Pathological typical view capture system based on artificial intelligence

Abstract

The invention provides a pathological typical view capturing system based on artificial intelligence. The pathological typical view capturing system comprises the following steps: firstly, receiving and processing a high-resolution full-slice image by using a slice data reading module, and cleaning image data to ensure the integrity and availability of the data; the image is optimized through a pathological image preprocessing module to improve the accuracy of subsequent analysis, and then pathological key features in the image are deeply mined through a deep learning feature extraction module by adopting an image feature mining model to lay a solid foundation for identifying a typical view; the pathological key features are subjected to graph neural network technology by a typical visual field identification module, and a typical visual field in a pathological image is accurately positioned; and finally, an identification result is displayed to a pathologist in a visual view screenshot manner through a typical view output module, so that further diagnosis and analysis are facilitated, the efficiency and accuracy of pathological image analysis are remarkably improved, and more reliable support is provided for the aspects of early diagnosis, prognosis evaluation, treatment effect monitoring and the like of diseases.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence recognition technology, and in particular to an artificial intelligence-based pathological typical visual field capture system. Background Art

[0002] In today's medical field, pathology plays a vital role. It provides important support for clinical diagnosis and treatment by studying the morphological and histological characteristics of tissues and cells. However, traditional pathological image analysis is often limited by manual operation and subjective judgment, especially the discovery and mining of typical fields of tissues and cells, which depends on the richness of experience, resulting in frequent uncertainty and error-prone problems.

[0003] With the rapid development of artificial intelligence technology, automated pathological image analysis based on deep learning and computer vision has gradually become possible. Existing pathological image analysis systems based on artificial intelligence mainly use deep learning technology, such as convolutional neural network (CNN), etc., to perform full-image automated analysis and diagnosis of pathological images. These systems usually learn the features in pathological images through training models, and perform tasks such as classification, segmentation or positioning. These systems have finished processing pathological images after predicting the lesions. However, the typical visual field cannot be pushed to the front end of the pathologist in time, and the screenshots of multiple typical visual fields cannot be automatically obtained and pushed, resulting in the pathologist being unable to locate and survey the lesion position and severity in time, and having to manually pay attention to and explore the typical areas of multiple pathological digital images to judge, and the manual operation is frequent, resulting in lower efficiency of reading films. Therefore, the present invention proposes a pathological typical visual field capture system based on artificial intelligence. Summary of the invention

[0004] The present invention provides a pathological typical visual field capture system based on artificial intelligence to solve the above problems.

[0005] The present invention provides a pathological typical visual field capture system based on artificial intelligence, comprising:

[0006] A slice data reading module is used to receive a full slice image and perform image data cleaning on the full slice image to obtain a valid full slice image;

[0007] Pathological image preprocessing module, used to optimize the effective full-slice image to obtain the optimized image;

[0008] Deep learning feature extraction module, which is used to extract key features of case pathology in optimized images based on pre-trained image feature mining models;

[0009] Typical field of view recognition module, used to determine the typical field of view of optimized images based on key pathological features through graph neural network;

[0010] The typical field of view output module is used to take a screenshot of the typical field of view and send the screenshot result directly to the target person.

[0011] Preferably, in a pathological typical visual field capture system based on artificial intelligence, the slice data reading module includes:

[0012] An image data recognition unit is used to obtain a grayscale image of the whole slice image and perform edge feature extraction on the grayscale image, and determine the number of intact tissue cells in the whole slice image based on the edge feature extraction result;

[0013] An image data cleaning unit, used to treat the whole slice image as an invalid image as a valid whole slice image when the number of intact tissue cells is greater than a preset value;

[0014] Otherwise, the whole slice image is regarded as an invalid image and is treated as an invalid whole slice image, and a whole slice image re-reading instruction is sent to the pathology database.

[0015] Preferably, in a pathological typical visual field capture system based on artificial intelligence, the pathological image preprocessing module includes:

[0016] A preliminary processing unit, used for inputting the effective full-slice image into a preset denoising model for denoising to obtain a denoised image;

[0017] An enhanced processing unit is used to obtain standard key features corresponding to the predicted pathological results of the effective full-slice image, and determine the morphological features corresponding to the diseased tissue cells and the cell-tissue connection relationship between the diseased tissue cells based on the standard key features;

[0018] According to the morphological features corresponding to the diseased tissue cells and the cell-tissue connection relationship between the diseased tissue cells, image details of the tissue cells in the denoised image are enhanced to obtain an optimized image;

[0019] The image screening unit is used to classify the optimized images based on typical field of view sampling features to obtain key optimized images and secondary optimized images.

