Image analysis methods, image analysis devices, classification devices, classification systems, control programs, recording media

Abstract

A classification device (1) is provided with: a single color component image generation unit (22) that generates, from a tissue image, a plurality of single color component images based on pixel values corresponding to a plurality of color components constituting the tissue image; a binarization unit (23) that generates a plurality of binarization images respectively from the single color component images; a feature number calculation unit (24) for calculating, for each of the binarization images, a first feature number indicating the number of hole-shaped regions comprising pixels of a second pixel value and surrounded by pixels of a first pixel value, a second feature number indicating the number of connected regions formed through connection of pixels of the first pixel value, and a third feature number that is a ratio between the first feature number and the second feature number; and a classification unit (25) that inputs, to a trained classification model, input data including the feature number calculated by the feature number calculation unit (24) and outputs a classification result relating to a change having occurred in the tissue.

Description

Technical Field

[0001] The present disclosure relates to an image analysis method, an image analysis apparatus, a classification apparatus, and a classification system, etc., which analyze a tissue image obtained by imaging a living body tissue to classify changes occurring in the tissue.

Background Art

[0002] By pathologically classifying changes occurring in a tissue, an appropriate treatment policy is determined according to the classification result. In recent years, with the progress of drug treatment, the importance of early detection and accurate classification of changes occurring in a tissue has increased.

[0003] Therefore, various image analysis methods by computers have been devised. In Non-Patent Document 1, a technique of applying artificial intelligence to image diagnosis of lung cancer is disclosed.

Prior Art Documents

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] A pathologist determines the presence or absence of changes occurring in a tissue from a tissue image obtained by imaging a patient's tissue, and classifies the changes occurring in the tissue. For example, based on a tissue image obtained by imaging a patient's lung, the pathologist determines whether the lung tissue of the patient is normal. And if it is not normal, the pathologist determines whether the change should be classified as either emphysema or a precancerous lesion of lung adenocarcinoma.

[0006] For example, the stage of cancer is evaluated based on the size and depth of invasion of the lesion. However, in the case of lung cancer (especially lung adenocarcinoma), it often occurs in the bronchiolar and alveolar regions, and while it is possible to evaluate the size of the lesion to some extent, it can be difficult to evaluate the depth of invasion. Even skilled pathologists may find it difficult to accurately classify precancerous lesions (or pre-invasive lesions) from histological images of lung tissue.

[0007] The image analysis method described in Non-Patent Document 1 was devised to improve this situation, but there is room for improvement in its accuracy in classifying precancerous lesions.

[0008] One aspect of this disclosure provides an image analysis method, image analysis device, classification device, etc., that accurately classifies changes occurring in biological tissue based on tissue images. [Means for solving the problem]

[0009] <1> To solve the above problems, an image analysis method according to one aspect of the present disclosure includes: a monochromatic component image generation step of generating a plurality of monochromatic component images based on pixel values ​​corresponding to each of a plurality of color components constituting a tissue image, from a tissue image in which the cell nuclei of cells contained in living tissue and components other than the cell nuclei are depicted in different hues; a binarization step of generating a plurality of binarized images with different binarization reference values ​​from each of the plurality of monochromatic component images; and for each of the plurality of binarized images generated from each of the plurality of monochromatic component images, the pixels of the second pixel value after binarization, surrounded by the pixels of the first pixel value after binarization. The method includes a feature number calculation step of calculating a first feature number indicating the number of regions with a hole shape, a second feature number indicating the number of connected regions formed by connecting pixels of the first pixel value, and a third feature number which is the ratio of the first feature number to the second feature number; and a classification step of inputting input data, which includes combinations of the first feature number, the second feature number, and the third feature number calculated for each of the binarized images generated from each of the plurality of monochromatic component images, into a classification model that models the correspondence between the first feature number, the second feature number, and the third feature number and a classification of changes occurring in the tissue, and outputting a classification result regarding changes occurring in the tissue shown in the tissue image.

[0010] <2> Furthermore, an image analysis device according to one aspect of the present disclosure includes: a monochromatic component image generation unit that generates a plurality of monochromatic component images based on pixel values ​​corresponding to each of a plurality of color components constituting a tissue image, from a tissue image in which the cell nuclei of cells contained in the tissue of a living organism and components other than the cell nuclei are depicted in different hues; a binarization unit that generates a plurality of binarized images with different binarization reference values ​​from each of the plurality of monochromatic component images; and a feature number calculation unit that calculates, for each of the plurality of binarized images generated from each of the plurality of monochromatic component images, a first feature number indicating the number of hole-shaped regions consisting of pixels of the second pixel value after binarization, surrounded by pixels of the first pixel value after binarization to a first pixel value and a second pixel value; a second feature number indicating the number of connected regions formed by connecting pixels of the first pixel value; and a third feature number which is the ratio of the first feature number to the second feature number.

[0011] <3> Furthermore, a classification device relating to one aspect of this disclosure is the above <2> The system includes a classification unit which obtains the first feature number, the second feature number, and the third feature number from the image analysis device described above, inputs the input data, which includes the first feature number, the second feature number, and the third feature number calculated for each of the binarized images generated from each of the plurality of monochromatic component images, into a classification model which models the correspondence between the first feature number, the second feature number, and the third feature number and the classification of changes occurring in the tissue, and outputs a classification result regarding changes occurring in the tissue shown in the tissue image.

[0012] <4> Alternatively, a classification device according to one aspect of the present disclosure includes: a monochromatic component image generation unit that generates a plurality of monochromatic component images based on pixel values ​​corresponding to each of a plurality of color components constituting a tissue image, from a tissue image in which the cell nuclei of cells contained in living tissue and components other than the cell nuclei are depicted in different hues; a binarization unit that generates a plurality of binarized images with different binarization reference values ​​from each of the plurality of monochromatic component images; and for each of the plurality of binarized images generated from each of the plurality of monochromatic component images, a hole shape consisting of pixels of the second pixel value after binarization, surrounded by pixels of the first pixel value after binarization. The system includes a feature number calculation unit that calculates a first feature number indicating the number of regions, a second feature number indicating the number of connected regions formed by connecting pixels of the first pixel value, and a third feature number which is the ratio of the first feature number to the second feature number; and a classification unit that inputs input data, including the first feature number, the second feature number, and the third feature number calculated for each of the binarized images generated from each of the plurality of monochromatic component images, into a classification model that models the correspondence between the first feature number, the second feature number, and the third feature number and a classification of changes occurring in the tissue, and outputs a classification result regarding changes occurring in the tissue shown in the tissue image.

[0013] <5> Furthermore, the classification system relating to one aspect of this disclosure is as described above. <2> The image analysis device described above, and <3> The system includes a classification device described above, an external device that transmits the tissue images to the image analysis device, and a presentation device that acquires the classification results output from the classification device and presents the classification results.

