CN116797588A

CN116797588A - Diagnostic method, device, equipment and storage medium for abnormal cells

Info

Publication number: CN116797588A
Application number: CN202310801324.4A
Authority: CN
Inventors: 车拴龙; 危桂坚; 李淑燕; 刘栋; 梁贺华; 姚南燕; 钟学军; 冯晓冬
Original assignee: Guangzhou Kingmed Diagnostics Central Co Ltd
Current assignee: Guangzhou Kingmed Diagnostics Central Co Ltd
Priority date: 2023-06-30
Filing date: 2023-06-30
Publication date: 2023-09-22

Abstract

The application discloses a diagnosis method, a device, equipment and a storage medium for abnormal cells, which comprise the following steps: carrying out digital scanning on the cell pathological section to be analyzed, and identifying cell clusters in the obtained digital pathological section images to obtain a cell cluster to be analyzed containing the cell clusters to be analyzed; inputting a first digital pathological section image to be analyzed into a first cell analysis model, and obtaining the probability that each cell mass to be analyzed is an abnormal pathological cell mass; screening a preset number of cell clusters to be analyzed to serve as suspected pathological cell clusters according to the probability; and displaying the digital pathological section image to be analyzed corresponding to the suspected pathological cell mass to the user. According to the application, abnormal pathological cells of the cell mass are judged through the neural network, so that the defect or deficiency of the prior art on cell mass diagnosis is overcome, the accuracy of pathological cell diagnosis is improved, the missed diagnosis rate and the misdiagnosis rate are reduced, and the pathological cell thinning diagnosis result is more accurate and convincing.

Description

Diagnostic method, device, equipment and storage medium for abnormal cells

Technical Field

The application relates to the technical field of artificial intelligence and medical diagnosis, in particular to a method, a device, equipment and a storage medium for diagnosing abnormal cells.

Background

With the rapid development of deep learning technology, artificial intelligence is increasingly used in the medical field. In the field of medical images, research and application of deep learning models is also under high-speed development. The deep learning model is used for carrying out auxiliary screening and auxiliary diagnosis on the cytopathology image, so that the cost and time of repeated labor of doctors can be greatly reduced, and the sensitivity and accuracy of screening various diseases such as cancers can be obviously improved.

At present, institutions at home and abroad are actively exploring and applying deep learning models to screening and diagnosing cancers. The existing industry cytology deep learning model has higher sensitivity to scattered cytopathy, but has lower sensitivity to other non-scattered cells. Leading to the improvement of the recognition and judgment capability of the cytopathology image.

Disclosure of Invention

The application mainly aims to provide a diagnosis method, device, equipment and storage medium for abnormal cells, which can solve the technical problems that the existing deep learning model is mainly used for identifying and judging scattered cells in the prior art, but the identification sensitivity of the non-scattered cells is low.

To achieve the above object, the present application provides, in a first aspect, a method for diagnosing abnormal cells, the method comprising:

carrying out digital scanning on a cell pathological section to be analyzed to obtain a plurality of digital pathological section images;

identifying cell clusters in the digital pathological section image to obtain a first digital pathological section image to be analyzed containing the cell clusters to be analyzed;

inputting a first digital pathological section image to be analyzed into a first cell analysis model, and analyzing cell clusters to be analyzed by using the first cell analysis model to obtain a first probability that each cell cluster to be analyzed is an abnormal pathological cell;

screening a first preset number of cell clusters to be analyzed to serve as suspected pathological cell clusters according to the first probability;

and displaying the first to-be-analyzed digital pathological section image corresponding to the suspected pathological cell mass to the user.

In order to achieve the above object, a second aspect of the present application provides a diagnostic device for abnormal cells, comprising:

the first scanning module is used for carrying out digital scanning on the cell pathological section to be analyzed to obtain a plurality of digital pathological section images;

the first target determining module is used for identifying the cell mass in the digital pathological section image to obtain a first digital pathological section image to be analyzed containing the cell mass to be analyzed;

The first cell analysis module is used for inputting a first digital pathological section image to be analyzed into a first cell analysis model, analyzing the cell mass to be analyzed by using the first cell analysis model, and obtaining a first probability that each cell mass to be analyzed is an abnormal pathological cell mass;

the suspected cell mass determining module is used for screening out a first preset number of cell masses to be analyzed as suspected pathological cell masses according to the first probability;

the first display module is used for displaying the first digital pathological section image to be analyzed corresponding to the suspected pathological cell mass to the user.

To achieve the above object, a third aspect of the present application provides a computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of:

inputting a first digital pathological section image to be analyzed into a first cell analysis model, and analyzing the cell mass to be analyzed by using the first cell analysis model to obtain a first probability that each cell mass to be analyzed is an abnormal pathological cell mass;

Screening a first preset number of cell clusters to be analyzed to serve as suspected pathological cell clusters according to the first probability; and displaying the first to-be-analyzed digital pathological section image corresponding to the suspected pathological cell mass to the user.

