CN115546163A

CN115546163A - Method, device, equipment and medium for identifying cervical exfoliated cell slide

Info

Publication number: CN115546163A
Application number: CN202211259379.9A
Authority: CN
Inventors: 佘银海; 程绍禹; 李昌众; 伍祥辰
Original assignee: Suzhou Ruiqian Technology Co ltd
Current assignee: Suzhou Ruiqian Technology Co ltd
Priority date: 2022-10-14
Filing date: 2022-10-14
Publication date: 2022-12-30

Abstract

The embodiment of the invention discloses a method, a device, equipment and a medium for identifying cervical exfoliated cell slides, wherein the method comprises the following steps: acquiring a first cervical exfoliated cell slide image acquired based on a first preset resolution, and identifying and segmenting a single cell image with the first preset resolution and a cell mass image with the first preset resolution; respectively carrying out cell classification and prediction on the single cell image with the first preset resolution and the cell mass image with the first preset resolution to obtain different types of positive cell prediction results; acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution, and identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image; and inputting the cell cluster image with the second preset resolution into the pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result. The embodiment of the invention realizes the rapid positioning of the cells and improves the accuracy and the efficiency of identifying the cervical exfoliated cell slide.

Description

Method, device, equipment and medium for identifying cervical exfoliated cell slide

Technical Field

The embodiment of the invention relates to the technical field of medical image processing, in particular to a method, a device, equipment and a medium for identifying cervical exfoliated cell slides.

Background

Thin-layer Cytology Test (TCT) or Liquid-Based Cytology Test (LCT) are one of the methods for screening cervical cancer. At present, the cervical exfoliated cell slide is mostly identified manually. However, the manual identification of the cervical exfoliated cell slide is inefficient, not only is time-consuming, but also the identification result depends on the experience of the doctor.

Disclosure of Invention

The embodiment of the invention provides a method, a device, equipment and a medium for identifying a cervical exfoliated cell slide, solves the problem of low manual identification efficiency of the cervical exfoliated cell slide, realizes rapid positioning of cells, and improves the accuracy and efficiency of identification of the cervical exfoliated cell slide.

In a first aspect, an embodiment of the present invention provides a method for identifying a cervical exfoliated cell slide, where the method includes:

acquiring a first cervical exfoliated cell slide image acquired based on a first preset resolution, and identifying and segmenting a single cell image with the first preset resolution and a cell cluster image with the first preset resolution in the first cervical exfoliated cell slide image;

respectively carrying out cell classification and prediction on the single cell image with the first preset resolution and the cell mass image with the first preset resolution to obtain different types of positive cell prediction results;

acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution based on the prediction results of different types of positive cells, and identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution;

and inputting the cell cluster image with the second preset resolution into the pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result.

In a second aspect, the present invention also provides an apparatus for identifying a cervical exfoliated cell slide, including:

the first image segmentation module is used for acquiring a first cervical exfoliated cell slide image acquired based on a first preset resolution, and identifying and segmenting a single cell image with the first preset resolution and a cell cluster image with the first preset resolution in the first cervical exfoliated cell slide image;

the first image classification module is used for respectively carrying out cell classification and prediction on the single cell image with the first preset resolution and the cell mass image with the first preset resolution to obtain different types of positive cell prediction results;

the second image segmentation module is used for acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution based on the prediction results of different types of positive cells, and identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution;

and the second image classification module is used for inputting the second preset resolution cell cluster image into the pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result.

In a third aspect, an embodiment of the present invention further provides a computer device, where the computer device includes:

one or more processors;

a memory for storing one or more programs;

when executed by one or more processors, cause the one or more processors to implement a method for identifying a cervical exfoliated cell slide as provided by any one of the embodiments of the present invention.

In a fourth aspect, embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored, where the program, when executed by a processor, implements the method for identifying a cervical exfoliated-cell slide as provided in any one of the embodiments of the present invention.

According to the technical scheme of the embodiment of the invention, a first cervix uteri exfoliated cell slide image acquired based on a first preset resolution is acquired, and a single cell image with a first preset resolution and a cell cluster image with a first preset resolution in the first cervix uteri exfoliated cell slide image are identified and segmented; respectively carrying out cell classification and prediction on the single cell image with the first preset resolution and the cell cluster image with the first preset resolution to obtain different types of positive cell prediction results; acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution based on the prediction results of different types of positive cells, and identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution; and inputting the cell cluster image with the second preset resolution into the pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result. The technical scheme of the embodiment of the invention solves the problem of low efficiency of manual identification of the cervical exfoliated cell slide, realizes quick positioning of cells, and improves the accuracy and efficiency of identification of the cervical exfoliated cell slide.

Drawings

Fig. 1 is a flowchart of a method for identifying a cervical exfoliated cell slide according to an embodiment of the present invention;

FIG. 2 is a flowchart of another method for identifying a cervical exfoliated cell slide according to an embodiment of the present invention;

FIG. 3 is a flow chart of cell classification and prediction for a first predetermined resolution single-cell image;

FIG. 4 is a flow chart of cell classification and prediction for a first predetermined resolution image of a cell mass;

FIG. 5 is a flow chart of another method for identifying a cervical exfoliated cell slide according to an embodiment of the present invention;

FIG. 6 is a flow chart of a method of identifying a cervical exfoliated cell slide;

FIG. 7 is a flow chart of a method for identifying a 10-fold under-the-mirror cervical exfoliated cell slide;

FIG. 8 is a flow chart of a 20-fold under-the-mirror cervical exfoliated cell slide identification method;

fig. 9 is a block diagram of a cervical exfoliated cell slide recognition apparatus provided by an embodiment of the present invention;

fig. 10 is a block diagram of a computer device according to an embodiment of the present invention.

Detailed Description

In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that the terms "first," "second," "third," and "fourth," etc. in the description and claims of the present invention and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

Fig. 1 is a flowchart of a method for identifying a cervical exfoliated cell slide according to an embodiment of the present invention, which is applicable to a scene of image identification of cervical exfoliated cells, and in particular, is more applicable to a case of identification of a cervical exfoliated cell slide in a TCT or LCT type slide making manner. The method can be executed by a cervical exfoliated cell slide recognition device, which can be realized by software and/or hardware and is integrated in a computer device with an application development function.

As shown in fig. 1, the method for identifying a cervical exfoliated cell slide according to this embodiment includes the following steps:

s110, acquiring a first cervical exfoliated cell slide image acquired based on a first preset resolution, and identifying and segmenting a single cell image with the first preset resolution and a cell cluster image with the first preset resolution in the first cervical exfoliated cell slide image.

Wherein the first preset resolution is a preset resolution for acquiring the image of the cervical exfoliated cell slide. For example, it may be the resolution of a high-throughput fully automated slide scanner, and by setting the resolution, the image of the cervical exfoliated cell slide at that resolution is acquired as the first image of the cervical exfoliated cell slide.

Optionally, the image of the cervical exfoliated cell slide is an image of the cervical exfoliated cell slide prepared in a TCT or LCT manner.

