Drug research and development assistance device, method for operating drug research and development assistance device, and program for operating drug research and development assistance device

Abstract

A drug development assistance device is provided with a processor that acquires a plurality of specimen images of tissue specimens in which a plurality of organs of a subject to be used in an evaluation test of a candidate substance are captured, and performs evaluation of the candidate substance on the basis of selection priority information in which a selection priority is set for each of the plurality of organs. A target specimen image in which a tissue specimen of one organ has been captured is selected from the plurality of specimen images, whether or not a morphological abnormality has occurred in the tissue specimen captured in the target specimen image is determined, and selection priority information is updated on the basis of the determination result of whether or not the morphological abnormality has occurred.

Description

Technical Field

[0001] The present invention relates to a drug development auxiliary device, a method for operating the drug development auxiliary device, and a procedure for operating the drug development auxiliary device. Background Technology

[0002] In the field of drug development, experiments are conducted to administer candidate drugs to subjects such as rats and evaluate the efficacy and toxicity of these candidate drugs. Such evaluation experiments utilize images of tissue specimens (brain, liver, heart, etc.) collected from the subjects through necropsy. More specifically, the images detect morphological abnormalities occurring in the tissue specimens captured in the specimen images. Previously, pathologists and other users would observe specimen images to detect areas presumed to have morphological abnormalities (hereinafter referred to as presumed morphological abnormalities). However, with recent advancements in image analysis technology, techniques have been developed that can automatically detect presumed morphological abnormalities without user intervention.

[0003] However, although techniques are used to automatically detect presumed morphological abnormalities, the final determination of whether an abnormality has actually occurred still rests with users such as pathologists. Furthermore, the number of specimen images processed in a single evaluation trial can be in the thousands. Therefore, the burden on users remains considerable.

[0004] Previously, as a method for effectively analyzing large numbers of images, techniques described in Japanese Patent Application Publication No. 2009-077800 and International Patent Publication No. 2018 / 008195 have been proposed. Japanese Patent Application Publication No. 2009-077800 describes a technique for processing multiple images captured in a time sequence, such as multiple images captured by a capsule endoscope. In Japanese Patent Application Publication No. 2009-077800, anomaly detection is performed on each segmented region of a first image among the multiple images captured in a time sequence. Then, based on the anomaly detection results of the first image, the order in which anomaly detection is performed on the segmented regions of the second image is set, and the segmented regions of the second image are analyzed after the first image has been analyzed. Specifically, based on the assumption that approximately the same area in the time-series images may contain anomalies, the segmented regions in the first image where anomalies were detected, and the surrounding segmented regions, are set to a higher order than other segmented regions.

[0005] International Publication No. 2018 / 008195 also describes a technique for processing multiple images captured using a capsule endoscope. In International Publication No. 2018 / 008195, the order of image processing is set for multiple image groups obtained by capturing images inside multiple subjects using a capsule endoscope, based on the number of images of interest. Specifically, the order of image groups with a relatively large number of images of interest is set higher than the order of image groups with a relatively small number of images of interest. Furthermore, paragraph

[0141] of International Publication No. 2018 / 008195 describes setting the image processing order based on the number of images of interest for image groups obtained from a single subject, i.e., image groups separated by organs such as the stomach, small intestine, and large intestine. Incidentally, images of interest include, for example, images with a high red component, images showing detected lesions, images with characteristic values ​​within a specified range, or images captured by the user. Summary of the Invention

[0006] The technical problem to be solved by the invention

[0007] In Japanese Patent Application Publication No. 2009-077800, the order is set based on the presumption that abnormalities may be captured in approximately the same area in images taken in a time series. Therefore, it cannot be applied to specimen images of tissue specimens taken at a time point such as after the evaluation test. Furthermore, while Japanese Patent Application Publication No. 2009-077800 can quickly detect abnormalities in a second image, if no abnormalities are found in the second image, the detection process becomes futile.

[0008] International Publication No. 2018 / 008195 requires identifying images of interest in each of the multiple image groups by calculating the red component or detecting lesions in all images of each group. When a user-captured image is designated as the image of interest, the user needs to spend time observing the image.

[0009] One embodiment of the present invention provides a drug development auxiliary device, a method for operating the drug development auxiliary device, and a procedure for operating the drug development auxiliary device, which can prioritize the analysis of specimen images that are considered to have morphological abnormalities in tissue specimens without causing processing load.

[0010] means for solving technical problems

[0011] This invention provides a drug development auxiliary device, which includes a processor. The processor acquires multiple specimen images of tissue specimens from multiple organs of a subject for evaluation tests of candidate substances. Based on selection priority information that sets a selection priority for each of the multiple organs, the processor selects a target specimen image of a tissue specimen from the multiple specimen images and determines whether morphological abnormalities have occurred in the tissue specimen captured in the target specimen image. Based on the determination result of whether morphological abnormalities have occurred, the selection priority information is updated.

[0012] Preferably, in the selection priority information, each of the multiple organs is assigned a probability of being selected as the object specimen image as the selection priority.

[0013] Preferably, if the processor determines that a morphological abnormality has occurred, it resets the selection priority of organs in the tissue specimen captured in the object specimen image to a higher level; if the processor determines that no morphological abnormality has occurred, it either does not change the selection priority of organs in the tissue specimen captured in the object specimen image or resets the selection priority to a lower level.

[0014] Preferably, the processor resets the selection priority of organs in the tissue specimen captured in the target specimen image based on the determination result, and also resets the selection priority of related organs that have a functional relationship with the organs in the tissue specimen captured in the target specimen image.

[0015] Preferably, the processor accepts the user's final judgment result on whether a morphological abnormality has actually occurred, and updates the selection priority information based on the final judgment result, in addition to the judgment result.

[0016] Preferably, if the processor determines that no morphological abnormality has occurred but the final determination is that a morphological abnormality has occurred, it resets the selection priority of organs in the tissue specimen captured in the object specimen image to a higher level. If the processor determines that a morphological abnormality has occurred but the final determination is that no morphological abnormality has occurred, it either does not change the selection priority of organs in the tissue specimen captured in the object specimen image or sets the selection priority to a lower level.

[0017] Preferably, the subjects are divided into multiple groups, and in the selection priority information, a selection priority is set for each of the multiple organs and for each of the multiple groups.

[0018] Preferably, the multiple groups include a dosing group that was given the candidate substance and a control group that was not given the candidate substance.

[0019] Preferably, the dosing group includes multiple sub-dosing groups with different dosages of the candidate substance.

[0020] Preferably, the selection priority information in the initial state includes a selection priority based on previously obtained insights.

[0021] Preferably, the processor detects presumed morphologically abnormal portions in a part of the object specimen image that are presumed to have morphological abnormalities, and makes a determination by comparing a value related to the number of presumed morphologically abnormal portions with a pre-set determination threshold.

[0022] Preferably, the processor processes each of the multiple patch images obtained by subdividing the object specimen image as a part, and performs the detection of presumed morphologically abnormal parts by comparing the feature quantity obtained by inputting the patch image into the machine learning model with the reference feature quantity obtained by inputting the reference patch image of the tissue specimen regarded as normal into the machine learning model.

[0023] This invention provides a method for operating a drug development auxiliary device, comprising: acquiring multiple specimen images of tissue specimens from multiple organs of a subject for evaluation tests of candidate substances; selecting a target specimen image of a tissue specimen from the multiple specimen images based on selection priority information that sets a selection priority for each of the multiple organs; determining whether morphological abnormalities have occurred in the tissue specimen captured in the target specimen image; and updating the selection priority information based on the determination result of whether morphological abnormalities have occurred.

[0024] This invention provides a working procedure for a drug development auxiliary device, wherein a computer performs the following processing: acquiring multiple specimen images of tissue specimens from multiple organs of a subject for evaluation tests of candidate substances; selecting a target specimen image of a tissue specimen from the multiple specimen images based on selection priority information that sets a selection priority for each of the multiple organs; determining whether morphological abnormalities have occurred in the tissue specimen captured in the target specimen image; and updating the selection priority information based on the determination result of whether morphological abnormalities have occurred.

[0025] Invention Effects

[0026] According to the technology of the present invention, a drug development auxiliary device, a method for operating the drug development auxiliary device, and a procedure for operating the drug development auxiliary device can be provided, which can prioritize the analysis of specimen images that are considered to have morphological abnormalities in tissue specimens without causing processing load. Attached Figure Description

[0027] Figure 1 It is a diagram showing the procedures, specimen images, and auxiliary devices for drug development in the evaluation test.

[0028] Figure 2 This is a graph showing the treatment group and the control group.

[0029] Figure 3 This is an example of a photograph showing tissue specimens of various organs.

[0030] Figure 4 This is a block diagram representing a computer that constitutes a drug development support device.

[0031] Figure 5 This is a block diagram representing the processing unit of the CPU in a drug development support device.

[0032] Figure 6 It is a graph representing selection probability information.

[0033] Figure 7 This is a diagram showing the processing of the selection section.

[0034] Figure 8 This is a diagram representing patch images obtained by subdividing a specimen image.

[0035] Figure 9 This is a diagram illustrating the extraction of features from a patch image using a feature extractor.

[0036] Figure 10 This is a diagram showing the structure of the feature extractor.

[0037] Figure 11 This is a diagram representing the processing of an autoencoder during the learning phase.

[0038] Figure 12 This is a diagram showing the composition of past control group and learning reference patch images.

[0039] Figure 13 This is a diagram illustrating the case where reference features are extracted from a reference patch image using a feature extractor.

[0040] Figure 14 It represents a graph of reference features drawn in the feature space, and a graph of the detection reference information.