To achieve the above object, a fourth aspect of the present application provides a computer apparatus including a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of:

The embodiment of the application has the following beneficial effects:

the application provides a method for distinguishing abnormal pathological cells of a cell mass through a neural network, which overcomes the defects or the shortages of the prior art on the diagnosis of the cell mass, further improves the accuracy of the diagnosis of the pathological cell, reduces the missed diagnosis rate and the misdiagnosis rate, and ensures that the diagnosis result of the pathological cell is more accurate and more convincing. Especially in the cervical cancer diagnosis field, the missing diagnosis rate of cervical abnormal cell clusters is reduced, the accurate screening rate of glandular epithelial lesion cell clusters and abnormal squamous epithelial cell clusters is improved, and the auxiliary diagnosis effect is improved.

Drawings

In order to more clearly illustrate the embodiments of the application or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Wherein:

FIG. 1 is a flow chart of a method for diagnosing abnormal cells according to an embodiment of the present application;

FIG. 2 is a graph showing the comparative effect of scattered cells and cell clusters in an embodiment of the present application;

FIG. 3 is a graph showing the comparative effects of different cell clusters according to the embodiment of the present application;

FIG. 4 is a graph showing the contrast effect of different digital pathological section images obtained by multi-layer digital scanning in an embodiment of the present application;

FIG. 5 is a block diagram showing a diagnostic apparatus for abnormal cells according to an embodiment of the present application;

fig. 6 is a block diagram of a computer device in an embodiment of the application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

In one embodiment, as shown in FIG. 1, a method of diagnosing abnormal cells is provided. The method can be applied to the terminal and the server. The diagnosis method of the abnormal cells specifically comprises the following steps:

s100: and carrying out digital scanning on the cell pathological section to be analyzed to obtain a plurality of digital pathological section images.

Specifically, a pathological scanner is utilized to scan the pathological section of the cell to be analyzed area by area, and each obtained digital pathological section image is an area image.

The cytopathological section to be analyzed can be any cytopathological sample, such as thyroid cells, urine cells, coelomic fluid cells, cervical cells and the like.

S200: and identifying the cell mass in the digital pathological section image to obtain a first digital pathological section image to be analyzed containing the cell mass to be analyzed.

Specifically, there are a plurality of obtained digital pathological section images, and cell clusters are not present in the possible digital pathological section images, and the cell clusters in the possible digital pathological section images do not meet the pixel requirements and other adverse factors. Therefore, it is necessary to pick out the digital pathological section image having the cell mass, and acquire the digital pathological section image satisfying the pixel requirements as much as possible as the first digital pathological section image to be analyzed. Thus, the idle work can be reduced, and the recognition efficiency of abnormal cell clusters can be accelerated.

FIG. 2 is a graph showing the comparative effect of the scattered cells and cell clusters in the example of the present application, referring to FIG. 2, 4 cells are used to form a cluster as one cell cluster, and a single cell is used as the scattered cell. Of course, the cell mass may be defined as a mass of 2 or more cells, i.e., a cell mass, and the present application is not limited thereto.

S300: inputting the first to-be-analyzed digital pathological section image into a first cell analysis model, and analyzing the to-be-analyzed cell clusters by using the first cell analysis model to obtain a first probability that each to-be-analyzed cell cluster is an abnormal pathological cell cluster.

Specifically, the first cell analysis model is a trained cell analysis model for identifying normal cell clusters and abnormal cell clusters, and outputting the probability that each cell cluster to be analyzed is a normal cell and the probability that each cell cluster is an abnormal pathological cell.

If the probability of being a normal cell is greater than the probability of being an abnormal diseased cell, the probability of the cell mass to be analyzed being a normal cell is greater; if the probability of being an abnormal diseased cell is greater than that of being a normal cell, the probability of the cell mass to be analyzed being an abnormal diseased cell is greater.

S400: and screening out a first preset number of cell clusters to be analyzed as suspected pathological cell clusters according to the first probability.

Specifically, the first probability of the selected suspected pathological cell mass is as high as possible, and the probability of the selected suspected pathological cell mass being an abnormal pathological cell is larger than that of the selected suspected pathological cell mass being a normal cell.

S500: and displaying the first to-be-analyzed digital pathological section image corresponding to the suspected pathological cell mass to the user.

Specifically, a first digital pathological section image to be analyzed corresponding to the suspected pathological cell mass is displayed to a user, so that the user can conveniently review and analyze the image.

For example, the first 20 suspected diseased cell clusters are preferably presented on the computer screen for review and analysis by a physician.

FIG. 3 is a graph showing the comparative effect of different cell clusters according to the embodiment of the present application, and referring to FIG. 3, cell cluster 1 in the upper left circle is a normal cervical tubular gland epithelial cell; cell mass 3 in the upper right circle is a suspected abnormal cell mass; the cell mass 4 in the lower left circle is AGC (atypical glandular epithelial cell), which is a diseased cell mass; cell mass 6 in the lower right circle is HSIL (high grade diseased squamous epithelial cells).

