Information processing system, information processing method, program, trained model, and method for generating a trained model.

Abstract

This invention provides an information processing system, method, program, trained model, and method for generating the same, which are all necessary for obtaining more interpretable scintigraphy results. [Solution] The method obtains an estimated medical image by inputting input data, including input medical images, into a trained model. The input medical image is image data showing the distribution of radioactivity in the body of a subject who has been administered a radiopharmaceutical. The estimated medical image is image data showing the estimated distribution of radioactivity in the body of a subject who has been administered a radiopharmaceutical, assuming that the subject's waiting time is a reference time. The trained model is trained by performing machine learning using a set of training data. The training data includes ground truth medical images showing the distribution of radioactivity in the body of a reference subject, taken after a reference time has elapsed since the administration of the radiopharmaceutical, and reference medical images showing the distribution of radioactivity in the body of a reference subject, taken after a reference waiting time different from the reference time has elapsed since the administration of the radiopharmaceutical.

Description

Technical Field

[0001] The present invention relates to an information processing system, an information processing method, a program, a learned model, and a method for generating a learned model.

Background Art

[0002] Patent Document 1 discloses a technique for appropriately processing medical images captured by different types of imaging devices in analysis using a learned model. The abnormal accumulation detection device according to this technique includes an input unit that inputs a bone scintigram of a subject captured by an imaging device, an inverse filtering processing unit that converts a device-specific bone scintigram subjected to noise removal processing unique to the imaging device into a medical image before noise removal, an abnormal accumulation detection processing unit that applies the bone scintigram before noise removal to a learned model for abnormal accumulation detection to infer abnormal accumulation, and an output unit that outputs data indicating an abnormal location.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Incidentally, when performing scintigraphy, a certain waiting time is required after administering a radiopharmaceutical to the subject of the examination until the radiopharmaceutical appropriately accumulates in the target organ. Since the image obtained by scintigraphy changes depending on the length of this waiting time, it is desirable that this waiting time be standardized to a predetermined standard time in order to compare it with the results of other examinations. However, due to various practical constraints such as operational constraints of the medical institutions performing the examinations and time constraints of the subjects, it is actually difficult to standardize the waiting time to this standard time, and such variability in waiting times may reduce the interpretability of scintigraphy results. [Means for solving the problem]

[0005] According to one aspect of the present invention, an information processing system comprising at least one processor, the processor configured to execute a program such that the following steps are performed, the image acquisition step involves inputting input data including an input medical image into a trained model to acquire an estimated medical image output from the trained model, the input medical image being image data showing the distribution of radioactivity in the body of a subject who has been administered a radiopharmaceutical, and being imaged after the subject's waiting time, which is a waiting period, has elapsed since the administration of the radiopharmaceutical to the subject, the subject's waiting time being different from a predetermined standard time, and the estimated medical image The system provides image data showing the estimated distribution of radioactivity within a subject who has been administered a radiopharmaceutical, assuming a standard waiting time for the subject to be tested. The trained model is trained by machine learning using a set of training data, and the training data includes ground truth medical images and reference medical images. The ground truth medical images are medical images showing the distribution of radioactivity within a reference subject's body, taken after a standard time has elapsed since the administration of the radiopharmaceutical, and the reference medical images are medical images showing the distribution of radioactivity within a reference subject's body, taken after a reference waiting time different from the standard time has elapsed since the administration of the radiopharmaceutical.

[0006] This configuration allows for more interpretable scintigraphy results. [Brief explanation of the drawing]

[0007] [Figure 1] This figure shows an example of the hardware configuration of an information processing device 1, which is an example of an information processing system. [Figure 2] This figure shows an example of medical images related to bone scintigraphy for each patient, broken down by waiting time for examination. [Figure 3] This is a block diagram showing the flow of information processing performed in the information processing device 1. [Figure 4] This is a conceptual diagram illustrating the training method for a pre-trained model and the method for collecting the training data used for training. [Modes for carrying out the invention]

[0008] 1. Hardware Configuration Figure 1 shows an example of the hardware configuration of an information processing device 1, which is an example of an information processing system. The information processing device 1 comprises a communication unit 11, a storage unit 12, at least one processor 13, a display unit 14, and an input unit 15, and these components are electrically connected within the information processing device 1 via a communication bus 10.

[0009] The communication unit 11 preferably uses wired communication methods such as USB, IEEE1394, Thunderbolt®, and wired LAN network communication, but may also include wireless LAN network communication, mobile communication such as 3G / LTE / 5G, and Bluetooth® communication as needed. In other words, it is more preferable to implement it as a collection of these multiple communication methods. That is, the information processing device 1 may communicate various information from the outside via the communication unit 11 and the network.

[0010] The storage unit 12 stores various types of information as defined above. This can be implemented, for example, as a storage device such as a solid-state drive (SSD) that stores various programs related to the information processing device 1 executed by the processor 13, or as memory such as random access memory (RAM) that stores temporarily necessary information (arguments, arrays, etc.) related to program calculations. The storage unit 12 stores various programs and variables related to the information processing device 1 executed by the processor 13.

