Positive detection method and device for cervical cells

Abstract

The present application relates to the technical field of cytological detection, and provides a positive detection method and device for cervical cells. The method comprises: obtaining abnormal cell detection results according to a cell detection model and a cervical cell small block image obtained by cropping the cervical cell full slice image by field of view; mapping the abnormal cell detection results to the corresponding position of the cervical cell full slice image to obtain a positive confidence distribution heat map; performing field of view sampling on the fields of view represented by several grids of the above heat map to obtain cells in several fields of view as suspicious positive cells; when the maximum confidence of the suspicious positive cell is greater than the high-risk cell threshold, determining that the detection result of the cervical cell full slice image is positive; when the maximum confidence of the suspicious positive cell is less than the high-risk cell threshold, inputting the cervical cell full slice image into a classification model to determine whether the cervical cell full slice image is positive. The use of this method can improve the accuracy of cervical cell positive detection.

Description

Technical Field

[0001] The present application relates to the technical field of cytological detection, and in particular to a method, device, computer equipment, computer-readable storage medium and computer program product for positive detection of cervical cells. Background Art

[0002] In recent years, the incidence and mortality of cervical cancer have gradually increased, and are showing a trend of younger people. Early screening and treatment are crucial to reducing the incidence and mortality of cervical cancer, among which cervical cell positive detection is the main screening method.

[0003] In traditional technology, cell-level annotation can be performed in advance on multiple fields of view of the whole cervical cell slice to obtain a detection model with the ability to locate positive cells, and then the whole cervical cell slice can be classified by aggregating the information of multiple positive cells to determine whether the whole cervical cell slice is positive. However, due to the complex morphology of cervical cells, the current detection model based on deep learning with good performance has a serious problem of model overconfidence. The error of the detection model will be directly accumulated in the information aggregation stage, affecting the accuracy of cervical cell positive detection. In addition, the currently commonly used information aggregation modules often use classification models with weak fitting capabilities, such as random forests, which have serious underfitting problems, resulting in low generalization performance of the model, further affecting the accuracy of cervical cell positive detection. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for positive detection of cervical cells in response to the above-mentioned technical problems.

[0005] In a first aspect, the present application provides a method for positive detection of cervical cells, comprising:

[0006] According to the pre-constructed cell detection model, abnormal cells in the small block image of cervical cells are detected to obtain a number of abnormal cell detection results; the small block image of cervical cells is obtained by cropping the full slice image of cervical cells field by field;

[0007] Mapping the plurality of abnormal cell detection results to corresponding positions of the full-slice image of cervical cells to obtain a positive confidence distribution heat map;

[0008] Performing visual field sampling on the visual fields represented by the plurality of grids of the positive confidence distribution heat map to obtain cells in the plurality of visual fields as suspected positive cells;

[0009] When the maximum confidence of the suspicious positive cells is greater than the high-risk cell threshold, determining that the detection result of the whole cervical cell slice image is positive;

[0010] When the maximum confidence of the suspicious positive cells is less than the high-risk cell threshold, the whole cervical cell slide image is input into a pre-built classification model to determine whether the whole cervical cell slide image is positive.

[0011] In one embodiment, mapping the plurality of abnormal cell detection results to corresponding positions of the full cervical cell slice image to obtain a positive confidence distribution heat map includes:

[0012] Mapping the confidences of the plurality of abnormal cell detection results to corresponding positions of the full-slice image of cervical cells to obtain an initial heat map;

[0013] Each grid in the initial thermodynamic map is interpolated and normalized to obtain a positive confidence distribution thermodynamic map.

[0014] In one embodiment, obtaining a high-risk cell threshold comprises:

[0015] Acquire a multi-center data verification set; the multi-center data verification set is cervical cell full-slice image sample data from multiple different sources;

[0016] Inferring the cell detection model according to the multi-center data validation set to obtain the classification probability of each cervical cell full-slice image sample;

[0017] According to the receiver operating characteristic curve and precision-recall curve, the classification results of the whole-slice image samples of cervical cells at different thresholds were evaluated, and the performance indicators at different thresholds were obtained;

[0018] A high-risk cell threshold is determined based on the performance indicators under the different thresholds.