Specifically, a collected cervical exfoliated cell slide image with a preset resolution is obtained and used as a first cervical exfoliated cell slide image, nuclei are extracted from the first cervical exfoliated cell slide image by means of low-pass filtering, high-pass filtering, binarization, morphological calculation and the like, then the characteristics of the nuclei, such as area characteristics, nuclear-to-cytoplasmic ratios, nuclear optical density and the like, are calculated, and finally single cells and cell clusters in the cervical exfoliated cell slide image are distinguished by setting corresponding threshold values according to the characteristics, so that a corresponding first preset resolution single cell image and a corresponding first preset resolution cell cluster image are obtained.

And S120, respectively carrying out cell classification and prediction on the single cell image with the first preset resolution and the cell mass image with the first preset resolution to obtain positive cell prediction results of different categories.

It can be understood that the single cell image with the first preset resolution and the cell cluster image with the first preset resolution need to be processed by a classification and prediction algorithm, so as to obtain the positive cell types corresponding to the single cells and the cell clusters, that is, the prediction results of the positive cells of different types.

Common classification and prediction algorithms include: the method is used for classifying the identification of the cervical exfoliated cell slides and identifying different types of positive cells by using algorithms such as an NBC (Naive Bayesian classification) algorithm, an LR (Logistic regression) algorithm, an ID3 algorithm, an SVM (Support Vector Machine) algorithm, a KNN (K-Nearest Neighbor) algorithm, an ANN (Artificial Neural Network) algorithm and the like.

Optionally, the positive cell classification prediction result comprises: high-grade squamous epithelial lesion types, low-grade squamous intraepithelial lesion types, atypical squamous epithelial cell types which cannot exclude high-grade squamous intraepithelial lesions, and undefined atypical squamous cell types, and the like are used for describing screening results of the positive cervical exfoliated cells.

S130, acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution based on the prediction results of different types of positive cells, and identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution.

Specifically, for the different types of positive cell prediction results of the first cervical exfoliated cell slide image, the first several types of positive cells and cell groups, for example, the first five types of corresponding positive cells and cell groups, can be selected according to the confidence, and the cervical exfoliated cell slide image of the positive cells can be acquired with higher resolution according to the coordinates of the selected different types of positive cells and cell groups or other information for determining the positions of the positive cells and cell groups, and can be used as the second cervical exfoliated cell slide image, so that the image acquisition range is narrowed, and the automatic identification and classification process is more efficient; and identifying and segmenting nuclei of the second preset resolution cell cluster image in the second cervical exfoliated cell slide image. The second preset resolution is higher than the first preset resolution, and the high resolution and the low resolution are combined, so that the accuracy and the efficiency of identifying the cervical exfoliated cell slide are improved.

And S140, inputting the cell cluster image with the second preset resolution into the pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result.

It can be understood that the cascade cell classification model is formed by cascading all classification submodels, so that all classification submodels need to be pre-trained on the basis of different types of positive cell clusters and negative cell clusters marked by doctors, then a cell cluster image with a second preset resolution is input into the pre-trained cascade cell classification model, and finally the classification result of the cascade cell classification model is used as the identification result of the target cervical exfoliated cell slide.

According to the technical scheme of the embodiment of the invention, a first cervix uteri exfoliated cell slide image acquired based on a first preset resolution is acquired, and a single cell image with a first preset resolution and a cell cluster image with a first preset resolution in the first cervix uteri exfoliated cell slide image are identified and segmented; respectively carrying out cell classification and prediction on the single cell image with the first preset resolution and the cell cluster image with the first preset resolution to obtain different types of positive cell prediction results; acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution based on different types of positive cell prediction results, and identifying and segmenting a second preset resolution cell group image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution; and inputting the cell cluster image with the second preset resolution into the pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result. According to the technical scheme of the embodiment of the invention, the problem of low manual identification efficiency of the cervical exfoliated cell slide is solved, the cells are quickly positioned, and the accuracy and efficiency of identification of the cervical exfoliated cell slide are improved.

Fig. 2 is a flowchart of another method for identifying a cervical exfoliated cell slide according to an embodiment of the present invention, which belongs to the same inventive concept as the method for identifying a cervical exfoliated cell slide according to the foregoing embodiment, and further describes a process for cell classification and prediction of a single cell image with a first preset resolution and a cell cluster image with a first preset resolution on the basis of the foregoing embodiment, where the method can be executed by a cervical exfoliated cell slide identifying apparatus, and the apparatus can be implemented by software and/or hardware and is integrated in a computer device with an application development function.

S210, acquiring a first cervical exfoliated cell slide image acquired based on a first preset resolution, and identifying and segmenting a single cell image with the first preset resolution and a cell cluster image with the first preset resolution in the first cervical exfoliated cell slide image.

S220, carrying out cell classification and prediction on the single cell image with the first preset resolution to obtain positive cell prediction results of different categories.

Fig. 3 is a flowchart of cell classification and prediction for a single cell image with a first preset resolution, and as shown in fig. 3, the cell classification and prediction for the single cell image with the first preset resolution is performed to obtain prediction results of different types of positive cells, including the following steps:

s2201, extracting at least one preset image feature in the single cell image with the first preset resolution.

The preset image features may include: HOG (Histogram of oriented gradients) features, LBP (Local Binary Pattern) texture features, morphological features, histogram of gray scale, and statistical features of gray scale.

The HOG feature is a feature descriptor used for object detection in computer vision and image processing. The HOG characteristics form characteristics by calculating and counting a gradient direction histogram of a local area of the image, and have good invariance to image translation, rotation and illumination.

Specific extraction steps of HOG features:

1. color and gamma normalization: to reduce the influence of the illumination factor, the whole image needs to be normalized first. In the texture intensity of the image, the local exposure contribution of the surface layer has a large proportion, so the compression processing can effectively reduce the local shadow and illumination change of the image.

2. Calculating the image gradient: and calculating the gradient of the horizontal coordinate and the vertical coordinate of the image, and calculating the gradient direction value of each pixel position according to the gradient. The common method is as follows: the image is processed in one direction or in both the horizontal and vertical directions using a one-dimensional discrete differential template.

3. Each pixel in the histogram cell unit that constructs the direction votes for a certain direction-based histogram channel. The voting is in a weighted voting mode, that is, each vote has a weight, and the weight is calculated according to the gradient amplitude of the pixel point. The magnitude itself or a function thereof may be used to represent the weight, or a function of the magnitude may be selected to represent the weight, such as the square root of the magnitude, the square of the magnitude, a truncated form of the magnitude, or the like. The cell units may be rectangular or star-shaped. Histogram channels are evenly distributed in the range of 0-1800 (undirected) or 0-3600 (directed).

4. The cell units are organized into large intervals: due to the change of local illumination and the change of foreground-background contrast, the change range of the gradient intensity is very large, the gradient intensity needs to be normalized, and the normalization can further compress illumination, shadow and edges. Specifically, the individual cell units are grouped into large, spatially connected compartments. Thus, the HOG descriptor becomes a vector composed of histogram components of all cell units in each bin.

