A method and system for cervical cell identification

Abstract

The application provides a cervical cell recognition method and system, and the method comprises the following steps: recognizing qualified field regions in a to-be-detected pathological image; extracting contour coordinate information of each to-be-detected cervical cell in the qualified field regions, segmenting and extracting a feature vector of each to-be-detected cervical cell; and based on the feature vector of each to-be-detected cervical cell, querying in a query set to obtain the category of each to-be-detected cervical cell. The application analyzes the image quality of each field region in the to-be-detected pathological image, thereby ensuring the reliability of the final recognition result output and avoiding prediction deviation and errors caused by low-quality data; each cervical cell in the qualified field region is recognized, thereby ensuring that the main to-be-detected target is not missed, learning the underlying common morphological features of various cervical cells, and reducing the workload of labeling; and through the query classification mode, only a small amount of abnormal cell labeling is required compared with the traditional method, and a satisfactory recognition accuracy can be achieved.

Description

Technical Field

[0001] This invention relates to the field of pathological cytology detection, and more specifically, to a method and system for identifying cervical cells. Background Technology

[0002] Currently, cervical cancer screening methods mainly include human papillomavirus (HPV) nucleic acid testing and cytological examination. Those who test positive for HPV usually require further cytological examination for triage. The current mainstream cytological examination method is liquid-based thin-layer cytology (TCT), which has high specificity, but its sensitivity largely depends on the cytologist's skill level and experience. The lack of experienced pathological cytologists is a widespread social problem. On the other hand, testing methods that rely entirely on manual slide interpretation also suffer from inefficiency and uncontrollable quality.

[0003] The rapidly developing artificial intelligence (AI) technology in recent years has the potential to overcome the aforementioned obstacles and significantly improve the efficiency of cervical cancer screening. Based on digital microscopy, algorithmic models can quickly locate and detect abnormal cervical cells according to their unique morphology, color, karyotype, and boundary characteristics, thereby enabling further interpretation of the sample (such as normal / abnormal detection or abnormal cell category determination). As a low-cost auxiliary diagnostic tool, this technology can overcome the time constraints and shortage of technical personnel associated with manual image interpretation, and avoid bias caused by subjective factors. It is conducive to implementing community-level cervical cancer screening in resource-scarce areas and also has broad market prospects.

[0004] Currently, several companies both domestically and internationally are developing TCT AI-assisted diagnostic products, but their technical approaches are highly similar, and the industry generally suffers from insufficient generalization performance. Specifically, a common solution involves collecting multi-center TCT pathology image files for large-scale annotation to create an abnormal cell dataset, and then training a supervised recognition or segmentation model for this type of abnormal cell. However, given the complexity of cervical cytology diagnostic standards, the diversity of abnormal cell types, and the non-standardization of pathology slides, these models often experience some performance degradation on samples from other medical institutions, thus affecting the quality of clinical diagnosis (e.g., false negatives due to missed detection of important abnormal cells). Furthermore, large-scale clinical cell image annotation also incurs significant time and economic costs (acquiring high-quality annotation data sources and the long-term commitment of senior cytologists). Summary of the Invention

[0005] This invention addresses the technical problems existing in the prior art by providing a method and system for cervical cell identification, which can overcome the inefficiency of requiring a large number of annotations for model training.

[0006] According to a first aspect of the present invention, a method for identifying cervical cells is provided, comprising:

[0007] Identify qualified fields of view in the pathological images to be examined;

[0008] Extract the contour coordinate information of each cervical cell to be tested in the qualified field of view area;

[0009] Based on the contour coordinate information of each cervical cell to be tested, the image of each cervical cell to be tested is segmented from the qualified field of view area;

[0010] Extract the feature vector of each cervical cell from each cervical cell image to be tested;

[0011] Based on the feature vector of each cervical cell to be tested, a query is performed in the query set to obtain the category of each cervical cell to be tested. The query set includes feature vectors of cells of multiple categories.

[0012] The screening and identification classification of the pathological image to be tested is determined based on the category of all cervical cells in the image.

