Data generating apparatus, data generating method, and computer program product

Abstract

Provided is a data generation apparatus including a display control section that causes an image to be displayed in a screen overlaid with a boundary line of a plurality of partial images generated by dividing the image into the plurality of partial images; an input section that accepts input of a label added to each of the plurality of partial images; and a generation section that generates learning data for causing a learning model to learn by associating each of the plurality of partial images with the label added to each of the plurality of partial images.

Description

Technical Field

[0001] This invention relates to a data generation apparatus, a data generation method, and a program. Background Technology

[0002] A system for pathological diagnosis using human intelligence to analyze pathological images is currently known. For example, Patent Document 1 describes using multiple annotated digital pathological images to enable a machine learning algorithm to learn, and using a recognition model generated through learning to identify whether abnormal image patterns correspond to pathological abnormalities.

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: Japanese Patent Publication No. 2019-525151 Summary of the Invention

[0006] The technical problem that the invention aims to solve

[0007] The training data requires a large number of pathological images, including labels indicating the presence or absence of abnormalities, for the learning model to learn from. However, since doctors need to judge whether the pathological images contain abnormalities, generating a large amount of training data places a significant burden on them. It should be noted that this same problem is not limited to pathological images but can occur with any type of image.

[0008] Here, the object of the present invention is to provide a technique that can more easily generate learning data related to images.

[0009] Technical solutions for solving the problem

[0010] One aspect of the present invention relates to a data generation apparatus comprising: a display control unit that overlays an image with the boundary lines of a plurality of local images generated by dividing the image into a plurality of images and displays the image on a screen; an input unit that accepts input of a label added to each of the plurality of local images; and a generation unit that generates learning data for a learning model to learn by establishing a correspondence between each of the plurality of local images and the label added to each of the plurality of local images.

[0011] Invention Effects

[0012] According to the present invention, a technique can be provided that enables the simpler generation of image-related learning data. Attached Figure Description

[0013] Figure 1 This is a diagram illustrating an example of an image processing system that determines whether there are abnormalities in a pathological image.

[0014] Figure 2 This is a diagram illustrating an example of the hardware configuration of a data generation device.

[0015] Figure 3 This is a diagram illustrating an example of the functional block configuration of a data generation device.

[0016] Figure 4 This is a flowchart illustrating an example of the processing sequence performed by the data generation device.

[0017] Figure 5 This is an example of a screen showing the process of adding tags.

[0018] Figure 6 This is a diagram used to illustrate the processing of pathological images. Detailed Implementation

[0019] Embodiments of the present invention will be described with reference to the accompanying drawings. It should be noted that components with the same reference numerals in each figure have the same or identical configuration. It should be noted that this embodiment is described using the case of generating learning data related to a pathological image based on an input pathological image as an example, but is not limited thereto. This embodiment can be applied to the generation of learning data based on various images. For example, it can also be applied to two-dimensional image data related to medical applications, such as ultrasound images, CT images, and endoscopic images. Furthermore, it can also be applied to images other than those used for medical purposes.

[0020] <System Composition>

[0021] Figure 1 This diagram illustrates an example of an image processing system 1 that determines whether an abnormality exists in a pathological image. The data generation device 10 generates learning data (teaching data) from the pathological image input to it, which is used to train a learning model in the diagnostic device 20. The diagnostic device 20 determines whether an abnormality exists in a pathological image using a learned model obtained by learning from the learning data generated by the data generation device 10. The data generation device 10 and the diagnostic device 20 communicate with each other via a wired or wireless communication network N.

[0022] The data generation device 10 displays a pathological image input into the data generation device 10 on a screen by overlapping the boundary lines of multiple local images, which are generated by dividing the pathological image into multiple images. Local images can also be referred to as tiles. The size of the local image (the number of pixels vertically and horizontally) is the same as the size of the learning data that can be input into the learning model. Furthermore, for each local image displayed on the screen, the data generation device 10 receives input from a pathologist or similar individual using the data generation device 10 (hereinafter referred to as "user") who adds pathology-related tags to the local image. In the following description, it is assumed that the tags are two types indicating the presence or absence of pathological abnormalities in the local image, but this embodiment is not limited to this. This embodiment can also be applied to cases where three or more tags are added.

