WO2021082433A1 - 一种数字病理图像质控的方法及装置 - Google Patents

一种数字病理图像质控的方法及装置 Download PDF

Info

Publication number
WO2021082433A1
WO2021082433A1 PCT/CN2020/093579 CN2020093579W WO2021082433A1 WO 2021082433 A1 WO2021082433 A1 WO 2021082433A1 CN 2020093579 W CN2020093579 W CN 2020093579W WO 2021082433 A1 WO2021082433 A1 WO 2021082433A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
pixel
processed
quality control
abnormal
Prior art date
Application number
PCT/CN2020/093579
Other languages
English (en)
French (fr)
Inventor
王佳平
南洋
李风仪
侯晓帅
谢春梅
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021082433A1 publication Critical patent/WO2021082433A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30168—Image quality inspection
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Definitions

  • This application relates to the field of image processing, and in particular to a method and device for quality control of digital pathological images.
  • Pathology quality control is the transformation of the pathology department from the traditional experience management model to the scientific management model. It is a key link to ensure high-quality service, high-quality medical treatment, high efficiency and low consumption. Pathological quality control through pathological slices has also become an important way. Among them, pathological slices are the final product of the pathological process. It tests the standardization of the process and the technician's film production skills. The technicians can understand the occurrence of the disease through the pathological slices. The development process, so as to make a diagnosis and even treatment of the disease. It can be said that high-quality pathological slices are the essential foundation and guarantee for correct pathological diagnosis. Improving the quality of pathological slices can help improve diagnostic efficiency and ensure diagnostic quality.
  • the present application provides a method and device for quality control of digital pathology images, in order to improve the quality management of digital pathology images, and realize the quantification and unification of digital pathology image quality control standards.
  • the first aspect of the embodiments of the present application provides a digital pathological image quality control method, which includes: obtaining the image to be processed of a specified layer of the digital pathological image according to the image format of the digital pathological image; Input the U-Net segmentation model to obtain a first abnormal probability matrix corresponding to each pixel in the image to be processed.
  • the first abnormal probability matrix includes a plurality of correlation values, and the plurality of correlation values represent the first abnormality probability matrix.
  • Each quality control label includes non-abnormal labels and N types of abnormal labels.
  • the U-Net segmentation model is trained based on image samples, so
  • the image sample includes at least a plurality of digital pathological image samples and the quality control label feature of each pixel in the digital pathological image sample, where N is a positive integer; according to each pixel in the image to be processed and each pixel
  • the relevant value of the quality control label determines the first quality control label of each pixel in the image to be processed; according to the first quality control label of each pixel in the image to be processed, the statistics in the image to be processed
  • the number of pixels corresponding to the N types of abnormal tags; the digital pathological image is classified according to the number of pixels corresponding to the N types of abnormal tags in the image to be processed.
  • the second aspect of the embodiments of the present application provides an apparatus for quality control of digital pathology images, including: an acquisition module for obtaining images to be processed in a designated layer of the digital pathology image according to the image format of the digital pathology image; and a learning module, It is used to input the image to be processed into the U-Net segmentation model to obtain a first abnormal probability matrix corresponding to each pixel in the image to be processed, and the first abnormal probability matrix includes a plurality of related values, so The multiple correlation values represent the correlation between the pixels corresponding to the first abnormal probability matrix and each quality control label.
  • Each quality control label includes a non-abnormal label and N types of abnormal labels.
  • the U-Net segmentation model is Based on image sample training, the image sample includes at least a plurality of digital pathological image samples and the quality control label feature of each pixel in the digital pathological image sample, N is a positive integer; the determination module is used to determine The correlation value of each pixel in the image to be processed and the quality control label determines the first quality control label of each pixel in the image to be processed; the statistics module is used to determine the first quality control label of each pixel in the image to be processed; The first quality control label of each pixel counts the number of pixels corresponding to the N types of abnormal tags in the image to be processed; the classification module is used to calculate the number of pixels corresponding to the N types of abnormal tags in the image to be processed The number of pixels classifies the digital pathological image.
  • the third aspect of the embodiments of the present application provides an electronic device, which includes a processor, a memory, and an input-output interface; the processor is respectively connected to the memory and the input-output interface, wherein the input-output interface Used for data interaction with a user, the memory is used for storing program code, and the processor is used for calling the program code to execute the digital pathology image quality control method as described in the first aspect of the embodiments of the present application.
  • the fourth aspect of the embodiments of the present application provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, the computer program includes program instructions, and the program instructions, when executed by a processor, execute as The digital pathological image quality control method described in the first aspect of the embodiments of the present application.
  • the embodiments of the application can realize the quantification of the quality control standards for digital pathology images, and perform unified storage through the pathology image management platform, making the management of digital pathology images more unified and efficient, and obtain unified quality control standards for digital pathology images, thereby improving the Digital pathology image management and quality control efficiency.
  • FIG. 1 is a schematic flowchart of a method for quality control of digital pathology images provided by an embodiment of the present application.
  • Fig. 2 is a schematic diagram of the output of a U-Net segmentation model provided by an embodiment of the present application.
  • Fig. 3 is a schematic diagram of output of an abnormal classifier provided by an embodiment of the present application.
  • Fig. 4 is a digital pathological image quality control device provided by an embodiment of the present application.
  • Fig. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • FIG. 1 is a schematic flowchart of a method for quality control of digital pathology images provided by an embodiment of the present application. As shown in Figure 1, the above method includes the following steps.
  • step S101 an image to be processed of a designated layer of the digital pathological image is obtained according to the image format of the digital pathological image.
  • the pathology image management platform After the pathology image management platform receives the digital pathology image uploaded by the user, it acquires the image format of the digital pathology image, determines the image level of the extracted specified layer according to the image format, and obtains the specified layer of the digital pathology image The image to be processed.
  • the pathology image management platform can determine the user's authority by logging in user accounts such as administrators and ordinary users.
  • the administrator can be the hospital headquarters, and the ordinary user can be a subordinate hospital of the hospital headquarters.
  • the subordinate hospital uses the scan Scan the pathological slice to obtain the digital pathological image of the pathological slice, and upload the digital pathological image to the pathological image management platform.
  • the pathological image management platform receives the digital pathological image, it obtains the image format of the digital pathological image. Format to get the image to be processed of the specified layer of the digital pathology image.
  • the digital pathology image has different image formats, such as .svs, .kfb, .ndpi, and .tif.
  • image formats such as .svs, .kfb, .ndpi, and .tif.
  • image levels that need to be extracted are different.
  • the image to be processed in the specified layer of the digital pathological image is extracted according to the preset image level corresponding to the image format.
  • the image format of the digital pathological image obtained by scanning the pathological slice through the scanner is different, and the layers composing the digital pathological image are also different.
  • the digital pathology image is composed of M layers, and the resolution is gradually reduced from the 0th layer, that is, the resolution of the 0th layer is the largest, and M is a positive integer, and the M of the digital pathology image of different image formats is different.
  • the digital pathology image in .ndpi format is obtained
  • the digital pathology image in .ndpi format is composed of ten layers, and the resolution of the layers from the 0th layer to the 9th layer is reduced, and M is 10 .
  • the preset image level corresponding to the digital pathological image in the .ndpi format is 4, the image to be processed at the fourth level of the digital pathological image is extracted, where the designated level here is the fourth level.
  • the image to be processed of the designated layer of the digital pathological image can be extracted by an image processing method, and the image processing method can be openslide, or the digital pathological image can be sampled directly by the resolution.
  • Step S102 Input the image to be processed into the U-Net segmentation model to obtain the first abnormal probability matrix of each pixel in the image to be processed.
  • the image to be processed is input into the U-Net segmentation model to obtain a first abnormal probability matrix corresponding to each pixel in the image to be processed.
  • the first abnormal probability matrix includes a plurality of correlation values. The value represents the correlation between the pixel corresponding to the first abnormal probability matrix and each quality control label.
  • Each quality control label includes non-abnormal labels and N types of abnormal labels.
  • the U-Net segmentation model is based on image sample training. Yes, the image sample includes at least multiple digital pathology image samples and the quality control label feature of each pixel in each digital pathology image sample, and N is a positive integer.
  • the image to be processed includes 20*20 pixels
  • 400 first abnormal probability matrices are obtained, and the first pixel corresponds to the first
  • the abnormal probability matrix contains (N+1) correlation values, and (N+1) correlation values are the correlations between the first pixel and each quality control label, where each pixel in the image to be processed corresponds to one
  • the first anomaly probability matrix that is, 20*20 first anomaly probability matrices are obtained in this example.
  • the image to be processed is input into the U-Net segmentation model to obtain the first abnormal probability matrix of each pixel in the image to be processed, based on the first abnormal probability matrix of each pixel in the image to be processed with the largest correlation value
  • the quality control label determines the abnormal display image. Specifically, if the quality control label with the largest correlation value with the pixel in the image to be processed is a non-abnormal label, then the abnormal display image is the same as the pixel in the image to be processed.
  • the gray value of the pixel is set to the preset background value.
  • the abnormal display image will be compared with the The gray value of pixels with the same pixel position in the processed image is set as a preset abnormal value, and the abnormal display image is determined according to each pixel in the set abnormal display image.
  • the preset background value may be 0, the preset abnormal value may be 255, or the preset background value is 255 and the preset abnormal value is 0.
  • the gray value of the preset background value and the preset abnormal value may also be 0, 1, or 1, 0, respectively, which are preset according to the gray value setting of the output image of the U-Net segmentation model.
  • the U-Net segmentation model is a fully convolutional neural network, which is an end-to-end network, that is, both input and output are images.
