Methods, devices, equipment, and media for identifying cervical exfoliated cells on glass slides.

Abstract

This invention discloses a method, apparatus, device, and medium for identifying cervical exfoliated cell slides. The method includes: acquiring a first cervical exfoliated cell slide image at a first preset resolution; identifying and segmenting a single-cell image and a cell cluster image at the first preset resolution; classifying and predicting cells in the single-cell image and cell cluster image at the first preset resolution, respectively, to obtain prediction results for different categories of positive cells; acquiring a second cervical exfoliated cell slide image at a second preset resolution; identifying and segmenting a cell cluster image at the second preset resolution in the second cervical exfoliated cell slide image; and inputting the cell cluster image at the second preset resolution into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide identification result. This invention enables rapid cell localization and improves the accuracy and efficiency of cervical exfoliated cell slide identification.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of medical image processing technology, and in particular to a method, apparatus, device and medium for identifying cervical exfoliated cell slides. Background Technology

[0002] Thinprep Cytologic Test (TCT) or Liquid-Based Cytology Test (LCT) is one of the methods for cervical cancer screening. Currently, the identification of cervical exfoliated cell slides is mostly done manually. However, manual identification of cervical exfoliated cell slides is inefficient, time-consuming, and the results depend heavily on the doctor's experience. Summary of the Invention

[0003] This invention provides a method, apparatus, device, and medium for identifying cervical exfoliated cells on slides, which solves the problem of low efficiency in manual identification of cervical exfoliated cells on slides, achieves rapid cell localization, and improves the accuracy and efficiency of cervical exfoliated cell identification on slides.

[0004] In a first aspect, embodiments of the present invention provide a method for identifying cervical exfoliated cells on a glass slide, the method comprising:

[0005] Acquire a first cervical exfoliated cell slide image based on a first preset resolution, and identify and segment the first preset resolution single cell image and the first preset resolution cell cluster image in the first cervical exfoliated cell slide image;

[0006] Cell classification and prediction were performed on single-cell images and cell cluster images at the first preset resolution, respectively, to obtain prediction results for different categories of positive cells.

[0007] Acquire second cervical exfoliated cell slide images at a second preset resolution based on prediction results of different categories of positive cells, identify and segment cell clusters in the second cervical exfoliated cell slide images at a second preset resolution, wherein the second preset resolution is higher than the first preset resolution;

[0008] The cell cluster image at the second preset resolution is input into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result.

[0009] Secondly, embodiments of the present invention also provide a cervical exfoliated cell slide identification device, the device comprising:

[0010] The first image segmentation module is used to acquire a first cervical exfoliated cell slide image based on a first preset resolution, and to identify and segment the first preset resolution single cell image and the first preset resolution cell cluster image in the first cervical exfoliated cell slide image.

[0011] The first image classification module is used to classify and predict cells in single-cell images and cell cluster images at a first preset resolution, respectively, and obtain prediction results of positive cells of different categories.

[0012] The second image segmentation module is used to acquire a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results of different types of positive cells, and to identify and segment cell cluster images at a second preset resolution in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution;

[0013] The second image classification module is used to input the cell cluster image at the second preset resolution into the pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result.

[0014] Thirdly, embodiments of the present invention also provide a computer device, the computer device comprising:

[0015] One or more processors;

[0016] Memory, used to store one or more programs;

[0017] When one or more programs are executed by one or more processors, the one or more processors implement the cervical exfoliated cell slide identification method provided in any embodiment of the present invention.

[0018] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the cervical exfoliated cell slide identification method as provided in any embodiment of the present invention.

[0019] The technical solution of this invention involves acquiring a first cervical exfoliated cell slide image at a first preset resolution, identifying and segmenting single-cell images and cell cluster images at the first preset resolution within the first cervical exfoliated cell slide image, classifying and predicting cells in both images to obtain prediction results for different categories of positive cells, acquiring a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results for different categories of positive cells, identifying and segmenting cell cluster images at the second preset resolution within the second cervical exfoliated cell slide image (where the second preset resolution is higher than the first preset resolution), and inputting the second preset resolution cell cluster images into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result. This technical solution solves the problem of low efficiency in manual recognition of cervical exfoliated cell slides, achieves rapid cell localization, and improves the accuracy and efficiency of cervical exfoliated cell slide recognition. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for identifying cervical exfoliated cells on a glass slide provided in an embodiment of the present invention;

[0021] Figure 2 This is a flowchart of another method for identifying cervical exfoliated cells on a glass slide provided in an embodiment of the present invention;

[0022] Figure 3 It is a flowchart for classifying and predicting cells in a single-cell image with a first preset resolution;

[0023] Figure 4 This is a flowchart for classifying and predicting cells in a cell cluster image with a first preset resolution.

[0024] Figure 5 This is a flowchart of another method for identifying cervical exfoliated cells on a glass slide provided in an embodiment of the present invention;

[0025] Figure 6 This is a flowchart of a method for identifying cervical exfoliated cells using glass slides;

[0026] Figure 7 This is a flowchart of a method for identifying cervical exfoliated cells under a 10x microscope;

[0027] Figure 8 This is a flowchart of a method for identifying cervical exfoliated cells under a 20x microscope;

[0028] Figure 9 This is a structural block diagram of a cervical exfoliated cell slide identification device provided in an embodiment of the present invention;

[0029] Figure 10This is a structural block diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "first," "second," "third," and "fourth," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] Figure 1 This is a flowchart illustrating a method for identifying cervical exfoliated cell slides according to an embodiment of the present invention. This embodiment is applicable to scenarios involving cervical exfoliated cell image recognition, and particularly suitable for identifying cervical exfoliated cell slides prepared using TCT or LCT methods. This method can be executed by a cervical exfoliated cell slide recognition device, which can be implemented in software and / or hardware and integrated into a computer device with application development capabilities.

[0033] like Figure 1 As shown, the cervical exfoliated cell slide identification method of this embodiment includes the following steps:

[0034] S110. Acquire a first cervical exfoliated cell slide image based on a first preset resolution, and identify and segment the first preset resolution single cell image and the first preset resolution cell cluster image in the first cervical exfoliated cell slide image.

[0035] The first preset resolution is a pre-set resolution for acquiring cervical exfoliated cell slide images. For example, it could be the resolution of a high-throughput fully automated slide scanner. By setting the resolution, the cervical exfoliated cell slide image at that resolution is acquired as the first cervical exfoliated cell slide image.

[0036] Optionally, the cervical exfoliated cell slide image is a cervical exfoliated cell slide image prepared by TCT or LCT.

[0037] Specifically, a cervical exfoliated cell slide image acquired at a preset resolution is used as the first cervical exfoliated cell slide image. The cell nuclei are extracted from the first cervical exfoliated cell slide image using methods such as low-pass filtering, high-pass filtering, binarization, and morphological calculation. Then, the area features, nucleocytoplasmic ratio, and optical density of the cell nuclei are calculated. Finally, based on the above features, corresponding thresholds are set to distinguish between single cells and cell clusters in the cervical exfoliated cell slide image, resulting in the corresponding first preset resolution single cell image and first preset resolution cell cluster image.

