Abnormal blood vessel detection system, abnormal blood vessel detection method, and abnormal blood vessel detection program

Abstract

To provide an abnormal blood vessel detection system, an abnormal blood vessel detection method, and an abnormal blood vessel detection program that can automatically detect abnormal blood vessels on a tissue section.SOLUTION: The abnormal blood vessel detection system 1 is a system for detecting abnormal blood vessels that exhibit abnormalities in the vascular pathway morphology on a tissue section, and includes an acquisition unit 11 that acquires an image D1 of a pathological tissue section of a solid cancer that has been immunostained, an extraction unit 12 that uses artificial intelligence to recognize the blood vessels on the acquired image D1 and extract only the blood vessels, and a determination unit 13 that uses artificial intelligence to determine whether or not the extracted blood vessels include abnormal blood vessels.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to an abnormal blood vessel detection system, an abnormal blood vessel detection method, and an abnormal blood vessel detection program for detecting abnormal blood vessels on a tissue section. [Background technology]

[0002] It is known that tumors proliferate blood vessels as they grow, and it has been reported that their morphology differs from that of normal blood vessels. Immunostaining is a method of specifically staining target proteins, and by targeting and staining proteins expressed in blood vessels, it is possible to make the blood vessels easier to see. Furthermore, it has been reported that the blood vessels themselves can be recognized using AI (artificial intelligence), and that by combining multiple immunostains and comparing their expression, it is possible to detect vascular abnormalities. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7253177 [Patent Document 2] Japanese Patent Publication No. 2022-112407 [Patent Document 3] Patent Publication No. 2021-002303 [Non-patent literature]

[0004] [Non-Patent Document 1] Eur Arch Otorhinolaryngol.2020;277(10):2893-2906.doi:10.1007 / s00405-020-06097-2. [Non-patent document 2] Journal of the Japanese Society of Pathology 111(1)254-255 March 2022, Double immunohistochemistry for tumor vascular identification and its application to breast cancer prognosis, Ning, Yin-Wing-Swe, Akihiro Matsukawa, Masayoshi Fujisawa Summary of the Invention [Problem to be solved by the invention]

[0005] Abnormalities in tumor vascular morphology are considered to be of high diagnostic value because they are related to cancer diagnosis and prognosis. Microvessel density (MVD), the number of newly formed blood vessels per unit area, has long been known as a method for analyzing abnormal tumor blood vessels in tissue sections. However, the reproducibility and clinical significance of MVD have not yet been established, and therapeutic interventions based on MVD have not been performed. Recently, a new definition of abnormal tumor blood vessels has been reported, focusing on the structural complexity of tumor blood vessels and including C-shaped vessels and abnormal branching vessels (Non-Patent Document 2). C-shaped vessels (vessels encircling the periphery of cancer cell nests by more than 180°) and abnormal branching vessels (vessels branching in four or more directions from a single point near cancer cell nests) have been defined as vessels with complex morphologies.

[0006] In breast cancer, blood vessels were stained with CD31 antibodies in tissue sections from 304 breast cancer patients who underwent surgery at Okayama University Hospital, and analysis of these blood vessels revealed that cases with microscopically detected C-shaped blood vessels or abnormal branching blood vessels had poor disease-free survival and overall survival. However, there is currently no method for automatically identifying abnormalities in the vascular course (C-shaped blood vessels and abnormal branching blood vessels) (see, for example, Patent Documents 1 to 3).

[0007] To automatically identify the morphology of abnormal blood vessels (C-shaped vessels and abnormal branching vessels) on tissue sections, it is first necessary to stain the vessels with immunostaining and have the machine recognize them. Previously, methods have been reported for identifying the presence of blood vessels using hematoxylin-eosin staining (HE staining), a common stain used in pathological examinations, but this method is unable to identify all blood vessels. Previously, it has been reported that blood vessels are stained with immunostaining and then identified by artificial intelligence, and colored vessels have been recognized as blood vessels. While this method makes it possible to determine whether or not blood vessels are present within the observation area, in order to detect vascular morphological abnormalities, it is necessary to devise a method that can recognize the vascular morphology, and no such technology exists to date.

