting each sample pathological image slice into a cell annotation model, and obtaining predicted cell category information and predicted cell position information corresponding to each cell entity in each sample pathological image slice generated by the cell annotation model;

[0131] Determine the predicted slice category corresponding to each sample pathology image slice according to the predicted cell category information and predicted cell position information corresponding to each cell entity in each sample pathology image slice;

[0132] The cell annotation model is trained based on the sample slice categories and predicted slice categories corresponding to each sample pathological image slice.

[0133] In practical applications, the sample slice category is a reference classification label generated based on the initial slice category and target slice category of the sample pathology image slice, which is used to guide the training of the model; the predicted slice category is a classification label automatically generated by the cell annotation model based on the input data, which is used to evaluate the performance of the model on the sample.

[0134] The sample slice category can be understood as a high-confidence label determined based on the initial classification and target classification results. Its role is to provide accurate and reliable guidance data for model training. For example, in a set of sample pathology image slices, if the initial category of a slice is "cancer cell area" and the target category is "cancer cell high density area", then the sample category of the slice can be determined as "cancer cell high density area" through rule fusion. This method can not only reduce the interference of inconsistent labels on model training, but also improve the model's learning effect on different pathological features, providing better adaptability for subsequent classification tasks.

[0135] The predicted slice category can be understood as the label automatically generated by the cell annotation model based on the slice input. Its role is to evaluate the model's ability to classify sample data and its understanding of pathological features. For example, after the model is trained, a pathological image slice is input into the model, and the predicted category output by the model may be "high-density cancer cell area". By comparing with the sample slice category, the classification accuracy and robustness of the model can be quantified, potential problems can be discovered, and the model can be further optimized. This process ensures the reliability of the model in real application scenarios, while providing automated support for the classification of new data.

[0136] It should be noted that, based on the initial slice category and target slice category corresponding to each sample pathology image slice, obtaining the sample slice category corresponding to each sample pathology image slice can be understood as generating a high-confidence sample classification label by integrating the initial classification results and the target classification requirements. Its significance lies in providing accurate and reliable data input for subsequent model training, thereby improving the training effect and the adaptability of the model. The specific method can be to directly compare the initial slice category and the target slice category. When the two are consistent, the category is set as the sample slice category; it can also be introduced by introducing a prompt word template, splicing the initial slice category and the target slice category information and inputting them into the slice verification module to generate a more accurate sample slice category; it can also be based on a rule fusion method, such as setting weights, calculating the initial slice category and the target slice category in proportion, and finally outputting the sample slice category, etc. This manual does not impose any restrictions on this.

[0137] It should be noted that, in order to ensure the effectiveness of model training, the method of determining the predicted slice category based on the category and position of the cell is the same as the method of obtaining the initial slice category mentioned above.

[0138] By applying the solution of the embodiments of this specification, by acquiring the target pathological image and generating pathological image slices containing cell entities, the labeling of cell categories and cell location information can be automatically realized, and the slice categories can be verified and optimized through the slice verification model, which significantly improves the accuracy and consistency of the labeling. The collaborative use of the cell labeling model and the slice verification model can dynamically adjust the labeling results to make the labeling rules more flexible and more adaptable. In addition, the iterative training process of sample slices is used to effectively optimize the model performance, while reducing the cost of large-scale cell labeling, and realizing fast and efficient processing and analysis of target images. This labeling method not only improves the degree of automation, but also meets the rapidly growing labeling needs in scientific research and clinical applications, while ensuring the efficiency and reliability of data updates.

[0139] Corresponding to the above method embodiment, this specification also provides a cell labeling method embodiment, see Figure 3 , Figure 3 A flow chart of a cell labeling method provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0140] Step 302: Acquire a pathological image to be processed, wherein the pathological image to be processed includes at least one cell entity.

[0141] Step 304: input the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method.

[0142] Step 306: Generate a cell annotation result corresponding to the pathological image to be processed according to the cell category information and cell position information corresponding to each cell entity.