[0019] In one embodiment, obtaining the maximum confidence level of the suspected positive cells includes:

[0020] Sort by confidence level of each suspected positive cell to obtain a confidence level sequence;

[0021] According to the confidence sequence, the maximum confidence of the suspected positive cells is obtained.

[0022] In one embodiment, obtaining a pre-built classification model includes:

[0023] Obtaining a difficult slice sample set based on the cervical cell whole slice samples with high difficulty in positive determination of the cell detection model;

[0024] The residual network is trained according to the difficult slice sample set to obtain a pre-built classification model.

[0025] In one embodiment, obtaining a whole section sample of cervical cells that is difficult to determine as positive by the cell detection model includes:

[0026] Obtain a collection of whole-section cervical cell samples;

[0027] According to the pre-built cell detection model, the confidence distribution of each cervical cell whole section sample is obtained;

[0028] According to the confidence distribution of each cervical cell whole section sample, an average confidence distribution is obtained;

[0029] According to the distribution distance between the confidence distribution of each cervical cell whole section sample and the average confidence distribution, a high-difference whole section sample is obtained to determine the cervical cell whole section sample with high difficulty in positive judgment of the cell detection model.

[0030] In a second aspect, the present application also provides a positive detection device for cervical cells, comprising:

[0031] The abnormal cell detection module is used to detect abnormal cells in a small block of cervical cell images according to a pre-built cell detection model, and obtain a number of abnormal cell detection results; the small block of cervical cell images are obtained by cropping the full slice image of cervical cells field by field;

[0032] A heat map acquisition module, used for mapping the plurality of abnormal cell detection results to corresponding positions of the full-slice image of cervical cells to obtain a positive confidence distribution heat map;

[0033] A suspected positive cell acquisition module is used to perform visual field sampling on the visual fields represented by several grids of the positive confidence distribution heat map to obtain cells in several visual fields as suspected positive cells;

[0034] A positive judgment module, used for determining that the detection result of the whole cervical cell slice image is positive when the maximum confidence of the suspicious positive cells is greater than the high-risk cell threshold;

[0035] The positive judgment module is also used to input the cervical cell whole section image into a pre-built classification model to determine whether the cervical cell whole section image is positive when the maximum confidence of the suspicious positive cells is less than the high-risk cell threshold.

[0036] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the above method.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program is executed by a processor to execute the above method.

[0038] In a fifth aspect, the present application further provides a computer program product, wherein the computer program product comprises a computer program, and the computer program is executed by a processor to execute the above method.

[0039] The above-mentioned positive detection method, device, computer equipment, computer-readable storage medium and computer program product of cervical cells detect abnormal cells in a small block image of cervical cells according to a pre-constructed cell detection model to obtain several abnormal cell detection results; the small block image of cervical cells is obtained by cropping the full slice image of cervical cells field by field; the several abnormal cell detection results are mapped to the corresponding positions of the full slice image of cervical cells to obtain a positive confidence distribution heat map; the fields of view represented by several grids of the positive confidence distribution heat map are sampled to obtain cells in several fields of view as suspected positive cells; when the maximum confidence of the suspected positive cell is greater than the high-risk cell threshold, the detection result of the full slice image of cervical cells is determined to be positive; when the maximum confidence of the suspected positive cell is less than the high-risk cell threshold, the full slice image of cervical cells is input into a pre-constructed classification model to determine whether the full slice image of cervical cells is positive. In the present application, when the maximum confidence level of suspected positive cells is greater than the high-risk cell threshold, the detection result of the whole cervical cell slice image is determined to be positive. When the maximum confidence level of suspected positive cells is less than the high-risk cell threshold, the whole cervical cell slice image is input into a pre-built classification model to determine whether the whole cervical cell slice image is positive. This can avoid the error of the cell detection model from being directly accumulated in the information aggregation stage of the classification model, and can improve the accuracy of cervical cell positive detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A diagram showing an application environment of a positive detection method for cervical cells in one embodiment;

