Drug discovery assistance device, method for operating drug discovery assistance device, and program for operating drug discovery assistance device

Abstract

This drug discovery assistance device is provided with a processor, wherein the processor: acquires a sample image in which a tissue sample of an organ of a subject provided to an evaluation test of a candidate substance is captured; detects, in a portion of the sample image, an estimated form abnormality portion in which a form abnormality is estimated to occur; derives determination reference information from the detection result of the estimated form abnormality portion; determines, on the basis of the determination reference information, whether the estimated form abnormality portion is overdetected or the estimated form abnormality portion is underdetected; and presents, to a user, first cause identification reference information contributing to the identification of a cause by which the estimated form abnormal part is overdetected when it is determined that the estimated form abnormal portion is overdetected, and presents, to the user, second cause identification reference information contributing to the identification of a cause by which the estimated form abnormal portion is underdetected when it is determined that the estimated form abnormal portion is underdetected.

Description

Drug discovery support device, drug discovery support device operation method, and drug discovery support device operation program

[0001] The technology of the present disclosure relates to a drug discovery support device, an operating method for the drug discovery support device, and an operating program for the drug discovery support device.

[0002] In the field of drug discovery, drug candidate substances are administered to test subjects such as rats to evaluate the efficacy and toxicity of the candidate substances. In such evaluation tests, specimen images of tissue specimens (brain specimens, liver specimens, heart specimens, etc.) collected by autopsy of the test subjects are used.

[0003] For example, Japanese Patent Application Laid-Open No. 2022-007281 discloses the following technology. Specifically, partial regions (e.g., regions corresponding to specific features occurring in a specific disease) are extracted from a specimen image to be analyzed by segmentation. Then, feature amounts (features output from a neural network, color features, shape features, etc.) that quantify characteristics such as the morphology of the tissue specimen depicted in the partial regions are calculated. Next, the specimen image with information about the partial regions superimposed is displayed to the user, and the user is prompted to specify partial regions belonging to the same category. Then, auxiliary information about the features related to the categories (e.g., the contribution of each feature when classifying the partial regions into categories) is generated and displayed to the user.

[0004] In the evaluation of drug candidate substances, morphological abnormalities occurring in tissue specimens shown in specimen images are detected. Conventionally, specimen images were examined by experts such as pathologists to detect areas where morphological abnormalities are suspected (hereinafter referred to as suspected morphological abnormalities). However, with recent advances in image analysis technology, techniques have been developed to automatically detect suspected morphological abnormalities without the need for experts.

[0005] When detecting suspected morphological abnormalities using image analysis in this way, it is important to note that there are cases where a part where no morphological abnormality has occurred is mistakenly judged to be a suspected morphological abnormality, resulting in an over-detection of the suspected morphological abnormality, and conversely, there are cases where a part where a morphological abnormality has occurred is mistakenly judged not to be a suspected morphological abnormality, resulting in an under-detection of the suspected morphological abnormality.

[0006] There are various possible reasons why an estimated morphological abnormality may be over-detected or under-detected. If the cause can be identified and appropriate measures can be taken, the accuracy of detecting morphological abnormalities occurring in tissue specimens shown in specimen images can be improved. However, such a technology has not been proposed in the past, including in JP 2022-007281 A.

[0007] One embodiment of the technology of the present disclosure provides a drug discovery support device, an operating method for the drug discovery support device, and an operating program for the drug discovery support device that can contribute to improving the accuracy of detecting morphological abnormalities occurring in tissue specimens shown in specimen images.

[0008] The drug discovery support device of the present disclosure includes a processor, which acquires a specimen image of a tissue specimen of an organ of a subject subjected to an evaluation test of a candidate substance, detects a presumed morphologically abnormal portion of the specimen image where a morphological abnormality is presumed to have occurred, derives determination reference information from the detection result of the presumed morphologically abnormal portion, and determines whether the presumed morphologically abnormal portion is over-detected or under-detected based on the determination reference information.If it is determined that the presumed morphologically abnormal portion is over-detected, the processor presents to the user first cause identification reference information that helps to identify the cause of the presumed morphologically abnormal portion being over-detected, and if it is determined that the presumed morphologically abnormal portion is under-detected, the processor presents to the user second cause identification reference information that helps to identify the cause of the presumed morphologically abnormal portion being under-detected.

[0009] It is preferable that the processor acquires one specimen image, detects an estimated morphologically abnormal part from the one specimen image, and derives a numerical value representing the spatial arrangement state of the estimated morphologically abnormal part in the one specimen image as the judgment reference information.

[0010] Preferably, the processor determines whether the estimated morphological abnormality portion is over-detected or under-detected by comparing the numerical value with a preset determination threshold.

[0011] The numerical value is preferably a numerical value relating to the number of suspected morphologically abnormal portions.

[0012] It is preferable that the processor acquires a plurality of specimen images, detects estimated morphologically abnormal portions from each of the plurality of specimen images, and derives, as judgment reference information, a distribution of numerical values ​​representing the spatial arrangement state of the estimated morphologically abnormal portions of each of the plurality of specimen images.

[0013] The multiple specimen images are images of tissue specimens from multiple subjects belonging to the same group, and the processor preferably detects outliers from the numerical values ​​that make up the distribution to determine whether the estimated morphological abnormality portion is over-detected or under-detected.

[0014] The multiple specimen images are images of tissue specimens of multiple subjects belonging to different groups, and the processor preferably derives a distribution for each of the different groups, detects outliers from the numerical values ​​that make up one of the multiple distributions derived for each of the different groups, and determines whether the estimated morphological abnormality portion is over-detected or under-detected by referring to a distribution other than the one distribution.

[0015] The different groups are preferably an administration group administered with the candidate substance and a control group not administered with the candidate substance, or an administration group administered with the candidate substance and an administration group administered with a candidate substance identical or similar to the candidate substance in a previous evaluation test.

[0016] The numerical value is preferably a numerical value relating to the number of suspected morphologically abnormal portions.

[0017] It is preferable that the processor acquires multiple specimen images of tissue specimens from multiple subjects belonging to different groups, detects estimated morphologically abnormal parts from each of the multiple specimen images, and derives, as judgment reference information, a representative numerical value representing the spatial arrangement state of each estimated morphologically abnormal part in each of the multiple specimen images for each different group.

[0018] It is preferable that the processor determines whether the estimated morphological abnormality portion is over-detected or under-detected by comparing the representative value for each different group with a predetermined ideal value for each different group.

[0019] The different groups are preferably a treatment group to which the candidate substance is administered and a control group to which the candidate substance is not administered.

[0020] Preferably, the administration group includes multiple sub-administration groups with different doses of the candidate substance.

[0021] The numerical value is preferably a numerical value relating to the number of suspected morphologically abnormal portions.

[0022] There are multiple types of artifacts that may be erroneously detected as suspected morphological abnormalities, and it is preferable that the processor identify the type of artifact that may have been erroneously detected as suspected morphological abnormalities and present information regarding the type to the user as first cause identification reference information and second cause identification reference information.

[0023] Preferably, the processor presents to the user, as the first cause identification reference information and the second cause identification reference information, information regarding the color of the specimen image and information regarding the color of a reference specimen image of a tissue specimen deemed to be normal.

[0024] When the processor determines that the estimated morphological abnormality parts have been detected excessively, it is preferable that the processor performs a clustering process to define the cluster to which each of the multiple estimated morphological abnormality parts detected from one specimen image belongs, and presents to the user, as first cause identification reference information, multiple cluster images that reflect the results of the clustering process and that are generated by processing the specimen image, and that make the multiple clusters identifiable in a display format that is preset for each of the multiple clusters.

[0025] Preferably, the processor suggests to the user and / or performs a suppression process to suppress over-detection or under-detection of the suspected morphological abnormality portion.

[0026] The suppression process is preferably at least one of a process of changing a detection threshold used to detect a presumed morphologically abnormal portion, a process of correcting the color of the specimen image, and a process of changing the resolution of the specimen image in detecting a presumed morphologically abnormal portion.

[0027] The processor preferably suggests to the user and / or performs an exclusion process for excluding from the evaluation test any portion of an artifact that may have been erroneously detected as a suspected morphological abnormality.

[0028] Preferably, the processor detects estimated morphological abnormalities by treating each of multiple patch images obtained by dividing the specimen image as a part, and comparing the features obtained by inputting the patch images into a machine learning model with reference features obtained by inputting a reference patch image of a tissue specimen considered to be normal into the machine learning model.

[0029] The operating method of the drug discovery support device disclosed herein includes acquiring a specimen image of a tissue specimen of an organ of a subject used in an evaluation test of a candidate substance, detecting a presumed morphologically abnormal portion of the specimen image where a morphological abnormality is presumed to have occurred, deriving determination reference information from the detection result of the presumed morphologically abnormal portion, determining whether the presumed morphologically abnormal portion is over-detected or under-detected based on the determination reference information, and, if it is determined that the presumed morphologically abnormal portion is over-detected, presenting to the user first cause identification reference information that helps to identify the cause of the presumed morphologically abnormal portion being over-detected, and, if it is determined that the presumed morphologically abnormal portion is under-detected, presenting to the user second cause identification reference information that helps to identify the cause of the presumed morphologically abnormal portion being under-detected.

[0030] The operating program of the drug discovery support device disclosed herein causes a computer to execute processes including acquiring a specimen image of a tissue specimen from an organ of a subject used in an evaluation test of a candidate substance, detecting a presumed morphologically abnormal portion from within the specimen image where a morphological abnormality is presumed to have occurred, deriving judgment reference information from the detection result of the presumed morphologically abnormal portion, determining based on the judgment reference information whether the presumed morphologically abnormal portion is over-detected or under-detected, and, if it is determined that the presumed morphologically abnormal portion is over-detected, presenting to the user first cause identification reference information that helps to identify the cause of the presumed morphologically abnormal portion being over-detected, and, if it is determined that the presumed morphologically abnormal portion is under-detected, presenting to the user second cause identification reference information that helps to identify the cause of the presumed morphologically abnormal portion being under-detected.

[0031] According to the technology disclosed herein, it is possible to provide a drug discovery support device, an operating method for a drug discovery support device, and an operating program for a drug discovery support device that can contribute to improving the accuracy of detecting morphological abnormalities occurring in tissue specimens shown in specimen images.

[0032] 1 is a diagram illustrating the steps of an evaluation test, specimen images, and a drug discovery support device. It is a diagram illustrating an administration group and a control group. It is a block diagram illustrating a computer constituting the drug discovery support device. It is a block diagram illustrating a processing unit of a CPU of the drug discovery support device. It is a diagram illustrating an original image of a specimen image and an image for analysis. It is a diagram illustrating patch images obtained by dividing a specimen image. It is a diagram illustrating how features are extracted from patch images by a feature extractor. It is a diagram illustrating the structure of a feature extractor. It is a diagram illustrating processing in the learning phase of an autoencoder. It is a diagram illustrating the structure of a past control group and a reference patch image for learning. It is a diagram illustrating how reference features are extracted from reference patch images by a feature extractor. It is a diagram illustrating a graph in which reference features are plotted in feature space, and detection reference information. It is a diagram illustrating the distance between the position of a feature and the representative position of a reference feature. It is a diagram illustrating processing by a detection unit and detection results. It is a diagram illustrating processing by a detection unit and detection results. It is a diagram illustrating processing by a derivation unit and determination reference information. It is a diagram illustrating processing by a determination unit and a generation unit and determination results. It is a diagram illustrating processing by a determination unit and a generation unit and determination results. It is a diagram illustrating generation reference information. It is a flowchart illustrating the processing procedure of the generation unit. It is a diagram illustrating processing by the generation unit. 1 is a diagram showing the processing of the generation unit and cause identification reference information. FIG. 1 is a diagram showing the processing of the generation unit and cause identification reference information. FIG. 2 is a diagram showing the processing of the generation unit and cause identification reference information. FIG. 3 is a diagram showing the processing of the generation unit. FIG. 4 is a diagram showing cause identification reference information. FIG. 5 is a diagram showing a target selection screen. FIG. 6 is a diagram showing an analysis result display screen. FIG. 7 is a diagram showing an information display screen. FIG. 8 is a diagram showing an information display screen. FIG. 9 is a diagram showing another example of a message prompting the implementation of improvement processing. FIG. 10 is a diagram showing another example of a message prompting the implementation of improvement processing. FIG. 11 is a flowchart showing the processing procedure of the drug discovery support device. FIG. 12 is a block diagram showing a processing unit of a CPU of the drug discovery support device of the second embodiment. FIG. 13 is a diagram showing the processing of a CPU and determination reference information of the third embodiment. FIG. 14 is a diagram showing another example of the processing of a CPU and determination reference information of the third embodiment. FIG. 15 is a diagram showing yet another example of the processing of a CPU and determination reference information of the third embodiment. FIG. 16 is a diagram showing yet another example of the processing of a CPU and determination reference information of the third embodiment.FIG. 10 is a diagram showing the processing of the determination unit of the fourth embodiment. FIG. 11 is a diagram showing another example of the processing of the derivation unit and the determination unit and the determination reference information. FIG. 12 is a diagram showing another example of the processing of the derivation unit and the determination unit and the determination reference information. FIG. 13 is a diagram showing yet another example of the processing of the derivation unit and the determination unit and the determination reference information. FIG. 14 is a diagram showing a fifth embodiment that handles specimen images obtained by photographing a slide specimen on which tissue specimens of multiple types of organs are placed. FIG. 15 is a diagram showing how a feature extractor for cardiac specimens extracts features from patch images obtained by dividing a cardiac specimen.