[0020] Preferably, in a pathological typical visual field capture system based on artificial intelligence, a deep learning feature extraction module includes:

[0021] A recognition target determination unit is used to obtain the predicted pathological results of the effective full-slice image, and based on the predicted pathological results, control the pre-trained image feature mining model to perform recognition target determination;

[0022] The feature extraction unit is used to perform global feature recognition of different degrees of precision on the optimized image based on the image label according to the recognition target, extract the distribution characteristics of pathological cells on the effective full-slice image, and obtain the key pathological features of the optimized image.

[0023] Preferably, in a pathological typical visual field capture system based on artificial intelligence, the deep learning feature extraction module further includes:

[0024] A feature verification unit is used to compare the pathological key features with the standard key features corresponding to the predicted pathological results to determine whether there is a difference between the two;

[0025] If it does not exist, the current predicted pathological result is judged to be normal;

[0026] Otherwise, based on the standard key features, a search is performed in the preset pathology database to determine similar pathologies, establish a confusing pathology dataset, and simultaneously obtain the key feature differences between the pathology key features and the standard key features;

[0027] Based on the differences in key features, pathology comparison is performed in easily confused pathology datasets to obtain the best matching result, and the predicted pathology result is calibrated based on the best matching result.

[0028] Preferably, in a pathological typical visual field capture system based on artificial intelligence, the feature extraction unit includes:

[0029] A label recognition unit, used to recognize the image label corresponding to the optimization image;

[0030] A first feature extraction unit, configured to extract key pathological features of the optimized image based on the image detail enhancement result when the optimized image is a secondary optimized image;

[0031] A second feature extraction unit is used to perform grayscale classification on the pixel points based on the pixel grayscale values ​​corresponding to each pixel point of the optimized image when the optimized image is an important optimized image;

[0032] The optimized image is layered based on the grayscale grading results to obtain multiple layers, and pixel distribution features are extracted for each layer. The pixel distribution features of multiple layers corresponding to the same optimized image are fused to obtain the pixel distribution features of the entire layer. The pixel features of the entire layer are compared with the features of normal cells to obtain the key pathological features.

[0033] Preferably, in a pathological typical visual field capture system based on artificial intelligence, the typical visual field recognition module includes:

[0034] A key area marking unit is used to identify and mark the pathological key feature expression position of the optimized image through a graph neural network to determine the pathological key feature expression area;

[0035] An automatic marking calculation unit, used to automatically mark cells in the pathological key feature expression area, and calculate the concentration degree of the pathological key cells corresponding to the marked area according to the marking results;

[0036] The typical area screening unit is used to screen the pathological key feature expression area based on the concentration degree, determine the typical visual field expression area, and mark the typical visual field expression area based on the rectangular frame.

[0037] Preferably, in a pathological typical visual field capture system based on artificial intelligence, the typical visual field output module includes:

[0038] A visual field screenshot unit, used to screenshot a typical visual field performance area to obtain a typical visual field;

[0039] A target determination unit, used to search a preset person list based on the pathological judgment result corresponding to the typical visual field to determine the target person;

[0040] The result sending unit is used to obtain a typical visual field and send the typical visual field to a target person.

[0041] Preferably, in a pathological typical visual field capture system based on artificial intelligence, the result sending unit includes:

[0042] The error processing subunit is used to obtain a pre-stored standard typical visual field corresponding to the original pathology prediction result when the pathology prediction result is a calibration result;

[0043] The infected cells in the typical field of view and the pre-stored standard typical field of view are marked separately, and a complete infected cell is randomly extracted from the typical field of view and the pre-stored standard typical field of view for focusing and magnification to obtain a set of infected cell comparison images;

[0044] Compare the lesion details based on the infected cell control image, determine the difference in cytopathic effect, and generate a confusion distinction report based on the difference in cytopathic effect and the infected cell control image;

[0045] A visual field processing subunit is used to highlight key features in a typical visual field based on differences in cytopathic changes, and to specially highlight the differences in cytopathic changes, thereby generating a hidden typical visual field and adding it to a confusing distinction report;

[0046] The visual field sending subunit is used to send the confusion distinction report as an important attachment to the target person simultaneously with the typical visual field.