[0014] Each aspect of the present disclosure of the image analysis device and the classification device may be implemented by a computer. In this case, a control program that enables the computer to implement the image analysis device and the classification device by operating the computer as the respective parts (software elements) of the image analysis device and the classification device, and a computer-readable recording medium on which the program is recorded, also fall within the scope of the present disclosure. [Effects of the Invention]

[0015] According to one aspect of the present disclosure, based on tissue images, changes that have occurred in a living body's tissue can be accurately classified.

Brief Description of the Drawings

[0016] [Figure 1] It is a diagram showing an example of a tissue image in which normal lung tissue is depicted. [Figure 2] It is a diagram showing an example of a tissue image in which lung tissue classified as emphysema is depicted. [Figure 3] It is a diagram showing an example of a tissue image in which lung tissue classified as atypical adenomatous hyperplasia is depicted. [Figure 4] It is a diagram showing an example of a tissue image in which lung tissue classified as squamous adenocarcinoma is depicted. [Figure 5] It is a diagram showing an example of a tissue image in which lung tissue classified as invasive adenocarcinoma is depicted. [Figure 6] It is a schematic diagram for explaining the Betti number in the concept of homology. [Figure 7] It is a functional block diagram showing a configuration example of a classification system including a classification device according to one aspect of the present disclosure. [Figure 8] It is a flowchart showing the flow of processing performed by the classification device. [Figure 9] It is a diagram showing an example of the data structure of training tissue images. [Figure 10] It is a functional block diagram showing an example of the main configuration of a classification device that generates a classification model. [Figure 11] It is a flowchart showing the flow of processing performed by the classification device to generate a classification model. [Figure 12] It is a diagram showing the classification accuracy by an image analysis method according to one aspect of the present disclosure. [Figure 13] It is a functional block diagram showing another configuration example of a classification system according to one aspect of the present disclosure. [Figure 14] It is a functional block diagram showing another configuration example of a classification system according to one aspect of the present disclosure.

Mode for Carrying Out the Invention

[0017]

Embodiment 1

[0018] (Technical idea of the present disclosure) First, the technical idea of the image analysis method according to one aspect of the present disclosure will be described below.

[0019] The inventors of the present disclosure analyzed a tissue image in which cell nuclei contained in a tissue (biological tissue) of a patient (subject) and a component different from the cell nuclei (for example, cytoplasm) are shown in different hues. As an example of the tissue image to be analyzed, the inventors of the present disclosure selected a tissue image obtained by imaging the lung tissue of a patient. Then, the inventors of the present disclosure described the topological arrangement of cell nuclei shown in the tissue image and applied the concept of homology to quantify the changes occurring in the tissue.

[0020] The inventors of the present disclosure focused on the fact that in a tissue image used for pathological diagnosis, which is an image of a biological tissue, the cell nuclei and the cytoplasm are shown in different hues. The inventors of the present disclosure extracted a partial image in which the region to be analyzed is shown from the tissue image, and generated a plurality of monochromatic component images based on the pixel values corresponding to each of the plurality of color components constituting the partial image. For example, if the tissue image is a color image represented by red (R), green (G), and blue (B), a monochromatic component image based on the pixel values corresponding to the R component, a monochromatic component image based on the pixel values corresponding to the G component, and a monochromatic component image based on the pixel values corresponding to the B component were generated.

[0021] Next, the inventors of this disclosure generated multiple binarized images from each of multiple monochromatic component images, each using a different binarization criterion. For each of the binarized images, the inventors calculated the 1-dimensional Betti number b1 (first feature number), the 0-dimensional Betti number b0 (second feature number), and the ratio R (third feature number) between the 1-dimensional Betti number b1 and the 0-dimensional Betti number b0. Here, the ratio R may be b1 / b0 or b0 / b1.

[0022] The inventors of this disclosure have found that changes occurring in tissues shown in tissue images can be accurately classified based on the calculated one-dimensional Betti number b1, zero-dimensional Betti number b0, and ratio R, and have invented an image analysis method according to one aspect of this disclosure. For example, by applying the image analysis method according to one aspect of this disclosure, it is possible to accurately classify precancerous lesions of emphysema and lung adenocarcinoma based on tissue images.

[0023] (Classification according to the stage of lung adenocarcinoma) Here, we will explain the classification of precancerous lesions of emphysema and lung adenocarcinoma using Figures 1 to 5.

[0024] The changes occurring in the lung tissue are classified into the following five categories (1) to (5), depending on the degree of the change and the stage of cancer progression. ·(1)normal (2) Emphysema ·(3) Atypical adenomatous hyperplasia (AAH) ·(4) Lepidic pattern of adenocarcinoma (LP) ·(5) Invasive adenocarcinoma (AC).

[0025] Figure 1 shows an example of a tissue image showing normal lung tissue. Figure 2 shows an example of a tissue image showing tissue classified as emphysema, and Figure 3 shows an example of a tissue image showing tissue classified as atypical adenomatous hyperplasia. Figure 4 shows an example of a tissue image showing tissue classified as squamous adenocarcinoma, and Figure 5 shows an example of a tissue image showing tissue classified as invasive adenocarcinoma. The resolution of each tissue image is 1600 x 1200 pixels.

[0026] The tissue images shown in Figures 1 to 5 are all 100x magnified images of specimens obtained from a patient's lung, which were paraffin-embedded, sliced ​​into thin sections, and then stained with HE (Hematoxylin-Eosin). HE staining is one method used to stain collected tissue samples, and it involves the combined use of hematoxylin and eosin. Hematoxylin stains the chromatin in the cell nucleus and the ribosomes in the cytoplasm blue-violet (first color). On the other hand, eosin stains the components of the cytoplasm and the extracellular matrix red (second color).

[0027] In classifying precancerous lesions of lung adenocarcinoma, accurately distinguishing between atypical adenomatous hyperplasia (shown in Figure 3), squamous adenocarcinoma (shown in Figure 4), and invasive adenocarcinoma (shown in Figure 5) is crucial for determining the patient's treatment plan. Correctly classifying atypical adenomatous hyperplasia is particularly important because it allows for the early detection of precancerous lesions of lung adenocarcinoma. However, accurately distinguishing between atypical adenomatous hyperplasia and squamous adenocarcinoma, for example, is not easy even for pathologists. By applying the image analysis method according to one aspect of this disclosure, it is possible to classify atypical adenomatous hyperplasia, squamous adenocarcinoma, and invasive adenocarcinoma quickly and accurately.

[0028] Furthermore, the image analysis method according to one aspect of this disclosure can analyze any tissue image in which the cell nuclei of cells contained in the tissue and components other than the cell nuclei are depicted in different hues. In other words, the tissue image to be analyzed is not limited to images of HE-stained tissue, as shown in Figures 1 to 5. The tissue image to be analyzed may, for example, be an image of tissue stained using any known staining method capable of staining the cell nuclei. Also, the tissue image to be analyzed may, but is not limited to, a color image to which RGB representation is applied, as shown in Figures 1 to 5. For example, the tissue image may be a color image to which any representation other than RGB representation is applied, for example, a color image represented by cyan (Cy), magenta (Mg), and yellow (Ye).