[0033] First Embodiment As shown in FIG. 1 as an example, a drug discovery support device 10 of the present disclosure is used to evaluate the efficacy and toxicity of a drug candidate substance 27 (see FIG. 2). The drug discovery support device 10 is, for example, a desktop personal computer, and includes a display 11 that displays various screens, and an input device 12 such as a keyboard, a mouse, a touch panel, and / or a microphone for voice input. The drug discovery support device 10 is installed, for example, in a drug development facility, and is operated by a user U, such as a drug development staff member involved in drug development at the drug development facility. The drug discovery staff member may include a pathologist or the like.

[0034] A specimen image 15 is input to the drug discovery support device 10. The specimen image 15 is an image used to evaluate the efficacy and toxicity of a candidate substance 27. The specimen image 15 is generated, for example, by the following procedure. First, a subject S, such as a rat, prepared for evaluation of the candidate substance 27 is autopsied, and multiple tissue specimens (hereinafter referred to as liver specimens LVS) of the subject S's organs, in this case, cross-sections of the liver LV, are collected. Next, the collected liver specimens LVS are attached one by one to glass slides 16, and then stained, in this case with hematoxylin and eosin dye. Next, the stained liver specimens LVS are covered with a cover glass 17 to complete a slide specimen 18. The slide specimen 18 is then placed in an imaging device 19, such as a digital optical microscope, and a specimen image 15 is captured by the imaging device 19. The specimen image 15 thus obtained captures the entire liver specimen LVS. For this reason, the specimen image 15 is called a WSI (Whole Slide Image). The specimen image 15 is assigned a specimen ID (Identification Data) for uniquely identifying the specimen S, a specimen image ID for uniquely identifying the specimen image 15, and the date and time of capture. The tissue specimen is also called a tissue section. The staining may be performed using hematoxylin dye alone or nuclear fast red dye, for example.

[0035] As an example, as shown in FIG. 2 , subjects S are divided into a treatment group 25 and a control group 26. The treatment group 25 is composed of multiple subjects S administered with a candidate substance 27. The treatment group 25 is further divided into a high treatment group 25H, a medium treatment group 25M, and a low treatment group 25L according to the dose of the candidate substance 27. By dividing the treatment group 25 into the high treatment group 25H, the medium treatment group 25M, and the low treatment group 25L in this manner, the effect of the dose of the candidate substance 27 on the subjects S can be determined. The high treatment group 25H, the medium treatment group 25M, and the low treatment group 25L are examples of "sub-treatment groups" according to the technology of the present disclosure. Note that the sub-treatment groups are not limited to the three groups of the high treatment group 25H, the medium treatment group 25M, and the low treatment group 25L illustrated, but may be two groups of the high treatment group 25H and the low treatment group 25L, or may be four or more groups.

[0036] Contrary to the administration group 25, the control group 26 is composed of a plurality of subjects S who were not administered the candidate substance 27. The number of subjects S constituting each of the high-administration group 25H, the medium-administration group 25M, and the low-administration group 25L is the same as the number of subjects S constituting the control group 26, e.g., about 5 to 10. The subjects S constituting each of the high-administration group 25H, the medium-administration group 25M, and the low-administration group 25L and the subjects S constituting the control group 26 have the same attributes and are reared in the same environment. Examples of the same attributes include the same age in weeks, the same sex, and / or the same genetic lineage. Examples of the same genetic lineage include the same ancestor five generations ago and / or the same gene sequence in a specific region. Examples of the same attributes also include the same composition ratio of age in weeks, the same composition ratio of sex (e.g., five males and five females), and / or the same composition ratio of genetic lineage. The same rearing environment means, for example, that the animals are fed the same food, that the temperature and humidity of the rearing space are the same, and / or that the size of the rearing space is the same, etc. The "same" in the same rearing environment refers to the same in the sense that it is the same not only completely the same, but also includes an error that is generally acceptable in the technical field to which the technology of the present disclosure belongs and does not contradict the spirit of the technology of the present disclosure.

[0037] Since multiple specimen images 15 are obtained from one subject S, the number of specimen images 15 obtained from each group is calculated by multiplying the number of specimens S by the number of specimens S. For example, if the number of specimen images 15 obtained from one subject S is 100 and the number of subjects S constituting each group is 10, then 100 x 10 = 1000 specimen images 15 are obtained from each group.

[0038] 3, the computer constituting the drug discovery support apparatus 10 includes, in addition to the display 11 and input device 12, a storage 30, a memory 31, a CPU (Central Processing Unit) 32, and a communication unit 33. These are interconnected via a bus line 34.

[0039] The storage 30 is a hard disk drive built into the computer constituting the drug discovery support apparatus 10 or connected via a cable or network. Alternatively, the storage 30 is a disk array with multiple hard disk drives connected in series. The storage 30 stores control programs such as an operating system, various application programs, and various data associated with these programs. Note that a solid state drive may be used instead of a hard disk drive.

[0040] The memory 31 is a work memory for the CPU 32 to execute processing. The CPU 32 loads programs stored in the storage 30 into the memory 31 and executes processing in accordance with the programs. In this way, the CPU 32 comprehensively controls each part of the computer. The CPU 32 is an example of a "processor" according to the technology of the present disclosure. Note that the memory 31 may be built into the CPU 32. The communication unit 33 controls the transmission of various information to and from external devices such as the imaging device 19.

[0041] As an example, as shown in FIG. 4 , an operating program 40 is stored in the storage 30 of the drug discovery support apparatus 10. The operating program 40 is an application program for causing a computer to function as the drug discovery support apparatus 10. In other words, the operating program 40 is an example of an "operating program for a drug discovery support apparatus" according to the technology of the present disclosure. The storage 30 also stores a feature extractor 41, detection reference information 42, a judgment threshold 43, generation reference information 44, and the like. The feature extractor 41 is an example of a "machine learning model" according to the technology of the present disclosure.

[0042] When the operating program 40 is started, the CPU 32 of the computer constituting the drug discovery support device 10 works in cooperation with the memory 31 and the like to function as a read / write (hereinafter abbreviated as RW (Read Write)) control unit 50, a detection unit 51, a derivation unit 52, a judgment unit 53, a generation unit 54, and a display control unit 55.

[0043] The RW control unit 50 controls the storage of various data in the storage 30 and the reading of various data from the storage 30. For example, the RW control unit 50 stores a specimen image 15 from the imaging device 19 in the storage 30. At this time, as shown in FIG. 5 as an example, the RW control unit 50 performs resolution conversion processing on the original specimen image 15 from the imaging device 19 (hereinafter referred to as an original image 15O) to generate a specimen image 15 for analysis (hereinafter referred to as an analysis image 15A). The original image 15O has a resolution equivalent to a 40x magnification, and the analysis image 15A has a resolution equivalent to a 20x magnification. In other words, the analysis image 15A has half the resolution of the original image 15O. Therefore, the analysis image 15A imposes a smaller load on the analysis process than the original image 15O. The RW control unit 50 associates the original image 15O and the analysis image 15A and stores them in the storage 30.

[0044] The RW control unit 50 acquires an analysis image 15A corresponding to a specification made by the user U via the input device 12 by reading it from the storage 30. The RW control unit 50 outputs the read analysis image 15A to the detection unit 51, the generation unit 54, and the display control unit 55. The analysis image 15A output from the RW control unit 50 to the detection unit 51, etc., is a target for detecting whether or not a morphological abnormality has occurred in the liver specimen LVS. Hereinafter, the analysis image 15A for detecting whether or not a morphological abnormality has occurred in the liver specimen LVS is referred to as a target specimen image 15T (see FIG. 6, etc.). Note that morphological abnormalities include lesions not observed in a normal liver specimen LVS, such as hyperplasia, infiltration, congestion, inflammation, tumor, canceration, proliferation, bleeding, and glycogen depletion.

[0045] The RW control unit 50 reads the feature extractor 41 and the detection reference information 42 from the storage 30, and outputs the read feature extractor 41 and the read detection reference information 42 to the detection unit 51. The RW control unit 50 also reads the determination threshold 43 from the storage 30, and outputs the read determination threshold 43 to the determination unit 53. The RW control unit 50 also reads the generation reference information 44 from the storage 30, and outputs the read generation reference information 44 to the generation unit 54.

[0046] The detection unit 51 detects an estimated morphologically abnormal portion, where a morphological abnormality is estimated to have occurred, from the portion of the target specimen image 15T using the feature extractor 41 and the detection reference information 42. The detection unit 51 outputs a detection result 60 of the estimated morphologically abnormal portion to the derivation unit 52.

[0047] The derivation unit 52 derives determination reference information 61 from the detection result 60. The derivation unit 52 outputs the determination reference information 61 to the determination unit 53. Although not shown in the figure, the derivation unit 52 also outputs the determination reference information 61 to the display control unit 55 as an analysis result.

[0048] The determination unit 53 determines whether the estimated morphological abnormality portions are over-detected or under-detected based on the determination threshold 43 and the determination reference information 61. The determination unit 53 outputs a determination result 62 indicating whether the estimated morphological abnormality portions are over-detected or under-detected to the generation unit 54. Note that "the estimated morphological abnormality portions are over-detected" means that more estimated morphological abnormality portions are detected than the number of true morphological abnormality portions. Also, "the estimated morphological abnormality portions are under-detected" means that fewer estimated morphological abnormality portions are detected than the number of true morphological abnormality portions.

[0049] The generating unit 54 generates cause identification reference information 63 based on the target specimen image 15T, the generated reference information 44, and the determination result 62. The generating unit 54 outputs the cause identification reference information 63 to the display control unit 55.

[0050] The display control unit 55 controls the display of various screens on the display 11. The various screens include a target designation screen 125 (see FIG. 27) for designating a target specimen image 15T, an analysis result display screen 135 (see FIG. 28), and an information display screen 145 (see FIGS. 29 to 31) for displaying the cause identification reference information 63. In addition to these processing units 50 to 55, the CPU 32 also includes an instruction receiving unit that receives various operation instructions from the input device 12.

[0051] As an example, as shown in FIG. 6 , the detection unit 51 recognizes the liver specimen LVS depicted in the target specimen image 15T using well-known image recognition technology and subdivides the recognized liver specimen LVS into multiple patch images 70. The patch images 70 have a predetermined size that can be handled by the feature extractor 41. The patch images 70 are sized to cover not only the morphologically abnormal portion but also the surrounding area. The detection unit 51 assigns a patch image ID 85 (see FIG. 14 , etc.) to the patch image 70. The detection unit 51 also associates information indicating which portion of the target specimen image 15T the patch image 70 is cut out from, i.e., position information 86 of the patch image 70 (see FIG. 14 , etc.), with the patch image ID 85. The patch image 70 is an example of a "portion of a specimen image" according to the technology of the present disclosure. Note that, although adjacent patch images 70 do not have overlapping areas in FIG. 6 , adjacent patch images 70 may partially overlap.

[0052] 7 , the detection unit 51 uses the feature extractor 41 to extract a feature 72 for each of a plurality of patch images 70 obtained by dividing the target specimen image 15T. Therefore, the number of feature amounts 72 is the same as the number of patch images 70.

[0053] As an example, as shown in FIG. 8 , the feature extractor 41 uses an encoder unit 76 of an autoencoder 75. The autoencoder 75 has a decoder unit 77 in addition to the encoder unit 76. A patch image 70 is input to the encoder unit 76. The encoder unit 76 converts the patch image 70 into a feature 72. The encoder unit 76 passes the feature 72 to the decoder unit 77. The decoder unit 77 generates a restored image 78 of the patch image 70 from the feature 72.

[0054] As is well known, the encoder unit 76 includes a convolution layer that performs convolution processing using a filter, a pooling layer that performs pooling processing such as maximum value pooling, and the like. The decoder unit 77 also includes a convolution layer that performs convolution processing using a filter, a pooling layer that performs pooling processing such as maximum value pooling, and the like. The encoder unit 76 extracts feature quantities 72 by repeating the convolution processing by the convolution layer and the pooling processing by the pooling layer multiple times on the input patch image 70. The extracted feature quantities 72 represent the shape and texture characteristics of the liver specimen LVS depicted in the patch image 70.

[0055] The feature 72 is a set of multiple numerical values. In other words, the feature 72 is multidimensional data. The number of dimensions of the feature 72 is, for example, 512, 1024, or 2048. The feature 72 and a reference feature 72R (see FIG. 11 ), which will be described later, have the same number of dimensions and can be compared in the same feature space 81 (see FIG. 12 , etc.).

[0056] As an example, as shown in FIG. 9 , in a learning phase before the encoder unit 76 is converted into the feature extractor 41, the autoencoder 75 receives a training reference patch image 70RL as input and is trained. The autoencoder 75 outputs a training reconstructed image 78L in response to the training reference patch image 70RL. A loss calculation is performed on the autoencoder 75 using a loss function based on the training reference patch image 70RL and the training reconstructed image 78L. Then, update settings are made for various coefficients of the autoencoder 75 (such as the filter coefficients of the convolution layer) according to the results of the loss calculation, and the autoencoder 75 is updated according to the update settings.

[0057] During the learning phase of the autoencoder 75, the above-described series of processes, including input of the learning reference patch image 70RL to the autoencoder 75, output of the learning reconstructed image 78L from the autoencoder 75, loss calculation, update setting, and updating of the autoencoder 75, are repeatedly performed while the learning reference patch image 70RL is replaced. The repetition of the above-described series of processes is terminated when the accuracy of restoration from the learning reference patch image 70RL to the learning reconstructed image 78L reaches a predetermined set level. The encoder unit 76 of the autoencoder 75 whose restoration accuracy has reached the set level is stored in the storage 30 of the drug discovery support apparatus 10 as the feature extractor 41. Note that learning may be terminated when the above-described series of processes has been repeated a set number of times, regardless of the accuracy of restoration from the learning reference patch image 70RL to the learning reconstructed image 78L.