[0013] According to a second aspect of the present invention, a cervical cell recognition system is provided, comprising:

[0014] The recognition module is used to identify qualified visual areas in the pathological images to be tested.

[0015] The first extraction module is used to extract the contour coordinate information of each cervical cell to be tested in the qualified field of view area;

[0016] The segmentation module is used to segment each cervical cell image from the qualified field of view region based on the contour coordinate information of each cervical cell to be tested;

[0017] The second extraction module is used to extract the feature vector of each cervical cell from each image of the cervical cells to be tested.

[0018] The category query module is used to query a query set based on the feature vector of each cervical cell to be tested, and obtain the category of each cervical cell to be tested. The query set includes feature vectors of cells of multiple categories. Based on the categories of all cervical cells to be tested in the pathological image to be tested, the screening and identification classification of the pathological image to be tested is determined.

[0019] This invention provides a cervical cell identification method and system. By analyzing the image quality of each field of view in the pathological image to be detected, the reliability of the final identification result output is ensured, avoiding prediction bias and errors caused by low-quality data. Identifying each cell in the qualified field of view ensures that the main targets to be tested are not missed. By learning the underlying common morphological features of various types of cervical cells, the workload of annotation is reduced. Through query classification, only a relatively small number of abnormal cells need to be annotated to achieve a satisfactory diagnostic accuracy compared to traditional methods. Attached Figure Description

[0020] Figure 1 A flowchart of a cervical cell identification method provided in an embodiment of the present invention;

[0021] Figure 2 This is a schematic diagram of the structure of a cervical cell recognition system provided in an embodiment of the present invention;

[0022] Figure 3 A schematic diagram of the hardware structure of a possible electronic device provided by the present invention;

[0023] Figure 4 This is a schematic diagram of the hardware structure of a possible computer-readable storage medium provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined with each other to form feasible technical solutions. Such combinations are not constrained by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0025] Figure 1 A flowchart of a cervical cell identification method provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method includes:

[0026] Step 1: Identify the qualified field of view in the pathological image to be tested.

[0027] In one embodiment of the present invention, identifying qualified visual field regions in a pathological image to be detected includes: dividing the pathological image to be detected into multiple visual field regions according to a fixed-size sliding window; analyzing the image quality of each visual field region based on an image quality control model to obtain a quality control result for each visual field region, wherein the quality control result includes qualified image quality and unqualified image quality of the visual field region.

[0028] Specifically, the pathological image to be tested is divided into multiple image blocks using a fixed sliding window. Each image block represents a field of view. Based on a trained image quality control model, the image quality of each field of view is analyzed to identify the fields of view with acceptable image quality and those with unacceptable image quality. Subsequently, only the fields of view with acceptable image quality are processed.

[0029] The process of constructing the image quality control model is as follows: High-resolution pyramid pathological image data of TCT samples (such as TMAP, SVS, or TIFF formats, ensuring sufficient richness in slide imaging and covering real-world usage scenarios, but without specific requirements for sample identification and classification) are collected. A fixed-size sliding window patch (1280*1280 pixels in this embodiment) is used to traverse the specified magnification imaging layers (20X in this embodiment) horizontally and vertically to obtain the patch data set D1, which includes a large number of image patches. Unqualified samples with significant image quality problems (such as blurry scans, large areas of dirt, etc., severely affecting diagnosis) are manually screened from the remaining clear patch samples. Then, positive samples with a total quantity close to that of the unqualified samples are randomly sampled from the remaining clear patch samples to jointly construct the training set D_qc of the quality control classification model (300 positive and 300 negative patch samples in this embodiment).

[0030] Based on the obtained dataset D_qc, train the image quality control model M_qc (in this embodiment, a ResNet34 network model is used), and save the model weight file after completion.

[0031] Each visual field region of the pathological image to be detected is input into the trained image quality control model, and the quality control results of each visual field region are output to obtain the qualified and unqualified visual field regions among these visual field regions.

[0032] Step 2: Extract the contour coordinate information of each cervical cell to be tested in the qualified field of view area.