[0029] The tissue image may be an image of tissue taken from a patient's body. In this specification, the example of analyzing a tissue image of a patient's lung is used, but the tissue images to be analyzed by the image analysis method according to one aspect of this disclosure are not limited to lung tissue images. The image analysis method according to one aspect of this disclosure can analyze tissue images of, for example, the prostate, mammary glands, gastrointestinal tract, liver, pancreas, and lymph nodes.

[0030] (Mathematical representation for analyzing tissue images) Next, we will describe a mathematical representation applied to analyze tissue images in an image analysis method according to one aspect of this disclosure.

[0031] In order to quantify and analyze changes occurring in an organization, an image analysis method according to one aspect of this disclosure applies the concept of homology to a binarized image. Homology is a branch of mathematics that facilitates analysis such as the combination of figures by algebraically replacing the morphological properties of figures.

[0032] Homology is a mathematical concept that represents the connectivity and contact between constituent elements. In tissue images, appropriate binarization criteria (also called binarization parameters) are set and the image is binarized. Then, the 0-dimensional Betti number and the 1-dimensional Betti number b1 are calculated from the binarized image. Using the calculated 0-dimensional Betti number b0 and 1-dimensional Betti number b1, it is possible to evaluate the degree of connectivity and contact between constituent elements of the tissue.

[0033] Betti numbers are topological suggestive numbers that are independent of the shape of a figure (for example, the components of an organization) and relate only to the contact and separation of figures. When a q-dimensional singular homology group is finitely generated, this q-dimensional singular homology group can be divided into a direct sum of a free abelian group and a finite abelian group. The class of this free abelian group is called a Betti number.

[0034] <0-dimensional Betti number b0> The 0-dimensional Betti number b0 is mathematically defined as follows: The number of connected components in a figure K (also called a 1-dimensional complex) made up of a finite number of line segments is called the 0-dimensional Betti number. "A figure made up of a finite number of points connected by a finite number of line segments is connected" means that it is possible to reach any other vertex from any vertex of this figure by traversing its edges.

[0035] In each of the multiple binarized images generated using different binarization reference values, the number of connected regions formed by connecting pixels with one of the pixel values ​​after binarization (for example, pixels with a value of 0 after binarization) is the 0-dimensional Betti number b0.

[0036] <1D Betti number b1> The one-dimensional Betti number b1 is mathematically defined as follows: The one-dimensional Betti number b1 of a figure K is r if the following conditions (1) and (2) are met: (1) For a figure K made up of a finite number of line segments (a connected one-dimensional complex), removing r open (endless) one-dimensional simplices (e.g., line segments) from figure K does not increase the number of connected components of figure K. (2) Removing any (r+1) open one-dimensional simplices from K makes K no longer connected (i.e., the number of connected components of K increases by 1).

[0037] In each of the multiple binarized images generated using different binarization reference values, the number of hole-shaped regions (e.g., with a pixel value of 255 after binarization) surrounded by pixels with one of the binarized pixel values ​​(e.g., with a pixel value of 0 after binarization) is the one-dimensional Betti number b1.

[0038] <Example figures: 0-dimensional Betti number b0 and 1-dimensional Betti number b1> Here, we will explain the 0-dimensional Betti number b0 and the 1-dimensional Betti number b1 in a binarized image using the exemplary figure shown in Figure 6. Figure 6 is a schematic diagram for explaining the Betti number in the concept of homology. In the case of figure M1 shown in Figure 6, there is one black region. Therefore, the 0-dimensional Betti number b0 of figure M1 is 1. Also, in the case of figure M1, there is one white region surrounded by a black region. Therefore, the 1-dimensional Betti number b1 of figure M1 is 1.

[0039] In the case of figure M2 shown in Figure 6, there are 2 black regions. Therefore, the 0-dimensional Betti number b0 of figure M2 is 2. Also, in the case of figure M2, there are 3 white regions surrounded by black regions. Therefore, the 1-dimensional Betti number b1 of figure M2 is 3.

[0040] In the case of a two-dimensional image, the 0-dimensional Betti number b0 is the number of groups of interconnected components, and the 1-dimensional Betti number b1 is the number of spaces enclosed by the outer edges of these interconnected components (hereinafter sometimes referred to as "hole-shaped regions"). The number of hole-shaped regions is the total number of "holes" present in the interconnected components.

[0041] (Configuration of classification system 100) Next, the configuration of the classification system 100 will be explained using Figure 7. Figure 7 is a block diagram showing an example of the configuration of a classification system 100 that includes a classification device 1 that performs an image analysis method according to one aspect of the present disclosure.

[0042] The classification system 100 includes a classification device 1, an external device 4 that transmits tissue images 31 to the classification device 1, and a presentation device 5 that acquires the classification results output from the classification device 1 and presents the classification results. Figure 7 shows an example in which medical institution H1 has introduced the classification system 100.

[0043] External device 4 may be, for example, a microscope equipped with imaging capabilities, or a computer connected to a microscope and capable of acquiring image data from the microscope. Alternatively, external device 4 may be a server device within medical institution H1 that stores and manages various medical image data and pathological image data.

[0044] The diagram illustrates an example where the classification device 1 acquires tissue images 31 from an external device 4 separate from the classification device 1, but the system is not limited to this. For example, the classification device 1 may be built into the external device 4.

[0045] The presentation device 5 may be a display and speaker capable of displaying information output from the classification device 1. In one example, the presentation device 5 may be a display provided by the classification device 1 or an external device 4. Alternatively, it may be a computer and tablet terminal used by pathologists, laboratory technicians, researchers, etc., belonging to medical institution H1.

[0046] The classification device 1 and the external device 4, and the classification device 1 and the presentation device 5 may be connected by wireless communication or by wired communication.

[0047] (Configuration of Classification Device 1) The classification device 1 comprises a control unit 2 and a storage unit 3. The storage unit 3 may store tissue images 31 and classification models 33. The classification models 33 will be described later.

[0048] The memory unit 3 may store not only the tissue image 31 and the classification model 33, but also control programs for each part executed by the control unit 2, OS programs, application programs, etc. The memory unit 3 may also store various data that the control unit 2 reads when executing these programs. The memory unit 3 is composed of a non-volatile storage device such as a hard disk or flash memory. In addition to the memory unit 3, there may be a volatile storage device such as RAM (Random Access Memory) used as a working area to temporarily hold data during the execution of the various programs mentioned above.

[0049] <Configuration of Control Unit 2> The control unit 2 may be composed of a control device such as a CPU (central processing unit) or a dedicated processor. Each part of the control unit 2 shown in Figure 7 can be realized by a control device such as a CPU reading a program stored in a storage unit 3, which is implemented as ROM (read-only memory), into RAM (random access memory) or the like, and executing it.