5. And (4) collecting HOG characteristics: and inputting the extracted HOG features into an SVM (Support Vector Machine) classifier, and searching a hyperplane as a decision function. The SVM is a generalized linear classifier for binary classification of data in a supervised learning mode.

The LBP is an algorithm for describing local texture features of the image, reflects the change condition of the texture around the pixel points of the image, and has the advantages of rotation invariance, gray scale invariance (no influence of illumination change), low calculation complexity and the like.

The specific extraction step of LBP characteristics: dividing the image into N-by-N windows, taking the pixel value at the center of the window as the pixel threshold value of the window, comparing the pixel value of the sub-window adjacent to the center of the window with the pixel threshold value of the window, if the pixel value of the sub-window is greater than the pixel threshold value of the window, marking the position of the sub-window as 1, otherwise, marking the position of the sub-window as 0, combining the 1 and 0 of each sub-window to generate the binary number of the corresponding N-by-N window, usually converting the binary number into a decimal number, wherein the decimal number is the LBP value of the pixel point at the center of the window, and reflecting the texture characteristic of the image by using the LBP value.

The morphological feature is a morphological feature of a nucleus, and since the morphology of the nucleus of the cervical cancer cell is changed, for example, the nucleus may be large, the ratio of nuclear plasma is increased, the nucleus is deformed, elongated, the nuclear side is jagged, the nucleus has a concave shape, a long bud, a lobular shape, a mulberry shape, or a meniscus shape, and thus, the feature such as the area, the perimeter, or the circularity of the nucleus is extracted as the morphological feature based on the division of the nucleus.

Wherein the segmentation of the nuclei comprises:

1. the coordinate system changes: the coordinate system of the image is transformed from the image coordinate system to the polar coordinate system, for example, from a rectangular coordinate system to the polar coordinate system.

2. Calculating the gradient of the image: the gradient calculation of the image is the speed of image change, and for the edge part of the image, the gray value change is large, and the gradient value is also large; for smoother parts of the image, the gray value variation is smaller, and the corresponding gradient value is also smaller. Strictly speaking, image gradient computation requires derivation, but image gradients generally get an approximation of the gradient (approximate derivative value) by computing the difference of pixel values.

3. And (3) calculating the shortest path based on dynamic planning: firstly, a topological sorting sequence in the graph is found and used as a recursion sequence, a shortest path can be obtained for all nodes of the topology in the graph, and since the dynamic programming is a global matching algorithm, pixels on the same polar line are optimized and solved, the boundary of a cell nucleus is extracted, and the influence caused by discontinuous edges due to noise and the like is avoided.

4. And reversely mapping the shortest path to an image coordinate system to obtain a closed area in the original image, wherein the area is a division area of a cell nucleus: and reversely mapping the shortest path from the polar coordinate system to an image coordinate system to obtain a closed region in the image, namely a cell nucleus region, and segmenting the region to obtain the cell nucleus.

Gray level histogram feature: the gray histogram is a function of gray level distribution, and is a statistic of gray level distribution in an image. The gray histogram is to count the occurrence frequency of all pixels in the digital image according to the size of the gray value. The gray histogram is a function of gray level, which represents the number of pixels in an image having a certain gray level, reflecting the frequency of occurrence of a certain gray level in the image. Specifically, the cell nucleus is taken as the center, the image is divided into five areas, namely, an upper left area, a lower left area, an upper right area, a lower right area and the cell nucleus, the gray level histograms are respectively counted and combined to form a final gray level histogram for describing the distribution characteristics of the cell nucleus and the cytoplasm.

Gray level statistical characteristics: the gray statistical characteristics reflect the distribution of image gray, and may be gray statistical characteristics such as a gray mean, a gray variance, a gray skewness, a gray kurtosis, a gray energy, and a gray entropy, for example. For example, the possible nucleus, cytoplasm and background region can be segmented by threshold segmentation. Specifically, a gray mean and a gray variance are respectively calculated, and a threshold is set for the gray mean and the gray variance of the cell nucleus and the cell cytoplasm and is used for distinguishing the cell nucleus and the cell cytoplasm; and setting a threshold value for the gray mean and the gray variance of the cytoplasm and the background region, and distinguishing the cytoplasm from the background region.

S2202, performing cell filtration based on at least one preset image characteristic to obtain a suspected positive cell image.

Specifically, one or more preset image features are extracted by using a cascade model, input into a corresponding SVM classifier, subjected to cell filtration, and sequentially filtered to obtain images of suspected positive cells.

Further, the method for inputting at least one preset image feature into the corresponding cascade cell filtering support vector machine classifier respectively comprises the following steps:

firstly, inputting gradient histogram features and gray statistic features in at least one preset image feature into a first support vector machine classifier to obtain a first classification result.

And then, inputting the texture features of the images in the first classification result into a second support vector machine classifier obtained by training on the basis of the first support vector machine classifier to obtain a second classification result.

And finally, inputting the morphological characteristics and the gray histogram characteristics of each image in the second classification result into a third support vector machine classifier obtained by training on the basis of the second support vector machine classifier.

Specifically, calculating a gradient histogram and gray statistical characteristics, inputting the gradient histogram and gray statistical characteristics into a first support vector machine classifier, and filtering blood cells; then calculating texture characteristics, inputting the texture characteristics into a second support vector machine classifier combined with the first support vector machine classifier, and filtering impurities and residual blood cells; and finally, calculating morphological characteristics and gray histogram characteristics of cell nuclei, inputting the morphological characteristics and gray histogram characteristics into a third support vector machine classifier combined with the second support vector machine classifier, and filtering negative cells, residual impurities and residual blood cells to obtain a suspected positive cell image.

S2203, inputting the suspected positive cell images into a preset single cell image cascade classification model to obtain positive cell prediction results of different categories.

The preset single cell image cascade classification model is a cascade model, and is used for classifying the suspected positive cell images to obtain different types of positive cells, for example, CNN (Convolutional Neural Network) can be used for classifying the suspected positive cell images.

The CNN is an artificial neural network, and the structure can be divided into 3 layers: the convolution layer is used for extracting features; the pooling layer is used for realizing downsampling; the full connectivity layer is used for classification. The CNN consists of one or more convolutional layers and a top fully-connected layer, also including associated weights and pooling layers.

Each convolution layer consists of a plurality of convolution units, and the parameters of each convolution unit are obtained through a back propagation algorithm. The convolution operation aims at extracting different input features, the first layer of convolution layer can only extract some low-level features such as edge, line, angle and other hierarchical features, and more layers of networks can iteratively extract complex features from the low-level features.

Pooling is actually a form of down-sampling. There are many different forms of non-linear pooling functions, where maximum pooling is the division of the input image into several rectangular regions, with a maximum output for each sub-region. The pooling layer will constantly reduce the spatial size of the data and hence the number of parameters and the amount of calculations will also decrease, which to some extent also controls the overfitting. Typically, pooling layers are inserted periodically between convolutional layers of the CNN. The pooling layer will typically act on each input feature separately and reduce its size. In addition to maximum pooling, other pooling functions may be used by the pooling layer, such as "average pooling" or the like.

