Intelligent virtual staining method and device for Epstein-Barr virus-related tumor pathological section images

Abstract

The present invention discloses an intelligent virtual staining method and device for Epstein-Barr virus-related tumor pathology slice images, and relates to the field of image intelligent processing technology. The method comprises: obtaining H&E images of in vitro tissue slices of Epstein-Barr virus-positive cells, performing EBER in situ hybridization on the slices after elution and staining, and then scanning them into slice images, performing image registration on the above two images, and obtaining training images after marking Epstein-Barr virus-positive cells; then obtaining images of in vitro tissue slices of Epstein-Barr virus-negative cells as comparison images, and training a virtual staining model that can intelligently distinguish and virtually stain Epstein-Barr virus-positive cells in H&E slice images based on the training images and the comparison images. The present invention can reduce the complexity of the operation process of artificial intelligence processing slice image technology, improve the intelligence level of artificial intelligence processing slice image technology, and replace the biological marker method with intelligent virtual staining, which has great scientific significance, social value and economic benefits.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent image processing, and in particular to an intelligent virtual staining method and device for Epstein-Barr virus-related tumor pathological section images. Background Art

[0002] Epstein-Barr virus-encoded small RNA (EBER) in situ hybridization has become an important method in the field of in vitro tissue section analysis. It can mark EB virus-positive cell nuclei in in vitro tissue pathological sections to confirm that the tissue is infected with EB virus. However, although EBER in situ hybridization is used every day in contemporary hospital pathology departments to detect EB virus in in vitro sections, this technology still has the following disadvantages: (1) The detection takes a long time (at least 3 working days); (2) It requires the consumption of in vitro tissue; (3) The detection cost is high, which increases the social medical care cost; (4) Some grassroots hospitals are unable to carry out this technology.

[0003] With the development of artificial intelligence technology, using artificial intelligence technology to process slice images and mark target positive cells in slice images has become an important development direction in the field of slice image processing. However, existing artificial intelligence technology for processing slice images often only uses artificial intelligence technology as an auxiliary means for biotechnology to mark target positive cells. For example, after processing tissue slices through a series of biological means to obtain slice images that have completed biological marking, artificial intelligence technology is used to identify and process the slice images and obtain the final marking results.

[0004] Practice shows that the existing AI technology for processing slice images still mainly relies on biomarker methods, and the degree of AI involvement is insufficient, resulting in the existing AI technology for processing slice images being relatively complex and not intelligent enough. Therefore, how to reduce the complexity of the AI ​​technology for processing slice images and improve its intelligence has become one of the issues that need to be addressed in the field of slice image processing. Summary of the invention

[0005] The present invention provides an intelligent virtual staining method and device for Epstein-Barr virus-related tumor pathology slice images, which can reduce the complexity of the operation process of artificial intelligence slice image processing technology and improve the intelligence level of artificial intelligence slice image processing technology, thereby accurately marking Epstein-Barr virus-positive cells in Epstein-Barr virus-related tumor pathology slice images and quickly making judgments.

[0006] In order to solve the above technical problems, the first aspect of the present invention discloses an intelligent virtual staining method for EB virus-related tumor pathological section images, the method comprising:

[0007] Acquiring an image of a first slice, wherein the first slice is an ex vivo tissue slice containing Epstein-Barr virus-positive cells;

[0008] Acquire an image of a second slice, where the second slice is obtained by performing EBER in situ hybridization on the first slice after washing and staining;

[0009] Obtaining a training image based on the image of the first slice and the image of the second slice, wherein the training image is an image in which all target EB virus-positive cells are marked on the image of the first slice;

[0010] acquiring an image of a third slice, wherein the third slice is an ex vivo tissue slice containing EB virus-negative cells;

[0011] Determine a comparison image according to the image of the third slice, wherein the comparison image is an image in which all target EB virus-negative cells are marked on the image of the third slice;

[0012] The preset original virtual staining model is trained using the training image and the comparison image to obtain a virtual staining model trained to convergence, wherein the virtual staining model is used to perform a virtual staining operation on the input EB virus-related tumor pathology section image, and the virtual staining operation includes:

[0013] The cell nuclei identified as EB virus positive cells in the EB virus related tumor pathological section image are represented by a first color, and the cell nuclei identified as EB virus negative cells in the EB virus related tumor pathological section image are represented by a second color.

[0014] The second aspect of the present invention discloses an intelligent virtual staining device for EB virus-related tumor pathological section images, the device comprising:

[0015] A first acquisition module is used to acquire an image of a first slice, where the first slice is an ex vivo tissue slice containing EB virus-positive cells;

[0016] A second acquisition module is used to acquire an image of a second slice, where the second slice is obtained by performing EBER in situ hybridization on the first slice after washing and staining;

[0017] A training image module, used to obtain a training image based on the image of the first slice and the image of the second slice, wherein the training image is an image in which all target EB virus cells are marked on the image of the first slice;

[0018] A third acquisition module is used to acquire an image of a third slice, where the third slice is an ex vivo tissue slice containing EB virus-negative cells;

[0019] A contrast image module, used to determine a contrast image based on the image of the third slice, wherein the contrast image is an image in which all target EB virus-negative cells are marked on the image of the third slice;

[0020] A model training module is used to train a preset original virtual staining model using the training image and the comparison image to obtain a virtual staining model trained to convergence, wherein the virtual staining model is used to perform a virtual staining operation on an input Epstein-Barr virus-related tumor pathology section image, wherein the virtual staining operation includes: representing the cell nuclei of cells identified as Epstein-Barr virus-positive cells in the Epstein-Barr virus-related tumor pathology section image with a first color, and representing the cell nuclei of cells identified as Epstein-Barr virus-negative cells in the Epstein-Barr virus-related tumor pathology section image with a second color.

[0021] As an optional implementation, in the second aspect of the present invention, the specific manner in which the training image module obtains the training image according to the image of the first slice and the image of the second slice includes:

[0022] According to the image of the second slice, determining the Epstein-Barr virus positive cells marked after EBER in situ hybridization in the image of the second slice;

[0023] The EB virus-positive cells in the image of the second slice are paired with the cells in the image of the first slice, and all target EB virus-positive cells corresponding to the EB virus-positive cells are determined from the image of the first slice to obtain a training image.

[0024] As an optional embodiment, in the second aspect of the present invention, the training image module pairs the EB virus-positive cells in the image of the second slice with the cells in the image of the first slice, and determines from the image of the first slice all target EB virus-positive cells corresponding to the EB virus-positive cells in a specific manner, including:

[0025] Converting the image of the first slice and the image of the second slice into grayscale images and performing normalization processing to obtain a grayscale image of the first slice and a grayscale image of the second slice;

[0026] Extracting a first registration image feature on the grayscale image of the first slice, and extracting a second registration image feature on the grayscale image of the second slice; wherein the registration image feature includes one or more of a key point, an edge, a curve, a surface, and an invariant moment on the corresponding grayscale image;

[0027] According to the first registered image feature and the second registered image feature, registering the grayscale image of the first slice and the grayscale image of the second slice to obtain an image registration result;

[0028] The image registration result is input into a pre-trained registration quality control model to obtain a registration quality assessment score output by the registration quality control model. If the registration quality assessment score is greater than or equal to a preset registration quality threshold, all target Epstein-Barr virus-positive cells corresponding to the Epstein-Barr virus-positive cells are determined from the image of the first slice based on the image registration result.

[0029] As an optional embodiment, in the second aspect of the present invention, the training image module determines the specific manner in which the EB virus-positive cells in the image of the second slice are marked after EBER in situ hybridization based on the image of the second slice, including:

[0030] Performing sub-pixel edge recognition processing on the image of the second slice to obtain sub-pixel edges on the image of the second slice; determining an EBER in situ hybridization staining area in the image of the second slice according to the sub-pixel edges;

[0031] Determining that all cells located in the EBER in situ hybridization staining area in the image of the second slice are Epstein-Barr virus positive cells;

[0032] The specific method in which the training image module performs sub-pixel edge recognition processing on the image of the second slice to obtain the sub-pixel edge on the image of the second slice includes:

[0033] For each pixel point on each horizontal pixel row on the image of the second slice, a preset edge recognition pixel array is constructed with the pixel point as a reference pixel point, wherein the edge recognition pixel array is a pixel row including a preset number of pixel points; a pixel value of each pixel point in the edge recognition pixel array is obtained, and it is determined whether the distribution of the pixel values ​​of all the pixel points in the edge recognition pixel array satisfies a preset distribution condition; if it is determined that the distribution of the pixel values ​​of all the pixel points in the edge recognition pixel array satisfies the preset distribution condition, the edge recognition pixel array is determined to be a target edge recognition pixel array;

[0034] For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, determining a sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the target edge recognition pixel array;

[0035] The sub-pixel edge on the image of the second slice is determined according to all the sub-pixel edge positions on the image of the second slice.