The diagnosis method of abnormal cells of the present embodiment can be applied to the auxiliary diagnosis of various diseases. Such as cervical cancer, etc. Cervical cancer is one of the common malignant tumors of women, and early detection and treatment can greatly improve the cure rate. The traditional cervical cancer screening method is to observe and diagnose scraped cells under a microscope through cervical scraping. The method needs a doctor with abundant experience to judge, and has the problems of high subjectivity and misdiagnosis rate. And cervical cancer screening is performed by using the deep learning model, so that the artificial interference and subjectivity can be reduced, and the screening accuracy is improved. When squamous epithelial cells are abnormal, scattered cells are used as main materials and cell clusters are used as auxiliary materials. When the glandular epithelial cells are abnormal, the cell clusters are mainly scattered on the cells rarely. The deep learning model in the present stage has low sensitivity to glandular epithelial cytopathy and high missed diagnosis rate. Sensitivity to cervical lesion cells and screening effect can be provided by this embodiment.

In the embodiment, abnormal pathological cell discrimination is carried out on the cell mass through the neural network, so that the defect or deficiency of the prior art on cell mass diagnosis is overcome, the accuracy of pathological cell diagnosis is further improved, the missed diagnosis rate and the misdiagnosis rate are reduced, and the pathological cell thinning diagnosis result is more accurate and convincing. Especially in the cervical cancer diagnosis field, the missing diagnosis rate of cervical abnormal cell clusters is reduced, the accurate screening rate of glandular epithelial lesion cell clusters and abnormal squamous epithelial cell clusters is improved, and the auxiliary diagnosis effect is improved.

In the embodiment, the deep learning model is utilized to carry out auxiliary screening and auxiliary diagnosis on the cytopathology image, so that the cost and time of repeated labor of doctors can be greatly reduced, and the sensitivity and accuracy of cancer screening are obviously improved.

In one embodiment, the first cell analysis model is trained by:

carrying out multi-layer digital scanning on a known cell pathological section to obtain a plurality of cell mass image sets, wherein each cell mass image set comprises a plurality of cell mass images under the same x-axis coordinate, y-axis coordinate and different z-axis coordinates, and the different cell mass image sets correspond to different x-axis coordinates and/or y-axis coordinates;

Constructing a first data set according to the cell mass image, wherein the label of the cell mass image comprises abnormal pathological cell mass or normal cell mass;

training the first pre-trained cell analysis model by using the first data set to obtain a first cell analysis model.

Specifically, taking cervical cells as an example, in the cervical cells, the consistency of labeling results for the glandular epithelial cell mass is lower, so that the scale of a training data set for training the glandular epithelial cell mass is small, a deep learning model is caused to generate systematic bias, the sensitivity to scattered squamous cell abnormal lesions is more and more sensitive, and the sensitivity to agglomerated glandular epithelial cell abnormal lesions is always lower; cell clusters tend to be natural three-dimensional multilaminate structures, or cells aggregate into three-dimensional multilaminate structures during the tableting process. At present, after digital pathological scanning, single-layer or low-layer scanning files are mostly used.

Cervical cells often exist in the form of scattered or clustered cells on pathological sections. The embodiment mainly focuses on the technical problem that the current deep learning model is high in lesion cell mass missed diagnosis rate. Therefore, a large number of image samples containing cell clusters need to be collected. Liquid-based cytopathology smears often present two-dimensional images on the two-dimensional XY axis on scanners and computer displays. In the real world, the single-layer image scanning is performed by the lens of the scanner after focusing on a single plane on the Z axis. If a scan of successive n layers is performed in the Z-axis, the cell mass can be represented on successive layers due to its large volume, and different areas of the cell mass may have a change in image sharpness (clearer or more blurred). Some cells may become larger or smaller due to problems with the focal plane. So that more detailed images of the cell mass can be obtained. Similar to the zooming of the camera, the details and contours of the players at different locations of the population on the football field can be seen, see in particular fig. 4. Multiple different cell mass images can be obtained with the Z axis changed under the same X and Y axes.

The images of the cell mass acquired on the same XY axis and on different Z axes are all classified as the same class. Thereby, the cell mass images of n different layers are obtained by marking on any Z axis once. The amount of annotation data is extended by a factor of n. The purpose of effectively expanding the sample is achieved.

According to the method, after the X-axis coordinate and/or the Y-axis coordinate are changed, the X-axis and the Y-axis are kept unchanged, and the Z-axis is changed again, so that a plurality of cell mass images of the round can be obtained and a cell mass image set of the round can be formed.

The first cell analysis model of this embodiment is used to distinguish between normal cell clusters and abnormal cell clusters, and therefore, the label of the cell cluster image includes abnormal diseased cell clusters or normal cell clusters.

In one embodiment, the image of the cell mass may include a plurality of cell masses in addition to the cell mass, and therefore, it is necessary to frame the cell mass in the image of the cell mass and calculate the coordinates of the frame of the cell mass, and at the same time, it is necessary to label each of the frame-selected cell masses as an abnormal diseased cell mass or a normal cell mass.

The method comprises the steps that when training is carried out, a first pre-training cell analysis model needs to learn the characteristics of a framed cell mass, predicts whether the framed cell mass is an abnormal pathological cell mass or a normal cell mass, calculates a loss function according to a prediction result and a real label, updates model parameters according to the loss function, selects a first sample from a first data set again, and inputs the first sample into the first pre-training cell analysis model until the model converges. The model convergence condition is that the preset iteration times are reached, or the loss function value is smaller than the loss threshold value.