[0033] In one embodiment of the present invention, extracting the contour coordinate information of each cervical cell to be tested in the qualified field of view includes:

[0034] Based on the pan-cell recognition model, the contour coordinate information of each cervical cell to be tested in the qualified field of view is extracted.

[0035] Training the pan-cell recognition model includes:

[0036] Multiple qualified visual fields are collected, and the positional contour of each target cell in each qualified visual field is annotated using an image annotation tool. The positional contour is a polygonal contour.

[0037] Obtain a training dataset, which includes multiple training samples, each of which includes a qualified field of view and the positional contour of each target cell therein;

[0038] The pan-cell recognition model is trained based on the training dataset.

[0039] Understandably, a generalized cell recognition model is constructed and trained, and the model is used to extract the contour coordinate information of each cervical cell to be tested in each qualified field of view.

[0040] The process of constructing a pan-cell recognition model includes:

[0041] Constructing the training dataset: Reuse the D1 dataset from the previous steps, and select a portion of the samples with clear and acceptable image quality for cell-level annotation (this embodiment uses 100 examples, covering the main cell morphology categories). Specifically, use an image annotation tool (labelme in this embodiment) to accurately annotate the contour of each target cell in the image sample with polygons, obtaining the corresponding mask, i.e., the coordinate contour information of each target cell. Combine the annotated image patches and the mask to form the training dataset D_det for the general cell recognition model.

[0042] The pan-cell recognition model M_det is trained based on the obtained training dataset D_det (this embodiment uses the Mask-RCNN network model), and the model weight file is saved after completion.

[0043] The trained pan-cell recognition model is used to obtain the positional coordinates and contour information of each cervical cell to be tested in each qualified field of view in the pathological image to be tested.

[0044] Step 3: Based on the contour coordinate information of each cervical cell to be tested, segment the image of each cervical cell to be tested from the qualified field of view area.

[0045] In one embodiment of the present invention, the step of segmenting each cervical cell image from the qualified field of view based on the contour coordinate information of each cervical cell to be tested further includes: segmenting a mask of each cervical cell to be tested from the qualified field of view based on the contour coordinate information of each cervical cell to be tested, and placing the mask of each cervical cell to be tested in the center of a square white background canvas of a set size according to the morphological proportion to obtain a single cell image.

[0046] Specifically, each cervical cell image is segmented from the original sample image using its contour coordinates. These individual cells are then placed in the center of a square canvas with a white background of appropriate size, according to their morphological proportions, to obtain a single-cell image dataset D_cell. This embodiment of the invention uses contour segmentation for precise cell image segmentation rather than location bounding boxes to avoid information interference caused by adjacent or overlapping cells within the location bounding box.

[0047] Step 4: Extract the feature vector of each cervical cell from each image of the cervical cells to be tested.

[0048] In one embodiment of the present invention, the step of extracting the feature vector of each cervical cell to be tested from each image of the cervical cells to be tested includes:

[0049] Each cervical cell image to be tested is input into the self-supervised basic model to obtain the feature vector of each cervical cell.

[0050] The training process of the self-supervised basic model includes:

[0051] Images of normal cervical cells and abnormal cervical cells were acquired separately and combined to form multiple sample images;

[0052] Extract qualified field of view regions from each sample image, and segment each target cell image within each qualified field of view region as a sample target cell image;

[0053] The self-supervised basic model is trained by extracting feature vectors from the target cell images of the samples.

[0054] Specifically, after segmenting each cervical cell image to be tested from the original image, each segmented cervical cell image to be tested is input into the self-supervised basic model to obtain the feature vector of each cervical cell to be tested.

[0055] It's important to note that the training of the self-supervised base model does not require labeling; it's an unsupervised learning process that learns the feature vectors of the target cells independently. Since the dataset doesn't need labeling, it saves a significant amount of labeling time.

[0056] The training dataset for the self-supervised basic model is the dataset D_cell formed by segmenting the target cell images in step 3 to train the self-supervised model M_cell (DINOv2 is used in this embodiment). After completion, the model weight file is saved (the inference output of this model is a vector with shape [1, 1024].

