Three-dimensional image prediction device, three-dimensional image learning device, three-dimensional image prediction method, three-dimensional image learning method, three-dimensional image prediction program, and three-dimensional image learning program

Abstract

The present invention improves accuracy of predicting pharmacokinetics at a specific action site of a drug. This three-dimensional image prediction device includes: a prediction unit that inputs, to a trained model that is trained by using training data including: three-dimensional information relating to a training subject animal and showing an untreated organ, tissue, or cell; chemical structure information relating to a training subject drug; a normalized value obtained by normalizing a value of each condition item at a time when the training subject drug is administered; and three-dimensional information relating to the training subject animal and showing presence of the training subject drug at each position in the organ, tissue, or cell after the training subject drug is administered, three-dimensional information relating to a prediction subject animal and showing an untreated organ, tissue, or cell, chemical structure information relating to a prediction subject drug, and a normalized value of each condition item for the prediction subject animal, to thereby predict three-dimensional information showing presence of the prediction subject drug at each position in the organ, tissue, or cell of the prediction subject animal; and a display unit that displays an image of the predicted three-dimensional information.

Description

Three-dimensional image prediction device, three-dimensional image learning device, three-dimensional image prediction method, three-dimensional image learning method, three-dimensional image prediction program, and three-dimensional image learning program

[0001] The present disclosure relates to a three-dimensional image prediction device, a three-dimensional image learning device, a three-dimensional image prediction method, a three-dimensional image learning method, a three-dimensional image prediction program, and a three-dimensional image learning program.

[0002] Development of prediction techniques is underway to predict the pharmacokinetics at the site of action when a drug is administered to a target animal in a predetermined region unit (e.g., organ, tissue, or cell unit). For example, when the site of action of a drug is in the brain, such prediction techniques are required to generate high-resolution three-dimensional image data of the site of action and then predict the pharmacokinetics that changes over time with high accuracy.

[0003] Hiroki R. Ueda et al., “Whole-Brain Profiling of Cells and Circuits in Mammals by Tissue Clearing and Light-Sheet Microscopy”, Neuron, Volume 106, Issue 3, 6 May 2020, Pages 369-387

[0004] When the subject animal is an experimental animal, the time-dependent distribution of a drug in the body can be evaluated, for example, by using histopathological analysis. However, when the subject animal is a human, such methods cannot be used, and extrapolating the results of experimental animal evaluations to humans requires predictions that take into account at least species differences. Furthermore, because factors affecting pharmacokinetics differ from person to person (i.e., there are individual differences), to achieve highly accurate predictions, predictions that take into account factors affecting pharmacokinetics are required.

[0005] The present disclosure aims to improve the accuracy of predictions when predicting the pharmacokinetics of a drug at its site of action.

[0006] Examples of aspects of the present disclosure include the following: [1] A three-dimensional image prediction device comprising: a prediction unit that predicts three-dimensional information indicating the presence of the prediction target drug at each position in the organs, tissues, or cells of a prediction target animal by inputting three-dimensional information of the organs, tissues, or cells of a prediction target animal in a drug-free state, chemical structure information of a prediction target drug, and the normalized values ​​for each condition item for the prediction target animal into a trained model trained using training data including three-dimensional information of a training target animal, the three-dimensional information being three-dimensional information of unmedicated organs, tissues, or cells, chemical structure information of a training target drug, and normalized values ​​for each condition item for the training target animal when the training target drug was administered, and three-dimensional information of the training target animal indicating the presence of the training target drug at each position in the organs, tissues, or cells after the training target drug has been administered; and a display unit that displays an image of the predicted three-dimensional information. [2] The three-dimensional image prediction device described in [1], further comprising a calculation unit that normalizes the values ​​for each condition item and calculates the normalized values. [2a] The calculation unit normalizes each of the values ​​for the condition items based on a range that each of the values ​​for the condition items can take. [2b] The three-dimensional image prediction device described in [1], wherein the three-dimensional information of the training subject animal, wherein the three-dimensional information of an unmedicated organ, tissue, or cell is three-dimensional information of the organ, tissue, or cell before the training subject drug is administered to an experimental animal, and the three-dimensional information of the training subject animal indicating the presence of the training subject drug at each position within the organ, tissue, or cell after the training subject drug is administered is three-dimensional information of the organ, tissue, or cell after the training subject drug is administered to an experimental animal. [2c] The three-dimensional image prediction device described in [2], wherein the condition items store the elapsed time since the training drug and the prediction target drug were administered, respectively, the type of disease that the training animal and the prediction target animal have, individual information of each of the training animal and the prediction target animal, information on the lifestyle habits of each of the training animal and the prediction target animal for a certain period before administration, and the administration method if each of the training drug and the prediction target drug is administered.[2d] The three-dimensional image prediction device described in [2c], wherein the individual information of the training subject animal and the prediction subject animal further includes the age, weight, and sex of the training subject animal and the prediction subject animal. [2e] The three-dimensional image prediction device described in [2c], wherein the information on lifestyle habits includes the amount of food eaten, dietary imbalance, amount of exercise, sleep time, and stress level for a certain period before the training subject drug and the prediction subject drug are administered, respectively. [2f] The three-dimensional image prediction device described in [2c], wherein the administration method includes any of the administration route, administration rate, administration interval, dosage, and shape of the administered training subject drug and the administered prediction subject drug, respectively. [2g] The three-dimensional information of the animal to be trained, which indicates the presence of the drug to be trained at each position within an organ, tissue, or cell after administration of the drug to be trained, is generated based on one or a combination of light-sheet microscope image data, CFT image data, confocal microscope image data, multiphoton microscope image data, CT image data, MRI image data, PET image data, and SPECT image data, which are image data after administration of the drug to be trained to an experimental animal. [3] The three-dimensional image prediction device of [1], wherein the organ, tissue, or cell is a brain, brain tissue, or brain cell. [4] A three-dimensional image learning device having: a generation unit that generates learning data including three-dimensional information of a training subject animal, the three-dimensional information being three-dimensional information of unmedicated organs, tissues, or cells, chemical structure information of a training subject drug, normalized values ​​obtained by normalizing the conditions under which the training subject drug was administered for each condition item, and three-dimensional information of the training subject animal, the three-dimensional information being three-dimensional information indicating the presence of the training subject drug at each position within the organs, tissues, or cells after the training subject drug has been administered; a learning unit that performs learning processing on a model using the generated learning data to generate a trained model; and a memory unit that stores the generated trained model.[5] A three-dimensional image prediction method in which a computer executes the following steps: a prediction step of predicting three-dimensional information indicating the presence of the prediction target drug at each position in the organs, tissues, or cells of a prediction target animal when the drug is not administered, chemical structure information of the prediction target drug, and normalized values ​​for each condition item for the prediction target animal, by inputting three-dimensional information of the training target animal's unmedicated organs, tissues, or cells, chemical structure information of the training target drug, and normalized values ​​for each condition item for the prediction target animal into a trained model trained using training data including three-dimensional information of a training target animal, the three-dimensional information being three-dimensional information of the training target animal when the drug is not administered, chemical structure information of the training target drug, and normalized values ​​for each condition item for the prediction target animal, the three-dimensional information being three-dimensional information of the training target animal when the drug is administered and indicating the presence of the training target drug at each position in the organs, tissues, or cells of the prediction target animal; and a display step of displaying the predicted three-dimensional information as an image. [6] A three-dimensional image prediction program for causing a computer to execute the following steps: a prediction step of predicting three-dimensional information indicating the presence of the prediction target drug at each position in the organs, tissues, or cells of a prediction target animal by inputting three-dimensional information of the prediction target animal's organs, tissues, or cells without administration of medication, chemical structure information of the prediction target drug, and normalized values ​​for each condition item for the prediction target animal into a trained model trained using training data including three-dimensional information of a training target animal, the three-dimensional information being three-dimensional information of unmedicated organs, tissues, or cells, chemical structure information of the training target drug, and normalized values ​​for each condition item for the prediction target animal when the training target drug is administered, and the three-dimensional information of the training target animal, the three-dimensional information being three-dimensional information indicating the presence of the training target drug at each position in the organs, tissues, or cells of the prediction target animal after administration of the training target drug; and a display step of displaying the predicted three-dimensional information as an image.[7] A three-dimensional image learning method in which a computer executes the following steps: a generation step of generating training data including three-dimensional information of a training subject animal, the three-dimensional information being three-dimensional information of unadministered organs, tissues, or cells, chemical structure information of a training subject drug, normalized values ​​obtained by normalizing for each condition item the conditions under which the training subject drug was administered, and three-dimensional information of the training subject animal, the three-dimensional information being three-dimensional information indicating the presence of the training subject drug at each position within the organs, tissues, or cells after the training subject drug has been administered; a learning step of performing a learning process on a model using the generated training data to generate a trained model; and a storage step of storing the generated trained model in a storage unit. [8] A three-dimensional image learning program for causing a computer to execute the following steps: a generation step of generating training data including three-dimensional information of a training animal, the three-dimensional information being three-dimensional information of unmedicated organs, tissues, or cells, chemical structure information of a training drug, normalized values ​​obtained by normalizing for each condition item the conditions under which the training drug was administered, and three-dimensional information of the training animal, the three-dimensional information being three-dimensional information indicating the presence of the training drug at each position within the organ, tissue, or cell after the training drug was administered; a learning step of performing a learning process on a model using the generated training data to generate a trained model; and a storage step of storing the generated trained model in a storage unit.