The model parameters at the completion of model training are those of the trained first cell analysis model.

Of course, as the data is accumulated, further iterative updates to the first cell analysis model may also be made to enhance the performance of the first cell analysis model.

The application of deep learning models in the medical image field presents several problems and challenges. For example, the number and quality of data sets have an important impact on the training and effectiveness of the model, but medical image data acquisition and labeling are difficult, and the size and diversity of data sets are limited. The cell mass labeling of the traditional random monolayer is often limited in labeling quantity, and is limited to the problem of random monolayer image definition, and the labeling quality is often poor. Therefore, the detection rate of abnormal cell clusters becomes one of the bottlenecks restricting the wide range of application of the artificial intelligent cervical cell pathology auxiliary screening system at the present stage.

According to the embodiment, based on a scanning imaging principle, continuous n-layer scanning is performed on the Z axis, a large number of cell mass images can be obtained, the labeling quantity and quality of the cell mass can be greatly improved, the expansion of a data set in the medical field is realized, and the performance of a neural network model is further improved.

In addition, the embodiment supplements a new method for identifying and analyzing the cell mass by using a neural network model on the basis of the existing scattered cell identification and interpretation, and improves the detection sensitivity of abnormal cell mass. Can greatly improve the abnormal detection rate of cervical gland epithelial cells and avoid the missed diagnosis of cervical adenocarcinoma which is a second common tumor in cervical cancer screening. The detection rate of the partially agglomerated squamous cell lesions is also improved.

In one embodiment, the method further comprises:

identifying scattered cells in the digital pathological section image to obtain a second digital pathological section image to be analyzed containing scattered cells to be analyzed;

inputting a second digital pathological section image to be analyzed into a second cell analysis model, and analyzing scattered cells to be analyzed by using the second cell analysis model to obtain a second probability that each scattered cell to be analyzed is an abnormal pathological cell;

screening a second preset number of scattered cells to be analyzed as suspected lesion scattered cells according to the second probability;

and displaying a second digital pathological section image to be analyzed corresponding to the suspected pathological change scattered cells to the user.

Specifically, there are a plurality of obtained digital pathological section images, and there are no scattered cells in the possible digital pathological section images, and the scattered cells in the possible digital pathological section images do not meet the pixel requirements, and other adverse factors. Therefore, it is necessary to pick out the digital pathological section image scattered on the cells, and acquire the digital pathological section image satisfying the pixel requirements as much as possible as the second digital pathological section image to be analyzed. Therefore, the idle work can be reduced, and the recognition efficiency of the abnormal scattered cells can be accelerated.

FIG. 2 is a graph showing the comparative effect of the scattered cells and cell clusters in the example of the present application, referring to FIG. 2, 4 cells are used to form a cluster as one cell cluster, and a single cell is used as the scattered cell. Of course, it may be defined as consisting of 2 cells and as scattered cells, and the present application is not limited in this regard.

The second cell analysis model is a trained cell analysis model and is used for identifying normal scattered cells and abnormal lesion scattered cells and outputting the probability that each cell mass to be analyzed is the normal scattered cells and the probability that each cell mass to be analyzed is the abnormal lesion scattered cells.

If the probability of being normal scattered cells is greater than the probability of being abnormal lesion scattered cells, the probability that the scattered cells to be analyzed are normal scattered cells is greater; if the probability of being scattered in the cells for an abnormal lesion is greater than the probability of being scattered in the cells for a normal lesion, the probability of being scattered in the cells for an abnormal lesion to be analyzed is greater.

The second probability of the selected suspected lesion scattered cells is as high as possible, and the probability of the selected suspected lesion scattered cells being abnormal lesion scattered cells is larger than the probability of the selected suspected lesion scattered cells being normal scattered cells.

And displaying a second digital pathological section image to be analyzed corresponding to the suspected pathological change scattered cells to the user, so that the user can conveniently audit and analyze the images.

For example, the first 50 most probable suspected lesions are preferably scattered on the cell for review and analysis by the physician.

In a specific embodiment, the first cell analysis model and the second cell analysis model are the same cell analysis model. The tag includes abnormal diseased cells, normal cells. The training sample includes images containing cell clusters, as well as images containing scattered cells. The abnormal pathological cell clusters or scattered cells are marked as abnormal pathological cells, and the normal cell clusters or scattered cells are marked as normal cells.

In the embodiment, the abnormal pathological change cells are judged while the abnormal pathological change cells are judged on the cell clusters, and whether the risk of cervical lesions exists or not is comprehensively estimated according to the judging results of the cell clusters and the scattered cells, so that the accuracy of cervical lesion diagnosis is further improved, and the missing rate is reduced.

In one embodiment, step S400 specifically includes:

screening out the cell mass to be analyzed, the first probability of which exceeds a probability threshold value, as a suspected pathological cell mass;

or alternatively, the process may be performed,

descending order is carried out on the first probability, and the cell mass to be analyzed corresponding to the first probability which is ranked lower than a first threshold value in the descending order is used as the suspected pathological cell mass;

Or alternatively, the process may be performed,

and carrying out ascending order on the first probability, and taking the cell mass to be analyzed corresponding to the first probability ranked higher than the second threshold value in the ascending order as the suspected pathological cell mass.