[0038] S120. Perform cell classification and prediction on the single-cell image and cell cluster image at the first preset resolution respectively, and obtain prediction results for different categories of positive cells.

[0039] Understandably, classification and prediction algorithms are needed to process the single-cell image and the cell cluster image at the first preset resolution, respectively, to obtain the positive cell type corresponding to each single cell and cell cluster, i.e., the prediction results of different categories of positive cells.

[0040] Commonly used classification and prediction algorithms include: NBC (Naive Bayesian Classifier), LR (Logistic Regression), ID3, SVM (Support Vector Machine), KNN (K-Nearest Neighbor), and ANN (Artificial Neural Network), which are used for the classification of cervical exfoliated cell slides to identify different categories of positive cells.

[0041] Optional, positive cell classification prediction results include: high-grade squamous intraepithelial lesion type, low-grade squamous intraepithelial lesion type, atypical squamous cell type that cannot rule out high-grade squamous intraepithelial lesion, and unclear atypical squamous cell type, etc., to describe the screening results of positive cervical exfoliated cells.

[0042] S130. Acquire a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results of different types of positive cells, identify and segment cell cluster images at a second preset resolution in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution.

[0043] Specifically, based on the predicted results of different categories of positive cells in the first cervical exfoliated cell slide image, the top few categories of positive cells and cell clusters can be selected according to confidence levels. For example, the top five categories of positive cells and cell clusters can be selected. Using the coordinates of the selected positive cells and cell clusters, or other information used to determine their location, a higher resolution cervical exfoliated cell slide image can be acquired as the second cervical exfoliated cell slide image. This narrows the image acquisition range and makes the automatic identification and classification process more efficient. The cell nuclei of cell clusters in the second cervical exfoliated cell slide image at a second preset resolution are identified and segmented. The second preset resolution is higher than the first preset resolution; the combination of high and low resolution improves the accuracy and efficiency of cervical exfoliated cell slide identification.

[0044] S140. Input the cell cluster image with the second preset resolution into the pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result.

[0045] Understandably, the cascaded cell classification model connects various sub-models. Therefore, it is necessary to first pre-train each sub-model based on the different types of positive and negative cell clusters marked by the doctor. Then, the cell cluster image at the second preset resolution is input into the pre-trained cascaded cell classification model. Finally, the classification result of the cascaded cell classification model is used as the result of the target cervical exfoliated cell slide recognition.

[0046] The technical solution of this invention involves acquiring a first cervical exfoliated cell slide image at a first preset resolution, identifying and segmenting single-cell images and cell cluster images at the first preset resolution within the first cervical exfoliated cell slide image, classifying and predicting cells in both images to obtain prediction results for different categories of positive cells, acquiring a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results for different categories of positive cells, identifying and segmenting cell cluster images at the second preset resolution within the second cervical exfoliated cell slide image (where the second preset resolution is higher than the first preset resolution), and inputting the second preset resolution cell cluster images into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result. This technical solution solves the problem of low efficiency in manual recognition of cervical exfoliated cell slides, achieves rapid cell localization, and improves the accuracy and efficiency of cervical exfoliated cell slide recognition.

[0047] Figure 2This is a flowchart of another method for identifying cervical exfoliated cells on a slide provided by an embodiment of the present invention. This embodiment belongs to the same inventive concept as the cervical exfoliated cell slide identification method in the above embodiments. Based on the above embodiments, it further describes the process of classifying and predicting cells in single-cell images and cell cluster images at a first preset resolution. This method can be executed by a cervical exfoliated cell slide identification device, which can be implemented by software and / or hardware and integrated into a computer device with application development capabilities.

[0048] S210. Acquire a first cervical exfoliated cell slide image based on a first preset resolution, and identify and segment the first preset resolution single cell image and the first preset resolution cell cluster image in the first cervical exfoliated cell slide image.

[0049] S220. Perform cell classification and prediction on the single-cell image with the first preset resolution to obtain prediction results for different categories of positive cells.

[0050] Figure 3 This is a flowchart for classifying and predicting cells in a single-cell image at a first preset resolution, such as... Figure 3 As shown, cell classification and prediction are performed on a single-cell image at a first preset resolution to obtain prediction results for different categories of positive cells, including the following steps:

[0051] S2201. Extract at least one preset image feature from a single-cell image at a first preset resolution.

[0052] Preset image features may include: HOG (Histogram of Oriented Gradient) features, LBP (Local Binary Pattern) texture features, morphological features, gray-level histograms, and gray-level statistical features.

[0053] Among them, HOG features are feature descriptors used for object detection in computer vision and image processing. HOG features are constructed by calculating and statistically analyzing the gradient orientation histograms of local image regions, and they exhibit good invariance to image translation, rotation, and illumination.

[0054] The specific steps for extracting HOG features are as follows:

[0055] 1. Color and Gamma Normalization: To reduce the impact of lighting factors, the entire image needs to be normalized first. In the texture intensity of an image, local surface exposure contributes a significant proportion; therefore, this compression process can effectively reduce local shadows and lighting variations in the image.

[0056] 2. Calculate image gradient: Calculate the gradient along the horizontal and vertical axes of the image, and then calculate the gradient direction value for each pixel. A common method is to use a one-dimensional discrete differential template to process the image in one direction or simultaneously in both the horizontal and vertical directions.

[0057] 3. Constructing an Oriented Histogram Cell: Each pixel in the cell votes for a specific orientation-based histogram channel. Voting is weighted, meaning each vote carries a weight calculated based on the pixel's gradient magnitude. This weight can be represented by the magnitude itself or a function of it, such as the square root, square, or truncated form of the magnitude. Cells can be rectangular or star-shaped. Histogram channels are evenly distributed within the range of 0-1800 (undirected) or 0-3600 (directed).

[0058] 4. Grouping cell units into large intervals: Due to variations in local illumination and foreground-background contrast, the gradient intensity varies greatly, necessitating gradient normalization. Normalization further compresses illumination, shadows, and edges. Specifically, cell units are grouped into large, spatially connected intervals. In this way, the HOG descriptor becomes a vector composed of the histogram components of all cell units within each interval.

[0059] 5. Collect HOG features: Input the extracted HOG features into an SVM (Support Vector Machine) classifier to find a hyperplane as the decision function. SVM is a type of generalized linear classifier that performs binary classification of data in a supervised learning manner.

[0060] LBP is an algorithm used to describe the local texture features of an image. It reflects the texture changes around the pixels in the image and has advantages such as rotation invariance, grayscale invariance (unaffected by changes in illumination) and low computational complexity.

[0061] The specific steps for extracting LBP features are as follows: Divide the image into N*N windows, and use the pixel value of the center pixel of the window as the pixel threshold of the window. Compare the pixel values ​​of the sub-windows adjacent to the center of the window with the pixel threshold of the window. If the pixel value of the sub-window is greater than the pixel threshold of the window, the position of the sub-window is marked as 1, otherwise it is marked as 0. The combination of 1 and 0 of each sub-window generates the binary number of the corresponding N*N window, which is usually converted into a decimal number. This decimal number is the LBP value of the center pixel of the window. The LBP value is used to reflect the texture features of the image.