[0143] In practical applications, the pathological images to be processed are pathological image data that have not yet been analyzed and processed. They usually come from pathological scanners or panoramic pathological images and are the input basis for pathological analysis tasks. The cell annotation results are the output information obtained after classifying and locating each cell entity in the pathological image, including the cell category label and spatial position information, which are important data to support pathological research and diagnosis.

[0144] The unprocessed pathological image can be understood as a kind of unprocessed pathological image data, which is mainly used as the input of the subsequent image segmentation, feature extraction and cell labeling model. This image data is usually obtained from the glass slide sample through a digital scanning device, has extremely high resolution, and can fully display the structure and details of the pathological tissue. Figure 4a As shown, Figure 4a It is a schematic diagram of a target pathological image provided by an embodiment of the present specification, which is a scan result of a whole piece of diseased tissue in cancer diagnosis.

[0145] The cell annotation results can be understood as the classification and location information of each cell entity in the pathological image through the model, which is a structured expression of the cell information in the image. Figure 4b As shown, Figure 4b This is a schematic diagram of a cell labeling result provided by an embodiment of the present specification, which is a picture of a scan result of a whole piece of diseased tissue in cancer diagnosis, in which different colors are used to mark cancer cells, immune cells and their specific locations and distribution patterns in the slice.

[0146] The above is a schematic scheme of a cell labeling method of this embodiment. It should be noted that the technical scheme of the cell labeling method and the technical scheme of the above-mentioned cell labeling model training method belong to the same concept, and the details not described in detail in the technical scheme of the cell labeling method can be referred to the description of the technical scheme of the above-mentioned cell labeling model training method.

[0147] By applying the solution of the embodiments of this specification, the pathological images to be processed are automatically labeled with cell entities, and the cell category information and cell location information generated by the cell annotation model are combined to achieve high accuracy and consistency of the annotation results. By using the cell annotation model optimized based on the training method, it is possible to significantly improve efficiency when processing large-scale pathological images and reduce the time and cost required for manual participation. In addition, by directly generating annotation results corresponding to cell entities, the overall process from image input to annotation output is optimized, ensuring the rapid responsiveness and flexible adaptability of the annotation process. This technology effectively meets the rapidly growing needs of scientific research and clinical annotation, supports efficient data processing and high-quality analysis, while enhancing the scalability of data in dynamically updated scenarios, and provides reliable technical support for annotation tasks in medical research and practice.

[0148] Corresponding to the above method embodiment, this specification also provides a computer-aided diagnosis method embodiment of a tumor, see Figure 5 , Figure 5 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.

[0149] Step 502: Acquire a pathological image to be processed, wherein the pathological image to be processed includes at least one cell entity.

[0150] Step 504: input the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method.

[0151] Step 506: Generate an image detection result corresponding to the pathological image to be processed according to the cell category information and cell position information corresponding to each cell entity.

[0152] In practical applications, image detection results are target analysis results generated based on cell category and location information in pathological images, which are used to provide specific conclusions on disease status, prognosis information or other pathological features, and are important outputs for diagnosis and scientific research. Image detection results can be understood as structured results generated by analyzing cell distribution and feature information in pathological images to support diagnostic decisions or scientific research conclusions. For example, in computer-aided diagnosis of cancer, image detection results may include the area of ​​the tumor region, cell classification statistics (such as the ratio of cancer cells to normal cells), and information on whether there are metastatic lesions. This result can not only help doctors quickly assess disease progression, but can also be used to design personalized treatment plans. In addition, the detection results may be further aggregated into global pathological image labels, such as whether specific treatment is needed or possible prognostic information, thereby providing clearer guidance for clinical practice.

[0153] 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 cell labeling 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 all be referred to the description of the technical scheme of the above-mentioned cell labeling method.

[0154] By applying the solution of the embodiments of this specification, by obtaining the pathological image to be processed and extracting the cell entity information contained therein, it is possible to automatically label the cell category and cell location information, reducing the time and manpower investment in the traditional manual labeling process, while significantly improving the labeling efficiency and accuracy. Combined with the use of the cell labeling model, high-quality image detection results can be quickly generated to ensure the consistency and reliability of the labeling. In addition, through the pre-training and optimization of the labeling model, it has dynamic adaptability, can maintain good performance in a variety of pathological image scenes, and meet the needs of different pathological analysis tasks. The application of this method not only accelerates the processing of pathological images, but also provides support for the intelligent management of large-scale pathological data, which helps to promote the widespread application of computer-aided diagnosis technology in clinical and scientific research fields.