[0058] The learning of the autoencoder 75 may be performed by the drug discovery support device 10, or may be performed by a device separate from the drug discovery support device 10. In the latter case, the feature extractor 41 is transmitted from the separate device to the drug discovery support device 10, and the feature extractor 41 is stored in the storage 30 by the RW control unit 50.

[0059] As an example, as shown in FIG. 10 , the learning reference patch image 70RL is supplied from multiple reference patch images 70R obtained by subdividing the reference specimen image 15R. The reference specimen image 15R is an image of a liver specimen LVS of a subject S in a past control group 26P. The past control group 26P is composed of multiple subjects S to whom a candidate substance was not administered in past evaluation tests. Therefore, the number of subjects S constituting the past control group 26P is significantly greater than the number of subjects S constituting the administration group 25 and the control group 26, for example, on the order of several hundred to several thousand. Like the specimen image 15, multiple reference specimen images 15R are obtained from one subject S. Therefore, the number of reference specimen images 15R obtained from the past control group 26P is calculated by multiplying the number obtained from one subject S by the number of subjects S. The liver specimen LVS of the subject S in the past control group 26P is an example of a "tissue specimen considered normal" according to the technology of the present disclosure. In addition to specimen image 15 of the liver specimen LVS of subject S in the past control group 26P, specimen image 15 of the liver specimen LVS that has been determined to be normal by an expert such as a pathologist in a past administration group consisting of multiple subjects S to whom a candidate substance was administered in a past evaluation test may also be used as reference specimen image 15R.

[0060] Next, we will explain how the detection reference information 42 is formed. First, as shown in Fig. 11 as an example, a feature extractor 41 is used to extract multiple reference features 72R from multiple reference patch images 70R based on all of the multiple reference specimen images 15R.

[0061] As an example, a graph 80 shown in FIG. 12 is a plot of multiple reference features 72R extracted in FIG. 11 in a feature space 81. The detection reference information 42 includes coordinates 82 in the feature space 81 of representative positions of the reference features 72R, indicated by crosses (hereinafter referred to as representative position coordinates). The representative position is, for example, the center point or average point of a distribution 83 of the reference features 72R. The detection reference information 42 also includes a detection threshold 84. Note that in FIG. 12, for convenience of explanation, the dimension of the feature space 81 is shown as two-dimensional, having axes D1 and D2. However, the actual dimension of the feature space 81 is, as described above, 512 or the like. Similarly, in subsequent figures such as FIG. 13, for convenience of explanation, the dimension of the feature space 81 is shown as two-dimensional.

[0062] Similar to the learning of the autoencoder 75, the representative position coordinates 82 of the detection reference information 42 may be derived by the drug discovery support apparatus 10, or may be derived by a device other than the drug discovery support apparatus 10. In the latter case, the representative position coordinates 82 are transmitted from the other device to the drug discovery support apparatus 10, and the RW control unit 50 stores the representative position coordinates 82 in the storage 30.

[0063] As an example, as shown in FIG. 13 , the detection unit 51 calculates the distance D in the feature space 81 between the representative position of the reference feature 72R represented by the representative position coordinates 82 of the detection reference information 42 and the position of the feature 72. The detection unit 51 calculates the distance D between the multiple feature values ​​72 extracted for each of the multiple patch images 70 obtained by subdividing a single target specimen image 15T. The distance D is the Mahalanobis distance. The distance D indicates the degree of deviation of the feature value 72 from the reference feature 72R, or more specifically, the degree of deviation of the liver specimen LVS depicted in the patch image 70 of the target specimen image 15T from a liver specimen LVS deemed normal. In other words, the greater the distance D, the more the liver specimen LVS depicted in the patch image 70 deviates from a liver specimen LVS deemed normal. Therefore, the greater the distance D, the higher the probability that a morphological abnormality has occurred in the liver specimen LVS depicted in the patch image 70.

[0064] Note that the distance D may be calculated as the average, median, or maximum value of the Euclidean distance between the position of the k-neighbor sample in the distribution 83 of the reference feature 72R and the position of the feature 72. Alternatively, instead of the distance D, a value obtained by subtracting from 1.0 the cosine similarity between a vector representing the representative position of the reference feature 72R and a vector representing the position of the feature 72 may be calculated. The cosine similarity takes a value between -1.0 and 1.0, and the larger the value, the more similar the orientations of the vectors are. Furthermore, instead of the distance D, a likelihood function such as negative logarithmic likelihood may be calculated as the deviation.

[0065] 14 and 15 , the detection unit 51 compares the calculated distance D with a detection threshold 84. As shown in FIG. 14 , if the distance D is less than the detection threshold 84, it is detected that no morphological abnormality has occurred in the liver specimen LVS captured in the patch image 70. The detection unit 51 outputs a detection result 60 indicating that no morphological abnormality has occurred in the liver specimen LVS captured in the patch image 70. The detection result 60 in this case includes a patch image ID 85 and position information 86.

[0066] On the other hand, as shown in FIG. 15 , if the distance D is equal to or greater than the detection threshold 84, the detection unit 51 detects that the liver specimen LVS captured in the patch image 70 has a morphological abnormality. The detection unit 51 outputs a detection result 60 indicating that the liver specimen LVS captured in the patch image 70 has a morphological abnormality. In this case, the detection result 60 includes a feature 72 in addition to a patch image ID 85 and position information 86. The portion of the patch image 70 detected as having a morphological abnormality corresponds to the "estimated morphological abnormality portion" according to the technology of the present disclosure. Note that the detection threshold 84 may be common to the high dose group 25H, the medium dose group 25M, the low dose group 25L, and the control group 26, or may be different among these groups. Alternatively, instead of the distance D, the presence or absence of a morphological abnormality in the liver specimen LVS captured in the patch image 70 may be detected by comparing the aforementioned cosine similarity or likelihood function with the detection threshold 84.

[0067] As shown in Figures 10 and 11, the reference feature 72R is a feature extracted from a reference patch image 70R obtained by subdividing a reference specimen image 15R depicting a liver specimen LVS of a subject S in the historical control group 26P. Because the subject S in the historical control group 26P is a subject S who was not administered a candidate substance, the liver specimen LVS depicted in the reference specimen image 15R does not exhibit morphological abnormalities at least due to the toxicity of the candidate substance. Therefore, the representative position of the reference feature 72R is regarded as the representative position of the feature of the specimen image 15 depicting a normal liver specimen LVS. Therefore, as described above, the distance D between the representative position of the reference feature 72R and the position of the feature 72 serves as an index indicating the degree to which the liver specimen LVS depicted in the patch image 70 differs from the normal liver specimen LVS. 14, for a patch image 70 in which the distance D is less than the detection threshold 84, the detection unit 51 determines that the liver specimen LVS depicted in the patch image 70 does not deviate from the normal liver specimen LVS and therefore determines that no morphological abnormality has occurred. On the other hand, for a patch image 70 in which the distance D is equal to or greater than the detection threshold 84, the detection unit 51 determines that the liver specimen LVS depicted in the patch image 70 deviates from the normal liver specimen LVS and therefore determines that a morphological abnormality has occurred.

[0068] As an example, as shown in FIG. 16 , the detection results 60 of all patch images 70 are input to the derivation unit 52. The derivation unit 52 counts the number of detection results 60 indicating that morphological abnormalities have occurred in the liver specimen LVS captured in the patch images 70, i.e., the number of estimated morphological abnormality portions. Then, the derivation unit 52 calculates the number ratio of patch images 70 detected as having morphological abnormalities, i.e., estimated morphological abnormality portions, by dividing the counted number by the total number of patch images 70. The derivation unit 52 outputs the calculated number ratio as determination reference information 61. The number ratio is an example of a "numeric value related to the number of estimated morphological abnormality portions" according to the technology of the present disclosure. Note that instead of the number ratio, the number of estimated morphological abnormality portions themselves may be output as determination reference information 61.

[0069] 17 and 18 , the determination threshold 43 includes a first determination threshold 431 and a second determination threshold 432. The determination unit 53 compares the ratio of the number of determination reference information 61 with the first determination threshold 431 and the second determination threshold 432.

[0070] 17 , when the count ratio of the determination reference information 61 is equal to or greater than the first determination threshold 431, the determination unit 53 determines that the patch image 70 detected as having a morphological abnormality, i.e., that the estimated morphological abnormality portion is detected as being oversized, and outputs a determination result 62 to that effect. In this case, the generation unit 54 generates, as the cause identification reference information 63, first cause identification reference information 631 that contributes to identifying the cause why the estimated morphological abnormality portion is detected as being oversized.

[0071] 18 , when the count ratio of the determination reference information 61 is equal to or less than the second determination threshold 432, the determination unit 53 determines that the patch image 70 detected as having a morphological abnormality, i.e., the estimated morphological abnormality portion, is underdetected, and outputs a determination result 62 to that effect. In this case, the generation unit 54 generates, as the cause identification reference information 63, second cause identification reference information 632 that contributes to identifying the cause why the estimated morphological abnormality portion is underdetected.

[0072] Although not shown in the figure, if the number ratio of the determination reference information 61 is neither greater than or equal to the first determination threshold 431 nor less than the second determination threshold 432, i.e., if the number ratio of the determination reference information 61 is greater than the second determination threshold 432 and less than the first determination threshold 431, the determination unit 53 outputs a patch image 70 detected as having a morphological abnormality, i.e., a determination result 62 indicating that the number of detected estimated morphological abnormality portions is appropriate. In this case, the generation unit 54 does not generate the cause identification reference information 63. Note that the first determination threshold 431 and the second determination threshold 432 may be common to the high administration group 25H, the medium administration group 25M, the low administration group 25L, and the control group 26, or may be different among these groups.

[0073] 19 , the generated reference information 44 includes an artifact feature set 88 and a representative reference specimen image 15RR. The artifact feature set 88 is a collection of feature sets 72A of multiple types of artifacts that may be erroneously detected as estimated morphologically abnormal portions (hereinafter referred to as artifact feature sets). The representative reference specimen image 15RR is an image that represents, among the multiple reference specimen images 15R, the color tone, particularly the degree of staining with hematoxylin-eosin dye.

[0074] Artifacts that may be erroneously detected as suspected morphological abnormalities occur when preparing the slide specimen 18 and when capturing the specimen image 15 with the imaging device 19. The artifacts include a first artifact caused by uneven thickness of the specimen (here, a liver specimen LVS), a second artifact caused by scalpel scratches made when the specimen was extracted, and a third artifact caused by sebum adhering to the specimen. Other artifacts include a fourth artifact caused by air getting into the cover glass 17, a fifth artifact caused by out-of-focus images captured with the imaging device 19, and a sixth artifact caused by foreign matter such as dust.

[0075] The artifact feature 72A includes a first artifact feature 72A1 that is the feature 72 of the first artifact, a second artifact feature 72A2 that is the feature 72 of the second artifact, and a third artifact feature 72A3 that is the feature 72 of the third artifact, etc. The artifact feature 72A also includes a fourth artifact feature 72A4 that is the feature 72 of the fourth artifact, a fifth artifact feature 72A5 that is the feature 72 of the fifth artifact, and a sixth artifact feature 72A6 that is the feature 72 of the sixth artifact, etc.

[0076] 20 to 22, the generation unit 54 generates the cause identification reference information 63. First, the generation unit 54 calculates the distance D in the feature space 81 between the position of the feature 72 of the patch image 70 in which a morphological abnormality has been detected and the position of each artifact feature 72A (step ST10).

[0077] The generation unit 54 compares the shortest distance SD among the calculated distances D with a preset distance threshold (step ST11). If the shortest distance SD is less than the distance threshold (YES in step ST11), the generation unit 54 identifies the artifact with the artifact feature 72A of the shortest distance SD as a type of artifact that may have been erroneously detected as an estimated morphological abnormality portion (step ST12). On the other hand, if the shortest distance SD is equal to or greater than the distance threshold (NO in step ST11), the generation unit 54 proceeds to step ST13.

[0078] The generation unit 54 continues the series of processes in steps ST10 to ST12 until they have been performed on all patch images 70 detected as having morphological abnormalities (NO in step ST13). When the series of processes in steps ST10 to ST12 have been performed on all patch images 70 detected as having morphological abnormalities (YES in step ST13), the generation unit 54 outputs the types of artifacts whose identified numbers are equal to or greater than a predetermined identification threshold 90 (see FIG. 22 ) and the identified numbers thereof as cause identification reference information 63 (first cause identification reference information 631 and second cause identification reference information 632) (step ST14). The types of artifacts and the identified numbers thereof are an example of "information related to type" according to the technology of the present disclosure.

[0079] 21 illustrates an example in which the distance D between the position of the feature 72 in the patch image 70 detected as having a morphological abnormality and the position of the second artifact feature 72A2 is the shortest distance SD, and the shortest distance SD is less than the distance threshold. The second artifact feature 72A2 is a feature of the second artifact caused by the scalpel scar. Therefore, in this case, the generation unit 54 identifies the scalpel scar as the type of artifact that may have been erroneously detected as an estimated morphological abnormality.

[0080] 22 illustrates an example in which the number of identified specimen thickness non-uniformities is 20, the number of identified scalpel scratches is 722, the number of identified sebum deposits and air intrusions is 0, ..., and the identification threshold 90 is 500, as shown in Table 91. In this case, the generation unit 54 outputs cause identification reference information 63 including scalpel scratches and the number of identified scalpel scratches 722. The identification threshold 90 may be common to the high dose group 25H, the medium dose group 25M, the low dose group 25L, and the control group 26, or may be different for these groups.

