Diagnosis result generating system and method

Abstract

The present invention discloses a system and a method thereof, which outputs the results of disease diagnosis based on biological tissue images in a form that can be understood by both machines and humans. According to one aspect of the present invention, the provided result generation system includes: an annotation information generation module, which generates annotation information reflecting the diagnosis result of whether there is a disease on a biological image slide, wherein the annotation information includes disease status information of each pixel of the slide; a contour extraction module, which extracts at least one contour from the annotation information; and an XML generation module, which generates an XML document of each contour line information of the extracted at least one contour.

Description

Technical Field

[0001] The present invention relates to a diagnosis result generation system and method thereof, and more particularly to a system and method thereof that can output the diagnosis result of a disease through biological tissue images in a form that can be understood by both machines and humans. Background Art

[0002] One of the main tasks of pathology or pathology department is to read the patient's biological images, and the diagnosis is to determine the status or symptoms of a specific disease. This diagnosis method has long relied on the rich experience and knowledge of medical staff.

[0003] Recently, with the development of machine learning, there are attempts to automate tasks such as image recognition or classification using computers. In particular, attempts are being made to use a type of machine learning, namely neural networks (for example, deep learning methods using convolutional neural networks (CNNs)) to automate diagnostic tasks that were previously performed by experienced medical personnel.

[0004] In particular, diagnosis through deep learning using neural networks (such as CNN) does not simply automate the rich experience and knowledge of senior medical personnel in the past, but finds characteristic factors and derives the desired answers through self-learning. Sometimes, it can even discover characteristics of disease factors in images that senior medical personnel are not aware of.

[0005] Typically, when diagnosing a disease through a biological image-based neural network, fragments of biological images, i.e., image patches (also called image tiles), are used. That is, medical personnel who are familiar with the image sub-patch will annotate the state of a specific disease (e.g., whether cancer is expressed), and use these annotated large amounts of image sub-patch as training data to learn the neural network. In this case, the neural network can be a convolutional neural network.

[0006] However, in this way, the learned neural network can only judge the disease state of the image sub-block according to the image features of the corresponding image sub-block. In fact, when judging the state of a specific biological tissue for a specific disease, it is necessary not only to consider the specific biological tissue itself, but also to consider the state of the surrounding tissue of the specific biological tissue (for example, shape, whether a specific pattern exists, etc.). However, the disadvantage of the traditional method is that it is not applicable to such situations.

[0007] On the other hand, in traditional learning, the color of the biological image or image patch itself will be input as input data. That is, the input data usually defined by the three channel values ​​of RGB will be used directly.

[0008] However, in this case, depending on the characteristics of the staining reagent used to stain the biological tissue corresponding to the biological image, the color of the stained tissue may vary, which will directly affect the learned neural network.

[0009] Therefore, it may be necessary to strengthen the learning method of the neural network to cope with non-root color features caused by staining, etc., which are not root tissue image features.

[0010] In addition, according to the diagnosis results of the image block unit, whether the disease is expressed or not is determined according to each image block. When the diagnosis results of each image block are directly visualized, the problem that the part that is not a tissue (cell tissue) may actually be visualized. Therefore, a visualization method that can accurately grasp the tissue part diagnosed as a disease may be required.

[0011] On the other hand, after the biological tissue is diagnosed, in order to facilitate the management of the results or share them with other related systems, the diagnosis results need to be organized in an appropriate manner and format.

[0012] *Previous Technical Literature

[0013] -Patent Literature

[0014] Korean Patent Publication No. 10-2016-0034814 “Client device accompanying neural network and system including the same”. Summary of the invention

[0015] Technical issues

[0016] The technical problem that the present invention aims to solve is to provide a neural network-based diagnostic system and method that uses the specific image block and surrounding image blocks for learning to further improve the accuracy, so as to judge the state of the disease related to the specific image block (for example, whether the disease is expressed or indicators showing the disease state, etc.).

[0017] In addition, a diagnosis system and method thereof are provided, wherein the neural network utilized has strong color features that are not root cause image features when diagnosing whether a disease is expressed or not.

[0018] In addition, a visualization method capable of accurately grasping the tissue site of disease expression and a neural network-based diagnosis system capable of executing the method are also provided.

[0019] In addition, a system and method are provided that can output the results of disease diagnosis based on images of biological tissues in a form that can be understood by both machines and humans.

[0020] Technical solutions

[0021] According to one aspect of the present invention, a diagnosis result generation system is provided, comprising: an annotation information generation module, which generates annotation information reflecting the diagnosis result of whether a disease exists on a biological image slide - wherein the annotation information includes disease status information of each pixel of the slide -; a contour extraction module, which extracts at least one contour from the annotation information; and an XML generation module, which generates an XML document including each contour line information of the at least one extracted contour.

[0022] In one embodiment, the disease state information may include normal, Gleason mode 3, Gleason mode 4 and Gleason mode 5.

[0023] In one embodiment, the XML document may further include the extracted disease status information of the at least one profile.

[0024] In one embodiment, it may be characterized in that the disease is prostate cancer.

[0025] In one embodiment, the diagnosis result generating system further includes a diagnosis system that uses the slide and the neural network to output a diagnosis result of whether the disease exists, and the annotation information generating module can generate the annotation information according to the diagnosis result output by the diagnosis system.

[0026] In one embodiment, the diagnostic system includes: an image block neural network, which generates an image block-level diagnostic result of whether the slide is divided into each preset image block of a preset size and whether there is a disease; a heat map generation module, which generates an image block-level heat map image corresponding to the slide based on the diagnosis result of each image block in the multiple image blocks contained in the slide; a tissue mask image generation module, which generates a tissue mask image corresponding to the slide based on the HSV (Hue-Saturation-Value) model corresponding to the slide; and a diagnostic result visualization module, which generates a diagnostic result corresponding to the slide based on the image block-level heat map image and the tissue mask image.

[0027] In one embodiment, the tissue mask generation module performs image binarization on the S space corresponding to the slide and generates a first binarization result (here, the S space corresponding to the slide refers to the saturation value space of the HSV model corresponding to the slide), performs image binarization on the `1-V space corresponding to the slide and generates a second binarization result (here, the V space corresponding to the slide is the nominal value space of the HSV module corresponding to the slide), and a tissue mask image corresponding to the slide can be generated based on the first binarization result and the second binarization result.

[0028] In one embodiment, the diagnosis result visualization module can apply a conditional random field to the image block level thermal map image and the tissue mask image to generate a disease diagnosis visualization image corresponding to the slide.