[0155] 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 6 , Figure 6 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.

[0156] Step 602: Acquire a pathological image to be processed, wherein the pathological image to be processed includes at least one cell entity.

[0157] Step 604: input the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method.

[0158] Step 606: Generate an image detection result corresponding to the pathological image to be processed according to the cell category information and cell position information corresponding to each cell entity.

[0159] 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.

[0160] By applying the solution of the embodiments of this specification, by inputting the pathological image to be processed into the cell annotation model deployed in the client, efficient extraction of cell category and location information in the image is achieved. By directly generating image detection results, the diagnosis process can be effectively shortened, and accurate auxiliary information can be quickly provided to doctors, reducing the impact of diagnostic efficiency caused by data transmission or delays. In addition, the application of this method on the client side makes the annotation and detection process more flexible, able to adapt to different hardware conditions and application scenarios, and provides reliable support for pathological analysis in edge computing environments, promoting the development of computer-aided diagnosis technology in a distributed and personalized direction.

[0161] 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 7 , Figure 7 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.

[0162] Step 702: receiving a tumor diagnosis request sent by a terminal device, wherein the tumor diagnosis request includes a pathological image to be processed, and the pathological image to be processed includes at least one cell entity.

[0163] Step 704: input the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method.

[0164] Step 706: Generate an image detection result corresponding to the pathological image to be processed according to the cell category information and cell position information corresponding to each cell entity, and send the image detection result to the terminal side device.

[0165] 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.

[0166] By applying the solution of the embodiments of this specification, efficient collaboration and automation of remote pathology analysis can be achieved by receiving tumor diagnosis requests sent by the end-side device on the cloud device and processing the pathological images to be processed. The cell annotation model is used to accurately classify and locate cell entities in pathological images, which not only improves the speed and accuracy of data processing, but also ensures the consistency and high quality of the test results. Feeding back the generated image detection results directly to the end-side device helps to realize the instant sharing of diagnostic information and reduce the time delay in the traditional analysis process. At the same time, based on the computing power of cloud devices, it can process large-scale pathological data and support more complex model optimization and task distribution, so that excellent performance can still be maintained in multi-device collaboration and high-concurrency tasks. The application of this diagnostic method broadens the scope of computer-aided diagnosis scenarios and provides strong technical support for telemedicine and intelligent pathological analysis.

[0167] Corresponding to the above method embodiment, this specification also provides a cell labeling method embodiment applied to a cloud device, see Figure 8 , Figure 8 A flow chart of a cell labeling method applied to a cloud device according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0168] Step 802: receiving a cell annotation request sent by a terminal device, wherein the cell annotation request includes a pathological image to be processed, and the pathological image to be processed includes at least one cell entity.

[0169] Step 804: input the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method.

[0170] Step 806: Generate a cell annotation result corresponding to the pathological image to be processed according to the cell category information and cell position information corresponding to each cell entity, and send the cell annotation result to the terminal device.

[0171] The above is a schematic scheme of a cell labeling method applied to a cloud device in this embodiment. It should be noted that the technical scheme of the cell labeling method applied to the cloud device belongs to the same concept as the technical scheme of the cell labeling method described above. For details not described in detail in the technical scheme of the cell labeling method applied to the cloud device, please refer to the description of the technical scheme of the cell labeling method described above.

[0172] By applying the solution of the embodiment of this specification, by receiving the cell annotation request sent by the end-side device and inputting the pathological image to be processed into the cell annotation model, the cell category information and cell location information can be quickly obtained, thereby realizing the accurate annotation of each cell entity in the pathological image. This method not only greatly improves the annotation efficiency through the automatic generation and feedback of cell annotation results, but also ensures the accuracy and consistency of the annotation results. Relying on the efficient computing power of cloud devices, it can quickly respond to the requests of the end-side devices, realizing the real-time and efficiency of the annotation process. In addition, the dynamic optimization characteristics of the cell annotation model make it highly adaptable and can meet the cell annotation needs in a variety of complex scenarios. At the same time, it reduces the time cost and resource overhead of large-scale annotation, providing flexible and reliable technical support for scientific research and clinical scenarios.