[0007] According to the present disclosure, it is possible to improve the accuracy of prediction when predicting the pharmacokinetics of a drug at a specific site of action.

[0008] FIG. 1 is a diagram illustrating an example of the system configuration of a pharmacokinetic analysis system. FIG. 2 is a diagram illustrating an overview of pharmacokinetic analysis processing by the pharmacokinetic analysis system. FIG. 3 is a diagram illustrating an overview of processing by a three-dimensional image learning device in the learning phase. FIG. 4 is a diagram illustrating an overview of processing by a three-dimensional image prediction device in the prediction phase. FIG. 5 is a diagram illustrating a specific example of structural information generation processing by the three-dimensional image learning device. FIG. 6 is a diagram illustrating a specific example of normalization processing by the three-dimensional image learning device. FIG. 7 is a diagram illustrating a specific example of three-dimensional information generation processing by a three-dimensional image generation device. FIG. 8 is a diagram illustrating an example of learning data. FIG. 9 is a diagram illustrating an example of the hardware configuration of a three-dimensional image learning device. FIG. 10 is a diagram illustrating an example of the functional configuration of a three-dimensional image learning device. FIG. 11 is an example of a flowchart illustrating the flow of learning processing by the three-dimensional image learning device. FIG. 12 is a diagram illustrating an example of a prediction target animal in the prediction phase. FIG. 13 is a diagram illustrating a specific example of normalization processing by the three-dimensional image prediction device. FIG. 14 is a diagram illustrating an example of the hardware configuration of a three-dimensional image prediction device. FIG. 15 is a diagram illustrating an example of the functional configuration of a three-dimensional image prediction device. FIG. 16 is an example of a flowchart showing the flow of prediction processing by the three-dimensional image prediction device.

[0009] Hereinafter, each embodiment will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0010] [First Embodiment] <System Configuration of Pharmacokinetic Analysis System> First, the system configuration of a pharmacokinetic analysis system including a three-dimensional image learning device and a three-dimensional image prediction device according to the first embodiment will be described. Fig. 1 is a diagram showing an example of the system configuration of a pharmacokinetic analysis system. As shown in Fig. 1, a pharmacokinetic analysis system 100 includes a three-dimensional image learning device 110, various measuring devices 120, a three-dimensional image generation device 130, and a three-dimensional image prediction device 140.

[0011] In the pharmacokinetic analysis system 100, an experimenter 150 first uses various measuring devices 120 to measure data for generating three-dimensional information on unadministered organs, tissues, or cells of a training subject animal 152. The experimenter 150 also generates pre-administration three-dimensional information from the measured data using a three-dimensional image generating device 130. The pre-administration three-dimensional information is three-dimensional image data on unadministered organs, tissues, or cells of the training subject animal 152.

[0012] Next, the experimenter 150 administers a drug 151, which is a drug to be learned, to an animal 152, which is a drug to be learned. The experimenter 150 also uses various measuring devices 120 to measure data (e.g., concentration data of the drug 151) indicating the presence of the drug 151 at each position in an organ, tissue, or cell at each elapsed time after the drug 151 is administered to the animal 152.

[0013] Next, the experimenter 150 generates post-administration three-dimensional information from the measured data using the three-dimensional image generating device 130. The post-administration three-dimensional information is three-dimensional image data (three-dimensional image data colored according to the concentration data of the drug 151) that indicates the presence of the drug 151 at each position in the organ, tissue, or cell at each elapsed time after administration.

[0014] Next, the experimenter 150 inputs the following information into the 3D image learning device 110: information on each condition item (an example of a factor that affects pharmacokinetics) when the drug 151 is administered to the animal 152 (condition item information to be learned); information on the administered drug 151 (drug information to be learned); and the pre-administration 3D information and post-administration 3D information generated by the 3D image generation device 130.

[0015] The learning target condition item information includes values ​​of each condition item, such as: "time elapsed" since the administration of the drug 151; "type of disease" of the animal 152 to which the drug 151 was administered; "individual information" of the animal 152 to which the drug 151 was administered; "information on lifestyle" of the animal 152 to which the drug 151 was administered for a certain period before administration; and "administration method" when the drug 151 was administered. Of these, the "individual information" of the animal 152 includes the age, weight, sex, etc. of the animal 152 to which the drug 151 was administered. Furthermore, the "information on lifestyle" includes the amount of food eaten, dietary imbalance, amount of exercise, sleep time, stress level, etc. of the animal 152 to which the drug 151 was administered. Furthermore, the "administration method" includes the administration route, administration rate, administration interval, dosage, and form of the drug 151 when the drug 151 was administered to the animal 152.

[0016] The various measuring devices 120 measure data for generating pre-administration three-dimensional information of the animal 152, which is a learning target animal. The various measuring devices 120 also measure data for generating post-administration three-dimensional information after the drug 151 is administered to the animal 152. The data measured by the various measuring devices 120 is image data of the animal 152, and the image data includes image data measured by, for example, a light-sheet microscope, CFT (Cryo-Fluorescence Tomography), a confocal microscope, CT (Computed Tomography), MRI (Magnetic Resonance Imaging), PET (Positron Emission Tomography), a multiphoton microscope, and SPECT (Single Photon Emission Computed Tomography).