[0036] As an optional implementation, in the second aspect of the present invention, the edge recognition pixel array includes a first pixel segment, a second pixel segment, and a third pixel segment arranged in sequence, wherein the first pixel segment is a pixel row including a preset first number of pixel points, the second pixel segment is a pixel row including a preset second number of pixel points, and the third pixel segment is a pixel row including a preset third number of pixel points;

[0037] Furthermore, the specific manner in which the training image module determines whether the distribution of pixel values ​​of all pixels in the edge recognition pixel array meets a preset distribution condition includes:

[0038] Determine whether the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determine whether the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determine whether the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition;

[0039] If it is determined that the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition, then the edge recognition pixel array is determined to be a target edge recognition pixel array;

[0040] Furthermore, the training image module determines the specific manner of the sub-pixel edge position for each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice according to the distribution of pixel values ​​of all pixels in the target edge recognition pixel array, including:

[0041] For each of the target edge recognition pixel arrays corresponding to each horizontal pixel row on the image of the second slice, the sub-pixel edge position is determined according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array.

[0042] As an optional implementation, in the second aspect of the present invention, the device further includes:

[0043] a pixel analysis module, for determining, for the edge recognition pixel array corresponding to each pixel point on each horizontal pixel row on the image of the second slice, a first pixel value mean and a first pixel value variance of the pixel points in the first pixel segment of the edge recognition pixel array, a second pixel value mean and a second pixel value variance of the pixel points in the second pixel segment of the edge recognition pixel array, and a third pixel value mean and a third pixel value variance of the pixel points in the third pixel segment of the edge recognition pixel array;

[0044] And, the specific manner in which the training image module determines whether the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determines whether the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determines whether the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition includes:

[0045] Determine whether the first pixel segment of the edge recognition pixel array satisfies that the difference between the first pixel value mean and the third pixel value mean is greater than a preset mean difference threshold and the first pixel value variance is less than a preset first variance threshold; whether the second pixel segment of the edge recognition pixel array satisfies that the second pixel value mean is between the first pixel value mean and the third pixel value mean and the second pixel value variance is greater than a preset second variance threshold; whether the third pixel segment of the edge recognition pixel array satisfies that the difference between the third pixel value mean and the second pixel value mean is greater than a preset mean difference threshold and the third pixel value variance is less than a preset third variance threshold;

[0046] Furthermore, the training image module determines the specific manner of the sub-pixel edge position for each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array, including:

[0047] For each of the target edge identification pixel arrays corresponding to each horizontal pixel row on the image of the second slice, determine the pixel value change slope of each pixel in the second pixel segment according to the distribution of pixel values ​​of all pixel points in the second pixel segment of the target edge identification pixel array; determine the pixel point with the largest pixel value change slope in the second pixel segment as the target pixel point, and determine the sub-pixel edge position according to the pixel value change slopes of the target pixel point and the neighboring pixel points adjacent to the target pixel point.

[0048] As an optional implementation, in the second aspect of the present invention, the device further includes:

[0049] a pixel fitting module, configured to fit the pixel values ​​of all the pixels in the first pixel segment of the edge recognition pixel array corresponding to each pixel point on each horizontal pixel row on the image of the second slice to a first straight line, fit the pixel values ​​of all the pixels in the second pixel segment of the edge recognition pixel array to a second straight line, and fit the pixel values ​​of all the pixels in the third pixel segment of the edge recognition pixel array to a third straight line;

[0050] And, the specific manner in which the training image module determines whether the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determines whether the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determines whether the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition includes:

[0051] Determine whether the absolute value of the slope of the first straight line is less than a preset first slope threshold, determine whether the slope of the second straight line is greater than a preset second slope threshold, and determine whether the slope of the third straight line is less than a preset first slope threshold;

[0052] Furthermore, the training image module determines the specific manner of the sub-pixel edge position for each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array, including:

[0053] For each of the target edge recognition pixel arrays corresponding to each horizontal pixel row on the image of the second slice, a pixel value change curve of all pixel points in the second pixel segment is fitted according to the distribution of pixel values ​​of all pixel points in the second pixel segment of the target edge recognition pixel array, and the position where the absolute value of the slope on the pixel value change curve is the largest is determined as the sub-pixel edge position.

[0054] The third aspect of the present invention discloses another intelligent virtual staining device for EB virus-related tumor pathological section images, the device comprising:

[0055] A memory storing executable program code;

[0056] a processor coupled to the memory;

[0057] The processor calls the executable program code stored in the memory to execute the intelligent virtual staining method for EB virus-related tumor pathological section images disclosed in the first aspect of the present invention.

[0058] The fourth aspect of the present invention discloses a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute the intelligent virtual staining method of EB virus-related tumor pathology section images disclosed in the first aspect of the present invention.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The intelligent virtual staining method of EB virus-related tumor pathological slice images in the present invention can mark all EB virus-positive cells on the image of the first slice through the image of the first slice and the image of the second slice, thereby obtaining a training image; at the same time, mark all EB virus-negative cells on the image of the third slice, thereby obtaining a comparison image, and finally use the training image and the comparison image to train the preset original virtual staining model to obtain a virtual staining model trained to convergence, which can mark the EB virus-positive cells on the input target pathological image and perform a virtual staining operation. In the present invention, the biomarker method is only used as an auxiliary means to obtain a more efficient training data set, thereby reducing the complexity of the operation process of the artificial intelligence processing slice image technology and improving the intelligence of the artificial intelligence processing slice image technology. And the intelligent virtual staining replaces the biomarker method, which has great scientific significance, social value and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0062] Figure 1 It is a flow chart of an intelligent virtual staining method for Epstein-Barr virus-related tumor pathological section images disclosed in an embodiment of the present invention;

[0063] Figure 2 It is a flow chart of another intelligent virtual staining method for Epstein-Barr virus-related tumor pathological section images disclosed in an embodiment of the present invention;

[0064] Figure 3 is a schematic diagram of an edge pixel array in an embodiment of the present invention;

[0065] Figure 4is a schematic diagram of a pixel segment in an embodiment of the present invention;

[0066] Figure 5 It is a structural schematic diagram of an intelligent virtual staining device for Epstein-Barr virus-related tumor pathological section images disclosed in an embodiment of the present invention;

[0067] Figure 6 It is a structural schematic diagram of another intelligent virtual staining device for Epstein-Barr virus-related tumor pathological section images disclosed in an embodiment of the present invention;

[0068] Figure 7 It is a structural schematic diagram of another intelligent virtual staining device for EB virus-related tumor pathology section images disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0069] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or end including a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or ends.

[0071] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0072] The present invention discloses an intelligent virtual staining method and device for Epstein-Barr virus-related tumor pathology slice images, wherein the biomarker method is only used as an auxiliary means to obtain a more efficient training data set, thereby reducing the complexity of the operation process of the artificial intelligence processing slice image technology and improving the intelligence level of the artificial intelligence processing slice image technology. The following are detailed descriptions.

[0073] Embodiment 1

[0074] See also Figure 1 , Figure 1 : is a flow chart of an intelligent virtual staining method for Epstein-Barr virus-related tumor pathology slice images disclosed in an embodiment of the present invention. Figure 1 The described intelligent virtual staining method for EB virus-related tumor pathology slice images can be applied to an intelligent virtual staining device for EB virus-related tumor pathology slice images, and the intelligent virtual staining device for EB virus-related tumor pathology slice images can be integrated in a cloud server or a local server, which is not limited in the embodiments of the present invention. Figure 1 As shown, the intelligent virtual staining method for EB virus-related tumor pathological section images may include the following operations:

[0075] 101. Acquire an image of a first slice.

[0076] In the embodiment of the present invention, the first slice is an ex vivo tissue slice containing target positive cells. For example, the first slice can be a tissue slice of Epstein-Barr virus-positive malignant tumor (including nasopharyngeal carcinoma, EBV-positive adenocarcinoma, EBV-positive diffuse large B-cell lymphoma and NK / T-cell lymphoma), and the target positive cells are Epstein-Barr virus-positive cells.

[0077] Epstein-Barr virus (EBV) is the abbreviation of human herpesvirus type 4. It is one of the causes of various malignant tumors (such as nasopharyngeal carcinoma, adenocarcinoma, diffuse large B-cell lymphoma and NK / T-cell lymphoma, etc.), and mainly infects epithelial cells and B lymphocytes in the human oropharynx. EBER is a small RNA encoded by EBV, which has a very high copy number in EBV-infected cells and is expressed at all stages of viral infection.