[0023] The data generation device 10 can generate learning data based on a single pathological image or multiple pathological images. When generating learning data from multiple pathological images, the data generation device 10 repeats the following process for each pathological image: dividing the pathological image into multiple local images to accept label input. When the user has completed adding labels to all pathological images, the data generation device 10 generates learning data by establishing a correspondence between the image data of each local image and the labels added to each local image. The generated learning data is then sent to the diagnostic device 20.

[0024] The diagnostic device 20 uses learning data sent from the data generation device 10 to train a learning model. A learning model refers to, for example, a neural network that can acquire predetermined capabilities through learning. The diagnostic device 20 inputs a pathological image of the diagnostic object into the learned model and, based on the output of the learned model, determines whether an abnormality exists in the pathological image.

[0025] It should be noted that, in Figure 1 In the example shown, the data generation device 10 is illustrated as an information processing device, but this embodiment is not limited to this. For example, the data generation device 10 may be composed of one or more physical servers, or it may be composed of a virtual server running on a hypervisor, or it may be composed of a cloud server.

[0026] <Hardware Configuration>

[0027] Figure 2This diagram illustrates an example of the hardware configuration of the data generation device 10. The data generation device 10 includes a processor 11 such as a CPU (Central Processing Unit) or GPU (Graphical Processing Unit), a storage device 12 such as a memory, HDD (Hard Disk Drive) and / or SSD (Solid State Drive), a communication interface 13 for wired or wireless communication, an input device 14 for accepting input operations, and an output device 15 for outputting information. The input device 14 may be, for example, a keyboard, a touch panel, a mouse, and / or a microphone. The output device 15 may be, for example, a display, a touch panel, and / or a speaker.

[0028] <Functional Block Structure>

[0029] Figure 3 This diagram illustrates an example of the functional block configuration of the data generation apparatus 10. The data generation apparatus 10 includes a storage unit 100, a display control unit 101, an input unit 102, a generation unit 103, and an image processing unit 104. The storage unit 100 can be implemented using the storage device 12 included in the data generation apparatus 10. Furthermore, the display control unit 101, the input unit 102, the generation unit 103, and the image processing unit 104 can be implemented by executing a program stored in the storage device 12 through the processor 11 of the data generation apparatus 10. In addition, this program can be stored in a storage medium. The storage medium storing the program can be a non-transitory computer-readable medium. There are no particular limitations on the non-transitory storage medium; for example, it can be a USB memory or a CD-ROM.

[0030] Storage unit 100 stores: a pathological image DB (image DB), which stores one or more pathological images (images) used to generate learning data; a tag DB, which stores tags added by the user to local images; and a learning data DB, which stores the generated learning data.

[0031] The display control unit 101 enables output devices such as displays 15 and other information processing devices that communicate with the data generation device 10 to display various screens involved in this embodiment. In addition, the display control unit 101 displays a screen in which the boundary lines of a local image overlap with a pathological image (image).

[0032] Input unit 102 accepts various inputs from the user. Additionally, input unit 102 accepts input of labels added to each of the plurality of partial images via input device 14. Input unit 102 stores the labels added to each partial image in a label DB. For example, input unit 102 can store a unique partial image ID corresponding to the label added to each partial image in the label DB. For example, label input can be performed by selecting the partial image for which the label is to be input from the partial images displayed on the screen and accepting the specification of the label to be added to that partial image.

[0033] In addition, when generating learning data based on multiple pathological images (multiple images), the input unit 102 can accept input of information related to the pathological type corresponding to the multiple pathological images (pathological type common to the multiple pathological images, image type corresponding to the multiple images).

[0034] The generation unit 103 generates learning data for the learning model by establishing a correspondence between each of a plurality of local images and a label added to each of the plurality of local images by the input unit 102. For example, the generation unit 103 obtains local image IDs and labels from the label database, and simultaneously extracts image data of the local images corresponding to the local image IDs from pathological images (images) stored in the pathological image database (image database). Then, the generation unit 103 generates learning data by combining the extracted image data of the local images with the labels corresponding to the local image IDs.

[0035] The image processing unit 104 changes the colors of multiple pathological images (multiple images) according to a color-changing method determined based on the pathological type (image type) input to the input unit 102. This color-changing method may include changing the RGB values ​​of each pixel in the multiple pathological images (multiple images) to standardized values, and randomly changing the RGB values ​​of each pixel in the multiple pathological images (multiple images).