[0062] Morphological characteristics refer to the morphological features of the cell nucleus, because the morphology of the cell nucleus in cervical cancer cells can change. For example, the nucleus may be large, the nuclear-cytoplasmic ratio may be increased, the nucleus may be deformed, elongated, have serrated edges, be concave, have buds, be lobed, be mulberry-shaped, or be crescent-shaped. Therefore, based on the segmentation of the cell nucleus, features such as the area, perimeter, or roundness of the cell nucleus are extracted as morphological characteristics.

[0063] The segmentation of the cell nucleus includes:

[0064] 1. Coordinate system transformation: Transform the coordinate system of an image from the image coordinate system to the polar coordinate system. For example, it can be transformed from the rectangular coordinate system to the polar coordinate system.

[0065] 2. Calculating the gradient of an image: The gradient of an image represents the rate of change of the image. Edges of an image experience larger changes in grayscale values, resulting in larger gradient values; smoother parts of the image experience smaller changes in grayscale values, leading to smaller gradient values. Strictly speaking, calculating the gradient requires finding the derivative, but it is generally approximated by calculating the difference between pixel values ​​(approximate derivative value).

[0066] 3. Shortest path calculation based on dynamic programming: First, the topological sorting order of the graph must be found as the recursive order. For all nodes in the graph, a shortest path can be obtained and all sub-paths are shortest paths. Since dynamic programming is a global matching algorithm, the pixels on the same epipolar line are optimized and solved to extract the boundary of the cell nucleus, avoiding the influence of discontinuous edges caused by noise, etc.

[0067] 4. Reverse map the shortest path to the image coordinate system to obtain a closed region in the original image. This region is the segmentation region of the cell nucleus: Reverse map the shortest path from the polar coordinate system to the image coordinate system to obtain a closed region in the image, which is the cell nucleus region. Segment this region to obtain the cell nucleus.

[0068] Gray-level histogram features: A gray-level histogram is a function of the gray-level distribution, representing a statistical representation of the gray-level distribution in an image. It calculates the frequency of occurrence of all pixels in a digital image according to their gray-level values. Specifically, using the cell nucleus as the center, the image is divided into five regions: upper left, lower left, upper right, lower right, and the cell nucleus. Gray-level histograms are calculated for each region and combined to form the final gray-level histogram, used to describe the distribution characteristics of the cell nucleus and cytoplasm.

[0069] Gray-level statistical features: Gray-level statistical features reflect the distribution of gray levels in an image. These can include gray-level mean, gray-level variance, gray-level skewness, gray-level kurtosis, gray-level energy, and gray-level entropy. For example, thresholding can be used to segment possible cell nuclei, cytoplasm, and background regions. Specifically, the gray-level mean and variance are calculated separately. Thresholds are set for the gray-level mean and variance of the cell nucleus and cytoplasm to distinguish them; similarly, thresholds are set for the gray-level mean and variance of the cytoplasm and background regions to distinguish them.

[0070] S2202. Cell filtering is performed based on at least one preset image feature to obtain a suspected positive cell image.

[0071] Specifically, a cascaded model is used to extract one or more preset image features, which are then input into the corresponding SVM classifier for cell filtering. Different types of cells are filtered sequentially to obtain images of suspected positive cells.

[0072] Furthermore, at least one preset image feature is input into the corresponding cascaded cell filter support vector machine classifier, including the following steps:

[0073] First, the gradient histogram feature and gray-level statistical feature from at least one preset image feature are input into the first support vector machine classifier to obtain the first classification result.

[0074] Then, the texture features of each image in the first classification result are input into the second support vector machine classifier trained on the basis of the first support vector machine classifier to obtain the second classification result.

[0075] Finally, the morphological features and grayscale histogram features of each image in the second classification result are input into the third support vector machine classifier trained on the basis of the second support vector machine classifier.

[0076] Specifically, the gradient histogram and gray-level statistical features are calculated and input into the first support vector machine classifier to filter blood cells; then the texture features are calculated and input into the second support vector machine classifier combined with the first support vector machine classifier to filter impurities and remaining blood cells; finally, the morphological features and gray-level histogram features of the cell nucleus are calculated and input into the third support vector machine classifier combined with the second support vector machine classifier to filter negative cells, remaining impurities, and remaining blood cells, thereby obtaining an image of suspected positive cells.

[0077] S2203. Input the suspected positive cell image into the preset single-cell image cascade classification model to obtain the prediction results of different categories of positive cells.

[0078] The preset single-cell image cascade classification model is a cascade model used to classify suspected positive cell images to obtain different types of positive cells. For example, it can use CNN (Convolutional Neural Network) to classify suspected positive cell images.

[0079] CNN is an artificial neural network with a structure consisting of three layers: convolutional layers for feature extraction, pooling layers for downsampling, and fully connected layers for classification. A CNN comprises one or more convolutional layers and a fully connected layer at the top, as well as associated weights and pooling layers.

[0080] Each convolutional layer consists of several convolutional units, and the parameters of each convolutional unit are obtained through the backpropagation algorithm. The purpose of convolution is to extract different features from the input. The first convolutional layer may only extract some low-level features such as edges, lines, and corners. More layers of the network can iteratively extract complex features from low-level features.

[0081] Pooling is essentially a form of downsampling. There are various non-linear pooling functions, among which max pooling divides the input image into several rectangular regions and outputs the maximum value for each sub-region. Pooling layers continuously reduce the spatial size of the data, thus decreasing the number of parameters and computational cost, which to some extent controls overfitting. Typically, pooling layers are periodically inserted between convolutional layers in CNNs. Pooling layers usually act on each input feature individually and reduce its size. Besides max pooling, other pooling functions can also be used, such as average pooling.

[0082] Each neuron in a fully connected layer is connected to all neurons in the previous layer, used to synthesize the features extracted earlier. Due to its fully connected nature, fully connected layers generally have the most parameters. In a CNN (Convolutional Neural Network) architecture, after multiple convolutional and pooling layers, one or more fully connected layers are connected. Fully connected layers are used to integrate local information with class discriminative power from convolutional or pooling layers. To improve the performance of CNN networks, the activation function of each neuron in a fully connected layer typically uses the ReLU function (linear rectification function). The output value of the last fully connected layer is passed to an output layer, which can be a Softmax layer, using Softmax logistic regression for classification.

[0083] The normalization exponential function, also known as the Softmax function, is the gradient logarithm normalization of a finite number of discrete probability distributions. The Softmax function is widely used in various probability-based multi-class classification methods, including multinomial logistic regression, multinomial linear discriminant analysis, Naive Bayes classifiers, and artificial neural networks. It is primarily used in multi-class classification problems because it maps data records in a database to a specific category, thus enabling its application in data prediction.

[0084] Specifically, the process of identifying and classifying suspected positive cell images using a pre-defined single-cell image cascade classification model includes the following steps:

[0085] First, the suspected positive cell image is input into the single-cell binary sub-model of the preset single-cell image cascade classification model to obtain the positive cell image.