[0042] Figure 2 It is a schematic diagram of the process of a positive detection method of cervical cells in one embodiment;

[0043] Figure 3A schematic diagram of the structure of a YOLOv8 model in one embodiment;

[0044] Figure 4 is a schematic flow chart of a positive detection method for cervical cells in another embodiment;

[0045] Figure 5 is a structural block diagram of a positive detection device for cervical cells in one embodiment;

[0046] Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] The present application embodiment provides a method for positive detection of cervical cells, which can be executed by a computer device, such as Figure 1 As shown, the computer device can obtain the whole cervical cell slice image, and then determine whether the whole cervical cell slice image is positive. It can be understood that the computer device can be implemented by a server, a terminal, or an interactive system between a terminal and a server. In this embodiment, the method includes Figure 2 The steps shown are:

[0049] Step S201, detecting abnormal cells in a small block of cervical cell images according to a pre-constructed cell detection model, and obtaining a number of abnormal cell detection results; the small block of cervical cell images are obtained by cropping the full slice image of cervical cells field by field.

[0050] According to Figure 3 The YOLOv8 model shown is trained to obtain a pre-built cell detection model. Figure 3 The backbone network (Backbone) of the YOLOv8 model is a multi-scale pyramid structure, which can be a cross-stage residual network CSPDarknet, which can extract image features at different scales; the neck (Neck) of the YOLOv8 model can be a feature pyramid network (FPN), which can efficiently mix multi-scale information by interactively fusing the multi-layer feature maps output by the backbone network; the detection head (Head) of the YOLOv8 model realizes the positioning and classification of abnormal cells through regression and classification.

[0051] A solution that integrates multiple homogeneous or heterogeneous cell detection models can be adopted to alleviate the problem of large differences in multi-center performance of cell detection models.

[0052] The cell detection model can be optimized according to the loss formula shown in formula (1).

[0053] (1)

[0054] in, is the Complete Intersection over Union Loss (CioULoss), is the cross entropy loss, and is a hyperparameter, is the variable focal loss (VariFocal Loss), and the calculation method is shown in formula (2).

[0055] (2)

[0056] Among them, q is the intersection-over-union (IoU) of the predicted box and the true box of the cell detection model. IoU is the intersection of the predicted box and the true box divided by the union of the two boxes, and p is the confidence distribution predicted by the cell detection model.

[0057] The whole slide imaging (WSI) of cervical cells can be cropped field by field to obtain a data set X= of N 1280×1280 cervical cell small block images. The cervical cell patch images can be input into the cell detection model one by one to obtain M abnormal cell detection results Boxes= .

[0058] Step S202, mapping a number of abnormal cell detection results to corresponding positions of the cervical cell full-slice image to obtain a positive confidence distribution heat map.

[0059] The abnormal cell test result includes the confidence level that the cervical cells are classified as abnormal.

[0060] The confidence levels of several abnormal cell detection results can be mapped to the corresponding positions of the full-slice image of cervical cells at the original scale to obtain a positive confidence distribution heat map.

[0061] Step S203 , sampling the visual fields represented by the plurality of grids of the positive confidence distribution heat map to obtain cells in the plurality of visual fields as suspicious positive cells.

[0062] The visual field represented by the top k grids in the highlighted area of ​​the positive confidence distribution heat map can be sampled to obtain the cells in the top k visual fields as the suspected positive cells Q= .

[0063] Step S204: when the maximum confidence level of the suspicious positive cells is greater than the high-risk cell threshold, the detection result of the whole cervical cell slice image is determined to be positive.

[0064] In the positive confidence distribution heat map, each field of view has a corresponding confidence, and the confidence corresponding to each field of view can be used as the confidence of the suspected positive cells corresponding to each field of view. The maximum confidence of the suspected positive cells can be obtained according to the confidence of each suspected positive cell.