Specifically, the probability threshold is configured according to the actual situation, which is not limited by the present application. The setting of the probability threshold determines a first preset number of selected suspected diseased cell clusters.

Or, performing descending order or ascending order on all the first probabilities, selecting a first preset number of maximum first probabilities according to the ordering result, and taking the cell mass to be analyzed corresponding to the selected maximum first probabilities as a suspected pathological cell mass.

The first threshold and the second threshold determine the first preset number, and the specific value is configured according to the actual situation, which is not limited by the present application.

According to the method, the most probable suspected lesion cell mass is screened out in a sorting mode or a threshold comparison mode according to the first probability of the abnormal lesion cell mass, so that a doctor can accurately judge whether the risk of cervical lesions exists according to the suspected lesion cell mass.

In one embodiment, step S200 includes:

identifying a cell mass in the digital pathological section image;

Performing definition analysis on the identified cell clusters to obtain definition of each cell cluster;

judging whether the cell mass meets a first preset definition requirement according to the definition;

if the first preset definition requirement is not met, re-executing the digital scanning until the first preset definition requirement is met;

and after the first preset definition requirement is met, taking the digital pathological section image corresponding to the cell mass meeting the second preset definition requirement as a first digital pathological section image to be analyzed.

Specifically, after the pathological section of the cell to be analyzed is digitally scanned, the cell mass is identified and segmented, and the segmentation is used for segmenting the cell mass or scattered cells in the image from the background, so that the cell mass and the scattered cells can be identified and distinguished conveniently.

For example, cell clusters having a diameter greater than 30 microns (average diameter of individual cells 10-30 microns) may be screened. As in fig. 2, a cell mass of 4 cells is taken as an example. For example, 1000 cell clusters were selected from the whole sheet and dispersed in 9000 cells. And (3) carrying out definition analysis on the screened cell clusters to obtain the definition of each cell cluster. The sharpness may be obtained by analyzing the gray scale of the edges of the cell mass. For example, the more blurred the edge, the more gradual the pixel gray value, and the lower the sharpness. Conversely, the sharper the edge, the more abrupt the pixel gray value, and the higher the sharpness. Of course, the image containing the cell mass may also be displayed with the definition determined by human beings.

The first preset sharpness requirement is specifically that the sharpness of the third preset number or preset proportion of cell clusters exceeds a sharpness threshold, or that the third preset number or preset proportion of cell clusters is determined to meet the sharpness requirement. The definition requirements, in particular the definition of the cell mass, can be met to the extent that the naked human eye can be used to interpret the details of the cell structure.

In one embodiment, the sharpness is ordered to select a third predetermined number of cell clusters of maximum sharpness, and rescanning is performed if some or all of the cell clusters of maximum sharpness do not meet the sharpness requirement.

For example, the first 30% of the cell mass is selected for clarity determination, and rescanning if the 30% of the cell mass is not available to the unaided human eye for interpretation of cell structural details.

If the cell mass meets the first preset definition requirement, selecting the cell mass meeting the second preset definition requirement from the cell mass, and taking the digital pathological section image corresponding to the selected cell mass as a first digital pathological section image to be analyzed.

The second predetermined definition requirement may be a fourth predetermined number or predetermined proportion of the cell clusters having the highest definition.

For example, a cell mass with a definition of the first 30% or the first 10% is selected as the cell mass to be analyzed.

FIG. 3 is a graph showing the comparative effect of different cell masses in the example of the present application, and FIG. 3 shows that cell mass 1 in the upper left circle is normal cervical tubular gland epithelial cells with a sharpness of 80%; cell mass 3 in the upper right circle is a suspected abnormal cell mass, and the definition is 50%; the cell mass 4 in the lower left circle is AGC (atypical glandular epithelial cell), and is lesion cell mass with definition of 90%; cell mass 6 in the lower right circle is HSIL (high grade diseased squamous epithelial cells) with a definition of 60%. The definition of the cell mass 5 in the circle at the lower left corner is 20%, and the cell mass is judged to be invisible, so that the definition requirement is not met; the cell mass 2 in the circle at the upper right corner has 20% of definition, and is judged to be invisible, and the definition requirement is not satisfied.

According to the embodiment, whether the photographed digital pathological section image meets the definition requirement is determined through definition, and under the condition that the photographed digital pathological section image does not meet the definition requirement, the photographed digital pathological section image is scanned again until the photographed digital pathological section image with high quality and definition is photographed, so that noise interference is reduced, the cervical lesion cell diagnosis result is more accurate, the convincing effect is achieved, and the missed diagnosis and misdiagnosis probability is reduced.

In one embodiment, determining whether the cell mass meets a first predetermined definition requirement based on the definition comprises:

counting the number of cell clusters with definition meeting the first preset definition requirement;

if the statistical number does not exceed the number threshold, or if the ratio of the statistical number to the total number of cell clusters does not exceed the ratio threshold, determining that the cell clusters do not meet the first preset definition requirement.

Specifically, the total number of cell clusters is the sum of the number of identified cell clusters. The definition meets a first predetermined definition requirement, specifically the definition of the cell mass exceeding a definition threshold.