[0081] As an example, as shown in FIG. 23 , the generation unit 54 generates a histogram 95 of pixel values ​​of pixels of each color—blue (B), green (G), and red (R)—from the target specimen image 15T. The generation unit 54 also generates a histogram 96 of pixel values ​​of pixels of each color—blue (B), green (G), and red (R)—from the representative reference specimen image 15RR. Furthermore, as shown in Table 97, the generation unit 54 calculates the Bhattacharyya distance between the B distributions in the target histogram 95 and the reference histogram 96, the Bhattacharyya distance between the G distributions, and the Bhattacharyya distance between the R distributions. The more similar the distributions are, the closer the Bhattacharyya distance is to 0. The generation unit 54 outputs the target histogram 95, the reference histogram 96, and the Bhattacharyya distance as cause identification reference information 63 (first cause identification reference information 631 and second cause identification reference information 632). The target histogram 95 is an example of "information about the color of the specimen image" according to the technology of the present disclosure. The reference histogram 96 is an example of "information about the color of the reference specimen image" according to the technology of the present disclosure. The Bhattacharyya distance is an example of "information about the color of the specimen image" and "information about the color of the reference specimen image" according to the technology of the present disclosure. Note that the reference histogram 96 may be stored as the generation reference information 44 instead of the representative reference specimen image 15RR.

[0082] 24 as an example, the generation unit 54 extracts a hematoxylin component 102T and an eosin component 103T of the target specimen image 15T from the distribution of pixel values ​​of the pixels of the target specimen image 15T in an RGB space 101 shown in a graph 100. Similarly, the generation unit 54 extracts a hematoxylin component 102RR and an eosin component 103RR of the representative reference specimen image 15RR.

[0083] The hematoxylin component 102T and eosin component 103T of the target specimen image 15T, and the hematoxylin component 102RR and eosin component 103RR of the representative reference specimen image 15RR, make it possible to visualize the difference between the staining conditions for the liver specimen LVS depicted in the target specimen image 15T and the staining conditions for the liver specimen LVS depicted in the representative reference specimen image 15RR. The generation unit 54 outputs the hematoxylin component 102T and eosin component 103T of the target specimen image 15T, and the hematoxylin component 102RR and eosin component 103RR of the representative reference specimen image 15RR as cause identification reference information 63 (first cause identification reference information 631 and second cause identification reference information 632).

[0084] The hematoxylin component 102T and the eosin component 103T of the target specimen image 15T are an example of "information about the color of the specimen image" according to the technology of the present disclosure. The hematoxylin component 102RR and the eosin component 103RR of the representative reference specimen image 15RR are an example of "information about the color of the reference specimen image" according to the technology of the present disclosure. Note that the hematoxylin component 102RR and the eosin component 103RR of the representative reference specimen image 15RR may be stored as the generation reference information 44, instead of the representative reference specimen image 15RR.

[0085] As an example, as shown in Fig. 25 , when the determining unit 53 determines that the estimated morphological abnormality portion is detected as being excessively large, the generating unit 54 performs a clustering process to define the cluster to which each feature 72 of the patch image 70 detected as having a morphological abnormality belongs. Fig. 25 shows an example in which the feature 72 is clustered into three clusters, cluster 1, cluster 2, and cluster 3. Note that, as shown in the example, some feature 72 do not belong to any cluster.

[0086] The generation unit 54 generates the clustering information 110. The clustering information 110 is information in which the cluster to which each patch image 70 belongs is registered for each patch image ID 85 of the patch image 70. A patch image 70 whose feature amount 72 does not belong to any cluster is not registered in the clustering information 110.

[0087] 26 , the generation unit 54 processes the target specimen image 15T in accordance with the clustering information 110 to generate cluster images 115, 116, and 117 as the first cause identification reference information 631. The cluster image 115 is an image corresponding to cluster 1. The cluster image 116 is an image corresponding to cluster 2. The cluster image 117 is an image corresponding to cluster 3.

[0088] The generation unit 54 generates cluster images 115 to 117 according to a display format 118 preset for each cluster. The display format 118, for example, displays cluster 1 in indigo, cluster 2 in yellow-green, and cluster 3 in gray. The generation unit 54 generates cluster image 115 by filling in the position (determined from the position information 86) of the patch image 70 of patch image ID 85, for which cluster 1 is registered in the clustering information 110, with indigo, in the target specimen image 15T. Similarly, the generation unit 54 generates cluster image 116 by filling in the position of the patch image 70 of patch image ID 85, for which cluster 2 is registered in the clustering information 110, with yellow-green, in the target specimen image 15T. Furthermore, the generation unit 54 generates cluster image 117 by filling in the position of the patch image 70 of patch image ID 85, for which cluster 3 is registered in the clustering information 110, with gray, in the target specimen image 15T. By changing the display color in this way, the cluster images 115 to 117 become images that enable identification of clusters 1 to 3. Note that the display format 118 may be configured so that the user U can freely change the settings.

[0089] The generation unit 54 generates a superimposed image 119 by superimposing the target specimen image 15T and at least one of the cluster images 115 to 117. Fig. 26 illustrates an example of the superimposed image 119 in which all of the cluster images 115 to 117 are superimposed on the target specimen image 15T.

[0090] As an example, as shown in FIG. 27 , the display control unit 55 controls the display of a target designation screen 125 on the display 11. When a user U issues an instruction to display a specimen image 15 via the input device 12, the display control unit 55 displays the target designation screen 125. A plurality of specimen images 15 are displayed side by side on the target designation screen 125. The plurality of specimen images 15 are specimen images 15 obtained from one of a plurality of subjects S constituting one of the groups selected in a pull-down menu 126 from the high-dose group 25H, the medium-dose group 25M, the low-dose group 25L, and the control group 26. FIG. 27 shows an example in which ten specimen images 15 with specimen image IDs "SI00001" to "SI00010" obtained from the subject S with the subject ID "R001" are displayed side by side. The subject ID is displayed as a pull-down menu 127, making it possible to switch the subject S whose specimen image 15 is displayed on the target designation screen 125.

[0091] The target designation screen 125 is a screen for designating one target specimen image 15T from among a plurality of specimen images 15. The target designation screen 125 has a selection frame 128 that can be moved between the specimen images 15. An analysis button 129 is also provided at the bottom of the target designation screen 125. The user U aligns the selection frame 128 with the desired specimen image 15 and then selects the analysis button 129. As a result, the specimen image 15 with the selection frame 128 aligned is set as the target specimen image 15T, and the detection unit 51 detects the estimated morphological abnormality portion, the derivation unit 52 derives the determination reference information 61, the determination unit 53 determines whether the estimated morphological abnormality portion has been detected as being oversized or undersized, and the generation unit 54 generates the cause identification reference information 63.

[0092] When the processing of each of the above processing units is completed, the display control unit 55 controls the display 11 to display an analysis result display screen 135, as shown in FIG. 28 , for example. The analysis result display screen 135 displays a target specimen image 15T. A first display area 136 and a second display area 137 are provided below the target specimen image 15T. The first display area 136 displays the ratio of the number of estimated morphological abnormality portions included in the determination reference information 61 from the derivation unit 52 as the analysis result. The second display area 137 displays a message informing the determination unit 53 that the estimated morphological abnormality portions are over- or under-detected if the determination unit 53 determines that the estimated morphological abnormality portions are over- or under-detected. FIG. 28 illustrates an example of a case in which the determination unit 53 determines that the estimated morphological abnormality portions are over-detected. Although not shown, if the determination unit 53 does not determine that the estimated morphological abnormality portions are over- or under-detected, the second display area 137 displays a message indicating that the number of estimated morphological abnormality portions detected is appropriate.

[0093] The second display area 137 is provided with an information display button 138. The information display button 138 is a button for displaying the cause identification reference information 63. When the confirmation button 139 is selected, the display control unit 55 erases the display of the analysis result display screen 135.

[0094] Here, in the target specimen image 15T on the analysis result display screen 135, the portion of the patch image 70 where a morphological abnormality has been detected may be displayed in a distinguishable manner by being filled in red, for example. In this case, the greater the difference between the distance D and the detection threshold 84, the darker the displayed color may be, or the color to be filled in may be changed depending on the type of artifact identified by the generation unit 54.

[0095] When the information display button 138 is selected, the display control unit 55 performs control to display an information display screen 145, as shown in Figures 29 to 31, on the display 11. The information display screen 145 is provided with a first display area 146, a second display area 147, a third display area 148, and a fourth display area 149. These first to fourth display areas 146 to 149 can be scrolled.

[0096] The first display area 146 displays the types and numbers of artifacts identified as possibly being erroneously detected as suspected morphological abnormalities, while the second display area 147 displays the target histogram 95, the reference histogram 96, and the Bhattacharyya distance table 97.

[0097] In addition, the second display area 147 displays a graph 100 including the hematoxylin component 102T and eosin component 103T of the target specimen image 15T and the hematoxylin component 102RR and eosin component 103RR of the representative reference specimen image 15RR, and a legend 150.

[0098] The third display area 148 displays a superimposed image 119 and a legend 151. Display switch buttons 152, 153, and 154 are provided below the legend 151. The display switch button 152 is a button for selecting whether or not to superimpose the cluster image 115 on the target specimen image 15T. The display switch button 153 is a button for selecting whether or not to superimpose the cluster image 116 on the target specimen image 15T. The display switch button 154 is a button for selecting whether or not to superimpose the cluster image 117 on the target specimen image 15T. Therefore, for example, when all of the display switch buttons 152 to 154 are selected as shown in the figure, the display control unit 55 displays a superimposed image 119 in which all of the cluster images 115 to 117 are superimposed on the target specimen image 15T. In this way, the display control unit 55 superimposes at least one of the multiple cluster images 115 to 117 on the target specimen image 15T. The information display screen 145 is initially displayed with all of the display switching buttons 152 to 154 selected.

[0099] By displaying the first cause identification reference information 631 or the second cause identification reference information 632 in the first display area 146 to the third display area 148 in this manner, the display control unit 55 presents the first cause identification reference information 631 or the second cause identification reference information 632 to the user U.

[0100] A message 155 is displayed in the fourth display area 149 to prompt the user U to perform improvement processing. The improvement processing includes suppression processing and exclusion processing. The suppression processing is processing to suppress over-detection or under-detection of estimated morphological abnormality parts. The exclusion processing is processing to exclude artifact parts that may have been erroneously detected as estimated morphological abnormality parts from the target of the evaluation test. Exclusion from the target of the evaluation test means that patch images 70 including artifacts that may have been erroneously detected as estimated morphological abnormality parts are excluded from the target of the calculation of the number ratio.

[0101] If an artifact type (here, a scalpel scratch) is identified as possibly being erroneously detected as a presumed morphological abnormality, a message 155A is displayed in the fourth display area 149, prompting the user U to perform an exclusion process. Furthermore, if the color difference between the target specimen image 15T and the representative reference specimen image 15RR is greater than or equal to a preset color threshold, a message 155B is displayed in the fourth display area 149, prompting the user U to perform a color correction process on the target specimen image 15T as a suppression process. Here, the color threshold is set for the Bhattacharyya distance (e.g., 0.5). Alternatively, the color threshold is set for the distance in the RGB space 101 between the hematoxylin component 102T and eosin component 103T of the target specimen image 15T and the hematoxylin component 102RR and eosin component 103RR of the representative reference specimen image 15RR. When the confirmation button 156 is selected, the display control unit 55 erases the display on the information display screen 145.

[0102] As an example, as shown in Figures 32 and 33, there are various variations of the message 155 that prompts the user U to perform the improvement process. The message 155C shown in Figure 32 is displayed when the determination unit 53 determines that the estimated morphological abnormality portion has been over-detected. Furthermore, the message 155C is displayed when there is no type of artifact identified as possibly having been erroneously detected as the estimated morphological abnormality portion, and the difference in color between the target specimen image 15T and the representative reference specimen image 15RR does not exceed the color threshold. The message 155C prompts the user U to perform a process to change the detection threshold 84 used to detect the estimated morphological abnormality portion. More specifically, the message 155C prompts the user U to set the detection threshold 84 more strictly. Setting the detection threshold 84 more strictly means resetting the detection threshold 84 to a higher value.

[0103] Messages 155D and 155E shown in FIG. 33 are displayed when the determination unit 53 determines that the estimated morphological abnormality portion has been underdetected. Messages 155D and 155E are also displayed when there is no artifact type identified as possibly having been erroneously detected as the estimated morphological abnormality portion, and the color difference between the target specimen image 15T and the representative reference specimen image 15RR does not exceed the color threshold. Like message 155C, message 155D urges user U to change the detection threshold 84 used to detect the estimated morphological abnormality portion. However, unlike message 155C, message 155D urges user U to loosen the setting of the detection threshold 84. Loosening the setting of the detection threshold 84 means resetting the detection threshold 84 to a lower value.

[0104] The message 155E urges the user U to change the resolution of the target specimen image 15T. More specifically, the message 155E urges the user U to increase the resolution of the target specimen image 15T. Increasing the resolution of the target specimen image 15T means using the original image 15O instead of the analysis image 15A as the target specimen image 15T.

[0105] By displaying the message 155 in the fourth display area 149 in this way, the display control unit 55 is suggesting to the user U the suppression process and / or the exclusion process.

[0106] Next, the operation of the above configuration will be described with reference to the flowchart shown in Fig. 34. First, when the operating program 40 is started in the drug discovery support apparatus 10, the CPU 32 of the drug discovery support apparatus 10 functions as an RW control unit 50, a detection unit 51, a derivation unit 52, a determination unit 53, a generation unit 54, and a display control unit 55, as shown in Fig. 4.