[0065] When the maximum confidence of the suspicious positive cells is greater than the high-risk cell threshold, it indicates that the probability of the presence of positive cells in the whole section image of cervical cells is relatively high. At this time, the detection result of the whole section image of cervical cells can be determined as positive.

[0066] Step S205, when the maximum confidence of the suspicious positive cells is less than the high-risk cell threshold, the whole cervical cell slide image is input into a pre-built classification model to determine whether the whole cervical cell slide image is positive.

[0067] A classification model can be trained based on the residual network model ResNet34 and the whole cervical cell slice sample with high difficulty in positive judgment of the cell detection model.

[0068] When the maximum confidence of the suspicious positive cells is less than the high-risk cell threshold, it indicates that the probability of positive cells in the whole cervical cell slice image is small. At this time, in order to improve the accuracy of cervical cell positive detection, the whole cervical cell slice image can be input into the pre-built classification model so that the classification model can extract the field-level features of all positions of the whole cervical cell slice image to obtain the feature data set F= , all feature information is aggregated by multi-instance learning to obtain the feature vector Fi of the cervical cell full slice image The feature vector is sent to the classification network in the classification model for classification to determine whether the cervical cell full slide image is positive. The confidence level can also be called a score.

[0069] In the above-mentioned positive detection method of cervical cells, when the maximum confidence of suspected positive cells is greater than the high-risk cell threshold, the detection result of the whole section image of cervical cells is determined to be positive; when the maximum confidence of suspected positive cells is less than the high-risk cell threshold, the whole section image of cervical cells is input into a pre-constructed classification model to determine whether the whole section image of cervical cells is positive. This can avoid the error of the cell detection model from being directly accumulated in the information convergence stage of the classification model, and can improve the accuracy of cervical cell positive detection.

[0070] In one of the embodiments, several abnormal cell detection results are mapped to corresponding positions of the full-slice image of cervical cells to obtain a positive confidence distribution heat map. The specific steps are as follows: the confidences in several abnormal cell detection results are mapped to corresponding positions of the full-slice image of cervical cells to obtain an initial heat map; each grid in the initial heat map is interpolated and normalized to obtain a positive confidence distribution heat map.

[0071] The confidences in several abnormal cell detection results can be mapped to the corresponding positions of the full-slice image of cervical cells at the original scale to obtain an initial heat map H; each grid in the initial heat map H can be interpolated and normalized to obtain a positive confidence distribution heat map.

[0072] In this embodiment, the confidence levels of several abnormal cell detection results are mapped to corresponding positions of the full-slice image of cervical cells to obtain an initial heat map, and the initial heat map is interpolated and normalized to obtain a more accurate positive confidence distribution heat map.

[0073] In one of the embodiments, a high-risk cell threshold is obtained by the following specific steps: obtaining a multi-center data verification set; the multi-center data verification set is cervical cell whole slide image sample data from multiple different sources; inferring the cell detection model based on the multi-center data verification set to obtain the classification probability of each cervical cell whole slide image sample; evaluating the classification results of cervical cell whole slide image samples under different thresholds based on the receiver operating characteristic curve and the precision-recall curve to obtain performance indicators under different thresholds; determining the high-risk cell threshold based on the performance indicators under different thresholds.

[0074] A multi-center data validation set can be obtained based on cervical cell whole-slice image sample data from multiple different sources, thereby ensuring sample diversity of the multi-center data validation set.

[0075] The cell detection model was inferred based on the multi-center data validation set to obtain the classification probability of each cervical cell whole-slice image sample; the classification results of cervical cell whole-slice image samples under different thresholds were evaluated based on the Receiver Operating Characteristic (ROC) curve and the Precision-Recall Curve, and the performance indicators under different thresholds were obtained; the performance indicators may be accuracy, sensitivity (true positive rate), specificity (true negative rate), and F1 score (F1 Score); based on the performance indicators under different thresholds, the threshold that makes the cell detection model have the highest specificity was selected as the high-risk cell threshold.