[0036] <Processing Order>

[0037] Figure 4This is a flowchart illustrating an example of the processing sequence performed by the data generation device 10. First, the input unit 102 receives pathological images from the user for generating learning data and saves the input pathological images in the pathological image DB (S10). To facilitate the determination of whether an abnormality exists, the pathological images are typically stained. Examples of staining methods include hematoxylin-eosin staining, PAS staining, May-Giemsa staining, Alcian blue staining, Papanicolaou staining, azocarboxymethyl red staining, elastica-van Giesson staining, and elastica-Masson staining. It should be noted that when multiple pathological images are input, the user inputs pathological images of the same pathological type. In the following description, it is assumed that the user inputs multiple pathological images.

[0038] Next, for the multiple pathological images input in the processing sequence of step S10, the input unit 102 receives input from the user regarding information related to the pathological type (S11). The input pathological type can be either "tumor" or "hypermutation". If each of the multiple pathological images is a tumor image, the user inputs "tumor" as the pathological type; if each of the multiple pathological images is a hypermutation image, the user inputs "hypermutation" as the pathological type.

[0039] Next, the image processing unit 104 changes the colors of multiple pathological images according to the color-changing method based on the input pathological type, following the processing sequence of step S11 (S12). For example, when the pathological type is "tumor," the color-changing method can be a method of changing the RGB values ​​of each pixel in the multiple pathological images to standardized values. First, for all pixels in the multiple pathological images, the image processing unit 104 calculates the average and standard deviation of the R (Red) value, the average and standard deviation of the G (Green) value, and the average and standard deviation of the B (Blue) value. Next, the image processing unit 104 standardizes the R value of each pixel in all pixels of the multiple pathological images using the average and standard deviation of the R values. Specifically, the standardized R value can be calculated using the formula "Standardized R value = (R value - Average R value) ÷ Standard deviation of R value." The standardized G value and standardized B value are calculated by performing the same calculation on the G and B values. As a result, the color tone of the multiple pathological images can be made uniform.

[0040] Furthermore, when the pathological type is "hypermutation," the color alteration method can be a method of randomly shifting the RGB values ​​of individual pixels in multiple pathological images for each pathological image. First, the image processing unit 104 randomly determines the number of RGB value shifts for each pathological image, and the determined values ​​for shifting the R, G, and B values. For example, when the number of RGB value shifts for the first pathological image is determined to be "5," the image processing unit 104 adds 5 to the R, G, and B values ​​of each pixel contained in that pathological image. Next, when the number of RGB value shifts for the second pathological image is determined to be "-2," the image processing unit 104 adds -2 to the R, G, and B values ​​of each pixel contained in that pathological image. The image processing unit 104 repeats the same processing for all pathological images. This allows for uneven color tones in multiple pathological images.

[0041] Next, the display control unit 101 selects a pathological image from the input plurality of pathological images and displays a screen on the selected pathological image where the boundary lines of the local images overlap (S13). Next, for each local image, the input unit 102 accepts input of a label to be added (S14). When the input plurality of pathological images are tumor images, the label added by the user to each local image is any one of the following labels: a label indicating that the local image is a tumor image, or a label indicating that the local image is not a tumor image. Furthermore, when the input plurality of pathological images are hypermutation images, the label added by the user to each local image is any one of the following labels: a label indicating that the local image is a hypermutation image, or a label indicating that the local image is not a hypermutation image.

[0042] The input unit 102 saves the received tags to the tag DB by establishing a correspondence between the tags and the local image IDs. When the tag input for each local image is not completed (S15-No), the data generation device 10 repeats the processing sequence of step S14. When the tag input for each local image is completed (S15-Yes), the generation unit 103 generates learning data by establishing a correspondence between the added tags and the image data of each local image for which tags have been added (S16). The determination of whether tag input is complete can be made by checking whether the input unit 102 detects that a predetermined button (e.g., a button to start generating learning data) on the screen has been pressed. Alternatively, it can be made by checking whether tags have been added to a predetermined number of local images. Furthermore, tags are added to a predetermined number of local images, and when the ratio of added tags (the ratio of each type of tag) is within a predetermined range (e.g., the ratio of tags representing tumor images to tags representing non-tumor images is approximately 1:4), it is automatically determined that tag addition is complete.