[0057] Step 5: Based on the feature vector of each cervical cell to be tested, query the query set to obtain the category of each cervical cell to be tested. The query set includes feature vectors of target cells of multiple categories.

[0058] In one embodiment of the present invention, the step of querying a query set based on the feature vector of each cervical cell to be tested to obtain the category of each cervical cell to be tested, wherein the query set includes feature vectors of target cells of multiple categories, including:

[0059] Calculate the similarity between the feature vector of any cervical cell image to be tested and the feature vectors of multiple target cells of each category in the query set, and take the category with the highest similarity as the category of the cervical cell to be tested.

[0060] The process begins by constructing a query set. This involves selecting typical positive cells of each type from the target cells and combining them with a randomly sampled portion of normal cells to construct the query set Q_cell. In this embodiment, assuming there are N cell types (one type for normal cervical cells, and N-1 types for abnormal cervical cells, for a total of N types), Q_cell contains N subsets. Each type includes m cells and m cell feature vectors. For example, each query category may use the feature vectors of 100 target cells.

[0061] Calculate the similarity between the feature vector of each cervical cell to be tested and the m target cells of each category in the query set, and obtain m similarities between the cervical cell to be tested and the m cells of each category. Select m' similarities from the m similarities that are greater than the preset similarity threshold, and calculate the average of the m' similarities to obtain the average similarity between the cervical cell to be tested and the cells of each category in the query set. Select the category with the highest average similarity as the category of the cervical cell to be tested.

[0062] Step 6: Determine the screening and identification classification of the pathological image to be tested based on the category of all cervical cells to be tested in the pathological image to be tested.

[0063] Specifically, the pathological image to be tested includes multiple cervical cells to be tested. According to step 5, the category of each cervical cell to be tested is obtained, and the category with the highest degree of abnormality among all the categories of cervical cells to be tested is taken as the abnormal category of cervical cells in the pathological image to be tested.

[0064] It should be noted that for non-compliant areas in the pathological images to be tested, the category also needs to be identified according to the above model. The specific steps include:

[0065] For each unqualified field of view in the pathological image to be tested, extract the contour coordinate information of each cervical cell to be tested in the unqualified field of view;

[0066] Based on the contour coordinate information of each cervical cell to be tested, the image of each cervical cell to be tested is segmented from the unqualified field of view area;

[0067] Extract the feature vector of each cervical cell from each cervical cell image to be tested;

[0068] Based on the feature vector of each cervical cell to be tested, a query is performed in the query set to obtain the category of each cervical cell to be tested. The query set includes feature vectors of cervical cells of multiple categories.

[0069] The category of the non-compliant field of view in the pathological image to be tested is determined based on the category of all cervical cells to be tested in all non-compliant field of view regions in the pathological image to be tested.

[0070] This process identifies the categories of acceptable and unacceptable fields of view in the pathological image to be examined. Finally, the quality control results for each field of view in the pathological image to be examined, along with the location outline and abnormality category of each cervical cell to be examined, are annotated on the image for visualization.

[0071] See Figure 2 This invention provides a cervical cell identification system according to an embodiment of the invention, the system comprising:

[0072] The recognition module 201 is used to identify qualified visual field areas in the pathological image to be detected;

[0073] The first extraction module 202 is used to extract the contour coordinate information of each cervical cell to be tested in the qualified field of view area;

[0074] Segmentation module 203 is used to segment each cervical cell image from the qualified field of view region based on the contour coordinate information of each cervical cell to be tested.

[0075] The second extraction module 204 is used to extract the feature vector of each cervical cell from each image of the cervical cells to be tested.

[0076] The category query module 205 is used to query a query set based on the feature vector of each cervical cell to be tested, and obtain the category of each cervical cell to be tested. The query set includes feature vectors of cells of multiple categories. Based on the categories of all cervical cells to be tested in the pathological image to be tested, the screening and identification classification of the pathological image to be tested is determined.