[0076] In this embodiment, the classification results of the whole-slice image samples of cervical cells at different thresholds are evaluated to obtain performance indicators at different thresholds. The high-risk cell threshold is determined based on the performance indicators at different thresholds, which can ensure the robustness of the cell detection model in real scenarios. Under the premise of ensuring the sensitivity of the cell detection model, the high-risk cell threshold that makes the cell detection model have the highest specificity can be screened out.

[0077] In one embodiment, the maximum confidence of the suspicious positive cells is obtained by the following specific steps: sorting the confidences of the suspicious positive cells to obtain a confidence sequence; and obtaining the maximum confidence of the suspicious positive cells according to the confidence sequence.

[0078] In the positive confidence distribution heat map, each field of view has a corresponding confidence, and the confidence corresponding to each field of view can be used as the confidence of the suspected positive cells corresponding to each field of view. According to the confidence of each suspected positive cell, it is sorted from large to small to obtain a confidence sequence; the confidence ranked first in the confidence sequence can be used as the maximum confidence of the suspected positive cell.

[0079] In this embodiment, the confidence levels of the suspicious positive cells are sorted to obtain a confidence level sequence, so as to obtain the maximum confidence level of the suspicious positive cells.

[0080] In one of the embodiments, a pre-built classification model is obtained, and the specific steps are as follows: a difficult slice sample set is obtained based on a whole slice sample of cervical cells that is difficult to determine as positive by a cell detection model; and a residual network is trained based on the difficult slice sample set to obtain a pre-built classification model.

[0081] The residual network can be ResNet34, which is a residual block (ResBlock) with 34 layers, and can achieve efficient information aggregation by residually connecting each layer.

[0082] A difficult slice sample set can be obtained based on the cervical cell whole slice samples with high difficulty in positive judgment of the cell detection model; the residual network can be trained based on the difficult slice sample set to obtain a pre-built classification model with good performance, so as to perform field-level feature extraction on the cervical cell whole slice images when the maximum confidence of suspicious positive cells is less than the high-risk cell threshold, and aggregate and classify all feature information through multi-instance learning to determine whether the cervical cell whole slice image is positive.

[0083] In this embodiment, based on the whole slice samples of cervical cells with high difficulty in positive judgment of the cell detection model, a difficult slice sample set is obtained to train the residual network to obtain a pre-built classification model with better performance, thereby improving the accuracy of cervical cell positive detection.

[0084] In one of the embodiments, whole-slice cervical cell samples that are difficult to determine as positive by a cell detection model are obtained, and the specific steps are as follows: obtain a set of whole-slice cervical cell samples; obtain the confidence distribution of each whole-slice cervical cell sample based on a pre-constructed cell detection model; obtain the average confidence distribution based on the confidence distribution of each whole-slice cervical cell sample; obtain high-difference whole-slice samples based on the distribution distance between the confidence distribution of each whole-slice cervical cell sample and the average confidence distribution, so as to determine whole-slice cervical cell samples that are difficult to determine as positive by a cell detection model.

[0085] You can get a collection of cervical cell slides Slides= According to the pre-built cell detection model, the whole cervical cell slice samples are traversed and tested to obtain the confidence distribution T= ; The average confidence distribution can be obtained based on the confidence distribution of each cervical cell whole section sample ,in is the mean, is the standard deviation.

[0086] The confidence distribution T = With the average confidence distribution The distribution distance between the confidence distribution of each cervical cell whole section sample and the average confidence distribution is calculated; the distribution difference can be described by the mean difference of the distribution; according to the distribution distance between the confidence distribution of each cervical cell whole section sample and the average confidence distribution, the top 5% of high-difference whole section samples can be selected as cervical cell whole section samples with high difficulty in positive judgment of the cell detection model, wherein the longer the distribution distance, the higher the difference.

[0087] In this embodiment, an average confidence distribution is obtained based on the confidence distribution of each cervical cell whole slice sample; a high-difference whole slice sample is obtained based on the distribution distance between the confidence distribution of each cervical cell whole slice sample and the average confidence distribution, so as to determine the cervical cell whole slice samples with high difficulty in positive judgment of the cell detection model, thereby obtaining a difficult slice sample set, training a classification model with better performance, and improving the accuracy of cervical cell positive detection.

