WO2021082434A1 - 一种异常细胞筛选方法、装置、电子设备和存储介质 - Google Patents

一种异常细胞筛选方法、装置、电子设备和存储介质 Download PDF

Info

Publication number
WO2021082434A1
WO2021082434A1 PCT/CN2020/093581 CN2020093581W WO2021082434A1 WO 2021082434 A1 WO2021082434 A1 WO 2021082434A1 CN 2020093581 W CN2020093581 W CN 2020093581W WO 2021082434 A1 WO2021082434 A1 WO 2021082434A1
Authority
WO
WIPO (PCT)
Prior art keywords
cervical cell
image
images
cervical
sub
Prior art date
Application number
PCT/CN2020/093581
Other languages
English (en)
French (fr)
Inventor
谢魏玮
郭冰雪
王季勇
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 WO2021082434A1 publication Critical patent/WO2021082434A1/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
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10056—Microscopic image
    • 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
    • G06T2207/30096—Tumor; Lesion

Definitions

  • This application relates to the technical field of neural networks, and in particular to a method, device, electronic equipment and storage medium for screening abnormal cells.
  • Cervical cancer is the most common gynecological malignant tumor, and its incidence has tended to be younger in recent years. There are 500,000 new cases and 274,000 deaths worldwide each year, of which 85% of cervical cancer deaths occur in low-surveillance cases. Low- and middle-income areas. Cervical cancer is currently the only cancer that can be detected and cured early. Therefore, early screening and diagnosis is a key link in the prevention and treatment of cervical cancer.
  • the embodiments of the present application provide a method, device, electronic equipment, and storage medium for screening abnormal cells. Implementing the embodiments of the present application can help improve the efficiency of screening abnormal cells.
  • the first aspect of the application provides a method for screening abnormal cells, including:
  • the multiple cervical cell sub-images are respectively input to the abnormal cell screening model to obtain multiple prediction results corresponding to the multiple cervical cell sub-images, wherein each cervical cell sub-image corresponds to a prediction result, and each prediction result It is used to indicate the abnormal cells included in each cervical cell image, and the abnormal cells are cells that have undergone lesions or cancerous changes on the basis of cervical cells;
  • the abnormal cell screening model includes a plurality of neural networks, the number of which is equal to the number of types of abnormal cells, and the plurality of neural networks are used to screen different abnormal cells.
  • the second aspect of the present application provides an abnormal cell screening device, including:
  • the first acquisition module is used to acquire cervical cell images
  • a segmentation module configured to segment the cervical cell image based on the gray values of pixels in the cervical cell image to obtain a plurality of cervical cell sub-images
  • the input module is used to input the multiple cervical cell sub-images into the abnormal cell screening model to obtain multiple prediction results corresponding to the multiple cervical cell sub-images, wherein each cervical cell sub-image corresponds to one prediction result ,
  • Each prediction result is used to indicate the abnormal cells included in each cervical cell image, and the abnormal cells are cells that undergo lesions or cancerous changes on the basis of cervical cells;
  • the selection module is configured to select multiple prediction results from the multiple prediction results as multiple first prediction results according to a preset selection strategy
  • a second acquisition module configured to acquire a plurality of first cervical cell sub-images corresponding to the plurality of first prediction results
  • a display module configured to display the plurality of first cervical cell sub-images on a display interface
  • the abnormal cell screening model includes a plurality of neural networks, the number of which is equal to the number of types of abnormal cells, and the plurality of neural networks are used to screen different abnormal cells.
  • the third aspect of the present application provides an electronic device for screening abnormal cells, including a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and are Generate instructions to be executed by the processor to execute the steps in any method of an abnormal cell screening method.
  • the fourth aspect of the present application provides a computer-readable storage medium, the computer-readable storage medium is used to store a computer program, and the stored computer program is executed by the processor to implement any one of the abnormal cell screening methods. The method described in the item.
  • the abnormal cell screening model includes a plurality of neural networks, the number of which is equal to the number of types of abnormal cells, and the plurality of neural networks are used to screen different abnormal cells, so that multiple types of abnormal cells can be screened out. Abnormal cells are more practical.
  • FIG. 1A is a schematic flowchart of a method for screening abnormal cells according to an embodiment of the application
  • FIG. 1B is a schematic diagram of a display interface provided by an embodiment of the application.
  • 2A is a schematic flow chart of another method for screening abnormal cells according to an embodiment of the application.
  • 2B is a schematic diagram of a coordinate system provided by an embodiment of this application.
  • 2C is a schematic diagram of a segmented shape provided by an embodiment of the present application.
  • 2D is a schematic diagram of a pixel diffusion provided by an embodiment of the present application.
  • Fig. 3 is a schematic flow chart of another method for screening abnormal cells provided by an embodiment of the application.
  • FIG. 4 is a schematic diagram of an abnormal cell screening device provided by an embodiment of the application.
  • FIG. 5 is a schematic diagram of a server structure of a hardware operating environment involved in an embodiment of the application.
  • the execution subject of the embodiment of the present application may be, for example, a server or a local data processing device.
  • the server may be, for example, a tablet computer, a notebook computer, a palmtop computer, a MID, a desktop computer, or other server devices. There are no restrictions in this application.
  • the acquiring an image of cervical cells includes: acquiring the image of cervical cells through a scanning device.
  • the scanning device may be a scanner, for example. It is understandable that a scanner is used to scan the cervical cell layer detected by liquid-based thin-layer cells to obtain cervical cell images.
  • the cervical cell image is an image of the cervical cell layer.
  • cervical cells are cells in a normal growth state.
  • the abnormal cell screening model includes a plurality of neural networks, the number of which is equal to the number of types of abnormal cells, and the plurality of neural networks are used to screen different abnormal cells.
  • the first neural network is one of the plurality of neural networks, and the first neural network is used to screen the first abnormality included in each cervical cell sub-image in the plurality of cervical cell sub-images Cell
  • the second neural network is another neural network different from the first neural network among the plurality of neural networks, and the second neural network is used to screen each of the plurality of cervical cell sub-images The second abnormal cell included in the cervical cell image.
  • the first abnormal cell is different from the second abnormal cell, the first abnormal cell is one kind of abnormal cells, and the second abnormal cell is another kind of abnormal cells.
  • the abnormal cells include at least one of the following: squamous cell carcinoma (SCC), high-grade squamous epithelial lesion (LSIL), and atypical high-grade squamous intraepithelial lesions cannot be ruled out Squamous cells (atypical squamous cells, cannot exclude high-gradesquamous intraepithelial lesion, ASC-H)), low-grade squamous epithelial lesions (LSIL), and atypical squamous cells of ambiguous significance ( atypical squamouscells of undP. termined significance, ASC-US), adenocarcinoma (AC), atypical adenocarcinoma (AGC), etc.
  • SCC squamous cell carcinoma
  • LSIL high-grade squamous epithelial lesion
  • ASC-H high-gradesquamous intraepithelial lesion
  • LSIL low-grade squamous epithelial les
  • multiple neural networks are respectively used to screen an abnormal cell included in each cervical cell sub-image in the multiple cervical cell sub-images. Specifically, one of the multiple neural networks is used to screen the squamous cell carcinoma included in each of the multiple cervical cell sub-images, and the other one of the multiple neural networks is used to filter multiple squamous cell carcinomas. Adenocarcinoma included in each of the cervical cell sub-images.
  • the preset selection strategy is determined according to a prediction selection operation, and the prediction selection operation includes the following steps: obtaining a prediction probability corresponding to each prediction result among the plurality of prediction results; The prediction probabilities corresponding to each prediction result of is numbered in descending order of the prediction probability to obtain multiple numbers; part numbers are selected from the multiple numbers, where the part numbers are the multiple numbers At least one number whose middle number is greater than the preset number; and the partial predicted probability corresponding to the partial numbers one-to-one is used as the preset selection strategy.
  • the preset number is set by the administrator. For example, if the multiple numbers are 10 numbers from 10-1, and the default number is 6, then the partial numbers are 3 numbers 7-10.
  • the selecting a plurality of prediction results from the plurality of prediction results as the plurality of first prediction results according to a preset selection strategy includes: selecting a plurality of prediction results from the plurality of prediction results according to the partial prediction probability The prediction result is used as a plurality of first prediction results.
  • FIG. 1B is a schematic diagram of a display interface provided by an embodiment of the application.
  • each rectangle with diagonal lines represents a first cervical cell image. It can be seen that multiple first cervical cell images are displayed on the display interface.
  • an image of cervical cells is acquired; the image of cervical cells is segmented based on the gray values of pixels in the image of cervical cells to obtain multiple images of cervical cells;
  • the sub-images are respectively input to the abnormal cell screening model to obtain multiple prediction results corresponding to the multiple cervical cell sub-images, wherein each cervical cell sub-image corresponds to a prediction result, and each prediction result is used to indicate each cervical cell Abnormal cells included in the sub-image, where the abnormal cells are cells that undergo pathological changes or cancerous changes on the basis of cervical cells; multiple prediction results are selected from the multiple prediction results as multiple first prediction results according to a preset selection strategy Acquire multiple first cervical cell sub-images corresponding to the multiple first prediction results; display the multiple first cervical cell sub-images on the display interface.
  • the abnormal cell screening model screens out the efficiency of abnormal cells.
  • the abnormal cell screening model includes a plurality of neural networks, the number of which is equal to the number of types of abnormal cells, and the plurality of neural networks are used to screen different abnormal cells, so that multiple types of abnormal cells can be screened. Abnormal cells are more practical.
  • an abnormal cell screening method provided in an embodiment of the present application may include:
  • the range of the gray value is 0-255.
  • the abnormal cells are different, and their corresponding gray values are also different.
  • the gray value corresponding to the squamous cell carcinoma is different from the gray value corresponding to the high-grade squamous epithelial lesion.
  • Figure 2B is a schematic diagram of a coordinate system provided by an embodiment of the application. It can be seen that the positive direction of the x-axis of the coordinate system is the right direction, that is, the transverse positive direction of the cervical cell image; The positive direction of the y-axis is the upward direction, that is, the positive longitudinal direction of the cervical cell image; the coordinate system is established on the cervical cell image based on the coordinate origin.
  • the segmenting the cervical cell image from the origin of the coordinates to obtain the plurality of cervical cell images includes: according to each of the cervical cell images The gray value corresponding to each pixel determines a plurality of gray difference values, and each gray difference value is the difference in gray value between each pixel and the corresponding adjacent pixel; The differences that fall within the same gray value interval are divided into one group to obtain multiple gray level groups; the multiple segments corresponding to the cervical cell image are determined from the coordinate origin according to the multiple gray level groups Shape; normalize the irregular shapes in the plurality of segmented shapes to obtain a plurality of regular first segmented shapes; set the plurality of first segmented shapes as the plurality of cervical cell sub-images .
  • the first pixel point is any pixel point in the cervical cell image
  • the plurality of pixels are determined according to the gray value corresponding to each pixel point in the cervical cell image.
  • the grayscale difference value includes: obtaining at least one grayscale value corresponding to at least one pixel point adjacent to the first pixel point; determining the grayscale value corresponding to the first pixel point and the at least one grayscale value respectively The grayscale difference between each grayscale value in to obtain at least one grayscale difference value.
  • the cervical cell image includes 9 pixels, and the 9 pixels are arranged in a square.
  • at least one pixel adjacent to the first pixel includes: The adjacent pixel above the pixel, the adjacent pixel below the first pixel, the adjacent pixel to the left of the first pixel, and the adjacent pixel to the right of the first pixel; those at the top right corner of the square
  • at least one pixel adjacent to the first pixel includes: the adjacent pixel below the first pixel, the adjacent pixel to the left of the first pixel, and the right of the first pixel.
  • the adjacent pixels of the square is the first pixel, at least one pixel adjacent to the first pixel includes: The adjacent pixel below the first pixel, the adjacent pixel to the left of the first pixel, and the right of the first pixel.
  • the at least one pixel point adjacent to the first pixel point includes at least one of the following: an adjacent pixel point above the first pixel point, an adjacent pixel point below the first pixel point, and a left side of the first pixel point And the adjacent pixel to the right of the first pixel.
  • At least one gray value corresponding to at least one pixel point adjacent to the first pixel is obtained; the gray value corresponding to the first pixel is determined to be the same as the at least one gray value.
  • the gray value difference between each gray value in the gray value to obtain at least one gray value difference to realize the determination of the gray value of the adjacent pixel point corresponding to the first pixel point and the first pixel point.
  • the difference value prepares for subsequent grouping according to the same gray value interval that the gray difference value falls into.
  • the multiple segmented shapes corresponding to the cervical cell image are determined from the coordinate origin based on the multiple gray scale groups, it is very likely that these segmented shapes are not regular shapes.
  • the regular shapes include, for example, rectangles, squares, and so on. Therefore, it is necessary to normalize the irregular shapes among the multiple segmented shapes to obtain the regular multiple first segmental shapes, so that the abnormal cell screening model can screen abnormal cells more quickly.
  • the irregular shape is a shape that does not include rectangles and squares
  • the normalization processing includes the following operations: determining a plurality of second pixel points corresponding to each of the plurality of segmented shapes, and each second pixel The gray value corresponding to the point is the first gray value; and the plurality of second pixel points corresponding to each of the plurality of segmented shapes are subjected to preset processing according to the first order.
  • the first sequence includes one of the following: an order in which the center point of each of the plurality of segmented shapes is closer to the origin of the coordinate, and each of the plurality of segmented shapes The order of the center point of the segmented shape from the farthest to the nearest to the origin of the coordinates.
  • the preset processing includes at least one of the following processing: removing redundant second pixel points in each segmented shape; collecting redundant second pixel points in the plurality of segmented shapes to obtain a plurality of third pixel points And using the plurality of third pixel points to fill the segmented shapes that need to be filled in the plurality of segmented shapes; and rearranging the plurality of second pixel points corresponding to each segmented shape.
  • the redundant second pixel point is determined according to the shape attribute corresponding to each segmented shape
