Information processing system, information processing device, information processing method, and program

Abstract

To provide an information processing system, an information processing apparatus, an information processing method and a program that generate highly accurate teacher data by performing annotation for learning data on the basis of medical information given to other medical image data corresponding to a subject.SOLUTION: An information processing system according to the present invention comprises: an acquisition unit that acquires first medical information corresponding to PCCT image data which is obtained using a photon counting type radiation detector for a subject and CT image data which is obtained using an energy integration type radiation detector for the subject; a teacher data acquisition unit that acquires first teacher data which is given a first correct label based on the first medical information for the CT image data; and a learning unit that learns a CT inference model which performs predetermined inference for the CT image data using the first teacher data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an information processing system, an information processing apparatus, an information processing method, and a program that generate high-accuracy training data.

Background Art

[0002] In general, training data used for learning an inference unit based on machine learning is generated with the knowledge of an annotator such as an expert for the data.

[0003] For example, an inference unit for inferring the presence or absence of lung cancer from medical image data such as CT images is generated by a learning process based on learning data annotated with correct labels indicating whether lung cancer is present, through observation of the CT image data by an expert.

[0004] Here, Patent Document 1 discloses that an annotator assigns correct labels to candidate training data.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Summary of Invention

Problems to be Solved by the Invention

[0006] However, with the technique described in Patent Document 1, the accuracy of the annotation corresponding to the correct label depends on the annotator. For example, when a feature region that affects inference is overlooked, or the knowledge related to the feature region is insufficient, the learning data may not be appropriately annotated.

[0007] The present invention aims to provide an information processing system, information processing device, information processing method, and program that generate highly accurate training data by annotating training data based on medical information assigned to other medical image data corresponding to a subject. [Means for solving the problem]

[0008] To solve the above problems, an information processing system according to one aspect of the present invention includes an acquisition unit that acquires first medical information corresponding to PCCT image data obtained from a subject using a photon counting type radiation detector, and CT image data obtained from the subject using an energy integrating type radiation detector, A training data acquisition unit acquires first training data to which a first correct label based on the first medical information is assigned to the CT image data, A learning unit that learns a CT inference model that performs predetermined inferences on CT image data using the aforementioned training data, It holds. [Effects of the Invention]

[0009] According to the present invention, highly accurate training data can be generated based on medical information assigned to other medical image data. [Brief explanation of the drawing]

[0010] [Figure 1] A diagram showing an example of the functional configuration of the information processing system according to the present invention. [Figure 2] A diagram showing an example of the hardware configuration of the information processing device of the present invention. [Figure 3] A diagram showing an example of the training data creation flow for an information processing system according to the first embodiment. [Figure 4] A diagram showing an example of the training data creation flow for an information processing system according to the first embodiment. [Figure 5] A schematic diagram showing an example of an inference model according to the second embodiment. [Figure 6]A diagram showing an example of the training data creation flow for an information processing system according to the second embodiment. [Figure 7] A diagram showing an example of the training data creation flow for an information processing system according to the second embodiment. [Figure 8] A schematic diagram showing an example of an inference model according to the third embodiment. [Figure 9] A diagram showing an example of the inference flow of an information processing system according to the third embodiment. [Figure 10] A diagram showing an example of the inference flow of an information processing system according to the third embodiment. [Figure 11] A diagram showing an example of the inference flow of an information processing system according to the third embodiment. [Modes for carrying out the invention]

[0011] The present invention is preferably applicable to sinogram data acquired by modality, medical image data generated by image data reconstruction from raw data, and the like.

[0012] Modalities can be applied to CT data and PCCT data, which are data acquired by, for example, X-ray CT scanners and photon counting CT (PCCT: Photon Counting Computed Tomography), as well as to CT image data and PCCT image data generated by image data reconstruction from the data.

[0013] PCCT image data is image data generated by reconstructing PCCT data obtained using a photon-counting type radiation detector using known reconstruction techniques. CT image data, on the other hand, is image data generated by reconstructing CT data obtained using an energy-integrating type radiation detector. Note that energy-integrating type radiation detectors and photon-counting type radiation detectors are different types of radiation detectors.

[0014] Hereinafter, an information processing system, information processing apparatuses (a learning apparatus and an inference apparatus) constituting the information processing system, an information processing method executed by each of the apparatuses, and a program will be described with reference to the drawings. Note that the description in the following embodiments is merely an example, and is not limited to the following embodiments.

[0015] [First Embodiment] Here, an information processing system 1 including a learning apparatus 2 and an inference apparatus 3 which are information processing apparatuses of the present invention will be described using Figure 1. The information processing system 1 according to the present invention is configured to include the learning apparatus 2, the inference apparatus 3, and a data server 4. Further, the information processing system 1 has an input unit 5 that receives input of information from a user to the information processing system 1, and a display unit 6 that receives display control from the information processing system 1 and displays information.