[0107] The imaging device 19 captures a specimen image 15 of a liver specimen LVS of the subject S. The specimen image 15 is output from the imaging device 19 to the drug discovery support device 10. In the drug discovery support device 10, the specimen image 15 from the imaging device 19 is stored in the storage 30 by the RW control unit 50. At this time, as shown in FIG. 5 , the RW control unit 50 performs resolution conversion processing on the original specimen image 15 from the imaging device 19, i.e., the original image 15O, to generate an analysis image 15A. Then, the original image 15O and the analysis image 15A are associated with each other and stored in the storage 30.

[0108] When the user U issues an instruction to display the specimen image 15 via the input device 12, the RW control unit 50 reads and acquires the specimen image 15 specified in the display instruction from the storage 30 (step ST100). The specimen image 15 is output from the RW control unit 50 to the display control unit 55. As shown in Fig. 27 , the specimen image 15 is displayed on the display 11 via the target designation screen 125 under the control of the display control unit 55 (step ST105).

[0109] When the user U aligns the selection frame 128 with the desired specimen image 15 on the target designation screen 125 and selects the analysis button 129 (YES in step ST110), the RW control unit 50 reads out and acquires the analysis image 15A of the specimen image 15 on which the selection frame 128 was aligned at that time from the storage 30 as a target specimen image 15T (step ST115). The target specimen image 15T is output from the RW control unit 50 to the detection unit 51, the generation unit 54, and the display control unit 55.

[0110] The RW control unit 50 reads the feature extractor 41 and the detection reference information 42 from the storage 30 , and outputs the read feature extractor 41 and the read detection reference information 42 to the detection unit 51 .

[0111] As shown in Fig. 6, the detection unit 51 subdivides the target specimen image 15T into a plurality of patch images 70. Subsequently, as shown in Fig. 7, the detection unit 51 extracts features 72 from the patch images 70 using the feature extractor 41.

[0112] 13 , the detection unit 51 calculates a distance D between the representative position of the reference feature 72R and the position of the feature 72. Then, as shown in FIGS. 14 and 15 , the detection unit 51 compares the distance D with a detection threshold 84, thereby detecting whether or not a morphological abnormality has occurred in the liver specimen LVS captured in the patch image 70 (step ST120). A detection result 60 indicating whether or not a morphological abnormality has occurred in the liver specimen LVS captured in the patch image 70 is output from the detection unit 51 to the derivation unit 52.

[0113] The process of extracting the feature 72 from the patch image 70, the process of calculating the distance D between the representative position of the reference feature 72R and the position of the feature 72, and the process of detecting whether or not a morphological abnormality has occurred in the liver specimen LVS shown in the patch image 70 are performed on all patch images 70 of the target specimen image 15T. After the above processes are performed on all patch images 70, the process proceeds to step ST125.

[0114] 16, the derivation unit 52 derives the determination reference information 61 from the detection result 60 (step ST125). The determination reference information 61 is output from the derivation unit 52 to the determination unit 53 and the display control unit 55.

[0115] 17 and 18 , the determination unit 53 determines whether the estimated morphological abnormality portion is overestimated or underestimated based on the determination threshold value 43 and the determination reference information 61 (step ST130). The determination result 62 is output from the determination unit 53 to the generation unit 54.

[0116] If the determination unit 53 determines that the estimated morphological abnormality portion has been detected as being oversized or undersized (YES in step ST135), the generation unit 54 generates first cause identification reference information 631 or second cause identification reference information 632 (step ST140), as shown in Figures 20 to 26. Specifically, as shown in Figures 20 to 22, the types of artifacts that may have been erroneously detected as the estimated morphological abnormality portion and the number of identified artifacts are generated as the first cause identification reference information 631 or the second cause identification reference information 632. Also, as shown in Figures 23 and 24, the target histogram 95, the reference histogram 96, the Bhattacharyya distance, the hematoxylin component 102T and the eosin component 103T of the target specimen image 15T, and the hematoxylin component 102RR and the eosin component 103RR of the representative reference specimen image 15RR are generated as the first cause identification reference information 631 or the second cause identification reference information 632. 25 and 26 , the cluster images 115 to 117 are generated as first cause identification reference information 631. The first cause identification reference information 631 or the second cause identification reference information 632 is output from the generation unit 54 to the display control unit 55.

[0117] Under the control of the display control unit 55, the analysis result display screen 135 shown in Fig. 28 is displayed on the display 11 (step ST145). When the information display button 138 is selected by the user U (YES in step ST150), the information display screen 145 shown in Figs. 29 to 31 is displayed on the display 11 under the control of the display control unit 55 (step ST155). As a result, the first cause identification reference information 631 or the second cause identification reference information 632 is presented to the user U. In addition, a suppression process and / or an exclusion process is suggested to the user U by a message 155.

[0118] As described above, the CPU 32 of the drug discovery support device 10 includes the RW control unit 50, the detection unit 51, the derivation unit 52, the determination unit 53, and the display control unit 55. The RW control unit 50 acquires the target specimen image 15T, which depicts a liver specimen LVS of a subject S subjected to an evaluation test for a candidate substance 27, by reading it from the storage 30. The detection unit 51 detects a patch image 70 (estimated morphological abnormality portion) estimated to have a morphological abnormality among patch images 70 obtained by dividing the target specimen image 15T. The derivation unit 52 derives determination reference information 61 from the detection result 60 of the estimated morphological abnormality portion. The determination unit 53 determines whether the estimated morphological abnormality portion is over-detected or under-detected based on the determination reference information 61. When the display control unit 55 determines that the estimated morphological abnormality portion has been overdetected, it presents to the user U first cause identification reference information 631 that contributes to identifying the cause of the estimated morphological abnormality portion being overdetected. On the other hand, when the display control unit 55 determines that the estimated morphological abnormality portion has been underdetected, it presents to the user U second cause identification reference information 632 that contributes to identifying the cause of the estimated morphological abnormality portion being underdetected.

[0119] The user U can view the first cause identification reference information 631 or the second cause identification reference information 632 to identify the cause of the estimated morphological abnormality being detected as being oversized or undersized, or can perform improvement processing in accordance with the message 155. This can contribute to improving the accuracy of detecting morphological abnormalities occurring in the liver specimen LVS shown in the specimen image 15.

[0120] As shown in Figure 4 and other figures, the RW control unit 50 acquires one target specimen image 15T by reading it from the storage 30. The detection unit 51 detects estimated morphologically abnormal parts from one target specimen image 15T. The derivation unit 52 derives, as the determination reference information 61, a number ratio, which is a numerical value representing the spatial arrangement state of estimated morphologically abnormal parts in one target specimen image 15T. This makes it possible to simply detect estimated morphologically abnormal parts and derive the numerical value.

[0121] 17 and 18, the determination unit 53 determines whether the estimated morphological abnormality portion is over-detected or under-detected by comparing the number ratio with a preset determination threshold 43. This allows the determination to be completed easily and accurately. Furthermore, the determination criteria are clear, eliminating the risk of erroneous determination.

[0122] 16, the numerical value is a numerical value relating to the number of estimated morphologically abnormal portions, more specifically, a ratio of the number of portions, which allows the numerical value to be derived easily.

[0123] 20 to 22, the generation unit 54 identifies the type of artifact that may have been erroneously detected as the estimated morphological abnormality portion. As shown in Fig. 29, the display control unit 55 presents information related to the type to the user U as first cause identification reference information 631 and second cause identification reference information 632. This allows the user U to determine whether the artifact is the cause of the estimated morphological abnormality portion being detected as being oversized or undersized.

[0124] 29 and 30 , the display control unit 55 presents to the user U the target histogram 95, the reference histogram 96, the Bhattacharyya distance table 97, the hematoxylin component 102T and the eosin component 103T of the target specimen image 15T, and the hematoxylin component 102RR and the eosin component 103RR of the representative reference specimen image 15RR as first cause identification reference information 631 and second cause identification reference information 632. This allows the user U to determine whether the color of the target specimen image 15T is the cause of the estimated morphological abnormality portion being detected as being oversized or undersized.

[0125] As shown in FIGS. 25 and 26 , if the generation unit 54 determines that the estimated morphological abnormality portions are overexposed, it performs a clustering process to define the clusters to which each of the estimated morphological abnormality portions detected from one target specimen image 15T belongs. As shown in FIG. 31 , the display control unit 55 presents to the user U, as first cause identification reference information 631, multiple cluster images 115-117, which are generated by processing the target specimen image 15T and reflect the results of the clustering process. The cluster images 115-117 are images that allow multiple clusters to be distinguished using a display format 118 preset for each cluster. This allows the user U to determine whether the determination that the estimated morphological abnormality portions are overexposed is appropriate. It is rare for there to be only one type of morphological abnormality; multiple types are likely to occur. Therefore, if the number of clusters is relatively small, for example, one, it can be determined that the determination that the estimated morphological abnormality portions are overexposed is appropriate. On the other hand, if the number of clusters is relatively large, it can be determined that the determination that the estimated morphological abnormality portion is over-detected is inappropriate.

[0126] The display control unit 55 uses a message 155 to suggest to the user U suppression processing that suppresses over-detection or under-detection of the estimated morphological abnormality portion. Specifically, the suppression processing is processing to change the detection threshold 84 of the estimated morphological abnormality portion, color correction processing of the target specimen image 15T, and processing to change the resolution of the target specimen image 15T. Therefore, by having the user U take appropriate suppression processing in accordance with the message 155, the detection accuracy of the morphological abnormality can be improved. Note that the suppression processing does not have to be all of the processing to change the detection threshold 84, color correction processing of the target specimen image 15T, and processing to change the resolution of the target specimen image 15T, but may be at least one of these processes.

[0127] Here, consider a case where the user U changes the resolution of the target specimen image 15T in accordance with the message 155E prompting the user to change the resolution of the target specimen image 15T, specifically, a case where the user decides to use the original image 15O instead of the analysis image 15A as the target specimen image 15T. In this case, the feature extractor 41 is switched to use the original image 15O. That is, two types of feature extractors 41 are prepared in advance: one for the analysis image 15A and one for the original image 15O.

[0128] Furthermore, the display control unit 55 suggests to the user U, by means of a message 155, an exclusion process for excluding from the evaluation test any part of an artifact that may have been erroneously detected as a part of an estimated morphological abnormality. Therefore, by having the user U take an appropriate exclusion process in accordance with the message 155, the reliability of the evaluation test can be improved.

[0129] 6, 7, and 13, the detection unit 51 treats each of the multiple patch images 70 obtained by dividing the target specimen image 15T as a portion. The detection unit 51 detects the estimated morphological abnormality portion by comparing a feature amount 72 obtained by inputting the patch image 70 to the feature amount extractor 41 with a reference feature amount 72R obtained by inputting a reference patch image 70R depicting a liver specimen LVS considered to be normal to the feature amount extractor 41. This makes it possible to easily and accurately detect the estimated morphological abnormality portion.

[0130] Second Embodiment In the first embodiment, the suppression process and the exclusion process are merely suggested to the user U by the message 155, but this is not limiting. As an example, as shown in FIG. 35 , the suppression process and the exclusion process may be performed by the CPU 32.

[0131] 35, the CPU of the drug discovery support apparatus of the second embodiment functions as a color correction processing unit 160 and a detection threshold change processing unit 161 in addition to the processing units 50 to 55 of the first embodiment (the determination unit 53, generation unit 54, and display control unit 55 are not shown in FIG. 35). The color correction processing unit 160 is provided between the RW control unit 50 and the detection unit 51. The detection threshold change processing unit 161 is connected to the detection unit 51.

[0132] The color correction processing unit 160 performs color correction processing on the target specimen image 15T when the color difference between the target specimen image 15T and the representative reference specimen image 15RR is equal to or greater than a preset color threshold. The color correction processing is processing to bring the color of the target specimen image 15T closer to the color of the representative reference specimen image 15RR. Specifically, the color correction processing is processing to bring the Bhattacharyya distance between the target histogram 95 and the reference histogram 96 closer to 0, and / or processing to bring the hematoxylin component 102T and the eosin component 103T of the target specimen image 15T closer to the hematoxylin component 102RR and the eosin component 103RR of the representative reference specimen image 15RR. The color correction processing unit 160 outputs the target specimen image 15T after the color correction processing to the detection unit 51, the generation unit 54, and the display control unit 55.

[0133] The detection threshold change processing unit 161 performs a process of changing the detection threshold 84 when the determination unit 53 determines that the estimated morphological abnormality portion has been detected as being oversized or undersized, there is no type of artifact identified as possibly having been erroneously detected as the estimated morphological abnormality portion, and the difference in color between the target specimen image 15T and the representative reference specimen image 15RR does not exceed the color threshold. Specifically, when it is determined that the estimated morphological abnormality portion has been detected as being oversized, the detection threshold change processing unit 161 resets the detection threshold 84 to be higher. On the other hand, when it is determined that the estimated morphological abnormality portion has been detected as being undersized, the detection threshold change processing unit 161 resets the detection threshold 84 to be lower. The degree to which the detection threshold 84 is increased or decreased may be uniform, or may be changed depending on the difference between the number ratio and the determination threshold 43, etc. When changing the degree depending on the difference between the number ratio and the determination threshold 43, the degree is increased when the difference between the number ratio and the determination threshold 43 is relatively large, and decreased when the difference is relatively small.

[0134] When the determining unit 53 determines that the estimated morphological abnormality portion has been underdetected, there is no type of artifact that has been identified as possibly having been erroneously detected as the estimated morphological abnormality portion, and the difference in color between the target specimen image 15T and the representative reference specimen image 15RR does not exceed the color threshold, the RW control unit 50 reads out the original image 15O as the target specimen image 15T from the storage 30 instead of the analysis image 15A. In other words, the RW control unit 50 performs processing to change the resolution of the target specimen image 15T.