[0029] According to another aspect of the present invention, a method for generating a diagnosis result is provided, comprising: an annotation information step of generating annotation information reflecting the diagnosis result of whether a disease exists on a biological image slide (wherein the annotation information includes disease status information of each pixel on the slide); a contour extraction step of extracting at least one contour from the annotation information; and an XML generation step of generating an XML document containing contour line information of at least one extracted contour.

[0030] In one embodiment, the diagnosis result generating method further includes: a diagnosis step of using the slide and the neural network to output the diagnosis result of whether the disease exists, and the annotation information generating step may include a step of generating the annotation information according to the diagnosis result output by the diagnosis system.

[0031] In one embodiment, the diagnosis step includes: an image block level diagnosis step, using a neural network to generate an image block level diagnosis result of whether the slide is divided into each preset image block of a preset size and whether there is a disease; a heat map generation step, based on the diagnosis result of each image block in the multiple image blocks contained in the slide, generating an image block level heat map image corresponding to the slide; a tissue mask image generation step, based on the HSV (Hue-Saturation-Value) model corresponding to the slide, generating a tissue mask image corresponding to the slide; and a diagnosis result visualization step, based on the image block level heat map image and the tissue mask image, generating a diagnosis result corresponding to the slide.

[0032] In one embodiment, the step of generating the tissue mask includes: a step of binarizing the S space corresponding to the slide and generating a first binarization result (here, the S space corresponding to the slide refers to the saturation value space of the HSV model corresponding to the slide); a step of binarizing the `1-V space corresponding to the slide and generating a second binarization result (wherein, the V space corresponding to the slide is the nominal value space of the HSV model corresponding to the slide); and a step of generating a tissue mask image corresponding to the slide based on the first binarization result and the second binarization result.

[0033] In one embodiment, the diagnosis result visualization step may apply a conditional random field to the image block level heat map image and the tissue mask image to generate a disease diagnosis visualization image corresponding to the slide.

[0034] According to another aspect of the present invention, there is provided a computer program which is installed in a data processing device and recorded on a medium in order to implement the method.

[0035] Effects of the Invention

[0036] According to the technical idea of ​​the present invention, when diagnosing a specific image block, what is considered is a macroscopic image block, that is, it includes not only the specific image block but also the surrounding image blocks, and provides a neural network that can judge the disease state of the specific image block, thereby effectively providing higher diagnostic accuracy.

[0037] In addition, according to the technical idea of ​​the present invention, the input data includes not only the original color value of the image block (for example, RGB 3-channel value), but also the grayscale value as input data, so it is possible to prevent the image features related to the disease from being ignored due to the color difference that may occur when only the grayscale value is used. At the same time, in the diagnosis of whether the disease is expressed or not, there may be strong features of the variables (variaton) caused by various factors of color, rather than the root image features.

[0038] In addition, according to the technical concept of the present invention, the effect produced is that within the image block for judging the disease expression, only the tissue part can be distinguished and visualized.

[0039] Furthermore, according to the technical concept of the present invention, the result of diagnosing a disease through a biological tissue image can be output in a form that can be understood by both machines and humans. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to better understand the figures referenced in the detailed description of the present invention, a brief description of each figure is provided.

[0041] Figure 1 The figure shows a schematic diagram of the system structure of the diagnosis result generating system involved in the technical concept of the present invention.

[0042] Figure 2a What is shown is a schematic diagram of the logical structure of the diagnosis result generation system involved in the technical concept of the present invention.

[0043] Figure 2b It is a logical structure diagram of a disease diagnosis system using a neural network involved in the technical idea of ​​the present invention.

[0044] Figure 3It is a schematic diagram of the hardware structure of the diagnosis result generating system involved in the embodiment of the present invention.

[0045] Figure 4a It is a graphic diagram that marks the diseased area on a slide.

[0046] Figure 4b It is a schematic diagram of an example of an XML document generated by a diagnosis result generation system according to an embodiment of the present invention.

[0047] FIG. 5 is a schematic diagram of an exemplary structure of a neural network involved in an embodiment of the present invention.

[0048] Figure 6 It is a schematic diagram of an exemplary structure of a neural network involved in another embodiment of the present invention.

[0049] Figure 7a is a schematic diagram of an example of a slide.

[0050] Figure 7b yes Figure 7a A schematic diagram of an example of a heat map in the slide.

[0051] Figure 7c About Figure 7a Schematic diagram of an example of a tissue mask image from a slide.

[0052] Figure 7d This is a schematic diagram showing an example of a visualization result image of a site diagnosed as a disease. DETAILED DESCRIPTION

[0053] The present invention is capable of various transformations and embodiments, and specific embodiments will be illustrated in the drawings and specifically described in the detailed description. However, this is not intended to limit the present invention to a specific implementation, but should be understood to include all transformations, equivalents or substitutes within the scope of the present invention. When describing the present invention, if it is considered that a detailed description of the known technology would obscure the main points of the present invention, the detailed description will be omitted.

[0054] The terms "first", "second", etc. can be used to describe various components, but the components cannot be limited by the terms. The terms are only used to distinguish one component from other components.

[0055] The terms used in the present application are only used to describe specific embodiments and are not intended to limit the present invention. An expression in the singular includes an expression in the plural unless there is an obviously different meaning in the context.

[0056] The terms "including" or "having" in this specification refer to the existence of features, numbers, steps, actions, components, parts or combinations thereof recorded in the specification, and should be understood as not excluding the existence or additional possibility of one or more other features, numbers, steps, actions, components, parts or combinations thereof in advance.

[0057] In addition, in this specification, when a component "transmits" data to another component, it means that the component can directly transmit the data to the other component or transmit the data to the other component through at least one other component. On the contrary, if any component "directly transmits" data to another component, it means that the data will be transmitted from the component to the other component, not through the other component.

[0058] Hereinafter, the present invention will be described in detail around the embodiments of the present invention with reference to the accompanying drawings. The same reference symbols provided in each figure represent the same components.

[0059] The diagnosis result generation method involved in the technical idea of ​​the present invention can be executed by a diagnosis result generation system.

[0060] Figure 1 The figure shows a schematic diagram of the system structure of the diagnosis result generating system involved in the technical concept of the present invention.