[0047] Preferably, in a pathological typical visual field capture system based on artificial intelligence, the pathological image preprocessing module further includes:

[0048] A field of view selection optimization unit, used to obtain a screenshot position of a typical field of view in a corresponding full-slice image and a slice sample sampling position corresponding to the full-slice image, and generate a first pathology capture feature;

[0049] and simultaneously obtain the second pathology capture feature corresponding to the historical typical of the same type, and generate a typical visual field capture feature set by combining the first pathology capture feature and the second pathology capture feature;

[0050] Based on the lesion degree corresponding to the pathological prediction result, the pathological capture features in the typical visual field capture feature set are classified to obtain typical visual field capture feature subsets corresponding to multiple lesion degrees;

[0051] Typical visual field sampling features are extracted for each typical visual field capture feature subset respectively, typical visual field sampling features corresponding to current diseases with different lesion degrees are determined, and the original typical visual field sampling features are updated.

[0052] Compared with the prior art, the present invention has at least the following beneficial effects:

[0053] The present invention first uses a slice data reading module to receive and process high-resolution full-slice images and cleans the image data to ensure the integrity and availability of the data; then the image is optimized by a pathological image preprocessing module, including denoising and enhancement, etc., to improve the accuracy of subsequent analysis. Thereafter, the image feature mining model is used to deeply mine the key pathological features in the image through a deep learning feature extraction module, laying a solid foundation for identifying typical visual fields; then the key pathological features are adopted by the typical visual field recognition module, and the typical visual field in the pathological image is accurately located through the graph neural network (GNN) technology; finally, the recognition results are presented to the pathologist in the form of an intuitive visual field screenshot through the typical visual field output module, which is convenient for further diagnosis and analysis. The design of the entire system ensures smooth cooperation between modules, and each module is the recipient and service provider of the output of the previous module, forming a coherent, efficient and automated pathological image analysis workflow, significantly improving the efficiency and accuracy of pathological image analysis, solving the problem of low efficiency of typical visual fields, and realizing the purpose of directly presenting the typical visual field in front of the corresponding pathologist terminal, effectively improving the efficiency of pathological analysis, and providing more reliable support for early diagnosis, prognosis evaluation and treatment effect monitoring of diseases.

[0054] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0055] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0057] Figure 1 A schematic diagram of a pathological typical visual field capture system based on artificial intelligence according to the present invention;

[0058] Figure 2 It is a schematic diagram of a slice data reading module of a pathological typical visual field capture system based on artificial intelligence of the present invention;

[0059] Figure 3 It is a schematic diagram of a pathological image preprocessing module of a pathological typical visual field capture system based on artificial intelligence of the present invention;

[0060] Figure 4 It is a schematic diagram of a deep learning feature extraction module of a pathological typical visual field capture system based on artificial intelligence of the present invention;

[0061] Figure 5 It is a schematic diagram of a typical visual field recognition module of a pathological typical visual field capture system based on artificial intelligence of the present invention;

[0062] Figure 6 This is a schematic diagram of a typical visual field output module of a pathological typical visual field capture system based on artificial intelligence in the present invention. DETAILED DESCRIPTION

[0063] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0064] Embodiment 1:

[0065] The present invention provides a pathological typical visual field capture system based on artificial intelligence, such as Figure 1 As shown, including:

[0066] A slice data reading module is used to receive a full slice image and perform image data cleaning on the full slice image to obtain a valid full slice image;

[0067] Pathological image preprocessing module, used to optimize the effective full-slice image to obtain the optimized image;

[0068] Deep learning feature extraction module, which is used to extract key features of case pathology in optimized images based on pre-trained image feature mining models;

[0069] Typical field of view recognition module, used to determine the typical field of view of optimized images based on key pathological features through graph neural network;

[0070] The typical field of view output module is used to take a screenshot of the typical field of view and send the screenshot result directly to the target person.

[0071] In this embodiment, when the burst is sent to the target person, the system will automatically perform backup storage to facilitate subsequent access and query.

[0072] In this embodiment, the image feature mining model is a convolutional neural network model.