[0088] In order to better understand the above method, an application example of the positive detection method of cervical cells of the present application is described in detail below. Figure 4 shown.

[0089] The whole slide imaging (WSI) of cervical cells can be cropped field by field through a sliding window to obtain a data set X= of N 1280×1280 cervical cell small block images. The cervical cell patch images can be input into the cell detection model one by one to obtain M abnormal cell detection results Boxes= ; The pre-built cell detection model is trained based on the YOLOv8 model.

[0090] Map the M abnormal cell detection results to the corresponding positions of the cervical cell full-slice image to obtain a positive confidence distribution heat map; perform field sampling on the fields of view represented by several grids of the positive confidence distribution heat map to obtain cells in several fields of view as suspected positive cells; obtain a confidence sequence based on the confidence of each suspected positive cell; use the confidence ranked first in the confidence sequence as the maximum confidence of the suspected positive cell; when the maximum confidence of the suspected positive cell is greater than the high-risk cell threshold, determine that the detection result of the cervical cell full-slice image is positive; when the maximum confidence of the suspected positive cell is less than the high-risk cell threshold, input the cervical cell full-slice image into a pre-constructed classification model to obtain a positive probability distribution to determine whether the cervical cell full-slice image is positive.

[0091] In the above-mentioned positive detection method of cervical cells, when the maximum confidence of suspected positive cells is greater than the high-risk cell threshold, the detection result of the whole section image of cervical cells is determined to be positive; when the maximum confidence of suspected positive cells is less than the high-risk cell threshold, the whole section image of cervical cells is input into a pre-constructed classification model to determine whether the whole section image of cervical cells is positive. This can avoid the error of the cell detection model from being directly accumulated in the information convergence stage of the classification model, and can improve the accuracy of cervical cell positive detection.

[0092] This application can test the positive detection performance of cervical cell whole-slice images on a large-scale multicenter cervical cell clinical dataset containing 9 descriptive (The Bethesda System, TBS) categories. It can create a field-level multicenter cervical cell sample test set containing 70,000 annotated images and a multicenter cervical cell clinical experimental slice dataset of nearly 10,000 images to evaluate the abnormal cell location classification ability and whole-slice image classification ability of the cervical cell positive detection method proposed in this application.

[0093] Since the boundaries of cervical cells are not clearly defined during the labeling process, the labeling boxes of different labelers may be quite different. Therefore, in the experiment, mAP@50 (Mean Average Precision at IoU threshold 0.50), which has slightly lower requirements on positioning accuracy, can be used as the main indicator for cervical cell positive cell detection. The performance indicator data of the classification model of the large-scale multi-center cervical cell clinical experiment are shown in Table 1, and the mAP@50 performance test results of the cell detection model on the positive category are shown in Table 2.

[0094] Table 1 Performance index data of cervical cell positive detection in large-scale multicenter cervical cell clinical trials

[0095]

[0096] Table 2 mAP@50 performance test results of cell detection model on positive category

[0097]

[0098] Among them, ASC-US is called Atypical Squamous Cells of Undetermined Significance in English, and its full name in Chinese is Atypical Squamous Cells of Undetermined Significance; LSIL is called Low-Grade Squamous Intraepithelial Lesion in English, and its full name in Chinese is Low-Grade Squamous Intraepithelial Lesion; ASC-H is called Atypical Squamous Cells, cannot rule out High-grade Squamous Intraepithelial Lesion in English, and its full name in Chinese is Atypical Squamous Cells that cannot rule out High-grade Squamous Intraepithelial Lesion; HSIL+ is called High-Grade Squamous Intraepithelial Lesion in English, and its full name in Chinese is High-Grade Squamous Intraepithelial Lesion; HSIL+ includes: AGC (Adenomatous Glandular Change, glandular hyperplasia), HSIL (High-grade Squamous Intraepithelial Lesion, high-grade squamous intraepithelial lesion), SCC (Squamous Cell Carcinoma (squamous cell carcinoma) and ADC (Adenocarcinoma).