[0061] Reference Figure 1 The diagnostic result generation system 100 involved in the technical idea of ​​the present invention can be installed on a preset server 10 to implement the technical idea of ​​the present invention. The server 10 refers to a data processing device with computing power capable of implementing the technical idea of ​​the present invention. A data processing device to which a client can be connected through a network, as well as any device capable of performing a specific service such as a personal computer and a mobile terminal can be defined as a server. Ordinary experts in the technical field of the present invention can easily infer this.

[0062] like Figure 3 As shown, the server 10 may include a processor 11 and a memory 12. The processor 11 may refer to a computing device capable of driving a program (12-1) that implements the technical concept of the present invention. The processor 11 may generate a diagnosis result of a specific disease (such as prostate cancer). The diagnosis result may include information about a specific disease site and / or disease status information of each site (such as the degree of progression of the disease, etc.) when a disease is present on a biological tissue image slide.

[0063] In addition, the processor 11 may utilize the program (12-1) and a neural network (12-2) for diagnosis. As described below, the neural network (12-2) may include an image block neural network for performing image block level diagnosis.

[0064] The memory 12 may refer to a data storage means for storing the program (12-1) and the neural network (12-2), and may also be implemented by a variety of storage means according to the embodiment. In addition, the memory 12 refers to not only the main memory included in the server 10, but also the temporary memory or memory that may be included in the processor 11.

[0065] Although the diagnosis result generating system 100 Figure 1 or Figure 3 What is shown in the figure is implemented by any one physical device, but ordinary experts in the technical field of the present invention will be able to easily infer that multiple physical devices can be organically combined according to the technical ideas of the present invention and as needed to implement the diagnosis result generating system 100.

[0066] In this specification, the generation of the diagnosis result by the diagnosis result generation system 100 may refer to a series of processes of generating an electronic document including annotation information of the part diagnosed as a disease. In particular, the diagnosis result generation system 100 may generate a document in the form of an extensible markup language (XML) such as annotation information embodying the diagnosis result. An XML document is a document encoded in a format that can be recognized by both humans and machines (computers), and can be implemented in a standardized manner through the World Wide Web Consortium (WWWC).

[0067] In one embodiment, the diagnosis result generation system 100 can also use biological tissue image slides to be diagnosed by humans or other systems, and then generate an XML document using the annotated parts that are judged to be diseased. In other words, the disease diagnosis itself is performed by humans or other systems, and only the function of outputting in the form of an XML document is performed by the diagnosis result generation system 100.

[0068] On the other hand, in another embodiment, the diagnosis result generating system 100 can directly diagnose the disease through the slides and output the diagnosis result in the form of an XML document.

[0069] When the diagnosis result generating system 100 directly performs diagnosis, the diagnosis result generating system 100 can execute a series of processes of outputting output data defined in this specification by inputting a biological image expressed by biological tissue, i.e., the entire image block of a slide or a partial image block of the slide, which will be explained later.

[0070] If the diagnosis result generation system 100 is included in a preset server 10 and implemented, the diagnosis result generation system 100 can also communicate with at least one client (e.g., 20, 20-1) that can be connected to the server 10. In this case, the client (e.g., 20, 20-1) can transmit a biological image to the diagnosis result generation system 100, and the diagnosis result generation system 100 can perform a diagnosis based on the technical idea of ​​the present invention on the transmitted biological image. Moreover, the diagnosis result can also be sent to the client (e.g., 20, 20-1).

[0071] Figure 2a It is a schematic diagram of the logical structure of a diagnosis result generating system involved in one embodiment of the present invention.

[0072] refer to Figure 2a , the diagnosis result generation system 100 may include an annotation information generation module 110, a contour extraction module 120, and an XML generation module 130. According to an embodiment, the diagnosis result generation system 100 may also include a diagnosis system 200. According to a nominal embodiment, some parts of the above components may not necessarily be components necessary for implementing the present invention, and according to an embodiment, the diagnosis result generation system 100 should likely include more components. For example, the diagnosis result generation system 100 may further include a communication module (not shown) for communicating with the client (e.g., 20, 20-1).

[0073] The diagnosis result generation system 100 may refer to a logical configuration of hardware resources and / or software required to implement the technical idea of ​​the present invention, and does not necessarily mean a physical component or a device. That is, the diagnosis result generation system 100 may refer to a logical combination of hardware and / or software for implementing the technical idea of ​​the present invention, and may be configured to be installed on devices spaced apart from each other, and to perform respective functions to implement a collection of logical components of the technical idea of ​​the present invention when necessary. In addition, the diagnosis result generation system 100 may also refer to a collection of components configured according to respective functions or roles for implementing the technical idea of ​​the present invention. The annotation information generation module 110, the contour extraction module 120, the XML generation module 130 and / or the diagnosis system 200 may be located in different physical devices, or in the same physical device. According to another embodiment, the combination of software and / or hardware that respectively constitute the annotation information generation module 110, the contour extraction module 120, the XML generation module 130 and / or the diagnosis system 200 may also be located in different physical devices, and the components located in different physical devices may be organically combined with each other and configured into the modules.

[0074] In addition, the module in this specification may refer to a functional and structural combination of hardware for executing the technical idea of ​​the present invention and software for driving the hardware. For example, the module may refer to a logical unit for executing a preset code and a hardware resource of the preset code, and does not necessarily refer to a physically connected code or a type of hardware, which can be easily deduced by ordinary experts in the technical field of the present invention.

[0075] The annotation information generating module 110 can generate annotation information reflecting the diagnosis result of whether there is a disease on the biological image-slide. At this time, the annotation information can include the disease status information of each pixel in the slide.

[0076] For example, if a slide is composed of N*M pixels, the annotation information corresponding to the slide is in the form of an N*M matrix, and each value in the matrix may have status information of a corresponding pixel.

[0077] On the other hand, the disease state may include normal, Glisson mode 3, Glisson mode 4, and Glisson mode 5, and the annotation information may have a value corresponding to a specific state. For example, each disease state information included in the annotation information may have a value of 0 when the corresponding pixel is normal, a value of 1 when it is Glisson mode 3, a value of 2 when it is Glisson mode 4, and a value of 3 when it is Glisson mode 5.

[0078] Figure 4a is a schematic diagram of an image where the diseased area is annotated on a slide. Figure 4a As shown, the disease area can be annotated with a specific color according to the disease state. The annotation information generation module 110 can set the value of the annotation information corresponding to each pixel in the annotated area to the value corresponding to the disease state of the corresponding pixel, thereby generating the annotation information.