[0173] See also Fig. 9 , Fig. 9 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;

[0174] The client 100 is used to send a tumor diagnosis request to the server 200, wherein the tumor diagnosis request carries a pathological image to be processed, and the pathological image to be processed includes at least one cell entity;

[0175] The server 200 is used to input the pathological image to be processed into the cell annotation model, obtain the cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the above-mentioned cell annotation model training method; generate the image detection result corresponding to the pathological image to be processed according to the cell category information and cell location information corresponding to each cell entity; and send the image detection result to the client 100;

[0176] The client 100 is also used to receive the image detection result sent by the server 200.

[0177] By applying the solution of the embodiments of this specification, by inputting the cell entity information in the pathological image to be processed into the cell annotation model, the cell category information and cell location information can be quickly obtained, effectively improving the automation and efficiency of cell annotation. Combined with the collaborative work of the client and the server, the diagnostic process can be dynamically optimized to ensure the accuracy and consistency of the annotation results. The real-time transmission and feedback mechanism of the image detection results is used to significantly shorten the processing cycle of pathological images, meeting the needs for rapid diagnosis in clinical and scientific research scenarios. At the same time, through the training and iterative update of the cell annotation model by the server, the model performance can be continuously optimized to adapt to the changing annotation needs and reduce the overall cost of large-scale annotation. This diagnostic system architecture not only improves the accuracy of annotation and analysis, but also enhances the adaptability to new pathological images, helping the development of precision medicine and pathology research.

[0178] 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.

[0179] 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 detection result based on the data stream and pushes the image detection result to other clients that have established communication.

[0180] 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.

[0181] 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.

[0182] 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 De l ivery Network), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0183] 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.

[0184] The following combination Fig.10Taking the application of the cell annotation model training method provided in this specification in computer-assisted prognosis prediction as an example, the cell annotation model training method is further described. Fig.10 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.

[0185] Step 1002: Segment each pathological image in the pathological image library into a plurality of pathological image slices in sequence.

[0186] Step 1004: Input the multiple pathological image slices obtained above into the cell annotation model, obtain the cell category information and cell position information of each cell entity in the slice image generated by the cell annotation model, and count the cell category information and cell position information of each cell entity in the above slice to analyze and obtain the initial slice category corresponding to each pathological image slice.

[0187] Step 1006: splice the multiple pathological image slices obtained above with the preset slice verification prompt word template, and input the spliced ​​information into the slice verification model to obtain the target slice category corresponding to each pathological image slice output by the slice verification model.

[0188] Step 1008: Pathological image slices with the same initial slice category and target slice category are used as sample images, and the corresponding slice category is used as the sample category.

[0189] Step 1010: further training the cell annotation model based on the sample pathology image slices composed of the sample images and sample categories.

[0190] Step 1012: Obtain a pathological image that needs to be analyzed for prognosis and input it into the trained cell annotation model to obtain the position information of each cell entity in the pathological image and the category information of each cell entity.

[0191] Step 1014: Obtain the prognostic analysis result corresponding to the sample pathological image based on the location information of the cell entity and the category information of each cell entity.

[0192] By applying the solution of the embodiments of this specification, by segmenting the pathological image into multiple pathological image slices containing cell entities, and using the cell annotation model to analyze the cell category and position information in the slice, the initial annotation results can be automatically generated to ensure the efficiency and consistency of large-scale annotation tasks. By combining with the slice verification prompt word template, the initial slice category and the output category of the slice verification model are dynamically compared and optimized, which not only improves the accuracy of the annotation results, but also enhances the adaptability of the annotation rules to different data environments. During the training process, the performance of the cell annotation model is further optimized using sample pathological image slices with consistent categories, achieving iterative improvement of the annotation system and reducing the overall annotation cost. At the same time, combined with the location information and category information generated by the cell annotation model, the prognostic analysis results of the sample pathological image can be quickly obtained, providing an efficient and accurate solution for complex image processing needs in scientific research and clinical practice, and promoting the in-depth application of pathological image data.