  • Input the image to be processed into the U-Net segmentation model use the activation function to down-sample the image to be processed after passing through the convolutional layer in the contraction path, extract the feature map of the image to be processed, and then perform up-sampling in the expansion path.
  • the corresponding feature map obtained in the contraction path is added each time upsampling, and finally the first abnormal probability matrix of each pixel in the image to be processed is extracted.
  • the U-Net segmentation model is a u-shaped structure, and it can be considered that the corresponding feature map in the contraction path added during upsampling is obtained during downsampling at the same level in the u-shaped structure.
  • the first anomaly probability matrix is an (N+1)-dimensional vector, that is, it includes (N+1) correlation values, which can be considered to include 512*512 (N+1)-dimensional vectors, and each ( The quality control label with the largest correlation value in the N+1) dimensional vector is determined as the quality control label of the pixel corresponding to the (N+1) dimensional vector.
  • the quality control label is a non-abnormal label Or any one of the N types of abnormal tags, binarize the gray value of the pixel, so as to realize the binarization of the image to be processed, and obtain the abnormal display image.
  • N types of abnormal labels include wrinkles, folds, knife marks, and bubbles.
  • N is 4, 0 means non-abnormal labels, 1 means wrinkles, 2 means folds, and 3 means abnormal labels.
  • the abnormal label of the knife mark and the abnormal label of 4 indicate the bubble.
  • the first abnormal probability matrix of each pixel is obtained, and the first abnormal probability matrix includes the corresponding pixel. Respectively and the relevant value of each quality control label.
  • the first probability matrix is [0.8, 0.1, 0, 0.05, 0.05].
  • the first probability matrix it can be known that the pixel at (0, 0) and the quality control label 0
  • the correlation value is the largest, and the quality control label 0 is a non-abnormal label
  • the gray value of the pixel at (0,0) in the abnormal display image is set to 255.
  • the pixel corresponding to the non-abnormal label is set to white
  • Set the pixel corresponding to the abnormal label to black where the black gray value is 0 and the white gray value is 255, until the gray value of each pixel in the abnormal display image is determined, and the abnormal display image is obtained.
  • FIG. 2 is a schematic diagram of the output of a U-Net segmentation model provided by an embodiment of the present application.
  • each abnormal label image sample is obtained by adding quality control label features to the corresponding digital pathology image sample, that is, includes the quality control label feature of each pixel in the corresponding digital pathology image sample.
  • the quality control label with the largest correlation value is the quality control label feature of the corresponding pixel in the abnormal label image sample, so as to obtain the trained U-Net segmentation model.
  • Step S103 Determine the first quality control label of each pixel in the image to be processed according to the correlation value of each pixel in the image to be processed and each quality control label.
  • the quality control label with the largest correlation value in the first abnormal probability matrix of each pixel in the image to be processed is determined as the first quality control label of each pixel in the image to be processed.
  • the quality control label with the largest correlation value is determined when it is determined in step S102 that each pixel is a non-abnormal label or any abnormal label in the N types of abnormal labels, and each pixel in the abnormal display image is obtained in step S102 At the same time as the gray value of the pixel, determine the first quality control label of the pixel.
  • Step S104 according to the first quality control label of each pixel in the image to be processed, count the number of pixels corresponding to the N types of abnormal labels in the image to be processed, and classify the digital pathology image.
  • the transfer weight can be determined according to the degree of influence of the abnormality corresponding to the abnormal label on the digital pathological image.
  • a pathology quality control standard can be set.
  • the pathology quality control standard is uniformly determined by pathology quality control experts, and can be regarded as a unified standard for classifying digital pathology images.
  • the pathology quality control standard can be preset to L, for example, when L is 100, it can be considered that the total quality control score of the digital pathology image is 100 points, and the number of pixels corresponding to the N types of abnormal labels in the image to be processed is calculated Obtain the data transfer volume of the image to be processed, and determine the quality control category of the digital pathology image corresponding to the image to be processed by calculating the difference between 100 and the data transfer volume; or, if L is 3, the digital pathology can be considered
  • the image includes three types of "excellent, medium, and poor".
  • the data transfer volume of the image to be processed is obtained according to the number of pixels corresponding to the N types of abnormal tags in the image to be processed, and the corresponding quality control category is determined according to the data transfer volume, for example, When the amount of data transfer is 0 ⁇ 3, it corresponds to “excellent”, when the amount of data transfer is 3 to 10, it corresponds to “medium”, and when the amount of data transfer is greater than 10, it corresponds to “poor”, etc.
  • the value of L can be set according to requirements .
  • the data transfer volume interval corresponding to the number of pixels of each abnormal tag in the image to be processed can be obtained to determine the data transfer volume of the image to be processed, such as when When the abnormal label "knife mark" has pixels 0 to 3, the corresponding data transfer amount is 0, etc., and the data transfer amount corresponding to different abnormal labels is summed or weighted summation or preset summation formula summation to obtain the image to be processed
  • the amount of data transfer or, directly based on the summation or weighted summation of the number of pixels of different abnormal tags to obtain the amount of data transfer of the image to be processed; or, determine the amount of data transfer based on the number of pixels of different abnormal tags in the percentage of the image to be processed The amount of data transfer of the image to be processed.
  • the gray value of pixels whose first quality control label is a non-abnormal label in the image to be processed can be set as a preset background value, and the gray values of pixels whose first quality control label is an abnormal label remain unchanged ,
  • the first image is obtained, and the first image is input to the anomaly classifier to obtain a second anomaly probability matrix corresponding to each pixel in the first image.
  • the second anomaly probability matrix includes a plurality of correlation values.
  • a correlation value represents the correlation between the corresponding pixel of the second abnormal probability matrix in the first image and each quality control label, and the abnormal classifier is based on each pixel in the multiple first image samples and digital pathology image samples
  • the quality control label feature training of each first image sample is the corresponding digital pathology image sample based on the correlation value of each pixel and each quality control label, and the maximum correlation value is the gray value of the pixel of the non-abnormal label Set to the preset background value; according to the second abnormal probability matrix of each pixel in the first image, determine the second quality control label of each pixel in the first image, and the second quality control label is the first The quality control label with the largest correlation value in the second abnormal probability matrix of each pixel in the image; according to the second quality control label of each pixel in the first image, determine the pixels corresponding to the N types of abnormal labels in the image to be processed The number of points.
  • the U-Net segmentation model and the anomaly classifier are both an end-to-end convolutional neural network.
  • the convolutional neural network When the convolutional neural network is training or predicting, it will essentially combine the The prediction environment around the pixel or pixel block processes the pixel or pixel block.
  • the gray value of the pixel corresponding to the non-abnormal label in the image to be processed is set as the preset background value
  • various abnormal labels The difference between the gray value of the corresponding pixel and the gray value of the pixel corresponding to the non-abnormal label will be large, so that when a single pixel corresponding to an abnormal label appears, the abnormal
  • the gray value of a pixel has a large difference with the gray value of adjacent pixels, and in general, there is no abnormal situation of a single pixel, so in the prediction, it will be provided by the adjacent pixels of the single pixel.
  • the information including environmental field information and detailed information, etc., filter the abnormal label of the single pixel.
  • the anomaly classifier extracts the characteristic information of the single pixel through up-sampling, and displays the environmental information of the single pixel through down-sampling.
  • the environmental information of the single pixel is provided by the information of adjacent pixels. Combining the feature information obtained by the up-sampling and the environmental information obtained by the down-sampling to determine whether the gray value and abnormal label of the single pixel are reasonable, and realize the filtering of the abnormal label of the single pixel by neighboring pixels.
  • the single pixel is obtained
  • the abnormal probability matrix includes the correlation value of the single pixel and each quality control label, and according to the pixel information of the neighboring pixels of the single pixel, it includes the gray value and the abnormal probability of the neighboring pixels
  • Matrix to obtain the environmental information of the single pixel the environmental information is used to indicate that the neighboring pixels of the single pixel are all non-abnormal pixels, and the abnormal probability matrix of the single pixel is adjusted through the environmental information, The correlation value between the single pixel and the non-abnormal label is increased, so as to realize the filtering of the abnormal label of the single pixel.
  • the abnormal classifier is trained on the basis of the U-Net segmentation model training, specifically after the U-Net segmentation model is trained, according to the U-Net segmentation model
  • the output result updates the digital pathological image sample to obtain the first image sample, and trains the abnormal classifier according to the first image sample and the quality control label feature of each pixel in the digital pathological image sample.
  • the image size of the image to be processed may be obtained, if the image size of the image to be processed is larger than the preset sliding window size ,
  • the sliding window of the image to be processed is divided into multiple second images by the preset sliding window size, the image size of the second image is not greater than the preset sliding window size, and the preset sliding window size is the number trained by the U-Net segmentation model
  • the size of the pathological image sample or the preset sliding window size set according to the requirements; input any second image of the multiple second images into the U-Net segmentation model to obtain the first abnormality of each pixel in any second image Probability matrix.
  • the position information of each second image in the plurality of second images is recorded, and the position information is the relative position of the corresponding second image in the image to be processed.
  • the relative position can be the pixel position of the first pixel in the upper left corner of each second image in the image to be processed, or the relative position between different second images. For example, (0, 0) means that the corresponding second image is in the image to be processed. Process the first row and first column of the image.
  • step S103 based on the first abnormal probability matrix of each pixel in any second image, determine the first quality control label of each pixel in any second image, based on each of the multiple second images The position information of the second image and the first quality control label of each pixel in any second image determine the first quality control label of each pixel in the image to be processed.
  • the digital pathological image, the abnormal display image, and the digital pathological image corresponding to the quality control label of each pixel in the image to be processed can be displayed to the user, so that the user can intuitively obtain the digital pathological image Quality situation.
  • the abnormal display image is shown as 201 in FIG. 2, and the quality control label of each pixel in the image to be processed may be as shown in the first quality control label image 202 in FIG. 2.
  • FIG. 3 is a schematic diagram of the output of an abnormality classifier provided in an embodiment of the present application.