[0041] Figure 15 It is a graph representing the distance between the position of a feature quantity and the representative position of a reference feature quantity.

[0042] Figure 16 This is a diagram showing the processing and testing results of the testing department.

[0043] Figure 17 This is a diagram showing the processing and testing results of the testing department.

[0044] Figure 18This is a diagram showing the processing and decision reference information of the decision-making unit.

[0045] Figure 19 This is a diagram showing the processing and determination results of the determination unit.

[0046] Figure 20 This is a diagram showing the processing and determination results of the determination unit.

[0047] Figure 21 This is a diagram representing the processing of the information update department.

[0048] Figure 22 This diagram represents the scenario where the information update department resets the selection probability to a higher level.

[0049] Figure 23 This is a diagram representing the processing of the information update department.

[0050] Figure 24 This diagram represents the scenario where the information update department resets the selection probability to a lower level.

[0051] Figure 25 This is a diagram representing the analysis and instruction screen.

[0052] Figure 26 This is a diagram representing the screen displaying the analysis results.

[0053] Figure 27 This is a flowchart illustrating the processing steps of a drug development support device.

[0054] Figure 28 This is a flowchart illustrating the processing steps of a drug development support device.

[0055] Figure 29 This is another diagram illustrating the processing of the information update department.

[0056] Figure 30 This is a diagram illustrating a second embodiment in which the selection probability of relevant organs that have a functional relationship with the organs of the tissue specimen captured in the object specimen image is reset.

[0057] Figure 31 This is a diagram of the analysis result display screen and the indicator receiving unit in the third embodiment, which represents the user's input on the final determination of whether a morphological abnormality has actually occurred.

[0058] Figure 32 This diagram illustrates the processing of the information update unit in the third embodiment.

[0059] Figure 33 This diagram illustrates the processing of the information update unit in the third embodiment.

[0060] Figure 34This is a diagram illustrating a fourth embodiment of processing a slide specimen containing tissue specimens to obtain a specimen image. Detailed Implementation

[0061] [First Implementation Method]

[0062] As an example, such as Figure 1 As shown, the drug development aid 10 of the present invention is used for candidate substances 27 of a drug (see reference). Figure 2 The drug development support device 10 is used for evaluating the efficacy and toxicity of the drug. It is, for example, a desktop personal computer equipped with a monitor 11 for displaying various screens, a keyboard, mouse, touch panel, and / or a microphone for voice input, and other input devices 12. The drug development support device 10 is installed in a pharmaceutical research facility and operated by users such as drug development personnel involved in the drug development process. These drug development personnel also include pathologists.

[0063] A specimen image 15 is input into the drug development support device 10. The specimen image 15 is used to evaluate the efficacy and toxicity of candidate substance 27. The specimen image 15 is generated, for example, through the following steps: First, a subject S, such as a rat, prepared for the evaluation of candidate substance 27, is dissected, and organs of the subject S are collected. Here, multiple tissue specimens of the liver LV (hereinafter referred to as liver specimens LVS) are collected in cross-section. Next, after each of the collected liver specimens LVS is pasted onto a glass slide 16, the liver specimens LVS are stained using hematoxylin-eosin. Then, the stained liver specimens LVS are covered with a coverslip 17 to prepare a glass slide specimen 18. Then, the glass slide specimen 18 is placed on a photographic device 19, such as a digital optical microscope, and the photographic device 19 captures the specimen image 15. The entire liver specimen LVS is captured in the resulting specimen image 15. Therefore, specimen image 15 is referred to as WSI (Whole Slide Image). Specimen image 15 is labeled with the subject ID (Identification Data) used to uniquely identify the subject S, the specimen image ID used to uniquely identify specimen image 15, and the date and time of capture. Furthermore, the tissue specimen is also referred to as a tissue section. Additionally, staining can be performed using hematoxylin alone, or using Nuclear Fast Red, etc.

[0064] As an example, such as Figure 2As shown, the subjects S were divided into a treatment group 25 and a control group 26. The treatment group 25 consisted of multiple subjects S to which candidate substance 27 was administered. The treatment group 25 was further divided into a high-dose group 25H, a medium-dose group 25M, and a low-dose group 25L according to the amount of candidate substance 27 administered. By dividing the treatment group 25 into high-dose group 25H, medium-dose group 25M, and low-dose group 25L, the effect of the dosage of candidate substance 27 on the subjects S can be clearly observed. The high-dose group 25H, medium-dose group 25M, and low-dose group 25L are examples of the "sub-dose groups" involved in the technology of this invention. Furthermore, the sub-dose groups are not limited to the three groups of high-dose group 25H, medium-dose group 25M, and low-dose group 25L shown, but may also be the two groups of high-dose group 25H and low-dose group 25L, or may be four or more groups.

[0065] The control group 26, unlike the treatment group 25, consisted of multiple subjects S that were not given candidate substance 27. The number of subjects S constituting the high-dose group 25H, medium-dose group 25M, and low-dose group 25L, and the number of subjects S constituting the control group 26, were, for example, approximately 5 to 10. The subjects S constituting the high-dose group 25H, medium-dose group 25M, and low-dose group 25L, and the subjects S constituting the control group 26, were subjects S with the same attributes and placed in the same rearing environment. Similar attributes include, for example, the same age, the same sex, and / or the same genetic system. The same genetic system includes, for example, the same ancestor five generations ago, and / or the same gene sequence in a specific region. Other similar attributes include the same age ratio, the same sex ratio (e.g., five males and five females), and / or the same genetic system ratio. The same rearing environment includes, for example, the same feed, the same temperature and humidity of the rearing space, and / or the same size of the rearing space. The term "same" in "same breeding environment" means not only completely identical, but also includes the degree of error that is generally permissible in the technical field to which the technology of this invention pertains, and does not violate the spirit of the technology of this invention.

[0066] Since multiple specimen images 15 are obtained from one subject S, the number of specimen images 15 obtained from each group is the number obtained from one subject S multiplied by the number of subjects 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 × 10 = 1000 specimen images 15 are obtained from each group.

[0067] As an example, such as Figure 3 As shown, specimen image 15, except for Figure 1 In addition to the liver specimen LVS shown, tissue specimens from multiple organs of the subject S were also photographed. Figure 3Examples of specimen images include images 15 of a heart (HS), a brain (BS), and a bone marrow (BMS). Specimen image 15 also includes images of tissue specimens from various organs such as the lungs, stomach, small intestine, large intestine, gallbladder, pancreas, spleen, and kidneys (see reference). Figure 6 The organ tissue specimens photographed as specimen images 15 are, for example, about 40 different types. Therefore, the total number of specimen images 15 obtained from a single subject S is approximately several hundred. Hereinafter, the collection of multiple specimen images 15 of multiple organs from multiple subjects S in each group will be referred to as specimen image group 15G (refer to...). Figure 5 ).

[0068] As an example, such as Figure 4 As shown, in addition to the aforementioned display 11 and input device 12, the computer constituting the drug development support device 10 also includes a storage device 30, a memory 31, a CPU (Central Processing Unit) 32, and a communication unit 33. They are interconnected via a bus 34.

[0069] Storage device 30 is a hard disk drive built into or connected via cable or network to the computer constituting the drug development support device 10. Alternatively, storage device 30 is a disk array that connects multiple hard disk drives. Storage device 30 stores control programs such as the operating system, various application programs, and various data associated with these programs. Furthermore, solid-state drives (SSDs) can be used instead of hard disk drives.

[0070] Memory 31 is a working memory used by CPU 32 for processing. CPU 32 loads the program stored in storage device 30 into memory 31 and executes the processing according to the program. Thus, CPU 32 uniformly controls all parts of the computer. CPU 32 is an example of a "processor" according to the technology of this invention. In addition, memory 31 may also be built into CPU 32. Communication unit 33 performs various information transmission control with external devices such as camera device 19.

[0071] As an example, such as Figure 5 As shown, the storage device 30 of the drug development support device 10 stores a working program 40. The working program 40 is an application program used to enable the computer to function as the drug development support device 10. That is, the working program 40 is an example of the "working program of a drug development support device" according to the technology of this invention. The storage device 30 also stores a feature extractor 41, detection reference information 42, a judgment threshold 43, and selection probability information 44, etc. The feature extractor 41 is an example of the "machine learning model" according to the technology of this invention.

[0072] When the working program 40 is started, the CPU 32 and memory 31 of the computer constituting the drug development auxiliary device 10 cooperate to function as a read / write (hereinafter referred to as RW) control unit 50, detection unit 51, judgment unit 52, information update unit 53, and display control unit 54. The RW control unit 50 includes a selection unit 55.

[0073] The RW control unit 50 controls the storage of various data in the storage device 30 and the reading of various data in the storage device 30. For example, the RW control unit 50 acquires a specimen image group 15G from the imaging device 19 and stores the acquired specimen image group 15G in the storage device 30.

[0074] The selection unit 55 of the RW control unit 50 selects a specimen image 15 of a tissue specimen from the specimen image group 15G based on the selection probability information 44. The selection unit 55 outputs the selected specimen image 15 to the detection unit 51 and the display control unit 54. The specimen image 15 output from the selection unit 55 to the detection unit 51, etc., is the object used to determine whether morphological abnormalities have occurred in the tissue specimen. Hereinafter, the specimen image 15 used to determine whether morphological abnormalities have occurred in the tissue specimen will be referred to as the object specimen image 15T. Furthermore, morphological abnormalities refer to lesions that are not visible in normal tissue specimens, such as hyperplasia, infiltration, congestion, inflammation, tumors, carcinogenesis, proliferation, hemorrhage, glycogen deficiency, etc.