[0193] Corresponding to the above method embodiment, this specification also provides a cell annotation model training device embodiment, Fig.11 FIG. 2 shows a schematic diagram of a cell labeling model training device provided by an embodiment of the present specification. Fig.11 As shown, the device comprises:

[0194] The slicing module 1102 is configured to acquire a target pathological image and generate at least one pathological image slice based on the target pathological image, wherein the pathological image slice includes at least one cell entity;

[0195] The initial classification module 1104 is configured to input each pathological image slice into the cell annotation model, obtain 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 determine 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;

[0196] The target classification module 1106 is configured to input each pathological image slice into a slice verification model, and obtain a target slice category corresponding to each pathological image slice generated by the slice verification model;

[0197] The training module 1108 is configured to train the cell annotation model according to the initial slice category and the target slice category corresponding to each pathological image slice.

[0198] Optionally, the cell annotation model includes a cell positioning module and a cell classification module;

[0199] The initial classification module 1104 is further configured to:

[0200] Determine a first pathological image slice, wherein the first pathological image slice is any one of the pathological image slices;

[0201] Inputting the first pathological image slice into the cell positioning module to obtain cell position information corresponding to each cell entity in the first pathological image slice generated by the cell positioning module;

[0202] The first pathological image slice and each cell position information are processed according to the cell classification module to obtain cell category information corresponding to each cell entity.

[0203] Optionally, the initial classification module 1104 is further configured to:

[0204] Determine the cell image corresponding to each cell entity based on the position information of each cell in the first pathological image slice;

[0205] Each cell image is input into the cell classification module to obtain the cell category information corresponding to each cell entity generated by the cell classification module.

[0206] Optionally, the initial classification module 1104 is further configured to:

[0207] Determine the cell distribution characteristic information 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;

[0208] Based on the cell distribution characteristic information corresponding to each pathological image slice, the initial slice category corresponding to each pathological image slice is determined.

[0209] Optionally, the slice verification model includes at least one slice verification module;

[0210] The target classification module 1106 is further configured to:

[0211] Determine a second pathological image slice, wherein the second pathological image slice is any one of the pathological image slices;

[0212] Inputting the second pathological image slice into each slice verification module to obtain the reference slice category generated by each slice verification module;

[0213] A target slice category corresponding to the second pathological image slice is determined according to each reference slice category.

[0214] Optionally, the reference slice category includes a first slice category and a second slice category;

[0215] The target classification module 1106 is further configured to:

[0216] determining a first slice category among the reference slice categories;

[0217] Acquire slice check module quantity information, and determine a check category determination threshold according to the slice check module quantity information;

[0218] When the number corresponding to each first slice category is greater than the verification category determination threshold, it is determined that the target slice category corresponding to the second pathological image slice is the first slice category.

[0219] Optionally, the training module 1108 is further configured to:

[0220] Determining at least one sample pathology image slice in each pathology image slice according to the initial slice category and the target slice category corresponding to each pathology image slice;

[0221] The cell annotation model is trained based on the pathological image slices of each sample.

[0222] Optionally, the training module 1108 is further configured to:

[0223] Determine a third pathological image slice, wherein the third pathological image slice is any one of the pathological image slices;

[0224] In a case where the initial slice category corresponding to the third pathological image slice is the same as the target slice category, the third pathological image slice is determined to be a sample pathological image slice.

[0225] Optionally, the training module 1108 is further configured to:

[0226] Based on the initial slice category and the target slice category corresponding to each sample pathological image slice, obtaining the sample slice category corresponding to each sample pathological image slice;

[0227] Inputting each sample pathological image slice into a cell annotation model, and obtaining predicted cell category information and predicted cell position information corresponding to each cell entity in each sample pathological image slice generated by the cell annotation model;

[0228] Determine the predicted slice category corresponding to each sample pathology image slice according to the predicted cell category information and predicted cell position information corresponding to each cell entity in each sample pathology image slice;

[0229] The cell annotation model is trained based on the sample slice categories and predicted slice categories corresponding to each sample pathological image slice.