[0073] The beneficial effects of the above technical solution are as follows: the present invention first uses the slice data reading module to receive and process the high-resolution full slice image and clean the image data to ensure the integrity and availability of the data; then the image is optimized by the pathological image preprocessing module, including denoising and enhancement, etc., to improve the accuracy of subsequent analysis, and then the image feature mining model is used by the deep learning feature extraction module to deeply mine the key pathological features in the image, laying a solid foundation for identifying the typical visual field; then the key pathological features are adopted by the typical visual field recognition module, and the typical visual field in the pathological image is accurately located by the graph neural network (GNN) technology; finally, the recognition results are presented to the pathologist in the form of an intuitive visual field screenshot through the typical visual field output module, which is convenient for further diagnosis and analysis. The design of the entire system ensures smooth cooperation between modules, and each module is the receiver and service provider of the output of the previous module, forming a coherent, efficient and automated pathological image analysis workflow, significantly improving the efficiency and accuracy of pathological image analysis, solving the problem of low efficiency of the typical visual field, and realizing the purpose of directly presenting the typical visual field in front of the corresponding pathologist terminal, effectively improving the efficiency of pathological analysis, and providing more reliable support for early diagnosis, prognosis evaluation and treatment effect monitoring of diseases.

[0074] Embodiment 2:

[0075] Based on Example 1, the slice data reading module, such as Figure 2 As shown, including:

[0076] An image data recognition unit is used to obtain a grayscale image of the whole slice image and perform edge feature extraction on the grayscale image, and determine the number of intact tissue cells in the whole slice image based on the edge feature extraction result;

[0077] An image data cleaning unit, used to treat the whole slice image as an invalid image as a valid whole slice image when the number of intact tissue cells is greater than a preset value;

[0078] Otherwise, the whole slice image is regarded as an invalid image and is treated as an invalid whole slice image, and a whole slice image re-reading instruction is sent to the pathology database.

[0079] Beneficial effects of the above technical solution: The present invention detects the validity of the received full-slice image through the image data recognition unit and the image data invasion unit to ensure the integrity and availability of the data.

[0080] Embodiment 3:

[0081] Based on Example 1, the pathological image preprocessing module, such as Figure 3 As shown, including:

[0082] A preliminary processing unit, used for inputting the effective full-slice image into a preset denoising model for denoising to obtain a denoised image;

[0083] An enhanced processing unit is used to obtain standard key features corresponding to the predicted pathological results of the effective full-slice image, and determine the morphological features corresponding to the diseased tissue cells and the cell-tissue connection relationship between the diseased tissue cells based on the standard key features;

[0084] According to the morphological features corresponding to the diseased tissue cells and the cell-tissue connection relationship between the diseased tissue cells, image details of the tissue cells in the denoised image are enhanced to obtain an optimized image;

[0085] The image screening unit is used to classify the optimized images based on typical field of view sampling features to obtain key optimized images and secondary optimized images.

[0086] In this embodiment, the typical field of view sampling feature refers to the optimal position of lesion sampling (i.e., slice sample sampling) corresponding to the typical field of view inferred based on the pathological prediction result of the lesion position corresponding to the full slice image currently being processed, and the distribution area on the full slice image where the typical field of view is located.

[0087] The beneficial effects of the above technical solution are as follows: the present invention performs denoising and enhancement processing on the effective full-section image through the preliminary processing unit and the enhanced processing unit, which is beneficial to improving the accuracy of typical field of view determination, wherein the enhanced processing unit obtains the standard key features corresponding to the predicted pathological results of the effective full-section image, determines the morphological features corresponding to the diseased tissue cells and the cell-tissue connection relationship between the diseased tissue cells based on the standard key features, and enhances the image details of the tissue cells in the denoised image, thereby obtaining an optimized image that meets the content information highlighted by the full-section image, which can better highlight the characteristics of the pathological cells in the image, and is beneficial to improving the accuracy of pathological key feature extraction; finally, the image screening unit classifies multiple optimized images to obtain important optimized images and secondary optimized images, which provide a basis for determining the recognition strategy of key features.

[0088] Embodiment 4:

[0089] Based on Example 1, a deep learning feature extraction module such as Figure 4 As shown, including:

[0090] A recognition target determination unit is used to obtain the predicted pathological results of the effective full-slice image, and based on the predicted pathological results, control the pre-trained image feature mining model to perform recognition target determination;

[0091] The feature extraction unit is used to perform global feature recognition of different degrees of precision on the optimized image based on the image label according to the recognition target, extract the distribution characteristics of pathological cells on the effective full-slice image, and obtain the key pathological features of the optimized image.