[0011] The processor 13 performs processing and control of the overall operation related to the information processing device 1. The processor 13 is, for example, a central processing unit (CPU) not shown. The processor 13 realizes various functions related to the information processing device 1 by reading predetermined programs stored in the memory unit 12. That is, information processing by software stored in the memory unit 12 is concretely realized by the processor 13, which is an example of hardware, and can be executed as each functional unit included in the processor 13. Note that the processor 13 is not limited to a single unit, and may be implemented with multiple processors 13 for each function, or a combination thereof.

[0012] The display unit 14 displays a graphical user interface (GUI) screen that can be operated by the user. Preferably, this is done by using a display device such as a CRT display, liquid crystal display, organic EL display, or plasma display, depending on the type of information processing device 1. The display unit 14 may be included in the housing of the information processing device 1 or it may be an external component.

[0013] The input unit 15 is configured to accept input from the user. The input unit 15 may be included in the housing of the information processing device 1 or it may be external. For example, the input unit 15 may be implemented as a touch panel integrated with the display unit 14. If it is a touch panel, the user can input tap operations, swipe operations, etc. Of course, instead of a touch panel, a switch button, mouse, QWERTY keyboard, voice recognition device, gesture detection device, gaze detection device, biosignal detection device, imaging device, etc. may be used. In other words, the input unit 15 accepts operation input made by the user. In response, the input unit 15 transmits a signal corresponding to the operation input to the processor 13 via the communication bus 10. The processor 13 can perform predetermined controls and calculations as needed.

[0014] Next, an example of the functions of the processor 13 will be described. The processor 13 is configured to acquire information from other devices. The processor 13 is configured to acquire various information by reading various information stored in the storage area, which is at least a part of the memory unit 12, and writing the read information to the work area, which is at least a part of the memory unit 12. The storage area is, for example, the area of ​​the memory unit 12 that is implemented as a storage device such as an SSD. The work area is, for example, the area that is implemented as memory such as RAM. The acquisition by the processor 13 includes acquiring the output results of each functional unit included in the processor 13.

[0015] The processor 13 is configured to display various types of information. This information can be presented to the user via the display unit 14 or other devices. In such cases, for example, the processor 13 controls the display unit 14 to display visual information such as screens, images including still images or videos, icons, and messages. The processor 13 may generate only rendering information for displaying the visual information on the display unit 14. The processor 13 may also present the outputted information to the user without going through the display unit 14 or other devices.

[0016] The information processing device 1 (particularly the processor 13) is configured to obtain an estimated medical image IM2 output from a trained model by inputting input data, including the input medical image IM1, into the trained model.

[0017] At least one trained model is stored in the memory unit 12 and is trained using machine learning to output an estimated medical image IM2 by inputting input data, including an input medical image, into the trained model. The trained model can be represented, for example, using a neural network model having an input layer, a hidden layer, and an output layer. In this embodiment, the trained model includes a first trained model M1 and a second trained model M2.

[0018] The first trained model M1 is configured to output an estimated medical image by correcting the input medical image to advance the patient's waiting time into the future, based on the input medical image in which the patient's waiting time is shorter than the reference time.

[0019] The second trained model M2 is configured to output an estimated medical image by correcting the input medical image to move the patient's waiting time back in time based on the input medical image where the patient's waiting time is longer than the reference time. In this embodiment, the first trained model M1 and the second trained model M2 are generated as separate trained models that are trained independently of each other.

[0020] A radiopharmaceutical is a pharmaceutical product that uses a radioisotope (RI). By administering it to the human body, it is prepared to selectively accumulate a nuclide that emits radiation (e.g., X-rays or γ-rays) in a desired site (bone, brain, heart, kidney, liver, etc.). The radiopharmaceutical circulates in the body after administration and accumulates in the desired site. Therefore, for the specific imaging method, any method can be adopted that uses a device capable of detecting the position dependence of radiation, such as a gamma camera, and visualizing it as an image. The medical image is preferably obtained particularly by bone scintigraphy or brain scintigraphy. For example, in this embodiment, the radiopharmaceutical is for performing bone scintigraphy, and the medical image such as the input medical image shows the result of bone scintigraphy.

[0021] The input medical image is image data (medical image) showing the distribution of radioactivity in the body of the examinee to whom the radiopharmaceutical has been administered. For example, the input medical image is taken after the passage of the examinee waiting time. The examinee waiting time is the waiting time from when the radiopharmaceutical is administered to the examinee until imaging, and is different from, for example, a predetermined reference time.

[0022] The estimated medical image is image data showing the estimated result of the distribution of radioactivity in the body of the examinee to whom the radiopharmaceutical has been administered when the examinee waiting time is assumed to be the reference time. That is, the learned model in this embodiment functions to estimate the medical image when the examinee waiting time is a predetermined reference time from the medical image of bone scintigraphy taken at an examinee waiting time different from the reference time. The reference time is, for example, a time recommended for obtaining a quantitative evaluation index related to bone scintigraphy, such as the bone scan index (BSI), artificial neural network (ANN) value, hot spot number (HSN), etc., for example, 3 hours. Note that the reference time is not limited to this and can be arbitrarily set according to the type of medical image, the disease to be diagnosed, the quantitative index used, etc.