  • the embodiments of the application realize the quantification of the quality control standards for digital pathology images, and perform unified storage through the pathology image management platform, so that the management of digital pathology images is more unified and efficient, and the unified quality control standards for digital pathology images are obtained, thereby improving the quality control standards for digital pathology images.
  • the efficiency of pathological image management and quality control it is possible to perform secondary abnormal label extraction on the image to be processed to mark the abnormal area as reasonably as possible. For abnormalities, it is almost impossible to have one or two pixel abnormalities. Through secondary abnormal label extraction, It can reduce the occurrence of mislabeling and improve the accuracy of quality control.
  • each pixel is processed so that the image size of the input and output images of the model are the same, so as to improve the accuracy of the image.
  • FIG. 4 is a digital pathological image quality control device provided by an embodiment of the present application.
  • the digital pathological image quality control device can be used for the electronic equipment in the embodiment corresponding to FIG. 1.
  • the digital pathological image quality control device 40 includes an acquisition module 401, a learning module 402, and a determination Module 403, statistics module 404 and classification module 405.
  • the acquiring module 401 is used to obtain the to-be-processed image of the designated layer of the digital pathological image according to the image format of the digital pathological image;
  • the learning module 402 is to input the to-be-processed image into the U-Net segmentation model to obtain the to-be-processed image Process a first abnormal probability matrix corresponding to each pixel in the image, where the first abnormal probability matrix includes a plurality of correlation values, and the plurality of correlation values represent that the pixel corresponding to the first abnormal probability matrix is associated with each Correlation of quality control labels, each of the quality control labels includes non-abnormal labels and N types of abnormal labels, the U-Net segmentation model is trained based on image samples, and the image samples include at least multiple digital pathological images
  • the quality control label feature of each pixel in the sample and the digital pathological image sample, N is a positive integer;
  • the determination module 403 is used to determine the correlation between each pixel in the image to be processed and each quality control label Value to determine the first quality control
  • the statistics module 404 includes: a setting unit 4041, configured to set the gray value of the pixel with the non-abnormal label as the first quality control label in the image to be processed as a preset background value to obtain the first Image; learning unit 4042, used to input the first image into the anomaly classifier to obtain a second anomaly probability matrix corresponding to each pixel in the first image, and the second anomaly probability matrix includes a plurality of Correlation values, the multiple correlation values characterizing the correlation between the corresponding pixels of the second abnormal probability matrix in the first image and the respective quality control labels, and the abnormal classifier is based on the first image sample And the quality control label feature training of each pixel in the digital pathological image sample, the first image sample is the digital pathological image sample based on the correlation value of each pixel and each quality control label, The gray value of the pixel with the maximum correlation value of the non-abnormal label is set to the preset background value; the determining unit 4043 is used to determine the second abnormality of each pixel in the first image
  • the device further includes: the obtaining module 401, which is also used to obtain the image size of the image to be processed; the sliding window module 406, which is used to if the image size of the image to be processed is larger than the preset sliding window size, Then the sliding window of the image to be processed is divided into a plurality of second images by the preset sliding window size, the image size of the second image is not larger than the preset sliding window size, and the preset sliding window size is all The size of the digital pathological image sample trained by the U-Net segmentation model; the learning module 402 is specifically configured to: input any second image among the plurality of second images into the U-Net segmentation model, Obtain the first abnormal probability matrix of each pixel in any of the second images.
  • the obtaining module 401 which is also used to obtain the image size of the image to be processed
  • the sliding window module 406 which is used to if the image size of the image to be processed is larger than the preset sliding window size, Then the sliding window of the image to be processed is divided into a
  • the device further includes: a recording module 407, configured to record the position information of each second image in the plurality of second images, where the position information is the relative position of the corresponding second image in the image to be processed.
  • Position the determining module 403 is specifically configured to: determine the first quality control of each pixel in any second image according to the first abnormal probability matrix of each pixel in the any second image Label; based on the position information of each second image in the plurality of second images and the first quality control label of each pixel in any of the second images, determine each pixel in the image to be processed The first quality control label.
  • the classification module 405 is specifically configured to: according to the number of pixels corresponding to each of the N types of abnormal tags in the to-be-processed image, and based on the transfer weight of the abnormal tags, divide the The number of pixels corresponding to the abnormal label is weighted and summed to determine the data transfer amount of the digital pathology image, and the digital pathology image is classified according to the data transfer amount.
  • the determining module 403 is specifically configured to: determine the quality control label with the largest correlation value in the first abnormal probability matrix of each pixel in the image to be processed as the value of each pixel in the image to be processed The first quality control label.
  • the learning module 402 is specifically configured to: input the to-be-processed image into the U-Net segmentation model to obtain the first abnormal probability matrix of each pixel in the to-be-processed image, based on the The quality control label with the largest correlation value in the first abnormal probability matrix of each pixel determines the abnormal display image.
  • the learning module 402 is specifically configured to: if the quality control label with the largest correlation value with the pixel in the image to be processed is the non-abnormal label, display the abnormal image
  • the gray value of the pixel with the same position as the pixel in the image to be processed is set as the preset background value; if the quality control label with the largest correlation value with the pixel in the image to be processed is the N types of abnormalities If any abnormal label in the label, set the gray value of the pixel in the abnormal display image that is the same as the pixel in the image to be processed as the preset abnormal value; display each of the images in accordance with the set abnormality The pixels determine the abnormal display image.
  • the device further includes: an output module 408 for outputting the classification result of the digital pathology image and the abnormal display image.
  • the above-mentioned device can execute the implementation provided by each step in the above-mentioned implementation provided in FIG. 1 through the above-mentioned modules to realize the functions implemented in the above-mentioned embodiments.
  • the above-mentioned method shown in FIG. 1 The corresponding description provided in each step in the embodiment will not be repeated here.
  • the embodiment of the present application provides a digital pathological image quality control device.
  • the above-mentioned device quantifies the digital pathological image quality control standards and stores them uniformly through a pathological image management platform, so that the management of digital pathological images is more unified and efficient, and Unified quality control standards for digital pathology images, thereby improving the management and quality control efficiency of digital pathology images.
  • FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • the electronic device in this embodiment may include: one or more processors 501, a memory 502, and a transceiver 503.
  • the aforementioned processor 501, memory 502, and transceiver 503 are connected via a bus 504.
  • the memory 502 is used to store a computer program, the computer program includes program instructions, the transceiver 503 is used to connect to a terminal device and perform data interaction with the above-mentioned electronic device; the processor 501 is used to execute the program instructions stored in the memory 502, and perform the following operations.
  • the image to be processed of the designated layer of the digital pathology image according to the image format of the digital pathology image; input the image to be processed into the U-Net segmentation model to obtain the first abnormal probability of each pixel in the image to be processed Matrix, the first abnormal probability matrix includes corresponding pixel points and the correlation value of each quality control label, each quality control label includes non-abnormal labels and N types of abnormal labels, the U-Net segmentation model is based on image samples According to training, the image sample includes at least a plurality of digital pathology image samples and the quality control label feature of each pixel in the digital pathology image sample, N is a positive integer; according to each pixel in the image to be processed Point and the correlation value of each quality control label, determine the first quality control label of each pixel in the image to be processed; according to the first quality control label of each pixel in the image to be processed, statistics The number of pixels corresponding to the N types of abnormal tags in the image to be processed; the digital pathological image is classified according to the number of pixels corresponding to the
  • the above-mentioned processor 501 may be a central processing unit (central processing unit, CPU), and the processor may also be other general-purpose processors, digital signal processors (digital signal processors, DSPs), and dedicated integrated processing units. Circuit (application specific integrated circuit, ASIC), ready-made programmable gate array (field-programmable gate array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.
  • the memory 502 may include a read-only memory and a random access memory, and provides instructions and data to the processor 501 and the transceiver 503. A part of the memory 502 may also include a non-volatile random access memory. For example, the memory 502 may also store device type information.
  • the above-mentioned electronic device can execute the implementation manner provided by each step in FIG. 1 through its built-in functional modules.
  • the implementation manner provided by each step in FIG. 1 above which will not be repeated here.
  • the embodiment of the present application provides an electronic device including a processor, a transceiver, and a memory.
  • the processor obtains computer instructions in the memory and executes each step of the method shown in FIG. 1 to achieve quality control of digital pathology images.
  • Standard quantification and unified storage through the pathology image management platform makes the management of digital pathology images more unified and efficient, and obtains unified quality control standards for digital pathology images, thereby improving the management and quality control efficiency of digital pathology images.
  • the embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and the computer program includes program instructions that, when executed by a processor, realize the numbers provided in each step in FIG. 1
  • the computer-readable storage medium may be a non-volatile readable storage medium or a volatile readable storage medium.
  • the foregoing computer-readable storage medium may be the digital pathology image quality control apparatus provided in any of the foregoing embodiments or the internal storage unit of the foregoing terminal device, such as a hard disk or memory of an electronic device.
  • the computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), or a secure digital (SD) card equipped on the electronic device. Flash card, etc.
  • the computer-readable storage medium may also include both an internal storage unit of the electronic device and an external storage device.
  • the computer-readable storage medium is used to store the computer program and other programs and data required by the electronic device.
  • the computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Investigating Or Analysing Biological Materials (AREA)
  • Image Processing (AREA)