[0079] The contour extraction module 120 can extract at least one contour from the annotation information. In the slide, a contour can be extracted for each individual area annotated as a disease.

[0080] On the other hand, the XML generation module 130 may generate an XML document including the slide diagnosis result (more specifically, the annotation result called disease area).

[0081] The generated XML document may include each contour line information of the at least one contour extracted by the contour extraction module 120. In addition, the XML document may further include each disease state information of the at least one contour extracted.

[0082] Figure 4b is a diagram of an example of the generated XML document. Figure 4b In the embodiment shown, each annotation area can be represented by a <annotation>The disease status information of the corresponding region can be specified as a parameter named class. On the other hand, each point constituting the contour line corresponding to the annotated region can be stored by at least one <coordinate> tag.

[0083] As described above, according to the technical concept of the present invention, the diagnostic results of the slide can be classified into different annotation areas and generated in the form of XML through a standardized format. XML documents are a document format that can be recognized by both humans and machines, so it is convenient for other diagnosticians to grasp the diagnostic results through it, and the diagnostic results can be easily shared with other systems.

[0084] On the other hand, as described above, the diagnosis result generation system 100 involved in an embodiment of the present invention can directly diagnose the disease. To this end, the diagnosis result generation system 100 can further include a diagnosis system 200 that uses a neural network for diagnosis. The diagnosis system 200 will be described in detail below.

[0085] According to one embodiment of the present invention, the diagnostic system 200 can perform image block level diagnosis. Image block level diagnosis can refer to segmenting the slide into image block units and diagnosing whether the segmented image blocks express the disease. Therefore, the diagnostic system 200 can receive input from the image blocks on the slide and output whether the disease is expressed on the corresponding image blocks. To this end, the neural network should be able to learn and implement.

[0086] On the other hand, according to the technical ideas of the present invention, the neural network that performs image block level diagnosis can not only use the image block for diagnosis, but also further consider the surrounding image blocks of the image block to perform diagnosis. These technical ideas were specifically disclosed in the Korean patent application applied by the present applicant (application number 10-2016-0168176, disease diagnosis system and method using neural network, hereinafter referred to as "previous application"). In this way, the accuracy of diagnosis can be improved when the very peripheral area, that is, the area surrounding it is considered at the same time compared with performing diagnosis by considering the area corresponding to the image block alone. In addition, according to the technical ideas of the present invention, not only on the surrounding image blocks of a specific image block, but also by further comprehensively considering the physical characteristics such as the position, density and cluster size of each image block on the entire slide, it is possible to more accurately determine whether there is a disease on the slide. The previous application is included as a reference material for the present invention, and its content can be regarded as the content contained in this specification.

[0087] On the other hand, the state information outputted according to the image block level diagnosis may refer to the probability information of whether a specific disease (e.g., a specific type of cancer) is expressed in the tissue corresponding to the image block. The diagnostic system 200 may determine that the image block is an image block expressing a disease (e.g., prostate cancer) when a probability above a specific standard value (threshold) occurs.

[0088] Of course, the information displayed by the diagnostic system 200 is not only the expression of a specific disease disclosed in the previous application, but also the degree of progression of the specific disease (or the probability corresponding to the degree of progression). For example, when the technical idea of ​​the present invention is applied to the diagnosis of prostate cancer, the "Gleason Pattern" or "Gleason Score", an indicator of the degree of progression of prostate cancer, can be included in the state information output by the neural network. For example, the Gleason score ranges from 2 to 5, and the larger the number, the more serious the degree of prostate cancer expression. Therefore, the state information may refer to the probability that the biological tissue corresponding to the image block that becomes the diagnostic object corresponds to a specific value of the Gleason score (for example, 3, 4 or 5).

[0089] There may be multiple state information. For example, the first state information may represent the probability of a Gleason score of 3, the second state information may represent the probability of a Gleason score of 4, and the third state information may represent the probability of a Gleason score of 5. The state channels corresponding to these first state information, second state information, and third state information may all be defined on the output layer. According to an embodiment, the state information may also be defined as the probability of representing a certain range of Gleason scores (e.g., 3 to 5, 4 to 5, etc.). In other words, one state information may correspond to multiple indicators expressing the state of disease progression.

[0090] In this case, when the state information of the neural network having a Gleason score of 3 or above exceeds a preset threshold, it can be determined that the image block is a disease image block, that is, an image block expressing a disease.

[0091] On the other hand, the threshold used by the neural network can be set in a variety of ways. According to the embodiment, multiple thresholds can be used. Of course, according to the threshold, a specific image block can be judged as an image block expressing a disease, that is, a disease image block, or can be judged as a normal image block.

[0092] According to the technical idea of ​​the present invention, the neural network may use multiple thresholds. In this case, the diagnosed disease image block may be different according to each value of the multiple thresholds.

[0093] On the other hand, the diagnostic system 200 can also visualize the performed image block level diagnostic results. In this specification, visualization of the diagnostic results may refer to giving a preset visual effect to the part of the diagnostic result judged to be a disease. The visual effect may mean the effect of visual perception. For example, the part judged to be a disease through visualization can be represented or emphasized with different colors.

[0094] The diagnosis system 200 can use the neural network involved in the technical concept of the present invention to perform image block level diagnosis. Of course, in order to perform such diagnosis, the process of learning the neural network can be performed first.

[0095] Therefore, the diagnostic system 200 may be a system that receives a neural network learned based on the technical concept of the present invention from the outside and performs diagnosis using the neural network, or may be a system that performs the learning of the neural network. In addition, the diagnostic system 100 may be a dedicated device made to realize the technical concept of the present invention, rather than a general-purpose data processing device. In this case, it may also have a means for scanning biological images, etc.

[0096] The neural network may have the characteristic of performing the diagnosis of the specific image block, that is, when diagnosing the specific image block, not only the image of the specific image block itself as disclosed in the previous application but also the image of at least one image block adjacent to the specific image block should be considered. Through these technical ideas, in order to diagnose the biological tissue corresponding to the specific image block, not only the biological tissue but also the state of the surrounding tissue of the biological tissue should be considered, which is very meaningful for the diagnosis of the disease and can effectively improve the accuracy. In addition, when the biological image is segmented into multiple image blocks, the different segmentation methods of the image blocks or the part of the biological tissue in which the segmented area is may have a huge impact on the diagnosis result.

[0097] Of course, as mentioned above, the neural network may not have the features disclosed in the previous application. In any case, the neural network can be a learned neural network for diagnosing each image block.

