Auxiliary screening system, application method and model of cervical liquid-based thin-layer cytology

Abstract

The present invention discloses an auxiliary screening system, application method and model of cervical liquid-based thin-layer cytology, wherein the system comprises: a collection module for collecting electronic images of massively scanned cervical liquid-based thin-layer cytology; a labeling module for identifying abnormal cells in each electronic image, outlining abnormal cells in the electronic image, and adding real labels to the abnormal cells; a training module for training the real labels corresponding to the abnormal cells to obtain a cell-level diagnostic model, and training the sheet-level diagnostic model using the information output by the cell-level diagnostic model; a diagnosis and screening module for diagnosing and screening cells using the sheet-level diagnostic model; and a diagnostic efficiency module for respectively obtaining comparative screening results, auxiliary screening results and diagnostic screening results, and respectively calculating the diagnostic efficiency of each result. The present invention can increase the recognition ability of the model, and can also further improve the recognition ability of the model, and improve the recognition accuracy and efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of model training, and in particular to an auxiliary screening system, an application method and a model for cervical liquid-based thin-layer cytology. Background Art

[0002] According to statistics, there are about 520,000 new cases of cervical cancer every year in the world, of which nearly 150,000 are in my country, and 20,000 to 30,000 women die from cervical cancer every year. In recent years, the incidence of cervical cancer in my country has been on the rise, and the age of onset is tending to be younger. According to clinical data, it takes about several years for cervical precancerous lesions to develop into cervical cancer. It is a preventable and curable disease. Therefore, it is very important to detect cervical precancerous lesions in time and stop their development into cervical cancer.

[0003] Due to the large number of patients and the lack of existing medical resources, it is even more difficult to implement early screening for cervical cancer in primary hospitals. With the development of artificial intelligence, intelligent screening using artificial intelligence diagnostic models and intelligent recognition calculations is one of the commonly used inspection methods, which can not only improve the uneven allocation of medical resources in my country, but also reduce medical costs and improve medical efficiency.

[0004] However, the artificial recognition models currently in use have the following technical problems. Due to the opacity of neural network technology, it is currently impossible to explain the results output by the artificial intelligence system, and therefore it is impossible to interpret the logical relationship within the black box of the neural network. As a result, when the recognition model faces unfamiliar data, it will reduce the recognition accuracy and efficiency. Summary of the invention

[0005] The present invention proposes an auxiliary screening system, application method and model for cervical liquid-based thin-layer cytology. The system can use labeled cells for detection and positioning, and then use the positioned cells for deep classification training to improve the accuracy of identification and screening.

[0006] A first aspect of an embodiment of the present invention provides an auxiliary screening system for cervical liquid-based thin-layer cytology, the system comprising:

[0007] A collection module, used for collecting a large number of scanned electronic images of cervical liquid-based thin-layer cytology, wherein each of the scanned electronic images includes a variety of different cells;

[0008] a labeling module, identifying abnormal cells in each of the electronic images, outlining the abnormal cells in the electronic images, and adding true labels to the abnormal cells;

[0009] A training module, used to perform model training on the real labels corresponding to the abnormal cells based on a preset deep learning algorithm to obtain a cell-level diagnostic model, and to use the information output by the cell-level diagnostic model as input information to train a slice-level diagnostic model;

[0010] A diagnosis and screening module, used for diagnosing and screening cells using the chip-level diagnosis model;

[0011] The diagnostic efficacy module is used to respectively obtain comparative screening results, auxiliary screening results and diagnostic screening results, and respectively calculate the diagnostic efficacy of the comparative screening results, the auxiliary screening results and the diagnostic screening results, wherein the comparative screening results, the auxiliary screening results and the diagnostic screening results are obtained by collecting data of batch standard patients and dividing them in equal proportions, and then performing diagnostic screening by comparative screeners, auxiliary screeners and the slice-level diagnostic model.

[0012] In a possible implementation manner of the first aspect, the marking module is further used to:

[0013] Performing image segmentation on each of the scanned electronic images to obtain a plurality of pixel-level images;

[0014] The cell shape of each pixel-level image is outlined to obtain abnormal cells, and the abnormal cells are diagnosed at the slice level and marked with real labels according to the preset TBS2014 grading reporting system.