[0077] It is understood that the cervical cell identification system provided by the present invention corresponds to the cervical cell identification method provided in the foregoing embodiments. The relevant technical features of the cervical cell identification system can be referred to the relevant technical features of the cervical cell identification method, and will not be repeated here.

[0078] Please see Figure 3 , Figure 3 A schematic diagram illustrating an embodiment of the electronic device provided in this invention. For example... Figure 3 As shown, an embodiment of the present invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it implements the steps of the cervical cell identification method.

[0079] Please see Figure 4 , Figure 4 This is a schematic diagram illustrating an embodiment of a computer-readable storage medium provided by the present invention. (See diagram below.) Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 411 is stored. When the computer program 411 is executed by a processor, it implements the steps of the cervical cell identification method.

[0080] This invention provides a cervical cell identification method and system. By analyzing the image quality of each field of view in the pathological image to be detected, the reliability of the final identification result is ensured, avoiding prediction bias and errors caused by low-quality data. Identifying each cell in a qualified field of view ensures that no major targets are missed. The self-supervised basic model learns the underlying common morphological features of various cervical cells without additional annotation, reducing the workload of annotation. Through query classification, satisfactory diagnostic accuracy can be achieved with only a relatively small number of abnormal cells compared to traditional methods.

[0081] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0082] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0087] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method of identifying cervical cells, characterized by, The method comprises the following steps: Identify qualified field of view regions in a to-be-detected pathological image; Extract contour coordinate information of each to-be-detected cervical cell in the qualified field of view regions; Segment each to-be-detected cervical cell image from the qualified field of view regions based on the contour coordinate information of each to-be-detected cervical cell; extract a feature vector of each to-be-detected cervical cell from each to-be-detected cervical cell image; query each to-be-detected cervical cell in a query set based on the feature vector of each to-be-detected cervical cell to obtain the category of each to-be-detected cervical cell, wherein the query set comprises feature vectors of cervical cells of multiple categories; determine the screening and identification classification of the to-be-detected pathological image according to the categories of all to-be-detected cervical cells in the to-be-detected pathological image; the identification of the qualified field of view regions in the to-be-detected pathological image comprises: dividing the to-be-detected pathological image into multiple field of view regions according to a fixed size sliding window; analyzing the image quality of each field of view region based on an image quality control model to obtain a quality control result of each field of view region, wherein the quality control result comprises that the image quality of the field of view region is qualified or unqualified; The query of each to-be-detected cervical cell in a query set based on the feature vector of each to-be-detected cervical cell to obtain the category of each to-be-detected cervical cell, wherein the query set comprises feature vectors of cells of multiple categories, comprises: calculating the similarity between the feature vector of any to-be-detected cervical cell image and the feature vectors of multiple target cells of each category in the query set, and taking the category with the highest similarity as the category of the any to-be-detected cervical cell; specifically, calculating the similarity between the feature vector of each to-be-detected cervical cell and m target cells of each category in the query set to obtain m similarities between the to-be-detected cervical cell and the m cells of each category, screening m' similarities greater than a preset similarity threshold from the m similarities, and averaging the m' similarities to obtain the average similarity between the to-be-detected cervical cell and the cells of each category in the query set, and taking the category with the highest average similarity as the category of the to-be-detected cervical cell.

2. The method according to claim 1, wherein, The extraction of the contour coordinate information of each to-be-detected cervical cell in the qualified field of view regions comprises: extracting the contour coordinate information of each to-be-detected cervical cell in the qualified field of view regions based on a general cell recognition model; wherein the training of the general cell recognition model comprises: collecting multiple qualified field of view regions, labeling the position contour of each target cell in each qualified field of view region based on an image labeling tool, and the position contour is a polygon contour; obtaining a training data set, wherein the training data set comprises multiple training samples, and each training sample comprises a qualified field of view region and the position contour of each target cell therein; training the general cell recognition model based on the training data set.