Each neuron of the full connection layer is connected with all neurons of the previous layer and is used for integrating the extracted features. The parameters of a fully connected layer are also typically the most due to its fully connected nature. In a CNN (Convolutional Neural Networks) structure, 1 or more than 1 fully connected layer is connected after passing through a plurality of Convolutional layers and pooling layers. The full connection layer is used for integrating local information with category distinctiveness in the convolution layer or the pooling layer. In order to improve the CNN network performance, a ReLU function (linear rectification function) is generally used as an excitation function of each neuron of the full connection layer. The output value of the last fully-connected layer is transmitted to an output, which can be a Softmax layer, and is classified by Softmax logistic regression.

The normalized exponential function, or Softmax function, is a gradient log normalization of finite term discrete probability distribution. The Softmax function is widely applied to various probability-based multi-classification problem methods including multi-term logistic regression, multi-term linear discriminant analysis, a naive bayes classifier, an artificial neural network and the like, is mainly applied to multi-classification problems, and can map data records in a database to a certain one of given categories, so that the Softmax function can be applied to data prediction.

Specifically, the process of identifying and classifying suspected positive cell images by a preset single cell image cascade classification model comprises the following steps:

firstly, inputting a suspected positive cell image into a single cell two-classification sub-model of a preset single cell image cascade classification model to obtain a positive cell image.

Specifically, a single-cell two-classification sub-model of a preset single-cell image cascade classification model is trained on the basis of positive single cells and negative cells marked by doctors, suspected positive cell images are input into the single-cell two-classification sub-model of the trained preset single-cell image cascade classification model, negative cells are filtered out, and positive cell images are obtained.

And then, inputting the positive cell image into a first cell type classification submodel cascaded with the single-cell two-classification submodel to obtain a first type cell classification result and a second type, third type and fourth type combined classification result.

The first type may be a HSIL (high-grade squamous lesion) type; the second type may be a LSIL (low-grade squamous intraepithelial lesion) type; the third type may be an ASH (epithelial high-grade squamous epithelial cell, not excluding atypical squamous epithelial cell types of high-grade squamous intraepithelial lesions), and the fourth type may be an ASU (epithelial cell of undefined atypical squamous cell) type, referring to squamous epithelial cells that can neither be diagnosed as infected, inflammatory or reactive changes, nor as precancerous lesions.

Specifically, the first type of cells are used as one type, the second type of cells, the third type of cells and the fourth type of cells are used as one type, the type positive cell images are input into a first cell type classification sub-model cascaded with a single-cell two-classification sub-model, the first type of cells are filtered, and images of the second type of cells, the third type of cells and the fourth type of cells are obtained.

And finally, inputting the combined classification results of the second type, the third type and the fourth type into a second cell type classification submodel cascaded with the first cell type classification submodel to respectively obtain three classification results of the second type, the third type and the fourth type.

Specifically, a second cell type classification submodel cascaded with the first cell type classification submodel is trained through images of a second cell type, a third cell type and a fourth cell type, and then the images of the second cell type, the third cell type and the fourth cell type are input into the trained second cell type classification submodel to obtain classification results of the second cell type, the third cell type and the fourth cell type.

And S230, inputting the cell mass image with the first preset resolution into a preset cell mass image cascade classification model to obtain positive cell prediction results of different categories.

Specifically, a preset cell cluster image cascade classification model is trained based on positive cell clusters and negative cells marked by doctors, and cell cluster images with a first preset resolution are input into the trained preset cell cluster image cascade classification model to obtain positive cell HSIL type cells, LSIL type cells, ASH type cells and ASU type cell prediction results.

Fig. 4 is a flowchart of cell classification and prediction performed on a first preset resolution cell cluster image, and as shown in fig. 4, the step of performing cell classification and prediction on the first preset resolution cell cluster image to obtain prediction results of different types of positive cells includes the following steps:

s2301, inputting the cell mass image with the first preset resolution into a first cell mass two-classification sub-model of the preset cell mass image cascade classification model to obtain a positive cell mass image with the first preset resolution.

Specifically, a preset cell cluster image cascade classification model is trained based on positive cell clusters and negative cells marked by a doctor, a cell cluster image with a first preset resolution is input into the trained preset cell cluster image cascade classification model, and negative cell clusters are filtered out to obtain a positive cell cluster image with the first preset resolution.

And S2302, inputting the first preset resolution positive cell mass image into a second cell mass two-classification sub-model cascaded with the first cell mass two-classification sub-model to obtain a first combined classification result of the first type and the third type and a second combined classification result of the second type and the fourth type.

Specifically, the first type and the third type are used as a first type and are used as a first combination, the second type and the fourth type are used as a first type and are used as a first combination, positive cell mass images corresponding to the two combinations are input into a second cell mass two-classification sub-model cascaded with the first cell mass two-classification sub-model to train the model, then the first preset resolution positive cell mass images are input into the trained second cell mass two-classification sub-model, and positive cell mass images corresponding to the first combination classification of the first type and the third type are obtained and are used as a first combination classification result, and positive cell mass images corresponding to the second combination classification of the second type and the fourth type are used as a second combination classification result.

S2303, inputting the first combined classification result into a third cell mass two-classification sub-model in cascade connection with the second cell mass two-classification sub-model to respectively obtain a first type and a third type of two-classification results, and inputting the second combined classification result into a fourth cell mass two-classification sub-model in cascade connection with the second cell mass two-classification sub-model to respectively obtain a second type and a fourth type of two-classification results.

Specifically, a positive cell mass image corresponding to first combined classification of a first type and a third type is used for training a third cell mass two-classification sub-model cascaded with a second cell mass two-classification sub-model, and then the first combined classification result is input into the trained third cell mass two-classification sub-model to obtain the classification results of the first type and the third type of cell mass; meanwhile, a fourth cell mass two-classification sub-model cascaded with the second cell mass two-classification sub-model is trained through a positive cell mass image corresponding to second combined classification of the second type and the fourth type, and then the second combined classification result is input into the fourth cell mass two-classification sub-model to be trained, so that the classification results of the second type and the fourth type of cell mass are obtained.

And S240, acquiring a second cervical exfoliated cell slide image which is acquired at a second preset resolution based on the prediction results of the different types of positive cells.

Firstly, selecting a preset number of cells of different categories according to the classification confidence degrees in the prediction results of the positive cells of different categories.

Specifically, acquiring second cervical exfoliated cell slide images acquired at a second preset resolution based on different types of positive cell prediction results, sorting each type of positive cell and/or cell group image from high to low according to a corresponding classification confidence degree in the different types of positive cell prediction results of the first cervical exfoliated cell slide images, and selecting the cells or cell groups with the preset number, such as the first 5 cells, in the confidence degree sorting of each type of positive cell image to obtain 20 cells corresponding to the four types of positive cells; or the first 1 cell mass, and 4 cell masses corresponding to the four types of positive cells are obtained.

Then, a set of Z-Stack images with a second preset resolution are focused and collected at a preset number of cell positions respectively.