[0098] Now, the neural network can accept an additional channel as input value for each pixel in the image block, which is different from the past. The additional channel may be the grayscale value of each pixel. Therefore, the neural network can accept input for each image block, and accept 3 channels of the original value (e.g., RGB) of each pixel contained in the image block as input, as well as an additional additional channel - the gray channel.

[0099] In this case, a powerful effect can be achieved when the color of biological images may change due to factors unrelated to disease-related image features (e.g., characteristics of the diagnostic agency, staining reagents, etc.). Of course, problems may occur when simply using the gray channel without using the original value, that is, when disease-related image features are reflected and expressed in color, this important information may not be reflected in the learning, and these problems can also be solved.

[0100] The neural network used by the diagnostic system 200 according to the technical concept of the present invention may include a microscopic neural network and a macroscopic neural network.

[0101] The micro neural network may refer to a network that uses a specific image block for learning and uses the image features of the image block itself to diagnose the image block.

[0102] The macro neural network may refer to a network that utilizes not only the image block, but also a macro image block including the image block and at least one adjacent image block of the image block for learning, and utilizes the overall image features of the macro image block to diagnose the image block through a series of processes.

[0103] Therefore, the characteristic of the neural network involved in the technical idea of ​​the present invention when diagnosing the specific image block is that when diagnosing the specific image block, not only the image of the specific image block itself is considered, but also the image of at least one adjacent image block of the specific image block is considered. Through this technical idea, in fact, when diagnosing the biological tissue corresponding to the specific image block, not only the biological tissue but also the state of the surrounding tissue of the biological tissue should be considered, which is very helpful for the diagnosis of the disease and can effectively improve the accuracy. In addition, when the biological image is divided into multiple image sub-blocks, the division method of the image sub-blocks or the position of the divided area in the biological tissue may have a huge impact on the diagnosis result.

[0104] In order to realize this technical concept, the diagnostic system 200 may logically have Figure 2b Same configuration.

[0105] Figure 2b It is a schematic diagram of the logical structure of a diagnostic system 200 for diagnosing a disease using a neural network according to an embodiment of the present invention.

[0106] refer to Figure 2b , the diagnostic system 200 may include a control module 210, a neural network module 220 storing a neural network, a preprocessing module 230, a heat map generation module 240, a tissue mask generation module 250, and a diagnostic result visualization module 260. According to an embodiment of the present invention, some of the components may not necessarily correspond to the necessary components required to implement the present invention, and the diagnostic system 200 involved in the embodiment may of course include more components.

[0107] The diagnostic system 200 may refer to a logical configuration of hardware resources (resources) and / or software required to implement the technical idea of ​​the present invention, and does not necessarily mean a physical component or a device. That is to say, the diagnostic system 200 may refer to a logical combination of hardware and / or software for implementing the technical idea of ​​the present invention, and when necessary, it may be configured to be installed on devices spaced apart from each other to perform their respective functions to implement a collection of logical components of the technical idea of ​​the present invention. In addition, the diagnostic system 200 may also refer to a collection of components configured according to their respective functions or roles to implement the technical idea of ​​the present invention. For example, the control module 210, the neural network module 220, the preprocessing module 230, the heat map generation module 240, the tissue mask generation module 250 and / or the diagnostic result visualization module 260 may be in different physical devices, or in the same physical device. According to another embodiment, the combination of software and / or hardware that respectively constitute the control module 210, the neural network module 220, the preprocessing module 230, the thermal map generation module 240, the tissue mask generation module 250 and / or the diagnosis result visualization module 260 may also be in different physical devices, and the components in the different physical devices may be organically combined with each other and configured into the modules.

[0108] The control module (210) includes functions and / or resources for controlling other configurations included in the diagnostic system 200 (for example, the neural network module 220, the preprocessing module 230, the thermal map generation module 240, the tissue mask generation module 250, the diagnostic result visualization block 260, etc.).

[0109] The control module 210 can use the neural network stored in the neural network module 120 to perform the image block level diagnosis involved in the technical idea of ​​the present invention. As mentioned above, the image block level diagnosis can be achieved by the neural network based on deep learning involved in the technical idea of ​​the present invention. That is, under the control of the control module 210, the neural network can generate an image block level diagnosis result of whether there is a disease in each preset image block of a preset size into which the slide is divided.

[0110] The neural network may refer to a collection of information expressing a series of design matters defining the neural network. In this specification, the neural network may be a convolutional neural network.

[0111] As is well known, the convolutional neural network may include an input layer, multiple hidden layers and an output layer. Each of the multiple hidden layers may include a convolutional layer and a pooling layer (or a subsampling layer).

[0112] Convolutional neural networks can define each layer by function, filter, convolution stride, weight factor, etc. In addition, the output layer can be defined as a fully connected feedforward layer.

[0113] The design matters of each layer constituting a convolutional neural network are well known. For example, the number of layers included in the multiple layers, the convolution function, the pooling function, and the excitation function used to define the multiple layers can use known functions, or can use functions defined separately for realizing the technical concept of the present invention.

[0114] According to an embodiment of the present invention, the neural network for performing image block level diagnosis uses the public DenseNet. In this case, the design may not only consider the specific image block that is the object of diagnosis, but also the surrounding image blocks, as disclosed in the previous application. In addition, various neural networks may be used. In any case, the neural network may be defined to receive input of a specific image block and output a feature value corresponding to the probability of disease expression of the specific image block.

[0115] The control module 210 can receive the input data of the neural network stored in the neural network module 220, that is, the learned neural network, that is, the input of each image block. At this time, as described above, the value with the gray channel value added to the original value can be input. Of course, the gray channel value can be obtained by converting the value of the pixel into a grayscale value. And the output data can be output after executing the calculation defined by the neural network, that is, the characteristic value corresponding to the disease expression probability corresponding to the image block. In addition, according to the embodiment, in order to perform the slide level diagnosis described later, it is also possible to output whether the image block expresses the disease based on whether the characteristic value is a preset threshold.

[0116] The neural network module 220 may store a neural network. The neural network may refer to a collection of information expressing a series of design items defining a neural network. In this specification, the neural network may be a convolutional neural network.

[0117] As is known to all, the convolutional neural network may include an input layer, multiple hidden layers, and an output layer. Each convolutional neural network of multiple hidden layers may define each layer by functions, filters, convolution strides, weight factors, etc. In addition, the output layer may be defined as a fully connected feedforward layer.