[0015] In a possible implementation manner of the first aspect, the training module is further used to:

[0016] Inputting the abnormal cell pathology marker into a preset deep learning algorithm;

[0017] Use the preset deep learning algorithm to perform feature learning and model training on the labeled abnormal cells to obtain a cell-level diagnostic model;

[0018] The information output by the cell-level diagnostic model is used as input to the preset random forest classifier to train and obtain the slice-level diagnostic model.

[0019] In a possible implementation manner of the first aspect, the preset deep learning algorithm is a Retina-ResNet18-Random forest algorithm.

[0020] In a possible implementation manner of the first aspect, the system further includes:

[0021] The module to be identified is used to obtain an image to be detected, input the image to be detected into the cell diagnosis model for screening to obtain abnormal cells to be determined, and input the abnormal cells to be determined into the slice-level diagnosis model to obtain a result to be determined;

[0022] A comparison module, used to obtain a recognition result input by a user, and compare the recognition result with the result to be determined to obtain a comparison parameter, wherein the recognition result is obtained after the user manually recognizes the image to be detected;

[0023] An adjustment training module is used to adjust and train the cell screening model using the comparison parameters.

[0024] A second aspect of the embodiments of the present invention provides a data application method of the auxiliary screening system based on the above-mentioned cervical liquid-based thin layer cytology, the method comprising:

[0025] Using the slice-level diagnostic model to perform numerical processing on the massive amount of scanning electronic images to obtain numerical data;

[0026] evaluating the data quality of the digitized data;

[0027] A screening database is established based on the data quality.

[0028] A third aspect of the embodiments of the present invention provides a data application device based on the auxiliary screening system of cervical liquid-based thin layer cytology as described above, the device comprising:

[0029] A digitization module, used for performing digitization processing on the massive scanning electronic images using the slice-level diagnostic model to obtain digitized data;

[0030] An evaluation module, used for evaluating the data quality of the numerical data;

[0031] An establishing module is used to establish a screening database based on the data quality.

[0032] A fourth aspect of the embodiments of the present invention provides an artificial intelligence cervical liquid-based thin-layer cytology diagnosis prediction model, wherein the diagnosis prediction model includes a pixel-level cell detection model unit and a WSI classification model unit;

[0033] The cell detection model unit is used to perform image segmentation, abnormal cell detection and positioning, and automatic labeling of abnormal cells on the scanned electronic image through an artificial intelligence deep learning algorithm, and output the labeled abnormal cells to the WSI classification model unit;

[0034] The WSI classification model unit is used to diagnose the scanned electronic image based on the annotated abnormal cells through an artificial intelligence deep learning algorithm.

[0035] A fifth aspect of an embodiment of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the data application method of the auxiliary screening system based on cervical liquid-based thin layer cytology as described above is implemented.

[0036] The sixth aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the data application method of the auxiliary screening system based on cervical liquid-based thin layer cytology as described above.

[0037] Compared with the prior art, the auxiliary screening system, application method and model of cervical liquid-based thin-layer cytology provided by the embodiment of the present invention have the following beneficial effects: the present invention can identify abnormal cells from a large number of scanned electronic images annotated by users, and determine the pathological markers corresponding to the abnormal cells, and finally use a preset deep learning algorithm to train the model of the pathological markers corresponding to the abnormal cells to obtain a cell-level diagnosis model and a slice-level diagnosis model. The present invention uses targeted abnormal cells for model recognition, which can increase the recognition ability of the model, and because the number required for training is huge, it can also further improve the recognition ability of the model, and improve the recognition accuracy and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a structural schematic diagram of an auxiliary screening system based on cervical liquid-based thin-layer cytology provided by an embodiment of the present invention;

[0039] Figure 2 is a diagram of the diagnostic efficacy of the model provided by an embodiment of the present invention in a validation cohort;

[0040] Figure 3 This is a comparison chart of the diagnostic efficacy of CAISS, screeners, and CAISS-assisted screeners in a prospective validation cohort of the model provided by an embodiment of the present invention;

[0041] Figure 4 is a comparison chart of the diagnostic efficacy of CAISS, screeners, and CAISS-assisted screeners in a randomized controlled trial of the model provided by an embodiment of the present invention;

[0042] Figure 5 is a diagram of the diagnostic performance of the model provided by an embodiment of the present invention for CAISS-assisted screeners in a prospective real-world cohort;