  • the shape attribute corresponding to each segmented shape is determined according to the process of each segmented shape tending to a regular shape.
  • a certain segmented shape of the multiple segmented shapes is a triangle. See FIG. 2C.
  • FIG. 2C is a schematic diagram of a segmented shape provided by an embodiment of the present application.
  • the redundant second pixel points in the triangle are directly removed to obtain a rectangle, that is, the regular first segment shape; the example in the middle is to rearrange the redundant second pixels in the triangle first, and then use other segmentation shapes
  • the extra second pixel points in the middle fill in the triangles that have been arranged, and the final rectangle is formed, which is the regular first segment shape; the last legend is obtained by rearranging the extra second pixel points in the triangle Rectangle, that is, the regular first dichotomy shape.
  • the first gray value is 0 or 255.
  • a plurality of gray-scale differences are determined according to the gray-scale values corresponding to each pixel in the cervical cell image, and each gray-scale difference is each pixel corresponding to each pixel.
  • the gray level group determines the multiple segmented shapes corresponding to the cervical cell image from the origin of the coordinates; normalizes irregular shapes in the multiple segmented shapes to obtain multiple regular first segmented shapes Setting the multiple first segment shapes as the multiple cervical cell sub-images realizes the determination of the multiple segmented shapes corresponding to the cervical cell image from the origin of the coordinates according to the gray-scale group, so that each segmented shape It is composed of pixels that fall into the same gray value interval in the gray difference value, so that the abnormal cell screening model can screen out abnormal cells faster, and also speeds up the efficiency of image segmentation
  • the segmenting the cervical cell image based on the gray values of pixels in the cervical cell image to obtain the plurality of cervical cell sub-images includes: obtaining all the cervical cell images.
  • the pixel point includes at least one pixel point, and each pixel point in the at least one pixel point corresponds to the same gray value;
  • the cervical cell image is segmented according to the multiple sets of pixels to obtain the plurality of cervical cells Sub-images, the multiple sets of pixel points correspond to the multiple cervical cell sub-images.
  • the gray value corresponding to each pixel in the cervical cell image is obtained; the gray value in the cervical cell image is determined according to the gray value corresponding to each pixel in the cervical cell image.
  • Groups of pixels with the same degree value each group of pixel points includes at least one pixel point, and each pixel point in the at least one pixel point corresponds to the same gray value; and the plurality of groups of pixel points are divided into Cervical cell image to obtain the multiple cervical cell images, and the multiple sets of pixel points correspond to the multiple cervical cell images, so as to realize the segmentation of the cervical cell image according to the same gray value for subsequent abnormal cell screening
  • the model screens out abnormal cells faster for preparation.
  • the segmenting the cervical cell image according to the multiple sets of pixels to obtain the multiple cervical cell sub-images includes: cutting according to the multiple sets of pixels Divide the cervical cell image to obtain a plurality of second cervical cell images, and the plurality of sets of pixels correspond to the plurality of second cervical cell images; determine that the outer contour of the plurality of second cervical cell sub-images is not At least one regular third cervical cell image; performing the following operations for each third cervical cell image in the at least one third cervical cell image to obtain the plurality of cervical cell images, including: determining The outer contour size corresponding to the currently processed third cervical cell sub-image; obtaining a template image according to the outer contour size corresponding to the currently processed third cervical cell sub-image to obtain the outer contour size corresponding to the template image; according to the template The outer contour size corresponding to the image performs pixel diffusion on the currently processed third cervical cell sub-image, until the outer contour size corresponding to the currently processed third cervical cell sub-image is the same as the outer contour size corresponding to the template image
  • the irregular outer contour does not include the contours whose outer contours are rectangles and squares.
  • the template image includes one of the following: rectangle and square.
  • the first gray value is 0 or 255.
  • Figure 2D is a schematic diagram of pixel diffusion provided by an embodiment of the present application.
  • the template image corresponding to the second cervical cell sub-image is a rectangle
  • the triangle is performed according to the outer contour size corresponding to the rectangle.
  • the pixels are diffused, and finally, the cervical cell sub-image corresponding to the second cervical cell image is a rectangle.
  • the cervical cell image is segmented according to the multiple sets of pixels to obtain multiple second cervical cell sub-images, and the multiple sets of pixels and the multiple second cervical cell sub-images Image correspondence; determine at least one third cervical cell image with an irregular outer contour in the plurality of second cervical cell images; execute for each third cervical cell image in the at least one third cervical cell image
  • the following operations to obtain the plurality of cervical cell images include: determining the outer contour size corresponding to the currently processed third cervical cell sub-image; obtaining a template image according to the outer contour size corresponding to the currently processed third cervical cell sub-image , To obtain the outer contour size corresponding to the template image; according to the outer contour size corresponding to the template image, perform pixel diffusion on the currently processed third cervical cell sub-image until the currently processed third cervical cell sub-image corresponds to the outer contour
  • pixel diffusion is stopped.
  • the pixel diffusion is diffused by using the first gray value, and pixel diffusion is performed by at least one third cervical cell sub-image with an irregular outer contour , So that all cervical cell images become rectangular or square images, so that the subsequent abnormal cell screening model can quickly screen out abnormal cells.
  • Fig. 3 is a schematic flow chart of a method for screening abnormal cells according to another embodiment of the application.
  • the method further includes :
  • each training subset includes a plurality of cervical cell sub-image sets with different degrees of brightness
  • Each set of cervical cell sub-images includes a plurality of second cervical cell sub-images with a degree of brightness, and each second cervical cell sub-image of the plurality of second cervical cell sub-images includes different abnormal cells;
  • the acquiring the training set includes: displaying the multiple third cervical cell sub-images on a marking interface; During the marking operation, mark the multiple third cervical cell sub-images corresponding to the multiple positions to obtain multiple fourth cervical cell sub-images corresponding to the multiple third cervical cell sub-images, each The fourth cervical cell image is an image after marking each third cervical cell image; the multiple fourth cervical cell images are respectively processed with multiple preset brightnesses to obtain each of the multiple training subsets.
  • the plurality of cervical cell sub-image sets included in each training subset, and the plurality of preset brightnesses correspond to the plurality of cervical cell sub-image sets; each training subset included in the plurality of training subsets The plurality of cervical cell sub-image sets are set as the training set.
  • the marking interface includes a plurality of marking display areas, and the plurality of marking display areas correspond to the plurality of third cervical cell sub-images.
  • the displaying the multiple third cervical cell images on the marking interface includes: displaying the multiple third cervical cell images on the multiple marking display areas on the marking interface. Further, a person with medical knowledge can view multiple third cervical cell sub-images in multiple mark display areas, and mark the multiple third cervical cell sub-images.
  • marking the plurality of third cervical cells corresponding to the plurality of positions Image to obtain multiple fourth cervical cell sub-images corresponding to the multiple third cervical cell images, including: when a marking operation for multiple positions on the marking interface is detected, tracking the multiple Multiple labeled trajectories on the multiple third cervical cell sub-images corresponding to each position; acquire multiple labeled regions formed by the multiple labeled trajectories; generate multiple labeled labels according to the multiple labeled regions; The multiple third cervical cell sub-images corresponding to the multiple positions generate the multiple fourth cervical cell sub-images including the multiple mark tags.
  • the first position is any one of the multiple positions
  • the third cervical cell corresponding to the first position is tracked
  • the marking track on the image acquiring the marking area formed by the marking track; generating a marking label according to the marking area; generating a fourth cervical cell including the marking label according to the third cervical cell sub-image corresponding to the first position Sub-image.
  • the plurality of third cervical cell sub-images correspond to the plurality of marking trajectories
  • the plurality of marking trajectories correspond to the plurality of marking regions
  • the plurality of marking regions correspond to the plurality of marking labels correspond.
  • the plurality of third cervical cell sub-images corresponding to the plurality of positions are tracked.
  • Marking trajectories; acquiring multiple marking areas formed by the multiple marking trajectories; generating multiple marking labels according to the multiple marking areas; generating the multiple third cervical cell sub-images corresponding to the multiple positions includes
  • the multiple fourth cervical cell images of the multiple marking labels realize that when the marking trajectories are different, the marked areas are also different, which in turn leads to different marking labels, which improves the uniqueness of the fourth cervical cell image marking labels. It improves the efficiency of model training.
  • multiple third cervical cell sub-images are displayed on the marking interface; when a marking operation for multiple positions on the marking interface is detected, all corresponding to the multiple positions are marked.
  • the image after the image is marked; the plurality of fourth cervical cell sub-images are respectively processed by using a plurality of preset brightnesses to obtain the plurality of cervical cell sub-images included in each training subset in a plurality of training subsets
  • the plurality of preset brightnesses correspond to the plurality of cervical cell sub-image sets; the plurality of cervical cell sub-image sets included in each training subset of the plurality of training subsets are set as the training Set, realize that after marking multiple third cervical cell sub-images on the marking interface, use multiple preset brightness to adjust the brightness of the multiple third cervical cell sub-images after marking to obtain the training set,
  • the multiple neural networks to be trained run on multiple processes respectively, and the multiple processes correspond to the multiple neural networks to be trained.
  • the training set includes multiple training subsets, the multiple training subsets correspond to the multiple neural networks, and each training subset includes a different degree of brightness
  • each cervical cell image set includes a plurality of second cervical cell images with a degree of brightness, and each second cervical cell image in the plurality of second cervical cell images
  • the image includes different abnormal cells; constructing multiple neural networks to be trained, the multiple neural networks to be trained corresponding to the multiple neural networks; training the training set based on the multiple neural networks to be trained,
  • the abnormal cell screening model can recognize cervical cell images taken under different ambient light, so as to optimize the abnormal cell screening model and enhance the practicability of the abnormal cell screening model.
  • an abnormal cell screening device 400 provided by an embodiment of the present application may include:
  • the first acquiring module 401 is used to acquire cervical cell images
  • the first acquiring module 401 is configured to acquire the cervical cell image through a scanning device.
  • the scanning device may be a scanner, for example. It is understandable that a scanner is used to scan the cervical cell layer detected by liquid-based thin-layer cells to obtain cervical cell images.
  • the cervical cell image is an image of the cervical cell layer.
  • a segmentation module 402 configured to segment the cervical cell image based on the gray values of pixels in the cervical cell image to obtain multiple cervical cell sub-images;
  • the segmentation module 402 is specifically configured to obtain the cervical cell image The gray value corresponding to each pixel in the cervical cell image; determine the pixel with the smallest gray value according to the gray value corresponding to each pixel in the cervical cell image; select any pixel from the pixel as a segmentation
  • the coordinate origin of the cervical cell image establishing a coordinate system on the cervical cell image based on the coordinate origin, wherein the coordinate system takes the lateral positive direction of the cervical cell image as the x-axis, and the The longitudinal positive direction of the cervical cell image is the y-axis; the cervical cell image is segmented from the origin of the coordinates to obtain the multiple cervical cell images.
  • the segmentation module 402 is specifically configured to correspond to each pixel in the cervical cell image.
  • Determine multiple gray-scale difference values each gray-scale difference value is the difference in gray-scale value between each pixel and the corresponding adjacent pixel; drop the multiple gray-scale differences Divide the same gray value interval into one group to obtain multiple gray level groups; determine the multiple segmented shapes corresponding to the cervical cell image based on the multiple gray level groups from the origin of the coordinates;
  • the irregular shapes in the plurality of segmented shapes are standardized to obtain a plurality of regular first segmented shapes; the plurality of first segmented shapes are set as the plurality of cervical cell images.
  • the segmentation module 402 is specifically configured to obtain the cervical cell image The gray value corresponding to each pixel in the cervical cell image; determine multiple groups of pixels with the same gray value in the cervical cell image according to the gray value corresponding to each pixel in the cervical cell image, and each group of pixels includes at least One pixel point, the gray value corresponding to each pixel point in the at least one pixel point is the same; the cervical cell image is segmented according to the multiple sets of pixels to obtain the multiple cervical cell sub-images, so The multiple sets of pixel points correspond to the multiple cervical cell images.
  • the segmentation module 402 is specifically configured to segment the cervical cell images according to the multiple sets of pixels.
  • the cervical cell image to obtain a plurality of second cervical cell images, and the plurality of sets of pixel points correspond to the plurality of second cervical cell images; determining that the outer contour of the plurality of second cervical cell sub-images is irregular At least one third cervical cell image; performing the following operations for each third cervical cell image in the at least one third cervical cell image to obtain the plurality of cervical cell images, including: determining the current process The outer contour size corresponding to the third cervical cell sub-image; obtain the template image according to the outer contour size corresponding to the currently processed third cervical cell sub-image to obtain the outer contour size corresponding to the template image; corresponding to the template image Pixel diffusion is performed on the currently processed third cervical cell sub-image until the current processed third cervical cell sub-image corresponds to the same outer contour size as the template image. The pixel diffusion is stopped, so
  • the input module 403 is configured to input the multiple cervical cell sub-images into the abnormal cell screening model to obtain multiple prediction results corresponding to the multiple cervical cell sub-images, wherein each cervical cell sub-image corresponds to a prediction As a result, each prediction result is used to indicate the abnormal cells included in each cervical cell sub-image, and the abnormal cells are cells that undergo lesions or cancerous changes on the basis of cervical cells;
  • cervical cells are cells in a normal growth state.
  • the abnormal cell screening model includes a plurality of neural networks, the number of which is equal to the number of types of abnormal cells, and the plurality of neural networks are used to screen different abnormal cells.