[0092] The beneficial effects of the above technical solution are as follows: the present invention predicts the pathological results of the effective full-slice image by the identification target determination unit, and then controls the pre-trained image feature mining model to perform identification target determination according to the cell morphology corresponding to the prediction result, which is beneficial to improving the accuracy of key feature extraction, and performs global feature recognition of different degrees of precision on the optimized image based on the image label according to the identification target through the feature extraction unit, extracts the distribution characteristics of pathological cells on the effective full-slice image, obtains the pathological key features of the optimized image, and realizes the selection of processing strategies for different full-slice images, which is beneficial to improving the accuracy and efficiency of pathological key feature extraction.

[0093] Embodiment 5:

[0094] Based on Example 4, the deep learning feature extraction module, such as Figure 4 As shown, it also includes:

[0095] A feature verification unit is used to compare the pathological key features with the standard key features corresponding to the predicted pathological results to determine whether there is a difference between the two;

[0096] If it does not exist, the current predicted pathological result is judged to be normal;

[0097] Otherwise, based on the standard key features, a search is performed in the preset pathology database to determine similar pathologies, establish a confusing pathology dataset, and simultaneously obtain the key feature differences between the pathology key features and the standard key features;

[0098] Based on the differences in key features, pathology comparison is performed in easily confused pathology datasets to obtain the best matching result, and the predicted pathology result is calibrated based on the best matching result.

[0099] In this embodiment, the easily confused pathology data set refers to a data set containing machine pathology analysis contents of multiple diseases that are easily misjudged.

[0100] The beneficial effects of the above technical solution are as follows: the present invention compares the pathological key features with the standard key features corresponding to the predicted pathological results through the feature verification unit to determine whether there are differences between the two, completes the verification of the pathologist's pathological prediction results, and when it is determined that the predicted pathological results are incorrect, based on the standard key features, searches are performed in the preset pathology database to determine similar pathologies, establish a confusing pathology data set, and simultaneously obtain the key feature differences between the pathological key features and the standard key features. Based on the key feature differences, pathology comparison is performed in the confusing pathology data set to obtain the best matching result, and the predicted pathology result is calibrated based on the best matching result, thereby realizing intelligent calibration of the pathology prediction results and providing a reliable basis for the pathologist's final pathological judgment.

[0101] Embodiment 6:

[0102] Based on Example 4, the feature extraction unit includes:

[0103] A label recognition unit, used to recognize the image label corresponding to the optimization image;

[0104] A first feature extraction unit, configured to extract key pathological features of the optimized image based on the image detail enhancement result when the optimized image is a secondary optimized image;

[0105] A second feature extraction unit is used to perform grayscale classification on the pixel points based on the pixel grayscale values ​​corresponding to each pixel point of the optimized image when the optimized image is an important optimized image;

[0106] The optimized image is layered based on the grayscale grading results to obtain multiple layers, and pixel distribution features are extracted for each layer. The pixel distribution features of multiple layers corresponding to the same optimized image are fused to obtain the pixel distribution features of the entire layer. The pixel features of the entire layer are compared with the features of normal cells to obtain the key pathological features.

[0107] The beneficial effects of the above technical solution are as follows: the present invention identifies image labels (including primary optimized image labels and secondary optimized image labels) through a flag recognition unit, and performs non-strategic pathological key feature extraction on the whole slice image according to the recognition results. When the optimized image is a secondary optimized image, pathological key feature extraction is performed on the optimized image based on the image detail enhancement result, which is beneficial to shorten the time consumption of pathological key feature extraction and improve the recognition efficiency of typical visual field; when the optimized image is an important optimized image, the pixel points are grayscale graded based on the pixel grayscale values ​​corresponding to each pixel point of the optimized image; the optimized image is layered based on the grayscale grading result to obtain multiple layers, and pixel distribution feature extraction is performed on each layer, and the pixel distribution features of multiple layers corresponding to the same optimized image are fused and generated to obtain full layer pixel distribution features, and the full layer pixel features are compared with normal cell features to obtain pathological key features, so as to realize layered extraction of pathological key features of the whole slice image, so as to better obtain image details and improve the precision and accuracy of pathological key feature extraction.

[0108] Embodiment 7:

[0109] Based on Example 1, a typical visual field recognition module, such as Figure 5 As shown, including:

[0110] A key area marking unit is used to identify and mark the pathological key feature expression position of the optimized image through a graph neural network to determine the pathological key feature expression area;

[0111] An automatic marking calculation unit, used to automatically mark cells in the pathological key feature expression area, and calculate the concentration degree of the pathological key cells corresponding to the marked area according to the marking results;

[0112] The typical area screening unit is used to screen the pathological key feature expression area based on the concentration degree, determine the typical visual field expression area, and mark the typical visual field expression area based on the rectangular frame.