[0017] The three-dimensional image generating device 130 generates post-administration three-dimensional image data, which is post-administration three-dimensional information, from image data measured by the various measuring devices 120. As described above, the post-administration three-dimensional image data generated by the three-dimensional image generating device 130 is, for example, three-dimensional image data for each elapsed time, colored according to the concentration of the drug 151 at each position in an organ, tissue, or cell.

[0018] In this embodiment, the site of action of the drug 151 is the brain, and the organ, tissue, or cell refers to the brain, brain tissue, or brain cell. Therefore, all three-dimensional image data in this embodiment refers to three-dimensional image data of the whole brain (also called whole-brain profiling data).

[0019] When the learning target condition item information and the learning target drug information are input, the three-dimensional image learning device 110 performs a learning process on a model (referred to as a three-dimensional image generation model). Specifically, when the information is input, the three-dimensional image learning device 110 performs a learning process on the three-dimensional image generation model so that post-administration three-dimensional image data can be predicted from pre-administration three-dimensional image data.

[0020] The trained model (referred to as the trained 3D image generation model) generated by the training process performed by the 3D image training device 110 is readably stored in the memory unit of the 3D image prediction device 140.

[0021] Next, in the pharmacokinetic analysis system 100, the designer 160 acquires information (prediction target drug information) about a drug 161 that is a drug to be predicted. The designer 160 also identifies an animal 162 as a prediction target animal for which prediction of 3D image data at each elapsed time when the drug 161 is administered is performed. The designer 160 then acquires information (prediction target condition item information) about each condition item (an example of a factor that affects pharmacokinetics) for the identified animal 162. The condition items included in the prediction target condition item information are the same as the condition items included in the learning target condition item information, and therefore will not be described here.

[0022] Next, the designer 160 acquires "standard three-dimensional information," which is standard three-dimensional information of drug-free organs, tissues, or cells of the animal 162. The standard three-dimensional information refers to: drug-free three-dimensional image data; and standard three-dimensional image data (standard three-dimensional image data) for the species to which the animal 162 belongs. The standard three-dimensional information is assumed to be generated in advance for each species of the animal to be predicted, for example.

[0023] Furthermore, the designer 160 inputs the following to the three-dimensional image prediction device 140: prediction target condition item information for the animal 162; prediction target drug information for the drug 161; and standard three-dimensional information.

[0024] The 3D image prediction device 140 stores the trained 3D image generation model in a storage unit. When prediction target condition item information, prediction target drug information, and standard 3D information are input, the 3D image prediction device 140 reads the trained 3D image generation model from the storage unit and executes a prediction process. As a result, the trained 3D image generation model predicts predicted 3D image data as predicted 3D information indicating the presence of the drug 161 at each elapsed time in the animal 162 after administration of the drug 161.

[0025] <Outline of Pharmacokinetic Analysis Processing> Next, we will explain an outline of the pharmacokinetic analysis processing by the pharmacokinetic analysis system 100. Fig. 2 is a diagram for explaining an outline of the pharmacokinetic analysis processing by the pharmacokinetic analysis system.

[0026] 2A shows an overview of the pharmacokinetic analysis process in the learning phase. As shown in FIG. 2A, in the learning phase, the 3D image learning device 110 normalizes the values ​​of each condition item included in the learning target condition item information by performing a normalization process on the learning target condition item information. This generates normalized condition item information in which the conditions when the learning target drug is administered are composed of normalized values ​​normalized for each condition item.

[0027] In the learning phase, the three-dimensional image learning device 110 performs a structural information generation process on the drug information to be learned to generate chemical structure information of the drug to be learned (information that identifies the chemical structure of each drug). In the learning phase, the three-dimensional image learning device 110 also acquires the pre-administration three-dimensional image data and the post-administration three-dimensional image data generated by the three-dimensional image generation device 130.

[0028] Furthermore, as shown in 2a of Figure 2, in the learning phase, the following are input to the 3D image generation model as input data: normalization condition item information; chemical structure information; and pre-administration 3D image data; and post-administration 3D image data are input as correct answer data.

[0029] As a result, the three-dimensional image generation model is trained to predict post-administration three-dimensional image data from pre-administration three-dimensional image data when normalization condition item information and chemical structure information are input.

[0030] 2B shows an overview of the pharmacokinetic analysis process in the prediction phase. In the prediction phase, the 3D image prediction device 140 performs a normalization process to normalize the values ​​of each condition item included in the prediction target condition item information, as shown in 2B of FIG. 2. This generates normalized condition item information in which the conditions for administering the prediction target drug are configured using normalized values ​​normalized for each condition item.

[0031] Furthermore, in the prediction phase, the three-dimensional image prediction device 140 performs a structural information generation process on the predicted target drug information to generate chemical structural information of the predicted target drug.

[0032] Also, in the prediction phase, the 3D image prediction device 140 obtains standard 3D image data.

[0033] 2B, in the prediction phase, the trained 3D image generation model receives, as input data, the following: normalization condition item information; chemical structure information; and standard 3D image data. As a result, the trained 3D image generation model predicts predicted 3D image data as predicted 3D information.

[0034] <Overview of Processing in the Learning Phase> Next, an overview of the processing of the 3D image learning device 110 in the learning phase will be described. Fig. 3 is a diagram showing an overview of the processing of the 3D image learning device in the learning phase. Of these, the left side of the page shows the distribution in the condition item space (multidimensional) of the learning target condition item information corresponding to each post-administration 3D image data input to the 3D image learning device 110. Here, it is assumed that the learning target animal is "experimental animal α" and the learning target drug is "drug A."

[0035] As described above, the learning target condition item information includes multiple condition items, so the condition item space is multidimensional, but for convenience it is shown as two-dimensional in the example of Fig. 3. One black circle in the condition item space represents post-administration three-dimensional image data of experimental animal α that was generated when drug A was administered to experimental animal α under conditions in which one of the condition items had a value different from the values ​​of the condition items of the other black circles.

[0036] As described above, in the learning phase, the 3D image learning device 110 normalizes the value of each condition item included in the learning target condition item information. In Fig. 3, the right side of the page shows the distribution of the normalized condition item information in the normalized condition item space (multidimensional).

[0037] In this way, by normalizing the values ​​of each condition item included in the condition item information to be learned and inputting them into the three-dimensional image generation model, the three-dimensional image learning device 110 can have the three-dimensional image generation model learn the relationship between: - each position within the normalized condition item space; and - the three-dimensional image data after administration.

[0038] As a result, the 3D image learning device 110 can: - appropriately reflect each condition item (an example of a factor that affects pharmacokinetics) included in the learning target condition item information as an individual difference in the learning of the 3D image generation model; - treat each condition item (an example of a factor that affects pharmacokinetics) included in the learning target condition item information equally regardless of the type of learning target animal (species difference).