  • the first neural network is one of the plurality of neural networks, and the first neural network is used to screen the first abnormality included in each cervical cell sub-image in the plurality of cervical cell sub-images Cell
  • the second neural network is another neural network different from the first neural network among the plurality of neural networks, and the second neural network is used to screen each of the plurality of cervical cell sub-images The second abnormal cell included in the cervical cell image.
  • the first abnormal cell is different from the second abnormal cell, the first abnormal cell is one kind of abnormal cells, and the second abnormal cell is another kind of abnormal cells.
  • the abnormal cells include at least one of the following: squamous cell carcinoma (SCC), high-grade squamous epithelial lesion (LSIL), and atypical high-grade squamous intraepithelial lesions cannot be ruled out Squamous cells (atypical squamous cells, cannot exclude high-gradesquamous intraepithelial lesion, ASC-H)), low-grade squamous epithelial lesions (LSIL), and atypical squamous cells of ambiguous significance ( atypical squamouscells of undP. termined significance, ASC-US), adenocarcinoma (AC), atypical adenocarcinoma (AGC), etc.
  • SCC squamous cell carcinoma
  • LSIL high-grade squamous epithelial lesion
  • ASC-H high-gradesquamous intraepithelial lesion
  • LSIL low-grade squamous epithelial les
  • multiple neural networks are respectively used to screen an abnormal cell included in each cervical cell sub-image in the multiple cervical cell sub-images. Specifically, one of the multiple neural networks is used to screen the squamous cell carcinoma included in each of the multiple cervical cell sub-images, and the other one of the multiple neural networks is used to filter multiple squamous cell carcinomas. Adenocarcinoma included in each of the cervical cell sub-images.
  • the abnormal cell screening device further includes processing Module, the processing module is used to obtain a training set, the training set includes a plurality of training subsets, the plurality of training subsets correspond to the plurality of neural networks, each training subset includes a different degree of brightness
  • the processing module is used to obtain a training set, the training set includes a plurality of training subsets, the plurality of training subsets correspond to the plurality of neural networks, each training subset includes a different degree of brightness
  • a plurality of cervical cell image sets each cervical cell image set includes a plurality of second cervical cell images with a degree of brightness, and each second cervical cell image in the plurality of second cervical cell images
  • the abnormal cells included are different; a plurality of neural networks to be trained are constructed, and the plurality of neural networks to be trained corresponds to the plurality of neural networks; the training set is trained based on the plurality of neural networks to be trained to Obtain the abnormal cell screening model.
  • the processing module is specifically configured to display the plurality of third cervical cell sub-images on a marking interface; when the markings for multiple positions on the marking interface are detected During operation, the multiple third cervical cell images corresponding to the multiple positions are marked to obtain multiple fourth cervical cell images corresponding to the multiple third cervical cell images, and each fourth cervix
  • the cell sub-image is an image after marking each third cervical cell sub-image; the multiple fourth cervical cell sub-images are respectively processed with multiple preset brightnesses to obtain each training in multiple training subsets
  • the plurality of cervical cell sub-image sets included in the subset, and the plurality of preset brightnesses correspond to the plurality of cervical cell sub-image sets; the plurality of training subsets included in each training subset
  • a plurality of cervical cell sub-image sets are set as the training set.
  • the selection module 404 is configured to select multiple prediction results from the multiple prediction results as multiple first prediction results according to a preset selection strategy
  • the preset selection strategy is determined according to a prediction selection operation, and the prediction selection operation includes the following steps: obtaining a prediction probability corresponding to each prediction result among the plurality of prediction results; The prediction probabilities corresponding to each prediction result of is numbered in descending order of the prediction probability to obtain multiple numbers; part numbers are selected from the multiple numbers, where the part numbers are the multiple numbers At least one number whose middle number is greater than the preset number; and the partial predicted probability corresponding to the partial numbers one-to-one is used as the preset selection strategy.
  • the preset number is set by the administrator. For example, if the multiple numbers are 10 numbers from 10-1, and the default number is 6, then the partial numbers are 3 numbers 7-10.
  • the selecting a plurality of prediction results from the plurality of prediction results as the plurality of first prediction results according to a preset selection strategy includes: selecting a plurality of prediction results from the plurality of prediction results according to the partial prediction probability The prediction result is used as a plurality of first prediction results.
  • the second acquiring module 405 is configured to acquire multiple first cervical cell sub-images corresponding to the multiple first prediction results
  • the display module 406 is configured to display the multiple first cervical cell sub-images on the display interface
  • FIG. 5 is a schematic diagram of a server structure of a hardware operating environment involved in an embodiment of the application.
  • An embodiment of the application provides an electronic device for pushing information, including a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured by The processor executes to execute instructions including steps in any information push method.
  • the server of the hardware operating environment involved in the embodiment of the present application may include:
  • the processor 501 is, for example, a CPU.
  • the memory 502 optionally, the memory may be a high-speed RAM memory, or a stable memory, such as a disk memory.
  • the communication interface 503 is used to implement connection and communication between the processor 501 and the memory 502.
  • FIG. 5 does not constitute a limitation to it, and may include more or less components than those shown in the figure, or a combination of certain components, or different component arrangements.
  • the memory 502 may include an operating system, a network communication module, and one or more programs.
  • the operating system is a program that manages and controls server hardware and software resources, one or more programs, and the operation of other software or programs.
  • the network communication module is used to implement communication between various components in the storage 502 and communication with other hardware and software in the server.
  • the processor 501 is configured to execute the personnel management program stored in the memory 502, and implement the following steps:
  • the multiple cervical cell sub-images are respectively input to the abnormal cell screening model to obtain multiple prediction results corresponding to the multiple cervical cell sub-images, wherein each cervical cell sub-image corresponds to a prediction result, and each prediction result It is used to indicate the abnormal cells included in each cervical cell image, and the abnormal cells are cells that have undergone lesions or cancerous changes on the basis of cervical cells;
  • the abnormal cell screening model includes a plurality of neural networks, the number of which is equal to the number of types of abnormal cells, and the plurality of neural networks are used to screen different abnormal cells.
  • the server involved in the present application please refer to each embodiment of the above abnormal cell screening method, which will not be repeated here.
  • This application also provides an electronic device for screening abnormal cells, including a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and are generated by
  • the processor executes instructions to execute the following steps: acquiring a cervical cell image; segmenting the cervical cell image based on the gray values of pixels in the cervical cell image to obtain a plurality of cervical cell images;
  • the multiple cervical cell sub-images are respectively input to the abnormal cell screening model to obtain multiple prediction results corresponding to the multiple cervical cell sub-images, wherein each cervical cell sub-image corresponds to a prediction result, and each prediction result is used for Indicate the abnormal cells included in each cervical cell sub-image, where the abnormal cells are cells that undergo pathological changes or cancerous changes on the basis of cervical cells; select multiple prediction results from the multiple prediction results according to a preset selection strategy.
  • a first prediction result acquiring a plurality of first cervical cell sub-images corresponding to the plurality of first prediction results; displaying the plurality of first cervical cell sub-images on a display interface; wherein the abnormal cell screening model
  • a plurality of neural networks are included, the number of the neural networks is equal to the number of the types of abnormal cells, and the plurality of neural networks are used to screen different abnormal cells.
  • the processor is configured to obtain each cervical cell image.
  • the gray value corresponding to each pixel determine the pixel with the smallest gray value according to the gray value corresponding to each pixel in the cervical cell image; select any pixel from the pixel as the segmentation
  • the coordinate origin of the cervical cell image a coordinate system is established on the cervical cell image based on the coordinate origin, wherein the coordinate system takes the horizontal positive direction of the cervical cell image as the x-axis, and the cervical cell image
  • the positive longitudinal direction of the image is the y-axis; the cervical cell image is segmented from the origin of the coordinates to obtain the multiple cervical cell sub-images.
  • the processor is configured to calculate a gray scale corresponding to each pixel in the cervical cell image.
  • the degree value determines a plurality of gray-scale differences, and each gray-scale difference is the difference in gray-scale value between each pixel and the corresponding neighboring pixel; and the multiple gray-scale differences fall into the same
  • the gray value interval is divided into one group to obtain multiple gray level groups; the multiple segmented shapes corresponding to the cervical cell image are determined from the coordinate origin according to the multiple gray level groups;
  • the irregular shapes in the three segmented shapes are standardized to obtain a plurality of regular first segmental shapes; and the plurality of first segmented shapes are set as the plurality of cervical cell sub-images.
  • the processor is configured to obtain each cervical cell image.
  • a gray value corresponding to each pixel determining multiple groups of pixels with the same gray value in the cervical cell image according to the gray value corresponding to each pixel in the cervical cell image, and each group of pixels includes at least one pixel Point, the gray value corresponding to each pixel point in the at least one pixel point is the same; the cervical cell image is segmented according to the multiple sets of pixels to obtain the multiple cervical cell sub-images, and the multiple The group of pixels corresponds to the plurality of cervical cell sub-images.
  • the processor is configured to segment the cervix according to the multiple sets of pixels Cell image to obtain a plurality of second cervical cell images, and the plurality of sets of pixels correspond to the plurality of second cervical cell images; determining at least one irregular outer contour of the plurality of second cervical cell images A third cervical cell image; performing the following operations for each third cervical cell image in the at least one third cervical cell image to obtain the plurality of cervical cell images, including: determining the currently processed third cervical cell image The outer contour size corresponding to the third cervical cell sub-image; the template image is obtained according to the outer contour size corresponding to the currently processed third cervical cell sub-image to obtain the outer contour size corresponding to the template image; according to the outer contour size corresponding to the template image The contour size performs pixel diffusion on the currently processed third cervical cell image until the outer contour size corresponding to the currently processed third cervical cell image is the same as the outer contour size corresponding to the template image, and the pixel diffusion
  • the processor is further configured to obtain A training set, the training set includes a plurality of training subsets, the plurality of training subsets correspond to the plurality of neural networks, each training subset includes a plurality of cervical cell sub-image sets with different degrees of brightness, each The set of cervical cell sub-images includes a plurality of second cervical cell images with a degree of brightness, and each of the plurality of second cervical cell sub-images includes different abnormal cells; constructing a plurality of waiting A trained neural network, where the multiple neural networks to be trained correspond to the multiple neural networks; and the training set is trained based on the multiple neural networks to be trained to obtain the abnormal cell screening model.
  • the processor when the training set is acquired, is configured to display the multiple third cervical cell sub-images on a marking interface; when a marking operation for multiple positions on the marking interface is detected , Marking the multiple third cervical cell images corresponding to the multiple positions to obtain multiple fourth cervical cell images corresponding to the multiple third cervical cell images, and each fourth cervical cell image
  • the image is an image after marking each third cervical cell sub-image; the multiple fourth cervical cell sub-images are respectively processed with multiple preset brightnesses to obtain each training subset in the multiple training subsets
  • the plurality of cervical cell sub-image sets included, and the plurality of preset brightnesses correspond to the plurality of cervical cell sub-image sets; the plurality of training subsets included in each training subset
  • the cervical cell sub-image set is set as the training set.
  • the first neural network is one of the plurality of neural networks, and the first neural network is used to filter the first cervical cell sub-image included in each of the plurality of cervical cell sub-images.
  • the second neural network is another neural network different from the first neural network among the plurality of neural networks, and the second neural network is used to screen each of the plurality of cervical cell sub-images A second abnormal cell included in the cervical cell image.
  • the first abnormal cell is different from the second abnormal cell
  • the first abnormal cell is one kind of abnormal cells
  • the second abnormal cell is another kind of abnormal cells.
  • the preset selection strategy is determined according to a prediction selection operation, and the prediction selection operation includes the following steps:
  • the prediction probability corresponding to each prediction result in the plurality of prediction results Obtain the prediction probability corresponding to each prediction result in the plurality of prediction results; number the prediction probability corresponding to each prediction result in the plurality of prediction results in descending order of the prediction probability to obtain the number A number; a partial number is selected from the plurality of numbers, where the part number is at least one number of the plurality of numbers that is greater than a preset number; the partial predicted probability corresponding to the part number one-to-one is taken as The preset selection strategy.
  • the present application also provides a computer-readable storage medium, the computer-readable storage medium is used to store a computer program, and the stored computer program is executed by the processor to implement the following steps:
  • each cervical cell sub-image corresponds to a prediction result, and each prediction result is used to indicate abnormal cells included in each cervical cell sub-image, and Abnormal cells are cells that undergo lesions or cancerous changes on the basis of cervical cells; select multiple prediction results from the multiple prediction results as multiple first prediction results according to a preset selection strategy; obtain the multiple first predictions Multiple first cervical cell sub-images corresponding to the result; displaying the multiple first cervical cell sub-images on the display interface;
  • the abnormal cell screening model includes a plurality of neural networks, the number of which is equal to the number of types of abnormal cells, and the plurality of neural networks are used to screen different abnormal cells.
  • the computer-readable storage medium may be non-volatile or volatile.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Quality & Reliability (AREA)
  • Epidemiology (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Pathology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Investigating Or Analysing Biological Materials (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