[0043] The generation unit 103 can divide the generated learning data into folders for each label and save them to the learning data DB. For example, image data of local images labeled with tumors and image data of local images labeled without tumors can be saved in different folders.

[0044] In the processing sequence described above, local images where the proportion of non-existent tissue is greater than a predetermined percentage (local images that meet predetermined conditions) can be excluded from the learning data. For example, in the processing sequence of step S14, for local images in multiple local images of each pathological image where a specific color portion (e.g., white portion) is greater than a predetermined percentage, the input unit 102 may not accept label input. Furthermore, in the processing sequence of step S16, the generation unit 103 may exclude local images among the multiple local images from which the input unit 102 did not accept label input in the generated learning data. Therefore, it is possible to suppress the generation of learning data that is unsuitable for the learning model.

[0045] Figure 5 This is an example of a screen showing the process of adding tags. Figure 5 This example illustrates a pathological image that includes tumor cells. Menu M10 specifies the type of pathological image entered by the user. Menu M11 specifies which pathological images to display on the screen when multiple pathological images have been entered. Figure 5 The example shows an input of 40 pathological images related to tumor cells, and the third pathological image is being displayed now.

[0046] Figure 5 In the display area W10, grid-like boundary lines are overlaid on a magnified image of a portion of the pathological image. Within the display area W10, a region surrounded by the boundary lines corresponds to a local image. Additionally, the overview display area W11 displays the overall image of the pathological image, and a display frame V11 represents a magnified area of ​​the pathological image displayed in the display area W10. The position and size of the display frame V11 can be freely changed, for example, by mouse operation. After receiving changes to the position and size of the display frame V11, the input unit 102 updates the magnified pathological image displayed in the display area W10 according to the changed position and size of the display frame V11.

[0047] The size of a local image (the number of pixels vertically and horizontally) can also be predetermined and cannot be changed arbitrarily by the user, as it is based on the learning model of the learning object. Figure 5 In the example, the size of the local image is displayed as 300 pixels vertically and 300 pixels horizontally.

[0048] Additionally, the display area W10 includes a display box T10 indicating the location of a partial image, which is the image from which the user receives label input. The position of the display box T10 can be freely changed by selecting the partial image from which the user wants to input a label. By displaying the display box T10, it is possible to identify which location of the partial image the user wants to add a label to.

[0049] In addition, for a local image where the label input has been completed, the information representing the input label is overlaid on the local image. Figure 5 In the example, for a local image labeled as a tumor cell image, the text "T" is displayed in the upper left corner; for a local image labeled as not a tumor cell image, the text "N" is displayed in the upper left corner.

[0050] The display control unit 101 can display the following two pieces of information on the screen where labels are added: information indicating the ratio of each label, which is the target value of the label added to each of the multiple local images; and information indicating the actual value of the ratio of each label added to the multiple local images. For example, suppose that when the ratio of tumor cell images to non-tumor images is 1:4 as training data, the recognition accuracy of the learning model becomes higher. Figure 5 In the example, as a target value, the reference display area A10, which shows the baseline for adding labels, displays a baseline such as "The target value for the label ratio of T to N is T:N = 1:4". This indicates that it is desired to add labels in a way that achieves a ratio of 1:4 between the number of local images with label T and the number of local images with label N. Additionally, the display area N10 displays information about the ratio of added labels as the actual value representing the ratio of each label added to multiple local images. In other words, Figure 5 In the example, while the desired ratio is 1:4 for the number of local images labeled T to the number of local images labeled N, at the current moment, the ratio is 1:2. Therefore, the user is able to recognize the need to search for local images that are not tumor cell images and add labels until the ratio displayed in the display area N10 reaches 1:4.

[0051] In addition, the display control unit 101 can display the following two types of information on the screen where labels are added: information indicating the target number of local images to be tagged, and information indicating the actual number of local images to which labels have been added. For example, suppose the following condition exists: when 2000 images are used as training data, the recognition accuracy of the learning model increases. Figure 5In the example, the target value is displayed in the annotation display area A10 as a baseline such as "The target number of patches to be tagged is 2000". This indicates that the expectation is for the user to add tags to 2000 local images. Additionally, the display area N11 shows the actual number of local images with tags added. Thus, the user can understand that further tags need to be added to 1000 local images.