[0075] The RW control unit 50 reads the feature extractor 41 and detection reference information 42 from the storage device 30, and outputs the read feature extractor 41 and detection reference information 42 to the detection unit 51. Additionally, the RW control unit 50 reads the determination threshold 43 from the storage device 30 and outputs the read determination threshold 43 to the determination unit 52. Furthermore, the RW control unit 50 reads the selection probability information 44 from the storage device 30 and outputs the read selection probability information 44 to the information update unit 53.

[0076] The detection unit 51 uses the feature extractor 41 and the detection reference information 42 to detect the presumed morphological abnormality portion of the sample image 15T that is presumed to have morphological abnormalities. The detection unit 51 outputs the detection result 60 of the presumed morphological abnormality portion to the determination unit 52 and the display control unit 54.

[0077] Based on the detection result 60, the determination unit 52 determines whether a morphological abnormality has occurred in the tissue specimen captured in the target specimen image 15T. The determination unit 52 outputs the determination result 61 of whether a morphological abnormality has occurred in the tissue specimen captured in the target specimen image 15T to the information update unit 53 and the display control unit 54.

[0078] Based on the judgment result 61, the information update department 53 updates (changes) the selection probability information 44.

[0079] The information update unit 53 outputs the updated selection probability information 44 to the RW control unit 50. The RW control unit 50 writes the updated selection probability information 44 back to the storage device 30.

[0080] The display control unit 54 controls the display of various screens on the display 11. Among these various screens is an analysis instruction screen 90 for inputting analysis instructions (see reference). Figure 25 ), and the analysis results display screen 100 (refer to) Figure 26 In addition to these processing units 50-55, the CPU 32 also includes an instruction receiving unit 114 (see reference 114) that receives various operation instructions from the input device 12. Figure 31 )wait.

[0081] As an example, such as Figure 6 As shown, the selection probability information 44 contains the following information: for each of the high-dose group 25H, medium-dose group 25M, low-dose group 25L, and control group 26, the selection probability of each of multiple organs is registered. The selection probability is the probability that a specimen image 15 of the tissue specimen of that organ in that group is selected as the target specimen image 15T. The selection probability is an example of the "selection priority" involved in the technology of this invention.

[0082] Regarding selection probabilities, for example, the brain in the high-dose group 25H had a selection probability of 5.0%, and the liver LV had a selection probability of 10.0%. Furthermore, the brain and trachea in the medium-dose group 25M, low-dose group 25L, and control group 26 all had selection probabilities of 0.1%. (Illustrations omitted). For other organs in the medium-dose group 25M, low-dose group 25L, and control group 26, the selection probability was also uniformly set to 0.1%. The sum of the selection probabilities for all organs in all groups is 100.0%. Moreover, the lower limit of the selection probability is, for example, 0.1%, and is not set to 0%.

[0083] Figure 6This represents the selection probability information 44 in the initial state before being updated by the information update unit 53. In this initial state selection probability information 44, the selection probability of the high-dose group 25H is set higher than that of the medium-dose group 25M, the low-dose group 25L, and the control group 26. This is based on the following insight 1: Generally, the incidence of morphological abnormalities is higher in the high-dose group 25H. Furthermore, in the initial state selection probability information 44 for the high-dose group 25H, liver LV is set to the highest selection probability of 10.0%. Moreover, the heart is set to 7.5%, the brain and bone marrow to 5.0%, and other organs are uniformly set to 2.0%. This is based on the following insight 2: In past evaluation tests of candidate substance 27 and similar candidate substances, the incidence of morphological abnormalities decreased in the order of liver LV, heart, brain, and bone marrow. Thus, the initial state selection probability information 44 contains selection probabilities based on previously obtained insights. Furthermore, similar candidate substances refer to candidate substances whose composition is similar to that of candidate substance 27.

[0084] Insights can be, as exemplified in Insights 1 and 2, related to groups or organs with a high incidence of morphological abnormalities; conversely, they can be related to groups or organs with a low incidence of morphological abnormalities. In the latter case, the selection probability of groups or organs with a low incidence of morphological abnormalities is set to be relatively low.

[0085] As an example, such as Figure 7 As shown, when selecting object specimen images 15T based on selection probability information 44, the selection unit 55 considers the multi-armed bandit problem as described below. The multi-armed bandit problem involves selecting object specimen images 15T such that the number M of images M determined by the determination unit 52 to have morphological abnormalities in the tissue specimen is maximized among the selected N object specimen images 15T. The selection unit 55 operates to solve the aforementioned multi-armed bandit problem, for example, using well-known sampling methods such as Thompson sampling.

[0086] In the following Figures 8-15 In this example, we will use the case where a liver specimen LVS image 15 is selected as the subject specimen image 15T for illustration.

[0087] As an example, such as Figure 8As shown, the detection unit 51 uses well-known image recognition technology to identify the liver specimen LVS captured in the object specimen image 15T, and subdivides the identified liver specimen LVS into multiple patch images 70. The patch images 70 have a pre-set size that can be processed by the feature extractor 41. Furthermore, the patch images 70 have a size that covers not only the morphologically abnormal parts but also their surrounding areas. The detection unit 51 labels the patch images 70 with patch image ID 85 (see reference). Figure 16 (etc.). Additionally, the detection unit 51 will store the location information 86 of the patch image 70 (refer to...). Figure 16 (etc.) is associated with patch image ID 85, and the location information 86 of patch image 70 indicates which position in the object specimen image 15T was cut out from by patch image 70. Patch image 70 is an example of a "part of the object specimen image" involved in the technology of this invention. Furthermore, in Figure 8 In the image, adjacent patch images 70 do not have overlapping regions, but adjacent patch images 70 may also have overlapping regions.

[0088] As an example, such as Figure 9 As shown, the detection unit 51 uses the feature extractor 41 to extract feature quantities 72 from the multiple patch images 70 obtained by subdividing the object specimen image 15T. Therefore, the number of feature quantities 72 is the same as the number of patch images 70.

[0089] As an example, such as Figure 10 As shown, the encoder section 76 of the autoencoder 75 is converted into a feature extractor 41. In addition to the encoder section 76, the autoencoder 75 also has a decoder section 77. A patch image 70 is input to the encoder section 76. The encoder section 76 converts the patch image 70 into feature values ​​72. The encoder section 76 then passes the feature values ​​72 to the decoder section 77. The decoder section 77 generates a restored image 78 of the patch image 70 based on the feature values ​​72.

[0090] As is well known, the encoder unit 76 has convolutional layers that perform convolutional processing using filters, and pooling layers that perform pooling processing such as max pooling. The decoder unit 77 is the same. The encoder unit 76 extracts feature quantity 72 by repeatedly performing convolutional processing based on convolutional layers and pooling processing based on pooling layers on the input patch image 70. The extracted feature quantity 72 represents the shape and texture features of the liver specimen LVS captured in the patch image 70.

[0091] Feature 72 is a set of multiple values. That is, feature 72 is multidimensional data. The dimension of feature 72 may be, for example, 512, 1024, or 2048. Feature 72, and the reference feature 72R (described later) Figure 13Having the same dimension, they can be in the same feature space 81 (refer to...) Figure 14 (etc.) are compared.

[0092] As an example, such as Figure 11 As shown, during the learning phase before the encoder unit 76 is converted into a feature extractor 41, the autoencoder 75 is trained by the input learning reference patch image 70RL. The autoencoder 75 outputs a learning restored image 78L in response to the input of the learning reference patch image 70RL. Based on these learning reference patch images 70RL and the learning restored image 78L, a loss calculation using a loss function is performed on the autoencoder 75. Then, based on the result of the loss calculation, various coefficients of the autoencoder 75 (such as the coefficients of the filters in the convolutional layers) are updated, and the autoencoder 75 is updated according to the updated settings.

[0093] During the learning phase of the autoencoder 75, the following series of processes are repeated: inputting the learning reference patch image 70RL to the autoencoder 75, outputting the learning restored image 78L from the autoencoder 75, loss calculation, setting update, and updating the autoencoder 75, while changing the learning reference patch image 70RL. This series of processes continues until the restoration accuracy from the learning reference patch image 70RL to the learning restored image 78L reaches a preset level. The encoder section 76 of the autoencoder 75, having achieved the preset restoration accuracy, is stored as a feature extractor 41 in the storage device 30 of the drug development support device 10. Alternatively, the learning can end when the above series of processes has been repeated a set number of times, regardless of the restoration accuracy from the learning reference patch image 70RL to the learning restored image 78L.

[0094] The learning of such an autoencoder 75 can be performed by the drug development aid 10 or by a device different from the drug development aid 10. In the latter case, the feature extractor 41 is sent from another device to the drug development aid 10, and the feature extractor 41 is stored in the storage device 30 by the RW control unit 50.

[0095] As an example, such as Figure 12As shown, the reference patch image 70RL is provided from multiple reference patch images 70R obtained by subdividing the reference specimen image 15R. The reference specimen image 15R is an image of the liver specimen LVS of the subject S in the past control group 26P. The past control group 26P consists of multiple subjects S that were not administered the candidate substance in past evaluation trials. Therefore, the number of subjects S constituting the past control group 26P is significantly greater than the number of subjects S constituting the drug administration group 25 and the control group 26, for example, by several hundred to several thousand. The reference specimen image 15R is also obtained from multiple subjects S in the same way as specimen image 15, thus the reference specimen image 15R is obtained from the past control group 26P by multiplying the number of subjects S by the number of reference specimens. 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 invention. Alternatively, in addition to using the specimen image 15 of the liver specimen LVS of the subject S in the past control group 26P, a specimen image 15 of the liver specimen LVS of the past drug administration group, which was judged to be normal by pathologists and other experts, can also be used as the reference specimen image 15R.