本申请涉及神经网络，公开了一种异常细胞筛选方法、装置、电子设备和存储介质，所述方法包括：获取宫颈细胞图像；基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果；根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；在显示界面上显示所述多个第一宫颈细胞子图像。实施本申请实施例，有利于提高异常细胞筛选效率。

Description

一种异常细胞筛选方法、装置、电子设备和存储介质
本申请要求于 2019年 10月 29日提交中国专利局、申请号为 2019110409559、发明名称为 “一种异常细胞筛选方法、装置、电子设备和存储介质”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
本申请涉及神经网络技术领域，尤其涉及一种异常细胞筛选方法、装置、电子设备和存储介质。
背景技术
宫颈癌是最常见的妇科恶性肿瘤，且近年来其发病率有低龄化趋势，全世界每年有50万新发病例和27.4万死亡病例，其中85％的宫颈癌死亡病例发生在普查率低的低中等收入地区。宫颈癌又是目前唯一可以早发现并治愈的癌症，因此早期的筛查和诊断是防治宫颈癌的关键环节。
目前，在各大体检中心和医院，发明人发现一般需要医生在显微镜下从成千上万个细胞中找出异常细胞，并依据异常细胞进行诊断。这种异常细胞筛选方式效率低。
发明内容
本申请实施例提供了一种异常细胞筛选方法、装置、电子设备和存储介质，实施本申请实施例，有利于提高异常细胞筛选效率。
本申请第一方面提供了一种异常细胞筛选方法，包括：
获取宫颈细胞图像；
基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；
将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；
根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；
获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；
在显示界面上显示所述多个第一宫颈细胞子图像；
其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
本申请第二方面提供了一种异常细胞筛选装置，包括：
第一获取模块，用于获取宫颈细胞图像；
切分模块，用于基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；
输入模块，用于将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；
选取模块，用于根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；
第二获取模块，用于获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；
显示模块，用于在显示界面上显示所述多个第一宫颈细胞子图像；
其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
本申请第三方面提供了一种异常细胞筛选的电子设备，包括处理器、存储器、通信接口以及一个或多个程序，其中，所述一个或多个程序被存储在所述存储器中，并且被生成由所述处理器执行，以执行一种异常细胞筛选方法任一项方法中的步骤的指令。
本申请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用于存储计算机程序，所述存储计算机程序被所述处理器执行，以实现一种异常细胞筛选方法任一项所述的方法。
可以看出，上述技术方案中，通过将宫颈细胞图像切分成多个宫颈细胞子图像，并将多个宫颈细胞子图像分别输入异常细胞筛选模型，避免了一张宫颈细胞图像过大导致筛选过程时间耗费多的情况，加快了异常细胞筛选模型筛选出异常细胞的效率。同时，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞，实现可以筛选出多种异常细胞，实用性更强。进一步的，通过将预测概率最高的多个第一预测结果筛选出来，并将多个第一预测结果对应的多个第一宫颈细胞子图像显示在界面上，更加便于医生查看，通过将出现异常细胞概率最高的宫颈细胞子图像展示给医生，减少了医生的工作量。
附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍。
其中：
图1A为本申请实施例提供的一种异常细胞筛选方法的流程示意图；
图1B为本申请实施例提供的一种显示界面的示意图；
图2A为本申请实施例提供的又一种异常细胞筛选方法的流程示意图；
图2B为本申请的实施例提供的一种坐标系示意图；
图2C是本申请实施例提供的一种切分形状的示意图；
图2D是本申请实施例提供的一种像素扩散的示意图；
图3为本申请实施例提供的又一种异常细胞筛选方法的流程示意图；
图4为本申请实施例提供的一种异常细胞筛选装置的示意图；
图5为本申请的实施例涉及的硬件运行环境的服务器结构示意图。
具体实施方式
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。
以下分别进行详细说明。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”是用于区别不同对象，而不是用于描述特定顺序。此外，术语“包括”和“具有”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、系统、产品或设备没有限定于已列出的步骤或单元，而是可选地还包括没有列出的步骤或单元，或可选地还包括对于这些过程、方法、产品或设备固有的其它步骤或单元。
首先，本申请实施例的执行主体例如可以是服务器，也可以是本地数据处理设备。其中，服务器例如可以是平板电脑、笔记本电脑、掌上电脑、MID、台式电脑或其他服务器设备。在本申请中不做限制。
101、获取宫颈细胞图像；
可选的，所述获取宫颈细胞图像，包括：通过扫描设备获取所述宫颈细胞图像。
其中，扫描设备例如可以是扫描仪。可以理解的，采用扫描仪扫描通过液基薄层细胞 检测的宫颈细胞层以得到宫颈细胞图像。
其中，宫颈细胞图像为宫颈细胞层的图像。
102、基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；
103、将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；
其中，宫颈细胞为处于正常生长状态的细胞。
其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
进一步的，第一神经网络是所述多个神经网络中的一个神经网络，所述第一神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第一异常细胞，第二神经网络是所述多个神经网络中的不同于所述第一神经网络的另一个神经网络，所述第二神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第二异常细胞。
其中，第一异常细胞与第二异常细胞不同，第一异常细胞是所述异常细胞中的一种细胞，第二异常细胞是所述异常细胞中的另一种细胞。
可选的，异常细胞至少包括以下一种：鳞癌(squamous cell carcinom，SCC)、高度鳞状上皮病变(high-grade squamous epithelial lesion，LSIL)、不能排除高级别鳞状上皮内病变的非典型鳞状细胞(atypical squamous cells,cannot exclude high-gradesquamous intraepithelial lesion，ASC-H)))、低度鳞状上皮病变(low-grade squamous epithelial lesion，LSIL)、意义不明确的非典型鳞状细胞(atypical squamouscells of undP.termined significance,ASC-US)、腺癌(adenocarcinoma，AC)、非典型腺癌(Atypical adenocarcinoma，AGC)等。
可以理解的，多个神经网络分别用于筛选多个宫颈细胞子图像中的每个宫颈细胞子图像包括的一种异常细胞。具体来说，多个神经网络中的某个神经网络用于筛选多个宫颈细胞子图像中的每个宫颈细胞子图像包括的鳞癌，多个神经网络中的另一个神经网络用于筛选多个宫颈细胞子图像中的每个宫颈细胞子图像包括的腺癌。
104、根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；
其中，所述预设选取策略是根据预测选取操作确定，所述预测选取操作包括以下步骤：获取所述多个预测结果中的每个预测结果对应的预测概率；将所述多个预测结果中的每个预测结果对应的预测概率按照预测概率从大到小的顺序进行编号，以得到多个编号；从所述多个编号中选取部分编号，其中，所述部分编号为所述多个编号中编号大于预设编号的至少一个编号；将所述部分编号一一对应的部分预测概率作为所述预设选取策略。
其中，预设编号由管理员设置。举例来说，多个编号为从10-1的10个编号，预设编号为6，那么部分编号为7-10这3个编号。
进一步的，所述根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果，包括：根据所述部分预测概率从所述多个预测结果中选取多个预测结果作为多个第一预测结果。
105、获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；
106、在显示界面上显示所述多个第一宫颈细胞子图像。
参见图1B，图1B为本申请实施例提供的一种显示界面的示意图，在该显示界面上，每个带斜线的矩形代表一个第一宫颈细胞子图像。可以看出，在该显示界面上显示了多个 第一宫颈细胞子图像。
可以看出，上述技术方案中，获取宫颈细胞图像；基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；在显示界面上显示所述多个第一宫颈细胞子图像。通过将宫颈细胞图像切分成多个宫颈细胞子图像，并将多个宫颈细胞子图像分别输入异常细胞筛选模型，避免了一张宫颈细胞图像过大导致筛选过程时间耗费多的情况，加快了异常细胞筛选模型筛选出异常细胞的效率。同时，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞，实现可以筛选出多种异常细胞，实用性更强。进一步的，通过将预测概率最高的多个第一预测结果筛选出来，并将多个第一预测结果对应的多个第一宫颈细胞子图像显示在界面上，更加便于医生查看，通过将出现异常细胞概率最高的宫颈细胞子图像展示给医生，减少了医生的工作量。
下面对从所述基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像的这一过程进行具体举例说明。
参见图2A，图2A为本申请的一个实施例提供的一种异常细胞筛选方法的流程示意图。其中，如图2A所示，本申请的一个实施例提供的一种异常细胞筛选方法可以包括：
201、获取所述宫颈细胞图像中每个像素点对应的灰度值；
其中，该灰度值的范围在0-255。
当宫颈细胞图像中包括异常细胞时，异常细胞不同，其对应的灰度值也不同。举例来说，当宫颈细胞图像包括鳞癌和高度鳞状上皮病变时，鳞癌对应的灰度值与高度鳞状上皮病变对应的灰度值不同。
202、根据所述宫颈细胞图像中每个像素点对应的灰度值确定灰度值最小的像素点；
203、从所述像素点中选取任意一个像素点作为切分所述宫颈细胞图像的坐标原点；
204、以所述坐标原点为基础在所述宫颈细胞图像上建立坐标系，其中，所述坐标系以所述宫颈细胞图像的横向正方向为x轴，以所述宫颈细胞图像的纵向正方向为y轴；
参见图2B，图2B为本申请的一个实施例提供的一种坐标系示意图，可以看出，该坐标系x轴的正向方向为向右方向，即宫颈细胞图像的横向正方向；该坐标系y轴的正向方向为向上方向，即宫颈细胞图像的纵向正方向；该坐标系是以坐标原点为基础在宫颈细胞图像上建立的。
205、从所述坐标原点出发切分所述宫颈细胞图像，以得到所述多个宫颈细胞子图像。
可选的，在一种可能的实施方式中，所述从所述坐标原点出发切分所述宫颈细胞图像，以得到所述多个宫颈细胞子图像，包括：根据所述宫颈细胞图像中每个像素点对应的灰度值确定多个灰度差值，每个灰度差值为每个像素点与对应的相邻像素点在灰度值上的差值；将所述多个灰度差值中落入相同灰度值区间的划分为一组，以得到多个灰度组；根据所述多个灰度组从所述坐标原点出发确定所述宫颈细胞图像对应的多个切分形状；将所述多个切分形状中非规则形状进行规范化处理，以得到规则的多个第一切分形状；将所述多个第一切分形状设置为所述多个宫颈细胞子图像。
可选的，在一种可能的实施方式中，第一像素点为所述宫颈细胞图像中任意一个像素点，所述根据所述宫颈细胞图像中每个像素点对应的灰度值确定多个灰度差值，包括：获取所述第一像素点相邻的至少一个像素点对应的至少一个灰度值；分别确定所述第一像素 点对应的灰度值与所述至少一个灰度值中的每个灰度值之间的灰度差值，以得到至少一个灰度差值。
举例来说，宫颈细胞图像包括9个像素点，这9个像素点排列成正方形，处于正方形中心的像素点为第一像素点时，第一像素点相邻的至少一个像素点包括：第一像素点上方的相邻像素点、第一像素点下方的相邻像素点、第一像素点左方的相邻像素点、第一像素点右方的相邻像素点；处于正方形右上角顶点的像素点为第一像素点时，第一像素点相邻的至少一个像素点包括：第一像素点下方的相邻像素点、第一像素点左方的相邻像素点、第一像素点右方的相邻像素点。
进一步的，所述第一像素点相邻的至少一个像素点至少包括以下一种：第一像素点上方的相邻像素点、第一像素点下方的相邻像素点、第一像素点左方的相邻像素点和第一像素点右方的相邻像素点。
可以看出，上述技术方案中，通过获取所述第一像素点相邻的至少一个像素点对应的至少一个灰度值；分别确定所述第一像素点对应的灰度值与所述至少一个灰度值中的每个灰度值之间的灰度差值，以得到至少一个灰度差值，实现确定第一像素点与第一像素点对应的相邻像素点在灰度值上的差值，为后续根据灰度差值落入的相同灰度值区间进行分组做准备。
可以理解的，根据所述多个灰度组从所述坐标原点出发确定所述宫颈细胞图像对应的多个切分形状时，很有可能这些切分形状不是规则形状。其中，规则形状例如包括：矩形、正方形等等。因此，需要将所述多个切分形状中非规则形状进行规范化处理，以得到规则的多个第一切分形状，从而让异常细胞筛选模型更快的筛选出异常细胞。