[0118] The design matters of each layer constituting a convolutional neural network are well known. For example, the number of layers included in the multiple layers, the convolution functions, pooling functions, and activation functions that define the multiple layers can use known functions, or can use functions that are individually defined to implement the technical concept of the present invention.

[0119] For example, the convolution function may be discrete convolution and the like. The pooling function may be max-pooling, average-pooling, etc. The activation function may be sigmoid, tanh, RELU (Rectified Linear Unit), etc.

[0120] After defining the design items of these convolutional neural networks, the convolutional neural network with the defined design items can be stored on a storage device. Furthermore, after the convolutional neural network is learned, the weight factor corresponding to each layer can be specific.

[0121] That is to say, the learning of the convolutional neural network can refer to a process determined by the weight factors of each layer. Moreover, after the convolutional neural network is learned, the learned convolutional neural network can accept input data on the input layer and output output data through the pre-defined output layer.

[0122] The neural network involved in the embodiments of the present invention can be defined by selecting any one or more of the well-known design items, or independent design items can be defined for the neural network.

[0123] The diagnostic module 110 may input input data into the neural network stored in the neural network module 220, ie, the learned neural network, and may output output data after executing the calculation defined by the neural network.

[0124] The preprocessing module 230 can use a neural network to preprocess the biological image required before diagnosis. For example, the preprocessing of the biological image may include an image segmentation process of segmenting the biological image (i.e., slide) into image blocks (image sub-blocks) of a preset size, and performing appropriate image processing in a manner suitable for the neural network as required, which can be easily deduced by ordinary experts in the technical field of the present invention.

[0125] On the other hand, the technical concept of the present invention relates to a neural network having the technical characteristics of including the microscopic neural network and the macroscopic neural network as described above. Such an example will be specifically described with reference to FIG4.

[0126] FIG. 4 is a schematic diagram of the structure of a neural network involved in an embodiment of the present invention.

[0127] 4 , the neural network 200 involved in the technical concept of the present invention includes a micro neural network and a macro neural network.

[0128] The micro neural network includes multiple layers 210 and an output layer 230. The multiple layers 210 include an input layer 211 and multiple hidden layers 212.

[0129] The macro neural network includes multiple layers 220 and the output layer 230. The multiple layers 220 include an input layer 221 and multiple hidden layers 222.

[0130] The micro neural network is defined to accept an input of a specific image block 30 and output a diagnosis result of the specific image block, namely, each output data defined in the output layer 230 .

[0131] In addition, the macro neural network is defined as accepting input including the specific image block 30 and at least one macro image block 40 among the adjacent image blocks including the specific image block 30, and outputting a diagnosis result of the specific image block.

[0132] That is to say, in order to output the diagnosis result of a specific image block 30, the neural network 200 involved in the technical idea of ​​the present invention can not only consider the image features of the specific image block 30, but also consider the image features of each image block adjacent to the specific image block 30 to output the diagnosis result.

[0133] Although the macro image block 40 shown in FIG. 4 is a 3×3 image block surrounding the image block, various embodiments are possible.

[0134] The output layer 230 can receive input of each output data of the first front layer (212-1) of the previous layer of the output layer 230 contained in the micro neural network and the second front layer (222-1) of the previous layer of the output layer 230 contained in the macro neural network, thereby outputting the output data defined in the output layer 230. The first front layer (212-1), the second front layer (222-1) and the output layer 230 can be fully connected.

[0135] The feedforward function defining the output layer 230 may use the input data input by the input layer as one of various functions that output the output data by the output layer 230 through the neural network 200 as a result.

[0136] Finally, when the neural network 200 learns to diagnose a specific image block 30, it simultaneously considers the image features of the specific image block 30 and the image features of the macro image block 40 containing the specific image block 30, and then outputs the output data of the output layer 230 corresponding to the label values ​​of a large amount of training data.

[0137] That is, in order to learn the neural network 200, a large amount of training data is used, and the large amount of training data may include a pair of specific image blocks 30 and macro image blocks 40. Moreover, the macro image block 40 may also be learned using the annotation information of the specific image block 30.

[0138] In this way, the learning of the neural network 200 can simultaneously consider the image features of the specific image block 30 and the macro image block 40 to output output data corresponding to the annotation information of the specific image block 30 .

[0139] Moreover, if the learned neural network 200 uses the target image block to be diagnosed and the macro image block corresponding to the target image block as input data of the input layers of the micro neural network and the macro neural network respectively, it can output the diagnosis result of the target image block, that is, the output data of the output layer 230.

[0140] like Figure 5a As shown, the output layer 230 may output the diagnosis result of the specific image block 30 to be diagnosed as output data. The diagnosis result may at least include information about the disease state of the specific image block 30. The information about the disease state may simply refer to information about whether a specific disease is expressed (or a probability value) in the specific image block 30. However, depending on the type of disease, the information about the disease state may also include more specific information showing the degree of disease progression.

[0141] The output layer does not only output whether the disease is expressed or not as disclosed in the previous application, but can be designed to output a variety of additional information. For example, it can include information showing the degree of disease development and / or correlation factor information of the degree of expression of the correlation factor associated with the value of the state channel. In this regard, since it has been disclosed in detail in the previous application, the detailed description will be omitted.

[0142] When using Figure 5a When the neural network 200 shown in FIG. 1 is used, there may also be a feature value that is not outputted in the output layer 230 after the output data of the output layer 230 is inputted and corresponds to the expression probability of the disease of the image block finally inputted. Figure 5a The layers shown in FIG.

[0143] Or, if Figure 5b As shown, the neural network can be designed to have Figure 5a The layer shown outputs a plurality of state channels and correlation factor channels, and a layer 240 outputs a feature value corresponding to the expression probability of the disease of the input image block.

[0144] According to another embodiment of the present invention, the neural network for image block level diagnosis can be designed to have a single path, rather than having two paths (one for each of the micro network and the macro network) as shown in FIG4 . Figure 6 Same as shown.

[0145] Figure 6 It is a schematic diagram of the structure of an example of a neural network involved in another embodiment of the present invention.

[0146] refer to Figure 6 As mentioned above, the neural network can be defined as receiving input of the image block unit and judging whether the input image block has a disease. At this time, as shown in the figure, the neural network can input data of 4 channels (for example, RGB channels and gray channels).