[0043] Figure 6 It is a flow chart of a data application method of an auxiliary screening system based on cervical liquid-based thin-layer cytology provided by an embodiment of the present invention;

[0044] Figure 7 It is a structural schematic diagram of a data application device of an auxiliary screening system based on cervical liquid-based thin-layer cytology provided by an embodiment of the present invention;

[0045] Figure 8 It is a schematic diagram of the structure of an artificial intelligence diagnosis and prediction model for cervical liquid-based thin-layer cytology provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The currently commonly used node update method has the following technical problems: since notifications and reminders are only made through message transmission interactions during the update process, and the interactive messages are easily tampered with, not only will update errors occur, but the accuracy and effectiveness of the routing table will also be reduced, thereby reducing the communication efficiency and reliability of the entire network.

[0048] In order to solve the above problems, a screening model training method for cervical liquid-based thin layer cells provided in the embodiment of the present application will be introduced and explained in detail through the following specific examples.

[0049] Reference Figure 1 , showing Figure 1 It is a structural schematic diagram of an auxiliary screening system based on cervical liquid-based thin-layer cytology provided by an embodiment of the present invention.

[0050] Wherein, as an example, the auxiliary screening system based on cervical liquid-based thin layer cytology may include:

[0051] The collection module 101 is used to collect a large number of scanned electronic images of cervical liquid-based thin-layer cytology, wherein each of the scanned electronic images includes a variety of different cells.

[0052] Among them, the scanned electronic image can be a TCT smear scanned by a fully automatic digital slide scanner (PRECICE600 Series, UNIC Technologies, INC.; KF-PRO-400-HI, Ningbo Jiangfeng Bio-Information Technology Co., Ltd.) to generate a pathological whole slide image (Whole Slide Image, WSI) with pixel sizes of 0.2529μm×0.2529μm and 0.2484μm×0.2484μm respectively.

[0053] The labeling module 102 identifies abnormal cells in each of the electronic images, outlines the abnormal cells in the electronic images, and adds real labels to the abnormal cells.

[0054] Among them, the pathological mark is the user's mark of the cell type in the image. The image may contain a variety of different cells, such as benign cells, diseased cells, etc. The user can mark each cell to determine the cell type in the image to facilitate subsequent model training.

[0055] Since each scanning electron image contains a variety of different cells, the corresponding abnormal cells can be determined from the scanning electron image, so that corresponding training can be performed based on the characteristics or contours of the abnormal cells, thereby improving the accuracy of recognition.

[0056] In order to improve the accuracy of annotation, in one embodiment, the annotation module is further used to:

[0057] Performing image segmentation on each of the scanned electronic images to obtain a plurality of pixel-level images;

[0058] The cell shape of each pixel-level image is outlined to obtain abnormal cells, and the abnormal cells are diagnosed at the slice level and marked with real labels according to the preset TBS2014 grading system.

[0059] In one embodiment, the area of ​​the scanning electron image is relatively large, while the area of ​​the cell is relatively small. One scanning electron image may contain thousands of cell images. If a single recognition is performed on the scanning electron image, it will increase the time consumption and reduce the processing efficiency.

[0060] To this end, the scanned electronic image can be segmented into multiple pixel-level images, the size of which can be pre-set, such as 5*2 square millimeters or 2*1 square millimeters. After segmentation, multiple pixel-level images can be recognized simultaneously to reduce recognition time and improve recognition efficiency.

[0061] In one embodiment, the features of various abnormal cells in each pixel-level image can be extracted and outlined, and then a cell detection model at the pixel level is established based on the features of various abnormal cells, so that the abnormal cells in the scanned electronic image can be detected and located through the cell detection model.

[0062] In an optional embodiment, the identified abnormal cells may include: negative for intraepithelial lesion or malignancy (NILM), atypical squamous cells of undetermined significance (ASC-US), atypical squamous cells, atypical squamous cells that cannot exclude HSIL (ASC-H), low grade squamous intra-epithelial lesion (LSIL), high grade squamous intraepithelial lesion (HSIL), adenocarcinoma glandular cell (AGC), etc.

[0063] After determining the abnormal cells in the scanned electronic image, the pathological markers corresponding to the abnormal cells can be obtained, and the accuracy of the recognition can be determined by the pathological markers. In addition, model training can be performed based on the pathological markers to improve the accuracy of the recognition.