[0052] Multiple local images can be generated by dividing the pathological image from a reference point (e.g., the top-left pixel) to the right and downwards at predetermined intervals (e.g., 300 pixels to the right, 300 pixels downwards, etc.). Not limited to this, multiple local images can also be generated by dividing the pathological image from a point offset by a predetermined distance along the left-right and up-down directions at predetermined intervals (e.g., 300 pixels to the right-left direction, 300 pixels to the up-down direction, etc.). For example, the pathological image can be generated by dividing it to the right and downwards at predetermined intervals, starting from a point (the top-left image) offset by a predetermined distance to the right and / or downwards, using this as a reference point. The deviation value specifying region M12 specifies the region where the reference point in the pathological image is moved a few pixels to the right. The deviation value specifying region M13 specifies the region where the reference point in the pathological image is moved a few pixels downwards. When a predetermined distance is specified via the input unit 102, the display control unit 101 divides the local image and displays it on the screen based on the point after moving the specified predetermined distance from the upper left point of the pathological image.

[0053] use Figure 6 Let's illustrate with a specific example. In the case where local images are generated by dividing the pathological image at predetermined intervals to the right and downwards, with the upper left corner as the reference point, local image P1 represents the upper left local image, and local image P6 represents the local image adjacent to the right of local image P1. Assuming this state, moving towards... Figure 5 The deviation values ​​for regions M12 and M13 were input as 50 pixels and 0 pixels, respectively. In this case, a local image P2 is generated based on the point offset 50 pixels to the right from the top left of the pathological image. Similarly, assuming the deviation values ​​are... Figure 5 The deviation values ​​for regions M12 and M13 were input as 100 pixels and 0 pixels, respectively. In this case, a local image P3 is generated based on the point offset 100 pixels to the right from the top left of the pathological image.

[0054] <Summary>

[0055] According to the implementation described above, the data generation apparatus 10 divides an image into multiple local images and displays them, and for each local image, receives label input from the user. This makes it easier to generate image-related learning data.

[0056] Furthermore, the data generation device 10 modifies the pixel values ​​related to the colors of the multiple pathological images according to a pixel value modification method determined based on the image type corresponding to the multiple images. Thus, depending on the image type, the hues of the multiple images can be modified to a more suitable hue for the learning model, enabling the generation of a learning model with higher recognition accuracy.

[0057] Furthermore, the data generation device 10 displays the target value and the actual value for the ratio of each type of label added to each local image. Thus, the user can add labels to each local image to make the actual value close to the target value, thereby effectively generating learning data that is used to generate a learning model with higher recognition accuracy.

[0058] Furthermore, when dividing an image into local images, the data generation device 10 can divide the image into local images based on any point (pixel) on the image. Thus, by repeating the following operation, even with a small number of input images, the user can generate a large amount of training data: after adding labels, shifting the reference point used to divide the image into local images, and then adding labels again.

[0059] The embodiments described above are provided to facilitate a better understanding of the present invention and should not be construed as limiting the invention. The flowcharts, sequences, various elements of the embodiments, and their configurations, materials, conditions, shapes, and dimensions described in the embodiments are not limited to the examples and can be appropriately modified. Furthermore, the configurations represented by different embodiments can be partially substituted for or combined with each other.

[0060] Explanation of reference numerals in the attached figures

[0061] 1…Image processing system; 10…Data generation device; 11…Processor; 12…Storage device; 13…Communication IF; 14…Input device; 15…Output device; 20…Diagnostic device; 100…Storage unit; 101…Display control unit; 102…Input unit; 103…Generation unit; 104…Image processing unit.

Claims

1. A data generation apparatus, comprising: The display control unit overlays the pathological image with the boundary lines of multiple local pathological images generated by dividing the pathological image into multiple images and displays them on the screen. The input unit accepts pathology-related labels added to each of the plurality of local pathology images; and The generation unit generates learning data for the learning model by establishing a correspondence between each of the plurality of local pathological images and the pathology-related labels added to each of the plurality of local pathological images. The input unit will uniquely identify the local pathological image ID of each of the plurality of local pathological images and establish a correspondence between the local pathological image ID and the pathology-related tag added to each of the plurality of local pathological images, and store them in the tag database. The generation unit obtains the local pathological image ID and pathology-related tags from the tag database, extracts image data of the local pathological image corresponding to the local pathological image ID from the pathological images stored in the pathological image database, and generates learning data by combining the extracted image data of the local pathological image with the pathology-related tags corresponding to the local pathological image ID. The pathological images include multiple pathological images. When the generation unit generates the learning data, the input unit accepts input information related to the pathology type corresponding to the plurality of pathology images. The display control unit displays both information indicating the target number of local pathological images to be tagged and information indicating the actual number of local pathological images with tags added.