[0086] Specifically, the single-cell binary classification sub-model of the preset single-cell image cascade classification model is trained based on the positive and negative single cells marked by the doctor. The suspected positive cell image is input into the trained single-cell binary classification sub-model of the preset single-cell image cascade classification model to filter out the negative cells and obtain the positive cell image.

[0087] Then, the positive cell images are input into the first cell type classification sub-model cascaded with the single-cell binary classification sub-model to obtain the classification results of the first type of cells and the combined classification results of the second, third and fourth types.

[0088] The first type can be HSIL (high-grade squamous intraepithelial lesion); the second type can be LSIL (low-grade squamous intraepithelial lesion); the third type can be ASH (atypical squamous cells, cannot exclude high-grade squamous intraepithelial lesion); and the fourth type can be ASU (atypical squamous cells of undP. determined significance), which refers to squamous epithelial cells that cannot be diagnosed as infection, inflammation, or reactive changes, nor as precancerous lesions.

[0089] Specifically, the first type of cells is treated as one category, and the second, third, and fourth types of cells are treated as another category. The images of positive cells of the above types are input into the first cell type classification sub-model cascaded with the single-cell binary classification sub-model to filter out the first type of cells and obtain the images of the second, third, and fourth types of cells.

[0090] Finally, the combined classification results of the second, third, and fourth types are input into the second cell type classification sub-model cascaded with the first cell type classification sub-model to obtain the three-classification results of the second, third, and fourth types, respectively.

[0091] Specifically, the second cell type classification sub-model, which is cascaded with the first cell type classification sub-model, is trained using images of the second, third, and fourth cell types. Then, the images of the second, third, and fourth cell types are input into the trained second cell type classification sub-model to obtain the classification results of the second, third, and fourth cell types.

[0092] S230. Input the cell cluster image with the first preset resolution into the preset cell cluster image cascade classification model to obtain the prediction results of positive cells of different categories.

[0093] Specifically, a pre-defined cell cluster image cascade classification model is trained based on positive and negative cell clusters marked by doctors. The cell cluster image at the first pre-defined resolution is input into the trained pre-defined cell cluster image cascade classification model to obtain prediction results for positive cells of HSIL, LSIL, ASH, and ASU types.

[0094] Figure 4 This is a flowchart for classifying and predicting cells in a cell cluster image with a first preset resolution, such as... Figure 4 As shown, the process of classifying and predicting cells in a cell cluster image at a first preset resolution to obtain prediction results for different categories of positive cells includes the following steps:

[0095] S2301. Input the cell cluster image at the first preset resolution into the first cell cluster binary sub-model of the preset cell cluster image cascade classification model to obtain the positive cell cluster image at the first preset resolution.

[0096] Specifically, a pre-defined cell cluster image cascade classification model is trained based on positive and negative cell clusters marked by doctors. The cell cluster image at the first pre-defined resolution is input into the trained pre-defined cell cluster image cascade classification model to filter out negative cell clusters and obtain the positive cell cluster image at the first pre-defined resolution.

[0097] S2302. Input the positive cell cluster image at the first preset resolution into the second cell cluster binary classification sub-model cascaded with the first cell cluster binary classification sub-model to obtain the first combination classification result of the first type and the third type, and the second combination classification result of the second type and the fourth type.

[0098] Specifically, the first and third types are grouped into one category, forming the first combination, and the second and fourth types are grouped into another category, also forming the first combination. The positive cell cluster images corresponding to the above two combinations are input into the second cell cluster binary classification sub-model cascaded with the first cell cluster binary classification sub-model to train the model. Then, the positive cell cluster images at the first preset resolution are input into the trained second cell cluster binary classification sub-model to obtain the positive cell cluster images corresponding to the first combination classification of the first and third types, which are used as the first combination classification results, and the positive cell cluster images corresponding to the second combination classification of the second and fourth types, which are used as the second combination classification results.

[0099] S2303. Input the first combination classification result into the third cell cluster binary classification sub-model cascaded with the second cell cluster binary classification sub-model to obtain the first type and the third type binary classification results respectively. Then input the second combination classification result into the fourth cell cluster binary classification sub-model cascaded with the second cell cluster binary classification sub-model to obtain the second type and the fourth type binary classification results respectively.

[0100] Specifically, the third cell cluster binary classification model, cascaded with the second cell cluster binary classification model, is trained using images of positive cell clusters corresponding to the first combination classification of the first and third types. Then, the classification result of the first combination is input into the trained third cell cluster binary classification model to obtain the classification results of the first and third types of cell clusters. Simultaneously, the fourth cell cluster binary classification model, cascaded with the second cell cluster binary classification model, is trained using images of positive cell clusters corresponding to the second combination classification of the second and fourth types. Then, the classification result of the second combination is input into the trained fourth cell cluster binary classification model to obtain the classification results of the second and fourth types of cell clusters.

[0101] S240. Acquire a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results of different types of positive cells.

[0102] First, select a preset number of cells for each category based on the classification confidence in the prediction results of different categories of positive cells.

[0103] Specifically, a second cervical exfoliated cell slide image is acquired at a second preset resolution based on the prediction results of different types of positive cells. Among the prediction results of different types of positive cells in the first cervical exfoliated cell slide image, the images of each type of positive cell and / or cell cluster are sorted from high to low according to the corresponding classification confidence. The first preset number of cells or cell clusters in the confidence ranking of each type of positive cell image are selected. For example, it can be the first 5 cells, resulting in 20 cells corresponding to the four types of positive cells; or it can be the first cell cluster, resulting in 4 cell clusters corresponding to the four types of positive cells.

[0104] Then, a set of Z-Stack images with a second preset resolution are acquired by focusing at each of the preset number of cell locations.

[0105] Specifically, the positions of the preset number of cells and cell clusters are read, such as coordinates or other information describing the positions of cells and cell clusters. A set of Z-Stack images with a second preset resolution is then captured at the preset number of cell and cell cluster positions.

[0106] Z-Stack is a depth-of-field fusion algorithm that analyzes the image sequence of non-planar objects acquired during continuous zooming of a microscope lens, extracts the clearly focused area in each frame of the sequence, and then focuses these areas according to their corresponding positions to acquire a set of Z-Stack images. For example, it can be 24 Z-Stack images of the same coordinates at the location corresponding to a positive cell.

[0107] Finally, the Z-Stack images from each group are fused to obtain the second cervical exfoliated cell slide image, wherein the number of the second cervical exfoliated cell slide images is consistent with the preset number.

[0108] Specifically, a set of Z-Stack images with a resolution of a second preset resolution is captured at the locations of a preset number of individual cells and a preset number of cell clusters. Pixel-level image fusion is then performed to form a fused image at the locations of the preset number of individual cells and a preset number of cell clusters, which serves as the second cervical exfoliated cell slide image.

[0109] S250. Identify and segment a cell cluster image with a second preset resolution in a second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution.

[0110] S260. Input the cell cluster image at the second preset resolution into the pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result.