[0008] The present invention is intended to solve these conventional problems, and aims to provide an abnormal blood vessel detection system, an abnormal blood vessel detection method, and an abnormal blood vessel detection program that can automatically detect abnormal blood vessels on tissue sections. [Means for solving the problem]

[0009] One aspect of the present invention is an abnormal blood vessel detection system that detects abnormal blood vessels that exhibit abnormalities in the vascular course morphology on tissue sections, and includes an acquisition unit that acquires images of pathological tissue sections of solid cancers that have been immunostained, an extraction unit that uses artificial intelligence to recognize blood vessels on the acquired images and extract only the blood vessels, and a determination unit that uses artificial intelligence to determine whether the extracted blood vessels include the abnormal blood vessels.

[0010] Another aspect of the present invention is an abnormal blood vessel detection method for detecting abnormal blood vessels that exhibit abnormalities in the vascular course morphology on a tissue section, in which a computer executes an acquisition step of acquiring an image of a pathological tissue section of a solid cancer that has been subjected to immunostaining, an extraction step of recognizing blood vessels on the acquired image using artificial intelligence and extracting only the blood vessels, and a determination step of determining using artificial intelligence whether the extracted blood vessels include the abnormal blood vessels.

[0011] Another aspect of the present invention is an abnormal blood vessel detection program for detecting abnormal blood vessels that exhibit abnormalities in the vascular course morphology on a tissue section, which causes a computer to execute an acquisition step of acquiring an image of a pathological tissue section of a solid cancer that has been immunostained, an extraction step of recognizing blood vessels on the acquired image using artificial intelligence and extracting only the blood vessels, and a determination step of determining using artificial intelligence whether the extracted blood vessels include the abnormal blood vessels. [Effects of the Invention]

[0012] According to the present invention, it is possible to provide an abnormal blood vessel detection system, an abnormal blood vessel detection method, and an abnormal blood vessel detection program that can automatically detect abnormal blood vessels on a tissue section. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a schematic diagram for explaining an overall image of an embodiment of the present invention; [Figure 2] 1 is a system configuration diagram illustrating an example of an abnormal blood vessel detection system according to an embodiment of the present invention. [Figure 3] FIG. 3 is a functional block diagram showing the main parts of the abnormal blood vessel detection system shown in FIG. 2. [Figure 4] 2 is a flowchart showing a specific example of step S2 shown in FIG. [Figure 5] FIG. 5 is a schematic diagram for explaining the process shown in FIG. 4. [Figure 6] 1. FIG. 4 is a diagram showing an example of blood vessel recognition in step S2 shown in FIG. [Figure 7] 2 is a flowchart showing a specific example of step S3 shown in FIG. 1. [Figure 8] FIG. 8 is a schematic diagram for explaining the processing shown in FIG. 7. [Figure 9] FIG. 2 is a diagram showing an example of an abnormal blood vessel running morphology used in the learning in step S3 shown in FIG. 1. [Figure 10] 1. FIG. 4 is a schematic diagram showing an example of displaying the results determined in step S3 shown in FIG. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The embodiments shown below are merely examples, and the present invention can be implemented in various forms with various modifications and improvements based on the knowledge of those skilled in the art. Note that components with the same reference numerals in this specification and drawings represent the same components.

[0015] ≪Overall picture≫ Fig. 1 is a schematic diagram for explaining an overall image of an embodiment of the present invention. The present invention enables the presence or absence of specific abnormal blood vessels (C-shaped blood vessels and abnormal branching blood vessels) that appear due to the influence of tumors to be determined even by those without specialized knowledge or experience.

[0016] In other words, in a typical diagnosis, a specimen is prepared from an organ removed from a patient, stained with conventional staining, a pathological diagnosis is made based on the specimen, and a treatment plan is decided based on the diagnostic results, as shown in Figure 1. Conventional staining is, for example, hematoxylin and eosin staining (HE staining).