Abstract

本申请实施例适用于人工智能领域中的图像检测，公开了一种数字病理图像质控的方法，包括：根据数字病理图像的图像格式得到该数字病理图像的指定层的待处理图像；将待处理图像输入U-Net分割模型，得到待处理图像中与每个像素点分别对应的第一异常概率矩阵，第一异常概率矩阵包括多个相关值；根据待处理图像中每个像素点与各个质控标签的相关值，确定待处理图像中每个像素点的第一质控标签；根据待处理图像中每个像素点的第一质控标签，统计待处理图像中N种异常标签对应的像素点数量，对数字病理图像进行分类。本申请还相应的提出了一种数字病理图像质控的装置。采用本申请，可以实现对数字病理图像质控标准的量化统一，提高质控效率。

Description

一种数字病理图像质控的方法及装置
本申请要求于2019年10月29日提交中国专利局、申请号为201911034430.4，发明名称为“一种数字病理图像质控的方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像处理领域，尤其涉及一种数字病理图像质控的方法及装置。
背景技术
病理质控是病理科由传统的经验管理模式向科学管理模式转变，是确保优质服务、优质医疗、高效低耗的关键环节。通过病理切片进行病理质控也已经成为一种重要的方式，其中，病理切片是病理流程的最终产品，它检验了流程的规范标准化和技术员的制片技能，技术员可以通过病理切片了解病变的发生发展过程，从而对病变做出诊断乃至治疗。可以说，高质量的病理切片是进行正确的病理诊断时至关重要的基础和保证，提高病理切片的质量可以有助于提高诊断效率，确保诊断质量。
目前已有一些质控平台可以满足日常质控中心对下级医院的远程监控，比如可以监控每百张病床病理医师数和每百张病床病理技术人员数等指标，但无法得到下级医院的病理全切片对应的数字病理图像的分析，这些病理切片需要定期派遣专家到下级医院进行抽查。发明人意识到由于下级医院众多，加上专家人数有限，而且得到病理切片及其对于通过病理切片生成的数字病理图像进行分析非常耗费时间和精力，再加上不同专家对于劣片的判别标准不尽相同，使得病理质控管理缺少统一的标准，增加病理质控管理的难度。
发明概述
技术问题
基于此，本申请提供了一种数字病理图像质控的方法及装置，以期提高对数字病理图像的质量管理，实现对数字病理图像质控标准的量化及统一。
问题的解决方案
技术解决方案
本申请实施例第一方面提供了一种数字病理图像质控的方法，其中，包括：根据数字病理图像的图像格式得到所述数字病理图像的指定层的待处理图像；将所述待处理图像输入U-Net分割模型，得到所述待处理图像中与每个像素点分别对应的第一异常概率矩阵，所述第一异常概率矩阵包括多个相关值，所述多个相关值表征该第一异常概率矩阵对应的像素点分别与各个质控标签的相关度，所述各个质控标签包括非异常标签及N种异常标签，所述U-Net分割模型是基于图像样本训练得到的，所述图像样本中至少包括多个数字病理图像样本及所述数字病理图像样本中每个像素点的质控标签特征，N为正整数；根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签；根据所述待处理图像中每个像素点的第一质控标签，统计所述待处理图像中所述N种异常标签对应的像素点数量；根据所述待处理图像中所述N种异常标签对应的像素点数量对所述数字病理图像进行分类。
本申请实施例第二方面提供了一种数字病理图像质控的装置，包括：获取模块，用于根据数字病理图像的图像格式得到所述数字病理图像的指定层的待处理图像；学习模块，用于将所述待处理图像输入U-Net分割模型，得到所述待处理图像中与每个像素点分别对应的第一异常概率矩阵，所述第一异常概率矩阵包括多个相关值，所述多个相关值表征该第一异常概率矩阵对应的像素点分别与各个质控标签的相关度，所述各个质控标签包括非异常标签及N种异常标签，所述U-Net分割模型是基于图像样本训练得到的，所述图像样本中至少包括多个数字病理图像样本及所述数字病理图像样本中每个像素点的质控标签特征，N为正整数；确定模块，用于根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签；统计模块，用于根据所述待处理图像中每个像素点的第一质控标签，统计所述待处理图像中所述N种异常标签对应的像素点数量；分类模块，用于根据所述待处理图像中所述N种异常标签对应的像素点数量对所述数字病理图像进行分类。
本申请实施例第三方面提供了一种电子设备，其中，包括处理器、存储器、输入输出接口；所述处理器分别与所述存储器和所述输入输出接口相连，其中， 所述输入输出接口用于与用户进行数据交互，所述存储器用于存储程序代码，所述处理器用于调用所述程序代码，以执行如本申请实施例中第一方面所述的数字病理图像质控方法。
本申请实施例第四方面提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序包括程序指令，所述程序指令当被处理器执行时，执行如本申请实施例中第一方面所述的数字病理图像质控方法。
发明的有益效果
有益效果
本申请实施例可以实现对数字病理图像质控标准的量化，并通过病理图像管理平台进行统一存储，使得数字病理图像的管理更加统一高效，并得到数字病理图像统一的质控标准，从而提高对数字病理图像的管理及质控效率。
对附图的简要说明
附图说明
图1是本申请实施例提供的一种数字病理图像质控的方法流程示意图。
图2是本申请实施例提供的一种U-Net分割模型输出示意图。
图3是本申请实施例提供的一种异常分类器输出示意图。
图4是本申请实施例提供的一种数字病理图像质控的装置。
图5是本申请实施例提供的电子设备的结构示意图。
发明实施例
本发明的实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
请参见图1，是本申请实施例提供的一种数字病理图像质控的方法流程示意图。如图1所示，上述方法包括如下步骤。
步骤S101，根据数字病理图像的图像格式得到数字病理图像的指定层的待处理 图像。
具体的，当病理图像管理平台接收到用户上传的数字病理图像后，获取该数字病理图像的图像格式，根据该图像格式确定该提取的指定层的图像层次，得到该数字病理图像的指定层的待处理图像。其中，该病理图像管理平台可以是通过管理员及普通用户等登陆用户账号确定使用用户的权限，例如，管理员可以是医院总部，普通用户可以是该医院总部的下属医院，当下属医院使用扫描仪扫描病理切片得到该病理切片的数字病理图像，将该数字病理图像上传至病理图像管理平台，当该病理图像管理平台接收到该数字病理图像后，获取该数字病理图像的图像格式，根据图像格式得到数字病理图像的指定层的待处理图像。
其中，该数字病理图像存在不同的图像格式，如.svs、.kfb、.ndpi及.tif等。对于不同的图像格式，所需要提取的图像层次不同，当获取到该数字病理图像的图像格式后，根据该图像格式对应的预设图像层次，提取该数字病理图像中指定层的待处理图像。具体的，通过扫描仪扫描病理切片得到的数字病理图像的图像格式不同，组成该数字病理图像的图层也就不同。其中，数字病理图像由M个图层组成，从第0层开始分辨率逐渐降低，即第0层的分辨率最大，M为正整数，不同图像格式的数字病理图像的M不同。例如，若获取到的是.ndpi格式的数字病理图像，该.ndpi格式的数字病理图像由十个图层组成，从第0层到第9层图层的分辨率降低，此时M为10。假定该.ndpi格式的数字病理图像对应的预设图像层次为4，则提取该数字病理图像第4层的待处理图像，其中，此处的指定层为第4层。其中，可以通过图像处理方法提取数字病理图像指定层的待处理图像，该图像处理方法可以是openslide，或者直接通过分辨率对该数字病理图像进行采样。
步骤S102，将待处理图像输入U-Net分割模型，得到待处理图像中每个像素点的第一异常概率矩阵。
具体的，将待处理图像输入U-Net分割模型，得到待处理图像中与每个像素点分别对应的第一异常概率矩阵，该第一异常概率矩阵包括多个相关值，所述多个相关值表征该第一异常概率矩阵对应的像素点分别与各个质控标签的相关度 ，该各个质控标签包括非异常标签及N种异常标签，其中，U-Net分割模型是基于图像样本训练得到的，图像样本中至少包括多个数字病理图像样本及每个数字病理图像样本中每个像素点的质控标签特征，N为正整数。举例来说，若该待处理图像中包括20*20个像素点，则将该待处理图像输入U-Net分割模型后，得到400个第一异常概率矩阵，第一个像素点对应的第一异常概率矩阵中包含(N+1)个相关值，(N+1)个相关值为第一个像素点分别与各个质控标签的相关度，其中，待处理图像中每个像素点对应一个第一异常概率矩阵，即在本例中得到20*20个第一异常概率矩阵。
具体的，将待处理图像输入U-Net分割模型，得到待处理图像中每个像素点的第一异常概率矩阵，基于待处理图像中每个像素点的第一异常概率矩阵中相关值最大的质控标签，确定异常显示图像，具体的，若与待处理图像中像素点的相关值最大的质控标签为非异常标签，则将该异常显示图像中与待处理图像中像素点位置相同的像素点的灰度值设置为预设背景值，若与待处理图像中像素点的相关值最大的质控标签为N种异常标签中的任一异常标签，则将该异常显示图像中与待处理图像中像素点位置相同的像素点的灰度值设置为预设异常值，根据设置的异常显示图像中的各个像素点确定异常显示图像。具体来说，是将该待处理图像进行二值化处理，该预设背景值可以为0，预设异常值可以为255，或者预设背景值为255，预设异常值为0。其中，预设背景值及预设异常值的灰度值也可以是分别取0、1或1、0，具体根据U-Net分割模型输出图像的灰度值设置进行预设。
其中，U-Net分割模型为一种全卷积神经网络，是一个端到端的网络，即输入输出均为图像。将待处理图像输入U-Net分割模型，在收缩路径中通过卷积层后采用激活函数对待处理图像进行下采样，提取该待处理图像的特征图，再在扩展路径中进行上采样，并在每次上采样时添加收缩路径中得到的对应特征图，最终实现对待处理图像中每个像素点的第一异常概率矩阵的提取。其中，该U-Net分割模型是u型结构，可以认为上采样时添加的收缩路径中对应特征图为该u型结构中同一层次下采样时得到的。