[0099] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0100] Based on the same inventive concept, the embodiment of the present application also provides a positive detection device for cervical cells for implementing the positive detection method for cervical cells involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of one or more positive detection devices for cervical cells provided below can refer to the limitations of the positive detection method for cervical cells above, and will not be repeated here.

[0101] In an exemplary embodiment, Figure 5 As shown, a positive detection device for cervical cells is provided, wherein:

[0102] The abnormal cell detection module 501 is used to detect abnormal cells in a small block of cervical cell images according to a pre-built cell detection model, and obtain a number of abnormal cell detection results; the small block of cervical cell images are obtained by cropping the full slice image of cervical cells field by field;

[0103] The heat map acquisition module 502 is used to map the plurality of abnormal cell detection results to corresponding positions of the cervical cell full slice image to obtain a positive confidence distribution heat map;

[0104] The suspected positive cell acquisition module 503 is used to perform visual field sampling on the visual fields represented by the plurality of grids of the positive confidence distribution heat map to obtain cells in the plurality of visual fields as suspected positive cells;

[0105] A positive judgment module 504 is used to determine that the detection result of the cervical cell full-slice image is positive when the maximum confidence of the suspicious positive cells is greater than the high-risk cell threshold;

[0106] The positive judgment module 504 is further configured to input the whole cervical cell slide image into a pre-built classification model to determine whether the whole cervical cell slide image is positive when the maximum confidence of the suspicious positive cells is less than the high-risk cell threshold.

[0107] In one of the embodiments, the heat map acquisition module 502 is further used to: map the confidences in the plurality of abnormal cell detection results to corresponding positions in the full cervical cell slice image to obtain an initial heat map; and perform interpolation and normalization processing on each grid in the initial heat map to obtain a positive confidence distribution heat map.

[0108] In one embodiment, the device also includes a high-risk cell threshold acquisition module, which is used to: obtain a multi-center data verification set; the multi-center data verification set is cervical cell whole slide image sample data from multiple different sources; infer the cell detection model based on the multi-center data verification set to obtain the classification probability of each cervical cell whole slide image sample; evaluate the classification results of cervical cell whole slide image samples under different thresholds based on the receiver operating characteristic curve and the precision-recall curve to obtain performance indicators under different thresholds; determine the high-risk cell threshold based on the performance indicators under the different thresholds.

[0109] In one of the embodiments, the device further includes a maximum confidence acquisition module, which is used to: sort the suspicious positive cells according to their confidences to obtain a confidence sequence; and obtain the maximum confidence of the suspicious positive cells according to the confidence sequence.

[0110] In one embodiment, the device also includes a classification model acquisition module, which is used to: obtain a difficult slice sample set based on a cervical cell whole slice sample with high difficulty in positive judgment of the cell detection model; train a residual network based on the difficult slice sample set to obtain a pre-constructed classification model.

[0111] In one embodiment, the device also includes a whole-slice sample acquisition module, which is used to: acquire a set of cervical cell whole-slice samples; obtain the confidence distribution of each cervical cell whole-slice sample based on a pre-constructed cell detection model; obtain the average confidence distribution based on the confidence distribution of each cervical cell whole-slice sample; obtain high-difference whole-slice samples based on the distribution distance between the confidence distribution of each cervical cell whole-slice sample and the average confidence distribution, so as to determine cervical cell whole-slice samples with high difficulty in positive judgment of the cell detection model.

[0112] Each module in the above-mentioned positive detection device for cervical cells can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0113] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data of an embodiment of a positive detection method for cervical cells. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a positive detection method for cervical cells is implemented.

[0114] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0115] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0116] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0117] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0118] 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 this application 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 relevant regulations.