[0111] The cascaded cell classification model includes a preset number of densely connected modules, a first-size convolutional layer, a max pooling layer, a max global average pooling layer, and a fully connected layer. Between each densely connected module, there are second-size convolutional layers with a first preset number of network layers and second-preset number of average pooling layers.

[0112] Optionally, the target cervical exfoliated cell slide identification results include: classification results of type 1, type 2, type 3 and type 4.

[0113] Specifically, the cascaded cell classification model is trained based on different types of positive and negative cells marked by doctors. Suspected positive cell images are input into the trained cascaded cell classification model, and negative cells are filtered out to obtain the classification results of positive cell images and the first, second, third, and fourth types of target cervical exfoliated cell slides.

[0114] The technical solution of this invention involves acquiring a first cervical exfoliated cell slide image at a first preset resolution, identifying and segmenting single-cell images and cell cluster images at the first preset resolution within the slide image, classifying and predicting cells in the single-cell images at the first preset resolution to obtain prediction results for different categories of positive cells, and simultaneously inputting the cell cluster images at the first preset resolution into a preset cell cluster image cascade classification model to obtain prediction results for different categories of positive cells. A second cervical exfoliated cell slide image is then acquired at a second preset resolution based on the prediction results for different categories of positive cells. The cell cluster images at the second preset resolution within the second cervical exfoliated cell slide image are identified and segmented, wherein the second preset resolution is higher than the first preset resolution. The cell cluster images at the second preset resolution are then input into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide identification result. This technical solution solves the problem of low efficiency in manual identification of cervical exfoliated cell slides, achieves rapid cell localization, and further improves the accuracy and efficiency of cervical exfoliated cell slide identification.

[0115] Figure 5 This is a flowchart illustrating another method for identifying cervical exfoliated cell slides according to an embodiment of the present invention. This embodiment belongs to the same inventive concept as the cervical exfoliated cell slide identification method in the above embodiments, and further describes the process of identifying and classifying cell cluster images at a second preset resolution using a cascaded cell classification model based on the above embodiments. This method can be executed by a cervical exfoliated cell slide identification device, which can be implemented by software and / or hardware and integrated into a computer device with application development capabilities.

[0116] like Figure 5As shown, the method for identifying cervical exfoliated cells on a slide includes the following steps:

[0117] S310. Acquire a first cervical exfoliated cell slide image based on a first preset resolution, and identify and segment the first preset resolution single cell image and the first preset resolution cell cluster image in the first cervical exfoliated cell slide image.

[0118] S320. Perform cell classification and prediction on the single-cell image and cell cluster image at the first preset resolution respectively, and obtain prediction results for different categories of positive cells.

[0119] S330. Acquire a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results of different types of positive cells, identify and segment cell cluster images at a second preset resolution in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution.

[0120] S340. Input the cell cluster image at the second preset resolution into the first binary sub-model of the cascaded cell classification model to obtain the positive cell cluster image at the second preset resolution.

[0121] Specifically, the first binary classification sub-model can be used to distinguish between positive cell clusters and negative cell clusters. It needs to be pre-trained based on the image of negative cell clusters. The cell cluster image at the second preset resolution is input into the pre-trained first classification sub-model to obtain and output the negative cell cluster image at the second preset resolution and the positive cell cluster image at the second preset resolution.

[0122] S350. Input the positive cell cluster image at the second preset resolution into the second binary sub-model cascaded with the first binary sub-model to obtain the first classification result composed of the first type and the third type, and the second classification result composed of the second type and the fourth type.

[0123] Specifically, the second binary classification sub-model is trained based on the positive cell cluster images of the first combination classification of the first and third types marked by doctors and the positive cell cluster images of the second combination classification of the second and fourth types. The positive cell cluster images of the second preset resolution are input into the trained second binary classification sub-model cascaded with the first binary classification sub-model to obtain and output the first classification result composed of the positive cell cluster images corresponding to the first and third types, and the second classification result composed of the positive cell cluster images corresponding to the second and fourth types.

[0124] S360. Input the first classification result into the third binary classification sub-model cascaded with the second binary classification sub-model to obtain the first type and the third type binary classification results respectively. Then input the second classification result into the fourth binary classification sub-model cascaded with the second binary classification sub-model to obtain the second type and the fourth type binary classification results respectively.

[0125] Specifically, a third binary classification sub-model is trained based on images of positive cell clusters corresponding to the first and third types, as labeled by the doctor. The first classification result at a second preset resolution is input into the trained third binary classification sub-model, which is cascaded with the second binary classification sub-model, to obtain and output binary classification results for the first and third types. Simultaneously, a fourth binary classification sub-model is trained based on images of positive cell clusters corresponding to the second and fourth types, as labeled by the doctor. The second classification result at a second preset resolution is input into the trained fourth binary classification sub-model, which is cascaded with the second binary classification sub-model, to obtain and output binary classification results for the second and fourth types. This simultaneous binary classification of the first and second classification results accelerates the identification speed of cervical exfoliated cell slides and improves the efficiency of cervical exfoliated cell slide identification.

[0126] S370. The above-mentioned binary classification results of the first type and the third type and the binary classification results of the second type and the fourth type are used as the target cervical exfoliated cell slide identification results.

[0127] The technical solution of this invention involves acquiring a first cervical exfoliated cell slide image at a first preset resolution, identifying and segmenting single-cell images and cell cluster images at the first preset resolution within the first cervical exfoliated cell slide image, classifying and predicting cells in both images to obtain prediction results for different categories of positive cells, acquiring a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results for different categories of positive cells, identifying and segmenting cell cluster images at the second preset resolution within the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution, and identifying and classifying the cell cluster images at the second preset resolution using a cascaded cell classification model to obtain binary classification results for first and third types and second and fourth types, which serve as the target cervical exfoliated cell slide identification results. This technical solution solves the problem of low efficiency in manual identification of cervical exfoliated cell slides, achieves rapid cell localization, and further improves the accuracy and efficiency of cervical exfoliated cell slide identification.

[0128] In one specific embodiment Figure 6 This is a flowchart of a method for identifying cervical exfoliated cells using glass slides. Figure 7 This is a flowchart of a method for identifying cervical exfoliated cells on a slide under a 10x microscope. Figure 8 This is a flowchart of a method for identifying cervical exfoliated cells on a slide under a 20x microscope.

[0129] The specific method for identifying cervical exfoliated cells on slides includes the following steps:

[0130] 1. Based on a high-throughput fully automated slide scanner, acquire cervical exfoliated cell slide images under a 10x microscope using TCT or LCT slide preparation methods.

[0131] 2. Preprocessing and segmentation of cell images. To improve image contrast and effectively segment cells, low-pass filtering, high-pass filtering, binarization, and morphological calculations are used to extract cell nuclei. Then, the area features, nucleocytoplasmic ratio, and optical density of the nuclei are calculated. Finally, based on the extracted features, an appropriate threshold is set to distinguish between single cells and cell clusters. Single cells proceed to step 3, and cell clusters proceed to step 5.