具体的，假定将图像尺寸为512*512的待处理图像输入到U-Net分割模型，与U- Net分割模型中的各个权重矩阵进行学习，得到该待处理图像中每个像素点对应的第一异常概率矩阵，该第一异常概率矩阵为(N+1)维向量，即包括(N+1)个相关值，可以认为包括512*512个(N+1)维向量，获取每一个(N+1)维向量中相关值最大的质控标签，将相关值最大的质控标签确定为该(N+1)维向量对应的像素点的质控标签，根据该质控标签是非异常标签或N种异常标签中任一种异常标签，对该像素点的灰度值进行二值化，从而实现对待处理图像的二值化处理，得到异常显示图像。举例来说，假定该N种异常标签包括褶皱、折叠、刀痕及气泡，此时N为4，0表示非异常标签，1表示褶皱这一异常标签、2表示折叠这一异常标签、3表示刀痕这一异常标签及4表示气泡这一异常标签，当将待处理图像输入U-Net分割模型后，得到每个像素点的第一异常概率矩阵，该第一异常概率矩阵包括对应像素点分别与各个质控标签的相关值。如对于(0，0)处像素点的第一概率矩阵为[0.8，0.1，0，0.05，0.05]，根据该第一概率矩阵可知(0，0)处的像素点与质控标签0的相关值最大，质控标签0为非异常标签，则将异常显示图像中(0，0)处的像素点的灰度值设置为255，此处是以将非异常标签对应像素点设置为白色，将异常标签对应像素点设置为黑色，其中，黑色灰度值为0，白色灰度值为255，直至确定异常显示图像中每个像素点的灰度值，得到异常显示图像。具体可以参见图2中异常显示图像201，图2是本申请实施例提供的一种U-Net分割模型输出示意图。通过对待处理图像中每个像素点的第一异常概率矩阵进行对比，得到每个像素点的第一异常概率矩阵中相关值最大的质控标签对应第一质控标签图202，对待处理图像进行二值化处理，确定异常显示图像201中每个像素点的灰度值，以得到异常显示图像201。
具体的，在对U-Net分割模型进行训练时，获取多个数字病理图像样本及每个数字病理图像样本中每个像素点的质控标签特征，具体是获取多个数字病理图像样本及多个异常标签图像样本，每个异常标签图像样本为对应数字病理图像样本添加质控标签特征后得到的，即包括对应数字病理图像样本中每个像素点的质控标签特征。将数字病理图像样本输入初始U-Net网络中进行训练，不断更新该初始U-Net网络中的权重矩阵及卷积参数，直至该U-Net网络输出的图像中每个像素点的异常概率矩阵中相关值最大的质控标签，为异常标签图像样本中 对应像素点的质控标签特征，从而得到训练后的U-Net分割模型。
步骤S103，根据待处理图像中每个像素点与各个质控标签的相关值，确定待处理图像中每个像素点的第一质控标签。
具体的，将待处理图像中每个像素点的第一异常概率矩阵中相关值最大的质控标签，确定为该待处理图像中每个像素点的第一质控标签。其中，该相关值最大的质控标签是在步骤S102中确定每个像素点为非异常标签或N种异常标签中任一异常标签时确定，在步骤S102中得到异常显示图像中每个像素点的灰度值的同时，确定该像素点的第一质控标签。具体参见图2中第一质控标签图202所示。
步骤S104，根据待处理图像中每个像素点的第一质控标签，统计待处理图像中N种异常标签对应的像素点数量，对数字病理图像进行分类。
具体的，根据待处理图像中每个像素点的第一质控标签，统计该待处理图像中N种异常标签分别对应的像素点数量，对数字病理图像进行分类。具体可以根据该待处理图像中N种异常标签分别对应的像素点数量，基于各种异常标签的转移权重，将每种异常标签对应的像素点的数量进行加权求和，确定该数字病理图像的数据转移量，根据数据转移量对数字病理图像进行分类。其中，该转移权重可以根据异常标签对应的异常对数字病理图像的效果影响程度确定。
其中，可以设定一个病理质控标准，该病理质控标准是经过病理质控专家所统一确定的，可以认为是一个用于对数字病理图像分类的统一标准。其中，该病理质控标准可以预设L类，如L为100时，可以认为该数字病理图像的质控总分为100分，通过对待处理图像中N种异常标签对应的像素点数量，计算得到该待处理图像的数据转移量，通过计算100与数据转移量的差值，确定该待处理图像对应的数字病理图像所属的质控类别；或者，若L为3时，可以认为该数字病理图像包括“优、中、差”三类，根据待处理图像中N种异常标签对应的像素点数量得到该待处理图像的数据转移量，根据该数据转移量确定对应的质控类别，例如，当数据转移量为0～3时对应“优”，当数据转移量为3～10时对应“中”，当数据转移量大于10时对应“差”等，L的值可以根据需求进行设定。其中，可以通过设置不同异常标签对应的不同数据转移量区间，获取到待处理图像中每种异常标签 的像素点数量对应的数据转移量区间，以确定该待处理图像的数据转移量，如当异常标签“刀痕”存在像素点0～3时，对应数据转移量0等，将不同异常标签对应的数据转移量进行求和或加权求和或预设求和公式求和得到待处理图像的数据转移量；或者，直接根据不同异常标签的像素点数量进行求和或加权求和得到待处理图像的数据转移量；或者，根据不同异常标签的像素点数量占待处理图像的百分比，确定该待处理图像的数据转移量。
可选的，可以将待处理图像中第一质控标签为非异常标签的像素点的灰度值设置为预设背景值，第一质控标签为异常标签的像素点的灰度值不变，得到第一图像，将该第一图像输入异常分类器，得到第一图像中与每个像素点分别对应的第二异常概率矩阵，该第二异常概率矩阵包括多个相关值，所述多个相关值表征该第二异常概率矩阵在第一图像中对应的像素点与各个质控标签的相关度，该异常分类器是基于多个第一图像样本及数字病理图像样本中每个像素点的质控标签特征训练得到的，每个第一图像样本是对应数字病理图像样本基于各个像素点分别与各个质控标签的相关值，将相关值最大为非异常标签的像素点的灰度值设置为预设背景值后得到的；根据第一图像中每个像素点的第二异常概率矩阵，确定第一图像中每个像素点的第二质控标签，第二质控标签为第一图像中每个像素点的第二异常概率矩阵中相关值最大的质控标签；根据第一图像中每个像素点的第二质控标签，确定待处理图像中N种异常标签分别对应的像素点数量。
其中，U-Net分割模型与异常分类器均为一种端到端的卷积神经网络，卷积神经网络在进行训练或预测时，本质上在处理某一像素点或像素块时，会结合该像素点或像素块周边的预测环境对该像素点或像素块进行处理，因此，在将待处理图像中非异常标签对应的像素点的灰度值设置为预设背景值后，各种异常标签对应的像素点的灰度值与非异常标签对应的像素点的灰度值差值会较大，使得在出现有单个像素点对应异常标签时，在输入异常分类器进行预测，会由于该单个像素点的灰度值与相邻像素点的灰度值差异较大，而一般情况下不会出现单个像素点异常的情况，从而在预测时，会通过该单个像素点相邻的像素点提供的信息，包括环境场信息及细节信息等，对该单个像素点的异常标签进 行过滤。具体的，异常分类器中通过上采样提取该单个像素点的特征信息，通过下采样展现该单个像素点的环境信息，该单个像素点的环境信息是通过相邻像素点的信息所提供的，结合上采样得到的特征信息及下采样得到的环境信息，以确定该单个像素点的灰度值及异常标签是否合理，实现通过相邻像素点对该单个像素点的异常标签的过滤。举例来说，当单个像素点的灰度值不为预设背景值，而该单个像素点相邻的像素点的灰度值均为预设背景值，在预测过程中，得到该单个像素点的异常概率矩阵，该异常概率矩阵包括该单个像素点与各个质控标签的相关值，并根据该单个像素点相邻像素点的像素点信息，包括相邻像素点的灰度值及异常概率矩阵，以得到该单个像素点的环境信息，该环境信息用于指示该单个像素点的相邻像素点均为非异常像素点，通过该环境信息对该单个像素点的异常概率矩阵进行调整，增加该单个像素点与非异常标签的相关值，从而实现对该单个像素点的异常标签的过滤。
其中，在可选的方式中，该异常分类器是在U-Net分割模型训练的基础上进行训练的，具体的是在对U-Net分割模型训练好后，根据该U-Net分割模型的输出结果对数字病理图像样本进行更新得到第一图像样本，根据第一图像样本及数字病理图像样本中每个像素点的质控标签特征对异常分类器进行训练。
可选的，在上述步骤S101至步骤S104的执行过程中，在步骤S101中得到待处理图像后，可以获取该待处理图像的图像尺寸，若该待处理图像的图像尺寸大于预设滑窗尺寸，则通过预设滑窗尺寸将待处理图像滑窗分割为多个第二图像，第二图像的图像尺寸不大于预设滑窗尺寸，预设滑窗尺寸为U-Net分割模型训练的数字病理图像样本的尺寸或者根据需求设置的预设滑窗尺寸；将多个第二图像中任一第二图像输入U-Net分割模型，得到任一第二图像中每个像素点的第一异常概率矩阵。其中，在将待处理图像分割为多个第二图像后，记录多个第二图像中每个第二图像的位置信息，该位置信息为对应第二图像在待处理图像中的相对位置，该相对位置可以为每个第二图像的左上角第一个像素点在待处理图像中的像素位置，或者是不同第二图像间的相对位置，如(0，0)表示对应第二图像在待处理图像中的第一行第一列。在步骤S103中，基于任一第二图像中每个像素点的第一异常概率矩阵，确定任一第二图像中每个像素点的第一质 控标签，基于多个第二图像中每个第二图像的位置信息及任一第二图像中每个像素点的第一质控标签，确定待处理图像中每个像素点的第一质控标签。
可选的，在步骤S104之后，可以向用户显示数字病理图像、异常显示图像及数字病理图像对应待处理图像中每个像素点的质控标签，以使用户可以直观的得到该数字病理图像的质量情况。该异常显示图像如图2中201所示，待处理图像中每个像素点的质控标签可以如图2中第一质控标签图202所示。其中，若将待处理图像中第一质控标签为非异常标签的像素点的灰度值设置为预设背景值，得到第一图像，将该第一图像输入异常分类器中进行学习，则会对待处理图像进行二次学习，以尽可能合理的标记出数字病理图像的异常区域，避免对数字病理图像的质控评分太低的情况。例如，在出现图2中异常显示图像201中误标记的两处质控标签为2及质控标签为3的像素点时，可以对误标记的像素点进行过滤，得到如图3中所示过滤后的异常显示图像301及第二质控标签图302，图3是本申请实施例提供的一种异常分类器输出示意图。