[0113] The beneficial effects of the above technical solution are as follows: the present invention first uses the key area marking unit to identify and mark the pathological key feature expression position of the optimized image through the graph neural network to determine the pathological key feature expression area; then the automatic marking calculation unit automatically marks the cells in the pathological key feature expression area, and calculates the concentration of the pathological key cells corresponding to the marked area according to the marking results, to ensure that the typical field of view can better express the cell pathological condition and improve the reliability of the typical field of view; finally, the typical area screening unit screens the pathological key feature expression area based on the concentration degree, determines the typical field of view expression area, and marks the typical field of view expression area based on the rectangular frame to realize the automatic recognition and precise positioning of the typical field of view of the whole slice image, effectively reducing the amount of film reading in the process of pathologists' pathological judgment, improving the efficiency of pathological judgment, and providing more reliable support for early diagnosis of the disease, prognosis evaluation and treatment effect monitoring.

[0114] Embodiment 8:

[0115] Based on Example 7, a typical visual field output module, such as Figure 6 As shown, including:

[0116] A visual field screenshot unit, used to screenshot a typical visual field performance area to obtain a typical visual field;

[0117] A target determination unit, used to search a preset person list based on the pathological judgment result corresponding to the typical visual field to determine the target person;

[0118] The result sending unit is used to obtain a typical visual field and send the typical visual field to a target person.

[0119] The beneficial effect of the above technical solution: after obtaining the typical visual field, the present invention can accurately send the typical visual field to the corresponding target person, which is convenient for further diagnosis and analysis by the pathologist.

[0120] Embodiment 9:

[0121] Based on Embodiment 7, the result sending unit includes:

[0122] The error processing subunit is used to obtain a pre-stored standard typical visual field corresponding to the original pathology prediction result when the pathology prediction result is a calibration result;

[0123] The infected cells in the typical field of view and the pre-stored standard typical field of view are marked separately, and a complete infected cell is randomly extracted from the typical field of view and the pre-stored standard typical field of view for focusing and magnification to obtain a set of infected cell comparison images;

[0124] Compare the lesion details based on the infected cell control image, determine the difference in cytopathic effect, and generate a confusion distinction report based on the difference in cytopathic effect and the infected cell control image;

[0125] A visual field processing subunit is used to highlight key features in a typical visual field based on differences in cytopathic changes, and to specially highlight differences in cytopathic changes, thereby generating a hidden typical visual field and adding it to a confusing distinction report;

[0126] The visual field sending subunit is used to send the confusion distinction report as an important attachment to the target person simultaneously with the typical visual field.

[0127] In this embodiment, hiding the typical visual field means that the special highlight mark will appear only when the target person clicks "display of difference characteristics". The special highlight mark refers to a highlight color mark that will appear only when the target person performs "display of difference characteristics".

[0128] The beneficial effects of the above technical solution are as follows: when the pathological prediction result is a calibration result, the present invention obtains the pre-stored standard typical visual field corresponding to the original pathological prediction result through the error processing subunit; the infected cells in the typical visual field and the pre-stored standard typical visual field are separately marked, and a complete infected cell is arbitrarily extracted from the typical visual field and the pre-stored standard typical visual field for focusing and magnification to obtain a set of infected cell comparison pictures; the lesion details are compared based on the infected cell comparison picture, the cell lesion difference is determined, and a confusion distinction report is generated based on the cell lesion difference and the infected cell comparison picture, which is conducive to assisting doctors to complete the accumulation of pathological judgment knowledge, and also provides a basis for pathologists to make the final diagnosis of the disease; then the key features in the typical visual field are highlighted based on the cell lesion difference by the visual field processing subunit, which is conducive to pathologists to quickly obtain pathological feature information in the typical visual field, determine the disease type, and specially highlight the cell lesion difference to generate a hidden typical visual field, which is added to the confusion distinction report, which can better assist pathologists to recognize the difference between the original pathological prediction result and the actual manager prediction result; finally, the visual field sending subunit sends the confusion distinction report as an important attachment to the target person at the same time as the typical visual field, realizing the rapid presentation of the typical visual field.