Specifically, the positions of the preset number of cells and cell groups are read, for example, information such as coordinates for describing the positions of the cells and cell groups is used, and a set of Z-Stack images with a second preset resolution are collected in a focused manner at the positions of the preset number of cells and cell groups.

The Z-Stack is a depth-of-field fusion algorithm, which analyzes an image sequence of a non-planar object acquired when a micro-lens zooms continuously, extracts regions which are clearly focused in each frame of image in the sequence, focuses the regions according to corresponding positions of the regions, and acquires a group of Z-Stack images, for example, 24Z-Stack images at the same coordinate corresponding to a positive cell.

And finally, fusing the Z-Stack images to obtain a second cervical exfoliated cell slide image, wherein the second cervical exfoliated cell slide image is consistent with the preset number.

Specifically, a set of Z-Stack images with a second preset resolution are collected at the preset number of cells and preset number of cell groups in a focusing manner, pixel-level image fusion is performed, and fusion images at the preset number of cells and preset number of cell groups are formed and serve as a second cervical exfoliated cell slide image.

And S250, identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution.

And S260, inputting the cell cluster image with the second preset resolution into the pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result.

The cascade cell classification model comprises dense connection modules with the preset module number, a first size convolutional layer, a maximum pooling layer, a maximum global average pooling layer and a full connection layer, and a second size convolutional layer with the first preset network layer number and an average pooling layer with the second preset network layer number are arranged among the dense connection modules.

Optionally, the identification result of the target cervical exfoliated cell slide includes: classification results of a first type, a second type, a third type, and a fourth type.

Specifically, a cascade cell classification model is trained based on different types of positive single cells and negative cells marked by doctors, images of suspected positive cells are input into the trained cascade cell classification model, the negative cells are filtered, and classification results of the positive cell images and classification results of the first type, the second type, the third type and the fourth type of the target cervical exfoliated cell slide are obtained.

According to the technical scheme of the embodiment of the invention, a first preset resolution single cell image and a first preset resolution cell cluster image in a first cervical exfoliated cell slide image are identified and segmented by acquiring the first cervical exfoliated cell slide image acquired based on a first preset resolution; and meanwhile, inputting the first preset resolution cell mass image into a preset cell mass image cascade classification model to obtain the prediction results of the positive cells of different types. Acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution based on the different categories of positive cell prediction results. And identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution. And inputting the cell cluster image with the second preset resolution into the pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result. The technical scheme of the embodiment of the invention solves the problem of low efficiency of manual identification of the cervical exfoliated cell slide, realizes quick positioning of cells, and further improves the accuracy and efficiency of identification of the cervical exfoliated cell slide.

Fig. 5 is a flowchart of another method for identifying a slide with exfoliated cervix cells according to an embodiment of the present invention, which belongs to the same inventive concept as the method for identifying a slide with exfoliated cervix cells according to the foregoing embodiment, and further describes a process of identifying and classifying a cell cluster image with a second preset resolution by using a cascade cell classification model on the basis of the foregoing embodiment. The method can be executed by a cervical exfoliated cell slide recognition device, which can be realized by software and/or hardware and is integrated in a computer device with an application development function.

As shown in fig. 5, the method for identifying the cervical exfoliated cell slide includes the following steps:

s310, acquiring a first cervical exfoliated cell slide image acquired based on a first preset resolution, and identifying and segmenting a single cell image with the first preset resolution and a cell cluster image with the first preset resolution in the first cervical exfoliated cell slide image.

S320, respectively carrying out cell classification and prediction on the single cell image with the first preset resolution and the cell mass image with the first preset resolution to obtain positive cell prediction results of different categories.

S330, acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution based on the prediction results of different types of positive cells, and identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution.

S340, inputting the cell mass image with the second preset resolution into a first secondary classification submodel of the cascade cell classification model to obtain a positive cell mass image with the second preset resolution.

Specifically, the first secondary classification submodel may be used to distinguish positive cell clusters from negative cell clusters, and the first secondary classification submodel needs to be pre-trained based on an image of the negative cell clusters; and inputting the cell mass image with the second preset resolution into the pre-trained first classification sub-model to obtain and output a negative cell mass image with the second preset resolution and a positive cell mass image with the second preset resolution.

S350, inputting the second preset resolution positive cell mass image into a second classification submodel cascaded with the first classification submodel to obtain a first classification result formed by the first type and the third type and a second classification result formed by the second type and the fourth type.

Specifically, a second binary classification sub-model is trained based on positive cell mass images marked by doctors and classified by a first combination of a first type and a third type and a second combination of positive cell mass images composed of a second type and a fourth type, the positive cell mass images with a second preset resolution are input into the trained second binary classification sub-model cascaded with the first binary classification sub-model, and a first classification result composed of the positive cell mass images corresponding to the first type and the third type and a second classification result composed of the positive cell mass images corresponding to the second type and the fourth type are obtained and output.

And S360, inputting the first classification result into a third classification submodel cascaded with the second classification submodel to respectively obtain a first type classification result and a third type classification result, and inputting the second classification result into a fourth classification submodel cascaded with the second classification submodel to respectively obtain a second type classification result and a fourth type classification result.

Specifically, a third classification submodel is trained based on positive cell cluster images marked by doctors and corresponding to a first type and a third type, a first classification result of a second preset resolution is input into the trained third classification submodel cascaded with the second classification submodel, and classification results of the first type and the third type are obtained and output; meanwhile, training a fourth classification submodel based on a positive cell cluster image which is marked by the doctor and corresponds to a second type and a fourth type, inputting a second classification result of a second preset resolution into the trained fourth classification submodel which is in cascade connection with the second classification submodel, and obtaining and outputting the second type and the fourth type of classification results. And meanwhile, the first classification result and the second classification result are subjected to secondary classification, so that the speed of identifying the cervical exfoliated cell slide is increased, and the efficiency of identifying the cervical exfoliated cell slide is improved.

And S370, taking the classification result of the first type and the third type and the classification result of the second type and the fourth type as the identification result of the target cervical exfoliated cell slide.

According to the technical scheme, a first cervical exfoliated cell slide image acquired based on a first preset resolution ratio is acquired, a first preset resolution ratio single cell image and a first preset resolution ratio cell cluster image in the first cervical exfoliated cell slide image are identified and segmented, cell classification and prediction are respectively carried out on the first preset resolution ratio single cell image and the first preset resolution ratio cell cluster image, positive cell prediction results of different types are obtained, a second cervical exfoliated cell slide image acquired based on the positive cell prediction results of different types and with a second preset resolution ratio is acquired, a second preset resolution ratio cell cluster image in the second cervical exfoliated cell slide image is identified and segmented, the second preset resolution ratio is higher than the first preset resolution ratio, and the second preset resolution ratio cell cluster image is identified and classified through a cascade cell classification model, so that a first type and a third type classification result and a second type and a fourth type classification result are obtained and serve as a target cervical exfoliated cell identification result. According to the technical scheme of the embodiment of the invention, the problem of low manual identification efficiency of the cervical exfoliated cell slide is solved, the cells are quickly positioned, and the accuracy and efficiency of identification of the cervical exfoliated cell slide are further improved.