[0230] By applying the solution of the embodiments of this specification, through the processing of pathological image slices, the cell annotation model and the slice verification model work together to achieve accurate annotation of the cell category and position in the target pathological image, and at the same time, by verifying and optimizing the slice category, the annotation results are more consistent and accurate. The introduction of the slice verification module helps to dynamically adjust and optimize the annotation rules during the classification process to ensure its strong adaptability and scalability. In addition, the screening and iterative training process of sample pathological image slices further improves the model performance, reduces the cost and complexity of large-scale annotation, and speeds up the speed and efficiency of pathological image analysis. This method can fully meet the rapidly growing scientific research and clinical needs, while reducing human intervention, improving the degree of annotation automation, and supporting the rapid updating and management of the database, promoting the sustainable development of pathological research and clinical applications.

[0231] The above is a schematic scheme of a cell labeling model training device of this embodiment. It should be noted that the technical scheme of the cell labeling model training device and the technical scheme of the cell labeling model training method mentioned above belong to the same concept, and the details not described in detail in the technical scheme of the cell labeling model training device can be referred to the description of the technical scheme of the cell labeling model training method mentioned above.

[0232] Fig.12 The block diagram of a computing device 1200 according to one embodiment of the present specification is shown. The components of the computing device 1200 include but are not limited to a memory 1210 and a processor 1220. The processor 1220 is connected to the memory 1210 via a bus 1230, and the database 1250 is used to store data.

[0233] The computing device 1200 also includes an access device 1240 that enables the computing device 1200 to communicate via one or more networks 1260. 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 1240 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 802.11 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).

[0234] In one embodiment of the present specification, the above components of the computing device 1200 and Fig.12 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.12 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.

[0235] The computing device 1200 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 1200 may also be a mobile or stationary server.

[0236] Among them, the processor 1220 is used to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned cell labeling model training method, cell labeling method and computer-aided diagnosis method of tumors.

[0237] 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 cell annotation model training method, cell annotation 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 cell annotation model training method, cell annotation method and computer-aided diagnosis method of tumors.

[0238] 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 cell labeling model training method, cell labeling method, and computer-aided diagnosis method for tumors.

[0239] 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 cell annotation model training method, cell annotation 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 cell annotation model training method, cell annotation method and computer-aided diagnosis method of tumors.

[0240] 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 cell annotation model training method, cell annotation method, and computer-aided diagnosis method for tumors.

[0241] 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 cell annotation model training method, cell annotation 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 cell annotation model training method, cell annotation method and computer-aided diagnosis method of tumors.

[0242] 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.

[0243] 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.

[0244] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed 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.

[0245] 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.

[0246] 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 cell labeling model training method, comprising: Acquire a target pathological image, and generate at least one pathological image slice based on the target pathological image, wherein the pathological image slice includes at least one cell entity; Input each pathological image slice into a cell annotation model, obtain cell category information and cell position information corresponding to each cell entity in each pathological image slice generated by the cell annotation model, and determine 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 to obtain a target slice category corresponding to each pathological image slice generated by the slice verification model; The cell annotation model is trained according to the initial slice category and the target slice category corresponding to each pathological image slice.

2. The method according to claim 1, wherein the cell annotation model comprises a cell positioning module and a cell classification module; Inputting each pathological image slice into the cell annotation model, obtaining the cell category information and cell location information corresponding to each cell entity in each pathological image slice generated by the cell annotation model, including: Determine a first pathological image slice, wherein the first pathological image slice is any one of the pathological image slices; Inputting the first pathological image slice into the cell positioning module to obtain cell position information corresponding to each cell entity in the first pathological image slice generated by the cell positioning module; The first pathological image slice and each cell position information are processed according to the cell classification module to obtain cell category information corresponding to each cell entity.

3. The method according to claim 2, wherein the cell classification module processes the first pathological image slice and each cell position information to obtain the cell category information corresponding to each cell entity, comprising: Determine the cell image corresponding to each cell entity based on the position information of each cell in the first pathological image slice; Each cell image is input into the cell classification module to obtain the cell category information corresponding to each cell entity generated by the cell classification module.