[0096] Next, the composition of the detection reference information 42 will be explained. First, as an example, such as Figure 13 As shown, using feature extractor 41, multiple reference features 72R are extracted from multiple reference patch images 70R, which are based on all of the multiple reference specimen images 15R.

[0097] As an example, Figure 14 The diagram 80 shown is plotted in the feature space 81. Figure 13 A graph of multiple reference feature quantities 72R extracted from the data. Detection reference information 42 includes the coordinates (hereinafter referred to as representative position coordinates) 82 of the representative position of the reference feature quantity 72R, indicated by an × symbol, in the feature quantity space 81. The representative position is, for example, the center point or average point of the distribution 83 of the reference feature quantity 72R. Additionally, detection reference information 42 also includes a detection threshold 84. Furthermore, in... Figure 14 For ease of explanation, the feature space 81 is assumed to be two-dimensional with axes D1 and D2, but the actual feature space 81 has 512 dimensions as described above. The following... Figure 15 Similarly, for ease of explanation, the dimension of the feature space 81 is represented by two dimensions.

[0098] Similar to the learning process of the automatic encoder 75, the representative position coordinates 82 of the detection reference information 42 can be derived from the drug development aid 10 or from a device different from the drug development aid 10. In the latter case, the representative position coordinates 82 are sent from another device to the drug development aid 10, and the representative position coordinates 82 are stored in the storage device 30 by the RW control unit 50.

[0099] As an example, such as Figure 15 As shown, the detection unit 51 calculates the distance D between the representative position of the reference feature quantity 72R, represented by the representative position coordinates 82 of the detection reference information 42, and the position of the feature quantity 72 in the feature quantity space 81. The detection unit 51 calculates the distance D of the extracted feature quantities 72 for each of the multiple patch images 70 obtained by subdividing a target specimen image 15T. The distance D is the Mahalanobis distance. The distance D represents the deviation of the feature quantity 72 from the reference feature quantity 72R, and more specifically, it represents the deviation of the liver specimen LVS in the patch image 70 taken from the liver specimen LVS considered normal. That is, it can be said that the larger the distance D, the more the liver specimen LVS taken in the patch image 70 deviates from the liver specimen LVS considered normal. Therefore, the larger the distance D, the higher the probability of morphological abnormalities occurring in the liver specimen LVS taken in the patch image 70.

[0100] Furthermore, the distance D can be calculated as any one of the mean, median, or maximum Euclidean distances between the positions of the k-nearest samples of the distribution 83 of the reference feature quantity 72R and the positions of the feature quantity 72. Alternatively, instead of distance D, the cosine similarity between the vector representing the representative position of the reference feature quantity 72R and the vector representing the position of the feature quantity 72 can be calculated from 1.0. The cosine similarity takes values ​​between -1.0 and 1.0; in other words, the larger the value, the more similar the vectors are in direction. Moreover, the negative log-likelihood or other likelihood functions can be calculated as the deviation instead of distance D.

[0101] (Illustrations omitted) For each organ, a feature extractor 41 and detection reference information 42 are prepared. The RW control unit 50 reads the feature extractor 41 and detection reference information 42 corresponding to the organs of the tissue specimen captured in the target specimen image 15T from the storage device 30 and outputs them to the detection unit 51. The detection unit 51 uses the feature extractor 41 and detection reference information 42 corresponding to the organs of the tissue specimen captured in the target specimen image 15T to perform the processing described above.

[0102] As an example, such as Figure 16 and Figure 17As shown, the detection unit 51 compares the calculated distance D with the detection threshold 84. Figure 16 As shown, when the distance D is below the detection threshold 84, it is detected that no morphological abnormalities have occurred in the tissue specimen captured in the patch image 70. The detection unit 51 outputs a detection result 60 indicating that no morphological abnormalities have occurred in the tissue specimen captured in the patch image 70. The detection result 60 at this time includes the patch image ID 85 and location information 86.

[0103] On the other hand, such as Figure 17 As shown, when the distance D is greater than or equal to the detection threshold 84, the detection unit 51 detects that a morphological abnormality has occurred in the tissue specimen captured in the patch image 70. The detection unit 51 outputs a detection result 60 indicating that a morphological abnormality has occurred in the tissue specimen captured in the patch image 70. The detection result 60 at this time includes feature quantity 72 in addition to the patch image ID 85 and position information 86. Thus, the portion of the patch image 70 that is detected as having a morphological abnormality in the tissue specimen corresponds to the "presumed morphological abnormality portion" involved in the technology of the present invention. Furthermore, the detection threshold 84 may be common in the high-dose group 25H, the medium-dose group 25M, the low-dose group 25L, and the control group 26, or it may be different in these groups. In addition, the detection threshold 84 may be common in multiple organs, or it may be different in these organs. Moreover, instead of the distance D, the detection of whether a morphological abnormality has occurred in the tissue specimen captured in the patch image 70 may be determined by comparing the cosine similarity or likelihood function described above with the detection threshold 84.

[0104] like Figure 12 and Figure 13 As shown, the reference characteristic 72R is a tissue specimen of subject S from the previous control group 26P that will be photographed (in... Figure 12 and Figure 13 The feature quantities extracted from the reference patch image 70R, obtained by subdividing the reference specimen image 15R (liver specimen LVS), are as follows. Because the subject S in the previous control group 26P was a subject S that was not given the candidate substance, at least no morphological abnormalities caused by the toxicity of the candidate substance occurred in the tissue specimen captured in the reference specimen image 15R. Therefore, the representative position of the reference feature quantity 72R is considered as the representative position of the feature quantity in the specimen image 15 capturing a normal tissue specimen. Therefore, as described above, the distance D between the representative position of the reference feature quantity 72R and the position of feature quantity 72 becomes an indicator of how much the tissue specimen captured in the patch image 70 deviates from the normal tissue specimen. Therefore, as... Figure 16 As shown, the detection unit 51 detects patch images 70 whose distance D is less than the detection threshold 84 as tissue specimens that have not deviated from normal tissue specimens, i.e., no morphological abnormalities have occurred. On the other hand, as Figure 17 As shown, the detection unit 51 detects the patch image 70, which is at a distance of D greater than the detection threshold 84, as a tissue specimen that deviates from the normal tissue specimen, i.e., morphological abnormality has occurred.

[0105] The detection results 60 of all patch images 70 are input to the determination unit 52. The determination unit 52 counts the number of detection results 60 indicating morphological abnormalities in the tissue specimens captured in the patch images 70, i.e., the number of presumed morphologically abnormal portions. Then, by dividing the count by the total number of patch images 70, the ratio of the number of patch images 70 detected as having morphological abnormalities, i.e., the ratio of the number of presumed morphologically abnormal portions, is calculated. As an example, such as Figure 18 As shown, the determination unit 52 derives the calculated number ratio as determination reference information 88. The number ratio is an example of a "value related to the number of presumed morphologically abnormal parts" involved in the technology of this invention. Alternatively, the number of presumed morphologically abnormal parts may be derived as determination reference information 88 instead of the number ratio.

[0106] As an example, such as Figure 19 and Figure 20 As shown, the determination unit 52 compares the ratio of the number of determination reference information 88 with the magnitude of the determination threshold 43. Figure 19 As shown, when the ratio of the number of reference information 88 is greater than or equal to the determination threshold 43, the determination unit 52 determines that a morphological abnormality has occurred in the tissue specimen captured in the object specimen image 15T, and outputs the determination result 61 of this intention.

[0107] On the other hand, such as Figure 20 As shown, when the ratio of the number of reference information 88 is lower than the determination threshold 43, the determination unit 52 determines that no morphological abnormalities have occurred in the tissue specimen captured in the target specimen image 15T, and outputs the determination result 61 indicating this. Furthermore, the determination threshold 43 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 it may differ among these groups. Additionally, the determination threshold 43 may be common to multiple organs, or it may differ among multiple organs.

[0108] As an example, such as Figure 21 and Figure 22 As shown, if the determination result 61 indicates that a morphological abnormality has occurred in the tissue specimen captured in the target specimen image 15T, the information update unit 53 resets the organ selection probability of the tissue specimen captured in the target specimen image 15T to a higher level. Figure 22In the example, an image 15T of the liver specimen LVS of the subject S in the high-dose group 25H, captured in the determination unit 52, is determined to have morphological abnormalities. Furthermore, an example is shown where the selection probability of the liver LV in the high-dose group 25H is reset to 11.0%, which is obtained by increasing the probability by 10% (+1.0%) from 10.0%.

[0109] On the other hand, as an example, such as Figure 23 and Figure 24 As shown, if the determination result 61 indicates that no morphological abnormalities have occurred in the tissue specimen captured in the target specimen image 15T, the information update unit 53 resets the organ selection probability of the tissue specimen captured in the target specimen image 15T to a lower value. Figure 24 In one example, the determination unit 52 determines that the tracheal specimen image 15T of the subject S in the high-dose group 25H, which was captured, does not show any morphological abnormalities. Furthermore, another example is the case where the selection probability of the trachea in the high-dose group 25H is reset to 1.8%, obtained by reducing it by 10% (-0.2%) from 2.0%. Here, the degree of increase or decrease in the selection probability (the difference between the selection probability before and after the setting) can be constant regardless of the group or organ, as exemplified, such as 10% of the selection probability before the setting, or it can be different between each group and / or each organ.