[0039] As a result, the three-dimensional image learning device 110 can generate a trained three-dimensional image generation model that can predict with high accuracy the pharmacokinetics of: - a prediction target animal of the same type as the training target animal, but a different individual from the training target animal (a prediction target animal with different values ​​for condition items from the training target animal), or - a prediction target animal with the same values ​​for condition items as the training target animal, but a different type from the training target animal.

[0040] <Outline of Processing in Prediction Phase> Next, an outline of the processing of the 3D image prediction device 140 in the prediction phase will be described. Fig. 4 is a diagram showing an outline of the processing of the 3D image prediction device in the prediction phase. Of these, the left side of 4a in Fig. 4 shows the distribution of normalization condition item information in the normalization condition item space (multidimensional), which is the same as the right side of Fig. 3.

[0041] Furthermore, the white circle on the right side of the page in 4a of Figure 4 indicates the position in the normalization condition item space when the predicted 3D image data is predicted by changing the normalization value of the normalization condition item information based on the standard 3D image data of animal α.

[0042] As described above, the trained 3D image generation model has learned the relationship between each position in the normalized condition item space and the post-administration 3D image data based on the input pre-administration 3D image data. Therefore, the trained 3D image generation model can accurately predict predicted 3D image data corresponding to the position in the normalized condition item space (multidimensional) of the normalized value according to individual differences based on the input standard 3D image data.

[0043] The left side of FIG. 4b shows the distribution of normalized condition item information in the normalized condition item space, and is the same as the right side of FIG.

[0044] Furthermore, the white square on the right side of 4b in Figure 4 indicates that the predicted 3D image data was predicted by changing the standard 3D image data from animal α to animal X under the same normalization condition item information, that is, the position of the normalization value in the normalization condition item space (multidimensional) was the same.

[0045] As described above, the trained 3D image generation model has learned the relationship between each position in the normalized condition item space and the post-administration 3D image data based on the input pre-administration 3D image data. Therefore, based on standard 3D image data according to the type of target animal, the trained 3D image generation model can accurately predict predicted 3D image data corresponding to positions in the normalized condition item space (multidimensional) of at least the same normalized value.

[0046] <Details of Processing by 3D Image Learning Device and 3D Image Generating Device> Next, details of processing by the 3D image learning device 110 and the 3D image generating device 130 will be described.

[0047] (1) Details of the Structural Information Generation Process First, a specific example of the structural information generation process by the three-dimensional image learning device 110 will be described. FIG. 5 is a diagram showing a specific example of the structural information generation process by the three-dimensional image learning device. In FIG. 5, reference numeral 510 indicates training target drug information including a list of drugs administered by the experimenter 150. The example in FIG. 5 indicates that "drug A," "drug B," ... "drug n," etc., have been administered to the training target animal 152.

[0048] 5, reference numeral 520 denotes chemical structure information generated by the 3D image learning device 110 through structural information generation processing for each drug included in the learning target drug information. The example in FIG. 5 shows how "chemical structure information 1" is generated from "drug A," "chemical structure information 2" is generated from "drug B," and so on, and "chemical structure information n" is generated from "drug n."

[0049] (2) Details of Normalization Processing Next, a specific example of normalization processing by the three-dimensional image learning device 110 will be described. Fig. 6 is a diagram showing a specific example of normalization processing by the three-dimensional image learning device.

[0050] 6, the three-dimensional image learning device 110 multiplies the "time elapsed since administration" (610) of the learning condition item information by a coefficient corresponding to the corresponding type of learning target animal (611). As a result, the three-dimensional image learning device 110 calculates a normalized value of the "time elapsed since administration" (610) (612).

[0051] The three-dimensional image learning device 110 also quantifies the "type of disease the administration subject has" (620) from the learning subject condition item information, for example, to a predetermined value for each disease (621). As a result, the three-dimensional image learning device 110 calculates a normalized value of the information on the "type of disease the administration subject has" (620) (622).

[0052] Furthermore, for the "individual information of the administration subject" (reference numeral 630) among the learning subject condition item information, the three-dimensional image learning device 110: - In the case of "age," for example, normalizes based on the average lifespan of the corresponding learning subject animal. - In the case of "weight," for example, normalizes based on the variation from the average weight of the corresponding learning subject animal. - In the case of "sex," for example, quantifies to a predetermined value for each sex (reference numeral 631).

[0053] As a result, the three-dimensional image learning device 110 calculates the normalized values ​​of the "age" and "weight" values ​​and the "gender" information (reference numeral 632).

[0054] Furthermore, the three-dimensional image learning device 110 normalizes the "lifestyle habits for a certain period before administration" (640) among the learning condition item information, for example, based on the variation from the average value of the corresponding learning target animal (641). As a result, the three-dimensional image learning device 110 calculates a normalized value of the "lifestyle habits for a certain period before administration" (640) (642).

[0055] Furthermore, for the "administration method" (reference numeral 650) among the learning condition item information, the 3D image learning device 110: - quantifies, for example, to a predetermined value, the "administration route" and "dosage form"; - multiplies, for example, a coefficient according to the type of the corresponding learning target animal, the "administration rate," the "administration interval," and the "dosage amount" (reference numeral 651).

[0056] As a result, the three-dimensional image learning device 110 calculates normalized values ​​for the information on the "administration route" and "dosage form," as well as the values ​​for the "administration rate," "administration interval," and "dosage amount" (symbol 652).

[0057] (3) Details of the 3D Information Generation Process Next, a specific example of the 3D information generation process by the 3D image generation device 130 will be described. Fig. 7 is a diagram showing a specific example of the 3D information generation process by the 3D image generation device.

[0058] The three-dimensional image generating device 130 acquires data from the various measuring devices 120 indicating the presence of the drug 151 at each position within the organ, tissue, or cell at each elapsed time, measured by the various measuring devices 120 when the drug 151 is administered to the animal 152.

[0059] The example in Figure 7 shows that at each elapsed time, the following data were measured by various measuring devices 120: Light-sheet microscope image data, CFT whole-body image data, super-resolution / confocal cellular level image data, CT / MRI / PET image data, and ISH (Immunohistochemistry) / IHC (in-situ Hybridization) / MERFISH (Multiplexed Error-Robust Fluorescence in situ Hybridization) image data.

[0060] The three-dimensional image generating device 130 also extracts image data of the brain from the image data at each elapsed time measured by the various measuring devices 120. The three-dimensional image generating device 130 also generates data indicating the presence of the drug 151 at each position in the brain at each elapsed time (e.g., concentration data of the drug 151) by combining one or more of the extracted intracerebral image data and complementing them with each other. Furthermore, the three-dimensional image generating device 130 generates post-administration three-dimensional image data at each elapsed time after administration of the drug 151 by applying a color scheme according to the concentration data of the drug 151 at each position in the brain at each elapsed time.

[0061] 7, post-administration three-dimensional image data 720_1, 720_2, 720_3, ..., 720_T indicate post-administration three-dimensional image data at each elapsed time, respectively. Also, in FIG. 7, reference numerals 730_1, 730_2, 730_3, ..., 730_T indicate RGB values ​​at each three-dimensional coordinate in the brain at each elapsed time, respectively.

[0062] The three-dimensional image generating device 130 extracts image data of the brain from image data at an elapsed time of 0 (before the drug 151 is administered), not shown. The three-dimensional image generating device 130 also generates pre-administration three-dimensional image data before the drug 151 is administered by combining one or more of the extracted intracerebral image data and complementing them with each other.