[0050] The control unit 2 analyzes the tissue image 31 to be analyzed, classifies the changes occurring in the tissue shown in the tissue image 31, and outputs the classification result. The control unit 2 comprises an image acquisition unit 21, a monochromatic component image generation unit 22, a binarization unit 23, a feature number calculation unit 24, a classification unit 25, and an output control unit 26.

[0051] [Image acquisition unit 21] The image acquisition unit 21 acquires a tissue image 31 from an external device 4. Here, if the tissue to be analyzed is the lung, the tissue image 31 may be an image of a lung tissue sample taken from the patient's body, taken at a predetermined magnification. The image acquisition unit 21 may also acquire a partial image from the external device 4 corresponding to a region extracted from the tissue image 31. The image acquisition unit 21 may store the acquired tissue image 31 in the storage unit 3.

[0052] The image acquisition unit 21 may be equipped with known image recognition and image processing functions. This allows the image acquisition unit 21 to extract the region containing the tissue to be analyzed from the tissue image 31, or to divide the tissue image 31 to generate multiple partial images. For example, the image acquisition unit 21 may be able to distinguish and extract the region containing the tissue from the surrounding region (for example, the region containing resin) in the tissue image 31.

[0053] The classification device 1 may be able to classify changes occurring in the tissue image 31 more accurately by analyzing a tissue image 31 with reduced image resolution than by analyzing the original tissue image 31 without changing its image resolution. More specifically, fine information captured in a high-resolution tissue image 31 may become "noise" that affects each of the processes described later. To block the influx of fine information captured in the original tissue image 31, for example, it is effective to deliberately reduce the resolution of the tissue image 31. Therefore, the image acquisition unit 21 may generate a tissue image 31 with reduced image resolution from the original tissue image 31.

[0054] The classification device 1 may generate multiple tissue images 31 with different image resolutions from the acquired original tissue image and use them for analysis. This is because analyzing multiple tissue images 31 with different image resolutions may allow for more accurate classification of changes occurring in the tissues depicted in the tissue images 31. Therefore, the image acquisition unit 21 may generate multiple tissue images 31 with different image resolutions for each acquired tissue image. In this specification, the process in which the image acquisition unit 21 generates multiple tissue images 31 with different image resolutions for each acquired tissue image and uses these multiple tissue images 31 for analysis is referred to as "multiscale analysis." On the other hand, the process in which the image acquisition unit 21 generates a single tissue image 31 adjusted to a predetermined image resolution for each tissue image and uses this tissue image 31 for analysis may be referred to as "single-scale analysis." The tissue images 31 generated by the image acquisition unit 21, with reduced image resolution from the original tissue image 31, are stored in the storage unit 3 and may be used for processing described later.

[0055] [Monochromatic component image generation unit 22] The monochromatic component image generation unit 22 generates multiple monochromatic component images from the tissue image 31, based on the pixel values ​​corresponding to each of the multiple color components that make up the tissue image 31.

[0056] For example, if the tissue image 31 is a color image to which RGB representation is applied, the monochromatic component image generation unit 22 generates three monochromatic component images from the tissue image 31 based on the pixel values ​​corresponding to the R component, G component, and B component, respectively. The monochromatic component image generation unit 22 may also be configured to generate a grayscale image (monochromatic component image) based on the pixel values ​​corresponding to the luminance component of the tissue image 31. In the case of multiscale analysis, the monochromatic component image generation unit 22 generates multiple monochromatic component images for each of the multiple tissue images 31 with different image resolutions, which are generated by the image acquisition unit 21.

[0057] [Binarization section 23] The binarization unit 23 performs a binarization process on the monochromatic component image, which has been separated into monochromatic components, and generates multiple binarized images with different binarization reference values.

[0058] The binarization unit 23 generates multiple binarized images with different binarization reference values ​​from each of the multiple monochromatic component images. In the binarization process, the binarization unit 23 converts pixels with pixel values ​​greater than the binarization reference value into white pixels and pixels with pixel values ​​less than or equal to the binarization reference value into black pixels. At this time, the binarization unit 23 performs the binarization process on the monochromatic component image each time the binarization reference value is changed, and generates multiple binarized images. In other words, the binarization unit 23 generates multiple binarized images with different binarization reference values ​​for all monochromatic component images generated from the tissue image 31.

[0059] In one example, the binarization unit 23 sets the binarization reference value in the range of 0 to 255. For example, if the binarization reference value is set to a pixel value of 100, the pixel values ​​of pixels with a pixel value of 100 or less become 0 as a result of the binarization process, and the pixel values ​​of pixels with a pixel value greater than 100 become 255 as a result of the binarization process. In one example, the binarization unit 23 may change the binarization reference value by 1 from 1 to 255 for each monochromatic component image to generate a binarized image of 255. Alternatively, the binarization unit 23 may change the binarization reference value by 1 from 2 to 254 for each monochromatic component image to generate a binarized image of 253. Alternatively, the binarization reference value may be changed by 5 from 2 to 254 for each monochromatic component image to generate a binarized image of 253. However, the binarization unit 23 may generate multiple binarized images by changing the binarization reference value for each monochromatic component image according to a desired rule.

[0060] [Feature Count Calculation Unit 24] The feature number calculation unit 24 calculates a one-dimensional Betti number b1 for each of the multiple binarized images, which represents the number of hole-shaped regions consisting of pixels of the other pixel value (hereinafter referred to as the second pixel value) after binarization, surrounded by pixels of one pixel value (hereinafter referred to as the first pixel value) after binarization. The feature number calculation unit 24 also calculates a zero-dimensional Betti number b0, which represents the number of connected regions formed by the connection of pixels of the first pixel value, and the ratio R of the one-dimensional Betti number b1 to the zero-dimensional Betti number b0. In the following example, the case where the ratio R is b1 / b0 is given, but the ratio R may also be b0 / b1.

[0061] The above-mentioned connected regions are, for example, areas where pixels with a value of 0 after binarization are adjacent to each other. Each connected region is surrounded by pixels with a value of 255 after binarization, and is an independent region from one another.

[0062] On the other hand, the hole-shaped regions described above are, for example, regions where pixels with a pixel value of 255 after binarization are adjacent to each other. Each hole-shaped region is surrounded by pixels with a pixel value of 0 after binarization, and is an independent region from one another.

[0063] For example, if a binarized image with 255 features is generated by changing the binarization reference value by 1 from 1 to 255 for each tissue image, the feature number calculation unit 24 calculates the 1D Bettsch number b1 of 255, the 0D Bettsch number b0 of 255, and the ratio R of 255.

[0064] The values ​​of the one-dimensional Bettsch number b1 and the zero-dimensional Bettsch number b0 calculated by the feature number calculation unit 24 depend on the magnification, resolution, and area of ​​the region captured in the tissue image 31 that were set when acquiring the tissue image 31. Therefore, it is desirable for the feature number calculation unit 24 to calculate the one-dimensional Bettsch number b1 and the zero-dimensional Bettsch number b0 for tissue images 31 that have the same magnification and resolution, and the same area of ​​the region captured.