[0110] In addition, such as Figure 21 and Figure 22 Thus, if the selection probability of a certain organ OA in a certain group of GA is reset to be higher, the information update unit 53 resets the selection probability of other organs OB in that group of GA to be lower in order to achieve a total selection probability of 100.0%. On the other hand, as Figure 23 and Figure 24 Therefore, if the selection probability of a certain organ OC in a certain group of GC is reset to a lower value, the information update unit 53 similarly resets the selection probability of other organs OD in that group of GC to a higher value in order to achieve a total selection probability of 100.0%. Organs OB and OD can be selected randomly, or organs with a weak relationship to organs OA and OC can be selected. Organs with a weak relationship are, for example, organs from different organ systems such as the digestive system, circulatory system, urinary system, reproductive system, and musculoskeletal system. For example, when organ OA is the stomach in the digestive system, the heart in the circulatory system is selected as organ OB. Organ OB and OD can be one organ or multiple organs. Furthermore, organ OB and OD can also be selected from groups other than group GA and group GC.

[0111] As an example, such as Figure 25As shown, for example, when user U starts the working procedure 40, the display control unit 54 controls the display of the analysis instruction screen 90 on the display 11. The analysis instruction screen 90 includes a drop-down menu 91 for selecting an evaluation test and an analysis button 92. After user U selects the desired evaluation test from the drop-down menu 91, they select the analysis button 92. This performs the following: the selection unit 55 selects the object specimen image 15T; the detection unit 51 detects the presumed morphological abnormalities; the determination unit 52 determines whether a morphological abnormality has occurred; and the information update unit 53 updates the selection probability information 44.

[0112] When the processing of each of the above-mentioned processing units is completed, for example, the display control unit 54 displays the information on the display 11. Figure 26 The analysis result display screen 100 is controlled as shown. The object specimen image 15T is displayed on the analysis result display screen 100. The object specimen image 15T can be moved in its display position on the analysis result display screen 100. Furthermore, the object specimen image 15T can be zoomed in, zoomed out, and rotated on the analysis result display screen 100. As shown in the cross-section, the display control unit 54 displays in color the portion of the patch image 70 detected in the detection unit 51 as a presumed morphological abnormality. Additionally, the display can be performed such that the greater the difference between the distance D and the detection threshold 84, the darker the color.

[0113] A display area 101 is provided at the bottom of the specimen image 15T. The display area 101 displays a message indicating the determination result 61 of the determination unit 52. If the determination result 61 indicates that a morphological abnormality has occurred in the tissue specimen captured in the specimen image 15T, as shown in the figure, a message further urging the user U to observe the specimen image 15T in detail is added. Additionally, the display area 101 also displays the ratio of the estimated number of morphologically abnormal portions calculated by the determination unit 52.

[0114] An image return button 102 and an image forward button 103 are also provided at the lower part of the display area 101. When the user U wants to return to the previous object specimen image 15T, he / she selects the image return button 102. Conversely, when the user U wants to move to the next object specimen image 15T, he / she selects the image forward button 103. When the image forward button 103 is selected, the selection unit 55 selects the object specimen image 15T again, the detection unit 51 detects the presumed morphological abnormality, the determination unit 52 determines whether a morphological abnormality has occurred, and the information update unit 53 updates the selection probability information 44. The display control unit 54 updates the display of the analysis result display screen 100 to the content corresponding to each process. In addition, when the end button 104 is selected, the display control unit 54 turns off the display of the information display screen 100.

[0115] Next, regarding the function of the above structure, as an example, refer to... Figure 27 and Figure 28 The flowchart shown will be used for explanation. First, when the working procedure 40 is started in the drug development support device 10, as follows... Figure 5 As shown, the CPU 32 of the drug development support device 10 functions as the RW control unit 50, detection unit 51, judgment unit 52, information update unit 53, and display control unit 54. The RW control unit 50 includes a selection unit 55.

[0116] The imaging device 19 captures specimen images 15 of tissue specimens from multiple organs of multiple subjects S in multiple groups. The resulting multiple specimen images 15, i.e., specimen image sets 15G, are output from the imaging device 19 to the drug development support device 10. In the drug development support device 10, the specimen image sets 15G from the imaging device 19 are acquired by the RW control unit 50 and stored in the storage device 30 (step ST100).

[0117] Under the control of the display control unit 54, the display is shown on the monitor 11. Figure 25 The analysis instruction screen 90 shown is shown (step ST105). User U selects the desired evaluation test from the drop-down menu 91. If the analysis button 92 is selected ("YES" in step ST110), as shown... Figure 7 As shown, the selection unit 55 selects a target specimen image 15T of a tissue specimen of an organ from the plurality of specimen images 15 constituting the specimen image group 15G based on the selection probability information 44 (step ST115). The target specimen image 15T is output from the selection unit 55 to the detection unit 51 and the display control unit 54.

[0118] The RW control unit 50 reads the feature extractor 41 and detection reference information 42 corresponding to the organs of the tissue specimen captured in the target specimen image 15T from the storage device 30, and outputs the read feature extractor 41 and detection reference information 42 to the detection unit 51.

[0119] like Figure 8 As shown, in the detection unit 51, the object specimen image 15T is subdivided into multiple patch images 70. Then, as... Figure 9 As shown, in the detection unit 51, feature quantity 72 is extracted from the patch image 70 using feature quantity extractor 41.

[0120] like Figure 15 As shown, in the detection unit 51, the distance D between the representative position of the reference feature quantity 72R and the position of the feature quantity 72 is calculated. Then, as... Figure 16 and Figure 17As shown, in the detection unit 51, the distance D and the detection threshold 84 are compared to detect whether morphological abnormalities have occurred in the tissue specimen captured in the patch image 70 (step ST120). The detection result 60 of whether morphological abnormalities have occurred in the tissue specimen captured in the patch image 70 is output from the detection unit 51 to the determination unit 52.

[0121] The process involves extracting feature quantity 72 from all patch images 70 of the object specimen image 15T, calculating the distance D between the representative position of the reference feature quantity 72R and the position of the feature quantity 72, and detecting whether morphological abnormalities have occurred in the tissue specimens captured in the patch images 70. After performing the above processing on all patch images 70, the process proceeds to step ST125.

[0122] like Figure 18 As shown, in the determination unit 52, determination reference information 88 is derived based on the detection result 60 (step ST125). Then, in the determination unit 52, based on the determination threshold 43 and the determination reference information 88, it is determined whether a morphological abnormality has occurred in the tissue specimen captured in the object specimen image 15T (step ST130). More specifically, as... Figure 19 and Figure 20 As shown, the ratio of the number of presumed morphological abnormalities contained in the determination reference information 88 is compared with the determination threshold 43. If the ratio is greater than or equal to the determination threshold 43 ("YES" in step ST135), the process proceeds to step ST140. On the other hand, if the ratio is less than or equal to the determination threshold 43 ("NO" in step ST135), the process proceeds to step ST140. Figure 28 Step ST150 in the process.

[0123] In step ST140, as Figure 19 As shown, in the determination unit 52, it is determined that a morphological abnormality has occurred in the tissue specimen captured in the object specimen image 15T. The determination result 61 is output from the determination unit 52 to the information update unit 53 and the display control unit 54. In this case, as... Figure 21 and Figure 22 As shown, the information update unit 53 resets the organ selection probability of the tissue specimen captured in the object specimen image 15T to a higher level (step ST145). The process proceeds to step ST160.

[0124] On the other hand, in step ST150, such as Figure 20 As shown, in the determination unit 52, it is determined that no morphological abnormalities have occurred in the tissue specimen captured in the object specimen image 15T. The determination result 61 is output from the determination unit 52 to the information update unit 53 and the display control unit 54. In this case, as... Figure 23 and Figure 24 As shown, the information update unit 53 resets the organ selection probability of the tissue specimen captured in the object specimen image 15T to a lower value (step ST155). Similar to step ST145, the process proceeds to step ST160.

[0125] In step ST160, under the control of the display control unit 54, the display is shown on the display 11. Figure 26 The analysis results are displayed on screen 100. Thus, the object specimen image 15T, the detection results of the presumed morphologically abnormal parts 60, and the determination results of whether morphological abnormalities have occurred 61 are available for user U to view.

[0126] In the analysis results display screen 100, if user U selects the image advance button 103 (which is "YES" in step ST165), the series of processes following step ST115 are performed again. Specifically, in step ST115, the next object specimen image 15T is selected based on the selection probability information 44 updated by the information update unit 53. The analysis results display screen 100 continues to be displayed until user U selects the end button 104 (which is "NO" in step ST170).

[0127] As described above, the CPU 32 of the drug development support device 10 includes an RW control unit 50, a selection unit 55, a determination unit 52, and an information update unit 53. The RW control unit 50 acquires multiple specimen images 15 (specimen image group 15G) of tissue specimens from multiple organs of the subject S for evaluation tests of candidate substances 27 used in the drug formulation. The selection unit 55 selects a target specimen image 15T of a tissue specimen from the multiple specimen images 15, based on selection probability information 44, which sets a selection probability for each of the multiple organs. The determination unit 52 determines whether morphological abnormalities have occurred in the tissue specimen captured in the target specimen image 15T. The information update unit 53 updates the selection probability information 44 based on the determination result 61 of whether morphological abnormalities have occurred.

[0128] Based on the determination result 61 of whether morphological abnormalities have occurred, the selection probability information 44 is appropriately updated. Therefore, the selected specimen image 15T based on this selection probability information 44 is also an image with a high probability of morphological abnormalities occurring in the tissue specimen. Thus, it is possible to prioritize the analysis of specimen images 15 that are considered to have morphological abnormalities in the tissue specimen without incurring processing load. As a result, the burden on user U to analyze a large number of specimen images 15 can be reduced.

[0129] like Figure 6As shown, the selection priority information assigns a probability to each of the multiple organs that it will be selected as the target specimen image 15T, i.e., selection probability information 44. Therefore, the selection probability information 44 can be easily updated by simply increasing or decreasing the selection probability.