[0063] <Explanation of Learning Data> Next, details of the learning data generated by the 3D image learning device 110 in the learning phase will be described. Fig. 8 is a diagram showing an example of learning data. As shown in Fig. 8, learning data 800 includes "input data" and "correct answer data" as information items.

[0064] The "input data" further includes "normalized condition item information," "chemical structure information," and "pre-administration three-dimensional information." The "normalized condition item information" stores normalized condition item information in which the values ​​of each condition item included in the learning target condition item information are configured using normalized values. The "chemical structure information" stores chemical structure information generated based on the learning target drug information. The "pre-administration three-dimensional information" stores pre-administration three-dimensional image data before the learning target drug is administered. The "correct answer data" stores post-administration three-dimensional image data at each elapsed time after the learning target drug is administered.

[0065] The example of training data 800 shown in Figure 8 shows training data including three-dimensional image data after administration at each elapsed time when drug 151 is administered to animal 152 under different normalization condition item information, with the type of drug to be trained fixed.

[0066] <Hardware Configuration of 3D Image Learning Device> Next, the hardware configuration of the 3D image learning device 110 will be described. Fig. 9 is a diagram showing an example of the hardware configuration of the 3D image learning device. As shown in Fig. 9, the 3D image learning device 110 has a processor 901, a memory 902, an auxiliary storage device 903, an I / F (Interface) device 904, a communication device 905, and a drive device 906. Note that each piece of hardware in the 3D image learning device 110 is connected to each other via a bus 907.

[0067] The processor 901 has various computing devices such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc. The processor 901 reads various programs (for example, a 3D image learning program) into the memory 902 and executes them.

[0068] The memory 902 has a main storage device such as a read-only memory (ROM) or a random access memory (RAM). The processor 901 and the memory 902 form a so-called computer, and the processor 901 executes various programs read onto the memory 902, thereby enabling the computer to realize various functions.

[0069] The auxiliary storage device 903 has storage such as a hard disk drive (HDD) or a solid state drive (SSD), and stores various programs and various data used when the various programs are executed by the processor 901.

[0070] The I / F device 904 is a connection device that connects an operation device 910 and a display device 911, which are examples of external devices, with the three-dimensional image learning device 110. The I / F device 904 accepts operations for the three-dimensional image learning device 110 (e.g., input operations of various information by the experimenter 150, etc.) via the operation device 910. The I / F device 904 also outputs results of the learning process by the three-dimensional image learning device 110 and displays them to the experimenter 150 via the display device 911.

[0071] The communication device 905 is a communication device for communicating with an external device 921 (in this embodiment, the 3D image generation device 130 and the 3D image prediction device 140) via a network 920.

[0072] The drive device 906 is a device for loading a recording medium 930. The recording medium 930 here includes media that record information optically, electrically, or magnetically, such as a CD-ROM, a flexible disk, a magneto-optical disk, etc. The recording medium 930 may also include semiconductor memory that records information electrically, such as a ROM, a flash memory, etc.

[0073] The various programs to be installed in the auxiliary storage device 903 are installed, for example, by setting the distributed recording medium 930 in the drive device 906 and reading the various programs recorded on the recording medium 930 by the drive device 906. Alternatively, the various programs to be installed in the auxiliary storage device 903 may be installed by being downloaded from the network 920 via the communication device 905.

[0074] <Functional Configuration of 3D Image Learning Device> Next, a description will be given of the functional configuration of the 3D image learning device 110. Fig. 10 is a diagram showing an example of the functional configuration of the 3D image learning device.

[0075] As described above, a 3D image learning program is installed in the 3D image learning device 110. When the program is executed, the 3D image learning device 110 functions as a generation unit 1010 and a learning unit 1020.

[0076] The generation unit 1010 further includes a calculation unit 1011 and a structural information generation unit 1012. The calculation unit 1011 normalizes the value of each condition item included in the learning target condition item information. For example, the calculation unit 1011 normalizes the value of each condition item included in the learning target condition information for the learning target animal = experimental animal α, and generates normalized condition item information = normalized condition item information 1.

[0077] The structural information generating unit 1012 performs structural information generation processing on the drug information to be learned to generate chemical structure information. For example, the structural information generating unit 1012 performs structural information generation processing on the drug A to be learned to generate chemical structure information 1.

[0078] The generation unit 1010 stores the normalization condition item information generated by the calculation unit 1011 and the chemical structure information generated by the structure information generation unit 1012 in the “normalization condition item information” or “chemical structure information” of the “input data” of the learning data 800.

[0079] The generation unit 1010 associates the normalization condition item information and chemical structure information with: pre-administration 3D image data generated by the 3D image generation device 130, which shows the drug to be trained = drug A before it is administered to the training subject animal = experimental animal α; and post-administration 3D image data generated by the 3D image generation device 130, which shows the drug to be trained = drug A after it is administered to the training subject animal = experimental animal α, and stores them in the training data storage unit 1040. For example, the generation unit 1010 associates normalization condition item information 1 and chemical structure information 1 with pre-administration 3D image data 1 for the training subject animal = experimental animal α and post-administration 3D image data 1 for the training subject animal = experimental animal α, and stores them in the training data 800. The pre-administration three-dimensional image data generated by the three-dimensional image generating device 130 is stored in the "pre-administration three-dimensional image data" of the "input data" of the training data 800, and the post-administration three-dimensional image data is stored in the "correct answer data" of the training data 800.

[0080] The learning unit 1020 further includes a 3D image generation model 1021 and a comparison / modification unit 1022 .

[0081] The three-dimensional image generation model 1021 inputs the "normalization condition item information," "chemical structure information," and "pre-administration three-dimensional image data" contained in the "input data" of the learning data 800, and outputs output data.

[0082] The comparison / change unit 1022 updates the model parameters of the 3D image generation model 1021 so that the output data output from the 3D image generation model 1021 approaches the "correct data" of the training data 800. In this way, by performing a training process on the 3D image generation model 1021 using the training data 800, the 3D image training device 110 can generate a trained 3D image generation model.

[0083] <Learning Process Flow by 3D Image Learning Device> Next, we will explain the learning process flow by the 3D image learning device 110. Fig. 11 is an example of a flowchart showing the learning process flow by the 3D image learning device.

[0084] In step S1101, the three-dimensional image learning device 110 acquires learning target condition item information and performs normalization processing to generate normalized condition item information.

[0085] In step S1102, the three-dimensional image learning device 110 acquires learning target drug information and performs structure information generation processing to generate chemical structure information.

[0086] In step S1103, the three-dimensional image learning device 110 acquires pre-administration three-dimensional information and post-administration three-dimensional information from the three-dimensional image generating device 130.

[0087] In step S1104, the three-dimensional image learning device 110 generates learning data and stores it in the learning data storage unit 1040.

[0088] In step S1105, the 3D image learning device 110 performs a learning process on the 3D image generation model 1021 using the generated learning data, and generates a trained 3D image generation model.

[0089] In step S1106, the 3D image learning device 110 determines whether or not to end the learning process. If it is determined in step S1106 that the learning process is to be continued (NO in step S1106), the process returns to step S1101.