[0064] In order to avoid missing identification, in an optional embodiment, the annotation module may also be used to:

[0065] Non-labeled cells are identified from the scanning electron image, and pathological markers are added to the non-labeled cells.

[0066] Since cells marked by the user may be missed, some cells may not be marked. In order to avoid omissions, non-marked cells in the scanned electronic image can be identified, and the cell state of the cell can be determined, and pathological marking can be performed based on the cell state.

[0067] The training module 103 is used to perform model training on the real labels corresponding to the abnormal cells based on a preset deep learning algorithm to obtain a cell-level diagnostic model, and to use the information output by the cell-level diagnostic model as input information to train a slice-level diagnostic model.

[0068] In one embodiment, different deep learning algorithms can be used to perform model training on the pathological markers corresponding to the abnormal cells to obtain a cell-level diagnostic model.

[0069] The training module is also used to:

[0070] Inputting the abnormal cell pathology marker into a preset deep learning algorithm;

[0071] Use the preset deep learning algorithm to perform feature learning and model training on the labeled abnormal cells to obtain a cell-level diagnostic model;

[0072] The information output by the cell-level diagnostic model is used as input to the preset random forest classifier to train and obtain the slice-level diagnostic model.

[0073] In one embodiment, the preset deep learning algorithm is a Retina-ResNet18-Random forest algorithm.

[0074] Optionally, the preset deep learning algorithm may also adopt algorithms such as random forest or deep neural networks [DNN].

[0075] The diagnosis and screening module 104 is used to perform cell diagnosis and screening using the slice-level diagnosis model.

[0076] After the chip-level diagnostic model is generated, the chip-level diagnostic model can be used to perform corresponding cell screening and diagnosis to determine the user's cell status.

[0077] The diagnostic efficacy module 105 is used to respectively obtain comparative screening results, auxiliary screening results and diagnostic screening results, and respectively calculate the diagnostic efficacy of the comparative screening results, the auxiliary screening results and the diagnostic screening results, wherein the comparative screening results, the auxiliary screening results and the diagnostic screening results are obtained by collecting data of batch standard patients and dividing them in equal proportions, and then performing diagnostic screening by comparative screeners, auxiliary screeners and the slice-level diagnostic model.

[0078] Among them, the comparative screeners are pathology technicians, and the auxiliary screeners are pathology technicians who use film-level diagnostic models to make pathological judgments.

[0079] The diagrammatic diagram of the comparison of diagnostic efficacies provided by an embodiment of the present invention can be used to calculate the diagnostic efficacies of the diagnostic results using Python software.

[0080] Reference Figure 2, showing the diagnostic efficacy of the model provided by an embodiment of the present invention in the validation cohort. The sensitivity of the cervical liquid-based thin-layer cytology-assisted screening system (CAISS) in the internal validation cohort of Sun Yat-sen Memorial Hospital of Sun Yat-sen University, the external validation cohort of Guangzhou Women and Children's Medical Center, and the external validation cohort of the Third Affiliated Hospital of Guangzhou Medical University were 90.6%, 90.2%, and 91.8%, respectively; the accuracy was 88.1%, 85.0%, and 84.3%, respectively; the specificity was higher than 80.0% in the above three validation cohorts (see Table 2 for detailed results). In the stratified analysis, the model was effective in ACS-US and above (sensitivities were 91.1%,

[0081] The sensitivity was 93.5%, 92.5% and AUC were 0.923, 0.929 and 0.879 respectively), LSIL and above (sensitivities were 92.8%, 96.2% and 95.5% and AUC were 0.950, 0.946 and 0.927 respectively) and HSIL patients (sensitivities were 98.0%, 94.4% and 88.9% and AUC were 0.960, 0.930 and 0.896 respectively).

[0082] Reference Figure 3 , showing a comparison of the diagnostic efficacy of CAISS, screeners, and CAISS-assisted screeners in the prospective validation cohort of the model provided by an embodiment of the present invention. In the prospective validation cohort of Sun Yat-sen Memorial Hospital of Sun Yat-sen University, the TCT diagnostic efficacy of screeners, CAISS, and CAISS-assisted screeners was compared, and the comparison results were as follows: The diagnostic efficacy of CAISS-assisted screeners in reading films (sensitivity: 99.1%, specificity: 99.6%, accuracy: 99.5%, AUC: 0.993) was better than that of screeners (sensitivity: 90.9%, specificity: 98.7%, accuracy: 98.4%, AUC: 0.948) or CAISS alone (sensitivity: 94.6%, specificity: 89.0%, accuracy: 89.2%, AUC: 0.947). The detailed results are shown in Table 2.