[0023] 2. Regarding the transition of the medical image with the change of the examinee waiting time Next, we will explain the changes in medical images in relation to changes in the waiting time of the subjects being examined. Figure 2 shows an example of medical images related to bone scintigraphy for each subject being examined, based on their waiting time. The medical images in Figure 2 show the results of bone scintigraphy for the subjects, and the numerical value representing the time indicated above each medical image represents the waiting time of the subject after administration of the radiopharmaceutical for bone scintigraphy. Among these, organizations such as the Japanese Society of Nuclear Medicine Technology recommend taking images 3 hours after administration of the radiopharmaceutical in order to calculate quantitative indicators such as BSI. Therefore, the medical image corresponding to "180 min" here is the "medical image taken at the reference time," that is, the medical image to be obtained using the trained model.

[0024] As shown in Figure 2, the density (i.e., the amount of radiation) in the bone area increases (especially before 120 minutes). One theory suggests that this is due to the process by which the radiopharmaceutical circulates throughout the body via the blood vessels and accumulates in the bones throughout the body for a period of time after it has been injected into the bloodstream. On the other hand, for example, medical images taken when the patient waiting time for examination is 300 minutes tend to show a decrease in the density (i.e., the amount of radiation) in the bone area overall compared to when the patient waiting time is 180 minutes. One theory suggests that this is partly due to the gradual decrease in the amount of radiation emitted from the radionuclides accumulated in the bones over time, due to factors such as the physical half-life of the radionuclides and their excretion by the body's metabolic functions. From this perspective, since the main factors related to the change in radiation dose may differ, it is preferable, as mentioned above, that at least one trained model includes a first trained model M1 and a second trained model M2, each trained independently.

[0025] 3. About the flow of information processing Figure 3 is a block diagram illustrating the flow of information processing performed in the information processing device 1. As an example, we will explain the case where two types of input data, D1a and D1b, are input to the information processing device 1. In this section, the waiting time for the subject to be examined is represented by T, and the reference time is represented by Tb. The medical image will be the result of bone scintigraphy, as in the previous section, and the reference time Tb will be 3 hours.

[0026] As shown in Figure 3, first, the processor 13, acting as an image acquisition unit, acquires input medical images IM1a and IM1b from users such as medical professionals.

[0027] Each of the input data sets D1a and D1b contains the input medical images IM1a and IM1b. Input medical image IM1a is a medical image of the subject taken during a subject waiting time T1a that is shorter than the reference time Tb. Input medical image IM1b is a medical image of the subject taken during a subject waiting time T1b that is longer than the reference time Tb (e.g., 4 hours). The medical image contains positional information (pixel coordinates) within the captured image and information about the gamma ray dose detected at each pixel coordinate. The higher the dose, the darker the color displayed at the corresponding position in the medical image.

[0028] Next, the processor 13, acting as a waiting time acquisition unit, acquires information regarding the patient waiting times T1a and T1b corresponding to the input medical images IM1a and IM1b. For the sake of explanation, the information regarding the patient waiting times T1a and T1b will be referred to as time information T1a and T1b. Time information T1a and T1b are information that shows the relationship between the reference time Tb and the patient waiting times T1a and T1b, respectively, and are, for example, the patient waiting times T1a and T1b themselves. Note that time information T1a and T1b may be the difference between the patient waiting times T1a and T1b and the reference time Tb, or it may simply be information that shows whether the patient waiting times T1a and T1b are longer or shorter than the reference time Tb (for example, the sign of the difference).

[0029] The processor 13 generates input data D1a and D1b by associating the acquired input medical images IM1a and IM1b with the corresponding time information T1a and T1b. If the input medical images IM1a and IM1b are already associated with the time information T1a and T1b, the processor 13 may acquire the associated input medical images IM1a and IM1b and the time information T1a and T1b as input data D1a and D1b. In other words, the processor 13 may also acquire information regarding the patient waiting times T1a and T1b corresponding to the input medical images IM1a and IM1b. In this case, the input data D1a and D1b may further include information regarding the patient waiting times T1a and T1b corresponding to the input medical images IM1a and IM1b. With this configuration, the circulation process of radiopharmaceuticals in the body can be modeled more accurately, and thus more accurate estimated medical images can be obtained. From another perspective, with this configuration, estimated medical images are output from a trained model based on input data D1a, D1b, which associates input medical images IM1a, IM1b with patient waiting times T1a, T1b. Therefore, compared to cases where patient waiting times T1a, T1b are not associated with input medical images IM1a, IM1b (for example, when patient waiting times T1a, T1b are unknown), estimated medical images that are more reliable and interpretable for the user can be output.