[0132] 3. Use a cascaded model to filter cells sequentially. First, calculate the gradient histogram and gray-level statistical features, and select the few most effective features as SVM filter 1 to filter blood cells. Then, calculate the texture features and combine them with SVM filter 1 to form SVM filter 2, which filters impurities. After grouping, calculate the morphological features and gray-level histogram features of the cell nucleus, and combine them with SVM filter 2 to form SVM filter 3, which filters the remaining blood cells, impurities, and negative cells. Finally, output the suspected positive cells to the single-cell CNN prediction module.

[0133] 3.1 HOG Characteristics

[0134] HOG is an operator used to describe the structural features of an image, exhibiting good invariance to image translation, rotation, and illumination. This invention designs a multi-scale HOG feature, the specific method of which is as follows:

[0135] 1) For the input cell image, take the center of the cell nucleus as the center point, cut out the middle 32*32 area, and calculate the HOG feature H1 once.

[0136] 2) In the same way, a 64*64 region of the cell image is cropped. At the same time, in order to speed up the calculation, the image is reduced to 32*32, and the HOG feature is calculated again as H2.

[0137] 3) Combine H1 and H2 to form the final HOG feature.

[0138] This multi-scale design enhances the role of the intermediate nucleus while also taking into account the influence of the cytoplasm, thus enabling better filtration of blood cells.

[0139] 3.2 LBP Characteristics

[0140] LBP texture feature vectors are generally represented by block-based LBP histograms of the image. The specific steps are as follows:

[0141] 1) Divide the image into 3*3 image sub-blocks and calculate the LBP value of each pixel in each sub-block.

[0142] 2) Perform histogram statistics on each sub-block to obtain the histograms of 3*3 image sub-blocks.

[0143] 3) Combine the histograms of the 3*3 sub-blocks to obtain the final image texture features.

[0144] 3.3 Morphological features based on cell nuclear segmentation

[0145] The morphology of the cell nucleus is a very important factor in determining whether a cell is positive. Therefore, based on the segmentation of the cell nucleus, features such as the area, perimeter, and roundness of the cell nucleus were extracted.

[0146] The methods for segmenting the cell nucleus are as follows:

[0147] 1) Transform the image coordinate system to the polar coordinate system;

[0148] 2) Calculate the gradient information of the image;

[0149] 3) Shortest path calculation based on dynamic programming;

[0150] 4) Reverse map the shortest path to the image coordinate system to obtain a closed region in the original image, which is the segmentation region of the cell nucleus.

[0151] 3.4 Gray-level histogram features

[0152] The image is divided into five regions: top left, bottom left, top right, bottom right, and center. Gray-level histograms are calculated for each region and then combined to form the final histogram.

[0153] 3.5 Gray-scale statistical characteristics

[0154] Gray-scale statistical features segment possible cell nuclei, cytoplasm, and background regions using threshold segmentation, and calculate their mean and variance respectively. This feature is not affected by cell nucleus size or location and can serve as a supplement to the features mentioned above.

[0155] 4. Build an 81-layer dense network (DenseNet-81) classification model based on the Tensorflow framework.

[0156] The DenseNet-81 network consists of 38 dense blocks, each containing one 1x1 convolutional layer and one 3x3 convolutional layer. Dense blocks are connected by three 1x1 convolutional layers and three 2x2 average pooling layers. Additionally, DenseNet-81 includes one 7x7 convolutional layer, one 3x3 max pooling layer, one 7x7 global average pooling layer, and one fully connected layer. Each convolutional layer contains 32 kernels, and the outputs all use ReLU activation. The fully connected layer is followed by a softmax function for classification.

[0157] The cascaded CNN prediction model was trained on 10 times the number of single cells, and the improved DenseNet-81 network was used for classification tasks at each level, with 64*64 single-channel images as the training data.

[0158] 4.1. Based on the doctor's labeling of positive and negative single cells, a binary classification model, namely single-cell CNN model 1, is trained to filter out negative cells and output positive cells to enter the next process.

[0159] 4.2 Treat positive cells HSIL as one class and LSIL, ASH, and ASU as another class, and train a binary classification model, i.e., single-cell CNN model 2, to filter out HSIL type cells and output LSIL, ASH, and ASU type cells to enter the next process. Finally, train a LSIL, ASH, and ASU tri-classification model, i.e., single-cell CNN model 3, to further classify them.

[0160] 4.3. Sort the four types of positive cells predicted above in descending order of confidence. Select the top 5 cells with the highest confidence in each type and record the cell location information. Finally, output the location information of 20 single cells to the 20x cell prediction module.

[0161] 5. Train a cascaded CNN prediction model for 10x cell clusters, and use an improved DenseNet-81 network for each level of classification task, with 128*128 single-channel images for training.

[0162] 5.1. Based on the doctor's labeling of positive cell clusters and negative cells, a binary classification model, namely cell cluster CNN model 1, is trained to filter out negative cells (Others) and output positive cells.

[0163] 5.2 Treat HSIL and ASH as one class and LSIL and ASU as another class, and train a binary classification model, namely cell cluster CNN model 2, to divide the above positive cell clusters into two classes; finally, train two binary classification models respectively: the model that distinguishes HSIL and ASH, namely cell cluster CNN model 3, and the model that distinguishes LSIL and ASU, namely cell cluster CNN model 4, to further classify the positive cell clusters.

[0164] 5.3. Sort the four types of positive cell clusters predicted above in descending order of confidence, select the cell cluster with the highest confidence in each type, and record the location information of the cell clusters. Finally, output the location information of the four cell clusters to the 20x cell prediction module.

[0165] 6. The scanner system automatically acquires the location information of the 20 cells and 4 cell clusters input in steps 4 and 5, and switches to a 20x microscope to acquire images of cervical exfoliated cells on a slide.

[0166] 6.1. At each location, a set of 24 Z-Stack images with the same field of view are acquired. Then, each set of Z-Stack images is fused pixel-wise to obtain a total of 24 high-resolution images. Then, the preprocessing and segmentation methods in step 2 are used to segment the cervical exfoliated cell slide images acquired with a 20x microscope.

[0167] 7. For the cell cluster images output in step 6 that are 20 times larger, train a cascaded CNN prediction model. For each level of classification task, an improved DenseNet-81 network is used, and the training data consists of 160*160 single-channel images.

[0168] 7.1. Based on the doctor-labeled positive and negative cell images, a binary classification model (i.e., 20 times the CNN model 1) is trained to filter out negative cells and output positive cell images.

[0169] 7.2 Treat positive cells HSIL and ASH as one class and LSIL and ASU as another class, train a binary classification model, i.e., 20 times the CNN model2, to classify the above positive cells into two classes.

[0170] 7.3. Train two binary classification models respectively: a model that distinguishes between HSIL and ASH (i.e., a 20x CNN model 3) and a model that distinguishes between LSIL and ASU (i.e., a 20x CNN model 4). Finally, further classify the positive cells.

[0171] 7.4 Sort each type of positive cells predicted above in descending order of confidence, and select the top 5 cells with the highest confidence in each type as positive cells to recommend to the doctor.