[0083] Reference Figure 4, showing a comparison of the diagnostic efficacy of CAISS, screeners, and CAISS-assisted screeners in a randomized controlled trial using the model provided by an embodiment of the present invention. In a prospective randomized controlled trial at Sun Yat-sen Memorial Hospital, the diagnostic efficacy of the CAISS-assisted screener cohort (sensitivity: 100%, specificity: 98.9%, accuracy: 99.0%, AUC: 0.995) was superior to that of pathologists (sensitivity: 87.5%, specificity: 98.4%, accuracy: 97.5%, AUC: 0.929) or the prediction using the cervical cell pathology artificial intelligence diagnostic prediction model alone (sensitivity: 93.8%, specificity: 85.4%, accuracy: 86.1%, AUC: 0.929). See Table 3 for detailed results.

[0084] Reference Figure 5 , showing the diagnostic efficacy of the model provided by an embodiment of the present invention in a prospective real-world cohort of CAISS-assisted screeners. In the prospective real-world cohort of Sun Yat-sen Memorial Hospital of Sun Yat-sen University, the diagnostic efficacy of the CAISS-assisted screener cohort was: sensitivity: 94.3%, specificity: 99.8%, accuracy: 99.5%, AUC: 0.970, and the detailed results are shown in Table 4.

[0085] After completing the model training, in order to further determine the recognition and screening effect of the model, the system may also include:

[0086] The module to be identified 106 is used to obtain an image to be detected, input the image to be detected into the cell diagnosis model to screen and obtain abnormal cells to be determined, and input the abnormal cells to be determined into the slice-level diagnosis model to obtain a result to be determined.

[0087] The image to be detected is a scanned electronic image collected in advance by a user for use in screening, and specifically, may be a large number of scanned electronic images.

[0088] A large number of scanning electron images can be input into the cell-level diagnostic model, which can diagnose the large number of scanning electron images to obtain abnormal cells to be determined. The abnormal cells to be determined are used as input to the slice-level diagnostic model, and the slice-level diagnostic model outputs the results to be determined.

[0089] The comparison module 107 is used to obtain the recognition result input by the user, and compare the recognition result with the result to be determined to obtain a comparison parameter, wherein the recognition result is obtained after the user manually recognizes the image to be detected.

[0090] The recognition result is obtained after the user manually identifies and screens a large number of scanned electronic images.

[0091] The manual screening results of the user and the screening results of the cell screening model can be compared to obtain a comparison parameter, which is a difference value in the number of abnormal cells of the visual mark.

[0092] The adjustment training module 108 is used to adjust and train the cell screening model using the comparison parameters.

[0093] Then, the cell screening model can be verified and adjusted according to the difference in the number of abnormal cells to further improve the screening accuracy of the cell screening model.

[0094] In this embodiment, the embodiment of the present invention provides an auxiliary screening system based on cervical liquid-based thin-layer cytology, and its beneficial effects are: the present invention can identify abnormal cells from a large number of scanned electronic images marked by users, and determine the pathological markers corresponding to the abnormal cells, use a preset deep learning algorithm to train the model for the pathological markers corresponding to the abnormal cells to obtain a cell-level diagnostic model, and then use the information output by the cell-level diagnostic model as input to train and obtain a sheet-level diagnostic model, and finally use the sheet-level diagnostic model to perform corresponding cell diagnosis and screening. The present invention uses targeted abnormal cells for model recognition, which can increase the recognition ability of the model, and because the number of training required is huge, it can also further improve the recognition ability of the model, and improve the accuracy and efficiency of recognition.

[0095] The embodiment of the present invention also provides a data application method of an auxiliary screening system based on cervical liquid-based thin layer cytology, see Figure 6 , showing a flow chart of a data application method of an auxiliary screening system based on cervical liquid-based thin layer cytology provided by an embodiment of the present invention.

[0096] The data application method is applicable to the auxiliary screening system of cervical liquid-based thin-layer cytology as described in the above embodiment.