[0017] In contrast, in an embodiment of the present invention, blood vessels are first stained with immunostaining (step S0). Because hematoxylin-eosin staining cannot identify all blood vessels, blood vessels are stained with immunostaining. Next, an image D1 of the immunostained pathological tissue section is acquired using a general-purpose microscope digital camera or a virtual slide scanner (step S1). Of course, the term "image" here refers to digitized image data. In image D1, the area indicated by reference numeral 100 is the nucleus, and the area indicated by reference numeral 101 (the dark-colored area) is the stained blood vessel. However, the dark-colored area may contain non-vascular objects. Therefore, in an embodiment of the present invention, only blood vessels are extracted using artificial intelligence using the method described below, and then abnormal blood vessels are identified using artificial intelligence (steps S2 and S3). This two-step process allows blood vessels to be correctly extracted from various structures, making it possible to correctly identify abnormal blood vessels 102 from various blood vessels. An image D3 showing the results of the analysis is displayed on a display device and used as reference information for determining a treatment plan.

[0018] <Abnormal Blood Vessel Detection System> FIG. 2 is a system configuration diagram illustrating an example of an abnormal blood vessel detection system 1 according to an embodiment of the present invention. As shown in FIG. 2, a trained pathologist 2 operates a microscope digital camera 3 equipped with an imager 4 to capture images of immunostained pathological tissue sections of solid cancer. The captured images D1 are transferred to the cloud and stored as big data in a cloud server 5. Learning data is generated based on this big data and provided to a high-performance PC 6. The high-performance PC 6 is a computer equipped with an artificial intelligence engine for big data. This artificial intelligence algorithm 7, such as a conditional GAN ​​(CGAN) or a convolutional neural network (CNN), is continuously improved via a management terminal 9 operated by a computer science researcher 8. The computer science researcher 8 can discuss and learn from the pathologist 2 to improve the artificial intelligence algorithm 7 based on the latest information.

[0019] FIG. 3 is a functional block diagram showing the main components of the abnormal blood vessel detection system 1 shown in FIG. 2. As shown in FIG. 3, the abnormal blood vessel detection system 1 functionally includes an acquisition unit 11, an extraction unit 12, a determination unit 13, a display unit 14, a storage unit 15, an operation unit 16, and a reading unit 17. The acquisition unit 11 acquires an image D1 of a pathological tissue section of a solid cancer that has been immunostained. The acquisition unit 11 may be the microscope digital camera 3 or a virtual slide scanner itself, or may be a functional unit that acquires the image D1 from the microscope digital camera 3 or the virtual slide scanner. The extraction unit 12 is a functional unit that uses artificial intelligence to recognize blood vessels in the acquired image D1 and extract an image D2 containing only blood vessels. The determination unit 13 is a functional unit that uses artificial intelligence to determine whether or not abnormal blood vessels are included in the extracted image D2 containing only blood vessels. These functional units can be implemented by a computer using a CPU or GPU to execute a program loaded into memory. The display unit 14 is a display device that displays the determination results. The storage unit 15 is a storage device that stores various data and programs. The operation unit 16 is a keyboard, mouse, etc. that are operated by a user such as a computer science researcher 8. The reading unit 17 is a device that reads data and programs recorded on a computer-readable recording medium. In addition, the computer may be provided with various other functional units that are provided in a general computer. Of course, each of these functional units may be realized by a plurality of physically different computers (for example, a high-performance PC 6, an administration terminal 9, a cloud server 5) or their peripheral devices.

[0020] <Abnormal blood vessel detection method> The abnormal blood vessel detection method according to the embodiment of the present invention will be described in more detail below with reference to Fig. 1. This abnormal blood vessel detection method includes steps S1 to S3, which will be described below.

[0021] In step S1, an image D1 of an immunostained pathological tissue section of a solid cancer is acquired. A general-purpose device such as a digital camera for a microscope or a scanner can be used to acquire the image D1. The captured file can be in any format or size, such as JPG, GIF, PND, or TIFF, and may be stored in WSI format.