[0135] Furthermore, when there is a type of artifact that has been identified as possibly having been erroneously detected as a part of an estimated morphological abnormality, the derivation unit 52 excludes the patch image 70 including the artifact that has been identified as possibly having been erroneously detected as a part of an estimated morphological abnormality from the target for calculating the number ratio. In other words, the derivation unit 52 performs an exclusion process to exclude the part of the artifact that has been possibly erroneously detected as a part of an estimated morphological abnormality from the target for the evaluation test.

[0136] As described above, in the second embodiment, the RW control unit 50, the color correction processing unit 160, and the detection threshold change processing unit 161 perform suppression processing to suppress overdetection or underdetection of the estimated morphological abnormality portion. Specifically, the suppression processing is processing to change the detection threshold 84 of the estimated morphological abnormality portion by the detection threshold change processing unit 161, processing to correct the color of the target specimen image 15T by the color correction processing unit 160, and processing to change the resolution of the target specimen image 15T by the RW control unit 50. This makes it possible to improve the accuracy of morphological abnormality detection without requiring any intervention from the user U.

[0137] Furthermore, the derivation unit 52 performs an exclusion process to exclude from the evaluation test any artifact portion that may have been erroneously detected as an estimated morphological abnormality portion. This makes it possible to improve the reliability of the evaluation test without bothering the user U.

[0138] As in the first embodiment, the suppression processing and / or exclusion processing may be performed after being suggested to the user U. In this case, the suppression processing and / or exclusion processing may be performed only when instructed by the user U, or the suppression processing and / or exclusion processing may be performed without waiting for an instruction from the user U.

[0139] [Third Embodiment] In the first embodiment described above, processing was performed using one target specimen image 15T, but in the third embodiment shown in Figures 36 to 39, as an example, processing is performed using multiple target specimen images 15T. Figures 36 and 37 show an example in which processing is performed using target specimen images 15T that are images of liver specimens LVS of multiple subjects S that belong to the same group. In contrast, Figures 38 and 39 show an example in which processing is performed using target specimen images 15T that are images of liver specimens LVS of multiple subjects S that belong to different groups.

[0140] In the case of Figure 36, the RW control unit 50 acquires target specimen images 15T, which depict liver specimens LVS of multiple subjects S belonging to the low-dose group 25L, by reading them from the storage 30. The detection unit 51 detects estimated morphologically abnormal portions from each of the multiple target specimen images 15T. Furthermore, the derivation unit 52 derives the number ratio for each of the multiple target specimen images 15T. The derivation unit 52 derives a distribution 166 of the number ratio as determination reference information 165.

[0141] The determination unit 53 detects outliers OL from the number ratios constituting the distribution 166. Detection of outliers OL can be performed using a method using the interquartile range, the Smirnoff-Grubbs test, clustering processing, or the like. In this case, since the population of the number ratio distribution 166 is the low-dose group 25L, it is considered that the majority of the population is concentrated in areas with relatively low values, and the outliers OL appear in areas with relatively high values. Therefore, the determination unit 53 determines that an excessive number of estimated morphological abnormalities have been detected in the target specimen image 15T whose number ratio is outliers OL.

[0142] 37 shows an example using target specimen images 15T that capture liver specimens LVS from multiple subjects S belonging to the high-dose group 25H. In this case, the population is the high-dose group 25H, so the distribution 166 of the number ratio is considered to be the opposite of the case in FIG. 36, with the majority of the samples concentrated in areas with relatively high values, and outliers OL appearing in areas with relatively low values. Therefore, the determining unit 53 determines that estimated morphological abnormality portions have been underdetected in target specimen images 15T with a number ratio of outliers OL.

[0143] In the case of Figure 38, the RW control unit 50 acquires, by reading from the storage 30, target specimen images 15T depicting liver specimens LVS of multiple subjects S belonging to the control group 26 and target specimen images 15T depicting liver specimens LVS of multiple subjects S belonging to the high-dose group 25H. The detection unit 51 detects estimated morphologically abnormal portions from each of the multiple target specimen images 15T. Furthermore, the derivation unit 52 derives the number ratios of each of the multiple target specimen images 15T. The derivation unit 52 derives, as determination reference information 170, a distribution 171 of the number ratios of the control group 26 and a distribution 172 of the number ratios of the high-dose group 25H.

[0144] The determination unit 53 detects outliers OL from the number ratios constituting the distribution 172 of the high-dose group 25H. The determination unit 53 also determines the validity of the outliers OL detected from the distribution 172 of the high-dose group 25H by referring to the distribution 171 of the control group 26. Specifically, the determination unit 53 calculates the difference between a representative value, such as the average or median, of the number ratios of the control group 26 and the outliers OL detected from the distribution 172 of the high-dose group 25H, and determines that the outliers OL are valid if the difference is less than a preset threshold. In this case, the determination unit 53 determines that the estimated morphological abnormality portion has been underdetected in the target specimen image 15T whose number ratio is the outlier OL.

[0145] 39 shows an example in which a target specimen image 15T depicting liver specimens LVS of multiple subjects S belonging to the past medium-dose group 25MP and a target specimen image 15T depicting liver specimens LVS of multiple subjects S belonging to the medium-dose group 25M are used. Here, the past medium-dose group 25MP refers to an administration group to which a candidate substance identical to or similar to the candidate substance 27 of the current evaluation test was administered in the same amount as the medium-dose group 25M in a past evaluation test. A similar candidate substance refers to a candidate substance similar in composition to the candidate substance 27. The derivation unit 52 derives, as the determination reference information 170, a distribution 171 of the number ratio of the past medium-dose group 25MP and a distribution 172 of the number ratio of the medium-dose group 25M. In this case, the distributions of the number ratios 171 and 172 are the past medium administration group 25MP and the medium administration group 25M, so the majority of values ​​are concentrated in the intermediate range, and the outlier OL appears in either the relatively low or high value range.

[0146] In this case, the determination unit 53 detects an outlier OL from the number ratio constituting the distribution 172 of the medium administration group 25M. Furthermore, the determination unit 53 determines the validity of the outlier OL detected from the distribution 172 of the medium administration group 25M by referring to the distribution 171 of the past medium administration group 25MP. Specifically, the determination unit 53 calculates the difference between a representative value such as the average or median of the number ratio of the past medium administration group 25MP and the outlier OL detected from the distribution 172 of the medium administration group 25M, and if the difference is equal to or greater than a preset threshold, it determines that the outlier OL is valid.

[0147] If the outlier OL is in a region with relatively low values ​​and is valid, the determination unit 53 determines that the estimated morphological abnormality portion is underdetected in the target specimen image 15T whose count ratio is the outlier OL. On the other hand, as shown in the figure, if the outlier OL is in a region with relatively high values ​​and is valid, the determination unit 53 determines that the estimated morphological abnormality portion is overdetected in the target specimen image 15T whose count ratio is the outlier OL. Note that although FIGS. 36 to 39 illustrate the case where there is only one outlier OL, the number of outliers OL is not limited to one. Multiple outliers OL may be detected.

[0148] As described above, in the third embodiment, the RW control unit 50 acquires multiple target specimen images 15T by reading them from the storage 30. The detection unit 51 detects estimated morphologically abnormal portions from each of the multiple target specimen images 15T. The derivation unit 52 derives, as the determination reference information 165 or 170, a count ratio distribution 166, 171, or 172, which is a numerical value representing the spatial arrangement state of the estimated morphologically abnormal portions in each of the multiple target specimen images 15T. Therefore, by comparing multiple target specimen images 15T instead of a single target specimen image 15T, it is possible to determine whether the estimated morphologically abnormal portion is over-detected or under-detected. This significantly reduces the time required to determine whether the estimated morphologically abnormal portion is over-detected or under-detected for each target specimen image 15T.

[0149] 36 and 37 , the plurality of target specimen images 15T are images of liver specimens LVS of a plurality of subjects S belonging to the same group. The determination unit 53 determines whether the estimated morphological abnormality portion is over-detected or under-detected by detecting outliers OL from the number ratio constituting the distribution 166. Therefore, it is possible to quickly determine whether the estimated morphological abnormality portion is over-detected or under-detected for the plurality of target specimen images 15T depicting liver specimens LVS of a plurality of subjects S belonging to the same group.

[0150] As shown in Figures 38 and 39, the multiple target specimen images 15T are images of liver specimens LVS of multiple subjects S belonging to different groups. The derivation unit 52 derives distributions 171 and 172 for each different group. The determination unit 53 detects an outlier OL from the ratio of the number of individuals constituting one distribution 172 of the multiple distributions 171 and 172 derived for each different group, and determines whether the estimated morphological abnormality portion is over-detected or under-detected by referring to a distribution 171 other than distribution 172. This makes it possible to detect a valid outlier OL and improve the validity of the determination of whether the estimated morphological abnormality portion is over-detected or under-detected.

[0151] As shown in FIG. 38 , the different groups are the administration group 25 to which candidate substance 27 was administered and the control group 26 to which candidate substance 27 was not administered. This combination allows the validity of an outlier OL detected from the number ratio distribution 172 of the administration group 25 to be determined by referring to the number ratio distribution 171 of the control group 26. Furthermore, as shown in FIG. 39 , the different groups are the administration group 25 to which candidate substance 27 was administered and a past administration group to which a candidate substance identical to or similar to candidate substance 27 was administered in a past evaluation test. This combination allows the validity of an outlier OL detected from the number ratio distribution 172 of the administration group 25 to be determined by referring to the number ratio distribution 171 of the past administration group.

[0152] The different groups may be, for example, the low dose group 25L and the medium dose group 25M, or the medium dose group 25M and the high dose group 25H. The different groups may also be, for example, the high dose group 25H and the previous high dose group. Furthermore, the different groups may be three or more groups, for example, the low dose group 25L, the high dose group 25H, and the control group 26.

[0153] 40 and 41 , a representative value of the number ratio for each group is derived, and the representative value for each group is compared with an ideal value of the number ratio that is preset for each group to determine whether the estimated morphological abnormality portion is over-detected or under-detected.

[0154] As shown in FIG. 40 , the RW control unit 50 acquires multiple target specimen images 15T, each representing a liver specimen LVS of multiple subjects S belonging to a low-dose group 25L, a medium-dose group 25M, a high-dose group 25H, and a control group 26, by reading them from the storage 30. The detection unit 51 detects estimated morphologically abnormal portions from each of the multiple target specimen images 15T. The derivation unit 52 derives the number ratio for each of the multiple target specimen images 15T. As the determination reference information 175, the derivation unit 52 derives the average value of the number ratio for each of the low-dose group 25L, the medium-dose group 25M, the high-dose group 25H, and the control group 26, as shown in Table 176. The average value of the number ratio is an example of a "numerical representative value" according to the technology of the present disclosure. Note that instead of the average value, a median, a mode, or the like may be used as the representative value.

[0155] As an example, as shown in FIG. 41 , the determination unit 53 compares the average value of the number ratio for each group with the ideal value of the number ratio preset for each group. If the average value is greater than the ideal value and the difference between the average value and the ideal value is equal to or greater than a preset determination threshold, the determination unit 53 determines that the estimated morphologically abnormal portion is over-detected in the target specimen image 15T for that group. Also, if the average value is smaller than the ideal value and the difference between the average value and the ideal value is equal to or greater than a determination threshold, the determination unit 53 determines that the estimated morphologically abnormal portion is under-detected in the target specimen image 15T for that group. FIG. 41 illustrates a case in which the estimated morphologically abnormal portion is over-detected in the target specimen image 15T for the low-dose group 25L and the control group 26. Also, FIG. 41 illustrates a case in which the estimated morphologically abnormal portion is under-detected in the target specimen image 15T for the medium-dose group 25M.

[0156] In this case, the display control unit 55 switchably displays on the information display screen 145 multiple cause identification reference information 63 for multiple target specimen images 15T of a group in which it has been determined that the estimated morphological abnormality portion has been detected as being oversized or undersized.

[0157] As described above, in the fourth embodiment, the RW control unit 50 acquires multiple target specimen images 15T, each representing a liver specimen LVS of multiple subjects S belonging to different groups, by reading them from the storage 30. The detection unit 51 detects estimated morphological abnormality portions from each of the multiple target specimen images 15T. The derivation unit 52 derives, as the determination reference information 175, an average value of the number ratio, which is a representative value of numerical values ​​representing the spatial arrangement state of the estimated morphological abnormality portions in each of the multiple target specimen images 15T for each different group. Therefore, by comparing multiple target specimen images 15T instead of a single target specimen image 15T, it is possible to determine whether the estimated morphological abnormality portion is over-detected or under-detected. This significantly reduces the time required to determine whether the estimated morphological abnormality portion is over-detected or under-detected for each target specimen image 15T.

[0158] 41 , the determination unit 53 determines whether the estimated morphological abnormality portion is over-detected or under-detected by comparing the average value of the number ratio for each different group with the ideal value of the number ratio preset for each different group. Therefore, it is possible to quickly determine whether the estimated morphological abnormality portion is over-detected or under-detected for multiple target specimen images 15T that capture liver specimens LVS of multiple subjects S belonging to one group.

[0159] The different groups are an administration group 25 to which the candidate substance 27 is administered, and a control group 26 to which the candidate substance 27 is not administered. Therefore, it is possible to quickly determine whether the estimated morphologically abnormal portion is over-detected or under-detected for the plurality of target specimen images 15T that capture the liver specimens LVS of the plurality of subjects S belonging to each of the administration group 25 and the control group 26.