[0065] An existing program can be used as the feature number calculation unit 24. CHomP is one example of such a program. CHomP is freeware compliant with the GNU (General Public License). However, it is not limited to this; any program that can calculate the 0-dimensional Betti number b0 and the 1-dimensional Betti number b1 related to an image may be used instead of CHomP.

[0066] [Classification section 25] The classification unit 25 inputs input data, including combinations of one-dimensional Betti number b1, zero-dimensional Betti number b0, and ratio R, into the classification model described below, and outputs a classification result regarding the changes occurring in the tissue shown in the tissue image 31. Here, the changes occurring in the tissue may be the degree of cell differentiation calculated based on the structural characteristics, arrangement, and invasion pattern of tumor cells occurring in the tissue. The classification model 33 models the correspondence between one-dimensional Betti number b1, zero-dimensional Betti number b0, and ratio R and the classification regarding the changes occurring in the tissue. That is, the classification model 33 is generated by machine learning using the combination of (1) and (2) below as training data. (1) A training tissue image 32 obtained by imaging tissue, wherein classification information is pre-assigned to the training tissue image 32, classifying the changes occurring in the tissue shown in the training tissue image 32. (2) The one-dimensional Betti number b1, the zero-dimensional Betti number b0, and the ratio R calculated for each of the multiple binarized images with different binarization reference values ​​generated from the training tissue image 32.

[0067] The machine learning process using training tissue images 32 to generate the classification model 33 will be explained later.

[0068] [Output control unit 26] The output control unit 26 causes the display device 5 to display information indicating the classification result output from the classification unit 25. Alternatively, the output control unit 26 may be configured to display the tissue image 31, which was the subject of analysis, to the display device 5 along with the information indicating the classification result.

[0069] Furthermore, the output control unit 26 may be configured to control the presentation device 5 so as to display the classification results for the regions extracted from the tissue image 31 at the positions corresponding to those regions in the tissue image 31. With this configuration, the classification device 1 can present the classification results output for the tissue image 31 and the positions of the regions corresponding to those classification results to users, including pathologists, laboratory technicians, and researchers.

[0070] The method of presenting the classification results to the user may be any desired configuration. For example, as shown in Figure 7, the presentation device 5 may present the classification results, or the classification results may be output from a printer (not shown) and a speaker (not shown).

[0071] (Processing performed by Classification Device 1) Next, we will explain the processes performed by the classification device 1 using Figure 8. Figure 8 is a flowchart showing an example of the processing flow performed by the classification device 1.

[0072] First, the image acquisition unit 21 acquires a tissue image 31 from the external device 4 (step S1). Here, the image acquisition unit 21 may change the image resolution of the tissue image 31 as needed. Alternatively, the image acquisition unit 21 may generate multiple tissue images 31 with different image resolutions for each acquired tissue image.

[0073] Next, the monochromatic component image generation unit 22 generates multiple monochromatic component images from the tissue image 31, based on the pixel values ​​corresponding to each of the multiple color components that make up the tissue image 31 (Step S2: Monochromatic Component Image Generation Step).

[0074] Next, the binarization unit 23 generates multiple binarized images with different binarization reference values ​​from each monochromatic component image (Step S3: Binarization step).

[0075] The feature number calculation unit 24 calculates the one-dimensional Betti number b1, the zero-dimensional Betti number b0, and the ratio R for each of the multiple binarized images (Step S4: Feature number calculation step).

[0076] Next, the classification unit 25 inputs the input data, which includes a combination of the one-dimensional Betti number b1, the zero-dimensional Betti number b0, and the ratio R, into the classification model 33 (step S5: classification step), and outputs the classification result regarding the changes occurring in the tissue shown in the tissue image 31 (step S6: classification step).

[0077] According to the above configuration, the classification device 1 generates multiple monochromatic component images based on the pixel values ​​corresponding to each of the multiple color components constituting the tissue image 31, and generates multiple binarized images with different binarization reference values ​​for each monochromatic component image. Next, the classification device 1 calculates the 1D Bettsch number b1, the 0D Bettsch number b0, and the ratio R of the 1D Bettsch number b1 and the 0D Bettsch number b0 for each binarized image. Then, the classification device 1 inputs the 1D Bettsch number b1, the 0D Bettsch number b0, and the ratio R as a set of data into the classification model 33 and outputs a classification result regarding the changes occurring in the tissue. As a result, the classification device 1 can accurately classify the changes occurring in the tissue shown in the tissue image 31.

[0078] When changes in living tissue are precancerous lesions such as emphysema and lung adenocarcinoma, they often manifest as changes in the uniformity of cell shape and size. By employing the above method, the properties of tissue images 31 taken from living tissue are mathematically analyzed using the concept of homology, and the changes occurring in the tissue are classified based on the results of this analysis. Since this classification result is output from a classification model 33 generated by machine learning using training tissue images 32 described later, it is based on the properties of the tissue images 31, similar to the results of a pathological diagnosis by a pathologist. Therefore, the classification result is as understandable and reliable as the judgment of a pathologist.

[0079] (Training tissue image 32) Training tissue images 32 may be used to generate a classification model 33. Figure 9 shows an example of the data structure of the training tissue images 32.

[0080] As shown in Figure 9, the training tissue image 32 includes tissue images of tissue taken from a patient's body, each assigned a training tissue image ID. Each tissue image in the training tissue image 32 shows the cell nuclei of the cells contained in the tissue and other components with different hues, and classification information indicating the classification of changes occurring in the tissue shown in the training tissue image 32 is pre-assigned by a medical professional. The classification information is the result of a determination made by a pathologist who has examined the tissue images in the training tissue image 32, and indicates the classification of changes occurring in the tissue shown in each tissue image.

[0081] (Configuration of the classification device 1, which has the function of generating a classification model 33) Next, the configuration of the classification device 1 while it is executing the learning algorithm to generate the classification model 33 will be explained using Figure 10. Figure 10 is a functional block diagram showing an example of the main components of the classification device 1 that generates the classification model 33. For the sake of explanation, components having the same function as those described in Figure 7 will be denoted by the same reference numerals, and their explanations will not be repeated.

[0082] Figure 10 shows an example in which training tissue images 32 are pre-stored in the memory unit 3 of the classification device 1, but it is not limited to this. For example, the image acquisition unit 21 shown in Figure 7 may be configured to acquire training tissue images 32 from an external device 4.

[0083] Furthermore, while the example shown in Figure 10 includes a function to generate a classification model 33, the system is not limited to this. For example, another computer different from the classification device 1 may be used to perform the above-described processing to generate the classification model 33. In this case, the classification model 33 generated by the other computer is stored in the storage unit 3 of the classification device 1, and the classification unit 25 can then use the classification model 33.