本申请实施例实现对数字病理图像质控标准的量化，并通过病理图像管理平台进行统一存储，使得数字病理图像的管理更加统一高效，并得到数字病理图像统一的质控标准，从而提高对数字病理图像的管理及质控效率。同时，可以通过对待处理图像进行二次异常标签提取，以尽可能合理的标记出异常区域，由于对于异常来说，几乎很难出现一两个像素点异常的情况，通过二次异常标签提取，可以减少误标记的情况出现，提高质控的准确性。同时，对模型训练时是对每一个像素点进行处理，使得模型的输入和输出的图像的图像尺寸相同，以提高图像的精度。
可选的，参见图4，图4是本申请实施例提供的一种数字病理图像质控的装置。如图4所示，该数字病理图像质控的装置可以用于上述图1所对应实施例中的电子设备，具体的该数字病理图像质控的装置40包括获取模块401、学习模块402、确定模块403、统计模块404及分类模块405。
获取模块401，用于根据数字病理图像的图像格式得到所述数字病理图像的指定层的待处理图像；学习模块402，用于将所述待处理图像输入U-Net分割模型，得到所述待处理图像中与每个像素点分别对应的第一异常概率矩阵，所述第 一异常概率矩阵包括多个相关值，所述多个相关值表征该第一异常概率矩阵对应的像素点分别与各个质控标签的相关度，所述各个质控标签包括非异常标签及N种异常标签，所述U-Net分割模型是基于图像样本训练得到的，所述图像样本中至少包括多个数字病理图像样本及所述数字病理图像样本中每个像素点的质控标签特征，N为正整数；确定模块403，用于根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签；统计模块404，用于根据所述待处理图像中每个像素点的第一质控标签，统计所述待处理图像中所述N种异常标签对应的像素点数量；分类模块405，用于根据所述待处理图像中所述N种异常标签对应的像素点数量对所述数字病理图像进行分类。
其中，所述统计模块404包括：设置单元4041，用于将所述待处理图像中第一质控标签为所述非异常标签的像素点的灰度值设置为预设背景值，得到第一图像；学习单元4042，用于将所述第一图像输入异常分类器，得到所述第一图像中与每个像素点分别对应的第二异常概率矩阵，所述第二异常概率矩阵包括多个相关值，所述多个相关值表征该第二异常概率矩阵在所述第一图像中对应的像素点分别与所述各个质控标签的相关度，所述异常分类器是基于第一图像样本及所述数字病理图像样本中每个像素点的质控标签特征训练得到的，所述第一图像样本是所述数字病理图像样本基于各个像素点分别与所述各个质控标签的相关值，将相关值最大为所述非异常标签的像素点的灰度值设置为所述预设背景值后得到的；确定单元4043，用于根据所述第一图像中每个像素点的第二异常概率矩阵，确定所述第一图像中每个像素点的第二质控标签，所述第二质控标签为所述第一图像中每个像素点的第二异常概率矩阵中相关值最大的质控标签；所述确定单元4043，还用于根据所述第一图像中每个像素点的第二质控标签，确定所述待处理图像中所述N种异常标签对应的像素点数量。
其中，所述装置还包括：所述获取模块401，还用于获取所述待处理图像的图像尺寸；滑窗模块406，用于若所述待处理图像的图像尺寸大于预设滑窗尺寸，则通过预设滑窗尺寸将所述待处理图像滑窗分割为多个第二图像，所述第二图像的图像尺寸不大于所述预设滑窗尺寸，所述预设滑窗尺寸为所述U-Net分割模 型训练的所述数字病理图像样本的尺寸；所述学习模块402，具体用于：将所述多个第二图像中任一第二图像输入所述U-Net分割模型，得到所述任一第二图像中每个像素点的第一异常概率矩阵。
其中，所述装置还包括：记录模块407，用于记录所述多个第二图像中每个第二图像的位置信息，所述位置信息为对应第二图像在所述待处理图像中的相对位置；所述确定模块403，具体用于：根据所述任一第二图像中每个像素点的第一异常概率矩阵，确定所述任一第二图像中每个像素点的第一质控标签；基于所述多个第二图像中每个第二图像的位置信息及所述任一第二图像中每个像素点的第一质控标签，确定所述待处理图像中每个像素点的第一质控标签。
其中，所述分类模块405具体用于：根据所述待处理图像中所述N种异常标签中每种异常标签对应的像素点的数量，基于所述异常标签的转移权重，将所述每种异常标签对应的像素点的数量进行加权求和，确定所述数字病理图像的数据转移量，根据所述数据转移量对所述数字病理图像进行分类。
其中，所述确定模块403具体用于：将所述待处理图像中每个像素点的第一异常概率矩阵中相关值最大的质控标签，确定为所述待处理图像中每个像素点的第一质控标签。
所述学习模块402，具体用于：将所述待处理图像输入所述U-Net分割模型，得到所述待处理图像中每个像素点的第一异常概率矩阵，基于所述待处理图像中每个像素点的第一异常概率矩阵中相关值最大的质控标签，确定异常显示图像。
在所述确定异常显示图像方面，所述学习模块402具体用于：若与所述待处理图像中像素点的相关值最大的质控标签为所述非异常标签，则将所述异常显示图像中与所述待处理图像中像素点位置相同的像素点的灰度值设置为预设背景值；若与所述待处理图像中像素点的相关值最大的质控标签为所述N种异常标签中的任一异常标签，则将所述异常显示图像中与所述待处理图像中像素点位置相同的像素点的灰度值设置为预设异常值；根据设置的异常显示图像中的各个像素点确定所述异常显示图像。
所述装置还包括：输出模块408，用于输出所述数字病理图像的分类结果及所 述异常显示图像。
具体实现中，上述装置可通过上述各个模块执行上述图1所提供的实现方式中各个步骤所提供的实现方式，实现上述各实施例中所实现的功能，具体可参见上述图1所示的方法实施例中各个步骤提供的相应描述，在此不再赘述。
本申请实施例提供了一种数字病理图像质控的装置，上述装置对数字病理图像质控标准的量化，并通过病理图像管理平台进行统一存储，使得数字病理图像的管理更加统一高效，并得到数字病理图像统一的质控标准，从而提高对数字病理图像的管理及质控效率。
参见图5，图5是本申请实施例提供的电子设备的结构示意图。如图5所示，本实施例中的电子设备可以包括：一个或多个处理器501、存储器502和收发器503。上述处理器501、存储器502和收发器503通过总线504连接。存储器502用于存储计算机程序，该计算机程序包括程序指令，收发器503用于连接终端设备，与上述电子设备进行数据交互；处理器501用于执行存储器502存储的程序指令，执行如下操作。
根据数字病理图像的图像格式得到所述数字病理图像的指定层的待处理图像；将所述待处理图像输入U-Net分割模型，得到所述待处理图像中每个像素点的第一异常概率矩阵，所述第一异常概率矩阵包括对应像素点分别与各个质控标签的相关值，所述各个质控标签包括非异常标签及N种异常标签，所述U-Net分割模型是基于图像样本训练得到的，所述图像样本中至少包括多个数字病理图像样本及所述数字病理图像样本中每个像素点的质控标签特征，N为正整数；根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签；根据所述待处理图像中每个像素点的第一质控标签，统计所述待处理图像中所述N种异常标签对应的像素点数量；根据所述待处理图像中所述N种异常标签对应的像素点数量对所述数字病理图像进行分类。
在一些可行的实施方式中，上述处理器501可以是中央处理单元(central processing unit，CPU)，该处理器还可以是其他通用处理器、数字信号处理器(digital signal  processor，DSP)、专用集成电路(applicationspecificintegrated circuit，ASIC)、现成可编程门阵列(field-programmable gate array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
该存储器502可以包括只读存储器和随机存取存储器，并向处理器501和收发器503提供指令和数据。存储器502的一部分还可以包括非易失性随机存取存储器。例如，存储器502还可以存储设备类型的信息。
具体实现中，上述电子设备可通过其内置的各个功能模块执行如上述图1各个步骤所提供的实现方式，具体可参见上述图1中各个步骤所提供的实现方式，在此不再赘述。
本申请实施例通过提供一种电子设备，包括：处理器、收发器、存储器，通过处理器获取存储器中的计算机指令，执行上述图1中所示方法的各个步骤，实现对数字病理图像质控标准的量化，并通过病理图像管理平台进行统一存储，使得数字病理图像的管理更加统一高效，并得到数字病理图像统一的质控标准，从而提高对数字病理图像的管理及质控效率。
本申请实施例还提供一种计算机可读存储介质，该计算机可读存储介质存储有计算机程序，该计算机程序包括程序指令，该程序指令被处理器执行时实现图1中各个步骤所提供的数字病理图像质控的方法，具体可参见上述图1中各个步骤所提供的实现方式，在此不再赘述。其中，该计算机可读存储介质可以是非易失性可读存储介质，也可以是易失性可读存储介质。
上述计算机可读存储介质可以是前述任一实施例提供的数字病理图像质控装置或者上述终端设备的内部存储单元，例如电子设备的硬盘或内存。该计算机可读存储介质也可以是该电子设备的外部存储设备，例如该电子设备上配备的插接式硬盘，智能存储卡(smart media card，SMC)，安全数字(secure digital，SD)卡，闪存卡(flash card)等。进一步地，该计算机可读存储介质还可以既包括该电子设备的内部存储单元也包括外部存储设备。该计算机可读存储介质用于存储该计算机程序以及该电子设备所需的其他程序和数据。该计算机可读存储介质还可以用于暂时地存储已经输出或者将要输出的数据。
本申请实施例的说明书和权利要求书及附图中的术语“第一”、“第二”等是用于区别不同对象，而非用于描述特定顺序。此外，术语“包括”以及它们任何变形，意图在于覆盖不排他的包含。例如包含了一系列步骤或单元的过程、方法、装置、产品或设备没有限定于已列出的步骤或模块，而是可选地还包括没有列出的步骤或模块，或可选地还包括对于这些过程、方法、装置、产品或设备固有的其他步骤单元。以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到变化或替换，都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以所述权利要求的保护范围为准。