In a specific embodiment, fig. 6 is a flow chart of a method for identifying a cervical exfoliated cell slide, fig. 7 is a flow chart of a method for identifying a 10-fold specular cervical exfoliated cell slide, and fig. 8 is a flow chart of a method for identifying a 20-fold specular cervical exfoliated cell slide.

The specific identification method of the cervical exfoliated cell slide comprises the following steps:

1. based on a high-throughput full-automatic slide scanner, images of the cervical exfoliated cell slides in a TCT type or LCT type slide making mode under a 10-time microscope are acquired.

2. And preprocessing and segmenting the cell image. In order to improve the image contrast and effectively segment cells, the cell nuclei are extracted by adopting the modes of low-pass filtering, high-pass filtering, binaryzation, morphological calculation and the like, then the area characteristics, the nuclear-to-cytoplasmic ratio, the optical density of the cell nuclei and the like of the cell nuclei are calculated, and finally a proper threshold value is set to distinguish single cells from cell clusters according to the extracted characteristics. Belonging to the step 3 of entering single cells and belonging to the step 5 of entering cell clusters.

3. Cells were sequentially filtered using a cascade model. Firstly, calculating a gradient histogram and gray statistical characteristics, selecting the most effective few characteristics as an SVM filter 1, and filtering blood cells; then calculating the combination of the texture characteristics and the SVM filter 1 into an SVM filter 2, and filtering impurities; and after grouping, calculating morphological characteristics and gray histogram characteristics of cell nuclei, combining the morphological characteristics and gray histogram characteristics with an SVM filter 2 to form an SVM filter 3, filtering residual blood cells, impurities and negative cells, and finally outputting the suspected positive cells to a single cell CNN prediction module.

3.1 HOG characteristics

The HOG is an operator used for describing the structural characteristics of the image, and has good invariance to image translation, rotation and illumination. The invention designs a multi-scale HOG characteristic, and the specific method comprises the following steps:

1) For the input cell image, the center of the cell nucleus is taken as a central point, the middle 32 × 32 area is cut out, and the HOG characteristic is calculated once and is taken as H1.

2) In the same way, a region of the cell image 64 x 64 is truncated, and for the purpose of speeding up the calculation, the image is reduced to 32 x 32, and the HOG feature is calculated again as H2.

3) Fusion of H1 and H2 combined into the final HOG signature.

The multi-scale design strengthens the function of the intermediate cell nucleus, simultaneously considers the influence of cytoplasm, and can better filter blood cells.

3.2 LBP characteristics

The LBP texture feature vector is generally represented by a block LBP histogram of an image, and the specific steps are as follows:

1) The image is divided into 3x3 image sub-blocks and the LBP value for each pixel in each sub-block is calculated.

2) And performing histogram statistics on each sub-block to obtain 3x3 histograms of the image sub-blocks.

3) And combining the histograms of the 3 × 3 sub-blocks to obtain the final image texture features.

3.3 morphological features based on Nuclear segmentation

The morphology of the cell nucleus is an important factor for judging whether the cell is positive, so that the characteristics of the cell nucleus such as area, perimeter, circularity and the like are extracted based on the division of the cell nucleus.

The segmentation method of the cell nucleus is as follows:

1) Transforming the image coordinate system to a polar coordinate system;

2) Calculating gradient information of the image;

3) Calculating the shortest path based on dynamic programming;

4) And mapping the shortest path to an image coordinate system in a reverse direction to obtain a closed area in the original image, wherein the closed area is the segmentation area of the cell nucleus.

3.4 Gray histogram feature

And dividing the image into five middle areas of upper left, lower left, upper right and lower right kernels, respectively counting the gray level histograms, and combining to form a final histogram.

3.5 statistical characteristics of gray level

The gray statistical characteristics are divided into possible cell nucleuses, cytoplasm and background areas by a threshold segmentation mode, and the mean value and the variance of the possible cell nucleuses, cytoplasm and background areas are respectively calculated, so that the characteristics are not influenced by the size and the position of the cell nucleuses and can be used as supplements of the characteristics.

4. And building an 81-layer dense connection network (DenseNet-81) classification model based on a Tensorflow framework.

The DenseNet-81 network comprises 38 Dense connection blocks (Dense blocks), each Dense connection block comprising 1x1 convolutional layer (confinement layer) and 1x 3 convolutional layer, the Dense connection blocks being transitioned between by 31 x1 convolutional layers and 3x 2 Average pooling layers (Average potential). In addition, denseNet-81 also contains 17 × 7 convolutional layer, 13 × 3 maximum pooling layer (Max pooling), 17 × 7 Global average pooling layer (Global averaging pooling), and 1 Fully connected layer (full connected layer), each convolutional layer contains 32 convolutional kernels, the output adopts the ReLU function as the activation function, and the Fully connected layer is classified by the normalized exponential function (softmax) as the activation function output.

The cascade CNN prediction model was trained for 10-fold single cells, and the classification task at each level employed a modified DenseNet-81 network, training a single-channel image with a number of 64 × 64.

4.1, training a two-classification model, namely a single cell CNN model 1, based on the marking of the positive single cells and the negative cells by the doctor, filtering out the negative cells, and outputting the positive cells to enter the next process.

4.2, taking positive cells HSIL as one class, taking LSIL, ASH and ASU as one class, training a two-classification model, namely the single-cell CNN model 2, filtering out HSIL type cells, outputting LSIL, ASH and ASU type cells to enter the next flow, finally training a LSIL, ASH and ASU three-classification model, namely the single-cell CNN model 3, and further classifying the single-cell CNN model.

4.3, sorting the four types of predicted positive cells from high to low according to the confidence level, selecting the first 5 cells with the highest confidence level in each type, simultaneously recording the position information of the cells and the like, and finally outputting the position information of 20 single cells to a 20-time cell prediction module.

5. The cascade CNN prediction model was trained on 10-fold cell clusters, and the classification task at each level employed a modified DenseNet-81 network, training 128 × 128 single-channel images.

And 5.1, training a two-classification model, namely a cell cluster CNN model 1, based on marking positive cell clusters and negative cells by doctors, filtering out negative cells (other), and outputting positive cells.

5.2, taking HSIL and ASH as one class, taking LSIL and ASU as one class, training a two-classification model, namely a cell cluster CNN model 2, and classifying the positive cell clusters into two classes; and finally, respectively training two classification models, namely a cell mass CNN model 3 for distinguishing HSIL and ASH, a cell mass CNN model 4 for distinguishing LSIL and ASU, and finally further classifying the positive cell masses.

And 5.3, sorting the predicted four types of positive cell clusters from high to low according to the confidence level, selecting 1 cell cluster with the highest confidence level of each type, simultaneously recording the position information of the cell clusters and the like, and finally outputting the position information of 4 cell clusters to a 20-time cell prediction module.