[0022] In step S2, artificial intelligence is used to recognize blood vessels in the acquired image D1, and an image D2 of only the blood vessels is extracted. For example, the vascular morphology alone is accurately output from a micrograph of a pathological tissue section that has been immunostained for blood vessels. The antibody used for immunostaining may be CD31, CD34, factor VIII, or another antibody that has some degree of specificity for blood vessels. In this embodiment, CD31 is preferred because it has the highest sensitivity and can completely depict the outline of blood vessels.

[0023] In step S3, artificial intelligence is used to determine whether or not abnormal blood vessels are included in the extracted image D2 of only blood vessels. For example, the presence or absence of abnormal blood vessels (C-shaped blood vessels, abnormal branching blood vessels) is determined based on the blood vessel pattern. The determination results are used as reference information for determining treatment plans.

[0024] <Details of Step S2> Fig. 4 is a flowchart showing a specific example of step S2. Here, the explanation is given assuming that there is an artificial intelligence that has performed machine learning using a CGAN or the like (step S20). For example, as shown in Fig. 4, when an image D1 acquired in step S1 is input (step S21), the image D1 is divided into regions of a predetermined size (step S22), blood vessels in each divided image are recognized by the artificial intelligence (step S23), the recognized blood vessels are restored and reconstructed to their original size (step S24), and an image D2 showing only the reconstructed blood vessel running pattern is output (step S25).

[0025] FIG. 5 is a schematic diagram for explaining the process shown in FIG. 4. Of course, this schematic diagram is merely an example. The artificial intelligence used to identify blood vessels is created by machine learning the annotation image DA created by a trained pathologist 2 using a neural network (CNN), a conditional generative adversarial network (CGAN), or the like. To achieve high-precision learning, it is necessary to divide the area for learning. The specimen size is 100x100μm to 600x600μm, and the area is divided into approximately 256x256 pixels for learning. For example, as shown in FIG. 5(A), the corresponding area of ​​the original image D1 (1200x1600 pixels) and the annotation image DA is extracted, and the 256x256 pixel images D1(1) and DA(1) are used as paired images for learning. Similarly, the entire area of ​​images D1 and DA is divided like tiles to generate paired images for learning. Dividing images D1 and DA like tiles in this way enables high-precision (high-resolution) learning.

[0026] The same applies to blood vessel recognition. That is, to enable highly accurate recognition of blood vessels, as shown in Figure 5(B), the original image D1 is divided into portions of approximately the same size as that used during learning, and the blood vessels on each divided image are recognized using artificial intelligence, and an image showing only the blood vessel running pattern is output (AI output image). Then, as shown in Figure 5(C), this AI output image is reconstructed and restored to image D2 of the same size as the original image D1. This makes it possible to accurately extract only the blood vessel running pattern from the original image D1 and output it as image D2 of the same size as the original image D1.

[0027] Figure 6 shows an example of blood vessel recognition in a microscopic image of a pathological section using AI with machine learning. Figure 6(A) shows the CD31 immunostained image D1 obtained in step S1. Figure 6(B) shows the learning annotation image DA created by trained pathologist 2. Figure 6(C) shows the AI ​​output image D2 output in step S2. Deep learning was performed in a Docker container environment on a computer equipped with an RTX4060Ti with 16GB of memory. CGAN was used to create the AI ​​that outputs the vascular annotation image DA. The vascular morphology was extracted from each segmented image, and the vascular morphology in the reconstructed image had a recall rate of 48%, specificity of 99%, precision of 64%, and F-measure of 51%, demonstrating that the system accurately depicts blood vessels by excluding nonspecific staining such as inflammatory cells.

[0028] <Details of Step S3> Fig. 7 is a flowchart showing a specific example of step S3. Here, too, the explanation will be given assuming that there is artificial intelligence that has performed machine learning using a CGAN or the like (step S30). For example, as shown in Fig. 7, when image D2 output in step S2 is input (step S31), the image is successively divided so that each divided image partially overlaps (step S32), and the artificial intelligence determines whether or not each divided image contains abnormal blood vessels (presence or absence of abnormalities in the running morphology of the blood vessels) (step S33), and the determination result is output (step S34).