Claims (20)

  1. 一种数字病理图像质控的方法，其中，包括：
    根据数字病理图像的图像格式得到所述数字病理图像的指定层的待处理图像；
    将所述待处理图像输入U-Net分割模型，得到所述待处理图像中与每个像素点分别对应的第一异常概率矩阵，所述第一异常概率矩阵包括多个相关值，所述多个相关值表征该第一异常概率矩阵对应的像素点分别与各个质控标签的相关度，所述各个质控标签包括非异常标签及N种异常标签，所述U-Net分割模型是基于图像样本训练得到的，所述图像样本中至少包括多个数字病理图像样本及所述数字病理图像样本中每个像素点的质控标签特征，N为正整数；
    根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签；
    根据所述待处理图像中每个像素点的第一质控标签，统计所述待处理图像中所述N种异常标签对应的像素点数量；
    根据所述待处理图像中所述N种异常标签对应的像素点数量对所述数字病理图像进行分类。
  2. 如权利要求1所述的方法，其中，所述根据所述待处理图像中每个像素点的第一质控标签，统计所述待处理图像中所述N种异常标签对应的像素点数量，包括：
    将所述待处理图像中第一质控标签为所述非异常标签的像素点的灰度值设置为预设背景值，得到第一图像；
    将所述第一图像输入异常分类器，得到所述第一图像中与每个像素点的第二异常概率矩阵，所述第二异常概率矩阵包括多个相关值，所述多个相关值表征该第二异常概率矩阵在所述第一图像中对应的像素点分别与所述各个质控标签的相关度，所述异常分类器是基于第一图像样本及所述数字病理图像样本中每个像素点的 质控标签特征训练得到的，所述第一图像样本是所述数字病理图像样本基于各个像素点分别与所述各个质控标签的相关值，将相关值最大为所述非异常标签的像素点的灰度值设置为所述预设背景值后得到的；
    根据所述第一图像中每个像素点的第二异常概率矩阵，确定所述第一图像中每个像素点的第二质控标签，所述第二质控标签为所述第一图像中每个像素点的第二异常概率矩阵中相关值最大的质控标签；
    根据所述第一图像中每个像素点的第二质控标签，确定所述待处理图像中所述N种异常标签对应的像素点数量。
  3. 如权利要求1所述的方法，其中，所述得到所述数字病理图像的指定层的待处理图像之后，还包括：
    获取所述待处理图像的图像尺寸，若所述待处理图像的图像尺寸大于预设滑窗尺寸，则通过预设滑窗尺寸将所述待处理图像滑窗分割为多个第二图像，所述第二图像的图像尺寸不大于所述预设滑窗尺寸，所述预设滑窗尺寸为所述U-Net分割模型训练的所述数字病理图像样本的尺寸；
    所述将所述待处理图像输入U-Net分割模型，得到所述待处理图像中每个像素点的第一异常概率矩阵，包括：
    将所述多个第二图像中任一第二图像输入所述U-Net分割模型，得到所述任一第二图像中每个像素点的第一异常概率矩阵。
  4. 如权利要求3所述的方法，其中，所述通过滑窗法将所述待处理图像分割为多个第二图像之后，还包括：
    记录所述多个第二图像中每个第二图像的位置信息，所述位置信息为对应第二图像在所述待处理图像中的相对位置；
    所述根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签，包括：
    根据所述任一第二图像中每个像素点的第一异常概率矩阵，确定所述任一第二图像中每个像素点的第一质控标签；
    基于所述多个第二图像中每个第二图像的位置信息及所述任一第二图像中每个像素点的第一质控标签，确定所述待处理图像中每个像素点的第一质控标签。
  5. 如权利要求1所述的方法，其中，所述根据所述待处理图像中所述N种异常标签对应的像素点数量对所述数字病理图像进行分类，包括：
    根据所述待处理图像中所述N种异常标签中每种异常标签对应的像素点的数量，基于所述异常标签的转移权重，将所述每种异常标签对应的像素点的数量进行加权求和，确定所述数字病理图像的数据转移量，根据所述数据转移量对所述数字病理图像进行分类。
  6. 如权利要求1所述的方法，其中，所述根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签，包括：
    将所述待处理图像中每个像素点的第一异常概率矩阵中相关值最大的质控标签，确定为所述待处理图像中每个像素点的第一质控标签。
  7. 如权利要求1所述的方法，其中，所述将所述待处理图像输入U-Net分割模型，得到所述待处理图像中每个像素点的第一异常概率矩阵，包括：
    将所述待处理图像输入所述U-Net分割模型，得到所述待处理图像中每个像素点的第一异常概率矩阵，基于所述待处理图像中每个像素点的第一异常概率矩阵中相关值最大的质控标签，确定异常显示图像。
  8. 如权利要求7所述的方法，其中，所述确定异常显示图像，包括：
    若与所述待处理图像中像素点的相关值最大的质控标签为所述非 异常标签，则将所述异常显示图像中与所述待处理图像中像素点位置相同的像素点的灰度值设置为预设背景值；
    若与所述待处理图像中像素点的相关值最大的质控标签为所述N种异常标签中的任一异常标签，则将所述异常显示图像中与所述待处理图像中像素点位置相同的像素点的灰度值设置为预设异常值；
    根据设置的异常显示图像中的各个像素点确定所述异常显示图像。
  9. 如权利要求7所述的方法，其中，所述方法之后，还包括：
    输出所述数字病理图像的分类结果及所述异常显示图像。
  10. 一种数字病理图像质控的装置，其中，所述装置包括：
    获取模块，用于根据数字病理图像的图像格式得到所述数字病理图像的指定层的待处理图像；
    学习模块，用于将所述待处理图像输入U-Net分割模型，得到所述待处理图像中与每个像素点分别对应的第一异常概率矩阵，所述第一异常概率矩阵包括多个相关值，所述多个相关值表征该第一异常概率矩阵对应的像素点分别与各个质控标签的相关度，所述各个质控标签包括非异常标签及N种异常标签，所述U-Net分割模型是基于图像样本训练得到的，所述图像样本中至少包括多个数字病理图像样本及所述数字病理图像样本中每个像素点的质控标签特征，N为正整数；
    确定模块，用于根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签；
    统计模块，用于根据所述待处理图像中每个像素点的第一质控标签，统计所述待处理图像中所述N种异常标签对应的像素点数量；
    分类模块，用于根据所述待处理图像中所述N种异常标签对应的像素点数量对所述数字病理图像进行分类。
  11. 如权利要求10所述的装置，其中，所述统计模块包括：
    设置单元，用于将所述待处理图像中第一质控标签为所述非异常标签的像素点的灰度值设置为预设背景值，得到第一图像；
    学习单元，用于将所述第一图像输入异常分类器，得到所述第一图像中与每个像素点分别对应的第二异常概率矩阵，所述第二异常概率矩阵包括多个相关值，所述多个相关值表征该第二异常概率矩阵在所述第一图像中对应的像素点分别与所述各个质控标签的相关度，所述异常分类器是基于第一图像样本及所述数字病理图像样本中每个像素点的质控标签特征训练得到的，所述第一图像样本是所述数字病理图像样本基于各个像素点分别与所述各个质控标签的相关值，将相关值最大为所述非异常标签的像素点的灰度值设置为所述预设背景值后得到的；
    确定单元，用于根据所述第一图像中每个像素点的第二异常概率矩阵，确定所述第一图像中每个像素点的第二质控标签，所述第二质控标签为所述第一图像中每个像素点的第二异常概率矩阵中相关值最大的质控标签；
    所述确定单元，还用于根据所述第一图像中每个像素点的第二质控标签，确定所述待处理图像中所述N种异常标签对应的像素点数量。
  12. 一种电子设备，其中，包括处理器、存储器、输入输出接口；
    所述处理器分别与所述存储器和所述输入输出接口相连，其中，所述输入输出接口用于与用户进行数据交互，所述存储器用于存储程序代码，所述处理器用于调用所述程序代码，以执行：
    根据数字病理图像的图像格式得到所述数字病理图像的指定层的待处理图像；
    将所述待处理图像输入U-Net分割模型，得到所述待处理图像中与每个像素点分别对应的第一异常概率矩阵，所述第一异常概率矩阵包括多个相关值，所述多个相关值表征该第一异常概率矩阵对 应的像素点分别与各个质控标签的相关度，所述各个质控标签包括非异常标签及N种异常标签，所述U-Net分割模型是基于图像样本训练得到的，所述图像样本中至少包括多个数字病理图像样本及所述数字病理图像样本中每个像素点的质控标签特征，N为正整数；
    根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签；
    根据所述待处理图像中每个像素点的第一质控标签，统计所述待处理图像中所述N种异常标签对应的像素点数量；
    根据所述待处理图像中所述N种异常标签对应的像素点数量对所述数字病理图像进行分类。
  13. 如权利要求12所述的电子设备，其中，在所述根据所述待处理图像中每个像素点的第一质控标签，统计所述待处理图像中所述N种异常标签对应的像素点数量方面，所述处理器具体用于：
    将所述待处理图像中第一质控标签为所述非异常标签的像素点的灰度值设置为预设背景值，得到第一图像；
    将所述第一图像输入异常分类器，得到所述第一图像中与每个像素点的第二异常概率矩阵，所述第二异常概率矩阵包括多个相关值，所述多个相关值表征该第二异常概率矩阵在所述第一图像中对应的像素点分别与所述各个质控标签的相关度，所述异常分类器是基于第一图像样本及所述数字病理图像样本中每个像素点的质控标签特征训练得到的，所述第一图像样本是所述数字病理图像样本基于各个像素点分别与所述各个质控标签的相关值，将相关值最大为所述非异常标签的像素点的灰度值设置为所述预设背景值后得到的；
    根据所述第一图像中每个像素点的第二异常概率矩阵，确定所述第一图像中每个像素点的第二质控标签，所述第二质控标签为所述第一图像中每个像素点的第二异常概率矩阵中相关值最大的质 控标签；
    根据所述第一图像中每个像素点的第二质控标签，确定所述待处理图像中所述N种异常标签对应的像素点数量。
  14. 如权利要求12所述的电子设备，其中，所述得到所述数字病理图像的指定层的待处理图像之后，所述处理器还用于：
    获取所述待处理图像的图像尺寸，若所述待处理图像的图像尺寸大于预设滑窗尺寸，则通过预设滑窗尺寸将所述待处理图像滑窗分割为多个第二图像，所述第二图像的图像尺寸不大于所述预设滑窗尺寸，所述预设滑窗尺寸为所述U-Net分割模型训练的所述数字病理图像样本的尺寸；
    在所述将所述待处理图像输入U-Net分割模型，得到所述待处理图像中每个像素点的第一异常概率矩阵方面，所述处理器具体用于：
    将所述多个第二图像中任一第二图像输入所述U-Net分割模型，得到所述任一第二图像中每个像素点的第一异常概率矩阵。
  15. 如权利要求14所述的电子设备，其中，所述通过滑窗法将所述待处理图像分割为多个第二图像之后，所述处理器还用于：
    记录所述多个第二图像中每个第二图像的位置信息，所述位置信息为对应第二图像在所述待处理图像中的相对位置；
    所述根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签，包括：
    根据所述任一第二图像中每个像素点的第一异常概率矩阵，确定所述任一第二图像中每个像素点的第一质控标签；
    基于所述多个第二图像中每个第二图像的位置信息及所述任一第二图像中每个像素点的第一质控标签，确定所述待处理图像中每个像素点的第一质控标签。
  16. 如权利要求12所述的电子设备，其中，在所述根据所述待处理图 像中所述N种异常标签对应的像素点数量对所述数字病理图像进行分类方面，所述处理器具体用于：
    根据所述待处理图像中所述N种异常标签中每种异常标签对应的像素点的数量，基于所述异常标签的转移权重，将所述每种异常标签对应的像素点的数量进行加权求和，确定所述数字病理图像的数据转移量，根据所述数据转移量对所述数字病理图像进行分类。
  17. 如权利要求12所述的电子设备，其中，在所述根据所述待处理图像中每个像素点与所述各个质控标签的相关值，确定所述待处理图像中每个像素点的第一质控标签方面，所述处理器具体用于：
    将所述待处理图像中每个像素点的第一异常概率矩阵中相关值最大的质控标签，确定为所述待处理图像中每个像素点的第一质控标签。
  18. 如权利要求12所述的电子设备，其中，在所述将所述待处理图像输入U-Net分割模型，得到所述待处理图像中每个像素点的第一异常概率矩阵方面，所述处理器具体用于：
    将所述待处理图像输入所述U-Net分割模型，得到所述待处理图像中每个像素点的第一异常概率矩阵，基于所述待处理图像中每个像素点的第一异常概率矩阵中相关值最大的质控标签，确定异常显示图像。
  19. 如权利要求18所述的电子设备，其中，在所述确定异常显示图像方面，所述处理器具体用于：
    若与所述待处理图像中像素点的相关值最大的质控标签为所述非异常标签，则将所述异常显示图像中与所述待处理图像中像素点位置相同的像素点的灰度值设置为预设背景值；
    若与所述待处理图像中像素点的相关值最大的质控标签为所述N种异常标签中的任一异常标签，则将所述异常显示图像中与所述待处理图像中像素点位置相同的像素点的灰度值设置为预设异常值 ；
    根据设置的异常显示图像中的各个像素点确定所述异常显示图像。
  20. 一种计算机可读存储介质，其中，所述计算机可读存储介质存储有计算机程序，所述计算机程序包括程序指令，所述程序指令当被处理器执行时，执行如权利要求1-9任一项所述的方法。
PCT/CN2020/093579 2019-10-29 2020-05-30 一种数字病理图像质控的方法及装置 WO2021082433A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911034430.4A CN111008953B (zh) 2019-10-29 2019-10-29 一种数字病理图像质控的方法及装置
CN201911034430.4 2019-10-29