[0078] In the embodiment of the present invention, the most advanced international scanning instrument (Leica digital pathology scanner is used to obtain the image of the first slice, and the scanning instrument model is Leica Aperio AT2), which has automatic focus function and multi-layer scanning function; wherein, the scanning parameters are set as follows: the scanning magnification is 40 times; the image color mode is RGB; storage: each scanned image is saved in the SVS format with full resolution and lossless compression.

[0079] In an embodiment of the present invention, step 101 may include: obtaining H&E sections of in vitro tissues of target positive cells. H&E sections (hematoxylin-eosin staining) are a staining technique widely used in the fields of pathology and histology to observe the morphological structure of tissues and cells. This staining method can clearly show the different components of tissues, thereby helping pathologists to make diagnoses and research.

[0080] 102. Acquire an image of a second slice.

[0081] In the embodiment of the present invention, the second slice is a slice after the first slice is washed and stained and then subjected to EBER in situ hybridization. In situ hybridization (ISH) is a technique that combines the principles of molecular biology, histochemistry and cytology. It can accurately analyze specific nucleic acid sequences in terms of qualitative, localization and quantitative analysis, and plays an important role in molecular pathology, gene analysis diagnosis and a variety of biomedical research. The basis of in situ hybridization is to utilize the property that there are complementary base sequences between single strands of nucleic acid molecules, and to form a specific nucleic acid hybridization molecule by pairing and combining a specific labeled exogenous nucleic acid (probe) with the DNA or RNA to be tested. This process requires fixing tissue, cell or chromosome samples, selecting a probe that is complementary to the target fragment, and using a marker combined with the probe to display the result. The application of in situ hybridization to detect EBER in paraffin sections is currently the most effective method for diagnosing EB virus infection. However, this technology requires consumption of tissue, is time-consuming, and is expensive, which affects the promotion and use of primary hospitals.

[0082] In the embodiment of the present invention, step 102 may include:

[0083] Place the back of the H&E slice obtained in step 1 on a heating table and heat at 130°C for about 10 seconds until the sealing glue dissolves, and use tweezers to push the cover glass to remove it;

[0084] Degumming of H&E sections was performed in xylene using three cylinders of xylene, each cylinder for about 10 min, until the sealing glue was completely removed by naked eye observation;

[0085] Wash the H&E sections in ethanol: wash with 3 cylinders of absolute ethanol, 2 cylinders of 95% ethanol, and 1 cylinder of 80% ethanol, each for 1-2 minutes; rinse with running water for 1 minute to remove the eosin;

[0086] Place the H&E sections in 1% hydrochloric acid ethanol for differentiation overnight (about 16 hours) until the hematoxylin becomes colorless, and then rinse in running water for 4 hours;

[0087] In situ hybridization detection: H&E sections were automatically stained using Dako Autostai ner, and the EBERs in situ hybridization detection kit was used according to the reagent instructions, DAB color development, hematoxylin counterstaining, and the second section was obtained;

[0088] The second slice is scanned into a digital image to obtain an image of the second slice.

[0089] 103. Obtain a training image according to the image of the first slice and the image of the second slice.

[0090] In the embodiment of the present invention, the training image is an image in which all target positive cells are marked on the image of the first slice. The second slice is a slice after washing and staining and subjected to EBER in situ hybridization, which can mark the target positive cells. By analyzing the image of the first slice and the image of the second slice, all target positive cells corresponding to EB virus positive cells can be determined from the image of the first slice, thereby obtaining a training image.

[0091] In an optional embodiment, step 103 may include:

[0092] According to the image of the second slice, determining the Epstein-Barr virus positive cells after EBER in situ hybridization in the image of the second slice;

[0093] The EB virus-positive cells in the image of the second slice are paired with the cells in the image of the first slice, and all target positive cells corresponding to the EB virus-positive cells are determined from the image of the first slice to obtain a training image.

[0094] It can be seen that this optional embodiment pairs the EB virus-positive cells in the image of the second slice with the cells in the image of the first slice, thereby accurately determining all target positive cells corresponding to the EB virus-positive cells in the first slice and obtaining a more effective training image.

[0095] 104. Acquire an image of a third slice.

[0096] In the embodiment of the present invention, the third slice is an in vitro tissue slice containing target negative cells. The third slice can be a tissue slice of Epstein-Barr virus-negative benign lesions and malignant tumors (including squamous cell carcinoma, adenocarcinoma, diffuse large B-cell lymphoma and chronic inflammation of nasopharyngeal mucosa), and the target negative cells are Epstein-Barr virus-negative cells.

[0097] 105. Determine a comparison image based on the image of the third slice.

[0098] In the embodiment of the present invention, the comparison image is an image in which all target negative cells are marked on the image of the third slice.

[0099] 106. Use the training image and the comparison image to train the preset original virtual staining model to obtain a virtual staining model that is trained to convergence.

[0100] In an embodiment of the present invention, the virtual staining model is used to perform a virtual staining operation on an input EB virus-related tumor pathology section image, and the virtual staining operation includes:

[0101] The cell nuclei identified as EB virus positive cells in the EB virus related tumor pathology section image are represented by the first color, and the cell nuclei identified as EB virus negative cells in the EB virus related tumor pathology section image are represented by the second color.

[0102] It can be seen that the intelligent virtual staining method for EB virus-related tumor pathological slice images in the embodiment of the present invention can mark all EB virus-positive cells on the image of the first slice through the image of the first slice and the image of the second slice, thereby obtaining a training image; at the same time, mark all EB virus-negative cells on the image of the third slice, thereby obtaining a comparison image, and finally use the training image and the comparison image to train the preset original virtual staining model to obtain a virtual staining model trained to convergence, which can mark the EB virus-positive cells on the input target pathological image and perform a virtual staining operation. In the present invention, the biomarker method is only used as an auxiliary means to obtain a more efficient training data set, thereby reducing the complexity of the operation process of the artificial intelligence processing slice image technology and improving the intelligence level of the artificial intelligence processing slice image technology.

[0103] In an optional embodiment, step 106 may include:

[0104] The DeepLabV3 architecture is used to train the model by using the training set of EB virus-positive tumor cells in the training images and the training set of EB virus-negative cells in the comparison images as supervisory signals, and then verifying it with unlabeled large sample data to obtain a virtual staining model that is trained to convergence. The specific method is as follows:

[0105] S601, prepare the data set: complete it according to the standard of 80% of the training set and 20% of the test set data.

[0106] S602, training: sending the training set to the DeepLabV3 model for three-fold cross-validation training.

[0107] S603, Verification: Use the test set to evaluate the model training effect. According to the evaluation results, determine whether the amount of training data is sufficient, whether the hyperparameters of model training are optimal, and perform iterative training to obtain the best model weights.

[0108] S604. Model evaluation indicators: Use different evaluation indicators (such as ROC curve, AUC, sensitivity, specificity, false positive rate, etc.) to evaluate the performance of the virtual staining model.

[0109] S605: Obtain a virtual coloring model trained to convergence.

[0110] In this optional embodiment, further optionally, the method may also include:

[0111] Input the H&E digital image to the virtual staining model, mark the target positive cells on the H&E digital image, and convert the output into an EBER in situ hybridization color image overlaid on the H&E digital image. In the EBER in situ hybridization color image, the nuclei of cells predicted to be EBV negative are indicated in blue, while the nuclei of cells predicted to be EBV positive are indicated in red. This makes it easier for users to use by displaying the target positive cells more intuitively.

[0112] In another optional embodiment, pairing the EB virus-positive cells in the image of the second slice with the cells in the image of the first slice, and determining all target EB virus-positive cells corresponding to the EB virus-positive cells from the image of the first slice may include:

[0113] Converting the image of the first slice and the image of the second slice into grayscale images and performing normalization processing to obtain a grayscale image of the first slice and a grayscale image of the second slice;

[0114] Extracting a first registration image feature on the grayscale image of the first slice, and extracting a second registration image feature on the grayscale image of the second slice;

[0115] According to the first registration image feature and the second registration image feature, registering the grayscale image of the first slice and the grayscale image of the second slice to obtain an image registration result;

[0116] The image registration result is input into a pre-trained registration quality control model to obtain the registration quality assessment score output by the registration quality control model. If the registration quality assessment score is greater than or equal to the preset registration quality threshold, all target EB virus-positive cells corresponding to EB virus-positive cells are determined from the image of the first slice based on the image registration result.

[0117] In this optional embodiment, the registered image features include one or more of key points, edges, curves, surfaces and invariant moments on the corresponding grayscale image.