[0119] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0120] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0121] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for positive detection of cervical cells, characterized in that: The method comprises: According to the pre-constructed cell detection model, abnormal cells in the small block image of cervical cells are detected to obtain a number of abnormal cell detection results; the small block image of cervical cells is obtained by cropping the full slice image of cervical cells field by field; Mapping the plurality of abnormal cell detection results to corresponding positions of the full-slice image of cervical cells to obtain a positive confidence distribution heat map; Performing visual field sampling on the visual fields represented by the plurality of grids of the positive confidence distribution heat map to obtain cells in the plurality of visual fields as suspected positive cells; When the maximum confidence of the suspicious positive cells is greater than the high-risk cell threshold, determining that the detection result of the whole cervical cell slice image is positive; When the maximum confidence of the suspicious positive cells is less than the high-risk cell threshold, the whole cervical cell slide image is input into a pre-built classification model to determine whether the whole cervical cell slide image is positive.

2. The method according to claim 1, characterized in that Mapping the plurality of abnormal cell detection results to corresponding positions of the full-slice image of cervical cells to obtain a positive confidence distribution heat map includes: Mapping the confidences of the plurality of abnormal cell detection results to corresponding positions of the full-slice image of cervical cells to obtain an initial heat map; Each grid in the initial thermodynamic map is interpolated and normalized to obtain a positive confidence distribution thermodynamic map.

3. The method according to claim 1, characterized in that Obtain high-risk cell thresholds, including: Acquire a multi-center data verification set; the multi-center data verification set is cervical cell full-slice image sample data from multiple different sources; Inferring the cell detection model according to the multi-center data validation set to obtain the classification probability of each cervical cell full-slice image sample; According to the receiver operating characteristic curve and precision-recall curve, the classification results of the whole-slice image samples of cervical cells at different thresholds were evaluated, and the performance indicators at different thresholds were obtained; A high-risk cell threshold is determined based on the performance indicators under the different thresholds.

4. The method according to claim 1, characterized in that: Obtain the highest confidence level for suspected positive cells, including: Sort by confidence level of each suspected positive cell to obtain a confidence level sequence; According to the confidence sequence, the maximum confidence of the suspected positive cells is obtained.

5. The method according to claim 1, characterized in that Get pre-built classification models, including: Obtaining a difficult slice sample set based on the cervical cell whole slice samples with high difficulty in positive determination of the cell detection model; The residual network is trained according to the difficult slice sample set to obtain a pre-built classification model.

6. The method according to claim 5, characterized in that Obtaining a whole section sample of cervical cells that is difficult to determine as positive by the cell detection model includes: Obtain a collection of whole cervical cell slide samples; According to the pre-built cell detection model, the confidence distribution of each cervical cell whole section sample is obtained; According to the confidence distribution of each cervical cell whole section sample, an average confidence distribution is obtained; According to the distribution distance between the confidence distribution of each cervical cell whole section sample and the average confidence distribution, a high-difference whole section sample is obtained to determine the cervical cell whole section sample with high difficulty in positive judgment of the cell detection model.

7. A positive detection device for cervical cells, characterized in that: The device comprises: The abnormal cell detection module is used to detect abnormal cells in a small block of cervical cell images according to a pre-built cell detection model, and obtain a number of abnormal cell detection results; the small block of cervical cell images are obtained by cropping the full slice image of cervical cells field by field; A heat map acquisition module, used for mapping the plurality of abnormal cell detection results to corresponding positions of the full-slice image of cervical cells to obtain a positive confidence distribution heat map; A suspected positive cell acquisition module is used to perform visual field sampling on the visual fields represented by several grids of the positive confidence distribution heat map to obtain cells in several visual fields as suspected positive cells; A positive judgment module, used for determining that the detection result of the whole cervical cell slice image is positive when the maximum confidence of the suspicious positive cells is greater than the high-risk cell threshold; The positive judgment module is also used to input the cervical cell whole section image into a pre-built classification model to determine whether the cervical cell whole section image is positive when the maximum confidence of the suspicious positive cells is less than the high-risk cell threshold.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images