[0160] The administration group 25 includes a low administration group 25L, a medium administration group 25M, and a high administration group 25H, which have different doses of the candidate substance 27. Therefore, it is possible to quickly determine whether the estimated morphological abnormality portion is over-detected or under-detected for the plurality of target specimen images 15T, which are images of liver specimens LVS of the plurality of subjects S belonging to the low administration group 25L, the medium administration group 25M, and the high administration group 25H, respectively.

[0161] The different groups may be two groups, namely, the administration group 25 and the control group 26. Alternatively, the different groups may be three groups, namely, the low administration group 25L, the medium administration group 25M, and the high administration group 25H, excluding the control group 26.

[0162] The processing of the fourth embodiment is performed to estimate a group likely to include target specimen images 15T in which the estimated morphologically abnormal portion is detected as being oversized or undersized. Then, the processing of the first or second embodiment may be performed on the target specimen images 15T in the estimated group to identify target specimen images 15T in which the estimated morphologically abnormal portion is detected as being oversized or undersized.

[0163] (Variation 1) When using one target specimen image 15T, the numerical value representing the spatial arrangement state of estimated morphological abnormality portions is not limited to a numerical value relating to the number of estimated morphological abnormality portions, such as the number ratio exemplified in each of the above embodiments. As an example, as shown in Figures 42 and 43 , the derivation unit 52 generates a distribution 180 of the distances from a reference point to the detected estimated morphological abnormality portions, and derives the variance σA of the distribution 180 as a numerical value representing the spatial arrangement state of the estimated morphological abnormality portions. The derived variance σA, together with the number ratio, may then be used as the determination reference information 181. The reference point may be, for example, the center point of the target specimen image 15T.

[0164] Artifacts that may be erroneously detected as estimated morphological abnormality portions are often concentrated in specific regions. Therefore, for a target specimen image 15T in which artifacts occur, the variance σA of the distribution 180 tends to be small. Therefore, when the number ratio is equal to or greater than the first judgment threshold 431, the determination unit 53 further uses the variance σA to make a determination. Specifically, the determination unit 53 compares the magnitude of the variance σA with a preset third judgment threshold 433. Then, as shown in FIG. 42 , when the variance σA is equal to or greater than the third judgment threshold 433, the determination unit 53 determines that the estimated morphological abnormality portion is not overexposed and outputs a determination result 62 to that effect. On the other hand, as shown in FIG. 43 , when the variance σA is less than the third judgment threshold 433, the determination unit 53 determines that the estimated morphological abnormality portion is overexposed and outputs a determination result 62 to that effect. Note that, instead of the variance σA, the standard deviation of the distribution 180 may be derived as a numerical value representing the spatial arrangement state of the estimated morphological abnormality portion.

[0165] 44, the derivation unit 52 performs a Fourier transform on the target specimen image 15T to derive a spatial frequency spectrum 190 of the target specimen image 15T, and derives a variance σB of the spatial frequency spectrum 190 as a numerical value representing the spatial arrangement state of the estimated morphological abnormality part. The derived variance σB may then be used as determination reference information 191 together with the number ratio.

[0166] The imaging device 19 captures the specimen image 15 by dividing it into multiple regions. Therefore, some specimen images 15 have noticeable periodic seams between the multiple regions. These seams are a type of artifact that may be erroneously detected as an estimated morphological abnormality. Therefore, for a target specimen image 15T in which artifacts such as the seams described above occur, the variance σB of the spatial frequency spectrum 190 tends to be small. Therefore, when the number ratio is equal to or greater than the first judgment threshold 431, the determination unit 53 further uses the variance σB to make a determination. Specifically, the determination unit 53 compares the magnitude of the variance σB with a predetermined fourth judgment threshold 434. Then, as indicated by the left arrow, if the variance σB is equal to or greater than the fourth judgment threshold 434, the determination unit 53 determines that the estimated morphological abnormality is not overexposed and outputs a determination result 62 to that effect. On the other hand, as indicated by the right arrow, if the variance σB is less than the fourth judgment threshold 434, the determination unit 53 determines that the estimated morphological abnormality is overexposed and outputs a determination result 62 to that effect. Note that, as a numerical value representing the spatial arrangement state of the estimated morphological abnormality parts, the standard deviation of the spatial frequency spectrum 190 may be derived instead of the variance σB. In this way, various numerical values ​​can be used as the numerical value representing the spatial arrangement state of the estimated morphological abnormality parts, in addition to a numerical value relating to the number of estimated morphological abnormality parts, such as a number ratio.

[0167] In addition to the reference patch image 70R depicting a liver specimen LVS deemed normal, a patch image 70 depicting a liver specimen LVS with a morphological abnormality may also be used as the training reference patch image 70RL. The patch image 70 depicting a liver specimen LVS with a morphological abnormality is obtained, for example, from a past administration group consisting of multiple subjects S to whom a candidate substance was administered in a past evaluation test. This allows the autoencoder 75, and therefore the feature extractor 41, to learn about liver specimens LVS with a wider variety of shape and texture characteristics. As a result, the feature extractor 41 can extract features 72 that better represent the shape and texture characteristics of the liver specimen LVS.

[0168] Note that the patch image 70 depicting the liver specimen LVS with morphological abnormality is not limited to one acquired from the subject S constituting the exemplified past administration group. Morphological abnormalities may also occur in the subject S constituting the past control group 26P. Therefore, as long as the patch image 70 depicts a liver specimen LVS with morphological abnormality, it does not matter whether the subject S is in the past control group 26P or the past administration group. Furthermore, the patch image 70 depicting the liver specimen LVS with morphological abnormality may be an image acquired from a subject S who has been intentionally induced to develop morphological abnormality by applying various stresses. Furthermore, the patch image 70 depicting the liver specimen LVS with morphological abnormality may be an image artificially created by processing a patch image 70 depicting a normal liver specimen LVS.

[0169] Instead of the encoder unit 76 of the autoencoder 75, an encoder unit of a convolutional neural network that outputs a class discrimination result in response to an input of a patch image 70 may be used as the feature extractor 41. The class discrimination result is a result of discriminating one type of morphological abnormality occurring in the liver specimen LVS shown in the patch image 70 from multiple types such as hyperplasia, infiltration, congestion, and inflammation.

[0170] Furthermore, the machine learning model substituted for the feature extractor 41 is not limited to the autoencoder 75 and the convolutional neural network illustrated in the examples. A generator of a generative adversarial network (GAN) may be substituted for the feature extractor 41. A machine learning model without a convolutional layer, such as a Vision Transformer (ViT), may also be substituted for the feature extractor 41.

[0171] Contrastive learning may be performed to reduce the distance between features derived from the same image in feature space and increase the distance between features derived from different images in feature space. Known examples of contrastive learning include learning methods such as SimCLR (A Simple Framework for Contrastive Learning of Visual Representations). Furthermore, learning methods such as BYOL (Bootstrap Your Own Latent) that do not use the above-described pairs of different images (also called negative samples) may also be used. Furthermore, constraints may be imposed on the distribution of extracted features, such as distribution on a unit sphere or distribution following a standard normal distribution.

[0172] [Fifth Embodiment] As an example, as shown in Fig. 45, the fifth embodiment deals with a slide specimen 195 in which tissue specimens of multiple organs are mounted on a single slide glass 16. Fig. 45 illustrates a case in which a heart specimen HS, a brain specimen BS, and a bone marrow specimen BMS are mounted in addition to a liver specimen LVS. In this case, the specimen image 15 shows the heart specimen HS, the brain specimen BS, and the bone marrow specimen BMS in addition to the liver specimen LVS.

[0173] The CPU of the drug discovery support apparatus of the fifth embodiment functions as an identification unit 196 in addition to the respective processing units 50 to 55 of the first embodiment. The identification unit 196 identifies the tissue specimen of each organ from the specimen image 15, for example, using a template for identifying the tissue specimen of each organ or a machine learning model. The identification unit 196 outputs coordinate information of boxes 197 to 200 surrounding the tissue specimen of each organ as the identification result. Box 197 is a box surrounding the heart specimen HS, and box 198 is a box surrounding the liver specimen LVS. Furthermore, box 199 is a box surrounding the brain specimen BS, and box 200 is a box surrounding the bone marrow specimen BMS.

[0174] 46 shows how the detection unit 51 subdivides the cardiac specimen HS in a frame 197 into a plurality of patch images 70 and extracts feature quantities 72 from the patch images 70 using a feature extractor 202 for the cardiac specimen HS. The detection unit 51 also extracts feature quantities 72 for the liver specimen LVS in a frame 198, the brain specimen BS in a frame 199, and the bone marrow specimen BMS in a frame 200 using dedicated feature extractors. That is, the detection unit 51 extracts feature quantities 72 for each of the tissue specimens of each organ identified by the identification unit 196. The subsequent processing is the same as the processing shown in the first embodiment and the like, except that a determination is made for each of the tissue specimens of each organ depicted in the target specimen image 15T, and therefore will not be illustrated or described again.

[0175] As described above, in the fifth embodiment, the specimen image 15 is an image obtained by capturing a specimen slide 195 on which tissue specimens from multiple organs are placed. The identification unit 196 identifies the tissue specimens of each organ from the specimen image 15. The detection unit 51 extracts the feature quantities 72 for each identified tissue specimen of each organ. This allows for support of the specimen slide 195 on which tissue specimens from multiple organs are placed. The specimen slide 195 on which tissue specimens from multiple organs are placed, as in this embodiment, is more common than the specimen slide 18 on which a tissue specimen from a single organ is placed, as in the first embodiment. This allows for processing that is more suited to general use. The user U may also define frames 197-200 representing the tissue specimens of each organ in the specimen image 15.

[0176] The feature amount 72 is not limited to that extracted by the feature amount extractor 41. It may be the average value, maximum value, minimum value, mode, or variance of the pixel values ​​of the patch image 70.

[0177] The organ is not limited to the liver LV, but may be the stomach, lung, small intestine, or large intestine. The subject S is not limited to a rat, but may be a mouse, guinea pig, gizzard shad, hamster, ferret, rabbit, dog, cat, monkey, or the like.

[0178] The drug discovery support device 10 may be a personal computer installed in a drug development facility as shown in FIG. 1, or may be a server computer installed in a data center independent of the drug development facility.

[0179] When the drug discovery support device 10 is configured as a server computer, the specimen image 15 is transmitted from a personal computer installed in each drug development facility to the server computer via a network such as the Internet. The server computer distributes various screens, such as the target selection screen 125, to the personal computer in the form of screen data for web distribution created using a markup language such as XML (Extensible Markup Language). The personal computer reproduces the screen to be displayed on the web browser based on the screen data and displays it on a display. Note that other data description languages, such as JSON (Javascript (registered trademark) Object Notation), may be used instead of XML.

[0180] The drug discovery support device 10 according to the technique of the present disclosure can be widely used across all stages of drug development, from the earliest stage of setting drug discovery targets to the final stage of clinical trials.

[0181] The candidate substance 27 is not limited to the example drug, but may be other chemical substances such as pesticides or radioactive materials.

[0182] The hardware configuration of the computer constituting the drug discovery support device 10 according to the technology of the present disclosure can be modified in various ways. For example, the drug discovery support device 10 can be configured with multiple computers separated as hardware in order to improve processing power and reliability. For example, the functions of the detection unit 51 and the derivation unit 52 and the functions of the determination unit 53 and the generation unit 54 can be distributed and performed by two computers. In this case, the drug discovery support device 10 is configured with two computers.

[0183] In this way, the hardware configuration of the computer of the drug discovery support apparatus 10 can be changed as appropriate depending on the required performance, such as processing power, safety, and reliability. Furthermore, not only the hardware, but also application programs such as the operating program 40 can be duplicated or stored in multiple storage devices in order to ensure safety and reliability.

[0184] In each of the above embodiments, the hardware structure of the processing units that perform various processes, such as the RW control unit 50, the detection unit 51, the derivation unit 52, the judgment unit 53, the generation unit 54, the display control unit 55, the color correction processing unit 160, the detection threshold change processing unit 161, and the identification unit 196, can be any of the various processors shown below. As described above, the various processors include the CPU 32, which is a general-purpose processor that executes software (operating program 40) and functions as various processing units, as well as programmable logic devices (PLDs) that are processors whose circuit configuration can be changed after manufacture, such as FPGAs (Field Programmable Gate Arrays), and dedicated electrical circuits that are processors having a circuit configuration designed specifically for executing specific processing, such as ASICs (Application Specific Integrated Circuits).

[0185] A single processing unit may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs and / or a combination of a CPU and an FPGA).Furthermore, multiple processing units may be configured with a single processor.

[0186] Examples of configuring multiple processing units with a single processor include, first, a form in which one processor is configured with a combination of one or more CPUs and software, as typified by computers such as client and server, and this processor functions as multiple processing units. Second, a form in which a processor is used to realize the functions of the entire system including multiple processing units with a single IC (Integrated Circuit) chip, as typified by systems on chips (SoCs). In this way, various processing units are configured using one or more of the above-mentioned various processors as a hardware structure.

[0187] Furthermore, more specifically, the hardware structure of these various processors can be an electric circuit (circuitry) that combines circuit elements such as semiconductor elements.

[0188] From the above description, the technology described in the following supplementary paragraphs can be understood.