[0118] In this optional embodiment, further optionally, the above steps may specifically be:

[0119] S301, image preprocessing: converting the image of the first slice and the image of the second slice into grayscale images, and normalizing the brightness values ​​of the grayscale images, so that pathological images of different staining types are as similar as possible in the grayscale domain.

[0120] S302, feature extraction: extract features of the first slice image and the second slice image using a VGG deep learning-based model.

[0121] S303, registration transformation: use the RANSAC algorithm to select suitable feature pairs from the features of the first slice image and the second slice image, and perform rigid registration and non-rigid registration using DeepFow.

[0122] S304, registration evaluation: Use DIC similarity coefficient (Dic, DSC), target matching error (TRE), relative target matching error (rTRE), Hausdorff distance (HD) and other evaluation indicators to evaluate the registration effect. If the registration effect does not meet the requirements, re-trigger step S301 until the registration effect meets the requirements.

[0123] S305 , determining all target positive cells corresponding to EB virus positive cells from the image of the first slice according to the registration result, and obtaining a training image.

[0124] It can be seen that this optional embodiment, through the improved image registration method and the addition of a registration quality control step, can obtain a more accurate registration result between the EB virus-positive cells in the image of the second slice and the image of the first slice, thereby more accurately determining all target positive cells corresponding to the EB virus-positive cells from the image of the first slice, thereby providing a more efficient and accurate training data set for the subsequent model training step.

[0125] In addition, during the registration process, due to the diversity and uncertainty between images, errors may occur in the registration results. Therefore, registration quality control is required, and the registration effect is automatically evaluated by training quality control CNN.

[0126] Therefore, in this optional embodiment, further optionally, step S304 may include:

[0127] S3041. Manual quality control: Pathologists evaluate the registration effect and collect samples with high and low registration quality as training data.

[0128] S3042, Classification CNN quality control: Train a classification quality control CNN model and perform corresponding tests and verification. Complete automatic evaluation of all registered slices through the quality control CNN model.

[0129] The positive cells in EBERs staining are the target of pathological intelligent annotation, so positive cell extraction and label transfer are required. Cell extraction mainly includes cell detection based on QuPath software, cell pairing of pathological images with different staining after registration, and positive and negative label assignment. The labeling of positive and negative cells can be set using color thresholds. However, in the extraction of positive and negative cells in pathological images, due to the diversity and uncertainty of cell size, shape, color, whether necrotic and other characteristics, there will be problems of false negatives and false positives, that is, unlabeled or incorrectly labeled cells; at the same time, because the accuracy of registration may have errors at the cellular level, it will also cause problems of missed labels and mislabeling. Therefore, in this optional embodiment, it is further optional to include false negative and false positive quality control, and the steps of training and labeling CNN, as follows:

[0130] The positive and negative cells that have successfully completed the registration are used as the training set to train the annotation CNN model, predict the unpaired cells, assign labels, and the doctors perform quality control on the labels assigned by the model.

[0131] The cell data that have passed quality control are added to the training set to improve the performance of the labeled CNN.

[0132] The trained annotation CNN can automatically assign cell labels.

[0133] In this optional embodiment, the digital pathology intelligent annotation process can also make comprehensive use of image registration algorithms and online deep learning and other technologies. In each link, the algorithm needs to be evaluated and improved according to the specific verification results, and the corresponding parameters need to be optimized and the model trained. By continuously optimizing the process and algorithm, the accuracy and efficiency of digital pathology intelligent annotation can be improved, providing better support for digital pathology research.

[0134] Embodiment 2

[0135] See also Figure 2 , Figure 2 1 is a flow chart of another intelligent virtual staining method for EB virus-related tumor pathology slice images disclosed in an embodiment of the present invention. Figure 2 The described intelligent virtual staining method for EB virus-related tumor pathology slice images can be applied to an intelligent virtual staining device for EB virus-related tumor pathology slice images, and the intelligent virtual staining device for EB virus-related tumor pathology slice images can be integrated in a cloud server or a local server, which is not limited in the embodiments of the present invention. Figure 2 As shown, the intelligent virtual staining method for EB virus-related tumor pathological section images may include the following operations:

[0136] 201. Acquire an image of a first slice.

[0137] 202. Acquire an image of a second slice.

[0138] In the embodiment of the present invention, for other descriptions of steps 201-202, please refer to the detailed description of steps 101-102 in the first embodiment, and the embodiment of the present invention will not be repeated here.

[0139] 203. For each pixel point on each horizontal pixel row on the image of the second slice, a preset edge recognition pixel array is constructed with the pixel point as a reference pixel point, where the edge recognition pixel array is a pixel row including a preset number of pixels; a pixel value of each pixel point in the edge recognition pixel array is obtained, and it is determined whether the distribution of the pixel values ​​of all the pixels in the edge recognition pixel array satisfies a preset distribution condition; if it is determined that the distribution of the pixel values ​​of all the pixels in the edge recognition pixel array satisfies the preset distribution condition, the edge recognition pixel array is determined to be a target edge recognition pixel array.

[0140] In the embodiment of the present invention, the sub-pixel edge refers to a technology for accurately locating the edge of an image in digital image processing, and its accuracy exceeds the single pixel resolution of the image. The edge recognition technology of the image can identify the edge of the colored area and the edge of a single cell from the image of the second slice. The traditional image edge recognition technology can only locate the edge on a certain pixel. For example, the left side of pixel A is a continuous bright area, and the right side of pixel A is a continuous dark area, then pixel A is determined to be an edge pixel. However, due to the insufficient accuracy during the image shooting process, the real edge should be at a certain position on the left or right side of pixel A, that is, at a certain position in the middle of pixel A and its adjacent pixels. This position is also called a sub-pixel position.

[0141] In the embodiment of the present invention, in order to identify the sub-pixel edge, an edge recognition pixel array is designed, such as Figure 3 As shown, the edge recognition pixel array in the embodiment of the present invention is a pixel row containing a preset number of pixel points, and the pixel row traverses from the first pixel in each row of pixels to the last pixel, that is, such an edge recognition pixel array is constructed with each pixel as a reference point. When the edge recognition pixel array moves from the leftmost side of a row of pixels to the rightmost side of a row of pixels, a number of virtual pixels are added to the rightmost side of the row of pixels, so that the edge recognition pixel array can traverse all pixels.

[0142] In an embodiment of the present invention, the pixel value of each pixel point in the edge recognition pixel array is obtained, and it is determined whether the distribution of the pixel values ​​of all the pixel points in the edge recognition pixel array meets the preset distribution conditions, which may include: obtaining the pixel value of each pixel point in the edge recognition pixel array and drawing a pixel value distribution curve, wherein the preset distribution condition is a pixel value distribution curve that meets the sub-pixel edge requirements, such as the curve having three curve segments with different trends, the appearance of peaks and troughs, the slope of the pixel value distribution curve exceeding a certain threshold, etc.

[0143] In the embodiment of the present invention, once it is determined that the distribution of pixel values ​​of all pixels in the edge recognition pixel array satisfies a preset distribution condition, the edge recognition pixel array is determined to be a target edge recognition pixel array, which indicates that the sub-pixel edge to be found is in the target edge recognition pixel array.

[0144] 204. For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, determine a sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the target edge recognition pixel array.

[0145] In the embodiment of the present invention, the sub-pixel edges to be found are all in the target edge recognition pixel array, and the sub-pixel edge position can be determined based on the distribution of pixel values ​​of all pixels in the target edge recognition pixel array. For example, the position with the largest pixel value change rate is determined as the sub-pixel edge position.

[0146] 205. Determine the sub-pixel edge on the image of the second slice according to all sub-pixel edge positions on the image of the second slice; determine the EBER in situ hybridization staining area in the image of the second slice according to the sub-pixel edge; determine that all cells in the EBER in situ hybridization staining area in the image of the second slice are Epstein-Barr virus positive cells;

[0147] In the embodiment of the present invention, the EBER in situ hybridization staining area and single cells identified by sub-pixel edges will be more precise, so that all cells located in the EBER in situ hybridization staining area in the image of the second slice can be more accurately determined as EB virus positive cells.

[0148] 206. Pairing the EB virus-positive cells in the image of the second slice with the cells in the image of the first slice, determining all target EB virus-positive cells corresponding to the EB virus-positive cells from the image of the first slice, and obtaining a training image;

[0149] 207. Acquire an image of a third slice.

[0150] 208. Determine a comparison image based on the image of the third slice.

[0151] 209. Use the training image and the comparison image to train the preset original virtual staining model to obtain a virtual staining model that is trained to convergence.