And counting the cell clusters with the definition exceeding the definition threshold to obtain the statistical quantity. If the statistical number does not exceed the number threshold, or the ratio of the statistical number to the total number of the cell clusters does not exceed the ratio threshold, determining that the cell clusters obtained by the round of scanning do not meet the first preset definition requirement, and needing to be scanned again.

If the statistical quantity exceeds the quantity threshold, or the ratio of the statistical quantity to the total quantity of the cell clusters exceeds the ratio threshold, the cell clusters obtained by the round of scanning are judged to meet the first preset definition requirement, and rescanning is not needed.

In this embodiment, by counting the cell clusters exceeding the sharpness threshold, whether the sharpness of the cell clusters meets the first preset sharpness requirement is accurately determined according to the statistical number or the ratio of the cell clusters in the total number, so as to determine whether rescanning is required. The embodiment rescans when the proportion of unclear cell clusters is too high due to scanning failure, ensures that high-quality clear digital pathological section images are acquired, reduces noise interference, ensures that the cervical lesion cell diagnosis result is more accurate, has persuasion and reduces the probability of missed diagnosis and misdiagnosis.

In one embodiment, identifying a cell mass in an image of a digital pathological section comprises:

digitally scanning known cytopathology sections to obtain a plurality of scattered cell images;

constructing a second data set according to the obtained cell mass image and the scattered cell image, wherein the label of the second data set comprises cell mass or scattered cells;

Training the pre-trained cell type analysis model by using the second data set to obtain a trained cell type analysis model;

and identifying scattered cells and cell clusters in the digital pathological section image of the cell pathological section to be analyzed by using the trained cell type analysis model.

In particular, cervical cells often exist in the form of scattered or clustered cells on pathological sections. The embodiment mainly focuses on the technical problem that the current deep learning model is high in lesion cell mass missed diagnosis rate. Therefore, a large number of image samples containing cell clusters need to be collected. Liquid-based cytopathology smears often present two-dimensional images on the two-dimensional XY axis on scanners and computer displays. In the real world, the single-layer image scanning is performed by the lens of the scanner after focusing on a single plane on the Z axis. If a scan of successive n layers is performed in the Z-axis, the cell mass can be represented on successive layers due to its large volume, and different areas of the cell mass may have a change in image sharpness (clearer or more blurred). Some cells may become larger or smaller due to problems with the focal plane. So that more detailed images of the cell mass can be obtained. Similar to the zooming of the camera, the details and contours of the players at different locations of the population on the football field can be seen, see in particular fig. 4. Multiple different cell mass images can be obtained with the Z axis changed under the same X and Y axes.

The cell type analysis model of this example is used to distinguish between cell clusters and scattered cells, so the label of the cell cluster image includes cell clusters and the label of the scattered cell image includes scattered cells.

In this example, a method for single-layer labeling of cytopathic effect and multi-layer XY layer expansion of a cell training set was established. And labeling few samples, and achieving the labeling effect of large samples. On the Z axis, the best definition layer can be combined with multiple layers of information, and finally the accuracy and consistency of labeling are improved.

In one embodiment, there may be other tissues in the image of the cell mass or the image of the scattered cells in addition to or in addition to the cell mass or the scattered cells, and thus, it is necessary to frame the cell mass and the scattered cells in the image, calculate the coordinates of the cell mass frame and the coordinates of the scattered cell frame, and label the cell mass of each frame as the cell mass and the scattered cells as the scattered cells.

The pre-trained cell type analysis model needs to learn the characteristics of the framed cell clusters or scattered cells during training, predicts whether the framed cell clusters or scattered cells are of the cell clusters or scattered cells, calculates a loss function according to a prediction result and a real label, updates model parameters according to the loss function, and selects a second sample from the second data set again to be input into the pre-trained cell type analysis model until the model converges. The model convergence condition is that the preset iteration times are reached, or the loss function value is smaller than the loss threshold value.

The model parameters at the completion of model training are those of the trained cell type analysis model.

Of course, as data is accumulated, further iterative updates to the trained cell type analysis model may also be made to enhance the performance of the trained cell type analysis model.

And identifying scattered cells and cell clusters in the digital pathological section image of the cell pathological section to be analyzed by using the trained cell type analysis model to obtain a first digital pathological section image to be analyzed containing the cell clusters to be analyzed and a second digital pathological section image to be analyzed containing the scattered cells to be analyzed.

The embodiment realizes the accurate distinction between the cell clusters and the scattered cells through the neural network model. So as to accurately and rapidly screen out the first to-be-analyzed digital pathological section image containing the cell mass to be analyzed, and improve the screening efficiency.

The application establishes a scheme for training, model establishment, identification and analysis of cell clusters. Based on the existing scattered cell identification and interpretation method, a new method for identifying and analyzing cell clusters is supplemented, so that the detection sensitivity of abnormal cell clusters is improved.