6. And (4) automatically acquiring the position information of the 20 cells and the 4 cell groups transmitted in the step (4) and the step (5) by the scanner system, and switching to a 20-fold mirror to acquire the image of the cervical exfoliated cell slide.

And 6.1, acquiring a group of 24Z-Stack images under the same visual field at each position, and then performing pixel-level fusion on each group of Z-Stack images to acquire 24 high-resolution images. And then, the pretreatment and segmentation method in the step 2 is utilized to segment the 20-fold lens collected cervical exfoliated cell slide image.

7. For the 20-fold cell mass image output in step 6, the cascade CNN prediction model is trained, and the classification task at each level adopts the improved DenseNet-81 network, training number is 160 × 160 of single-channel images.

7.1, training a two-classification model, namely a 20-time CNN model 1, based on the doctor marked positive cell and negative cell images, filtering out negative cells, and outputting positive cell images.

7.2, taking HSIL and ASH as a positive cell and LSIL and ASU as a positive cell, training a two-classification model, namely a 20-time CNN model 2, and classifying the positive cell into two types.

7.3, respectively training two classification models, namely a model for distinguishing HSIL from ASH, namely a 20-fold CNN model 3, a model for distinguishing LSIL from ASU, namely a 20-fold CNN model 4, and finally further classifying the positive cells.

7.4, sorting each type of the predicted positive cells according to confidence degree from high to low, selecting the first 5 cells with the highest confidence degree of each type, and recommending the cells to a doctor as the positive cells.

Fig. 9 is a block diagram of a structure of a cervical exfoliated cell slide recognition apparatus according to an embodiment of the present invention, which is applicable to a scene of image recognition of cervical exfoliated cells, and in particular, is more applicable to a situation of recognition of a cervical exfoliated cell slide in a TCT or LCT type slide making manner. The device can be realized by software and/or hardware, and is integrated in a computer device with application development function.

As shown in fig. 9, the cervical exfoliated cell slide recognizing apparatus includes: a first image segmentation module 310, a first image classification module 320, a second image segmentation module 330, and a second image classification module 340.

The first image segmentation module 310 is configured to acquire a first cervical exfoliated cell slide image acquired based on a first preset resolution, and identify and segment a single cell image with the first preset resolution and a cell cluster image with the first preset resolution in the first cervical exfoliated cell slide image; the first image classification module 320 is configured to perform cell classification and prediction on the single cell image with the first preset resolution and the cell cluster image with the first preset resolution respectively to obtain different types of positive cell prediction results; the second image segmentation module 330 is configured to obtain a second cervical exfoliated cell slide image acquired at a second preset resolution based on the prediction results of different types of positive cells, and identify and segment a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, where the second preset resolution is higher than the first preset resolution; and the second image classification module 340 is configured to input the cell cluster image with the second preset resolution into the pre-trained cascade cell classification model, so as to obtain a target cervical exfoliated cell slide recognition result.

According to the technical scheme of the embodiment of the invention, a first cervix uteri exfoliated cell slide image acquired based on a first preset resolution is acquired, and a single cell image with a first preset resolution and a cell cluster image with a first preset resolution in the first cervix uteri exfoliated cell slide image are identified and segmented; respectively carrying out cell classification and prediction on the single cell image with the first preset resolution and the cell mass image with the first preset resolution to obtain different types of positive cell prediction results; acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution based on the prediction results of different types of positive cells, and identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution; and inputting the cell cluster image with the second preset resolution into the pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result. The technical scheme of the embodiment of the invention solves the problem of low efficiency of manual identification of the cervical exfoliated cell slide, realizes quick positioning of cells, and improves the accuracy and efficiency of identification of the cervical exfoliated cell slide.

Optionally, the first image classification module 320 is configured to:

extracting at least one preset image feature in the single-cell image with the first preset resolution;

performing cell filtration based on at least one preset image characteristic to obtain a suspected positive cell image;

inputting the suspected positive cell images into a preset single cell image cascade classification model to obtain positive cell prediction results of different categories.

Optionally, the first image classification module 320 is further configured to: and respectively inputting at least one preset image characteristic into corresponding cascade cell filtration support vector machine classifiers for cell filtration.

Optionally, the first image classification module 320 is further configured to:

inputting gradient histogram features and gray statistic features in at least one preset image feature into a first support vector machine classifier to obtain a first classification result;

inputting the texture features of each image in the first classification result into a second support vector machine classifier obtained by training on the basis of the first support vector machine classifier to obtain a second classification result;

and inputting the morphological characteristics and the gray histogram characteristics of each image in the second classification result into a third support vector machine classifier obtained by training on the basis of the second support vector machine classifier.

Optionally, the first image classification module 320 is further configured to: and inputting the cell mass image with the first preset resolution into a preset cell mass image cascade classification model to obtain positive cell prediction results of different categories.

Optionally, the second image segmentation module 330 is configured to:

selecting a preset number of cells of different categories according to the classification confidence degrees in the prediction results of the positive cells of different categories;

respectively focusing and collecting a group of Z-Stack images with the resolution being a second preset resolution at each cell position with a preset number;

and fusing the Z-Stack images to obtain a second cervical exfoliated cell slide image, wherein the second cervical exfoliated cell slide image is consistent with the preset number.

Optionally, the first image classification module 320 is further configured to:

inputting the suspected positive cell image into a single-cell two-classification sub-model of a preset single-cell image cascade classification model to obtain a positive cell image;

inputting the positive cell image into a first cell type classification submodel cascaded with a single-cell two-classification submodel to obtain a first type cell classification result and a second type, a third type and a fourth type combined classification result;

and inputting the combined classification results of the second type, the third type and the fourth type into a second cell type classification submodel cascaded with the first cell type classification submodel to respectively obtain the results of the second type, the third type and the fourth type.

Optionally, the first image classification module 320 is further configured to:

inputting the cell mass image with the first preset resolution into a first cell mass two-classification sub-model of a preset cell mass image cascade classification model to obtain a positive cell mass image with the first preset resolution;

inputting the first preset resolution positive cell mass image into a second cell mass two-classification submodel cascaded with the first cell mass two-classification submodel to obtain a first combined classification result of the first type and the third type and a second combined classification result of the second type and the fourth type;

and inputting the first combined classification result into a third cell mass two-classification submodel cascaded with the second cell mass two-classification submodel to respectively obtain a first type and a third type of two-classification results, and inputting the second combined classification result into a fourth cell mass two-classification submodel cascaded with the second cell mass two-classification submodel to respectively obtain a second type and a fourth type of two-classification results.

Optionally, the second image classification module 330 is further configured to:

inputting the cell mass image with the second preset resolution into a first secondary classification submodel of the cascade cell classification model to obtain a positive cell mass image with the second preset resolution;

inputting the second preset resolution positive cell mass image into a second classification submodel cascaded with the first classification submodel to obtain a first classification result formed by the first type and the third type and a second classification result formed by the second type and the fourth type;

and inputting the first classification result into a third classification submodel cascaded with the second classification submodel to respectively obtain a first type and a third type classification result, and inputting the second classification result into a fourth classification submodel cascaded with the second classification submodel to respectively obtain a second type and a fourth type classification result.