其中，非规则形状为不包括矩形和正方形的形状，所述规范化处理包括以下操作：确定所述多个切分形状中每个切分形状对应的多个第二像素点，每个第二像素点对应的灰度值为第一灰度值；按照第一顺序对所述多个切分形状中每个切分形状对应的多个第二像素点进行预设处理。
其中，所述第一顺序包括以下一种：所述多个切分形状中每个切分形状的中心点离所述坐标原点从近到远的顺序和所述多个切分形状中每个切分形状的中心点离所述坐标原点从远到近的顺序。
进一步的，所述预设处理包括以下至少一种处理：去除每个切分形状中多余第二像素点；收集所述多个切分形状中多余第二像素点以得到多个第三像素点以及采用所述多个第三像素点对所述多个切分形状中需要进行填补的切分形状进行填补；和将每个切分形状对应的多个第二像素点进行重新排布。其中，多余第二像素点是根据每个切分形状对应的形状属性确定的，每个切分形状对应的形状属性是根据每个切分形状趋于规则形状的这个过程中确定。
举例来说，所述多个切分形状中某个切分形状是三角形，参见图2C，图2C是本申请实施例提供的一种切分形状的示意图，其中，最上面的图例是直接将三角形中多余的第二像素点直接去掉，从而得到矩形，也就是规则的第一切分形状；中间的图例是先将三角形中多余的第二像素点进行重新排布，再采用其他切分形状中多余第二像素点对已经排布好的三角形进行填补，最后形成的矩形，也就是规则的第一切分形状；最后一个图例是将三角形中多余的第二像素点进行重新排布得到的矩形，也就是规则的第一切分形状。
其中，第一灰度值为0或255。
可选的，上述技术方案中，根据所述宫颈细胞图像中每个像素点对应的灰度值确定多个灰度差值，每个灰度差值为每个像素点与每个像素点对应的相邻像素点在灰度值上的差值；将所述多个灰度差值中落入相同灰度值区间的划分为一组，以得到多个灰度组；根据所述多个灰度组从所述坐标原点出发确定所述宫颈细胞图像对应的多个切分形状；将所述 多个切分形状中非规则形状进行规范化处理，以得到规则的多个第一切分形状；将所述多个第一切分形状设置为所述多个宫颈细胞子图像，实现了按照灰度组从坐标原点出发确定宫颈细胞图像对应的多个切分形状，让每个切分形状由灰度差值中落入相同灰度值区间的像素点构成，让异常细胞筛选模型更快的筛选出异常细胞，也加快了切分图像的效率。同时，将多个切分形状中非规则形状进行规范化处理，构造出更适应与异常细胞筛选模型的输入数据，加快了筛选效率。
可以看出，在一种可能的实施方式中，所述基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像，包括：获取所述宫颈细胞图像中每个像素点对应的灰度值；根据所述宫颈细胞图像中每个像素点对应的灰度值确定所述宫颈细胞图像中灰度值相同的多组像素点，每组像素点包括至少一个像素点，所述至少一个像素点中的每个像素点对应的灰度值均相同；根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像，所述多组像素点与所述多个宫颈细胞子图像对应。
可以看出，上述技术方案中，获取所述宫颈细胞图像中每个像素点对应的灰度值；根据所述宫颈细胞图像中每个像素点对应的灰度值确定所述宫颈细胞图像中灰度值相同的多组像素点，每组像素点包括至少一个像素点，所述至少一个像素点中的每个像素点对应的灰度值均相同；根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像，所述多组像素点与所述多个宫颈细胞子图像对应，从而根据相同灰度值实现宫颈细胞图像的切分，为后续异常细胞筛选模型更快的筛选出异常细胞做准备。
可选的，在一种可能的实施方式中，所述根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像，包括：根据所述多组像素点切分所述宫颈细胞图像以得到多个第二宫颈细胞子图像，所述多组像素点与所述多个第二宫颈细胞子图像对应；确定所述多个第二宫颈细胞子图像中外轮廓非规则的至少一个第三宫颈细胞子图像；针对所述至少一个第三宫颈细胞子图像中的每个第三宫颈细胞子图像执行以下操作，以得到所述多个宫颈细胞子图像，包括：确定当前处理的第三宫颈细胞子图像对应的外轮廓尺寸；根据当前处理的第三宫颈细胞子图像对应的外轮廓尺寸获取模板图像，以得到所述模板图像对应的外轮廓尺寸；根据所述模板图像对应的外轮廓尺寸对当前处理的第三宫颈细胞子图像进行像素扩散，直到当前处理的第三宫颈细胞子图像对应的外轮廓尺寸与所述模板图像对应的外轮廓尺寸相同时停止像素扩散，所述像素扩散是采用第一灰度值进行扩散的。
其中，外轮廓非规则为不包括外轮廓为矩形和正方形的轮廓。其中，模板图像包括以下一种：矩形和正方形。
其中，第一灰度值为0或255。
举例来说，多个第二宫颈细胞子图像中的某个第二宫颈细胞子图像是三角形，那么，该第二宫颈细胞子图像对应的外轮廓为三角形，进一步的，该第二宫颈细胞子图像对应的外轮廓就是非规则的。可以理解的，参见图2D，图2D是本申请实施例提供的一种像素扩散的示意图，该第二宫颈细胞子图像对应的模板图像为矩形时，则根据矩形对应的外轮廓尺寸对三角形进行像素扩散，最后，第二宫颈细胞子图像对应的宫颈细胞子图像为矩形。
可以看出，上述技术方案中，根据所述多组像素点切分所述宫颈细胞图像以得到多个第二宫颈细胞子图像，所述多组像素点与所述多个第二宫颈细胞子图像对应；确定所述多个第二宫颈细胞子图像中外轮廓非规则的至少一个第三宫颈细胞子图像；针对所述至少一个第三宫颈细胞子图像中的每个第三宫颈细胞子图像执行以下操作，以得到所述多个宫颈细胞子图像，包括：确定当前处理的第三宫颈细胞子图像对应的外轮廓尺寸；根据当前处理的第三宫颈细胞子图像对应的外轮廓尺寸获取模板图像，以得到所述模板图像对应的外轮廓尺寸；根据所述模板图像对应的外轮廓尺寸对当前处理的第三宫颈细胞子图像进行像素扩散，直到当前处理的第三宫颈细胞子图像对应的外轮廓尺寸与所述模板图像对应的外 轮廓尺寸相同时停止像素扩散，所述像素扩散是采用第一灰度值进行扩散的，通过对外轮廓非规则的至少一个第三宫颈细胞子图像进行像素扩散，让所有的宫颈细胞子图像均变成矩形或正方形的图像，为后续异常细胞筛选模型更快的筛选出异常细胞做准备。
参见图3，图3为本申请的又一个实施例提供的一种异常细胞筛选方法的流程示意图。其中，如图3所示，在所述将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果之前，所述方法还包括：
301、获取训练集，所述训练集包括多个训练子集，所述多个训练子集与所述多个神经网络对应，每个训练子集包括不同明暗程度的多个宫颈细胞子图像集合，每个宫颈细胞子图像集合包括一种明暗程度的多个第二宫颈细胞子图像，所述多个第二宫颈细胞子图像中的每个第二宫颈细胞子图像包括的异常细胞不同；
可选的，在一种可能的实施方式中，所述获取训练集，包括：在标记界面上显示所述多个第三宫颈细胞子图像；在检测到针对所述标记界面上的多个位置的标记操作时，标记所述多个位置对应的所述多个第三宫颈细胞子图像，以得到所述多个第三宫颈细胞子图像对应的多个第四宫颈细胞子图像，每个第四宫颈细胞子图像是对每个第三宫颈细胞子图像进行标记后的图像；采用多个预设亮度分别对所述多个第四宫颈细胞子图像进行处理，以得到多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合，所述多个预设亮度与所述多个宫颈细胞子图像集合对应；将所述多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合设置为所述训练集。
其中，所述标记界面包括多个标记显示区，所述多个标记显示区与所述多个第三宫颈细胞子图像对应。所述在标记界面上显示多个第三宫颈细胞子图像，包括：在标记界面上的所述多个标记显示区显示所述多个第三宫颈细胞子图像。进一步的，拥有医护知识的人员可以在多个标记显示区查看多个第三宫颈细胞子图像，并对多个第三宫颈细胞子图像进行标记。
可选的，在一种可能的实施方式中，所述在检测到针对所述标记界面上的多个位置的标记操作时，标记所述多个位置对应的所述多个第三宫颈细胞子图像，以得到所述多个第三宫颈细胞子图像对应的多个第四宫颈细胞子图像，包括：在检测到针对所述标记界面上的多个位置的标记操作时，跟踪在所述多个位置对应的所述多个第三宫颈细胞子图像上的多个标记轨迹；获取所述多个标记轨迹形成的多个标记区域；根据所述多个标记区域生成多个标记标签；根据所述多个位置对应的所述多个第三宫颈细胞子图像生成包括所述多个标记标签的所述多个第四宫颈细胞子图像。
其中，第一位置为所述多个位置中的任意一个位置，在检测到所述标记界面上的所述第一位置的标记操作时，跟踪在所述第一位置对应的第三宫颈细胞子图像上的标记轨迹；获取所述标记轨迹形成的标记区域；根据所述标记区域生成标记标签；根据所述第一位置对应的第三宫颈细胞子图像生成包括所述标记标签的第四宫颈细胞子图像。
进一步的，所述多个第三宫颈细胞子图像与所述多个标记轨迹对应，所述多个标记轨迹与所述多个标记区域对应，所述多个标记区域与所述多个标记标签对应。
可以看出，上述技术方案中，在检测到针对所述标记界面上的多个位置的标记操作时，跟踪在所述多个位置对应的所述多个第三宫颈细胞子图像上的多个标记轨迹；获取所述多个标记轨迹形成的多个标记区域；根据所述多个标记区域生成多个标记标签；根据所述多个位置对应的所述多个第三宫颈细胞子图像生成包括所述多个标记标签的所述多个第四宫颈细胞子图像，实现了在标记轨迹不同时，标记区域也不同，进而导致标记标签也不同，提高了第四宫颈细胞子图像标记标签的唯一性，提高了模型训练效率。
可以看出，上述技术方案中，在标记界面上显示多个第三宫颈细胞子图像；在检测到针对所述标记界面上的多个位置的标记操作时，标记所述多个位置对应的所述多个第三宫 颈细胞子图像，以得到所述多个第三宫颈细胞子图像对应的多个第四宫颈细胞子图像，每个第四宫颈细胞子图像是对每个第三宫颈细胞子图像进行标记后的图像；采用多个预设亮度分别对所述多个第四宫颈细胞子图像进行处理，以得到多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合，所述多个预设亮度与所述多个宫颈细胞子图像集合对应；将所述多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合设置为所述训练集，实现了在标记界面上标记多个第三宫颈细胞子图像后，采用多个预设亮度对标记后的多个第三宫颈细胞子图像进行亮度调节，以得到训练集，为后续异常细胞筛选模型可以识别出不同环境光下拍摄的宫颈细胞图像做准备。
302、构建多个待训练的神经网络，所述多个待训练的神经网络与所述多个神经网络对应；
其中，所述多个待训练的神经网络分别运行在多个进程上，所述多个进程与所述多个待训练的神经网络对应。
303、基于所述多个待训练的神经网络训练所述训练集，以得到所述异常细胞筛选模型。
可以看出，上述技术方案中，通过获取训练集，所述训练集包括多个训练子集，所述多个训练子集与所述多个神经网络对应，每个训练子集包括不同明暗程度的多个宫颈细胞子图像集合，每个宫颈细胞子图像集合包括一种明暗程度的多个第二宫颈细胞子图像，所述多个第二宫颈细胞子图像中的每个第二宫颈细胞子图像包括的异常细胞不同；构建多个待训练的神经网络，所述多个待训练的神经网络与所述多个神经网络对应；基于所述多个待训练的神经网络训练所述训练集，以得到所述异常细胞筛选模型，让异常细胞筛选模型可以识别出不同环境光下拍摄的宫颈细胞图像，实现对异常细胞筛选模型的优化，增强了异常细胞筛选模型的实用性。
参见图4，图4为本申请的一个实施例提供的一种异常细胞筛选装置的示意图。其中，如图4所示，本申请的一个实施例提供的一种异常细胞筛选装置400可以包括：
第一获取模块401，用于获取宫颈细胞图像；
可选的，第一获取模块401，用于通过扫描设备获取所述宫颈细胞图像。
其中，扫描设备例如可以是扫描仪。可以理解的，采用扫描仪扫描通过液基薄层细胞检测的宫颈细胞层以得到宫颈细胞图像。
其中，宫颈细胞图像为宫颈细胞层的图像。
切分模块402，用于基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；
可选的，所述基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像时，切分模块402，具体用于获取所述宫颈细胞图像中每个像素点对应的灰度值；根据所述宫颈细胞图像中每个像素点对应的灰度值确定灰度值最小的像素点；从所述像素点中选取任意一个像素点作为切分所述宫颈细胞图像的坐标原点；以所述坐标原点为基础在所述宫颈细胞图像上建立坐标系，其中，所述坐标系以所述宫颈细胞图像的横向正方向为x轴，以所述宫颈细胞图像的纵向正方向为y轴；从所述坐标原点出发切分所述宫颈细胞图像，以得到所述多个宫颈细胞子图像。
可选的，所述从所述坐标原点出发切分所述宫颈细胞图像，以得到多个宫颈细胞子图像时，切分模块402，具体用于根据所述宫颈细胞图像中每个像素点对应的灰度值确定多个灰度差值，每个灰度差值为每个像素点与对应的相邻像素点在灰度值上的差值；将所述多个灰度差值中落入相同灰度值区间的划分为一组，以得到多个灰度组；根据所述多个灰度组从所述坐标原点出发确定所述宫颈细胞图像对应的多个切分形状；将所述多个切分形状中非规则形状进行规范化处理，以得到规则的多个第一切分形状；将所述多个第一切分形状设置为所述多个宫颈细胞子图像。
可选的，所述基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像时，切分模块402，具体用于获取所述宫颈细胞图像中每个像素点对应的灰度值；根据所述宫颈细胞图像中每个像素点对应的灰度值确定所述宫颈细胞图像中灰度值相同的多组像素点，每组像素点包括至少一个像素点，所述至少一个像素点中的每个像素点对应的灰度值均相同；根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像，所述多组像素点与所述多个宫颈细胞子图像对应。
可选的，所述根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像时，切分模块402，具体用于根据所述多组像素点切分所述宫颈细胞图像以得到多个第二宫颈细胞子图像，所述多组像素点与所述多个第二宫颈细胞子图像对应；确定所述多个第二宫颈细胞子图像中外轮廓非规则的至少一个第三宫颈细胞子图像；针对所述至少一个第三宫颈细胞子图像中的每个第三宫颈细胞子图像执行以下操作，以得到所述多个宫颈细胞子图像，包括：确定当前处理的第三宫颈细胞子图像对应的外轮廓尺寸；根据当前处理的第三宫颈细胞子图像对应的外轮廓尺寸获取模板图像，以得到所述模板图像对应的外轮廓尺寸；根据所述模板图像对应的外轮廓尺寸对当前处理的第三宫颈细胞子图像进行像素扩散，直到当前处理的第三宫颈细胞子图像对应的外轮廓尺寸与所述模板图像对应的外轮廓尺寸相同时停止像素扩散，所述像素扩散是采用第一灰度值进行扩散的。