The method is particularly applied to the field of cervical cancer screening diagnosis, and based on the existing traditional cervical cell pathology deep learning model, the combined analysis strategy of the cell mass deep learning model is supplemented and added, so that the condition that the cervical cell pathology deep learning model in the present stage leaks diagnosis of glandular epithelial lesions is improved. Can greatly improve the abnormal detection rate of cervical gland epithelial cells and avoid the missed diagnosis of cervical adenocarcinoma, which is the second common tumor in cervical cancer screening. The detection rate of the partially agglomerated squamous cell lesions is also improved. The cell mass labeling of the traditional random monolayer is often limited in labeling quantity, and is limited to the problem of random monolayer image definition, and the labeling quality is often poor. The scheme can greatly improve the labeling quantity and quality of the cell mass. And the consistency and accuracy of labeling are improved. And the quantity of the labeling training sets for expanding the cell clusters is improved by using fewer labeling cell clusters.

Referring to fig. 5, the present application also provides a diagnostic apparatus for abnormal cells, the apparatus comprising:

the first scanning module 100 is configured to digitally scan a cytopathological section to be analyzed to obtain a plurality of digital pathological section images;

the first target determining module 200 is configured to identify a cell mass in the digital pathological section image, so as to obtain a first digital pathological section image to be analyzed including the cell mass to be analyzed;

the first cell analysis module 300 is configured to input a first digital pathological section image to be analyzed into a first cell analysis model, analyze a cell mass to be analyzed by using the first cell analysis model, and obtain a first probability that each cell mass to be analyzed is an abnormal pathological cell mass;

the suspected cell mass determination module 400 is configured to screen a first preset number of cell masses to be analyzed as suspected pathological cell masses according to a first probability;

the first display module 500 is configured to display a first digital pathological section image to be analyzed corresponding to the suspected pathological cell mass to a user.

In one embodiment, the apparatus further comprises:

the second scanning module is used for carrying out multi-layer digital scanning on the known cell pathological section to obtain a plurality of cell mass image sets, wherein each cell mass image set comprises a plurality of cell mass images under the same x-axis coordinate, y-axis coordinate and different z-axis coordinates, and the different cell mass image sets correspond to different x-axis coordinates and/or y-axis coordinates;

A first data set construction module for constructing a first data set from the cell mass image, wherein the label of the cell mass image comprises an abnormal pathological cell mass or a normal cell mass;

and the first training module is used for training the first pre-training cell analysis model by using the first data set to obtain the first cell analysis model.

In one embodiment, the apparatus further comprises:

the second target determining module is used for identifying scattered cells in the digital pathological section image to obtain a second digital pathological section image to be analyzed, wherein the second digital pathological section image contains scattered cells to be analyzed;

the second cell analysis module is used for inputting a second digital pathological section image to be analyzed into a second cell analysis model, analyzing scattered cells to be analyzed by using the second cell analysis model, and obtaining a second probability that each scattered cell to be analyzed is an abnormal pathological cell;

the suspected scattered cell determining module is used for screening out a second preset number of scattered cells to be analyzed as suspected lesion scattered cells according to the second probability;

the second display module is used for displaying a second digital pathological section image to be analyzed corresponding to the suspected pathological change scattered cells to the user.

In one embodiment, the suspected cell mass determination module specifically comprises:

the first screening module is used for screening out the cell mass to be analyzed, the first probability of which exceeds the probability threshold value, as the suspected pathological cell mass;

or alternatively, the process may be performed,

the second screening module is used for descending order of the first probability and taking the cell mass to be analyzed corresponding to the first probability which is ranked lower than a first threshold value in the descending order as a suspected lesion cell mass;

or alternatively, the process may be performed,

and the third screening module is used for carrying out ascending order on the first probability, and taking the cell mass to be analyzed corresponding to the first probability ranked higher than the second threshold value in the ascending order as the suspected pathological cell mass.

In one embodiment, the first targeting module 200 specifically includes:

the cell distinguishing module is used for identifying cell clusters in the digital pathological section image;

the definition acquisition module is used for carrying out definition analysis on the identified cell clusters to obtain definition of each cell cluster;

the definition checking module is used for judging whether the cell mass meets a first preset definition requirement according to definition;

the rescanning module is used for re-executing the digital scanning until the first preset definition requirement is met if the first preset definition requirement is not met;

And the fourth screening module is used for taking the digital pathological section image corresponding to the cell mass meeting the second preset definition requirement as a first digital pathological section image to be analyzed after the first preset definition requirement is met.

In one embodiment, the sharpness checking module specifically includes:

the counting module is used for counting the number of cell clusters with definition meeting the first preset definition requirement;

and the judging module is used for judging that the cell mass does not meet the first preset definition requirement if the statistical quantity does not exceed the quantity threshold value or the ratio of the statistical quantity to the total quantity of the cell mass does not exceed the ratio threshold value.

In one embodiment, the cell differentiation module specifically comprises:

the third scanning module is used for carrying out multi-layer digital scanning on the known cell pathological section to obtain a plurality of cell mass image sets, wherein each cell mass image set comprises a plurality of cell mass images under the same x-axis coordinate, y-axis coordinate and different z-axis coordinates, and the different cell mass image sets correspond to different x-axis coordinates and/or y-axis coordinates;

the fourth scanning module is used for carrying out digital scanning on known cell pathological sections to obtain a plurality of scattered cell images;

The second data set construction module is used for constructing a second data set according to the obtained cell mass image and the scattered cell image, wherein the label of the second data set comprises cell mass or scattered cells;

the second training module is used for training the pre-trained cell type analysis model by utilizing a second data set to obtain a trained cell type analysis model;

and the subcellular differentiation module is used for identifying scattered cells and cell clusters in the digital pathological section image of the cell pathological section to be analyzed by using the trained cell type analysis model.