[0130] Furthermore, the selection priority is not limited to the illustrated selection probability, but can also be an order. In this case, the information update unit 53 updates the selection priority information by increasing or decreasing the order based on the determination result 61.

[0131] like Figure 21 and Figure 22 As shown, when the determination result 61 indicates a morphological abnormality, the information update unit 53 resets the selection probability of organs in the tissue specimen captured in the target specimen image 15T to a higher level. On the other hand, as... Figure 23 and Figure 24 As shown, when the determination result 61 indicates that no morphological abnormality has occurred, the information update unit 53 resets the selection probability of organs in the tissue specimen captured in the target specimen image 15T to a lower value. Therefore, the selection probability of organs with morphological abnormalities gradually increases, while the selection probability of organs without morphological abnormalities gradually decreases. Thus, the probability of selecting specimen images 15 considered to have morphological abnormalities in tissue specimens as target specimen images 15T can be further increased. As a result, the burden on user U to analyze a large number of specimen images 15 can be further reduced.

[0132] like Figure 2 As shown, the subjects S were divided into multiple groups. More specifically, the multiple groups included a treatment group 25 which was administered candidate substance 27, and a control group 26 which was not administered candidate substance 27. Furthermore, the treatment group 25 included a high-dose group 25H, a medium-dose group 25M, and a low-dose group 25L, with different dosages of candidate substance 27. Additionally, as... Figure 6 As shown, in the selection probability information 44, a selection probability is set for each of the multiple organs and for each of the multiple groups. Therefore, for example, the selection probability of the high-dose group 25H can be set higher than the selection probability of other groups, and the selection probability can be set according to the group. Therefore, the probability of selecting the specimen image 15, which is considered to have morphological abnormalities in the tissue specimen, as the target specimen image 15T can be further increased.

[0133] like Figure 6As shown, the selection probability information 44 in the initial state contains a selection probability based on previously obtained insights. Therefore, it is possible to select the specimen image 15 that matches the insights as the target specimen image 15T from the very beginning. Furthermore, since the selection probability based on the insights is updated, even in situations other than the initial state, it is possible to select the specimen image 15 that matches the insights to some extent as the target specimen image 15T.

[0134] Furthermore, the selection probability information 44 in the initial state can be uniformly set to the same value, regardless of the group or organ. In this case, the object specimen image 15T initially selected by the selection unit 55 is random. Alternatively, the selection of the initial object specimen image 15T can also be left to the user U.

[0135] like Figure 16 and Figure 17 As shown, the detection unit 51 detects portions of the object specimen image 15T that are presumed to have morphological abnormalities in the tissue specimen. For example... Figure 19 and Figure 20 As shown, the determination unit 52 determines whether a morphological abnormality has occurred in the tissue specimen by comparing a value related to the number of presumed morphologically abnormal portions, i.e., the number ratio, with a preset determination threshold 43. Therefore, it is possible to determine with higher accuracy whether a morphological abnormality has occurred in the tissue specimen in the object specimen image 15T. Furthermore, the determination criteria are clear, eliminating the possibility of determination errors. In addition, a machine learning model that outputs a determination result 61 based on the input of the object specimen image 15T can be used to determine whether a morphological abnormality has occurred in the tissue specimen captured in the object specimen image 15T.

[0136] like Figure 8 , Figure 9 ,and Figure 15 As shown, the detection unit 51 processes each of the multiple patch images 70 obtained by subdividing the object specimen image 15T as a part. The detection unit 51 performs the detection of presumed morphologically abnormal parts by comparing the feature quantity 72 obtained by inputting the patch image 70 into the feature quantity extractor 41 with the reference feature quantity 72R obtained by inputting the reference patch image 70R, which is regarded as a normal tissue specimen, into the feature quantity extractor 41. Therefore, the detection of presumed morphologically abnormal parts can be performed easily and with high accuracy.

[0137] (Modified Example)

[0138] The information update unit 53, in cases where no morphological abnormalities have occurred in the tissue specimens captured in the target specimen image 15T (judgment result 61), processes the data by resetting the organ selection probability of the tissue specimens captured in the target specimen image 15T to a lower value, but is not limited to this. For example, such as... Figure 29 As shown, the information update unit 53 can also be configured to not change the organ selection probability of the tissue specimen captured in the target specimen image 15T if the determination result 61 is that no morphological abnormalities have occurred in the tissue specimen captured in the target specimen image 15T. Not changing the selection probability means not updating the selection probability information 44. In this case, the selection unit 55 selects the next target specimen image 15T based on the same selection probability information 44 as before. In this way, the processing of updating the selection probability information 44 can be omitted, thus further reducing the processing load.

[0139] The opportunity for user U to select object specimen image 15T can also be set during the selection unit 55's selection of object specimen image 15T based on selection probability information 44. Furthermore, the selection of object specimen image 15T by selection unit 55 can be restricted as follows: Object specimen image 15T is not selected from the medium-dose group 25M and the low-dose group 25L until a morphological abnormality is determined to have occurred in the object specimen image 15T of the tissue specimen of subject S in the high-dose group 25H.

[0140] [Second Implementation]

[0141] In the first embodiment described above, only the selection probability of organs in the tissue specimens captured in the object specimen image 15T was reset, but it is not limited to this. As an example, such as... Figure 30 As shown in the second embodiment, in addition to the organs of the tissue specimen captured in the target specimen image 15T, the selection probability of related organs that have a functional relationship with the organs of the tissue specimen captured in the target specimen image 15T can also be reset. Figure 30 The example illustrates a case where, in the determination unit 52, the liver specimen image 15T of the subject S (LVS) from the high-dose group 25H is determined to have morphological abnormalities. Furthermore, it illustrates a case where the selection probability of the liver from the high-dose group 25H is reset to 11.0%, which is obtained by increasing the probability by 10% (+1.0%) from 10.0%, and the selection probabilities of the esophagus, stomach, small intestine, large intestine, bile duct, and pancreas from the high-dose group 25H are reset to 2.2%, which is obtained by increasing the probability by 10% (+0.2%) from 2.0%.

[0142] The liver, along with the esophagus, stomach, small intestine, large intestine, bile duct, and pancreas, belongs to the digestive system and is a functionally related organ. Other related organs, besides the digestive system, include: the circulatory system (trachea, lungs, heart, aorta, veins, and lymphatic vessels); the urinary system (kidneys, ureters, and bladder); the reproductive system (testes or ovaries and reproductive organs); and the musculoskeletal system (femur, pectoral muscles, and bone marrow). Furthermore, omitting illustrations, if the determination unit 52 determines that no morphological abnormality has occurred, the information update unit 53 resets the selection probability of the organs and related organs of the tissue specimen captured in the target specimen image 15T to a lower value.

[0143] Based on experience, when a morphological abnormality occurs in an organ, the probability that the abnormality will also affect related organs is relatively high. Therefore, as in this second embodiment, if the selection probability of related organs that have a functional relationship with the organs of the tissue specimens captured in the target specimen image 15T is reset in addition to the organs of the tissue specimens captured in the target specimen image 15T, the probability of selecting the specimen image 15, which is considered to have morphological abnormalities in the tissue specimen, as the target specimen image 15T can be further increased.

[0144] [Third Implementation Method]

[0145] As an example, such as Figure 31 As shown, the analysis result display screen 110 of the third embodiment also has a display area 111 at the lower part of the display area 101. In the display area 111, a final determination result 120 is displayed, prompting the user U to input whether a morphological abnormality has actually occurred in the tissue specimen captured in the object specimen image 15T (see reference). Figure 32 and Figure 33 The message is displayed. In addition, in the display area 111, a first input button 112 and a second input button 113 are displayed for the user U to input the final judgment result 120.

[0146] Based on the observation of the object specimen image 15T, if it is determined that a morphological abnormality has occurred in the tissue specimen captured in the object specimen image 15T, the user U selects the first input button 112. Conversely, if it is determined that no morphological abnormality has occurred in the tissue specimen captured in the object specimen image 15T, the user U selects the second input button 113. The selection indication of the first input button 112 and the second input button 113, i.e., the final determination result 120, is received by the indication receiving unit 114.

[0147] As an example, such as Figure 32 and Figure 33As shown, in addition to the judgment result 61, the information update unit 53 also updates the selection probability information 44 based on the final judgment result 120. More specifically, as... Figure 32 As shown, if the determination result 61 indicates that no morphological abnormalities have occurred in the tissue specimen captured in the target specimen image 15T, but the final determination result 120 indicates that morphological abnormalities have occurred in the tissue specimen captured in the target specimen image 15T, the information update unit 53 resets the organ selection probability of the tissue specimen captured in the target specimen image 15T to a higher level. On the other hand, as... Figure 33 As shown, if the determination result 61 indicates that morphological abnormalities have occurred in the tissue specimen captured in the target specimen image 15T, but the final determination result 120 indicates that no morphological abnormalities have occurred in the tissue specimen captured in the target specimen image 15T, the information update unit 53 resets the selection probability of the organ in the tissue specimen captured in the target specimen image 15T to a lower value. That is, the information update unit 53 prioritizes the user U's final determination result 120 more than the determination result 61 of the determination unit 52, and updates the selection probability information 44 accordingly. This allows the user U's final determination result 120 to be reflected in the update of the selection probability information 44. A selection probability that better matches the user U's determination can be set.

[0148] Furthermore, omitting illustrations, when both judgment result 61 and final judgment result 120 indicate morphological abnormalities in the tissue specimens captured in the target specimen image 15T, the information update unit 53 resets the organ selection probability of the tissue specimens captured in the target specimen image 15T to a higher level. Conversely, also omitting illustrations, when both judgment result 61 and final judgment result 120 indicate no morphological abnormalities in the tissue specimens captured in the target specimen image 15T, the information update unit 53 resets the organ selection probability of the tissue specimens captured in the target specimen image 15T to a lower level.