[0030] Next, the processor 13 performs a determination process based on the input data D1a and D1b. The determination process determines whether to input the input data D1a and D1b to the first trained model M1 or the second trained model M2. Here, the processor 13 inputs the input data to the first trained model M1 if the acquired time information indicates that the subject's waiting time is shorter than the reference time Tb. On the other hand, the processor 13 inputs the input data to the second trained model M2 if the acquired time information indicates that the subject's waiting time is longer than the reference time Tb. With this configuration, even if the factors contributing to the change in the distribution of radioactivity may differ depending on whether the waiting time is shorter or longer than the reference time Tb, the possibility of obtaining a more reliable estimated medical image IM2 can be increased by changing the trained model to which the input data D1a and D1b are input based on the magnitude of the waiting time relative to the reference time Tb. Here, since (T1a-Tb)<0 and (T1b-Tb)>0, the processor 13 inputs the input data D1a to the first trained model M1 and the input data D1b to the second trained model M2. As a result, output data D2, including the estimated medical image IM2, is output from the trained model into which the input data was received (the first trained model M1 if input data D1a was obtained, and the second trained model M2 if input data D1b was obtained). In other words, the processor 13 obtains the output estimated medical image IM2 by inputting the input data into either the first trained model M1 or the second trained model M2 based on the acquired time information T1a and T1b. With this configuration, the factors contributing to the change in the distribution of radioactivity may differ depending on whether the waiting time is shorter or longer than the reference time Tb. Therefore, by inputting the input data D1a and D1b into the trained model based on the reference time Tb, it is possible to increase the likelihood of obtaining a more reliable estimated medical image IM2.

[0031] The processor 13 may calculate the patient waiting time corresponding to the estimated medical image IM2 based on the output estimated medical image IM2 and obtain it as output data D2. For example, the processor 13 may extract features that correlate with the patient waiting time from the estimated medical image IM2 output from the trained model using an extractor capable of extracting such features, and output an estimated result of the patient waiting time based on these features. Ideally, this estimation result will be in agreement with the reference time Tb, but there may be some discrepancies. Therefore, this estimation result can be used as an indicator to evaluate the reliability of the estimated medical image IM2.

[0032] Furthermore, the processor 13 may calculate evaluation indices related to scintigraphy based on the output estimated medical image IM2 and acquire them as output data D2. The evaluation indices related to the disease can be arbitrary, such as BSI, ANN, or HSN. These evaluation indices can be used, for example, as reference information when diagnosing a disease.

[0033] With the above configuration, even if there is variation in the waiting time before imaging, an estimated medical image IM2 corresponding to a predetermined reference time Tb can be obtained, thus enabling the acquisition of more interpretable scintigraphy results within the practical constraints of medical institutions, etc. The output data D2 is presented to the user, for example, via the display unit 14.

[0034] 4. Learning Methods Next, we will describe an example of a method for collecting training data to generate the above-mentioned pre-trained model, and a method for training using the collected set of training data. Figure 4 is a conceptual diagram illustrating the training method for the pre-trained model and the method for collecting training data used for training.

[0035] First, the pre-trained model described above is trained by machine learning using a set of training data that includes ground truth medical images and reference medical images, which have been collected in advance. The reference medical image is a medical image showing the distribution of radioactivity within the body of a reference subject, taken after a reference waiting time different from the reference time Tb has elapsed since the administration of a radiopharmaceutical. The ground truth medical image is a medical image showing the distribution of radioactivity within the body of a reference subject, taken after a reference time Tb has elapsed since the administration of a radiopharmaceutical. Thus, the pre-trained model is defined by pairs of image data corresponding to each other: "reference waiting time, reference medical image" and "reference time Tb, ground truth medical image". The reference waiting time can be longer or shorter than the reference time Tb, as long as it is different from the reference time Tb. Here, the reference medical image and ground truth medical image are obtained by taking multiple images from the same reference subject after a single administration of a radiopharmaceutical.

[0036] In this embodiment, since at least one trained model includes a first trained model M1 and a second trained model M2, the content of the training data for each model and the method for collecting it will be described.

[0037] The first trained model M1 is trained using the first training data set TD1. The first training data set TD1 includes the first reference medical image, which is a medical image showing the distribution of radioactivity within a reference subject's body, taken after a reference waiting time shorter than the reference time Tb following the administration of a radiopharmaceutical.

[0038] The second trained model M2 is trained using the second set of training data TD2. The second training data TD2 includes the second reference medical image IMG3, which is a medical image showing the distribution of radioactivity within a reference subject's body, taken after a reference waiting time longer than the reference time Tb following the administration of a radiopharmaceutical.

[0039] In this embodiment, the training data (first training data TD1 and second training data TD2) each further includes information regarding the reference waiting time corresponding to the reference medical image. In Figure 4, a reference waiting time shorter than the reference time Tb is denoted as T1, the first reference medical image is denoted as IMG1, and the correct medical image corresponding to the first reference medical image is denoted as IMG2. Also in Figure 4, a reference waiting time longer than the reference time Tb is denoted as T2, the second reference medical image is denoted as IMG3, and the correct medical image corresponding to the second reference medical image is denoted as IMG4. The letters IMG1 to IMG4 actually represent image data. Here, the information regarding the reference waiting time can be defined in various forms, as with the time information T1a and T1b described above. It should be noted that conventionally, waiting times when obtaining such medical images were merely used as a guideline from the administration of radiopharmaceuticals until imaging equipment such as gamma cameras became usable, and were not associated with medical images in general medical institutions.