[0172] Figure 9This is a structural block diagram of a cervical exfoliated cell slide recognition device provided in an embodiment of the present invention. This embodiment is applicable to scenarios involving cervical exfoliated cell image recognition, and in particular, it is more suitable for recognizing cervical exfoliated cell slides prepared using TCT or LCT methods. The device can be implemented in software and / or hardware and integrated into a computer device with application development capabilities.

[0173] like Figure 9 As shown, the cervical exfoliated cell slide recognition device 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.

[0174] The first image segmentation module 310 is used to acquire a first cervical exfoliated cell slide image acquired at a first preset resolution, and to identify and segment the first preset resolution single-cell image and the first preset resolution cell cluster image in the first cervical exfoliated cell slide image; the first image classification module 320 is used to classify and predict cells in the first preset resolution single-cell image and the first preset resolution cell cluster image respectively, and to obtain prediction results for different categories of positive cells; the second image segmentation module 330 is used to acquire a second cervical exfoliated cell slide image acquired at a second preset resolution based on the prediction results for different categories of positive cells, and to identify and segment the 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; the second image classification module 340 is used to input the second preset resolution cell cluster image into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result.

[0175] The technical solution of this invention involves acquiring a first cervical exfoliated cell slide image at a first preset resolution, identifying and segmenting single-cell images and cell cluster images at the first preset resolution within the first cervical exfoliated cell slide image, classifying and predicting cells in both images to obtain prediction results for different categories of positive cells, acquiring a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results for different categories of positive cells, identifying and segmenting cell cluster images at the second preset resolution within the second cervical exfoliated cell slide image (where the second preset resolution is higher than the first preset resolution), and inputting the second preset resolution cell cluster images into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result. This technical solution solves the problem of low efficiency in manual recognition of cervical exfoliated cell slides, achieves rapid cell localization, and improves the accuracy and efficiency of cervical exfoliated cell slide recognition.

[0176] Optionally, the first image classification module 320 is used for:

[0177] Extract at least one preset image feature from a single-cell image at a first preset resolution;

[0178] Cell filtering is performed based on at least one preset image feature to obtain images of suspected positive cells;

[0179] Images of suspected positive cells are input into a pre-defined single-cell image cascade classification model to obtain prediction results for different categories of positive cells.

[0180] Optionally, the first image classification module 320 is further configured to: input at least one preset image feature into the corresponding cascaded cell filtering support vector machine classifier for cell filtering.

[0181] Optionally, the first image classification module 320 is also used for:

[0182] The gradient histogram feature and gray-level statistical feature from at least one preset image feature are input into the first support vector machine classifier to obtain the first classification result;

[0183] The texture features of each image in the first classification result are input into the second support vector machine classifier trained on the basis of the first support vector machine classifier to obtain the second classification result;

[0184] The morphological features and grayscale histogram features of each image in the second classification result are input into the third support vector machine classifier trained on the basis of the second support vector machine classifier.

[0185] Optionally, the first image classification module 320 is further configured to: input a cell cluster image of a first preset resolution into a preset cell cluster image cascade classification model to obtain prediction results of positive cells of different categories.

[0186] Optionally, the second image segmentation module 330 is used for:

[0187] Select a preset number of cells for each category based on the classification confidence level in the prediction results of different categories of positive cells;

[0188] A set of Z-Stack images with a second preset resolution are acquired by focusing at each cell location of a preset number of locations;

[0189] The Z-Stack images from each group are fused to obtain a second cervical exfoliated cell slide image, wherein the number of second cervical exfoliated cell slide images is consistent with the preset number.

[0190] Optionally, the first image classification module 320 is also used for:

[0191] The suspected positive cell image is input into the single-cell binary sub-model of the preset single-cell image cascade classification model to obtain the positive cell image;

[0192] The positive cell image is input into the first cell type classification sub-model cascaded with the single-cell binary classification sub-model to obtain the classification results of the first type of cell and the combined classification results of the second, third and fourth types.

[0193] The combined classification results of the second, third, and fourth types are input into the second cell type classification sub-model cascaded with the first cell type classification sub-model to obtain the three-classification results of the second, third, and fourth types, respectively.

[0194] Optionally, the first image classification module 320 is also used for:

[0195] The cell cluster image at the first preset resolution is input into the first cell cluster binary sub-model of the preset cell cluster image cascade classification model to obtain the positive cell cluster image at the first preset resolution.

[0196] The positive cell cluster image at the first preset resolution is input into the second cell cluster binary classification sub-model cascaded with the first cell cluster binary classification sub-model to obtain the first combination classification result of the first type and the third type, and the second combination classification result of the second type and the fourth type.

[0197] The first combination classification result is input into the third cell cluster binary sub-model cascaded with the second cell cluster binary sub-model to obtain the first type and the third type binary classification results respectively. The second combination classification result is input into the fourth cell cluster binary sub-model cascaded with the second cell cluster binary sub-model to obtain the second type and the fourth type binary classification results respectively.

[0198] Optionally, the second image classification module 330 is also used for:

[0199] The cell cluster image at the second preset resolution is input into the first binary sub-model of the cascaded cell classification model to obtain the positive cell cluster image at the second preset resolution.

[0200] The image of positive cell clusters at the second preset resolution is input into the second binary classification sub-model cascaded with the first binary classification sub-model to obtain the first classification result composed of the first type and the third type, and the second classification result composed of the second type and the fourth type.

[0201] The first classification result is input into the third binary classification sub-model, which is cascaded with the second binary classification sub-model, to obtain the first type and the third type binary classification results respectively. The second classification result is input into the fourth binary classification sub-model, which is cascaded with the second binary classification sub-model, to obtain the second type and the fourth type binary classification results respectively.

[0202] The cervical exfoliated cell slide identification device provided in the embodiments of the present invention can execute the cervical exfoliated cell slide identification method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0203] Figure 10 This is a structural block diagram of a computer device provided in an embodiment of the present invention, showing a structural block diagram of a computer device 10 that can be used to implement an embodiment of the present invention.

[0204] Computer equipment is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Computer equipment can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, 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 merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0205] like Figure 10 As shown, the computer device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer programs stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the computer device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0206] Multiple components in computer device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of monitors, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows computer device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0207] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various cervical exfoliated cell slide identification methods and processes described above, wherein the methods include:

[0208] Acquire a first cervical exfoliated cell slide image based on a first preset resolution, and identify and segment the first preset resolution single cell image and the first preset resolution cell cluster image in the first cervical exfoliated cell slide image;

[0209] Cell classification and prediction were performed on single-cell images and cell cluster images at the first preset resolution, respectively, to obtain prediction results for different categories of positive cells.

[0210] Acquire second cervical exfoliated cell slide images at a second preset resolution based on prediction results of different categories of positive cells, identify and segment cell clusters in the second cervical exfoliated cell slide images at a second preset resolution, wherein the second preset resolution is higher than the first preset resolution;

[0211] The cell cluster image at the second preset resolution is input into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result.

[0212] In some embodiments, the cervical exfoliated cell slide identification method may be implemented as a computer program tangibly contained 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 on computer device 10 via ROM 12 and / or 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, processor 11 may be configured to perform the cervical exfoliated cell slide identification method by any other suitable means (e.g., by means of firmware).