[0149] Even if morphological abnormalities occur in the tissue specimen captured in image 15T (judgment result 61), but no morphological abnormalities occur in the tissue specimen captured in image 15T (final judgment result 120), this method can still be applied. Figure 29 The variation shown does not change the organ selection probability of the tissue specimen captured in the object specimen image 15T.

[0150] The groups for which the selection probability is set can be either the drug treatment group 25 or the control group 26. Alternatively, the groups for which the selection probability is set can be three groups: a high-dose group 25H, a medium-dose group 25M, and a low-dose group 25L, excluding the control group 26. That is, the control group 26 can be excluded from the options of the subject specimen image 15T.

[0151] In addition to capturing a reference patch image 70R of a tissue specimen considered normal, a patch image 70 of a tissue specimen exhibiting morphological abnormalities can be used as a learning reference patch image 70RL. The patch image 70 of the tissue specimen exhibiting morphological abnormalities is obtained, for example, from a past drug administration group consisting of multiple subjects S who were administered the candidate substance in past evaluation trials. Thus, the autoencoder 75 and the feature extractor 41 can learn from tissue specimens with more diverse shape and texture features. As a result, the feature extractor 41 can extract feature quantities 72 that better represent the shape and texture of the tissue specimen.

[0152] Furthermore, the patch image 70 depicting a tissue specimen with morphological abnormalities is not limited to the illustrated patch image 70 obtained from the subject S constituting the previous drug administration group. Morphological abnormalities may also occur in the subject S constituting the previous control group 26P. Therefore, as long as the patch image 70 depicts a tissue specimen with morphological abnormalities, the subject S can be either the previous control group 26P or the previous drug administration group. Moreover, the patch image 70 depicting a tissue specimen with morphological abnormalities can also be an image obtained from a subject S that has been intentionally subjected to various pressures to induce morphological abnormalities. Additionally, the patch image 70 depicting a tissue specimen with morphological abnormalities can also be an artificially created image by processing a patch image 70 depicting a normal tissue specimen.

[0153] Alternatively, the encoder section 76 of the autoencoder 75 can be replaced by a feature extractor 41, which is a convolutional neural network that outputs a classification result based on the input of the patch image 70. The classification result is, for example, the result of identifying the type of a morphological abnormality occurring in the tissue specimen captured in the patch image 70 based on multiple categories such as cell proliferation, infiltration, congestion, and inflammation.

[0154] Furthermore, the machine learning model used as feature extractor 41 is not limited to the illustrated autoencoder 75 and convolutional neural networks. Generators from Generative Adversarial Networks (GANs) can also be used as feature extractors 41. Machine learning models without convolutional layers, such as Vision Transformer (ViT), can also be used as feature extractors 41.

[0155] Contrastive learning can also be performed where features from the same image are brought closer together in the feature space, while features from different images are made farther apart. Examples of contrastive learning methods include SimCLR (A Simple Framework for Contrastive Learning of Visual Representations). Alternatively, self-supervised learning methods such as BYOL (Bootstrap Your Own Latent) can be used, where the different image pairs (also called negative samples) are not used. Furthermore, restrictions can be imposed on the distribution of the extracted features, such as a distribution on a unit sphere or mimicking a standard normal distribution.

[0156] In the analysis results display screen 100 or 110, the group, organ, subject ID, and image ID of the next potentially selected object specimen images 15T can be displayed in advance. Additionally, the estimated time required for the analysis of the remaining portions of each group and / or organ can be displayed. The estimated time can be derived from the number of specimen images 15 previously selected as object specimen images 15T, the total number of specimen images 15 in each group and / or organ, and the average time required for the analysis of one object specimen image 15T.

[0157] [Fourth Implementation Method]

[0158] As an example, such as Figure 34 As shown, in the fourth embodiment, a slide specimen 125 containing tissue specimens of various organs mounted on a slide 16 is processed. Figure 34 The image illustrates a case where a liver specimen (LVS), a heart specimen (HS), a brain specimen (BS), and a bone marrow specimen (BMS) are placed. In this case, the liver specimen (LVS), heart specimen (HS), brain specimen (BS), and bone marrow specimen (BMS) are captured in specimen image 15.

[0159] In addition to functioning as the processing units 50-55 of the first embodiment, the CPU of the drug development support device in the fourth embodiment also functions as the recognition unit 126. The recognition unit 126 identifies the tissue specimens of each organ from the specimen image 15, for example, using a template for identifying tissue specimens of each organ or a machine learning model. The recognition unit 126 outputs the coordinate information of the frames 127-130 surrounding the tissue specimens of each organ as the recognition result. Frame 127 is the frame surrounding the heart specimen HS, and frame 128 is the frame surrounding the liver specimen LVS. Furthermore, frame 129 is the frame surrounding the brain specimen BS, and frame 130 is the frame surrounding the bone marrow specimen BMS.

[0160] The detection unit 51 uses a dedicated feature extractor 41 to extract feature quantities 72 from each tissue specimen within each frame 127-130. That is, the detection unit 51 extracts the feature quantities 72 from each tissue specimen of each organ identified by the identification unit 126. Subsequent processing is the same as that shown in the first embodiment described above; therefore, illustrations and descriptions are omitted.

[0161] Thus, in the fourth embodiment, the specimen image 15 is an image obtained by photographing a slide specimen 125 containing tissue specimens of multiple organs. The recognition unit 126 identifies the tissue specimens of each organ from such a specimen image 15. The detection unit 51 extracts the feature values ​​72 of each identified tissue specimen of each organ. Therefore, it is also possible to handle slide specimens 125 containing tissue specimens of multiple organs. In terms of slide specimens, slide specimens 125 containing tissue specimens of multiple organs, as in this embodiment, are more common than slide specimens 18 containing tissue specimens of one organ as in the first embodiment described above. Therefore, processing that conforms to more general applications is possible. Furthermore, the frames 127 to 130 representing the tissue specimens of each organ in the specimen image 15 can be defined by the user U's operation.

[0162] Feature 72 is not limited to the features extracted by feature extractor 41. It can also be the average, maximum, minimum, mode, or variance of the pixel values ​​of the patch image 70.

[0163] The subject S is not limited to rats. It can also be mice, guinea pigs, gerbils, hamsters, ferrets, rabbits, dogs, cats, or monkeys, etc.

[0164] Drug development support device 10 can be as follows Figure 1 The personal computer shown can be located in a pharmaceutical research and development facility, or it can be a server computer located in a data center independent of the pharmaceutical research and development facility.

[0165] In the case where the drug development support device 10 consists of a server computer, specimen images 15 are sent from personal computers located in various drug development facilities to the server computer via a network such as the Internet. The server computer distributes various screens, such as analysis instruction screens 90, to the personal computers in the form of web page distribution screen data created using a markup language such as XML (Extensible Markup Language). The personal computers reproduce the screens displayed on the web browser based on the screen data and display them on the monitor. Alternatively, other data description languages ​​such as JSON (Javascript Object Notation) can be used instead of XML.

[0166] The drug development support device 10 involved in the present invention can be widely used throughout the entire drug development process, from the initial stage of setting drug development goals to the final stage of clinical trials.

[0167] Candidate substance 27 is not limited to the illustrated pharmaceuticals. It could also be other chemical substances such as pesticides or radioactive materials.

[0168] The hardware structure of the computer constituting the drug development support device 10 according to the technology of the present invention can be modified in various ways. For example, to improve processing power and reliability, the drug development support device 10 can be composed of multiple computers that are separate hardware components. For example, two computers can be assigned the functions of the detection unit 51 and the determination unit 52, and the function of the information update unit 53. In this case, the drug development support device 10 is composed of two computers.

[0169] In this way, the hardware structure of the computer in the drug development support device 10 can be appropriately modified according to the performance requirements such as processing power, security, and reliability. Moreover, to ensure security and reliability, it is not limited to hardware; application programs such as the working program 40 can also be duplicated or distributed and stored in multiple storage devices.

[0170] In the above embodiments, for example, as the hardware structure of the processing units that perform various processes, such as the RW control unit 50, detection unit 51, determination unit 52, information update unit 53, display control unit 54, selection unit 55, instruction receiving unit 114, and identification unit 126, various processors as shown below can be used. Among the various processors, in addition to the general-purpose processor, i.e., CPU 32, which executes software (working program 40) as described above to perform the functions of various processing units, processors that can have their circuit structure changed after manufacturing, such as FPGA (Field Programmable Gate Array), i.e., Programmable Logic Device (PLD), and processors with circuit structures specially designed for performing specific processes, such as ASIC (Application Specific Integrated Circuit), i.e., dedicated circuits, are also included.

[0171] A processing unit can consist of one of these various processors, or it can consist of 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). Alternatively, multiple processing units can be composed of a single processor.

[0172] As examples of a single processor comprising multiple processing units, firstly, there is a method, exemplified by computers such as client machines and servers, where a single processor, composed of one or more CPUs and software, functions as multiple processing units. Secondly, there is a method, exemplified by System-on-Chip (SoC), which uses a single integrated circuit (IC) chip to implement the overall system functionality including multiple processing units. In this way, various processing units are constructed using one or more of the aforementioned processors as hardware structures.

[0173] Moreover, the hardware architecture of these various processors, more specifically, can utilize circuits composed of circuit elements such as semiconductor elements.

[0174] Based on the above records, one can master the techniques described in the following notes.

[0175] [Note 1]

[0176] A drug development aid device, wherein,

[0177] Equipped with a processor

[0178] The processor

[0179] Acquire multiple specimen images of tissue specimens from multiple organs of the subject for use in evaluation tests of candidate substances.