输入模块403，用于将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；
其中，宫颈细胞为处于正常生长状态的细胞。
其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
进一步的，第一神经网络是所述多个神经网络中的一个神经网络，所述第一神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第一异常细胞，第二神经网络是所述多个神经网络中的不同于所述第一神经网络的另一个神经网络，所述第二神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第二异常细胞。
其中，第一异常细胞与第二异常细胞不同，第一异常细胞是所述异常细胞中的一种细胞，第二异常细胞是所述异常细胞中的另一种细胞。
可选的，异常细胞至少包括以下一种：鳞癌(squamous cell carcinom，SCC)、高度鳞状上皮病变(high-grade squamous epithelial lesion，LSIL)、不能排除高级别鳞状上皮内病变的非典型鳞状细胞(atypical squamous cells,cannot exclude high-gradesquamous intraepithelial lesion，ASC-H)))、低度鳞状上皮病变(low-grade squamous epithelial lesion，LSIL)、意义不明确的非典型鳞状细胞(atypical squamouscells of undP.termined significance,ASC-US)、腺癌(adenocarcinoma，AC)、非典型腺癌(Atypical adenocarcinoma，AGC)等。
可以理解的，多个神经网络分别用于筛选多个宫颈细胞子图像中的每个宫颈细胞子图像包括的一种异常细胞。具体来说，多个神经网络中的某个神经网络用于筛选多个宫颈细胞子图像中的每个宫颈细胞子图像包括的鳞癌，多个神经网络中的另一个神经网络用于筛选多个宫颈细胞子图像中的每个宫颈细胞子图像包括的腺癌。
可选的，在所述将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果之前，所述异常细胞筛选装置还包括处理模块，所述处理模块，用于获取训练集，所述训练集包括多个训练子集，所述多个训练子集与所述多个神经网络对应，每个训练子集包括不同明暗程度的多个宫颈细胞子图像集合，每个 宫颈细胞子图像集合包括一种明暗程度的多个第二宫颈细胞子图像，所述多个第二宫颈细胞子图像中的每个第二宫颈细胞子图像包括的异常细胞不同；构建多个待训练的神经网络，所述多个待训练的神经网络与所述多个神经网络对应；基于所述多个待训练的神经网络训练所述训练集，以得到所述异常细胞筛选模型。
可选的，所述获取训练集时，所述处理模块，具体用于在标记界面上显示所述多个第三宫颈细胞子图像；在检测到针对所述标记界面上的多个位置的标记操作时，标记所述多个位置对应的所述多个第三宫颈细胞子图像，以得到所述多个第三宫颈细胞子图像对应的多个第四宫颈细胞子图像，每个第四宫颈细胞子图像是对每个第三宫颈细胞子图像进行标记后的图像；采用多个预设亮度分别对所述多个第四宫颈细胞子图像进行处理，以得到多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合，所述多个预设亮度与所述多个宫颈细胞子图像集合对应；将所述多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合设置为所述训练集。
选取模块404，用于根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；
其中，所述预设选取策略是根据预测选取操作确定，所述预测选取操作包括以下步骤：获取所述多个预测结果中的每个预测结果对应的预测概率；将所述多个预测结果中的每个预测结果对应的预测概率按照预测概率从大到小的顺序进行编号，以得到多个编号；从所述多个编号中选取部分编号，其中，所述部分编号为所述多个编号中编号大于预设编号的至少一个编号；将所述部分编号一一对应的部分预测概率作为所述预设选取策略。
其中，预设编号由管理员设置。举例来说，多个编号为从10-1的10个编号，预设编号为6，那么部分编号为7-10这3个编号。
进一步的，所述根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果，包括：根据所述部分预测概率从所述多个预测结果中选取多个预测结果作为多个第一预测结果。
第二获取模块405，用于获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；
显示模块406，用于在显示界面上显示所述多个第一宫颈细胞子图像；
参见图5，图5为本申请的实施例涉及的硬件运行环境的服务器结构示意图。
本申请实施例提供了一种信息推送的电子设备，包括处理器、存储器、通信接口以及一个或多个程序，其中，所述一个或多个程序被存储在所述存储器中，并且被配置由所述处理器执行，以执行包括任一项信息推送方法中的步骤的指令。其中，如图5所示，本申请的实施例涉及的硬件运行环境的服务器可以包括：
处理器501，例如CPU。
存储器502，可选的，存储器可以为高速RAM存储器，也可以是稳定的存储器，例如磁盘存储器。
通信接口503，用于实现处理器501和存储器502之间的连接通信。
本领域技术人员可以理解，图5中示出的服务器的结构并不构成对其的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
如图5所示，存储器502中可以包括操作系统、网络通信模块以及一个或多个程序。操作系统是管理和控制服务器硬件和软件资源的程序，一个或多个程序以及其他软件或程序的运行。网络通信模块用于实现存储器502内部各组件之间的通信，以及与服务器内部其他硬件和软件之间通信。
在图5所示的服务器中，处理器501用于执行存储器502中存储的人员管理的程序，实现以下步骤：
获取宫颈细胞图像；
基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；
将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；
根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；
获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；
在显示界面上显示所述多个第一宫颈细胞子图像；
其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。本申请涉及的服务器的具体实施可参见上述异常细胞筛选方法的各实施例，在此不做赘述。
本申请还提供了一种异常细胞筛选的电子设备，包括处理器、存储器、通信接口以及一个或多个程序，其中，所述一个或多个程序被存储在所述存储器中，并且被生成由所述处理器执行，以执行以下的步骤的指令：获取宫颈细胞图像；基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；在显示界面上显示所述多个第一宫颈细胞子图像；其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
可选的，在基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像时，所述处理器，用于获取所述宫颈细胞图像中每个像素点对应的灰度值；根据所述宫颈细胞图像中每个像素点对应的灰度值确定灰度值最小的像素点；从所述像素点中选取任意一个像素点作为切分所述宫颈细胞图像的坐标原点；以所述坐标原点为基础在所述宫颈细胞图像上建立坐标系，其中，所述坐标系以所述宫颈细胞图像的横向正方向为x轴，以所述宫颈细胞图像的纵向正方向为y轴；从所述坐标原点出发切分所述宫颈细胞图像，以得到所述多个宫颈细胞子图像。
可选的，在从所述坐标原点出发切分所述宫颈细胞图像，以得到多个宫颈细胞子图像时，所述处理器，用于根据所述宫颈细胞图像中每个像素点对应的灰度值确定多个灰度差值，每个灰度差值为每个像素点与对应的相邻像素点在灰度值上的差值；将所述多个灰度差值中落入相同灰度值区间的划分为一组，以得到多个灰度组；根据所述多个灰度组从所述坐标原点出发确定所述宫颈细胞图像对应的多个切分形状；将所述多个切分形状中非规则形状进行规范化处理，以得到规则的多个第一切分形状；将所述多个第一切分形状设置为所述多个宫颈细胞子图像。
可选的，在基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像时，所述处理器，用于获取所述宫颈细胞图像中每个像素点对应的灰度值；根据所述宫颈细胞图像中每个像素点对应的灰度值确定所述宫颈细胞图像中灰度值相同的多组像素点，每组像素点包括至少一个像素点，所述至少一个像素点中的每个像素点对应的灰度值均相同；根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像，所述多组像素点与所述多个宫颈细胞子图像对应。
可选的，在根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图 像时，所述处理器，用于根据所述多组像素点切分所述宫颈细胞图像以得到多个第二宫颈细胞子图像，所述多组像素点与所述多个第二宫颈细胞子图像对应；确定所述多个第二宫颈细胞子图像中外轮廓非规则的至少一个第三宫颈细胞子图像；针对所述至少一个第三宫颈细胞子图像中的每个第三宫颈细胞子图像执行以下操作，以得到所述多个宫颈细胞子图像，包括：确定当前处理的第三宫颈细胞子图像对应的外轮廓尺寸；根据当前处理的第三宫颈细胞子图像对应的外轮廓尺寸获取模板图像，以得到所述模板图像对应的外轮廓尺寸；根据所述模板图像对应的外轮廓尺寸对当前处理的第三宫颈细胞子图像进行像素扩散，直到当前处理的第三宫颈细胞子图像对应的外轮廓尺寸与所述模板图像对应的外轮廓尺寸相同时停止像素扩散，所述像素扩散是采用第一灰度值进行扩散的。
可选的，在所述将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果之前，所述处理器，还用于获取训练集，所述训练集包括多个训练子集，所述多个训练子集与所述多个神经网络对应，每个训练子集包括不同明暗程度的多个宫颈细胞子图像集合，每个宫颈细胞子图像集合包括一种明暗程度的多个第二宫颈细胞子图像，所述多个第二宫颈细胞子图像中的每个第二宫颈细胞子图像包括的异常细胞不同；构建多个待训练的神经网络，所述多个待训练的神经网络与所述多个神经网络对应；基于所述多个待训练的神经网络训练所述训练集，以得到所述异常细胞筛选模型。
可选的，在获取训练集时，所述处理器，用于在标记界面上显示所述多个第三宫颈细胞子图像；在检测到针对所述标记界面上的多个位置的标记操作时，标记所述多个位置对应的所述多个第三宫颈细胞子图像，以得到所述多个第三宫颈细胞子图像对应的多个第四宫颈细胞子图像，每个第四宫颈细胞子图像是对每个第三宫颈细胞子图像进行标记后的图像；采用多个预设亮度分别对所述多个第四宫颈细胞子图像进行处理，以得到多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合，所述多个预设亮度与所述多个宫颈细胞子图像集合对应；将所述多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合设置为所述训练集。
可选的，第一神经网络是所述多个神经网络中的一个神经网络，所述第一神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第一异常细胞，第二神经网络是所述多个神经网络中的不同于所述第一神经网络的另一个神经网络，所述第二神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第二异常细胞。其中，第一异常细胞与第二异常细胞不同，第一异常细胞是所述异常细胞中的一种细胞，第二异常细胞是所述异常细胞中的另一种细胞。
可选的，所述预设选取策略是根据预测选取操作确定，所述预测选取操作包括以下步骤：
获取所述多个预测结果中的每个预测结果对应的预测概率；将所述多个预测结果中的每个预测结果对应的预测概率按照预测概率从大到小的顺序进行编号，以得到多个编号；从所述多个编号中选取部分编号，其中，所述部分编号为所述多个编号中编号大于预设编号的至少一个编号；将所述部分编号一一对应的部分预测概率作为所述预设选取策略。
本申请还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用于存储计算机程序，所述存储计算机程序被所述处理器执行，以实现以下步骤：
获取宫颈细胞图像；基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；根据预设选取策略从所述多个预测结果中选 取多个预测结果作为多个第一预测结果；获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；在显示界面上显示所述多个第一宫颈细胞子图像；
其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
本申请涉及的计算机可读存储介质的具体实施可参见上述异常细胞筛选方法的各实施例，在此不做赘述。
其中，所述计算机可读存储介质可以是非易失性，也可以是易失性。
需要说明的是，对于前述的各方法实施例，为了简单描述，故将其都表述为一系列的动作组合，但是本领域技术人员应所述知悉，本申请并不受所描述的动作顺序的限制，因为依据本申请，某些步骤可以采用其他顺序或者同时进行。其次，本领域技术人员也应所述知悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作和模块并不一定是本申请所必须的。
以上所述，以上实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的范围。