The device for identifying the cervical exfoliated cell slide provided by the embodiment of the invention can execute the method for identifying the cervical exfoliated cell slide provided by any embodiment of the invention, and has corresponding functional modules and beneficial effects of the execution method.

Fig. 10 is a block diagram of a computer device according to an embodiment of the present invention, which shows a block diagram of a computer device 10 that can be used to implement an embodiment of the present invention.

Computer devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The computer device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the inventions described and/or claimed herein.

As shown in fig. 10, the computer device 10 includes at least one processor 11, and a memory communicatively connected to the at least one processor 11, such as a Read Only Memory (ROM) 12, a Random Access Memory (RAM) 13, and the like, wherein the memory stores a computer program executable by the at least one processor, and the processor 11 may perform various appropriate actions and processes according to the computer program stored in the Read Only Memory (ROM) 12 or the computer program loaded from the storage unit 18 into the Random Access Memory (RAM) 13. In the RAM 13, various programs and data necessary for the operation of the computer device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other via a bus 14. An input/output (I/O) interface 15 is also connected to bus 14.

A number of components in the computer device 10 are connected to the I/O interface 15, including: an input unit 16 such as a keyboard, a mouse, or the like; an output unit 17 such as various types of displays, speakers, and the like; a storage unit 18 such as a magnetic disk, optical disk, or the like; and a communication unit 19 such as a network card, modem, wireless communication transceiver, etc. The communication unit 19 allows the computer device 10 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.

The processor 11 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various processors running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, or the like. Processor 11 performs the various cervical exfoliated cell slide identification methods and processes described above, wherein the methods include:

In some embodiments, the cervical exfoliated cell slide identification method may be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and/or installed onto the computer device 10 via the ROM 12 and/or the communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the cervical exfoliated cell slide identification method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the cervical exfoliated cell slide identification method by any other suitable means (e.g., by way of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Computer programs for implementing the methods of the present invention can be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions/acts specified in the flowchart and/or block diagram block or blocks to be performed. A computer program can execute entirely on a machine, partly on a machine, as a stand-alone software package partly on a machine and partly on a remote machine or entirely on a remote machine or server.

In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer readable storage medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, the computer readable storage medium may be a machine readable signal medium. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer device. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), blockchain networks, and the internet.

The computing system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical host and VPS service are overcome.

It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present invention may be executed in parallel, sequentially, or in different orders, and are not limited herein as long as the desired results of the technical solution of the present invention can be achieved.

The above-described embodiments should not be construed as limiting the scope of the invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying a cervical exfoliated cell slide is characterized by comprising the following steps:

respectively carrying out cell classification and prediction on the single cell image with the first preset resolution and the cell cluster image with the first preset resolution to obtain different types of positive cell prediction results;

acquiring a second cervical exfoliated cell slide image acquired at a second preset resolution based on the prediction results of the different types of positive cells, and identifying and segmenting a second preset resolution cell cluster image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution;

and inputting the second preset resolution cell cluster image into a pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result.

2. The method of claim 1, wherein the classifying and predicting the cells of the single-cell image with the first preset resolution to obtain the prediction results of different types of positive cells comprises:

performing cell filtration based on the at least one preset image feature to obtain a suspected positive cell image;

and inputting the suspected positive cell image into a preset single cell image cascade classification model to obtain the prediction results of the different types of positive cells.

3. The method of claim 2, wherein the performing cell filtering based on the at least one preset image feature comprises:

and respectively inputting the at least one preset image characteristic into corresponding cascade cell filtration support vector machine classifiers for cell filtration.

4. The method according to claim 3, wherein the inputting the at least one preset image feature into the corresponding cascaded cell filtering support vector machine classifiers respectively comprises:

inputting the gradient histogram feature and the gray statistical feature in the at least one preset image feature into a first support vector machine classifier to obtain a first classification result;

5. The method according to claim 1, wherein the classifying and predicting the cells of the first preset resolution cell mass image to obtain the prediction results of different types of positive cells comprises:

and inputting the cell mass image with the first preset resolution into a preset cell mass image cascade classification model to obtain the prediction results of the different types of positive cells.

6. The method of claim 1, wherein said obtaining a second cervical exfoliated cell slide image acquired at a second preset resolution based on said different category positive cell prediction results comprises:

focusing and collecting a group of Z-Stack images with the resolution being the second preset resolution at each cell position of the preset number respectively;

7. The method of claim 1, wherein the cascaded cell classification model comprises a predetermined number of dense connection modules, a first size convolutional layer, a maximum pooling layer, a maximum global average pooling layer and a full connection layer, and a second size convolutional layer with the first predetermined number of network layers and an average pooling layer with the second predetermined number of network layers are disposed between the dense connection modules.

8. The method according to claim 2, wherein the process of identifying and classifying the suspected positive cell image by the preset single cell image cascade classification model comprises:

inputting the suspected positive cell image into a single-cell two-classification sub-model of the preset single-cell image cascade classification model to obtain a positive cell image;

inputting the positive cell image into a first cell type classification submodel cascaded with the single-cell two-classification submodel to obtain a first type cell classification result and a second type, a third type and a fourth type combined classification result;

9. The method of claim 5, wherein the process of the preset cell mass image cascade classification model for identifying and classifying the cell mass image with the first preset resolution comprises:

inputting the cell mass image with the first preset resolution into a first cell mass two-classification sub-model of the preset cell mass image cascade classification model to obtain a positive cell mass image with the first preset resolution;

10. The method according to any one of claims 1 to 9, wherein the cascade cell classification model performs a process of identifying and classifying the second preset resolution cell mass image, including:

inputting the second preset resolution cell mass image into a first secondary classification submodel of the cascade cell classification model to obtain a second preset resolution positive cell mass image;

inputting the second preset resolution positive cell mass image into a second classification submodel cascaded with the first classification submodel to obtain a first classification result composed of the first type and the third type and a second classification result composed of the second type and the fourth type;

11. An apparatus for identifying a cervical exfoliated cell slide, comprising:

the first image classification module is used for respectively carrying out cell classification and prediction on the first preset resolution single cell image and the first preset resolution cell mass image to obtain different types of positive cell prediction results;

the second image segmentation module is used for acquiring a second cervical exfoliated cell slide image which is acquired at a second preset resolution based on the prediction results of the different types of positive cells, and identifying and segmenting a second preset resolution cell group image in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution;

and the second image classification module is used for inputting the second preset resolution cell cluster image into a pre-trained cascade cell classification model to obtain a target cervical exfoliated cell slide recognition result.

12. A computer device, characterized in that the computer device comprises:

one or more processors;

a memory for storing one or more programs;

when executed by the one or more processors, cause the one or more processors to implement the method of cervical exfoliated cell slide identification recited in any one of claims 1 to 10.

13. A computer-readable storage medium, on which a computer program is stored, which program, when executed by a processor, carries out the method for identifying a cervical exfoliated-cell slide as claimed in any one of claims 1 to 10.