[0147] The input data can be defined as output data through multiple layers such as convolutional layers and maximum pooling, that is, whether the input image block is a disease image block. This neural network can use a well-known neural network using a Dense Net model. And it can be seen that at this time, the neural network involved in the technical idea of ​​the present invention has an additional 1×1 convolution compared to the original Dense Net model, which has the effect of confirming the internal feature map through it. In addition, the Signoid function is used as an excitation function, but various excitation functions can be used.

[0148] Neural networks that can perform image patch-level diagnosis in a variety of other ways can be defined, which can be easily deduced by ordinary experts in the field of the present invention.

[0149] On the other hand, according to the technical concept of the present invention, only the tissue part is distinguished and visualized in the image block determined to be disease expression, which will be described in more detail below.

[0150] Referring again to FIG. 2 , the heat map generation module 240 may generate an image block level heat map image corresponding to the slide based on the diagnosis result of each of the plurality of image blocks included in the slide.

[0151] The heat map generation module 240 can mark the disease image block according to the diagnosis result of each image block. In the image block level diagnosis, each image block is classified according to whether it is a disease, so that the entire disease image block can be marked as a disease. The heat map generation module 240 can generate Figure 7a For example, the heat map generation module 240 can generate the following slide heat map: Figure 7b The heat map shown. Figure 7b As shown, the image block-level diagnosis results are visualized in the form of a heat map, so that the diseased area can appear in a grid shape.

[0152] At the same time, the tissue mask generation module 250 can generate a tissue mask image on the slide. For example, the tissue mask generation module 250 can generate Figure 6 Tissue mask image shown in c.

[0153] The method for generating a tissue mask image by the tissue mask generating module 250 has the following features.

[0154] The following is a more specific description: The tissue mask generation module 250 may generate a tissue mask image corresponding to the slide based on a hue saturation value (HSV) model corresponding to the slide.

[0155] In one embodiment, the tissue mask generation module 250 may perform image binarization on the S space corresponding to the slide to generate the first binarization result. At this time, the S space corresponding to the slide is a space composed of saturation values ​​of the HSV model corresponding to the slide.

[0156] The tissue mask generation module 250 may use Otsu Threshholding as an image binarization method. Otsu Threshholding is a clustering-based image thresholding technique used in the field of computer vision or image processing.

[0157] In addition, the tissue mask generation module 250 can generate a second binarization result by performing image binarization on the space `1-V corresponding to the slide. At this time, the V space corresponding to the slide is a space composed of the nominal values ​​(Value) of the HSV model corresponding to the slide (i.e., a matrix with a size of w×h, and the values ​​(nominal values) of the V channel (w is the width of the image, h is the height of the image)), and the 1-V space is a space minus the value of the V channel from the matrix filled with 1 of size w×h.

[0158] On the other hand, the tissue mask generating module 250 may generate a tissue mask image corresponding to the slide based on the first binarization result and the second binarization result.

[0159] The first binarization result and the second binarization result may include a binary value (e.g., 0 or 1 or 0 or 255) corresponding to each pixel of the slide. The tissue mask generation module 150 corresponds to each pixel of the slide. When the binary value of the first binarization result corresponding to the pixel or the binary value of the first binarization result corresponding to the pixel is 1 (or 255), the pixel on the tissue mask image corresponding to the pixel can be determined as a tissue pixel (a pixel of the tissue). Otherwise (i.e., when the binary value is 0), the pixel on the tissue mask image corresponding to the pixel can be determined as a non-tissue pixel (a pixel of a non-tissue), thereby generating a tissue mask image corresponding to the slide. In summary, the tissue mask generation module 250 can generate a tissue mask image by performing a logical sum operation on the image binarization result of the S space and the image binarization result of the 1-V space.

[0160] On the other hand, the diagnosis result visualization module 260 may generate a disease diagnosis visualization image corresponding to the slide based on the image block level thermal map image and the tissue mask image. For example, the diagnosis result visualization module 260 may generate a disease diagnosis visualization image such as Figure 7d as shown in .

[0161] In one embodiment, the diagnosis result visualization module 260 may apply a conditional random field to the image block level heat map image and the tissue mask image to generate a disease diagnosis visualization image corresponding to the slide. In particular, the number of labels of the conditional random field is 2, and is characterized in that pixels below a threshold value among pixels contained in the image block level heat map image are converted to 0.

[0162] A more detailed description is as follows.

[0163] The diagnosis result visualization module 260 can input the original tissue image (i.e., slide or image block), the tissue mask image generated by the tissue mask generation module 250, and the image block level heat map image generated by the heat map generation module 240. From the size of each input value, the original tissue image is RGB 3 channels, so the size is w×h×3, the size of the tissue mask image is w×h, and the image block level heat map image is also w×h. On the other hand, the diagnosis result visualization module 260 can convert the image block level heat map image of w×h into w×h×2. At this time, w×h×[0] contains the original image block level heat map image, and w×h×[1] contains 1-each pixel value of the image block level heat map image. Afterwards, the softmax function is taken for DMEN=3, and each pixel constitutes a probability space for showing the disease (such as prostate cancer). According to this probability, the CRF algorithm will affect the shape of the actual tissue image and deform it, so that the disease image is visualized to conform to the shape of the actual tissue.

[0164] In particular, in a preferred embodiment of the present invention, the tissue mask generation module 150 can very accurately visualize the part diagnosed as diseased on the slide by performing a logical sum operation on the image binarization result of the S space and the image binarization result of the 1-V space, while applying a tissue mask image generation method and a conditional random field method.

[0165] An example of the final visualization result Figure 7d and a heatmap in the shape of a grid - Figure 7b By comparison, it can be seen that Figure 7d The shape is more blurred.

[0166] On the other hand, the diagnostic result visualization method involved in the technical concept of the present invention can also be applied to the diagnostic results of the general classification method that obtains the diagnostic results in image blocks. In the case of general classification in image block units, even parts that are not tissues are displayed as diseases (see Figure 7a ), but by applying the diagnostic result visualization method involved in the technical idea of ​​the present invention, only the tissue part diagnosed as diseased will be clearly distinguished from other parts.

[0167] On the one hand, the disease diagnosis visualization image generated by the diagnosis result visualization module 260 can be provided by the aforementioned annotation information generation module 110, and the contour extraction module 120 and the XML generation module 130 run the program to generate an XML document including the diagnosis result.