[0097] Wherein, as an example, the data application method of the auxiliary screening system based on cervical liquid-based thin layer cytology may include:

[0098] S21, using the slice-level diagnostic model to perform digital processing on the massive amount of scanning electronic images to obtain digital data.

[0099] The digitization process may be to convert the scanned electronic image into a numerical value in a corresponding format.

[0100] S22: Evaluate the data quality of the digitized data.

[0101] The data quality may be an indicator value of the number of abnormal cells included in the scanning electron image. If the number of abnormal cells included in the scanning electron image is large, the data quality is high.

[0102] S23. Establishing a screening database based on the data quality.

[0103] In this embodiment, the embodiment of the present invention provides a data application method for an auxiliary screening system based on cervical liquid-based thin layer cytology, which has the beneficial effect of: by establishing a screening database, the user can select scanning electronic images with different data qualities from the screening database for training when conducting model training, thereby generating cell screening models with different accuracy rates to facilitate the user to perform different screening operations later.

[0104] The embodiment of the present invention also provides a data application device for an auxiliary screening system based on cervical liquid-based thin-layer cytology, see Figure 7 , showing a schematic structural diagram of a data application device of an auxiliary screening system based on cervical liquid-based thin-layer cytology provided by an embodiment of the present invention;

[0105] Wherein, as an example, the data application device of the auxiliary screening system based on cervical liquid-based thin layer cytology may include:

[0106] A digitization module 301 is used to perform digitization processing on the massive scanning electronic images using the slice-level diagnostic model to obtain digitized data;

[0107] An evaluation module 302, used for evaluating the data quality of the numerical data;

[0108] The establishing module 303 is used to establish a screening database based on the data quality.

[0109] The present invention also provides an artificial intelligence cervical liquid-based thin-layer cytology diagnosis prediction model, see Figure 8 , showing a schematic structural diagram of an artificial intelligence cervical liquid-based thin-layer cytology diagnostic prediction model provided by an embodiment of the present invention.

[0110] Wherein, as an example, the diagnosis prediction model based on artificial intelligence cervical liquid-based thin layer cytology includes:

[0111] Pixel-level cell detection model unit and WSI classification model unit;

[0112] The cell detection model unit is used to perform image segmentation, abnormal cell detection and positioning, and automatic labeling of abnormal cells on the scanned electronic image through an artificial intelligence deep learning algorithm, and output the labeled abnormal cells to the WSI classification model unit;

[0113] The WSI classification model unit is used to diagnose the scanned electronic image based on the annotated abnormal cells through an artificial intelligence deep learning algorithm.

[0114] Furthermore, an embodiment of the present application also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the data application method of the auxiliary screening system based on cervical liquid-based thin layer cytology as described in the above embodiment is implemented.

[0115] Furthermore, an embodiment of the present application also provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the data application method of the auxiliary screening system based on cervical liquid-based thin layer cytology as described in the above embodiment.

[0116] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A data application method for an auxiliary screening system based on cervical liquid-based thin-layer cytology, characterized in that: The auxiliary screening system of cervical liquid-based thin-layer cytology includes: A collection module, used for collecting a large number of scanned electronic images of cervical liquid-based thin-layer cytology, wherein each scanned electronic image includes a variety of different cells; a labeling module, identifying abnormal cells in each electronic image, outlining the abnormal cells in the electronic image, and adding true labels to the abnormal cells; A training module, used to perform model training on the real labels corresponding to the abnormal cells based on a preset deep learning algorithm to obtain a cell-level diagnostic model, and to use the information output by the cell-level diagnostic model as input information to train a slice-level diagnostic model; A diagnosis and screening module, used for diagnosing and screening cells using the chip-level diagnosis model; A diagnostic efficacy module, for respectively obtaining comparative screening results, auxiliary screening results and diagnostic screening results, and respectively calculating the diagnostic efficacy of the comparative screening results, the auxiliary screening results and the diagnostic screening results, wherein the comparative screening results, the auxiliary screening results and the diagnostic screening results are obtained by collecting data of batch standard patients and dividing them in equal proportions, and then performing diagnostic screening by comparative screeners, auxiliary screeners and the slice-level diagnostic model respectively; The method comprises: Using the slice-level diagnostic model to digitize a large number of scanning electronic images to obtain digitized data; evaluating the data quality of the digitized data; establishing a screening database based on the data quality; The annotation module is also used for: Performing image segmentation on each of the scanned electronic images to obtain a plurality of pixel-level images; Delineating the cell shape of each pixel-level image to obtain abnormal cells, and performing slice-level diagnosis on the abnormal cells and marking true labels according to the preset TBS 2014 grading system; The training module is also used to: The abnormal cell pathology markers are input into a preset deep learning algorithm; Use the preset deep learning algorithm to perform feature learning and model training on the labeled abnormal cells to obtain a cell-level diagnostic model; The information outputted by the cell-level diagnostic model is used as input to a preset random forest classifier to train and obtain a slice-level diagnostic model; The preset deep learning algorithm is the Retina-ResNet18-Random forest algorithm; The auxiliary screening system of cervical liquid-based thin-layer cytology also includes: The module to be identified is used to obtain an image to be detected, input the image to be detected into the cell-level diagnostic model for screening to obtain abnormal cells to be determined, and input the abnormal cells to be determined into the slice-level diagnostic model to obtain a result to be determined; A comparison module, used to obtain a recognition result input by a user, and compare the recognition result with the result to be determined to obtain a comparison parameter, wherein the recognition result is obtained after the user manually recognizes the image to be detected; The adjustment training module is used to adjust and train the cell screening model using the comparison parameters.