Claims (20)

  1. 一种异常细胞筛选方法，其中，包括：
    获取宫颈细胞图像；
    基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；
    将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；
    根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；
    获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；
    在显示界面上显示所述多个第一宫颈细胞子图像；
    其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
  2. 根据权利要求1所述的方法，其中，所述基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像，包括：
    获取所述宫颈细胞图像中每个像素点对应的灰度值；
    根据所述宫颈细胞图像中每个像素点对应的灰度值确定灰度值最小的像素点；
    从所述像素点中选取任意一个像素点作为切分所述宫颈细胞图像的坐标原点；
    以所述坐标原点为基础在所述宫颈细胞图像上建立坐标系，其中，所述坐标系以所述宫颈细胞图像的横向正方向为x轴，以所述宫颈细胞图像的纵向正方向为y轴；
    从所述坐标原点出发切分所述宫颈细胞图像，以得到所述多个宫颈细胞子图像。
  3. 根据权利要求2所述的方法，其中，所述从所述坐标原点出发切分所述宫颈细胞图像，以得到多个宫颈细胞子图像，包括：
    根据所述宫颈细胞图像中每个像素点对应的灰度值确定多个灰度差值，每个灰度差值为每个像素点与对应的相邻像素点在灰度值上的差值；
    将所述多个灰度差值中落入相同灰度值区间的划分为一组，以得到多个灰度组；
    根据所述多个灰度组从所述坐标原点出发确定所述宫颈细胞图像对应的多个切分形状；
    将所述多个切分形状中非规则形状进行规范化处理，以得到规则的多个第一切分形状；
    将所述多个第一切分形状设置为所述多个宫颈细胞子图像。
  4. 根据权利要求1所述的方法，其中，所述基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像，包括：
    获取所述宫颈细胞图像中每个像素点对应的灰度值；
    根据所述宫颈细胞图像中每个像素点对应的灰度值确定所述宫颈细胞图像中灰度值相同的多组像素点，每组像素点包括至少一个像素点，所述至少一个像素点中的每个像素点对应的灰度值均相同；
    根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像，所述多组像素点与所述多个宫颈细胞子图像对应。
  5. 根据权利要求4所述的方法，其中，所述根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像，包括：
    根据所述多组像素点切分所述宫颈细胞图像以得到多个第二宫颈细胞子图像，所述多组像素点与所述多个第二宫颈细胞子图像对应；
    确定所述多个第二宫颈细胞子图像中外轮廓非规则的至少一个第三宫颈细胞子图像；
    针对所述至少一个第三宫颈细胞子图像中的每个第三宫颈细胞子图像执行以下操作，以得到所述多个宫颈细胞子图像，包括：
    确定当前处理的第三宫颈细胞子图像对应的外轮廓尺寸；根据当前处理的第三宫颈细胞子图像对应的外轮廓尺寸获取模板图像，以得到所述模板图像对应的外轮廓尺寸；根据所述模板图像对应的外轮廓尺寸对当前处理的第三宫颈细胞子图像进行像素扩散，直到当前处理的第三宫颈细胞子图像对应的外轮廓尺寸与所述模板图像对应的外轮廓尺寸相同时停止像素扩散，所述像素扩散是采用第一灰度值进行扩散的。
  6. 根据权利要求1所述的方法，其中，在所述将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果之前，所述方法还包括：
    获取训练集，所述训练集包括多个训练子集，所述多个训练子集与所述多个神经网络对应，每个训练子集包括不同明暗程度的多个宫颈细胞子图像集合，每个宫颈细胞子图像集合包括一种明暗程度的多个第二宫颈细胞子图像，所述多个第二宫颈细胞子图像中的每个第二宫颈细胞子图像包括的异常细胞不同；
    构建多个待训练的神经网络，所述多个待训练的神经网络与所述多个神经网络对应；
    基于所述多个待训练的神经网络训练所述训练集，以得到所述异常细胞筛选模型。
  7. 根据权利要求6所述的方法，其中，所述获取训练集，包括：
    在标记界面上显示所述多个第三宫颈细胞子图像；
    在检测到针对所述标记界面上的多个位置的标记操作时，标记所述多个位置对应的所述多个第三宫颈细胞子图像，以得到所述多个第三宫颈细胞子图像对应的多个第四宫颈细胞子图像，每个第四宫颈细胞子图像是对每个第三宫颈细胞子图像进行标记后的图像；
    采用多个预设亮度分别对所述多个第四宫颈细胞子图像进行处理，以得到多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合，所述多个预设亮度与所述多个宫颈细胞子图像集合对应；
    将所述多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合设置为所述训练集。
  8. 根据权利要求1-7任意一项所述的方法，其中，第一神经网络是所述多个神经网络中的一个神经网络，所述第一神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第一异常细胞，第二神经网络是所述多个神经网络中的不同于所述第一神经网络的另一个神经网络，所述第二神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第二异常细胞；
    其中，第一异常细胞与第二异常细胞不同，第一异常细胞是所述异常细胞中的一种细胞，第二异常细胞是所述异常细胞中的另一种细胞。
  9. 根据权利要求1所述的方法，其中，所述预设选取策略是根据预测选取操作确定，所述预测选取操作包括以下步骤：
    获取所述多个预测结果中的每个预测结果对应的预测概率；
    将所述多个预测结果中的每个预测结果对应的预测概率按照预测概率从大到小的顺序进行编号，以得到多个编号；
    从所述多个编号中选取部分编号，其中，所述部分编号为所述多个编号中编号大于预设编号的至少一个编号；
    将所述部分编号一一对应的部分预测概率作为所述预设选取策略。
  10. 一种异常细胞筛选装置，其中，包括：
    第一获取模块，用于获取宫颈细胞图像；
    切分模块，用于基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得 到多个宫颈细胞子图像；
    输入模块，用于将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；
    选取模块，用于根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；
    第二获取模块，用于获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；
    显示模块，用于在显示界面上显示所述多个第一宫颈细胞子图像；
    其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
  11. 一种异常细胞筛选的电子设备，其中，包括处理器、存储器、通信接口以及一个或多个程序，其中，所述一个或多个程序被存储在所述存储器中，并且被生成由所述处理器执行，以执行以下的步骤的指令：
    获取宫颈细胞图像；
    基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；
    将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；
    根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；
    获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；
    在显示界面上显示所述多个第一宫颈细胞子图像；
    其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
  12. 根据权利要求11所述的设备，其中，在基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像时，所述处理器，用于获取所述宫颈细胞图像中每个像素点对应的灰度值；根据所述宫颈细胞图像中每个像素点对应的灰度值确定灰度值最小的像素点；从所述像素点中选取任意一个像素点作为切分所述宫颈细胞图像的坐标原点；以所述坐标原点为基础在所述宫颈细胞图像上建立坐标系，其中，所述坐标系以所述宫颈细胞图像的横向正方向为x轴，以所述宫颈细胞图像的纵向正方向为y轴；从所述坐标原点出发切分所述宫颈细胞图像，以得到所述多个宫颈细胞子图像。
  13. 根据权利要求12所述的设备，其中，在从所述坐标原点出发切分所述宫颈细胞图像，以得到多个宫颈细胞子图像时，所述处理器，用于根据所述宫颈细胞图像中每个像素点对应的灰度值确定多个灰度差值，每个灰度差值为每个像素点与对应的相邻像素点在灰度值上的差值；将所述多个灰度差值中落入相同灰度值区间的划分为一组，以得到多个灰度组；根据所述多个灰度组从所述坐标原点出发确定所述宫颈细胞图像对应的多个切分形状；将所述多个切分形状中非规则形状进行规范化处理，以得到规则的多个第一切分形状；将所述多个第一切分形状设置为所述多个宫颈细胞子图像。
  14. 根据权利要求11所述的设备，其中，在基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像时，所述处理器，用于获取所述宫颈细胞图像中每个像素点对应的灰度值；根据所述宫颈细胞图像中每个像素点对应的灰度值确定所述宫颈细胞图像中灰度值相同的多组像素点，每组像素点包括至少一个像素点，所 述至少一个像素点中的每个像素点对应的灰度值均相同；根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像，所述多组像素点与所述多个宫颈细胞子图像对应。
  15. 根据权利要求14所述的设备，其中，在根据所述多组像素点切分所述宫颈细胞图像以得到所述多个宫颈细胞子图像时，所述处理器，用于根据所述多组像素点切分所述宫颈细胞图像以得到多个第二宫颈细胞子图像，所述多组像素点与所述多个第二宫颈细胞子图像对应；确定所述多个第二宫颈细胞子图像中外轮廓非规则的至少一个第三宫颈细胞子图像；针对所述至少一个第三宫颈细胞子图像中的每个第三宫颈细胞子图像执行以下操作，以得到所述多个宫颈细胞子图像，包括：确定当前处理的第三宫颈细胞子图像对应的外轮廓尺寸；根据当前处理的第三宫颈细胞子图像对应的外轮廓尺寸获取模板图像，以得到所述模板图像对应的外轮廓尺寸；根据所述模板图像对应的外轮廓尺寸对当前处理的第三宫颈细胞子图像进行像素扩散，直到当前处理的第三宫颈细胞子图像对应的外轮廓尺寸与所述模板图像对应的外轮廓尺寸相同时停止像素扩散，所述像素扩散是采用第一灰度值进行扩散的。
  16. 根据权利要求14所述的设备，其中，在所述将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果之前，所述处理器，还用于获取训练集，所述训练集包括多个训练子集，所述多个训练子集与所述多个神经网络对应，每个训练子集包括不同明暗程度的多个宫颈细胞子图像集合，每个宫颈细胞子图像集合包括一种明暗程度的多个第二宫颈细胞子图像，所述多个第二宫颈细胞子图像中的每个第二宫颈细胞子图像包括的异常细胞不同；构建多个待训练的神经网络，所述多个待训练的神经网络与所述多个神经网络对应；基于所述多个待训练的神经网络训练所述训练集，以得到所述异常细胞筛选模型。
  17. 根据权利要求16所述的设备，其中，在获取训练集时，所述处理器，用于在标记界面上显示所述多个第三宫颈细胞子图像；在检测到针对所述标记界面上的多个位置的标记操作时，标记所述多个位置对应的所述多个第三宫颈细胞子图像，以得到所述多个第三宫颈细胞子图像对应的多个第四宫颈细胞子图像，每个第四宫颈细胞子图像是对每个第三宫颈细胞子图像进行标记后的图像；采用多个预设亮度分别对所述多个第四宫颈细胞子图像进行处理，以得到多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合，所述多个预设亮度与所述多个宫颈细胞子图像集合对应；将所述多个训练子集中每个训练子集包括的所述多个宫颈细胞子图像集合设置为所述训练集。
  18. 根据权利要求11-17任意一项所述的设备，其中，第一神经网络是所述多个神经网络中的一个神经网络，所述第一神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第一异常细胞，第二神经网络是所述多个神经网络中的不同于所述第一神经网络的另一个神经网络，所述第二神经网络用于筛选所述多个宫颈细胞子图像中的每个宫颈细胞子图像包括的第二异常细胞；其中，第一异常细胞与第二异常细胞不同，第一异常细胞是所述异常细胞中的一种细胞，第二异常细胞是所述异常细胞中的另一种细胞。
  19. 根据权利要求11所述的设备，其中，所述预设选取策略是根据预测选取操作确定，所述预测选取操作包括以下步骤：
    获取所述多个预测结果中的每个预测结果对应的预测概率；将所述多个预测结果中的每个预测结果对应的预测概率按照预测概率从大到小的顺序进行编号，以得到多个编号；从所述多个编号中选取部分编号，其中，所述部分编号为所述多个编号中编号大于预设编号的至少一个编号；将所述部分编号一一对应的部分预测概率作为所述预设选取策略。
  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用于存储计算机程序，所述存储计算机程序被所述处理器执行，以实现以下步骤：
    获取宫颈细胞图像；基于所述宫颈细胞图像中像素点的灰度值切分所述宫颈细胞图像以得到多个宫颈细胞子图像；将所述多个宫颈细胞子图像分别输入异常细胞筛选模型，以得到所述多个宫颈细胞子图像对应的多个预测结果，其中，每个宫颈细胞子图像对应一个预测结果，每个预测结果用于指示每个宫颈细胞子图像包括的异常细胞，所述异常细胞为在宫颈细胞的基础上发生病变或癌变的细胞；根据预设选取策略从所述多个预测结果中选取多个预测结果作为多个第一预测结果；获取所述多个第一预测结果对应的多个第一宫颈细胞子图像；在显示界面上显示所述多个第一宫颈细胞子图像；
    其中，所述异常细胞筛选模型包括多个神经网络，所述神经网络的数量与所述异常细胞的种类数量相等，所述多个神经网络用于筛选不同的异常细胞。
PCT/CN2020/093581 2019-10-29 2020-05-30 一种异常细胞筛选方法、装置、电子设备和存储介质 WO2021082434A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911040955.9A CN110992303B (zh) 2019-10-29 2019-10-29 一种异常细胞筛选方法、装置、电子设备和存储介质
CN201911040955.9 2019-10-29