[0152] In the embodiment of the present invention, for other descriptions of steps 207-209, please refer to the detailed description of steps 104-106 in the first embodiment, and the embodiment of the present invention will not be repeated.

[0153] It can be seen that by implementing the intelligent virtual staining method for Epstein-Barr virus-related tumor pathology slice images in the embodiment of the present invention, the EBER in situ hybridization staining area and single cells identified by sub-pixel edges will be more accurate, thereby more accurately judging all cells located in the EBER in situ hybridization staining area in the image of the second slice, obtaining a more accurate and efficient training data set, and thereby improving the accuracy of artificial intelligence processing slice image technology.

[0154] In an optional embodiment, the edge recognition pixel array includes a first pixel segment, a second pixel segment, and a third pixel segment arranged in sequence, wherein the first pixel segment is a pixel row including a preset first number of pixel points, the second pixel segment is a pixel row including a preset second number of pixel points, and the third pixel segment is a pixel row including a preset third number of pixel points;

[0155] And, determining whether the distribution of pixel values ​​of all pixels in the edge recognition pixel array meets a preset distribution condition, including:

[0156] Determine whether the distribution of pixel values ​​of all pixels in a first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determine whether the distribution of pixel values ​​of all pixels in a second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determine whether the distribution of pixel values ​​of all pixels in a third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition;

[0157] If it is determined that the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition, then the edge recognition pixel array is determined to be a target edge recognition pixel array;

[0158] And, for each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, determining the sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the target edge recognition pixel array, including:

[0159] For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, the sub-pixel edge position is determined according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array.

[0160] In this optional embodiment, the left side of the sub-pixel edge may be an uncolored area, the right side is a colored area, and the middle is a transition area from the uncolored area to the colored area, wherein the sub-pixel edge is located in the transition area. The pixel value distributions of the colored area and the uncolored area are different, for example, the pixel value distributions of the colored area and the uncolored area are similar to uniform distributions, and the difference between the mean pixel value of the colored area and the mean pixel value of the uncolored area is greater than a preset threshold; the pixel value distribution of the transition area generally has a change process, for example, it can be fitted as a curve with a slope from small to large and then to small.

[0161] Therefore, according to the above-mentioned sub-pixel edge position distribution law, the embodiment of the present invention designs a three-stage method for finding the sub-pixel edge position, such as Figure 4 As shown, this optional embodiment obtains the target edge recognition pixel array by analyzing the distribution of pixel values ​​in the first pixel segment, the second pixel segment, and the third pixel segment; and then determines the sub-pixel edge position according to the distribution of pixel values ​​of all pixel points in the second pixel segment. It can be seen that this optional embodiment proposes a sub-pixel edge position search method that is more in line with the sub-pixel position distribution law, and can more accurately and reasonably identify the sub-pixel edge position, so as to more accurately determine all cells located in the EBER in situ hybridization staining area in the image of the second slice, obtain a more accurate and efficient training data set, and then improve the accuracy of artificial intelligence processing slice image technology.

[0162] In another optional embodiment, the method may further include:

[0163] For each pixel point on each horizontal pixel row on the image of the second slice, the edge identification pixel array corresponding to the edge identification pixel array is determined, the first pixel value mean and the first pixel value variance of the pixel points in the first pixel segment of the edge identification pixel array are determined, the second pixel value mean and the second pixel value variance of the pixel points in the second pixel segment of the edge identification pixel array are determined, and the third pixel value mean and the third pixel value variance of the pixel points in the third pixel segment of the edge identification pixel array are determined;

[0164] And, determining whether the distribution of pixel values ​​of all pixels in a first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determining whether the distribution of pixel values ​​of all pixels in a second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determining whether the distribution of pixel values ​​of all pixels in a third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition, including:

[0165] Determine whether a first pixel segment of the edge recognition pixel array satisfies that a difference between a first pixel value mean and a third pixel value mean is greater than a preset mean difference threshold and a first pixel value variance is less than a preset first variance threshold; whether a second pixel segment of the edge recognition pixel array satisfies that a second pixel value mean is between the first pixel value mean and the third pixel value mean and a second pixel value variance is greater than a preset second variance threshold; whether a third pixel segment of the edge recognition pixel array satisfies that a difference between a third pixel value mean and a second pixel value mean is greater than a preset mean difference threshold and a third pixel value variance is less than a preset third variance threshold;

[0166] And, for each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, determining the sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array, including:

[0167] For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, the pixel value change slope of each pixel point in the second pixel segment is determined according to the distribution of pixel values ​​of all pixel points in the second pixel segment of the target edge recognition pixel array; the pixel point with the largest pixel value change slope in the second pixel segment is determined as the target pixel point, and the sub-pixel edge position is determined according to the pixel value change slopes of the target pixel point and the neighboring pixel points adjacent to the target pixel point.

[0168] In this optional embodiment, for the pixels in the first pixel segment, the second pixel segment and the third pixel segment, the pixel value mean and the pixel value variance are calculated. The left side of the sub-pixel edge may be an uncolored area, the right side is a colored area, and the middle is a transition area from the uncolored area to the colored area, wherein the sub-pixel edge position is in the transition area. The mean of the pixel values ​​of the colored area and the uncolored area should exceed a certain value, and the pixel values ​​in the colored area and the pixel values ​​in the uncolored area should be evenly distributed, and the variance can just reflect whether the pixel values ​​are evenly distributed. When the variance is large, it means that the distribution of the pixel values ​​is more dispersed and not evenly distributed. On the other hand, the transition area is a transition area from the colored area to the uncolored area, so its pixel value should be able to reflect the process of changing from one value to another, which requires that the variance of the pixel values ​​in the transition area is greater than a certain value and its mean is between the mean of the pixel values ​​in the uncolored area and the mean of the pixel values ​​in the colored area.

[0169] Therefore, in this optional embodiment, only when it is determined that the difference between the pixel mean in the first pixel segment and the pixel mean in the third pixel segment of the edge recognition pixel array exceeds a certain value, the pixel mean in the first pixel segment is between the pixel mean in the first pixel segment and the pixel mean in the third pixel segment, and the pixel variance in the first pixel segment and the pixel variance in the second pixel segment are less than a certain value, and the pixel variance in the first pixel segment is greater than a certain value, can the edge recognition pixel array at this time be determined to be the target edge recognition pixel array.

[0170] In this optional embodiment, the sub-pixel edge is within the second pixel segment, that is, within the transition region, and the sub-pixel edge position is the position where the pixel value changes most dramatically, and is also near the pixel point with the largest slope of the pixel value change. Therefore, this optional embodiment first calculates the pixel value change slope of each pixel point in the second pixel segment, then finds the pixel point with the largest pixel value change slope, and finally determines the sub-pixel edge position based on the pixel value change slopes of the two pixel points to the left and right of the pixel point with the largest pixel value change slope. For example: it is determined that the pixel point with coordinate 4 is the pixel point with the largest pixel value change slope, and its slope is 8, and the pixel point with coordinate 3 has a pixel value change slope of 6, and the pixel point with coordinate 5 has a pixel value change slope of 5, then the sub-pixel edge position can be determined as:

[0171]

[0172] Then the position of the sub-pixel edge is 3.95.

[0173] It can be seen that this optional embodiment, by analyzing the mean value and variance of the pixel values ​​in the first pixel segment, the second pixel segment and the third pixel segment, can further improve the accuracy of finding the sub-pixel edge position, thereby more accurately determining all cells located in the EBER in situ hybridization staining area in the image of the second slice, obtaining a more accurate and efficient training data set, and thereby improving the accuracy of artificial intelligence processing of slice images.

[0174] In yet another optional embodiment, the method may further include:

[0175] For the edge recognition pixel array corresponding to each pixel point on each horizontal pixel row on the image of the second slice, the pixel values ​​of all the pixel points in the first pixel segment of the edge recognition pixel array are fitted to the first straight line, the pixel values ​​of all the pixel points in the second pixel segment of the edge recognition pixel array are fitted to the second straight line, and the pixel values ​​of all the pixel points in the third pixel segment of the edge recognition pixel array are fitted to the third straight line;

[0176] And, determining whether the distribution of pixel values ​​of all pixels in a first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determining whether the distribution of pixel values ​​of all pixels in a second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determining whether the distribution of pixel values ​​of all pixels in a third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition, including:

[0177] Determine whether the absolute value of the slope of the first straight line is less than a preset first slope threshold, determine whether the slope of the second straight line is greater than a preset second slope threshold, and determine whether the slope of the third straight line is less than a preset first slope threshold;

[0178] And, for each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, determining the sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array, including:

[0179] For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, a pixel value change curve of all pixel points in the second pixel segment is fitted according to the distribution of pixel values ​​of all pixel points in the second pixel segment of the target edge recognition pixel array, and the position with the largest absolute value of the slope on the pixel value change curve is determined as the sub-pixel edge position.