FIG. 6 illustrates an internal block diagram of a computer device in one embodiment. The computer device may specifically be a terminal or a server. As shown in fig. 6, the computer device includes a processor, a memory, and a network interface connected by a system bus. The memory includes a nonvolatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system, and may also store a computer program which, when executed by a processor, causes the processor to implement the steps of the method embodiments described above. The internal memory may also have stored therein a computer program which, when executed by a processor, causes the processor to perform the steps of the method embodiments described above. It will be appreciated by those skilled in the art that the structure shown in FIG. 6 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

In one embodiment, a computer device is provided comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of:

In one embodiment, a computer readable storage medium is provided, storing a computer program which, when executed by a processor, causes the processor to perform the steps of:

Those skilled in the art will appreciate that the processes implementing all or part of the methods of the above embodiments may be implemented by a computer program for instructing relevant hardware, and the program may be stored in a non-volatile computer readable storage medium, and the program may include the processes of the embodiments of the methods as above when executed. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The foregoing examples illustrate only a few embodiments of the application, which are described in detail and are not to be construed as limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of protection of the present application is to be determined by the appended claims.

Claims

1. A method for diagnosing an abnormal cell, the method comprising:

identifying the cell mass in the digital pathological section image to obtain a first digital pathological section image to be analyzed containing the cell mass to be analyzed;

inputting the first to-be-analyzed digital pathological section image into a first cell analysis model, and analyzing the to-be-analyzed cell clusters by using the first cell analysis model to obtain a first probability that each to-be-analyzed cell cluster is an abnormal pathological cell cluster;

Screening out a first preset number of cell clusters to be analyzed as suspected pathological cell clusters according to the first probability;

and displaying the first to-be-analyzed digital pathological section image corresponding to the suspected pathological cell mass to a user.

2. The method of claim 1, wherein the first cell analysis model is trained by:

carrying out multi-layer digital scanning on known cell pathological sections to obtain a plurality of cell mass image sets, wherein each cell mass image set comprises a plurality of cell mass images under the same x-axis coordinate, y-axis coordinate and different z-axis coordinates, and different cell mass image sets correspond to different x-axis coordinates and/or y-axis coordinates;

constructing a first data set according to the cell mass image, wherein the label of the cell mass image comprises an abnormal pathological cell mass or a normal cell mass;

and training a first pre-trained cell analysis model by using the first data set to obtain the first cell analysis model.

3. The method according to claim 1, wherein the method further comprises:

Inputting the second digital pathological section image to be analyzed into a second cell analysis model, and analyzing the scattered cells to be analyzed by using the second cell analysis model to obtain a second probability that each scattered cell to be analyzed is an abnormal pathological cell;

screening out a second preset number of scattered cells to be analyzed as suspected lesion scattered cells according to the second probability;

and displaying a second digital pathological section image to be analyzed corresponding to the suspected pathological change scattered cells to a user.

4. The method of claim 1, wherein screening a first predetermined number of cell clusters to be analyzed as suspected diseased cell clusters according to the first probability comprises:

or alternatively, the process may be performed,

the first probability is subjected to descending order, and the cell mass to be analyzed corresponding to the first probability which is ranked lower than a first threshold in the descending order is used as a suspected pathological cell mass;

or alternatively, the process may be performed,

5. The method of claim 1, wherein identifying the cell mass in the digital pathological section image results in a first digital pathological section image to be analyzed comprising the cell mass to be analyzed, comprising:

identifying a cell mass in the digital pathological section image;

if the first preset definition requirement is not met, re-executing digital scanning until the first preset definition requirement is met;

6. The method of claim 5, wherein said determining whether the cell mass meets a first predetermined definition requirement based on the definition comprises:

7. The method of claim 1 or 5, wherein said identifying a cell mass in said digital pathological section image comprises:

training a pre-trained cell type analysis model by using the second data set to obtain a trained cell type analysis model;

and identifying scattered cells and cell clusters in the digital pathological section image of the pathological section of the cell to be analyzed by using the trained cell type analysis model.

8. A diagnostic device for abnormal cells, the device comprising:

the first cell analysis module is used for inputting the first digital pathological section image to be analyzed into a first cell analysis model, analyzing the cell mass to be analyzed by using the first cell analysis model, and obtaining a first probability that each cell mass to be analyzed is an abnormal pathological cell mass;

the suspected cell mass determining module is used for screening a first preset number of cell masses to be analyzed to be the suspected pathological cell masses according to the first probability;

the first display module is used for displaying the first digital pathological section image to be analyzed corresponding to the suspected pathological cell mass to a user.

9. A computer readable storage medium storing a computer program, which when executed by a processor causes the processor to perform the steps of the method according to any one of claims 1 to 7.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program which, when executed by the processor, causes the processor to perform the steps of the method of any of claims 1 to 7.