[0213] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0214] Computer programs used to implement the methods of the present invention may 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, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0215] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0216] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer device having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer device. Other types of devices can also be used to provide interaction with the 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 can be received in any form (including sound input, voice input, or tactile input).

[0217] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0218] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0219] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0220] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for identifying cervical exfoliated cells on a glass slide, characterized in that, include: Acquire a first cervical exfoliated cell slide image based on a first preset resolution, and identify and segment the first preset resolution single cell image and the first preset resolution cell cluster image in the first cervical exfoliated cell slide image; Cell classification and prediction were performed on the first preset resolution single-cell image and the first preset resolution cell cluster image respectively to obtain prediction results of different categories of positive cells; Acquire a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results of the different categories of positive cells, identify and segment cell clusters in the second cervical exfoliated cell slide image at a second preset resolution, wherein the second preset resolution is higher than the first preset resolution; The cell cluster image at the second preset resolution is input into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result; wherein, the cascaded cell classification model includes a preset number of densely connected modules, a first-size convolutional layer, a max pooling layer, a max global average pooling layer, and a fully connected layer, and a second-size convolutional layer and a second-preset number of average pooling layers are set between each of the densely connected modules; The acquisition of a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results of the different categories of positive cells includes: Based on the classification confidence scores in the prediction results of different categories of positive cells, a preset number of cells of different categories are selected; A set of Z-Stack images with a resolution of the second preset resolution are captured by focusing at each of the preset number of cell locations; The Z-Stack images from each group are fused to obtain the second cervical exfoliated cell slide image, wherein the number of the second cervical exfoliated cell slide images is consistent with the preset number.

2. The method according to claim 1, characterized in that, The step of classifying and predicting cells in the first preset resolution single-cell image to obtain prediction results for different categories of positive cells includes: Extract at least one preset image feature from the first preset resolution single-cell image; Cell filtering is performed based on at least one preset image feature to obtain suspected positive cell images; The suspected positive cell images are input into a preset single-cell image cascade classification model to obtain prediction results for different categories of positive cells.

3. The method according to claim 2, characterized in that, The cell filtering based on the at least one preset image feature includes: The at least one preset image feature is input into the corresponding cascaded cell filtering support vector machine classifier for cell filtering.

4. The method according to claim 3, characterized in that, The step of inputting the at least one preset image feature into the corresponding cascaded cell filter support vector machine classifier includes: The gradient histogram feature and grayscale statistical feature from the at least one preset image feature are input into the first support vector machine classifier to obtain the first classification result; The texture features of each image in the first classification result are input into the second support vector machine classifier trained on the basis of the first support vector machine classifier to obtain the second classification result; The morphological features and grayscale histogram features of each image in the second classification result are input into the third support vector machine classifier trained on the basis of the second support vector machine classifier.

5. The method according to claim 1, characterized in that, The step of classifying and predicting cells in the first preset resolution cell cluster image to obtain prediction results for different categories of positive cells includes: The first preset resolution cell cluster image is input into a preset cell cluster image cascade classification model to obtain the prediction results of the different categories of positive cells.

6. The method according to claim 2, characterized in that, The process by which the preset single-cell image cascade classification model identifies and classifies the suspected positive cell images includes: The suspected positive cell image is input into the single-cell binary sub-model of the preset single-cell image cascade classification model to obtain the positive cell image. The positive cell image is input into the first cell type classification sub-model cascaded with the single-cell binary classification sub-model to obtain the classification results of the first type of cell and the combined classification results of the second, third and fourth types. The combined classification results of the second, third, and fourth types are input into the second cell type classification sub-model cascaded with the first cell type classification sub-model to obtain the three-classification results of the second, third, and fourth types, respectively.

7. The method according to claim 5, characterized in that, The process by which the preset cell cluster image cascade classification model identifies and classifies the first preset resolution cell cluster image includes: The first preset resolution cell cluster image is input into the first cell cluster binary sub-model of the preset cell cluster image cascade classification model to obtain the first preset resolution positive cell cluster image. The first preset resolution positive cell cluster image is input into the second cell cluster binary classification sub-model cascaded with the first cell cluster binary classification sub-model to obtain the first combination classification result of the first type and the third type, and the second combination classification result of the second type and the fourth type. The first combined classification result is input into the third cell cluster binary classification sub-model cascaded with the second cell cluster binary classification sub-model to obtain binary classification results of the first type and the third type, respectively. The second combined classification result is input into the fourth cell cluster binary classification sub-model cascaded with the second cell cluster binary classification sub-model to obtain binary classification results of the second type and the fourth type, respectively.

8. The method according to any one of claims 1-7, characterized in that, The process by which the cascaded cell classification model identifies and classifies the cell cluster image at the second preset resolution includes: The second preset resolution cell cluster image is input into the first binary sub-model of the cascaded cell classification model to obtain the second preset resolution positive cell cluster image; The image of positive cell clusters at the second preset resolution is input into the second binary classification sub-model cascaded with the first binary classification sub-model 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. The first classification result is input into the third binary classification sub-model cascaded with the second binary classification sub-model to obtain the first type and the third type binary classification results respectively. The second classification result is then input into the fourth binary classification sub-model cascaded with the second binary classification sub-model to obtain the second type and the fourth type binary classification results respectively.

9. A cervical exfoliated cell slide identification device, characterized in that, include: The first image segmentation module is used to acquire a first cervical exfoliated cell slide image based on a first preset resolution, and to identify and segment the first preset resolution single cell image and the first preset resolution cell cluster image in the first cervical exfoliated cell slide image. The first image classification module is used to classify and predict cells in the first preset resolution single cell image and the first preset resolution cell cluster image respectively, and obtain prediction results of positive cells of different categories. The second image segmentation module is used to acquire a second cervical exfoliated cell slide image at a second preset resolution based on the prediction results of the different types of positive cells, and to identify and segment cell cluster images at a second preset resolution in the second cervical exfoliated cell slide image, wherein the second preset resolution is higher than the first preset resolution; The second image classification module is used to input the cell cluster image at the second preset resolution into a pre-trained cascaded cell classification model to obtain the target cervical exfoliated cell slide recognition result; wherein, the cascaded cell classification model includes a preset number of densely connected modules, a first-size convolutional layer, a max pooling layer, a max global average pooling layer, and a fully connected layer, and a first preset number of second-size convolutional layers and a second preset number of average pooling layers are set between each of the densely connected modules; The second image segmentation module is used for: Based on the classification confidence scores in the prediction results of different categories of positive cells, a preset number of cells of different categories are selected; A set of Z-Stack images with a resolution of the second preset resolution are captured by focusing at each of the preset number of cell locations; The Z-Stack images from each group are fused to obtain the second cervical exfoliated cell slide image, wherein the number of the second cervical exfoliated cell slide images is consistent with the preset number.

10. A computer device, characterized in that, The computer device includes: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the cervical exfoliated cell slide identification method as described in any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the cervical exfoliated cell slide identification method as described in any one of claims 1-8.

Images