[0180] In this optional embodiment, for the pixels in the first pixel segment, the second pixel segment and the third pixel segment, a straight line of pixel values ​​is fitted. The left side of the sub-pixel edge may be an uncolored area, the right side is a colored area, and the middle is a transition area from the uncolored area to the colored area, wherein the sub-pixel edge position is within the transition area. The mean of the pixel values ​​of the colored area and the uncolored area should exceed a certain value, and the pixel values ​​of the colored area and the pixel values ​​of the uncolored area should be evenly distributed, therefore, the straight line fitted by the pixel values ​​of the colored area and the uncolored area should be an approximately horizontal straight line, that is, the slope of the straight line should be less than a certain value. On the other hand, the transition area is a transition area from the colored area to the uncolored area, so its pixel values ​​should be able to reflect the process of changing from one value to another, which requires that the slope of the straight line fitted by the pixel values ​​in the transition area should be greater than a certain value.

[0181] Therefore, in this optional embodiment, only when it is determined that the absolute value of the slope of the first straight line is less than the preset first slope threshold, the slope of the second straight line is greater than the preset second slope threshold, and the slope of the third straight line is less than the preset first slope threshold, can the edge recognition pixel array at this time be determined to be the target edge recognition pixel array.

[0182] In this optional embodiment, the sub-pixel edge is within the second pixel segment, that is, within the transition region, and the sub-pixel edge position is the position where the pixel value changes most dramatically. Therefore, this optional embodiment fits the pixel value change curve of all pixels in the second pixel segment, and determines that the position where the absolute value of the slope on the pixel value change curve is the largest is the sub-pixel edge position.

[0183] It can be seen that this optional embodiment, by fitting the pixel values ​​in the first pixel segment, the second pixel segment and the third pixel segment into a straight line, can more accurately determine the target edge recognition pixel array; then by fitting the pixel values ​​of all pixel points in the second pixel segment into a curve, determining the position on the curve with the largest absolute value of the slope as the sub-pixel edge position, the accuracy of finding the sub-pixel edge position can be further improved, thereby more accurately determining all cells located in the EBER in situ hybridization staining area in the image of the second slice, obtaining a more accurate and efficient training data set, and thereby improving the accuracy of artificial intelligence processing of slice images.

[0184] Embodiment 3

[0185] See also Figure 5 , Figure 5 Schematic diagram of the structure of an intelligent virtual staining device for Epstein-Barr virus-related tumor pathology slice images disclosed in an embodiment of the present invention. Figure 5 As shown, the intelligent virtual staining device for EB virus-related tumor pathological slice images may include:

[0186] A first acquisition module 301 is used to acquire an image of a first slice, where the first slice is an ex vivo tissue slice containing EB virus-positive cells;

[0187] The second acquisition module 302 is used to acquire an image of a second slice, where the second slice is a slice obtained by performing EBER in situ hybridization on the first slice after washing and staining;

[0188] A training image module 303 is used to obtain a training image based on the image of the first slice and the image of the second slice, wherein the training image is an image in which all target EB virus cells are marked on the image of the first slice;

[0189] A third acquisition module 304 is used to acquire an image of a third slice, where the third slice is an ex vivo tissue slice containing EB virus-negative cells;

[0190] A comparison image module 305 is used to determine a comparison image based on the image of the third slice, where the comparison image is an image in which all target EB virus-negative cells are marked on the image of the third slice;

[0191] The model training module 306 is used to train the preset original virtual staining model using the training image and the comparison image to obtain a virtual staining model that has been trained to convergence. The virtual staining model is used to perform a virtual staining operation on the input EB virus-related tumor pathology section image. The virtual staining operation includes: representing the cell nuclei of EB virus-positive cells in the EB virus-related tumor pathology section image with a first color, and representing the cell nuclei of EB virus-negative cells in the EB virus-related tumor pathology section image with a second color.

[0192] It can be seen that the intelligent virtual staining device for EB virus-related tumor pathological slice images in the embodiment of the present invention can mark all EB virus-positive cells on the image of the first slice through the image of the first slice and the image of the second slice, thereby obtaining a training image; at the same time, mark all EB virus-negative cells on the image of the third slice, thereby obtaining a comparison image, and finally use the training image and the comparison image to train the preset original virtual staining model to obtain a virtual staining model trained to convergence, which can mark the EB virus-positive cells on the input target pathological image and perform a virtual staining operation. In the present invention, the biomarker method is only used as an auxiliary means to obtain a more efficient training data set, thereby reducing the complexity of the operation process of the artificial intelligence processing slice image technology and improving the intelligence level of the artificial intelligence processing slice image technology.

[0193] In an optional embodiment, the specific manner in which the training image module 303 obtains the training image according to the image of the first slice and the image of the second slice may include:

[0194] According to the image of the second slice, determining the Epstein-Barr virus positive cells marked after EBER in situ hybridization in the image of the second slice;

[0195] The EB virus-positive cells in the image of the second slice are paired with the cells in the image of the first slice, and all target EB virus-positive cells corresponding to the EB virus-positive cells are determined from the image of the first slice to obtain a training image.

[0196] It can be seen that this optional embodiment pairs the EB virus-positive cells in the image of the second slice with the cells in the image of the first slice, thereby accurately determining all target EB virus-positive cells corresponding to the EB virus-positive cells in the first slice and obtaining a more effective training image.

[0197] In another optional embodiment, the training image module 303 pairs the EB virus-positive cells in the image of the second slice with the cells in the image of the first slice, and the specific method of determining all target EB virus-positive cells corresponding to the EB virus-positive cells from the image of the first slice may include:

[0198] Converting the image of the first slice and the image of the second slice into grayscale images and performing normalization processing to obtain a grayscale image of the first slice and a grayscale image of the second slice;

[0199] Extracting a first registration image feature on the grayscale image of the first slice, and extracting a second registration image feature on the grayscale image of the second slice; wherein the registration image feature includes one or more of a key point, an edge, a curve, a surface, and an invariant moment on the corresponding grayscale image;

[0200] According to the first registration image feature and the second registration image feature, registering the grayscale image of the first slice and the grayscale image of the second slice to obtain an image registration result;

[0201] The image registration result is input into a pre-trained registration quality control model to obtain the registration quality assessment score output by the registration quality control model. If the registration quality assessment score is greater than or equal to the preset registration quality threshold, all target EB virus-positive cells corresponding to EB virus-positive cells are determined from the image of the first slice based on the image registration result.

[0202] It can be seen that this optional embodiment, through the improved image registration method and the addition of a registration quality control step, can obtain a more accurate registration result between the EB virus-positive cells in the image of the second slice and the image of the first slice, thereby more accurately determining all target positive cells corresponding to the EB virus-positive cells from the image of the first slice, thereby providing a more efficient and accurate training data set for the subsequent model training step.

[0203] In another optional embodiment, the training image module 303 determines the specific manner of the EB virus positive cells in the image of the second slice after EBER in situ hybridization based on the image of the second slice, which may include:

[0204] Performing sub-pixel edge recognition processing on the image of the second slice to obtain sub-pixel edges on the image of the second slice; determining an EBER in situ hybridization staining area in the image of the second slice according to the sub-pixel edges;

[0205] Determine that all cells located in the EBER in situ hybridization staining area in the image of the second slice are EB virus positive cells;

[0206] The specific method in which the training image module 303 performs sub-pixel edge recognition processing on the image of the second slice and obtains the sub-pixel edge on the image of the second slice may include:

[0207] For each pixel point on each horizontal pixel row on the image of the second slice, taking the pixel point as a reference pixel point, constructing a preset edge recognition pixel array, the edge recognition pixel array being a pixel row including a preset number of pixel points; obtaining a pixel value of each pixel point in the edge recognition pixel array, and determining whether the distribution of the pixel values ​​of all the pixel points in the edge recognition pixel array satisfies a preset distribution condition, and if it is determined that the distribution of the pixel values ​​of all the pixel points in the edge recognition pixel array satisfies the preset distribution condition, determining the edge recognition pixel array as a target edge recognition pixel array;

[0208] For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, determining a sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the target edge recognition pixel array;

[0209] The sub-pixel edge on the image of the second slice is determined according to all sub-pixel edge positions on the image of the second slice.

[0210] It can be seen that the intelligent virtual staining device for EB virus-related tumor pathology slice images in the embodiment of the present invention can more accurately identify the EBER in situ hybridization staining area and single cells through sub-pixel edges, thereby more accurately judging all cells located in the EBER in situ hybridization staining area in the image of the second slice, obtaining a more accurate and efficient training data set, and thus improving the accuracy of artificial intelligence processing slice image technology.