2. The data generation apparatus according to claim 1, wherein, The data generation apparatus further includes a modification unit that modifies pixel values ​​related to the colors of the plurality of pathological images according to a modification method determined based on the pathological type input to the input unit.

3. The data generation apparatus according to claim 2, wherein, The modification method includes: a method for changing the RGB values ​​of each pixel in the plurality of pathological images to standardized values, and a method for randomly changing the RGB values ​​of each pixel in the plurality of pathological images.

4. The data generation apparatus according to any one of claims 1 to 3, wherein, The display control unit displays the following two types of information: information representing the ratio of each pathology-related label as a target value of the pathology-related label added to each of the plurality of local pathology images; and information representing the actual value of the ratio of each pathology-related label added to the plurality of local pathology images.

5. The data generation apparatus according to any one of claims 1 to 3, wherein, For local pathological images among the plurality of local pathological images that meet predetermined conditions, the input unit does not accept input of pathology-related labels. The generation unit excludes local pathological images from the learning data that did not receive input of pathology-related labels among the plurality of local pathological images.

6. The data generation apparatus according to any one of claims 1 to 3, wherein, The plurality of local pathological images are generated by dividing the pathological image at predetermined intervals in the right-left and up-down directions, using a point obtained by offsetting a predetermined distance from a point that serves as a reference point within the pathological image. The input unit accepts the specification of the predetermined distance.

7. A data generation method, executed by a data generation device, the data generation method comprising the following steps: The pathological image is overlapped with the boundary lines of multiple local pathological images generated by dividing the pathological image into multiple images and displayed on the screen; Accepts input of pathology-related labels for each of the plurality of local pathology images; and By establishing a correspondence between each of the plurality of local pathological images and the pathology-related labels added to each of the plurality of local pathological images, learning data for the learning model to learn is generated. In the step of receiving the input, the local pathological image ID that uniquely identifies each of the plurality of local pathological images is correspondingly stored in the tag database with the pathology-related tag added to each of the plurality of local pathological images. In the step of generating the learning data, the local pathological image ID and pathology-related tags are obtained from the tag database; image data of the local pathological image corresponding to the local pathological image ID is extracted from the pathological images stored in the pathological image database; and the learning data is generated by combining the extracted image data of the local pathological image with the pathology-related tags corresponding to the local pathological image ID. The pathological images include multiple pathological images. In the data generation method, In generating the learning data, input is accepted that relates to the pathological type corresponding to the plurality of pathological images. The system displays both a target value representing the number of local pathological images to be tagged and an actual value representing the number of local pathological images that have been tagged.

8. A computer program product comprising a program that causes a computer to perform the following steps: The pathological image is overlapped with the boundary lines of multiple local pathological images generated by dividing the pathological image into multiple images and displayed on the screen; Accepts input of pathology-related labels for each of the plurality of local pathology images; and By establishing a correspondence between each of the plurality of local pathological images and the pathology-related labels added to each of the plurality of local pathological images, learning data for the learning model to learn is generated. In the step of receiving the input, the local pathological image ID that uniquely identifies each of the plurality of local pathological images is correspondingly stored in the tag database with the pathology-related tag added to each of the plurality of local pathological images. In the step of generating the learning data, the local pathological image ID and pathology-related tags are obtained from the tag database; image data of the local pathological image corresponding to the local pathological image ID is extracted from the pathological images stored in the pathological image database; and the learning data is generated by combining the extracted image data of the local pathological image with the pathology-related tags corresponding to the local pathological image ID. The pathological images include multiple pathological images. The program causes the computer to perform the following steps: In generating the learning data, input is accepted that relates to the pathological type corresponding to the plurality of pathological images. In the step of overlapping and displaying a pathological image with the boundary lines of multiple local pathological images generated by dividing the pathological image into multiple images, both information representing the target value of the number of local pathological images to be tagged and information representing the actual value of the number of local pathological images with tags are displayed.

Images