[0040] Next, the process for generating (in other words, training) the pre-trained model described above will be explained using Figure 4. First, a group of reference subjects collected in advance are classified into two test groups, Group A and Group B. It is preferable that each reference subject is classified in a way that does not introduce physical bias into each test group.

[0041] Next, each of the reference subjects in group A is administered a radiopharmaceutical, and the first imaging is performed after the first reference waiting time T1 has elapsed. Subsequently, without administering any further radiopharmaceuticals, a second imaging is performed after the reference time Tb has elapsed since the administration of the radiopharmaceutical. The first imaging yields the first reference medical image IMG1, and the second imaging yields the ground truth medical image IMG2. By associating these two images, the first training data TD1 is prepared.

[0042] Meanwhile, each of the reference subjects in group B is administered a radiopharmaceutical, and the first image is taken after a reference time Tb has elapsed since the administration of the radiopharmaceutical. Then, without administering any further radiopharmaceuticals, the second image is taken after a second reference waiting time T2 has elapsed. The first image yields the ground truth medical image IMG4, and the second image yields the second reference medical image IMG3. By associating these two images, the second training data TD2 is prepared. In this way, as a preparation step, a set of training data is prepared by associating ground truth medical images obtained by taking multiple images after administering a radiopharmaceutical to the same reference subject with the reference medical image. Here, the first training data TD1 is obtained by taking multiple images of the reference subject belonging to group A, and the second training data TD2 is obtained by taking multiple images of the reference subject belonging to group B. By performing this for each reference subject, a set of the first training data TD1 and a set of the second training data TD2 are obtained. Furthermore, the preparation of training data is not limited to dividing the subjects into multiple groups such as Group A and Group B. For example, a reference waiting time may be assigned to each of the pre-collected reference subjects so that the sample size relative to the reference waiting time shows the desired distribution, and multiple images may be taken for each person.

[0043] In this embodiment, a reference waiting time is further associated with the reference medical image obtained through the multiple imaging described above. This prepares sets of training data TD1 and TD2. With this configuration, it is possible to increase the likelihood of obtaining an estimated medical image IM2 with higher reliability based on the waiting time. The number of data points in each of the first training data set TD1 and the second training data set TD2 can be set appropriately within a realistic range.

[0044] Next, as a generation step, machine learning is performed using a set of training data so that the correct medical images are used as correct labels, and a trained model is generated. The specific machine learning algorithm is arbitrary, but for example, an algorithm that applies supervised learning to a Generative Adversarial Network (GAN) can be used. When applying supervised learning to a GAN, for example, processor 13 performs supervised training of the discriminator using a set of training data in which reference medical images are incorrect and correct medical images are correct, while simultaneously training the generator and discriminator of the trained model so that when a reference medical image is input to the generator, an estimated medical image is output, the discriminator discriminates the estimated medical image, and outputs the medical image that has been determined to be captured at reference time Tb. Examples of such algorithms include Pix2Pix and conditional GAN ​​(CGAN).

[0045] Furthermore, creating a pre-trained model is not limited to creating a new pre-trained model; it can also include updating a pre-trained model, such as through transfer learning of an existing model or fine-tuning of a pre-trained model.

[0046] [others] The above embodiments may be provided, for example, in the following forms.

[0047] The processor 13 may further acquire information on the metabolic function of the subject of examination and information on the dose decay pattern, such as information on the radiopharmaceutical, and correct the input medical image based on this information. Information on metabolic function can be arbitrary, such as age, sex, basal metabolic rate, and values ​​related to excretion (especially urination). Information on radiopharmaceuticals may include arbitrary information that may be related to the half-life, such as the name of the radiopharmaceutical, the dosage, and the type of nuclide. The processor 13 may input this information as one of the elements of the input data into a trained model to obtain an estimated medical image that has been further corrected using the information on the decay pattern. Alternatively, the processor 13 may input the above-mentioned input medical image (and time information) into a trained model and further correct the output estimated medical image using the information on the decay pattern. With such a configuration, the circulation process of the radiopharmaceutical in the body according to the waiting time can be modeled with greater accuracy, and a more accurate estimated medical image IM2 can be obtained.

[0048] In the above embodiment, the processor 13 outputs an estimated medical image by inputting the patient waiting time as input data in addition to the input medical image to the trained models M1 and M2. However, it is not necessary to input the patient waiting time to the trained models M1 and M2. In this case, the patient waiting time may be used, for example, as a parameter to determine which of the first trained model M1 and the second trained model M2 to input the input data to. In this case, each trained model M1 and M2 may be trained to extract features that have a correlation with the waiting time based on the input medical image using an extractor configured to extract such features, and to output an estimated medical image using these features and the input medical image.

[0049] In the above embodiment, the memory unit 12 stored both the first trained model M1 and the second trained model M2 as multiple trained models, but it may also store only one of the first trained model M1 or the second trained model M2. That is, at least one trained model may include either the first trained model M1 or the second trained model M2. With such a configuration, it is possible to use a trained model suitable for the waiting time taken, thereby improving the accuracy of the estimated medical image IM2.