[0090] On the other hand, if it is determined in step S1106 that the learning process is to be ended (YES in step S1106), the process proceeds to step S1107.

[0091] In step S1107, the 3D image learning device 110 stores the learned 3D image generation model in the storage unit of the 3D image prediction device 140, and ends the learning process.

[0092] <Details of Processing by 3D Image Prediction Device> Next, details of processing by the 3D image prediction device 140 will be described.

[0093] (1) Specific Example of Prediction Target Animal First, a description will be given of a prediction target animal for which predicted 3D information is predicted in the prediction phase by the 3D image prediction device 140. Fig. 12 is a diagram showing an example of a prediction target animal in the prediction phase.

[0094] 12 , in this embodiment, the animals to be predicted are "rodents," "NHPs," and "humans." Therefore, when performing prediction processing in the prediction phase, the 3D image prediction device 140 acquires in advance: standard 3D image data of rodents, standard 3D image data of NHPs (Non-Human Primates), and standard 3D image data of humans.

[0095] (2) Details of Normalization Processing Next, a specific example of normalization processing by the 3D image prediction device 140 will be described. Fig. 13 is a diagram showing a specific example of normalization processing by the 3D image prediction device. Note that the normalization processing by the 3D image prediction device 140 is generally the same as the normalization processing by the 3D image learning device 110, and the only difference is whether the normalization processing is based on a training target animal or a prediction target animal.

[0096] 13 , the 3D image prediction device 140 multiplies the “time elapsed since administration” (reference numeral 1310) among the prediction target condition item information by a coefficient corresponding to the corresponding type of animal to be predicted (reference numeral 1311). As a result, the 3D image prediction device 140 calculates a normalized value of the “time elapsed since administration” (reference numeral 1310) (reference numeral 1312).

[0097] Furthermore, the 3D image prediction device 140 quantifies the "type of disease the administration target has" (reference numeral 1320) among the prediction target condition item information, for example, to a predetermined value for each disease (reference numeral 1321). As a result, the 3D image prediction device 140 calculates a normalized value of the information on the "type of disease the administration target has" (reference numeral 1320) (reference numeral 1322).

[0098] Furthermore, for the "individual information of the administration target" (symbol 1330) among the prediction target condition item information, the 3D image prediction device 140: - In the case of "age", for example, normalizes based on the average lifespan of the corresponding prediction target animal. - In the case of "weight", for example, normalizes based on the variation from the average weight of the corresponding prediction target animal. - In the case of "sex", for example, quantifies to a predetermined value for each sex (symbol 1331).

[0099] As a result, the 3D image prediction device 140 calculates the normalized values ​​of the "age" and "weight" values ​​and the "gender" information (reference numeral 1332).

[0100] Furthermore, the three-dimensional image prediction device 140 normalizes the "lifestyle habits for a certain period before administration" (reference numeral 1340) among the prediction target condition item information, for example, based on the variation from the average value of the corresponding prediction target animal (reference numeral 1341). As a result, the three-dimensional image prediction device 140 calculates a normalized value of the information on the "lifestyle habits for a certain period before administration" (reference numeral 1340) (reference numeral 1342).

[0101] Furthermore, for the "administration method" (symbol 1350) among the learning condition item information, the 3D image prediction device 140: - quantifies, for example, to a predetermined value, the "administration route" and "dosage form"; - multiplies, for example, a coefficient according to the type of animal to be predicted, the "administration rate," "administration interval," and "dosage amount" (symbol 1351).

[0102] As a result, the 3D image prediction device 140 calculates normalized values ​​of the information on the "administration route" and "dosage form", as well as the values ​​of the "administration rate", "administration interval", and "dosage amount" (symbol 1352).

[0103] <Hardware Configuration of 3D Image Prediction Device> Next, the hardware configuration of the 3D image prediction device 140 will be described. Fig. 14 is a diagram showing an example of the hardware configuration of a 3D image prediction device. As shown in Fig. 14, the hardware configuration of the 3D image prediction device 140 is generally the same as the hardware configuration of the 3D image learning device 110, and therefore differences from the hardware configuration of the 3D image learning device 110 will be described here.

[0104] The processor 1401 reads and executes various programs (for example, a 3D image prediction program) onto the memory 1402. When the processor 1401 executes the various programs read onto the memory 1402, the computer formed by the processor 1401 and the memory 1402 realizes the functions described below.

[0105] <Functional Configuration of 3D Image Prediction Device> Next, a description will be given of the functional configuration of the 3D image prediction device 140. Fig. 15 is a diagram showing an example of the functional configuration of the 3D image prediction device.

[0106] As described above, a 3D image prediction program is installed in the 3D image prediction device 140. By executing this program, the 3D image prediction device 140 functions as a calculation unit 1510 (or 1511), a structural information generation unit 1520 (or 1521), a prediction unit 1530, and a display unit 1540.

[0107] The calculation unit 1510 or the calculation unit 1511 normalizes the value of each condition item included in the prediction target condition item information. For example, the calculation unit 1510 normalizes the value of each condition item included in the prediction target condition information for the prediction target animal = animal α, and generates normalized condition item information 1. Furthermore, for example, the calculation unit 1511 normalizes the value of each condition item included in the prediction target condition information for the prediction target animal = animal X, and generates normalized condition item information 2.

[0108] The structural information generating unit 1520 or the structural information generating unit 1521 performs structural information generation processing on the predicted target drug information to generate chemical structure information. For example, the structural information generating unit 1520 performs structural information generation processing on the predicted target drug = drug A to generate chemical structure information 1. Furthermore, the structural information generating unit 1521 performs structural information generation processing on the predicted target drug = drug A to generate chemical structure information 1.

[0109] The prediction unit 1530 reads out a trained 3D image generation model 1531 that was generated by the training process performed by the 3D image training device 110 and stored in the storage unit 1550. The prediction unit 1530 also performs the following operations on the trained 3D image generation model 1531 read out from the storage unit 1550, for example: inputting normalization condition item information 1 generated by the calculation unit 1510, inputting chemical structure information 1 generated by the structure information generation unit 1520, and inputting standard 3D image data of animal α, thereby causing the trained 3D image generation model 1531 to perform a prediction process. As a result, the trained 3D image generation model 1531 predicts predicted 3D image data 1 at each elapsed time for the predicted target animal = animal α when the prediction target drug = drug A is administered.

[0110] Furthermore, the prediction unit 1530 inputs, for example, the normalization condition item information 1 generated by the calculation unit 1511, the chemical structure information 1 generated by the structure information generation unit 1520, and the standard 3D image data of animal X to the trained 3D image generation model 1531 read out from the storage unit 1550, thereby causing the trained 3D image generation model 1531 to execute a prediction process. As a result, the trained 3D image generation model 1531 predicts predicted 3D image data 2 at each elapsed time of the predicted target animal = animal X when the predicted target drug = drug A is administered.

[0111] The display unit 1540 displays, as an image, the predicted 3D image data predicted by the trained 3D image generation model 1531. For example, the display unit 1540 displays, as an image, the predicted 3D image data 1 of the prediction target animal = animal α at each elapsed time. Alternatively, the display unit 1540 displays, as an image, the predicted 3D image data 2 of the prediction target animal = animal X at each elapsed time.