[0180] Based on selection priority information that assigns a selection priority to each of the plurality of organs, an object specimen image of a tissue specimen of one organ is selected from the plurality of specimen images.

[0181] A determination is made as to whether morphological abnormalities have occurred in the tissue specimens captured in the object specimen images.

[0182] Based on the determination result of whether the morphological abnormality has occurred, the selection priority information is updated.

[0183] [Note 2]

[0184] According to the drug development auxiliary device described in Appendix 1, wherein,

[0185] In the selection priority information, each of the plurality of organs is assigned a probability of being selected as the object specimen image as the selection priority.

[0186] [Note 3]

[0187] According to the drug development auxiliary device described in Appendix 1 or 2, wherein,

[0188] The processor

[0189] If the determination result indicates that the morphological abnormality has occurred, the selection priority of the organs in the tissue specimen captured in the object specimen image is reset to a higher level.

[0190] If the determination result is that no morphological abnormality has occurred, the selection priority of the organs of the tissue specimen captured in the object specimen image shall not be changed, or the selection priority shall be reset to a lower value.

[0191] [Note 4]

[0192] The drug development auxiliary device according to any one of notes 1 to 3, wherein,

[0193] The processor

[0194] Based on the determination result, the selection priority of organs in the tissue specimens captured in the object specimen image is reset, and the selection priority of related organs that have a functional relationship with the organs in the tissue specimens captured in the object specimen image is also reset.

[0195] [Note 5]

[0196] The drug development auxiliary device according to any one of notes 1 to 4, wherein,

[0197] The processor accepts the user's final judgment result regarding whether the morphological anomaly has actually occurred.

[0198] In addition to the determination result, the selection priority information is also updated based on the final determination result.

[0199] [Note 6]

[0200] According to the drug development auxiliary device described in Appendix 5, wherein...

[0201] The processor

[0202] If the determination result is that the morphological abnormality has not occurred, but the final determination result is that the morphological abnormality has occurred, the selection priority of the organs in the tissue specimens captured in the object specimen images will be reset to a higher level.

[0203] If the determination result indicates that the morphological abnormality has occurred, but the final determination result indicates that the morphological abnormality has not occurred, the selection priority of the organs of the tissue specimen captured in the object specimen image is not changed, or the selection priority is reset to a lower value.

[0204] [Note 7]

[0205] The drug development auxiliary device according to any one of notes 1 to 6, wherein,

[0206] The subjects were divided into multiple groups.

[0207] In the selection priority information, a selection priority is set for each of the plurality of organs and for each of the plurality of groups.

[0208] [Note 8]

[0209] According to the drug development auxiliary device described in Appendix 7, wherein...

[0210] The multiple groups include a dosing group that was given the candidate substance and a control group that was not given the candidate substance.

[0211] [Note 9]

[0212] According to the drug development auxiliary device described in Appendix 8, wherein...

[0213] The dosing group includes multiple sub-dosing groups with different dosages of the candidate substance.

[0214] [Note 10]

[0215] The drug development auxiliary device according to any one of notes 1 to 9, wherein,

[0216] In the initial state, the selection priority information is set with selection priorities based on previously obtained insights.

[0217] [Note 11]

[0218] The drug development auxiliary device according to any one of appendices 1 to 10, wherein,

[0219] The processor

[0220] Detect the portion of the object specimen image that is presumed to have the morphological abnormality.

[0221] The determination is made by comparing a value related to the number of the presumed abnormal morphological parts with a pre-set determination threshold.

[0222] [Note 12]

[0223] According to the drug development auxiliary device described in Appendix 11, wherein,

[0224] The processor

[0225] Each of the multiple patch images obtained by subdividing the object specimen image is treated as the part.

[0226] The presumed morphological abnormality is detected by comparing the feature quantity obtained by inputting the patch image into the machine learning model with the reference feature quantity obtained by inputting a reference patch image of a tissue specimen considered normal into the machine learning model.

[0227] The technology of the present invention can also be appropriately combined with the various embodiments and / or variations described above. Furthermore, it is of course not limited to the embodiments described above; various structures can be adopted as long as they do not depart from the spirit of the invention. Moreover, the technology of the present invention relates not only to programs but also to storage media for non-transitory program storage.

[0228] The descriptions and illustrations above are detailed explanations of the parts related to the technology of this invention, and are merely one example of the technology of this invention. For example, the descriptions related to the structure, function, effect, and effect described above are only one example of the structure, function, effect, and effect of the parts related to the technology of this invention. Therefore, it goes without saying that, without departing from the spirit of the technology of this invention, unnecessary parts may be deleted, new elements may be added, or replacements may be made to the descriptions and illustrations above. In addition, to avoid complexity and to make the parts related to the technology of this invention easier to understand, descriptions related to common technical knowledge that do not require special explanation based on the technology that enables the implementation of this invention have been omitted from the descriptions and illustrations above.

[0229] In this specification, "A and / or B" is synonymous with "at least one of A and B". That is, "A and / or B" means that it can be only A, only B, or a combination of A and B. In addition, in this specification, when three or more items are connected by "and / or", the same idea applies as "A and / or B".

[0230] All documents, patent applications and technical specifications described herein are incorporated herein by reference to the same extent as the specific documents, patent applications and technical specifications described separately therein.

Claims

1. A drug development auxiliary device, wherein, Equipped with a processor The processor performs the following processing: Acquire multiple specimen images of tissue specimens from multiple organs of the subject for use in evaluation tests of candidate substances. Based on selection priority information that assigns a selection priority to each of the plurality of organs, an object specimen image of a tissue specimen of one organ is selected from the plurality of specimen images. A determination is made as to whether morphological abnormalities have occurred in the tissue specimens captured in the object specimen images. Based on the determination result of whether the morphological abnormality has occurred, the selection priority information is updated.

2. The drug development auxiliary device according to claim 1, wherein, In the selection priority information, each of the plurality of organs is assigned a probability of being selected as the object specimen image as the selection priority.

3. The drug development auxiliary device according to claim 1, wherein, The processor performs the following processing: If the determination result indicates that the morphological abnormality has occurred, the selection priority of the organs in the tissue specimen captured in the object specimen image is reset to be higher; If the determination result is that no morphological abnormality has occurred, the selection priority of the organs of the tissue specimen captured in the object specimen image is not changed, or the selection priority is reset to a lower value.

4. The drug development auxiliary device according to claim 1, wherein, The processor performs the following processing: Based on the determination result, the selection priority of organs in the tissue specimens captured in the object specimen image is reset, and the selection priority of related organs that have a functional relationship with the organs in the tissue specimens captured in the object specimen image is also reset.

5. The drug development auxiliary device according to claim 1, wherein, The processor accepts the user's final judgment result regarding whether the morphological anomaly has actually occurred. In addition to the determination result, the processor also updates the selection priority information based on the final determination result.

6. The drug development auxiliary device according to claim 5, wherein, The processor performs the following processing: If the determination result is that the morphological abnormality has not occurred, but the final determination result is that the morphological abnormality has occurred, the selection priority of the organs in the tissue specimens captured in the object specimen images will be reset to a higher level. If the determination result indicates that the morphological abnormality has occurred, but the final determination result indicates that the morphological abnormality has not occurred, the selection priority of the organs of the tissue specimen captured in the object specimen image is not changed, or the selection priority is reset to a lower value.

7. The drug development auxiliary device according to claim 1, wherein, The subjects were divided into multiple groups. In the selection priority information, a selection priority is set for each of the plurality of organs and for each of the plurality of groups.

8. The drug development auxiliary device according to claim 7, wherein, The multiple groups include a dosing group that was given the candidate substance and a control group that was not given the candidate substance.

9. The drug development auxiliary device according to claim 8, wherein, The dosing group includes multiple sub-dosing groups with different dosages of the candidate substance.

10. The drug development auxiliary device according to claim 1, wherein, In the initial state, the selection priority information is set with selection priorities based on previously obtained insights.

11. The drug development auxiliary device according to claim 1, wherein, The processor performs the following processing: Detect the portion of the object specimen image that is presumed to have the morphological abnormality. The determination is made by comparing a value related to the number of the presumed abnormal morphological parts with a pre-set determination threshold.

12. The drug development auxiliary device according to claim 11, wherein, The processor performs the following processing: Each of the multiple patch images obtained by subdividing the object specimen image is treated as the part. The presumed morphological abnormality is detected by comparing the feature quantity obtained by inputting the patch image into the machine learning model with the reference feature quantity obtained by inputting a reference patch image of a tissue specimen considered normal into the machine learning model.

13. A method for operating a drug development auxiliary device, wherein, Includes the following steps: Acquire multiple specimen images of tissue specimens from multiple organs of the subject for evaluation tests of candidate substances; Based on selection priority information that sets a selection priority for each of the plurality of organs, an object specimen image of a tissue specimen of an organ is selected from the plurality of specimen images; Determine whether morphological abnormalities have occurred in the tissue specimens captured in the object specimen images; as well as Based on the determination result of whether the morphological abnormality has occurred, the selection priority information is updated.

14. A working procedure for a drug development support device, the working procedure causing a computer to perform processing including the following steps: Acquire multiple specimen images of tissue specimens from multiple organs of the subject for evaluation tests of candidate substances; Based on selection priority information that sets a selection priority for each of the plurality of organs, an object specimen image of a tissue specimen of an organ is selected from the plurality of specimen images; Determine whether morphological abnormalities have occurred in the tissue specimens captured in the object specimen images; as well as Based on the determination result of whether the morphological abnormality has occurred, the selection priority information is updated.

Images