[0129] Embodiment 10:

[0130] Based on Example 7, the pathological image preprocessing module further includes:

[0131] A field of view selection optimization unit, used to obtain a screenshot position of a typical field of view in a corresponding full-slice image and a slice sample sampling position corresponding to the full-slice image, and generate a first pathology capture feature;

[0132] and simultaneously obtain the second pathology capture feature corresponding to the historical typical of the same type, and generate a typical visual field capture feature set by combining the first pathology capture feature and the second pathology capture feature;

[0133] Based on the lesion degree corresponding to the pathological prediction result, the pathological capture features in the typical visual field capture feature set are classified to obtain typical visual field capture feature subsets corresponding to multiple lesion degrees;

[0134] Typical visual field sampling features are extracted for each typical visual field capture feature subset respectively, typical visual field sampling features corresponding to current diseases with different lesion degrees are determined, and the original typical visual field sampling features are updated.

[0135] The beneficial effects of the above technical solution are as follows: the present invention generates a first pathology capture feature by obtaining the screenshot position of the typical field of view in the corresponding full-slice image and the slice sample sampling position corresponding to the full-slice image through the field of view selection optimization unit, and simultaneously obtains the second pathology capture feature corresponding to the same type of historical typical, and generates a typical field of view capture feature set with the first pathology capture feature and the second pathology capture feature; based on the degree of lesions corresponding to the pathological prediction result, the pathology capture features in the typical field of view capture feature set are classified to obtain typical field of view capture feature subsets corresponding to multiple degrees of lesions; typical field of view sampling features are extracted for each typical field of view capture feature subset respectively, the typical field of view sampling features corresponding to the current disease with different degrees of lesions are determined, and the original typical field of view sampling features are updated, thereby realizing automatic learning of the full-slice image prediction of the typical field of view of the current type of disease, which is conducive to adjusting and optimizing the accuracy of image classification, thereby improving the accuracy of typical field of view capture.

[0136] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A pathological typical visual field capture system based on artificial intelligence, characterized in that: include: A slice data reading module is used to receive a full slice image and perform image data cleaning on the full slice image to obtain a valid full slice image; Pathological image preprocessing module, used to optimize the effective full-slice image to obtain the optimized image; Deep learning feature extraction module, which is used to extract key features of case pathology in optimized images based on pre-trained image feature mining models; Typical field of view recognition module, used to determine the typical field of view of optimized images based on key pathological features through graph neural network; The typical field of view output module is used to take a screenshot of the typical field of view and send the screenshot result directly to the target person.

2. The artificial intelligence-based pathological typical visual field capture system according to claim 1, characterized in that: Slice data reading module, including: An image data recognition unit is used to obtain a grayscale image of the whole slice image and perform edge feature extraction on the grayscale image, and determine the number of intact tissue cells in the whole slice image based on the edge feature extraction result; An image data cleaning unit, used to treat the whole slice image as an invalid image as a valid whole slice image when the number of intact tissue cells is greater than a preset value; Otherwise, the whole slice image is regarded as an invalid image and is treated as an invalid whole slice image, and a whole slice image re-reading instruction is sent to the pathology database.

3. The artificial intelligence-based pathological typical visual field capture system according to claim 1, characterized in that: Pathological image preprocessing module, including: A preliminary processing unit, used for inputting the effective full-slice image into a preset denoising model for denoising to obtain a denoised image; An enhanced processing unit is used to obtain standard key features corresponding to the predicted pathological results of the effective full-slice image, and determine the morphological features corresponding to the diseased tissue cells and the cell-tissue connection relationship between the diseased tissue cells based on the standard key features; According to the morphological features corresponding to the diseased tissue cells and the cell-tissue connection relationship between the diseased tissue cells, image details of the tissue cells in the denoised image are enhanced to obtain an optimized image; The image screening unit is used to classify the optimized images based on typical field of view sampling features to obtain key optimized images and secondary optimized images.

4. The artificial intelligence-based pathological typical visual field capture system according to claim 1, characterized in that: Deep learning feature extraction module, including: A recognition target determination unit is used to obtain the predicted pathological results of the effective full-slice image, and based on the predicted pathological results, control the pre-trained image feature mining model to perform recognition target determination; The feature extraction unit is used to perform global feature recognition of different degrees of precision on the optimized image based on the image label according to the recognition target, extract the distribution characteristics of pathological cells on the effective full-slice image, and obtain the key pathological features of the optimized image.