3. The method of claim 1, wherein the method further comprises: The method further includes: based on the contour coordinate information of each to-be-tested cervical cell, segmenting each to-be-tested cervical cell image from the qualified field region; and based on the contour coordinate information of each to-be-tested cervical cell, segmenting a mask of each to-be-tested cervical cell from the qualified field region, and placing the mask of each to-be-tested cervical cell in the middle of a square white background canvas of a set size according to a shape ratio to obtain a single cell image.

4. The method of claim 1, wherein the method further comprises: The method further includes: inputting each to-be-tested cervical cell image into a self-supervised basic model to obtain a feature vector of each to-be-tested cervical cell; and a training process of the self-supervised basic model includes: obtaining normal cervical cell sample images and abnormal cervical cell sample images respectively to form a plurality of sample images; segmenting each target cell image in each sample image to obtain sample target cell images; and training the self-supervised basic model based on the sample target cell images to extract feature vectors of the sample target cell images.

5. The method of claim 1, wherein the method further comprises: The method further includes: determining a screening identification classification of the to-be-tested pathological image according to the categories of all to-be-tested cervical cells in the to-be-tested pathological image.

6. The method of claim 1, wherein the method further comprises: The method further includes: extracting contour coordinate information of each to-be-tested cervical cell in each unqualified field region of the to-be-tested pathological image; segmenting each to-be-tested cervical cell image from the unqualified field region based on the contour coordinate information of each to-be-tested cervical cell; 7. The method according to claim 1 or 6, wherein the method is a method for identifying cervical cells, characterized by, extracting a feature vector of each to-be-tested cervical cell from each to-be-tested cervical cell image; 8. A cervical cell identification system, comprising: determining a screening identification classification of each unqualified field region of the to-be-tested pathological image according to the categories of all to-be-tested cervical cells in all unqualified field regions of the to-be-tested pathological image. The method further includes: after determining the screening identification classification of the to-be-tested pathological image according to the categories of all to-be-tested cervical cells in the to-be-tested pathological image, annotating a quality control result of each field region of the to-be-tested pathological image, a position contour of each to-be-tested cervical cell, and a corresponding cervical cell abnormal category on the to-be-tested pathological image, and performing visual display. The method further includes: a recognition module configured to recognize a qualified field region in a to-be-tested pathological image; a first extraction module configured to extract contour coordinate information of each to-be-tested cervical cell in the qualified field region; a segmentation module configured to segment each to-be-tested cervical cell image from the qualified field region based on the contour coordinate information of each to-be-tested cervical cell; and The second extraction module is configured to extract a feature vector of each of the to-be-tested cervical cells from each of the to-be-tested cervical cell images; The category query module is configured to query a query set based on the feature vector of each of the to-be-tested cervical cells to obtain a category of each of the to-be-tested cervical cells, wherein the query set includes feature vectors of cells of multiple categories; and determine a screening and identification category of the to-be-tested pathological image according to the categories of all the to-be-tested cervical cells in the to-be-tested pathological image; and the identifying a qualified field region in the to-be-tested pathological image includes: dividing the to-be-tested pathological image into multiple field regions according to a fixed size sliding window; and analyzing image quality of each of the field regions based on an image quality control model to obtain a quality control result of each of the field regions, wherein the quality control result includes that the field region image quality is qualified or the field region image quality is unqualified. The querying the query set based on the feature vector of each of the to-be-tested cervical cells to obtain the category of each of the to-be-tested cervical cells, wherein the query set includes feature vectors of cells of multiple categories, includes: calculating a similarity between the feature vector of any one of the to-be-tested cervical cell images and feature vectors of multiple target cells of each of the categories in the query set, and taking a category with the highest similarity as the category of the any one of the to-be-tested cervical cells; specifically, calculating a similarity between the feature vector of each of the to-be-tested cervical cells and m target cells of each of the categories in the query set to obtain m similarities between the to-be-tested cervical cell and the m cells of each of the categories; screening m' similarities greater than a preset similarity threshold from the m similarities; averaging the m' similarities to obtain an average similarity between the to-be-tested cervical cell and cells of each of the categories in the query set; and taking a category with the highest average similarity as the category of the to-be-tested cervical cell.

Images