[0084] The control unit 2 of the classification device 1, which is executing a learning algorithm for generating a classification model 33, comprises a monochromatic image generation unit 22, a binarization unit 23, a feature number calculation unit 24, a classification unit 25, and a classification model generation unit 27.

[0085] [Classification model generation unit 27] The classification model generation unit 27 executes a machine learning algorithm using the training tissue images 32 on the classification model candidates to generate a classification model 33 (trained). The classification model 33 (trained) is stored in the memory unit 3.

[0086] In machine learning for generating classification models 33, known machine learning algorithms suitable for data classification can be applied. For example, perceptrons, logistic regression, k-nearest neighbors, support vector machines, decision trees, random forests, or gradient boosting may be used to generate classification models 33.

[0087] (Process to generate classification model 33) Next, the process of generating the classification model 33 will be explained using Figure 11. Figure 11 is a flowchart showing an example of the processing flow performed by the classification device 1 to generate the classification model 33.

[0088] First, the classification model generation unit 27 reads the training tissue images 32 from the storage unit 3 (step S11) and selects an unselected training tissue image 32 (for example, in Figure 9, the tissue image with training tissue image ID "T1") from the training tissue images 32 (step S12). Here, the image resolution of the training tissue images 32 may be changed to a predetermined image resolution in advance, but is not limited to this. For example, the classification model generation unit 27 may be configured to change the image resolution of the training tissue images 32 as needed.

[0089] Next, the monochromatic component image generation unit 22 generates multiple monochromatic component images from the tissue image selected by the classification model generation unit 27, based on the pixel values ​​corresponding to each of the multiple color components that make up the tissue image (step S13).

[0090] Next, the binarization unit 23 generates multiple binarized images for each of the multiple monochromatic component images, each with a different binarization reference value (step S14).

[0091] Next, the feature number calculation unit 24 calculates the one-dimensional Betti number b1, the zero-dimensional Betti number b0, and the ratio R for each of the multiple binarized images (step S15).

[0092] Next, the classification unit 25 inputs the input data, including the one-dimensional Betti number b1, the zero-dimensional Betti number b0, and the ratio R, into the classification model candidate (step S16), and outputs a classification result that classifies the changes occurring in the tissue shown in the tissue image (step S17).

[0093] Next, the classification model generation unit 27 compares the classification result output from the classification unit 25 with the classification information corresponding to the tissue image selected in step S12 and calculates the error (step S18). The classification model generation unit 27 also updates the classification model candidate output in step S18 so that the calculated error is minimized (step S19).

[0094] If not all tissue images included in the training tissue images 32 have been selected in step S12 (NO in step S20), the classification model generation unit 27 returns to step S12 and selects an unselected tissue image from the training tissue images 32. For example, if a tissue image with training tissue image ID "T1" is selected, then the tissue image with training tissue image ID "T2" is selected next (see Figure 9).

[0095] On the other hand, if all tissue images included in the training tissue images 32 have already been selected in step S12 (YES in step S20), the classification model generation unit 27 stores the current classification model candidate as classification model 33 (trained) in the memory unit 3 (step S21).

[0096] The classification model generated in this way can output highly accurate classification results regarding changes occurring in the tissues contained in a tissue image (for example, tissue image 31 in Figure 7) in response to inputting a one-dimensional Betti number b1, a zero-dimensional Betti number b0, and a ratio R calculated for a given tissue image (for example, tissue image 31 in Figure 7).

[0097] Alternatively, the classification model 33 may be generated by having a computer different from the classification device 1 perform the processing shown in Figure 11. In this case, the trained classification model 33 may be installed on the classification device 1.

[0098] (Evaluation of classification accuracy) The results of evaluating the classification accuracy when an image analysis method according to one aspect of this disclosure is applied to lung tissue images 31 and precancerous lesions of emphysema and lung adenocarcinoma are classified based on each tissue image 31 will be explained with reference to Figure 12. Figure 12 is a diagram showing the classification accuracy by the image analysis method according to one aspect of this disclosure.

[0099] To evaluate the accuracy of the classification, 94 lung tissue images 31, each labeled with classification results by a specialist pathologist, were used. Of the 94 tissue images 31, 20 were classified as emphysema, 20 as normal, 23 as atypical adenomatous hyperplasia, 19 as squamous adenocarcinoma, and 12 as invasive adenocarcinoma.

[0100] The analytical accuracy shown in Figure 12 is the result of comparing the classification results by pathologists with the classification results output from the classification model 33 for 94 lung tissue images 31. The analytical accuracy shown in Figure 12 is the result obtained by applying multiscale analysis including the process described below.

[0101] The image acquisition unit 21 generated tissue images 31 with image resolutions of 1600×1200 (no change in image resolution), 1800×600, 400×300, and 200×150 from each of the 94 tissue images 31 (image resolution 1600×1200). Then, the monochromatic component image generation unit 22 generated four monochromatic component images from each of the generated tissue images 31: three monochromatic component images based on the pixel values ​​corresponding to the R, G, and B components, and a grayscale image based on the pixel values ​​corresponding to the luminance component of each tissue image 31. Next, the binarization unit 23 performed a binarization process on these four monochromatic component images, generating multiple binarized images with different binarization reference values. The feature number calculation unit 24 calculated the one-dimensional Betti number b1, the zero-dimensional Betti number b0, and the ratio R for each of the multiple binarized images. The classification unit 25 inputs input data, including combinations of one-dimensional Betti number b1, zero-dimensional Betti number b0, and ratio R, into the classification model described later, and outputs classification results regarding the changes occurring in the tissues depicted in each of the 94 tissue images 31.

[0102] As shown in Figure 12, the classification device 1 classified tissue images 31 that had been classified as emphysema by a pathologist with a 90% accuracy rate (=18 / 20), and classified tissue images 31 that had been classified as normal lung by a pathologist with a 100% accuracy rate (=20 / 20).

[0103] Furthermore, classification device 1 classified tissue image 31, which had been classified by a pathologist as atypical adenomatous hyperplasia, with an accuracy rate of 78.3% (≒18 / 23). As mentioned above, correctly classifying atypical adenomatous hyperplasia is important because it means the early detection of precancerous lesions of lung adenocarcinoma. The probability of classification device 1 misclassifying tissue image 31, which had been classified by a pathologist as atypical adenomatous hyperplasia, as normal lung tissue was only 4.3%.

[0104] Furthermore, classification device 1 classified tissue images 31 that had been classified as scaly adenocarcinoma by pathologists with a correct accuracy of 68.4% (=13 / 19), and classified tissue images 31 that had been classified as invasive adenocarcinoma by pathologists with a correct accuracy of 83.3% (≒10 / 12).

[0105] As shown in Figure 12, the classification results obtained by applying the image analysis method according to one aspect of this disclosure show a high degree of agreement with the classification results by pathologists, indicating that it is possible to accurately classify precancerous lesions of emphysema and lung adenocarcinoma.