[0112] <Flow of Prediction Processing by 3D Image Prediction Device> Next, we will explain the flow of prediction processing by the 3D image prediction device 140. Fig. 16 is an example of a flowchart showing the flow of prediction processing by the 3D image prediction device.

[0113] In step S1601, the 3D image prediction device 140 acquires prediction target condition item information and performs normalization processing to generate normalized condition item information.

[0114] In step S1602, the three-dimensional image prediction device 140 acquires the prediction target drug information and performs a structure information generation process to generate chemical structure information.

[0115] In step S1603, the 3D image prediction device 140 acquires standard 3D information of the animal to be predicted.

[0116] In step S1604, the 3D image prediction device 140 inputs the normalization condition item information, chemical structure information, and standard 3D information of the animal to be predicted into the trained 3D image generation model 1531. As a result, the trained 3D image generation model 1531 executes the prediction process.

[0117] In step S1605, the 3D image prediction device 140 displays the predicted 3D information of the prediction target animal at each elapsed time as an image.

[0118] In step S1606, the 3D image prediction device 140 determines whether or not to end the prediction process. If it is determined in step S1606 that the prediction process should be continued (NO in step S1606), the process returns to step S1601.

[0119] On the other hand, if it is determined in step S1606 that the prediction process is to be ended, the prediction process is ended.

[0120] <Summary> As is clear from the above explanation, the 3D image learning device 110 according to the first embodiment: - acquires 3D image data of a training animal, including pre-administration 3D image data before the administration of a training drug, chemical structure information of the training drug, and normalized values ​​obtained by normalizing the conditions under which the training drug was administered for each condition item; - acquires 3D image data of a training animal, including post-administration 3D image data showing the presence of the training drug at each position after the training drug is administered; - generates training data using the pre-administration 3D image data before the administration of the training drug, chemical structure information, and normalized values ​​as input data and the post-administration 3D image data after the administration of the training drug as correct answer data; - performs a learning process on a 3D image generation model using the generated training data to generate a trained 3D image generation model.

[0121] In this way, the 3D image learning device 110 according to the first embodiment trains the 3D image generation model on the relationship between the normalized values ​​obtained by normalizing the factors affecting pharmacokinetics and the post-administration 3D image data. This allows the 3D image learning device 110 according to the first embodiment to reflect individual differences in the learning of the 3D image generation model. Furthermore, the 3D image learning device 110 according to the first embodiment can handle factors affecting pharmacokinetics equally regardless of species differences.

[0122] Furthermore, the 3D image prediction device 140 according to the first embodiment: causes the trained 3D image generation model to perform a prediction process by inputting standard 3D image data of the predicted target animal, chemical structure information of the predicted target drug, and normalized values ​​obtained by normalizing the conditions for administering the trained drug for each condition item. As a result, the trained 3D image generation model predicts predicted 3D image data indicating the presence of the predicted target drug at each position of the predicted target animal. The predicted predicted 3D image data is displayed as an image.

[0123] In this way, the 3D image prediction device 140 according to the first embodiment predicts predicted 3D image data by inputting normalized factors that affect pharmacokinetics and inputting standard 3D image data. As a result, the 3D image prediction device 140 according to the first embodiment can make predictions according to individual differences as well as species differences when predicting pharmacokinetics.

[0124] As a result, the three-dimensional image prediction device 140 according to the first embodiment can improve the prediction accuracy when predicting the pharmacokinetics of a drug at its site of action.

[0125] [Second Embodiment] In the first embodiment, the organ, tissue, or cell is described as being a brain, brain tissue, or brain cell, but the organ, tissue, or cell is not limited to a brain, brain tissue, or brain cell. It may be an organ, tissue, or cell other than the brain. Examples of such organs, tissues, or cells include liver (e.g., hepatic parenchymal cells, Kupffer cells, hepatic stellate cells, sinusoidal endothelial cells, bile duct epithelial cells), kidney (e.g., renal tubular epithelial cells, vascular endothelial cells, mesangial cells), adrenal gland (e.g., adrenal microvascular endothelial cells, adrenal cortical cells, chromaffin cells), muscle (e.g., myocytes, nerve cells, glial cells), heart (e.g., cardiac myocytes, vascular endothelial cells, vascular smooth muscle cells, fibroblasts), lung (e.g., alveolar epithelial cells, vascular endothelial cells, pericytes, pulmonary fibroblasts), spinal cord (e.g., nerve cells, glial cells), pancreas (e.g., α cells, β cells, δ cells), small intestine (e.g., absorptive epithelial cells, goblet cells, endocrine cells, Paneth cells), large intestine (e.g., absorptive epithelial cells, goblet cells, endocrine cells, Paneth cells), and stomach (e.g., surface mucous cells, glandular cells, endocrine cells). , spleen (e.g., granulocytes, monocytes, macrophages, dendritic cells, NK cells, T cells, and B cells), bladder (e.g., transitional epithelial cells, bladder cells), testis (e.g., Leydig cells, Sertoli cells, spermatogonia), ovary (e.g., oocytes, luteal cells), skin (e.g., epidermal cells, keratinocytes, spinous cells, granulocytes, keratinocytes), thymus (e.g., thymocytes, thymic epithelial cells), thyroid (e.g., follicular cells, parafollicular cells), bone (e.g., osteocytes, osteoblasts, osteoclasts), bone marrow (e.g., mesenchymal stem cells, hematopoietic stem cells), fat (e.g., white adipocytes, brown adipocytes), eye (e.g., cone cells, rod cells), esophagus (e.g., surface epithelial cells), lymph nodes (e.g., granulocytes, monocytes, macrophages, dendritic cells, NK cells, T cells, and B cells), lymphatic vessels, blood vessels (e.g., vascular endothelial cells).

[0126] In addition, in the first embodiment, the condition items shown in FIG. 2 are listed as factors that affect pharmacokinetics, but the factors that affect pharmacokinetics may include condition items other than the condition items shown in FIG. 2.

[0127] In the first embodiment, the normalization process is performed using the normalization method shown in Fig. 6 or 13, but normalization processes other than the normalization method shown in Fig. 6 or 13 may be performed. For example, the calculation unit 1011 and the calculation unit 1510 may normalize the values ​​of the condition items based on the ranges (maximum values) that the values ​​of the condition items can take.

[0128] In the first embodiment, the three-dimensional image generating device 130 uses the image data shown in Fig. 7 to generate the pre-administration three-dimensional image data and the post-administration three-dimensional image data. However, the image data used to generate the pre-administration three-dimensional image data and the post-administration three-dimensional image data is not limited to these.

[0129] Furthermore, in the first embodiment, a case has been described in which standard 3D image data is input to the 3D image prediction device 140. However, when pre-administration 3D image data can be generated for the predicted target animal, the pre-administration 3D image data may be input instead of the standard 3D image data.

[0130] In the first embodiment, when the type of the learning target animal is different from the type of the prediction target animal, the prediction target condition item information is input with approximately the same value as the learning target condition item information. However, even when the type of the learning target animal is different from the type of the prediction target animal, a value different from the learning target condition item information may be input as the prediction target condition item information.

[0131] Furthermore, in the first embodiment, the case where the learning target drug and the prediction target drug are the same has been described, but the learning target drug and the prediction target drug may be different.