2. A data application device for an auxiliary screening system based on cervical liquid-based thin-layer cytology, characterized in that: The auxiliary screening system of cervical liquid-based thin-layer cytology includes: A collection module, used for collecting a large number of scanned electronic images of cervical liquid-based thin-layer cytology, wherein each scanned electronic image includes a variety of different cells; a labeling module, identifying abnormal cells in each electronic image, outlining the abnormal cells in the electronic image, and adding true labels to the abnormal cells; A training module, used to perform model training on the real labels corresponding to the abnormal cells based on a preset deep learning algorithm to obtain a cell-level diagnostic model, and to use the information output by the cell-level diagnostic model as input information to train a slice-level diagnostic model; A diagnosis and screening module, used for diagnosing and screening cells using the chip-level diagnosis model; A diagnostic efficacy module, for respectively obtaining comparative screening results, auxiliary screening results and diagnostic screening results, and respectively calculating the diagnostic efficacy of the comparative screening results, the auxiliary screening results and the diagnostic screening results, wherein the comparative screening results, the auxiliary screening results and the diagnostic screening results are obtained by collecting data of batch standard patients and dividing them in equal proportions, and then performing diagnostic screening by comparative screeners, auxiliary screeners and the slice-level diagnostic model respectively; The device comprises: A digitization module, used to perform digitization processing on a large number of scanning electronic images using the slice-level diagnostic model to obtain digitized data; An evaluation module, used for evaluating the data quality of the numerical data; An establishment module is used to establish a screening database based on the data quality; The annotation module is also used for: Performing image segmentation on each of the scanned electronic images to obtain a plurality of pixel-level images; Delineating the cell shape of each pixel-level image to obtain abnormal cells, and performing slice-level diagnosis on the abnormal cells and marking true labels according to the preset TBS 2014 grading system; The training module is also used to: The abnormal cell pathology markers are input into a preset deep learning algorithm; Use the preset deep learning algorithm to perform feature learning and model training on the labeled abnormal cells to obtain a cell-level diagnostic model; The information outputted by the cell-level diagnostic model is used as input to a preset random forest classifier to train and obtain a slice-level diagnostic model; The preset deep learning algorithm is the Retina-ResNet18-Random forest algorithm; The auxiliary screening system of cervical liquid-based thin-layer cytology also includes: The module to be identified is used to obtain an image to be detected, input the image to be detected into the cell-level diagnostic model for screening to obtain abnormal cells to be determined, and input the abnormal cells to be determined into the slice-level diagnostic model to obtain a result to be determined; A comparison module, used to obtain a recognition result input by a user, and compare the recognition result with the result to be determined to obtain a comparison parameter, wherein the recognition result is obtained after the user manually recognizes the image to be detected; The adjustment training module is used to adjust and train the cell screening model using the comparison parameters.

3. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the data application method of the auxiliary screening system based on cervical liquid-based thin layer cytology as claimed in claim 1 is implemented.

4. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the data application method of the auxiliary screening system based on cervical liquid-based thin layer cytology as claimed in claim 1.

Images