[0029] FIG. 8 is a schematic diagram for explaining the process shown in FIG. 7. Of course, this schematic diagram is merely an example. In step S3, as in step S2, the artificial intelligence used to identify abnormal blood vessels is created by machine learning using a CGAN or the like on annotation images DA created by a trained pathologist 2. Furthermore, for highly accurate learning, it is necessary to divide the area for learning. The size of the specimen is determined by dividing an area of ​​100x100 μm to 600x600 μm on the specimen into approximately 256x256 pixels for learning. For example, as shown in FIG. 8(A), an annotation image DAs containing only normal blood vessels and an annotation image DAi containing abnormal blood vessels are prepared. Images of blood vessels contained in these annotation images DAs and DAi are cut out and placed within a 244x244 pixel area. As a result, the images are classified into a group image Gs of normal blood vessels and a group image Gi of abnormal blood vessels, and these group images Gs and Gi are used as training data. The extracted blood vessel images were randomly rotated within a range of -30 degrees to +30 degrees, then randomly flipped horizontally and then randomly flipped vertically, and training data was increased to perform learning.

[0030] Image D2 is also divided when determining abnormalities in the blood vessel vein pattern. However, in order to accurately determine abnormalities in the blood vessel vein pattern, it is preferable to divide image D2 into small overlapping images as shown in FIG. 8(B) and continuously determine each divided image, rather than simply dividing image D2 like tiles. This is because simply dividing image D2 like tiles may result in the blood vessel vein pattern being divided, making it difficult to accurately determine the abnormality. If image D2 is divided into small overlapping images, if an abnormality in the blood vessel vein pattern is found, the entire vein pattern will be included in one of the divided images, allowing for more accurate determination of abnormalities in the blood vessel vein pattern. The degree of overlap is not particularly limited, but in order to accurately determine abnormalities in the blood vessel vein pattern, it is preferable to overlap each divided image so that the center point O of each divided image is included in the other divided images, as shown in FIG. 8(B).

[0031] Figure 9 shows group images Gi of abnormal blood vessels (C-shaped vessels and abnormal branching vessels) annotated by trained pathologist 2. In addition, blood vessels other than these abnormal vessels were similarly annotated as normal blood vessels, and these group images Gs and Gi were used as training data. Deep learning was performed in a Docker container environment on a computer equipped with an RTX4060Ti with 16GB of memory. To create the AI ​​that outputs the annotated blood vessel images DA, we used pix2pix (https: / / github.com / affinelayer / pix2pix-tensorflow), which implements CGAN. The segmented images were reconstructed, and if even one image in the consecutively judged images was found to have an abnormal vascular morphology, the original image D1 was judged to contain an abnormal vascular morphology. Fifteen types of annotation images DAs containing only normal blood vessels were tested, and 13 were correctly judged as normal (86% accuracy rate). In addition, when 15 types of annotation images DAi containing abnormal blood vessels were tested, 14 were correctly judged to be abnormal (accuracy rate 93%).

[0032] <Example of judgment result display> FIG. 10 is a schematic diagram showing an example of the display of the results of the determination made in step S3. FIG. 10(A) shows the case where abnormal blood vessels are present, and FIG. 10(B) shows the case where abnormal blood vessels are not present. As shown in FIG. 10, a heat map image M may be created in which the likelihood of the determination made in step S3 is expressed by differences in color (including "differences in color intensity"), and the heat map image M may be superimposed on image D2, and the superimposed image D3 may be displayed on the display unit 14. The "likelihood of determination" may also be rephrased as "the likelihood of an abnormality." The image on which the heat map image M is superimposed may be image D1 instead of image D2. This not only makes it possible to immediately find characteristic areas of tissue, but also to immediately grasp the likelihood of the abnormality.

[0033] As described above, by using the abnormal blood vessel detection system 1 according to the embodiment of the present invention, it becomes possible to determine the presence or absence of specific abnormal blood vessels (C-shaped blood vessels, abnormal branching blood vessels) that appear due to the influence of tumors, even if one does not have specialized knowledge or experience. Since abnormalities in the vascular morphology of tumors are related to the diagnosis and prognosis of cancer, this invention can be said to have high diagnostic value.