5. The artificial intelligence-based pathological typical visual field capture system according to claim 4, characterized in that: The deep learning feature extraction module also includes: A feature verification unit is used to compare the pathological key features with the standard key features corresponding to the predicted pathological results to determine whether there is a difference between the two; If it does not exist, the current predicted pathological result is judged to be normal; Otherwise, based on the standard key features, a search is performed in the preset pathology database to determine similar pathologies, establish a confusing pathology dataset, and simultaneously obtain the key feature differences between the pathology key features and the standard key features; Based on the differences in key features, pathology comparison is performed in easily confused pathology datasets to obtain the best matching result, and the predicted pathology result is calibrated based on the best matching result.

6. The artificial intelligence-based pathological typical visual field capture system according to claim 4, characterized in that: Feature extraction unit, including: A label recognition unit, used to recognize the image label corresponding to the optimization image; A first feature extraction unit, configured to extract key pathological features of the optimized image based on the image detail enhancement result when the optimized image is a secondary optimized image; A second feature extraction unit is used to perform grayscale classification on the pixel points based on the pixel grayscale values ​​corresponding to each pixel point of the optimized image when the optimized image is an important optimized image; The optimized image is layered based on the grayscale grading results to obtain multiple layers, and pixel distribution features are extracted for each layer. The pixel distribution features of multiple layers corresponding to the same optimized image are fused to obtain the pixel distribution features of the entire layer. The pixel features of the entire layer are compared with the features of normal cells to obtain the key pathological features.

7. The artificial intelligence-based pathological typical visual field capture system according to claim 1, characterized in that: Typical vision recognition modules include: A key area marking unit is used to identify and mark the pathological key feature expression position of the optimized image through a graph neural network to determine the pathological key feature expression area; An automatic marking calculation unit, used to automatically mark cells in the pathological key feature expression area, and calculate the concentration degree of the pathological key cells corresponding to the marked area according to the marking results; The typical area screening unit is used to screen the pathological key feature expression area based on the concentration degree, determine the typical visual field expression area, and mark the typical visual field expression area based on the rectangular frame.

8. The artificial intelligence-based pathological typical visual field capture system according to claim 7, characterized in that: Typical field of view output modules include: A visual field screenshot unit, used to screenshot a typical visual field performance area to obtain a typical visual field; A target determination unit, used to search a preset person list based on the pathological judgment result corresponding to the typical visual field to determine the target person; The result sending unit is used to obtain a typical visual field and send the typical visual field to a target person.

9. The artificial intelligence-based pathological typical visual field capture system according to claim 7, characterized in that: The result sending unit comprises: The error processing subunit is used to obtain a pre-stored standard typical visual field corresponding to the original pathology prediction result when the pathology prediction result is a calibration result; The infected cells in the typical field of view and the pre-stored standard typical field of view are marked separately, and a complete infected cell is randomly extracted from the typical field of view and the pre-stored standard typical field of view for focusing and magnification to obtain a set of infected cell comparison images; Compare the lesion details based on the infected cell control image, determine the difference in cytopathic effect, and generate a confusion distinction report based on the difference in cytopathic effect and the infected cell control image; A visual field processing subunit is used to highlight key features in a typical visual field based on differences in cytopathic changes, and to specially highlight differences in cytopathic changes, thereby generating a hidden typical visual field and adding it to a confusing distinction report; The visual field sending subunit is used to send the confusion distinction report as an important attachment to the target person simultaneously with the typical visual field.

10. The artificial intelligence-based pathological typical visual field capture system according to claim 9, characterized in that: The pathological image preprocessing module also includes: A field of view selection optimization unit, used to obtain a screenshot position of a typical field of view in a corresponding full-slice image and a slice sample sampling position corresponding to the full-slice image, and generate a first pathology capture feature; and simultaneously obtain the second pathology capture feature corresponding to the historical typical of the same type, and generate a typical visual field capture feature set by combining the first pathology capture feature and the second pathology capture feature; Based on the lesion degree corresponding to the pathological prediction result, the pathological capture features in the typical visual field capture feature set are classified to obtain typical visual field capture feature subsets corresponding to multiple lesion degrees; Typical visual field sampling features are extracted for each typical visual field capture feature subset respectively, typical visual field sampling features corresponding to current diseases with different lesion degrees are determined, and the original typical visual field sampling features are updated.

Images