4. The method according to claim 1, 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, comprising: Determine the cell distribution characteristic information 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; Based on the cell distribution characteristic information corresponding to each pathological image slice, the initial slice category corresponding to each pathological image slice is determined.

5. The method of claim 1, wherein the slice verification model comprises at least one slice verification module; Inputting each pathological image slice into a slice verification model, and obtaining a target slice category corresponding to each pathological image slice generated by the slice verification model, including: Determine a second pathological image slice, wherein the second pathological image slice is any one of the pathological image slices; Inputting the second pathological image slice into each slice verification module to obtain the reference slice category generated by each slice verification module; A target slice category corresponding to the second pathological image slice is determined according to each reference slice category.

6. The method according to claim 5, determining the target slice category corresponding to the second pathological image slice according to each reference slice category, comprising: determining a first slice category among the reference slice categories; Acquire slice check module quantity information, and determine a check category determination threshold according to the slice check module quantity information; When the number corresponding to each first slice category is greater than the verification category determination threshold, it is determined that the target slice category corresponding to the second pathological image slice is the first slice category.

7. The method according to claim 1, training the cell annotation model according to the initial slice category and the target slice category corresponding to each pathological image slice, comprises: Determining at least one sample pathology image slice in each pathology image slice according to the initial slice category and the target slice category corresponding to each pathology image slice; The cell annotation model is trained based on the pathological image slices of each sample.

8. The method according to claim 7, determining at least one sample pathology image slice in each pathology image slice according to the initial slice category and the target slice category corresponding to each pathology image slice, comprising: Determine a third pathological image slice, wherein the third pathological image slice is any one of the pathological image slices; In a case where the initial slice category corresponding to the third pathological image slice is the same as the target slice category, the third pathological image slice is determined to be a sample pathological image slice.

9. The method according to claim 7, wherein training the cell annotation model based on each sample pathological image slice comprises: Based on the initial slice category and the target slice category corresponding to each sample pathological image slice, obtaining the sample slice category corresponding to each sample pathological image slice; Inputting each sample pathological image slice into a cell annotation model, and obtaining predicted cell category information and predicted cell position information corresponding to each cell entity in each sample pathological image slice generated by the cell annotation model; Determine the predicted slice category corresponding to each sample pathology image slice according to the predicted cell category information and predicted cell position information corresponding to each cell entity in each sample pathology image slice; The cell annotation model is trained based on the sample slice categories and predicted slice categories corresponding to each sample pathological image slice.

10. A cell labeling method, comprising: Acquiring a pathological image to be processed, wherein the pathological image to be processed includes at least one cell entity; Inputting the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the cell annotation model training method according to any one of claims 1 to 9; According to the cell category information and cell position information corresponding to each cell entity, a cell annotation result corresponding to the pathological image to be processed is generated.

11. A computer-aided diagnosis method for tumors, comprising: Acquiring a pathological image to be processed, wherein the pathological image to be processed includes at least one cell entity; Inputting the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the cell annotation model training method according to any one of claims 1 to 9; According to the cell category information and cell position information corresponding to each cell entity, an image detection result corresponding to the pathological image to be processed is generated.

12. A computer-aided diagnosis method for tumors, applied to a client, comprising: Acquiring a pathological image to be processed, wherein the pathological image to be processed includes at least one cell entity; Inputting the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the cell annotation model training method according to any one of claims 1 to 9; According to the cell category information and cell position information corresponding to each cell entity, an image detection result corresponding to the pathological image to be processed is generated.

13. 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 includes a pathological image to be processed, and the pathological image to be processed includes at least one cell entity; Inputting the pathological image to be processed into a cell annotation model to obtain cell category information and cell location information corresponding to each cell entity generated by the cell annotation model, wherein the cell annotation model is trained based on the cell annotation model training method according to any one of claims 1 to 9; According to the cell category information and cell position information corresponding to each cell entity, an image detection result corresponding to the pathological image to be processed is generated, and the image detection result is sent to the terminal side device.

14. 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 according to any one of claims 1 to 13 are implemented.

15. 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 13.

16. 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 13.

Images