[0106] [Embodiment 2] Other embodiments of this disclosure are described below. For the sake of clarity, components having the same function as those described in the above embodiments are denoted by the same reference numerals, and their descriptions are not repeated.

[0107] Figure 7 shows an example in which medical institution H1 has introduced the classification system 100, but is not limited to this. For example, the classification device 1a may be connected to an external device 4 of medical institution H2 via a communication network 50 so as to be able to communicate. A classification system 100a employing such a configuration will be explained with reference to Figure 13. Figure 13 is a functional block diagram showing an example of the configuration of a classification system 100a according to one aspect of this disclosure.

[0108] As shown in Figure 13, the classification device 1a is equipped with a communication unit 6 that functions as a communication interface with the external device 4 of the medical institution H2. This allows the image acquisition unit 21 to acquire tissue images 31 from the external device 4 of the medical institution H2 via the communication network.

[0109] Furthermore, the classification device 1a transmits the classification results output from the classification unit 25 to an external device 4 via the communication network 50.

[0110] The classification device 1a may be connected to external devices 4 of multiple medical institutions in a manner that enables communication. In this case, each tissue image 31 transmitted from medical institution H2 to the classification device 1a may be assigned an image ID that identifies each tissue image 31, and a classification number (patient ID) unique to the patient (e.g., patient) from whom the tissue shown in the tissue image 31 was taken. Furthermore, each tissue image 31 may be assigned a medical institution ID that identifies the transmitting medical institution H2.

[0111] By adopting this configuration, the classification device 1a can provide the classification results obtained by analyzing tissue images 31 acquired from each of multiple medical institutions to each medical institution that transmitted the image data of the tissue images 31. For example, the administrator managing the classification device 1a may charge each medical institution a predetermined fee as compensation for the service of providing classification results estimated from the acquired tissue images 31.

[0112] [Embodiment 3] Other embodiments of this disclosure are described below. For the sake of clarity, components having the same function as those described in the above embodiments are denoted by the same reference numerals, and their descriptions are not repeated.

[0113] The classification device 1 shown in Figure 7 and the classification device 1a shown in Figure 13 have both image analysis functions for tissue images 31 and classification functions using a classification model 33. However, the system is not limited to this configuration. For example, the functions of classification devices 1 and 1a may be realized by combining an image analysis device 1A, which includes an image acquisition unit 21, a monochromatic component image generation unit 22, a binarization unit 23, and a feature number calculation unit 24, with a classification device 1B, which includes a control unit 2B. The control unit 2B includes a classification unit 25. A classification system 100b employing such a configuration will be explained with reference to Figure 14. Figure 14 is a functional block diagram showing an example configuration of a classification system 100b according to one aspect of the present disclosure.

[0114] As shown in Figure 14, the image analysis device 1A is equipped with a communication unit 6A that functions as a communication interface with the external device 4 of the medical institution H2 and the classification device 1B. This allows the image acquisition unit 21 to acquire tissue images 31 from the external device 4 of the medical institution H2 via the communication network.

[0115] Furthermore, the image analysis device 1A transmits the first feature count, second feature count, and third feature count calculated by the feature count calculation unit 24 to the classification device 1B via the communication network 50.

[0116] The classification device 1B may be connected to multiple image analysis devices 1A in a communicative manner. In this case, the number of features (including the first, second, and third feature counts) transmitted from the image analysis device 1A to the classification device 1B may be assigned various IDs. These various IDs may include a classification number (patient ID) unique to the patient from whom the tissue shown in the tissue image 31 to be analyzed was taken, a medical institution ID indicating the medical institution H2 that transmitted each tissue image 31, and a device ID unique to the image analysis device 1A that performed the image analysis.

[0117] With this configuration, the image analysis device 1A analyzes tissue images 31 acquired from each of multiple medical institutions, calculates a predetermined number of features, and transmits them to the classification device 1B. The classification device 1B outputs a classification result using the number of features acquired from the image analysis device 1A, and can provide this classification result to each medical institution that transmitted the image data of the tissue images 31. For example, the administrator managing the classification device 1B may charge each medical institution a predetermined fee as compensation for the service of providing the classification results classified from the acquired tissue images 31.

[0118] [Variation] Classification device 1B may distribute a computer program (hereinafter referred to as "image analysis app") to a computer (for example, presentation device 5) deployed within medical institution H2, enabling the computer to function as image analysis device 1A. A computer on which the image analysis app is installed can function as image analysis device 1A. In this case, for example, classification device 1B may send a notice requesting payment to the computer that received and installed the image analysis app. This allows the administrator managing classification device 1B to receive a predetermined fee from medical institution H2 as payment for the service of providing the image analysis app. The notice requesting payment may also be sent to the credit card company or other entity with which the user of the computer that received and installed the image analysis app has a contract.

[0119] Furthermore, the number of features (including the first, second, and third feature counts) transmitted from a computer located within medical institution H2, where the image analysis application is installed, to the classification device 1B may be assigned various IDs. These IDs may include a classification number (patient ID) unique to the patient from whom the tissue shown in the tissue image 31 subject to analysis was taken, a medical institution ID indicating medical institution H2, the source of each tissue image 31, and a device ID unique to the image analysis device 1A that performed the image analysis.

[0120] With this configuration, medical institution H2 does not need to transmit the tissue images 31 to an external location (for example, an image analysis device 1A). Medical institution H2 can use an image analysis application to analyze each tissue image 31, calculate the number of first, second, and third features from each tissue image 31, and transmit these to the classification device 1B.

[0121] Since the tissue image 31 contains patient diagnostic information, when transmitting the tissue image 31 outside of medical institution H2, care must be taken to protect personal information. Adopting this configuration eliminates the need to transmit the tissue image 31 outside of medical institution H2. Furthermore, this configuration reduces the communication load compared to transmitting the tissue image 31 itself.

[0122] [Examples of implementation using software] The control block (particularly the control unit 2) of the classification device 1 may be implemented by logic circuits (hardware) formed on an integrated circuit (IC chip) or the like, or by software.

[0123] In the latter case, the classification device 1 includes a computer that executes instructions for a program, which is software that realizes each function. This computer includes, for example, one or more processors and a computer-readable recording medium that stores the program. The object of this disclosure is achieved when the processor reads the program from the recording medium and executes it in the computer. For example, a CPU (Central Processing Unit) can be used as the processor. As the recording medium, a "tangible medium that is not temporary," such as ROM (Read Only Memory), can be used, as well as tape, disk, card, semiconductor memory, programmable logic circuit, etc. It may also further include RAM (Random Access Memory) for deploying the program. Furthermore, the program may be supplied to the computer via any transmission medium capable of transmitting the program (such as a communication network or broadcast wave). One aspect of this disclosure can also be realized in the form of a data signal embedded in a carrier wave, in which the program is embodied by electronic transmission.