[0211] In yet another optional embodiment, the edge recognition pixel array includes a first pixel segment, a second pixel segment, and a third pixel segment arranged in sequence, wherein the first pixel segment is a pixel row including a preset first number of pixel points, the second pixel segment is a pixel row including a preset second number of pixel points, and the third pixel segment is a pixel row including a preset third number of pixel points;

[0212] Furthermore, the specific manner in which the training image module 303 determines whether the distribution of pixel values ​​of all pixels in the edge recognition pixel array meets the preset distribution condition may include:

[0213] Determine whether the distribution of pixel values ​​of all pixels in a first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determine whether the distribution of pixel values ​​of all pixels in a second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determine whether the distribution of pixel values ​​of all pixels in a third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition;

[0214] If it is determined that the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition, then the edge recognition pixel array is determined to be a target edge recognition pixel array;

[0215] Furthermore, the training image module 303 determines the sub-pixel edge position of each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice according to the distribution of pixel values ​​of all pixels in the target edge recognition pixel array, which may include:

[0216] For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, the sub-pixel edge position is determined according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array.

[0217] It can be seen that this optional embodiment can more accurately and reasonably identify the sub-pixel edge position, thereby more accurately determining all cells located within the EBER in situ hybridization staining area in the image of the second slice, obtaining a more accurate and efficient training data set, and thereby improving the accuracy of artificial intelligence processing of slice image technology.

[0218] In yet another optional embodiment, Figure 6 As shown, the device may also include:

[0219] The pixel analysis module 307 is used to determine, for each pixel point on each horizontal pixel row on the image of the second slice, a first pixel value mean and a first pixel value variance of the pixel points in a first pixel segment of the edge identification pixel array, determine a second pixel value mean and a second pixel value variance of the pixel points in a second pixel segment of the edge identification pixel array, and determine a third pixel value mean and a third pixel value variance of the pixel points in a third pixel segment of the edge identification pixel array;

[0220] Furthermore, the specific manner in which the training image module 303 determines whether the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determines whether the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determines whether the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition may include:

[0221] Determine whether a first pixel segment of the edge recognition pixel array satisfies that a difference between a first pixel value mean and a third pixel value mean is greater than a preset mean difference threshold and a first pixel value variance is less than a preset first variance threshold; whether a second pixel segment of the edge recognition pixel array satisfies that a second pixel value mean is between the first pixel value mean and the third pixel value mean and a second pixel value variance is greater than a preset second variance threshold; whether a third pixel segment of the edge recognition pixel array satisfies that a difference between a third pixel value mean and a second pixel value mean is greater than a preset mean difference threshold and a third pixel value variance is less than a preset third variance threshold;

[0222] Furthermore, the training image module 303 determines the sub-pixel edge position of each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array, which may include:

[0223] For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, the pixel value change slope of each pixel point in the second pixel segment is determined according to the distribution of pixel values ​​of all pixel points in the second pixel segment of the target edge recognition pixel array; the pixel point with the largest pixel value change slope in the second pixel segment is determined as the target pixel point, and the sub-pixel edge position is determined according to the pixel value change slopes of the target pixel point and the neighboring pixel points adjacent to the target pixel point.

[0224] It can be seen that this optional embodiment, by analyzing the mean value and variance of the pixel values ​​in the first pixel segment, the second pixel segment and the third pixel segment, can further improve the accuracy of finding the sub-pixel edge position, thereby more accurately determining all cells located in the EBER in situ hybridization staining area in the image of the second slice, obtaining a more accurate and efficient training data set, and thereby improving the accuracy of artificial intelligence processing of slice images.

[0225] In another optional embodiment, Figure 6 As shown, the device may also include:

[0226] A pixel fitting module 308 is used to fit the edge recognition pixel array corresponding to each pixel point on each horizontal pixel row on the image of the second slice, and to fit the pixel values ​​of all the pixel points in a first pixel segment of the edge recognition pixel array to a first straight line, to fit the pixel values ​​of all the pixel points in a second pixel segment of the edge recognition pixel array to a second straight line, and to fit the pixel values ​​of all the pixel points in a third pixel segment of the edge recognition pixel array to a third straight line;

[0227] Furthermore, the specific manner in which the training image module 303 determines whether the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determines whether the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determines whether the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition may include:

[0228] Determine whether the absolute value of the slope of the first straight line is less than a preset first slope threshold, determine whether the slope of the second straight line is greater than a preset second slope threshold, and determine whether the slope of the third straight line is less than a preset first slope threshold;

[0229] Furthermore, the training image module 303 determines the sub-pixel edge position of each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array, which may include:

[0230] For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, a pixel value change curve of all pixel points in the second pixel segment is fitted according to the distribution of pixel values ​​of all pixel points in the second pixel segment of the target edge recognition pixel array, and the position with the largest absolute value of the slope on the pixel value change curve is determined as the sub-pixel edge position.

[0231] It can be seen that this optional embodiment, by fitting the pixel values ​​in the first pixel segment, the second pixel segment and the third pixel segment into a straight line, can more accurately determine the target edge recognition pixel array; then by fitting the pixel values ​​of all pixel points in the second pixel segment into a curve, determining the position on the curve with the largest absolute value of the slope as the sub-pixel edge position, the accuracy of finding the sub-pixel edge position can be further improved, thereby more accurately determining all cells located in the EBER in situ hybridization staining area in the image of the second slice, obtaining a more accurate and efficient training data set, and thereby improving the accuracy of artificial intelligence processing of slice images.

[0232] Embodiment 4

[0233] See also Figure 7 , Figure 7 FIG. 1 is a schematic diagram of the structure of another intelligent virtual staining device for EB virus-related tumor pathology slice images disclosed in an embodiment of the present invention. Figure 7 As shown, the intelligent virtual staining device for EB virus-related tumor pathological slice images may include:

[0234] A memory 401 storing executable program codes;

[0235] a processor 402 coupled to the memory 401;

[0236] The processor 402 calls the executable program code stored in the memory 401 to execute the steps of the intelligent virtual staining method for EB virus-related tumor pathology section images described in the first embodiment of the present invention or the second embodiment of the present invention.

[0237] Embodiment 5

[0238] An embodiment of the present invention discloses a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute the steps in the intelligent virtual staining method of Epstein-Barr virus-related tumor pathology section images described in Embodiment 1 or Embodiment 2 of the present invention.

[0239] The device embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, i.e., they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

Claims

1. An intelligent virtual staining method for Epstein-Barr virus-related tumor pathological section images, characterized in that: The method comprises: Acquiring an image of a first slice, wherein the first slice is an ex vivo tissue slice containing Epstein-Barr virus-positive cells; Acquire an image of a second slice, where the second slice is obtained by performing EBER in situ hybridization on the first slice after washing and staining; Obtaining a training image based on the image of the first slice and the image of the second slice, wherein the training image is an image in which all target EB virus-positive cells are marked on the image of the first slice; acquiring an image of a third slice, wherein the third slice is an ex vivo tissue slice containing EB virus-negative cells; Determine a comparison image according to the image of the third slice, wherein the comparison image is an image in which all target EB virus-negative cells are marked on the image of the third slice; The preset original virtual staining model is trained using the training image and the comparison image to obtain a virtual staining model trained to convergence, wherein the virtual staining model is used to perform a virtual staining operation on the input EB virus-related tumor pathology section image, and the virtual staining operation includes: The cell nuclei identified as EB virus positive cells in the EB virus related tumor pathological section image are represented by a first color, and the cell nuclei identified as EB virus negative cells in the EB virus related tumor pathological section image are represented by a second color.

2. The intelligent virtual staining method for EB virus-related tumor pathological section images according to claim 1, characterized in that: The step of obtaining a training image according to the first slice image and the second slice image comprises: According to the image of the second slice, determining the Epstein-Barr virus positive cells marked after EBER in situ hybridization in the image of the second slice; The EB virus-positive cells in the image of the second slice are paired with the cells in the image of the first slice, and all target EB virus-positive cells corresponding to the EB virus-positive cells are determined from the image of the first slice to obtain a training image.