[0050] In the above embodiment, the case where there is at least one trained model is described as two trained models M1 and M2, but the number of trained models is not limited to two. For example, there may be three or more trained models. In this case, sets of training data may be created so that the waiting time ranges of the training data used for each trained model do not overlap, and each trained model may be trained. This makes it possible to obtain more accurate estimated medical images. Alternatively, there may be only one trained model.

[0051] If the trained models include a first trained model M1 and a second trained model M2, the processor 13 may acquire information (e.g., time information T1a) regarding the patient waiting time corresponding to the input medical images IM1a and IM1b, input one input data (e.g., input data D1a) containing this time information to each of the trained models M1 and M2, and acquire estimated medical images IM2 output from the first trained model M1 and the second trained model M2, respectively. In this case, the processor 13 may present all of the acquired estimated medical images IM2 to the user and have the user identify the estimated medical image that is closer to the reference time Tb, or the processor 13 may estimate the patient waiting time of the estimated medical images IM2 and present only the estimated medical images whose estimation result is closer to the reference time Tb to the user. In other words, the processor 13 may obtain the output estimated medical image IM2 by inputting the input data D1a and D1b, which include the acquired subject waiting time T1a and the input medical image IM1a, into at least one of the first trained model M1 and the second trained model M2. With such a configuration, the factors contributing to the change in the distribution of radioactivity may differ depending on whether the waiting time is shorter or longer than the reference time Tb. Therefore, by inputting the input data D1a and D1b into the trained model based on the reference time Tb, it is possible to increase the likelihood of obtaining an estimated medical image IM2 with higher reliability.

[0052] In the above embodiment, the first trained model M1 and the second trained model M2 were separate trained models, but they may be implemented as a single trained model. For example, the trained model may be configured to output an estimated medical image IM2 by inputting input data D1, which includes an input medical image, to the trained model, regardless of the length of the patient's waiting time relative to the reference time, by performing machine learning using the entire data set, which includes the first set of training data and the second set of training data.

[0053] In the above embodiment, the training data includes reference waiting time, and machine learning of the trained models M1 and M2 was performed using the reference waiting time. However, the reference waiting time does not necessarily have to be used for machine learning of the trained models. For example, a feature that has a correlation with the previously trained reference waiting time may be extracted from the reference medical images included in the training data using an extractor capable of extracting such feature, and machine learning of the trained models M1 and M2 may be performed using these feature and the reference medical images.

[0054] An information processing system may consist of one or more devices or components. Therefore, an information processing system is not limited to the information processing device 1 alone as described above, but may include other devices other than the information processing device 1 (for example, an information processing server that directly exchanges information with a trained model, or a server for training the trained model).

[0055] The embodiments described above are not limited to an information processing system, but may also be an information processing method or a program. An information processing method includes each step of an information processing system. A program causes at least one computer to execute each step of an information processing system. Alternatively, the information processing system may be provided as a trained model that functions as a kind of program. The trained model causes at least one computer to function to output an estimated medical image output from the trained model by taking input data, including an input medical image, as input data.

[0056] The above-mentioned information processing system, etc., may be provided in any of the following forms.

[0057] (1) An information processing system comprising at least one processor, wherein the processor is configured to execute a program such that the following steps are performed, the image acquisition step involves inputting input data including an input medical image into a trained model to acquire an estimated medical image output from the trained model, the input medical image being image data showing the distribution of radioactivity in the body of a subject who has been administered a radiopharmaceutical, the image being taken after the subject has been administered the radiopharmaceutical and the subject has been waiting for a waiting period, which is a waiting period for the subject, and the subject waiting period is different from a predetermined standard time, and the estimated medical image is the subject The system provides image data showing the estimated distribution of radioactivity within the body of the subject who has been administered the radiopharmaceutical, assuming the subject waiting time is the standard time, the trained model is trained by machine learning using a set of training data, the training data includes a ground truth medical image and a reference medical image, the ground truth medical image is a medical image showing the distribution of radioactivity within the body of the reference subject, taken after the standard time has elapsed since the administration of the radiopharmaceutical, and the reference medical image is a medical image showing the distribution of radioactivity within the body of the reference subject, taken after a reference waiting time different from the standard time has elapsed since the administration of the radiopharmaceutical.

[0058] With this configuration, even if there is variation in the waiting time before imaging, it is possible to obtain estimated medical images corresponding to a predetermined reference time, thus enabling the acquisition of more interpretable scintigraphy results within the practical constraints of medical institutions and other facilities.

[0059] (2) The information processing system described in (1) above, wherein the trained model includes a first trained model or a second trained model, the first trained model is trained using a first set of training data, the first trained model includes a first reference medical image as the reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time shorter than the reference time has elapsed from the administration of the radiopharmaceutical, and the second trained model is trained using a second set of training data, the second reference medical image as the reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time longer than the reference time has elapsed from the administration of the radiopharmaceutical.