[0132] In the first embodiment, the 3D image learning device 110 and the 3D image prediction device 140 are configured as separate devices. However, the 3D image learning device 110 and the 3D image prediction device 140 may be configured as an integrated device.

[0133] In the first embodiment, the three-dimensional image generation device 130 and the three-dimensional image learning device 110 are configured as separate devices. However, the three-dimensional image learning device 110 and the three-dimensional image generation device 130 may be configured as an integrated device.

[0134] Furthermore, in the first embodiment, the 3D image learning device 110 and the 3D image prediction device 140 are described as being realized by a single computer, but they may also be realized by multiple computers, taking the form of distributed computing.

[0135] Furthermore, in the first embodiment, a method of downloading and installing the 3D image learning program or the 3D image prediction program to the auxiliary storage device 903 was mentioned as an example of a method of installing the program via a network (not shown). Although no particular reference was made to the download source, when installing using such a method, the download source may be, for example, a server device that stores the 3D image learning program or the 3D image prediction program in an accessible manner. Furthermore, the server device may be, for example, a device that downloads the 3D image learning program or the 3D image prediction program via the network 920 or 1420, subject to a fee. In other words, the server device may be a device that provides a service of providing the 3D image learning program or the 3D image prediction program on the cloud.

[0136] Furthermore, although the above embodiments do not mention details of the 3D image learning model, the 3D image learning model used in the above embodiments may be a machine learning model including deep learning, and may be, for example, any of the following: - RNN (Recurrent Neural Network), - LSTM (Long Short-Term Memory), - CNN (Convolutional Neural Network), - R-CNN (Region-based Convolutional Neural Network), - GAN (Generative Adversarial Network), - VAE (Variational Autoencoder), - various diffusion models, etc.

[0137] The present invention is not limited to the configurations described in the above embodiments, but may be combined with other elements, etc. These aspects can be changed without departing from the spirit of the present invention, and can be appropriately determined depending on the application form.

[0138] This application claims priority based on Japanese Patent Application No. 2023-145120, filed on September 7, 2023, the entire contents of which are incorporated herein by reference.

[0139] 100: Pharmacokinetic analysis system 110: 3D image learning device 120: Various measuring devices 130: 3D image generation device 140: 3D image prediction device 151: Drug 152: Animal 161: Drug 162: Animal 800: Learning data 1010: Generation unit 1011: Calculation unit 1012: Structural information generation unit 1020: Learning unit 1021: 3D image generation model 1510, 1511: Calculation unit 1520, 1521: Structural information generation unit 1530: Prediction unit 1531: Trained 3D image generation model 1540: Display unit

Claims

1. A three-dimensional image prediction device having: a prediction unit that predicts three-dimensional information indicating the presence of the predicted target drug at each position within the organ, tissue or cell of the prediction target animal by inputting three-dimensional information of the unmedicated organs, tissues or cells of a prediction target animal, chemical structure information of the prediction target drug, and normalized values ​​for each condition item for the prediction target animal into a trained model trained using training data including three-dimensional information of a training target animal, the three-dimensional information being of an unmedicated organ, tissue or cell, chemical structure information of the training target drug, and normalized values ​​for each condition item for the prediction target animal when the training target drug was administered, and three-dimensional information of the training target animal, the three-dimensional information being of an unmedicated organ, tissue or cell, chemical structure information of the training target drug, and normalized values ​​for each condition item for the prediction target animal; and a display unit that displays an image of the predicted three-dimensional information.

2. The three-dimensional image prediction device according to claim 1, further comprising a calculation unit that normalizes the values ​​of the respective condition items and calculates normalized values.

3. The three-dimensional image prediction device according to claim 1, wherein the organ, tissue or cell is a brain, brain tissue or brain cell.

4. A three-dimensional image learning device having: a generation unit that generates learning data including three-dimensional information of a training subject animal, the three-dimensional information being three-dimensional information of unmedicated organs, tissues or cells, chemical structure information of the training subject drug, normalized values ​​obtained by normalizing the conditions under which the training subject drug was administered for each condition item, and three-dimensional information of the training subject animal, the three-dimensional information being indicative of the presence of the training subject drug at each position within the organ, tissue or cell after the training subject drug was administered; a learning unit that uses the generated learning data to perform learning processing on a model and generate a trained model; and a memory unit that stores the generated trained model.

5. A three-dimensional image prediction method in which a computer executes the following steps: a prediction step of predicting three-dimensional information indicating the presence of the predicted target drug at each position within the organ, tissue or cell of a prediction target animal by inputting three-dimensional information of a prediction target animal's unmedicated organs, tissues or cells, chemical structure information of a prediction target drug, and normalized values ​​for each condition item for the prediction target animal into a trained model trained using training data including three-dimensional information of a training target animal, the three-dimensional information being three-dimensional information of an unmedicated organ, tissue or cell, chemical structure information of a training target drug, and normalized values ​​for each condition item for the prediction target animal when the training target drug is administered, the three-dimensional information being three-dimensional information indicating the presence of the training target drug at each position within the organ, tissue or cell of the prediction target animal after the training target drug is administered; and a display step of displaying an image of the predicted three-dimensional information.

6. A three-dimensional image prediction program for causing a computer to execute the following steps: a prediction step of predicting three-dimensional information indicating the presence of the predicted target drug at each position within the organ, tissue or cell of a prediction target animal by inputting three-dimensional information of a prediction target animal's unmedicated organs, tissues or cells, chemical structure information of a prediction target drug, and normalized values ​​for each condition item for the prediction target animal into a trained model trained using training data including three-dimensional information of a training target animal, the three-dimensional information being three-dimensional information of an unmedicated organ, tissue or cell, chemical structure information of a training target drug, and normalized values ​​for each condition item for the prediction target animal when the training target drug is administered, the three-dimensional information being three-dimensional information indicating the presence of the training target drug at each position within the organ, tissue or cell of a prediction target animal; and a display step of displaying an image of the predicted three-dimensional information.

7. A three-dimensional image learning method executed by a computer, comprising: a generation step of generating training data including three-dimensional information of a training subject animal, the three-dimensional information being three-dimensional information of an unmedicated organ, tissue or cell, chemical structure information of the training subject drug, normalized values ​​obtained by normalizing the conditions under which the training subject drug was administered for each condition item, and three-dimensional information of the training subject animal, the three-dimensional information being indicative of the presence of the training subject drug at each position within the organ, tissue or cell after the training subject drug was administered; a learning step of performing a learning process on a model using the generated training data to generate a trained model; and a storage step of storing the generated trained model in a storage unit.

8. A three-dimensional image learning program for causing a computer to execute the following steps: a generation process for generating training data including three-dimensional information of a training subject animal, the three-dimensional information being three-dimensional information of an unmedicated organ, tissue or cell, chemical structure information of the training subject drug, normalized values ​​obtained by normalizing the conditions under which the training subject drug was administered for each condition item, and three-dimensional information of the training subject animal, the three-dimensional information being three-dimensional information indicating the presence of the training subject drug at each position within the organ, tissue or cell after the training subject drug was administered; a learning process for performing a learning process on a model using the generated training data to generate a trained model; and a storage process for storing the generated trained model in a storage unit.

Images