[0034] 4 (step S2) illustrates the case where the divided images are reconstructed to the size of the original image, but instead of reconstructing the entire image as described above, partial reconstruction may be performed. For example, it is also possible to reconstruct only characteristic regions of tissues contained in the original image.

[0035] 7 (step S3), the segmented images may be reconstructed to the size of the original image, or only the characteristic regions may be partially reconstructed. The object to be determined in step S3 for the presence of abnormal blood vessels may be an individual segmented image, a partially reconstructed image, or an image reconstructed to the size of the original image.

[0036] In addition, in the above description, a case where a heat map is displayed (see FIG. 10) has been exemplified, but the display example of the determination result is not limited to this. For example, abnormal blood vessels may be displayed as presence / absence (text), graphics, or numerical values ​​(number of abnormal blood vessels, position information, etc.). In addition, the determination result may be output as voice, or may be transferred to another device and input into an application program running on that device.

[0037] Furthermore, although the above description exemplifies the case of detecting abnormal blood vessels, the detection target is not limited to this. That is, the abnormal blood vessel detection system 1 according to the embodiment of the present invention can detect not only blood vessels with abnormal shapes but also lymphatic vessels with abnormal shapes. It may also be applicable to various other cases where it is necessary to distinguish between objects with distinctive shapes. This disclosure will likely lead to the identification of various alternative embodiments and operational techniques in the future.

[0038] Characteristic configurations and effects of embodiments of the present invention As described above, the abnormal blood vessel detection system 1 according to the embodiment of the present invention is a system for detecting abnormal blood vessels that exhibit abnormalities in the vascular morphology on a tissue section, and includes an acquisition unit 11 that acquires an image D1 of a pathological tissue section of a solid cancer that has been subjected to immunostaining, an extraction unit 12 that uses artificial intelligence to recognize blood vessels on the acquired image D1 and extract only the blood vessels, and a determination unit 13 that uses artificial intelligence to determine whether or not abnormal blood vessels are included among the extracted blood vessels. This enables automatic detection of abnormal blood vessels on a tissue section, making it possible for even a non-trained pathologist 2 to diagnose the presence or absence of abnormal blood vessel morphology (C-shaped blood vessels, abnormal branching blood vessels).

[0039] Furthermore, it is desirable that the extraction unit 12 divides the acquired image D1 into regions of a predetermined size, recognizes blood vessels on each divided image, and reconstructs the recognized blood vessels. This allows the image to be recognized with high resolution, making it possible to accurately extract only the running morphology of blood vessels from the original image D1.

[0040] Furthermore, it is desirable that the determination unit 13 divides the extracted image D2 into regions of a predetermined size and determines whether or not abnormal blood vessels are included in each divided image. This allows the image to be determined with high resolution, making it possible to accurately determine abnormalities in the running pattern.

[0041] Furthermore, it is desirable for the determination unit 13 to continuously divide the extracted image D2 so that each divided image partially overlaps. This ensures that if an abnormality in the driving pattern is detected, the entire driving pattern will be included in one of the divided images, making it possible to more accurately determine whether the driving pattern is abnormal.

[0042] Furthermore, a display unit 14 is provided to display the results of the determination, and it is desirable that the display unit 14 displays a heat map in which the certainty of the determination is expressed by different colors. This not only enables abnormal blood vessels to be found immediately, but also makes it possible to immediately grasp the likelihood of the abnormality.

[0043] Furthermore, it is desirable to use CD31 as the antibody for immunostaining. Although any antibody with a certain degree of specificity for blood vessels, such as CD31, CD34, or factor VIII, can be used for immunostaining, CD31 is the most sensitive and can completely depict the outline of blood vessels.

[0044] Abnormal blood vessels include either C-shaped blood vessels, in which blood vessels surround the outer periphery of cancer cell nests by more than 180°, or abnormal branching blood vessels, in which blood vessels branch out in four or more directions from a single point near cancer cell nests.