[0060] This configuration allows for the use of a pre-trained model suited to the waiting time, thereby improving the accuracy of estimated medical images.

[0061] (3) In the information processing system described in (1) or (2) above, the training data further includes information relating to the reference waiting time corresponding to the reference medical image, the waiting time acquisition step acquires information relating to the patient waiting time corresponding to the input medical image, and the input data further includes information relating to the patient waiting time corresponding to the input medical image.

[0062] This configuration allows for a more accurate modeling of the circulation process of radiopharmaceuticals within the body, thereby enabling the acquisition of more accurate estimated medical images.

[0063] (4) In the information processing system described in any one of (1) to (3) above, the trained model includes a first trained model and a second trained model, the first trained model is trained using a first set of training data, the first trained model includes a first reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time shorter than the reference time from the administration of the radiopharmaceutical, and the second trained model is trained using a first set of training data, the first reference medical image being taken after a reference waiting time shorter than the reference time from the administration of the radiopharmaceutical A system that is trained using a second set of training data, which includes a second reference medical image, a medical image showing the distribution of radioactivity within the body of the reference subject, captured after the aforementioned reference waiting time has elapsed; further, in the waiting time acquisition step, information regarding the subject's waiting time corresponding to the input medical image is acquired; and in the image acquisition step, based on the acquired information regarding the subject's waiting time, the system acquires the output estimated medical image by inputting the input data into either the first trained model or the second trained model.

[0064] With this configuration, the factors contributing to changes in radioactivity distribution may differ depending on whether the waiting time is shorter or longer than the reference time. Therefore, by inputting input data based on the reference time into a trained model, it is possible to increase the likelihood of obtaining more reliable estimated medical images.

[0065] (5) In the information processing system described in (4) above, in the image acquisition step, if the acquired information regarding the waiting time of the person to be examined indicates that the waiting time of the person to be examined is shorter than the reference time, the input data is input to the first trained model, and if the acquired information regarding the waiting time of the person to be examined indicates that the waiting time of the person to be examined is longer than the reference time, the input data is input to the second trained model.

[0066] With this configuration, the factors contributing to changes in the distribution of radioactivity may differ depending on whether the waiting time is shorter or longer than the reference time. Therefore, by changing the trained model to which the input data is received based on the magnitude of the waiting time relative to the reference time, it is possible to increase the likelihood of obtaining more reliable estimated medical images.

[0067] (6) In the information processing system described in any one of (1) to (5) above, the trained model includes a first trained model and a second trained model, the first trained model is trained using a first set of training data, the first trained model includes a first reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time shorter than the reference time has elapsed from the administration of the radiopharmaceutical, and the second trained model is trained using a first set of training data, the first reference medical image includes a first reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time shorter than the reference time has elapsed from the administration of the radiopharmaceutical, A system that is trained using a second set of training data, which includes a second reference medical image, a medical image showing the distribution of radioactivity within the body of the reference subject, captured after the elapsed reference waiting time; further, in the waiting time acquisition step, information regarding the subject's waiting time corresponding to the input medical image is acquired; and in the image acquisition step, the system acquires the output estimated medical image by inputting the input data, which includes the acquired subject's waiting time and the input medical image, into at least one of the first trained model and the second trained model.

[0068] With this configuration, the factors contributing to changes in radioactivity distribution may differ depending on whether the waiting time is shorter or longer than the reference time. Therefore, by inputting input data based on the reference time into a trained model, it is possible to increase the likelihood of obtaining more reliable estimated medical images.

[0069] (7) An information processing method comprising each step of the information processing system described in any one of (1) to (6) above.

[0070] (8) A program that causes at least one computer to perform each step of the information processing system described in any one of (1) to (6) above.

[0071] (9) A trained model, wherein at least one computer is configured to output an estimated medical image output from the trained model by inputting input data including an input medical image, and the trained model is the trained model described in any one of (1) to (6) above.

[0072] With this configuration, even if there is variation in the waiting time before imaging, it is possible to obtain estimated medical images corresponding to a predetermined reference time, thus enabling the acquisition of more interpretable scintigraphy results within the practical constraints of medical institutions and other facilities.

[0073] (10) A method for generating the trained model described in (9) above, comprising the following steps: a preparation step, which involves associating the ground truth medical images obtained by taking multiple images of the same reference subject after administration of the radiopharmaceutical with the reference medical images to prepare the set of training data; and a generation step, which involves generating the trained model by performing machine learning using the set of training data so that the ground truth medical images are used as ground truth labels.

[0074] (11) A method for generating a trained model as described in (10) above, wherein the preparation step further involves associating the reference waiting time with the reference medical image to prepare the set of training data.

[0075] This configuration increases the likelihood of obtaining more reliable estimated medical images based on the waiting time. Of course, this is not always the case.