[0189] [Supplementary Item 1] A drug discovery support device comprising a processor, wherein the processor acquires a specimen image depicting a tissue specimen of an organ of a subject subjected to an evaluation test for a candidate substance, detects a presumed morphologically abnormal portion of the specimen image where a morphological abnormality is presumed to have occurred, derives determination reference information from the detection result of the presumed morphologically abnormal portion, and determines whether the presumed morphologically abnormal portion is over-detected or under-detected based on the determination reference information, and if it is determined that the presumed morphologically abnormal portion is over-detected, presents to a user first cause identification reference information that contributes to identifying a cause why the presumed morphologically abnormal portion is over-detected, and if it is determined that the presumed morphologically abnormal portion is under-detected, presents to a user second cause identification reference information that contributes to identifying a cause why the presumed morphologically abnormal portion is under-detected. [Supplementary Item 2] The drug discovery support device according to Supplementary Item 1, wherein the processor acquires one of the specimen images, detects the estimated morphological abnormality portion from the one specimen image, and derives, as the determination reference information, a numerical value representing the spatial arrangement state of the estimated morphological abnormality portion in the one specimen image. [Supplementary Item 3] The drug discovery support device according to Supplementary Item 2, wherein the processor compares the numerical value with a preset determination threshold to determine whether the estimated morphological abnormality portion has been overdetected or underdetected. [Supplementary Item 4] The drug discovery support device according to Supplementary Item 2 or Supplementary Item 3, wherein the numerical value is a numerical value relating to the number of the estimated morphological abnormality portions. [Supplementary Item 5] The drug discovery support device according to Supplementary Item 1, wherein the processor acquires a plurality of the specimen images, detects the estimated morphological abnormality portion from each of the plurality of specimen images, and derives, as the determination reference information, a distribution of numerical values ​​representing the spatial arrangement state of the estimated morphological abnormality portion in each of the plurality of specimen images. [Supplementary Item 6] The drug discovery support device according to Supplementary Item 5, wherein the plurality of specimen images are images of tissue specimens of a plurality of subjects belonging to the same group, and the processor detects outliers from the numerical values ​​constituting the distribution to determine whether the estimated morphological abnormality portion is over-detected or under-detected.[Supplementary Item 7] The drug discovery support device according to Supplementary Item 5 or 6, wherein the plurality of specimen images are images of tissue specimens of a plurality of subjects belonging to different groups, and the processor derives the distribution for each of the different groups, detects outliers from the numerical values ​​constituting one of the plurality of distributions derived for each of the different groups, and determines whether the estimated morphological abnormality portion is over-detected or under-detected by referring to a distribution other than the one distribution. [Supplementary Item 8] The drug discovery support device according to Supplementary Item 7, wherein the different groups are an administration group to which the candidate substance is administered and a control group to which the candidate substance is not administered, or an administration group to which a candidate substance identical to or similar to the candidate substance was administered in a previous evaluation test. [Supplementary Item 9] The drug discovery support device according to any one of Supplementary Item 5 to Supplementary Item 8, wherein the numerical value is a numerical value related to the number of estimated morphological abnormality portions. [Supplementary Item 10] The drug discovery support device according to any one of Supplementary Items 1 to 9, wherein the processor acquires a plurality of specimen images depicting tissue specimens of a plurality of subjects belonging to different groups, detects the estimated morphological abnormality portion from each of the plurality of specimen images, and derives, for each of the different groups, a representative value of a numerical value representing a spatial arrangement state of the estimated morphological abnormality portion in each of the plurality of specimen images as the determination reference information. [Supplementary Item 11] The drug discovery support device according to Supplementary Item 10, wherein the processor compares the representative value for each of the different groups with an ideal numerical value set in advance for each of the different groups to determine whether the estimated morphological abnormality portion is over-detected or under-detected. [Supplementary Item 12] The drug discovery support device according to Supplementary Item 10 or 11, wherein the different groups are an administration group to which the candidate substance is administered and a control group to which the candidate substance is not administered. [Supplementary Item 13] The drug discovery support device according to Supplementary Item 12, wherein the administration group includes a plurality of sub-administration groups each having a different dose of the candidate substance. [Supplementary Item 14] The drug discovery support device according to any one of Supplementary Items 10 to 13, wherein the numerical value is a numerical value relating to the number of the estimated morphologically abnormal portions.[Supplementary Item 15] The drug discovery support device according to any one of Supplementary Items 1 to 14, wherein there are multiple types of artifacts that may be erroneously detected as the suspected morphological abnormality portion, and the processor identifies the type of the artifact that may have been erroneously detected as the suspected morphological abnormality portion, and presents information on the type to the user as the first cause identification reference information and the second cause identification reference information. [Supplementary Item 16] The drug discovery support device according to any one of Supplementary Items 1 to 15, wherein the processor presents information on the color of the specimen image and information on the color of a reference specimen image of a tissue specimen deemed to be normal to the user as the first cause identification reference information and the second cause identification reference information. [Supplementary Item 17] The drug discovery support device according to any one of Supplementary Items 1 to 16, wherein the processor, when determining that the estimated morphological abnormality portions are overdetected, performs a clustering process to define a cluster to which each of the plurality of estimated morphological abnormality portions detected from one of the specimen images belongs, and presents to the user, as the first cause identification reference information, a plurality of cluster images that are generated by processing the specimen image and reflect the results of the clustering process, and in which the plurality of clusters are distinguishable in a display format that is preset for each of the plurality of clusters. [Supplementary Item 18] The drug discovery support device according to any one of Supplementary Items 1 to 17, wherein the processor suggests to the user and / or performs suppression process to suppress overdetection or underdetection of the estimated morphological abnormality portions. [Supplementary Item 19] The drug discovery support device according to Supplementary Item 18, wherein the suppression processing is at least one of: processing to change a detection threshold used in detecting the estimated morphological abnormality portion; processing to correct a color of the specimen image; and processing to change a resolution of the specimen image in detecting the estimated morphological abnormality portion. [Supplementary Item 20] The drug discovery support device according to any one of Supplementary Items 1 to 19, wherein the processor suggests to the user and / or performs exclusion processing to exclude from the evaluation test an artifact portion that may have been erroneously detected as the estimated morphological abnormality portion.[Supplementary Item 21] The drug discovery support device according to any one of Supplementary Items 1 to 20, wherein the processor treats each of a plurality of patch images obtained by dividing the specimen image as the portion, and detects the estimated morphological abnormality portion by comparing a feature obtained by inputting the patch image into a machine learning model with a reference feature obtained by inputting a reference patch image of a tissue specimen considered to be normal into the machine learning model.

[0190] The technology of the present disclosure can be appropriately combined with the various embodiments and / or various modified examples described above. Furthermore, it is not limited to the above-described embodiments, and various configurations can be adopted without departing from the spirit of the present disclosure. Furthermore, the technology of the present disclosure extends not only to programs but also to storage media that non-temporarily store programs.

[0191] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[0192] In this specification, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed by connecting them with "and / or."

[0193] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

Claims

1. A drug discovery support device comprising a processor, which acquires a specimen image of a tissue specimen from an organ of a subject subjected to an evaluation test of a candidate substance, detects a presumed morphological abnormality portion among portions of the specimen image where a morphological abnormality is presumed to have occurred, derives judgment reference information from a detection result of the presumed morphological abnormality portion, and determines whether the presumed morphological abnormality portion has been overdetected or underdetected based on the judgment reference information, and if it is determined that the presumed morphological abnormality portion is overdetected, presents to a user first cause identification reference information that helps to identify a cause why the presumed morphological abnormality portion is overdetected, and if it is determined that the presumed morphological abnormality portion is underdetected, presents to a user second cause identification reference information that helps to identify a cause why the presumed morphological abnormality portion is underdetected.

2. The drug discovery support device of claim 1, wherein the processor acquires one of the specimen images, detects the estimated morphological abnormality portion from the one of the specimen images, and derives a numerical value representing the spatial arrangement state of the estimated morphological abnormality portion in the one of the specimen images as the judgment reference information.

3. The drug discovery support device of claim 2, wherein the processor determines whether the estimated morphological abnormality portion is over-detected or under-detected by comparing the numerical value with a predetermined judgment threshold.

4. A drug discovery support device as described in claim 2, wherein the numerical value is a numerical value related to the number of the estimated morphologically abnormal parts.

5. The drug discovery support device of claim 1, wherein the processor acquires a plurality of the specimen images, detects the estimated morphological abnormality portion from each of the plurality of the specimen images, and derives, as the judgment reference information, a distribution of numerical values ​​representing the spatial arrangement state of the estimated morphological abnormality portion in each of the plurality of the specimen images.

6. The drug discovery support device described in claim 5, wherein the multiple specimen images are images of tissue specimens from multiple subjects belonging to the same group, and the processor detects outliers from the numerical values ​​that make up the distribution to determine whether the estimated morphological abnormality is over-detected or under-detected.

7. The drug discovery support device of claim 5, wherein the multiple specimen images are images of tissue specimens of multiple subjects belonging to different groups, and the processor derives the distribution for each of the different groups, detects outliers from the numerical values ​​that constitute one of the multiple distributions derived for each of the different groups, and determines whether the estimated morphological abnormality is over-detected or under-detected by referring to a distribution other than the one distribution.

8. A drug discovery support device as described in claim 7, wherein the different groups are an administration group to which the candidate substance is administered and a control group to which the candidate substance is not administered, or an administration group to which the candidate substance is administered and an administration group to which a candidate substance identical or similar to the candidate substance was administered in a previous evaluation test.

9. A drug discovery support device as described in claim 5, wherein the numerical value is a numerical value related to the number of the estimated morphologically abnormal parts.

10. The drug discovery support device of claim 1, wherein the processor acquires a plurality of specimen images of tissue specimens of a plurality of subjects belonging to different groups, detects the estimated morphological abnormality portion from each of the plurality of specimen images, and derives, as the judgment reference information, a numerical representative value representing the spatial arrangement state of the estimated morphological abnormality portion in each of the plurality of specimen images for each of the different groups.

11. The drug discovery support device described in claim 10, wherein the processor determines whether the estimated morphological abnormality portion is over-detected or under-detected by comparing the representative value for each of the different groups with an ideal value of the numerical value that is preset for each of the different groups.

12. A drug discovery support device as described in claim 10, wherein the different groups are an administration group to which the candidate substance is administered and a control group to which the candidate substance is not administered.

13. A drug discovery support device as described in claim 12, wherein the administration group includes a plurality of sub-administration groups each having a different dose of the candidate substance.

14. A drug discovery support device as described in claim 10, wherein the numerical value is a numerical value related to the number of the estimated morphologically abnormal parts.

15. The drug discovery support device of claim 1, wherein there are multiple types of artifacts that may be erroneously detected as the suspected morphological abnormality, and the processor identifies the type of the artifact that may have been erroneously detected as the suspected morphological abnormality, and presents information regarding the type to the user as the first cause identification reference information and the second cause identification reference information.

16. The drug discovery support device of claim 1, wherein the processor presents to the user, as the first cause identification reference information and the second cause identification reference information, information regarding the color of the specimen image and information regarding the color of a reference specimen image depicting a tissue specimen considered to be normal.

17. The drug discovery support device of claim 1, wherein, when the processor determines that the estimated morphological abnormality portion has been over-detected, the processor performs a clustering process to define a cluster to which each of the multiple estimated morphological abnormality portions detected from one of the specimen images belongs, and presents to the user multiple cluster images reflecting the results of the clustering process, which are generated by processing the specimen image, as the first cause identification reference information, and in which the multiple clusters can be distinguished using a display format preset for each of the multiple clusters.

18. A drug discovery support device as described in claim 1, wherein the processor suggests to the user and / or performs a suppression process for suppressing over-detection or under-detection of the estimated morphological abnormality portion.

19. The drug discovery support device described in claim 18, wherein the suppression processing is at least one of: a processing for changing a detection threshold used for detecting the estimated morphological abnormality portion; a processing for correcting the color of the specimen image; and a processing for changing the resolution of the specimen image in detecting the estimated morphological abnormality portion.

20. A drug discovery support device as described in claim 1, wherein the processor proposes to the user and / or performs an exclusion process to exclude from the evaluation test any artifact portion that may have been erroneously detected as the suspected morphological abnormal portion.

21. The drug discovery support device of claim 1, wherein the processor detects the estimated morphological abnormality part by treating each of a plurality of patch images obtained by dividing the specimen image as the part, and comparing features obtained by inputting the patch image into a machine learning model with reference features obtained by inputting a reference patch image of a tissue specimen considered to be normal into the machine learning model.

22. A method for operating a drug discovery support device, comprising: acquiring a specimen image depicting a tissue specimen of an organ of a subject subjected to an evaluation test of a candidate substance; detecting a presumed morphological abnormality portion among portions of the specimen image where a morphological abnormality is presumed to have occurred; deriving judgment reference information from a result of the detection of the presumed morphological abnormality portion; determining whether the presumed morphological abnormality portion has been overdetected or underdetected based on the judgment reference information; when it is determined that the presumed morphological abnormality portion has been overdetected, presenting to a user first cause identification reference information that contributes to identifying a cause why the presumed morphological abnormality portion is overdetected; and when it is determined that the presumed morphological abnormality portion is underdetected, presenting to a user second cause identification reference information that contributes to identifying a cause why the presumed morphological abnormality portion is underdetected.

23. An operating program for a drug discovery support device that causes a computer to execute processes including: acquiring a specimen image depicting a tissue specimen of an organ of a subject used in an evaluation test of a candidate substance; detecting a presumed morphological abnormality portion among portions of the specimen image where a morphological abnormality is presumed to have occurred; deriving judgment reference information from the detection result of the presumed morphological abnormality portion; determining whether the presumed morphological abnormality portion has been overdetected or underdetected based on the judgment reference information; presenting to a user first cause identification reference information that contributes to identifying a cause for the presumed morphological abnormality portion being overdetected when it is determined that the presumed morphological abnormality portion is overdetected; and presenting to a user second cause identification reference information that contributes to identifying a cause for the presumed morphological abnormality portion being underdetected when it is determined that the presumed morphological abnormality portion is underdetected.

Images