[0168] On the other hand, this specification mainly illustrates an example in which the technical idea of ​​the present invention is applicable to prostate cancer, but it is also necessary to apply the technical idea of ​​the present invention to other diseases. It is necessary to consider not only the specific tissue but also the state of the surrounding tissue of the tissue to diagnose the specific tissue, so as to make a more accurate diagnosis and visualize the diagnosis results. This can be easily inferred by ordinary experts in the technical field of the present invention.

[0169] On the other hand, according to an embodiment, the diagnostic system 100 may include a processor and a memory storing a program executed by the processor. The processor may include a single-core CPU or a multi-core CPU. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magneto-optical disk storage devices, flash memory devices or other non-volatile solid-state memory devices. The access of the processor and other components to the memory may be controlled by a memory controller.

[0170] On the other hand, the neural network diagnostic method involved in the embodiment of the present invention can be implemented in the form of computer-readable program commands and stored in a computer-readable recording medium, and the control program and target program involved in the embodiment of the present invention can also be stored in a computer-readable recording medium. Computer-readable recording media include all types of recording devices that store computer system-readable data.

[0171] The program commands recorded on the recording medium may be specially designed and configured for the present invention, or may be well known and used by those skilled in the art in the software field.

[0172] Computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floppy disks, and specially configured hardware devices that store and execute program commands such as ROM, RAM, flash memory, etc. In addition, computer-readable recording media are dispersed in networked computer systems, and computer-readable codes can be stored and executed in a distributed manner.

[0173] Examples of program commands include not only machine language codes created by a compiler, but also high-level language codes that can be executed by a computer using a device that electronically processes information, such as an interpreter.

[0174] The hardware device may be configured as one or more software modules to perform the operations of the present invention, and vice versa.

[0175] The foregoing description of the present invention is for illustration only, and a person with common knowledge in the technical field to which the present invention belongs can understand that it can be easily changed to other specific forms without changing the technical ideas or essential features of the present invention. Therefore, the above-mentioned embodiments should be understood as being examples in all aspects and are not limited thereto. For example, each component described as a single form can be implemented separately, and similarly, each component described as dispersed can also be implemented in a combined form.

[0176] The scope of the present invention is as shown in the scope of the patent claims described below, and is not reflected by the above detailed description. The meaning and scope of the patent claims and all changes or deformations derived from their unified concept should be interpreted as being included in the scope of the present invention.

[0177] Possible industrial applications

[0178] The present invention can be used for "a diagnosis result generating system and method thereof".< / annotation>

Claims

1. A diagnostic result generating system, comprising: An annotation information generation module generates annotation information reflecting the diagnosis result of whether a disease exists on the biological image slide, wherein the annotation information includes disease status information of each pixel of the slide; a contour extraction module, extracting at least one contour from the disease state information of the pixels of the annotation information; and An XML generation module generates an XML document containing each contour line information of the at least one extracted contour; The diagnosis result generating system further comprises a diagnosis system for outputting a diagnosis result of whether the disease exists using the slide and the neural network, The diagnostic system comprises: An image block neural network generates an image block level diagnosis result of whether the disease exists in each preset image block of a preset size divided into the slide; a heat map generation module, which generates an image block level heat map image corresponding to the slide based on the diagnosis result of each image block in the plurality of image blocks included in the slide; a tissue mask generating module, which generates a tissue mask image corresponding to the slide based on the HSV model corresponding to the slide; and The diagnosis result visualization module can apply a conditional random field to the image block level heat map image and the tissue mask image to generate a disease diagnosis visualization image corresponding to the slide.

2. The diagnostic result generating system according to claim 1, The disease state information may include normal, Gleason mode 3, Gleason mode 4, and Gleason mode 5.

3. The diagnostic result generating system according to claim 2, The XML document further includes the extracted disease status information of the at least one profile.

4. The diagnostic result generating system according to claim 1, characterized in that: The disease is prostate cancer.

5. The diagnostic result generating system according to claim 1, The annotation information generating module may generate the annotation information according to the diagnosis result output by the diagnosis system.

6. The diagnostic result generating system according to claim 1, The tissue mask generating module, The image is binarized for the S space corresponding to the slide and a first binarization result is generated; the S space corresponding to the slide refers to the saturation value space of the HSV model corresponding to the slide, The image is binarized for the 1-V space corresponding to the slide and a second binarization result is generated; the V space corresponding to the slide is the nominal value space of the HSV module corresponding to the slide, A tissue mask image corresponding to the slide may be generated based on the first binarization result and the second binarization result.

7. A method for generating a diagnosis result, comprising: Annotation information step, generating annotation information reflecting the diagnosis result of whether the disease exists on the biological image slide; wherein the annotation information includes the disease status information of each pixel on the slide; a contour extraction step of extracting at least one contour from the disease state information of the pixels of the annotation information; and An XML generating step, generating an XML document containing each contour line information of the at least one extracted contour; The diagnostic result generating method further comprises: A diagnostic step of using the slide and the neural network to output a diagnosis result of whether the disease exists, The diagnostic steps include: An image block level diagnosis step, using a neural network to generate an image block level diagnosis result of whether the disease exists in each preset image block of a preset size divided from the slide; a heat map generating step, generating an image block level heat map image corresponding to the slide based on the diagnosis result of each image block in the plurality of image blocks included in the slide; a tissue mask generating step, generating a tissue mask image corresponding to the slide based on the HSV model corresponding to the slide; and The diagnosis result visualization step may apply a conditional random field to the image block level heat map image and the tissue mask image, thereby generating a disease diagnosis visualization image corresponding to the slide.

8. The method for generating a diagnosis result according to claim 7, The disease status information includes normal, Gleason pattern 3, Gleason pattern 4 and Gleason pattern 5.

9. The method for generating a diagnosis result according to claim 8, The XML document further includes each disease state information of the at least one profile extracted.

10. The method for generating a diagnosis result according to claim 7, The annotation information generating step includes the step of generating the annotation information according to the diagnosis result output by the diagnosis system.

11. The method for generating a diagnosis result according to claim 7, The tissue mask generating step comprises: A step of performing image binarization on the S space corresponding to the slide and generating a first binarization result; The S space corresponding to the slide refers to the saturation value space of the HSV model corresponding to the slide; The step of performing image binarization on the 1-V space corresponding to the slide and generating a second binarization result; the V space corresponding to the slide is the nominal value space of the HSV model corresponding to the slide; as well as A step of generating a tissue mask image corresponding to the slide based on the first binarization result and the second binarization result.

12. A computer storage medium installed in a data processing device, for recording a computer program for executing the method according to any one of claims 7 to 11.

Images