[0076] Finally, while various embodiments relating to this disclosure have been described, these are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be implemented in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. [Explanation of Symbols]

[0077] 1: Information Processing Device 10: Communications bus 11: Communications Department 12: Storage section 13: Processor 14:Display section 15: Input section D1a: Input data D1b: Input data D2: Output data IM1a: Input medical image IM1b: Input medical image IM2: Estimated medical image IMG1: First reference medical image IMG2: Correct medical image IMG3: Second reference medical image IMG4: Correct medical image M1: First pre-trained model M2: Second pre-trained model T1a, T1b: Time information T1: First reference wait time T2: Second reference wait time TD1: First training data TD2: Second training data Tb: Reference time

Claims

1. An information processing system, Equipped with at least one processor, The processor is configured to execute a program such that the following steps are performed: In the image acquisition step, the input data, including the input medical image, is input to the trained model, thereby acquiring the estimated medical image output from the trained model. The aforementioned input medical image is image data showing the distribution of radioactivity within the body of a subject who has been administered a radiopharmaceutical, and is captured after the subject has received the radiopharmaceutical and the waiting period has elapsed. The waiting time for the aforementioned test subjects differs from the predetermined standard time. The estimated medical image is image data showing the estimated distribution of radioactivity within the body of the subject who was administered the radiopharmaceutical, assuming that the subject's waiting time is the standard time. The aforementioned trained model is trained by performing machine learning using a set of training data. The aforementioned training data includes correct medical images and reference medical images. The aforementioned correct medical image is a medical image showing the distribution of radioactivity within the body of a reference subject, taken after the administration of a radiopharmaceutical and the subsequent time interval. The system is characterized in that the reference medical image is a medical image showing the distribution of radioactivity within the body of the reference subject, taken after a reference waiting period different from the standard time following the administration of a radiopharmaceutical.

2. In the information processing system described in claim 1, The aforementioned trained model includes a first trained model or a second trained model. The first trained model is trained using a first set of training data, which includes a first reference medical image as the reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time shorter than the reference time has elapsed since the administration of the radiopharmaceutical. The second trained model is trained using a second set of training data, the second reference medical image being a medical image showing the distribution of radioactivity within the body of the reference subject, which is taken after a reference waiting time longer than the reference time has elapsed since the administration of the radiopharmaceutical.

3. In the information processing system described in claim 1, The training data further includes information regarding the reference waiting time corresponding to the reference medical image, In the waiting time acquisition step, information regarding the waiting time of the subject being examined, corresponding to the input medical image, is acquired. The system further includes information regarding the waiting time of the subject being examined, corresponding to the input medical image, as part of the input data.

4. In the information processing system described in claim 1, The aforementioned trained model includes a first trained model and a second trained model. The first trained model is trained using a first set of training data, which includes a first reference medical image as the reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time shorter than the reference time has elapsed since the administration of the radiopharmaceutical. The second trained model is trained using a second set of training data, which includes a second reference medical image as the reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time longer than the reference time has elapsed since the administration of the radiopharmaceutical. Furthermore, in the waiting time acquisition step, information regarding the waiting time of the subject being examined, corresponding to the input medical image, is acquired. In the image acquisition step, the system acquires the output estimated medical image by inputting the input data into either the first trained model or the second trained model, based on the acquired information regarding the waiting time of the person being examined.

5. In the information processing system described in claim 4, In the image acquisition step, If the acquired information regarding the waiting time of the person being tested indicates that the waiting time of the person being tested is shorter than the reference time, the input data is input to the first trained model. A system that inputs the input data into the second trained model when the acquired information regarding the waiting time of the person being tested indicates that the waiting time of the person being tested is longer than the reference time.

6. In the information processing system described in claim 1, The aforementioned trained model includes a first trained model and a second trained model. The first trained model is trained using a first set of training data, which includes a first reference medical image as the reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time shorter than the reference time has elapsed since the administration of the radiopharmaceutical. The second trained model is trained using a second set of training data, which includes a second reference medical image as the reference medical image, which is a medical image showing the distribution of radioactivity in the body of the reference subject, taken after a reference waiting time longer than the reference time has elapsed since the administration of the radiopharmaceutical. Furthermore, in the waiting time acquisition step, information regarding the waiting time of the subject being examined, corresponding to the input medical image, is acquired. In the image acquisition step, the system acquires the output estimated medical image by inputting the input data, which includes the acquired waiting time of the person to be examined and the input medical image, into at least one of the first trained model and the second trained model.

7. Information processing method, A method comprising each step of the information processing system described in any one of claims 1 to 6.

8. It is a program, A program that causes at least one computer to perform each step of the information processing system described in any one of claims 1 to 6.

9. It is a pre-trained model, To enable at least one computer to function in such a way that it outputs an estimated medical image, which is output from the trained model, by taking input data including an input medical image as input data, The trained model is a trained model described in any one of claims 1 to 6.

10. A method for generating a trained model according to claim 9, The following steps are included: In the preparation step, the training data set is prepared by associating the ground truth medical images obtained by taking multiple images after administering the radiopharmaceutical to the same reference subject with the reference medical images. A method for generating a trained model by performing machine learning using the set of training data so that the correct medical images are used as correct labels in the generation step.

11. In a method for generating a trained model according to claim 10, A method for preparing the set of training data in the preparation step, by further associating the reference waiting time with the reference medical image.

Images