[0124] This disclosure is not limited to the embodiments described above, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of this disclosure. [Explanation of symbols]

[0125] 1, 1a, 1B classification device 1A Image analysis device 4 External equipment 5 Presentation device 22 Monochromatic component image generation unit 23 Binarization section 24 Feature Count Calculation Unit 25 Classification Department S2 Monochromatic component image generation step S3 Binarization step S4 Feature Count Calculation Step S5, S6 Classification Steps

Claims

1. A monochromatic component image generation step, which generates multiple monochromatic component images based on pixel values ​​corresponding to each of the multiple color components constituting the tissue image, from a tissue image in which the cell nuclei of cells contained in living tissue and components other than the cell nuclei are depicted with different hues, A binarization step that generates multiple binarized images with different binarization reference values ​​from each of the multiple monochromatic component images, A feature number calculation step for each of the plurality of binarized images generated from each of the plurality of monochromatic component images, which calculates a first feature number indicating the number of hole-shaped regions consisting of pixels of the second pixel value after binarization, surrounded by pixels of the first pixel value after binarization to a first pixel value and a second pixel value; a second feature number indicating the number of connected regions formed by the connection of pixels of the first pixel value; and a third feature number which is the ratio of the first feature number to the second feature number. A classification step includes input data, which includes combinations of the first, second, and third feature numbers calculated for each of the binarized images generated from each of the plurality of monochromatic component images, to a classification model that models the correspondence between the first, second, and third feature numbers and the classification of changes occurring in the tissue, and outputting a classification result regarding changes occurring in the tissue shown in the tissue image. Image analysis methods.

2. The tissue image is an image of the tissue in which the cell nuclei are stained with a component capable of staining with a first color, and the cytoplasm is stained with a second color different from the first color. The image analysis method according to claim 1.

3. The classification model is generated by machine learning using the following as training data: (1) training tissue images in which the cell nuclei of cells contained in the tissue and components other than the cell nuclei are depicted in different hues, and which are pre-assigned by medical professionals with classification information indicating the classification of changes occurring in the tissue depicted in the training tissue images; and (2) data including combinations of the first feature number, the second feature number, and the third feature number calculated in the feature number calculation step from the training tissue images. The image analysis method according to claim 1.

4. The aforementioned classification models were generated using perceptrons, logistic regression, k-nearest neighbors, support vector machines, decision trees, random forests, or gradient boosting. The image analysis method according to claim 3.

5. The aforementioned tissue image is an image taken from the tissue sample taken from the subject's body. The image analysis method according to any one of claims 1 to 4.

6. The aforementioned tissue is lung tissue, In the classification step, if a change has occurred in the lung tissue, the change is classified as emphysema, atypical adenomatous hyperplasia, squamous adenocarcinoma, or invasive adenocarcinoma. The image analysis method according to claim 5.

7. A monochromatic component image generation unit that generates a plurality of monochromatic component images based on pixel values ​​corresponding to each of a plurality of color components constituting a tissue image, from a tissue image in which the cell nuclei of cells contained in the tissue of a living organism and components other than the cell nuclei are depicted in different hues; a binarization unit that generates a plurality of binarized images with different binarization reference values ​​from each of the plurality of monochromatic component images; and for each of the plurality of binarized images generated from each of the plurality of monochromatic component images, a first feature number indicating the number of hole-shaped regions consisting of pixels of the second pixel value after binarization, surrounded by pixels of the first pixel value after binarization, and the pixels of the first pixel value are connected An image analysis device comprising: a feature number calculation unit that calculates a second feature number indicating the number of connected regions and a third feature number which is the ratio of the first feature number to the second feature number; a classification unit that obtains the first feature number, the second feature number and the third feature number from an image analysis device, inputs input data including the first feature number, the second feature number and the third feature number calculated for each of the binarized images generated from each of the plurality of monochromatic component images into a classification model that models the correspondence between the first feature number, the second feature number and the third feature number and the classification of changes occurring in the tissue, and outputs a classification result regarding changes occurring in the tissue image. Equipped with, Classification device.

8. A monochromatic component image generation unit generates multiple monochromatic component images based on pixel values ​​corresponding to each of the multiple color components constituting the tissue image, from a tissue image in which the cell nuclei of cells contained in living tissue and other components other than the cell nuclei are depicted with different hues. A binarization unit that generates multiple binarized images with different binarization reference values ​​from each of the multiple monochromatic component images, A feature number calculation unit calculates, for each of the plurality of binarized images generated from each of the plurality of monochromatic component images, a first feature number indicating the number of hole-shaped regions consisting of pixels of the second pixel value after binarization, surrounded by pixels of the first pixel value after binarization into a first pixel value and a second pixel value; a second feature number indicating the number of connected regions formed by the connection of pixels of the first pixel value; and a third feature number which is the ratio of the first feature number to the second feature number. A classification unit inputs input data, including the first feature number, second feature number, and third feature number calculated for each of the binarized images generated from each of the plurality of monochromatic component images, into a classification model that models the correspondence between the first feature number, second feature number, and third feature number and the classification of changes occurring in the tissue, and outputs a classification result regarding changes occurring in the tissue shown in the tissue image. Equipped with, Classification device.

9. A computer program for calculating the first feature count, the second feature count, and the third feature count from the tissue image is distributed. The classification device according to claim 7 or 8.

10. A notice is sent to the computer on which the aforementioned computer program is installed, requesting payment for providing the computer program. The classification device according to claim 9.

11. An image analysis device comprising: a monochromatic component image generation unit that generates a plurality of monochromatic component images based on pixel values ​​corresponding to each of a plurality of color components constituting a tissue image, from a tissue image in which the cell nuclei of cells contained in the tissue of a living organism and components other than the cell nuclei are depicted in different hues; a binarization unit that generates a plurality of binarized images with different binarization reference values ​​from each of the plurality of monochromatic component images; and a feature number calculation unit that calculates, for each of the plurality of binarized images generated from each of the plurality of monochromatic component images, a first feature number indicating the number of hole-shaped regions consisting of pixels of the second pixel value after binarization, surrounded by pixels of the first pixel value after binarization to a first pixel value and a second pixel value; a second feature number indicating the number of connected regions formed by the connection of pixels of the first pixel value; and a third feature number which is the ratio of the first feature number to the second feature number. A classification device comprising: an image analysis device that acquires the first feature number, the second feature number, and the third feature number, inputs the input data including the first feature number, the second feature number, and the third feature number calculated for each of the binarized images generated from each of the plurality of monochromatic component images into a classification model that models the correspondence between the first feature number, the second feature number, and the third feature number and the classification of changes occurring in the tissue, and outputs a classification result regarding changes occurring in the tissue shown in the tissue image; An external device that transmits the tissue image to the image analysis device, The presenting device includes a device that acquires the classification results output from the classification device and presents the classification results, Classification system.

12. A control program for causing a computer to function as a classification device according to claim 7 or 8, wherein the control program causes the computer to function as the classification unit.

13. A computer-readable recording medium that stores the control program described in claim 12.

Images