Publications (1)

Publication Number Publication Date
WO2021082433A1 true WO2021082433A1 (zh) 2021-05-06

Family

ID=70111669

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/093579 WO2021082433A1 (zh) 2019-10-29 2020-05-30 一种数字病理图像质控的方法及装置

Country Status (2)

Country Link
CN (1) CN111008953B (zh)
WO (1) WO2021082433A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113516671A (zh) * 2021-08-06 2021-10-19 重庆邮电大学 一种基于U-net和注意力机制的婴幼儿脑组织分割方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111008953B (zh) * 2019-10-29 2023-10-31 平安科技（深圳）有限公司 一种数字病理图像质控的方法及装置
CN111595850B (zh) * 2020-04-27 2022-07-15 平安科技（深圳）有限公司 切片缺陷检测方法、电子装置及可读存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8401256B2 (en) * 2006-11-13 2013-03-19 General Electric Company Systems and methods for an interactive PACS image display test image
CN108010015A (zh) * 2017-11-07 2018-05-08 深圳市金城保密技术有限公司 一种指静脉影像质量评价方法及其系统
CN108074239A (zh) * 2017-12-30 2018-05-25 中国传媒大学 一种基于先验感知质量特征图的无参考图像质量客观评价方法
CN109191457A (zh) * 2018-09-21 2019-01-11 中国人民解放军总医院 一种病理图像质量有效性识别方法
CN111008953A (zh) * 2019-10-29 2020-04-14 平安科技（深圳）有限公司 一种数字病理图像质控的方法及装置

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9424504B2 (en) * 2014-09-15 2016-08-23 Paypal, Inc. Combining a QR code and an image
EP3570753A4 (en) * 2017-02-23 2020-10-21 Google LLC METHOD AND SYSTEM TO AID IN PATHOLOGIST IDENTIFICATION OF TUMOR CELLS IN ENLARGED TISSUE IMAGES
CN107451998B (zh) * 2017-08-08 2020-11-06 北京大恒普信医疗技术有限公司 一种眼底图像质量控制方法
WO2019108888A1 (en) * 2017-11-30 2019-06-06 The Research Foundation For The State University Of New York SYSTEM AND METHOD TO QUANTIFY TUMOR-INFILTRATING LYMPHOCYTES (TILs) FOR CLINICAL PATHOLOGY ANALYSIS
CN108629777A (zh) * 2018-04-19 2018-10-09 麦克奥迪（厦门）医疗诊断系统有限公司 一种数字病理全切片图像病变区域自动分割方法
CN109785306A (zh) * 2019-01-09 2019-05-21 上海联影医疗科技有限公司 器官勾画方法、装置、计算机设备和存储介质
CN110120040B (zh) * 2019-05-13 2021-09-14 广州锟元方青医疗科技有限公司 切片图像处理方法、装置、计算机设备和存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8401256B2 (en) * 2006-11-13 2013-03-19 General Electric Company Systems and methods for an interactive PACS image display test image
CN108010015A (zh) * 2017-11-07 2018-05-08 深圳市金城保密技术有限公司 一种指静脉影像质量评价方法及其系统
CN108074239A (zh) * 2017-12-30 2018-05-25 中国传媒大学 一种基于先验感知质量特征图的无参考图像质量客观评价方法
CN109191457A (zh) * 2018-09-21 2019-01-11 中国人民解放军总医院 一种病理图像质量有效性识别方法
CN111008953A (zh) * 2019-10-29 2020-04-14 平安科技（深圳）有限公司 一种数字病理图像质控的方法及装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YANG WEI-SHAN , YANG YAN , SHAO FENG: "No reference quality assessment method for retinal images based on blood vessel segmentation", JOURNAL OF OPTOELECTRONICS·LASER, vol. 30, no. 5, 31 May 2019 (2019-05-31), pages 536 - 545, XP055809150, ISSN: 1005-0086, DOI: 10.16136/j.joel.2019.05.0283 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113516671A (zh) * 2021-08-06 2021-10-19 重庆邮电大学 一种基于U-net和注意力机制的婴幼儿脑组织分割方法
CN113516671B (zh) * 2021-08-06 2022-07-01 重庆邮电大学 一种基于U-net和注意力机制的婴幼儿脑组织图像分割方法

Also Published As

Publication number Publication date
CN111008953A (zh) 2020-04-14
CN111008953B (zh) 2023-10-31

Similar Documents

Publication Publication Date Title
CN110232719B (zh) 一种医学图像的分类方法、模型训练方法和服务器
WO2021082433A1 (zh) 一种数字病理图像质控的方法及装置
CN107742100B (zh) 一种考生身份验证方法及终端设备
US9286537B2 (en) System and method for classifying a skin infection
CN110705583A (zh) 细胞检测模型训练方法、装置、计算机设备及存储介质
CN108304816B (zh) 身份识别方法、装置、存储介质及电子设备
CN111652869B (zh) 基于深度学习的板坯空洞识别方法、系统、介质及终端
CN113159147A (zh) 基于神经网络的图像识别方法、装置、电子设备
US20190236738A1 (en) System and method for detection of identity fraud
CN112419202B (zh) 基于大数据及深度学习的野生动物图像自动识别系统
CN108256527A (zh) 一种基于端到端全卷积网络的皮肤病变多类语义分割方法
WO2021114818A1 (zh) 基于傅里叶变换的oct图像质量评估方法、系统及装置
US11721023B1 (en) Distinguishing a disease state from a non-disease state in an image
CN112036295A (zh) 票据图像处理方法、装置、存储介质及电子设备
CN110796039B (zh) 一种面部瑕疵检测方法、装置、电子设备及存储介质
Venkatesvara Rao et al. Real-time video object detection and classification using hybrid texture feature extraction
CN113743378A (zh) 一种基于视频的火情监测方法和装置
CN111369557B (zh) 图像处理方法、装置、计算设备和存储介质
CN117036947A (zh) 基于图像识别的农险出险预警方法、装置、设备及介质
CN116959099A (zh) 一种基于时空图卷积神经网络的异常行为识别方法
CN110909706A (zh) 一种昼伏夜出人员判定方法、装置、电子设备及存储介质
CN111179226B (zh) 视野图的识别方法、装置及计算机存储介质
CN111402185A (zh) 一种图像检测方法及装置
CN110827261B (zh) 图像质量检测方法及装置、存储介质及电子设备
CN114612897A (zh) 一种智能果蔬称重打票方法、装置、电子设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20883122

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20883122

Country of ref document: EP

Kind code of ref document: A1