3. The intelligent virtual staining method for EB virus-related tumor pathological section images according to claim 2, characterized in that: The step of pairing the EB virus-positive cells in the image of the second slice with the cells in the image of the first slice, and determining all target EB virus-positive cells corresponding to the EB virus-positive cells from the image of the first slice, comprises: Converting the image of the first slice and the image of the second slice into grayscale images and performing normalization processing to obtain a grayscale image of the first slice and a grayscale image of the second slice; Extracting a first registration image feature on the grayscale image of the first slice, and extracting a second registration image feature on the grayscale image of the second slice; wherein the registration image feature includes one or more of a key point, an edge, a curve, a surface, and an invariant moment on the corresponding grayscale image; Registering the grayscale image of the first slice and the grayscale image of the second slice according to the first registered image feature and the second registered image feature to obtain an image registration result; The image registration result is input into a pre-trained registration quality control model to obtain a registration quality assessment score output by the registration quality control model. If the registration quality assessment score is greater than or equal to a preset registration quality threshold, all target Epstein-Barr virus-positive cells corresponding to the Epstein-Barr virus-positive cells are determined from the image of the first slice based on the image registration result.

4. The intelligent virtual staining method for EB virus-related tumor pathological section images according to claim 2, characterized in that: Determining, based on the image of the second slice, Epstein-Barr virus-positive cells that are marked after EBER in situ hybridization in the image of the second slice comprises: Performing sub-pixel edge recognition processing on the image of the second slice to obtain sub-pixel edges on the image of the second slice; determining an EBER in situ hybridization staining area in the image of the second slice according to the sub-pixel edges; Determining that all cells located in the EBER in situ hybridization staining area in the image of the second slice are Epstein-Barr virus positive cells; The performing of sub-pixel edge recognition processing on the image of the second slice to obtain the sub-pixel edge on the image of the second slice specifically includes: For each pixel point on each horizontal pixel row on the image of the second slice, a preset edge recognition pixel array is constructed with the pixel point as a reference pixel point, wherein the edge recognition pixel array is a pixel row including a preset number of pixel points; a pixel value of each pixel point in the edge recognition pixel array is obtained, and it is determined whether the distribution of the pixel values ​​of all the pixel points in the edge recognition pixel array satisfies a preset distribution condition; if it is determined that the distribution of the pixel values ​​of all the pixel points in the edge recognition pixel array satisfies the preset distribution condition, the edge recognition pixel array is determined to be a target edge recognition pixel array; For each target edge recognition pixel array corresponding to each horizontal pixel row on the image of the second slice, determining a sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the target edge recognition pixel array; The sub-pixel edge on the image of the second slice is determined according to all the sub-pixel edge positions on the image of the second slice.

5. The intelligent virtual staining method for EB virus-related tumor pathological section images according to claim 4, characterized in that: The edge recognition pixel array includes a first pixel segment, a second pixel segment, and a third pixel segment arranged in sequence, wherein the first pixel segment is a pixel row including a preset first number of pixel points, the second pixel segment is a pixel row including a preset second number of pixel points, and the third pixel segment is a pixel row including a preset third number of pixel points; And, the determining whether the distribution of pixel values ​​of all pixels in the edge recognition pixel array meets a preset distribution condition includes: Determine whether the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determine whether the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determine whether the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition; If it is determined that the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition, then the edge recognition pixel array is determined to be a target edge recognition pixel array; And, for each of the target edge recognition pixel arrays corresponding to each horizontal pixel row on the image of the second slice, determining the sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the target edge recognition pixel array comprises: For each of the target edge recognition pixel arrays corresponding to each horizontal pixel row on the image of the second slice, the sub-pixel edge position is determined according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array.

6. The intelligent virtual staining method for EB virus-related tumor pathological section images according to claim 5, characterized in that: The method further comprises: For each pixel point on each horizontal pixel row on the image of the second slice, the edge identification pixel array corresponding to the edge identification pixel array determines a first pixel value mean and a first pixel value variance of the pixel points in the first pixel segment of the edge identification pixel array, determines a second pixel value mean and a second pixel value variance of the pixel points in the second pixel segment of the edge identification pixel array, and determines a third pixel value mean and a third pixel value variance of the pixel points in the third pixel segment of the edge identification pixel array; Furthermore, the step of determining whether the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determining whether the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determining whether the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition comprises: Determine whether the first pixel segment of the edge recognition pixel array satisfies that the difference between the first pixel value mean and the third pixel value mean is greater than a preset mean difference threshold and the first pixel value variance is less than a preset first variance threshold; whether the second pixel segment of the edge recognition pixel array satisfies that the second pixel value mean is between the first pixel value mean and the third pixel value mean and the second pixel value variance is greater than a preset second variance threshold; whether the third pixel segment of the edge recognition pixel array satisfies that the difference between the third pixel value mean and the second pixel value mean is greater than a preset mean difference threshold and the third pixel value variance is less than a preset third variance threshold; And, for each of the target edge recognition pixel arrays corresponding to each horizontal pixel row on the image of the second slice, determining the sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array, including: For each of the target edge identification pixel arrays corresponding to each horizontal pixel row on the image of the second slice, determine the pixel value change slope of each pixel in the second pixel segment according to the distribution of pixel values ​​of all pixel points in the second pixel segment of the target edge identification pixel array; determine the pixel point with the largest pixel value change slope in the second pixel segment as the target pixel point, and determine the sub-pixel edge position according to the pixel value change slopes of the target pixel point and the neighboring pixel points adjacent to the target pixel point.

7. The intelligent virtual staining method for EB virus-related tumor pathological section images according to claim 5, characterized in that: The method further comprises: For the edge recognition pixel array corresponding to each pixel point on each horizontal pixel row on the image of the second slice, pixel values ​​of all pixel points in the first pixel segment of the edge recognition pixel array are fitted to form a first straight line, pixel values ​​of all pixel points in the second pixel segment of the edge recognition pixel array are fitted to form a second straight line, and pixel values ​​of all pixel points in the third pixel segment of the edge recognition pixel array are fitted to form a third straight line; Furthermore, the step of determining whether the distribution of pixel values ​​of all pixels in the first pixel segment of the edge recognition pixel array satisfies a preset first distribution condition, determining whether the distribution of pixel values ​​of all pixels in the second pixel segment of the edge recognition pixel array satisfies a preset second distribution condition, and determining whether the distribution of pixel values ​​of all pixels in the third pixel segment of the edge recognition pixel array satisfies a preset third distribution condition comprises: Determine whether the absolute value of the slope of the first straight line is less than a preset first slope threshold, determine whether the slope of the second straight line is greater than a preset second slope threshold, and determine whether the slope of the third straight line is less than a preset first slope threshold; And, for each of the target edge recognition pixel arrays corresponding to each horizontal pixel row on the image of the second slice, determining the sub-pixel edge position according to the distribution of pixel values ​​of all pixels in the second pixel segment of the target edge recognition pixel array, including: For each of the target edge recognition pixel arrays corresponding to each horizontal pixel row on the image of the second slice, a pixel value change curve of all pixel points in the second pixel segment is fitted according to the distribution of pixel values ​​of all pixel points in the second pixel segment of the target edge recognition pixel array, and the position where the absolute value of the slope on the pixel value change curve is the largest is determined as the sub-pixel edge position.

8. An intelligent virtual staining device for Epstein-Barr virus-related tumor pathological slice images, characterized in that: The device comprises: A first acquisition module is used to acquire an image of a first slice, where the first slice is an ex vivo tissue slice containing EB virus-positive cells; A second acquisition module is used to acquire an image of a second slice, where the second slice is obtained by performing EBER in situ hybridization on the first slice after washing and staining; A training image module, used for obtaining a training image according to the image of the first slice and the image of the second slice, wherein the training image is an image in which all target EB virus cells are marked on the image of the first slice; A third acquisition module is used to acquire an image of a third slice, where the third slice is an ex vivo tissue slice containing EB virus-negative cells; A contrast image module, used to determine a contrast image based on the image of the third slice, wherein the contrast image is an image in which all target EB virus-negative cells are marked on the image of the third slice; A model training module is used to train a preset original virtual staining model using the training image and the comparison image to obtain a virtual staining model trained to convergence, wherein the virtual staining model is used to perform a virtual staining operation on an input Epstein-Barr virus-related tumor pathology section image, wherein the virtual staining operation includes: representing the cell nuclei of cells identified as Epstein-Barr virus-positive cells in the Epstein-Barr virus-related tumor pathology section image with a first color, and representing the cell nuclei of cells identified as Epstein-Barr virus-negative cells in the Epstein-Barr virus-related tumor pathology section image with a second color.

9. An intelligent virtual staining device for Epstein-Barr virus-related tumor pathological slice images, characterized in that: The device includes: a memory storing executable program code; a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the intelligent virtual staining method for EB virus-related tumor pathology section images as described in any one of claims 1-7.

10. A computer storage medium, characterized in that: The computer storage medium stores computer instructions, which, when called, are used to execute the intelligent virtual staining method for EB virus-related tumor pathology section images as described in any one of claims 1-7.

Images