[0045] In addition, solid cancers include any one of breast cancer, oral cancer, pharyngeal cancer, esophageal cancer, gastric cancer, colon cancer, liver cancer, bile duct cancer, pancreatic cancer, lung cancer, cervical cancer, uterine cancer, ovarian cancer, kidney cancer, ureter cancer, bladder cancer, prostate cancer, thyroid cancer, and skin cancer.

[0046] The present invention can also be realized as an abnormal blood vessel detection method in which each step corresponds to each of the characteristic functional units of the abnormal blood vessel detection system 1, or as an abnormal blood vessel detection program for causing a computer to execute those steps. Needless to say, such a program can be installed in a computer via a computer-readable recording medium or a network such as the Internet. [Explanation of symbols]

[0047] 1. Abnormal blood vessel detection system 11 Acquisition Department 12 Extraction part 13 Judgment section 14 Display section 15 Storage Unit

Claims

1. An abnormal blood vessel detection system that detects abnormal blood vessels that show abnormalities in the vascular course morphology on tissue slices, comprising: an acquisition unit that acquires images of pathological tissue slices of solid cancer that have been immunostained; an extraction unit that uses artificial intelligence to recognize blood vessels on the acquired images and extract only the blood vessels; and a determination unit that uses artificial intelligence to determine whether the extracted blood vessels include the abnormal blood vessels.

2. 2. The abnormal blood vessel detection system according to claim 1, wherein the extraction unit divides the acquired image into regions of a predetermined size, recognizes blood vessels on each divided image, and reconstructs the recognized blood vessels.

3. 3. The abnormal blood vessel detection system according to claim 2, wherein the determination unit divides the extracted image into regions of a predetermined size and determines whether the abnormal blood vessel is included in each divided image.

4. The abnormal blood vessel detection system according to claim 3 , wherein the determining unit successively divides the extracted image so that each divided image partially overlaps with another.

5. The abnormal blood vessel detection system according to claim 4 , further comprising a display unit that displays the judgment result, wherein the display unit displays a heat map in which the certainty of the judgment is expressed by different colors.

6. The abnormal blood vessel detection system according to claim 1, wherein the antibody used in the immunostaining is CD31.

7. The abnormal blood vessel detection system of claim 1, wherein the abnormal blood vessels include either a C-shaped blood vessel in which the blood vessel surrounds the outer periphery of the cancer cell nest by more than 180°, or an abnormal branching blood vessel in which the blood vessel branches in four or more directions from one point near the cancer cell nest.

8. 2. The abnormal blood vessel detection system of claim 1, wherein the solid cancer comprises any one of breast cancer, oral cancer, pharyngeal cancer, esophageal cancer, gastric cancer, colon cancer, liver cancer, bile duct cancer, pancreatic cancer, lung cancer, cervical cancer, uterine cancer, ovarian cancer, kidney cancer, ureter cancer, bladder cancer, prostate cancer, thyroid cancer, and skin cancer.

9. An abnormal blood vessel detection method for detecting abnormal blood vessels that exhibit abnormalities in the vascular course morphology on a tissue section, the abnormal blood vessel detection method comprising the steps of: an acquisition step in which a computer acquires an image of a pathological tissue section of a solid cancer that has been immunostained; an extraction step in which the computer uses artificial intelligence to recognize the blood vessels on the acquired image and extract only the blood vessels; and a determination step in which the computer uses artificial intelligence to determine whether the abnormal blood vessels are included in the extracted blood vessels.

10. An abnormal blood vessel detection program for detecting abnormal blood vessels that exhibit abnormalities in the vascular course morphology on a tissue section, the abnormal blood vessel detection program causing a computer to execute the following steps: an acquisition step for acquiring an image of a pathological tissue section of a solid cancer that has been immunostained; an extraction step for recognizing blood vessels on the acquired image using artificial intelligence and extracting only the blood vessels; and a determination step for determining using artificial intelligence whether the extracted blood vessels include the abnormal blood vessels.

Images