Publications (1)

Publication Number Publication Date
WO2021082434A1 true WO2021082434A1 (zh) 2021-05-06

Family

ID=70082495

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/093581 WO2021082434A1 (zh) 2019-10-29 2020-05-30 一种异常细胞筛选方法、装置、电子设备和存储介质

Country Status (2)

Country Link
CN (1) CN110992303B (zh)
WO (1) WO2021082434A1 (zh)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110415212A (zh) * 2019-06-18 2019-11-05 平安科技（深圳）有限公司 异常细胞检测方法、装置及计算机可读存储介质
CN110992303B (zh) * 2019-10-29 2023-12-22 平安科技（深圳）有限公司 一种异常细胞筛选方法、装置、电子设备和存储介质
CN111951271B (zh) * 2020-06-30 2023-12-15 杭州依图医疗技术有限公司 一种识别病理图像中癌细胞的方法及装置
CN111967465A (zh) * 2020-07-07 2020-11-20 广州金域医学检验中心有限公司 肿瘤细胞含量评估方法、系统、计算机设备及存储介质
CN112037862B (zh) * 2020-08-26 2021-11-30 深圳太力生物技术有限责任公司 基于卷积神经网络的细胞筛选方法和装置
CN112184733B (zh) * 2020-09-30 2024-07-23 平安科技（深圳）有限公司 一种宫颈异常细胞检测装置及方法
CN113111926A (zh) * 2021-03-31 2021-07-13 南京华晟医学检验实验室有限公司 基于tct制片的异常宫颈血细胞筛选方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2008005426A2 (en) * 2006-06-30 2008-01-10 University Of South Florida Computer-aided pathological diagnosis system
CN105095865A (zh) * 2015-07-17 2015-11-25 广西师范大学 基于有向加权复杂网络的宫颈细胞识别方法及宫颈细胞识别装置
CN107742299A (zh) * 2017-11-28 2018-02-27 中国联合网络通信集团有限公司 一种图像分割方法和装置
CN110120040A (zh) * 2019-05-13 2019-08-13 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
CN110211108A (zh) * 2019-05-29 2019-09-06 武汉兰丁医学高科技有限公司 一种新型的基于Feulgen染色方法的异常宫颈细胞自动识别方法
CN110689518A (zh) * 2019-08-15 2020-01-14 平安科技（深圳）有限公司 宫颈细胞图像筛查方法、装置、计算机设备及存储介质
CN110992303A (zh) * 2019-10-29 2020-04-10 平安科技（深圳）有限公司 一种异常细胞筛选方法、装置、电子设备和存储介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
MY195031A (en) * 2015-05-25 2023-01-04 Natarajan Adarsh Automated Stainer and a Method Thereof
US10803582B2 (en) * 2016-07-04 2020-10-13 Nec Corporation Image diagnosis learning device, image diagnosis device, image diagnosis method, and recording medium for storing program
US10354122B1 (en) * 2018-03-02 2019-07-16 Hong Kong Applied Science and Technology Research Institute Company Limited Using masks to improve classification performance of convolutional neural networks with applications to cancer-cell screening
CN109872306B (zh) * 2019-01-28 2021-01-08 腾讯科技（深圳）有限公司 医学图像分割方法、装置和存储介质

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2008005426A2 (en) * 2006-06-30 2008-01-10 University Of South Florida Computer-aided pathological diagnosis system
CN105095865A (zh) * 2015-07-17 2015-11-25 广西师范大学 基于有向加权复杂网络的宫颈细胞识别方法及宫颈细胞识别装置
CN107742299A (zh) * 2017-11-28 2018-02-27 中国联合网络通信集团有限公司 一种图像分割方法和装置
CN110120040A (zh) * 2019-05-13 2019-08-13 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
CN110211108A (zh) * 2019-05-29 2019-09-06 武汉兰丁医学高科技有限公司 一种新型的基于Feulgen染色方法的异常宫颈细胞自动识别方法
CN110689518A (zh) * 2019-08-15 2020-01-14 平安科技（深圳）有限公司 宫颈细胞图像筛查方法、装置、计算机设备及存储介质
CN110992303A (zh) * 2019-10-29 2020-04-10 平安科技（深圳）有限公司 一种异常细胞筛选方法、装置、电子设备和存储介质

Also Published As

Publication number Publication date
CN110992303A (zh) 2020-04-10
CN110992303B (zh) 2023-12-22

Similar Documents

Publication Publication Date Title
WO2021082434A1 (zh) 一种异常细胞筛选方法、装置、电子设备和存储介质
US20220092772A1 (en) Interpretation and Quantification of Emergency Features on Head Computed Tomography
CN107067402B (zh) 医疗图像处理装置及其乳房图像处理方法
CN110705583B (zh) 细胞检测模型训练方法、装置、计算机设备及存储介质
Qian et al. CephaNet: An improved faster R-CNN for cephalometric landmark detection
WO2021217851A1 (zh) 异常细胞自动标注方法、装置、电子设备及存储介质
US10650221B2 (en) Systems and methods for comprehensive multi-assay tissue analysis
US9058650B2 (en) Methods, apparatuses, and computer program products for identifying a region of interest within a mammogram image
Jamil et al. Melanoma segmentation using bio-medical image analysis for smarter mobile healthcare
Nateghi et al. A deep learning approach for mitosis detection: application in tumor proliferation prediction from whole slide images
WO2020077961A1 (zh) 一种乳腺影像病灶识别的方法及装置
TWI587844B (zh) 醫療影像處理裝置及其乳房影像處理方法
WO2021139447A1 (zh) 一种宫颈异常细胞检测装置及方法
US20220245802A1 (en) Systems and methods for classifying biomedical image data using a graph neural network
CN113313680B (zh) 一种结直肠癌病理图像预后辅助预测方法及系统
CN109815974A (zh) 一种细胞病理玻片分类方法、系统、设备、存储介质
CN111339899B (zh) 导管特征获取方法、装置、设备、介质和智能显微镜
Shanker et al. Brain tumor segmentation of normal and lesion tissues using hybrid clustering and hierarchical centroid shape descriptor
CN113096080A (zh) 图像分析方法及系统
US20170367677A1 (en) Analysis method for breast image and electronic apparatus using the same
CN113378790A (zh) 视点定位方法、装置、电子设备和计算机可读存储介质
TWM527991U (zh) 醫療影像處理裝置
CN111179226B (zh) 视野图的识别方法、装置及计算机存储介质
CN111815591A (zh) 基于ct图像的肺结节检测方法
Bafna et al. A heuristic approach for automated mammogram orientation in computer aided diagnosis

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: 20882847

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: 20882847

Country of ref document: EP

Kind code of ref document: A1