[0016] Further, each apparatus is configured to be connectable via a network. Further, each apparatus and each functional configuration of the information processing system 1 may be configured by a single apparatus, or may be configured by a plurality of apparatuses, and the administrator or implementer of the apparatus that executes the function of the learning apparatus and the apparatus that executes the function of the inference apparatus may be different. Further, each apparatus may be connected by wire.

[0017] The information processing system 1 includes a learning apparatus 2 that learns a CT inference model based on first medical information corresponding to PCCT image data obtained by using a photon-counting radiation detector for a subject, and an inference apparatus 3 that executes inference processing on inference target data using the CT inference model learned by the learning apparatus 2.

[0018] Here, the learning apparatus 2 has an acquisition unit 21 that acquires first medical information corresponding to PCCT image data obtained by using a photon-counting radiation detector for a subject, and CT image data obtained by using an energy-integrating radiation detector for the subject.

[0019] Furthermore, the learning device 2 includes a training data acquisition unit 22 that acquires first training data in which a first correct label based on first medical information is assigned to CT image data, and a determination unit 23 that compares the first medical information with second medical information based on CT image data to determine whether or not to use the first training data for training the CT inference model.

[0020] Furthermore, the learning device 2 includes a learning unit 24 that uses training data to train a CT inference model that performs predetermined inferences on CT image data.

[0021] The acquisition unit 21 acquires first medical information corresponding to PCCT image data obtained using a photon counting type radiation detector and CT image data obtained using an energy integrating type radiation detector from the subject. Here, PCCT image data is integrated image data or energy-discriminated image data, which is data after energy discrimination, generated by reconstructing medical data obtained using a photon counting type radiation detector from the subject. The first medical information is document information such as a medical report generated by a physician, such as a reading of the image or a medical interview, or a medical report generated by image processing, based on the PCCT image data. Here, the first medical information may be acquired using a PCCT inference model. The PCCT inference model is an inference model learned based on training data to which a first correct label is assigned to the PCCT image data. The PCCT inference model may be generated based on deep learning or machine learning techniques.

[0022] The training data acquisition unit 22 acquires training data to which first correct labels based on first medical information corresponding to PCCT image data are assigned. The training data acquisition unit 22 extracts information that will become the correct labels from the document information constituting the first medical information using known techniques such as natural language processing, sets it as the first correct label for CT image data of the same subject, and acquires training data in which CT image data and the first correct label based on the first medical information are paired.

[0023] The determination unit 23, when it has first medical information and second medical information for CT image data, compares the two pieces of information to determine whether or not to use the first training data based on the first medical information for training the CT inference model. This determination may be omitted as appropriate depending on the presence or absence of second medical information and user settings. Here, the second medical information is medical information attached to CT image data, and is a medical report generated by a physician through image interpretation or interview, or document information described in a medical report generated by image processing on CT image data.

[0024] The learning unit 24 obtains information about the CT's inference model from the data server 4. Here, the information about the inference model includes the model's structure data, and if the inference model is a pre-trained model, it also includes the model's parameter information. Furthermore, if there is training data used for pre-training the inference model, that training data may also be obtained.

[0025] The inference model is implemented using known deep learning or machine learning techniques. For example, any inference model generated based on labeled training data, such as deep learning techniques like CNNs and Vision Transformers, or supervised or semi-supervised learning techniques like SVMs and RFs, can be combined as an ensemble model.

[0026] The learning unit 24 applies the acquired first training data to the inference model and performs a learning process to determine the parameters for the inference model. If the inference model is a pre-trained inference model, additional training may be performed on the inference model. Alternatively, if the training data used during pre-training can be obtained, the parameters for the inference model may be newly learned using training data that combines this training data with the first training data.

[0027] Once the learning process for the inference model is complete, the learning unit 24 sends the parameter information to the data server 4.

[0028] Here, the inference device 3 in the information processing system 1 performs inference processing using the inference model learned by the learning device 2. Specifically, it includes an inference target data acquisition unit 31 that acquires medical image data to be inferred, and a model acquisition unit 32 that acquires a CT inference model learned using training data consisting of pairs of ground truth labels based on medical information assigned to PCCT image data obtained using a photon counting type radiation detector for a subject, and CT image data obtained using an energy integrating type radiation detector for the subject. The inference device 3 further includes an inference unit 33 that performs predetermined inference on the medical image data to be inferred using the CT inference model, and a display control unit 34 that displays the inference results from the inference unit 33 on the display unit 6.

[0029] Data server 4 stores various types of data. For example, it stores medical image data such as training data and data to be inferred, as well as information on various inference models and parameters of those inference models, and transmits data via the network in response to requests from each device.

[0030] The input unit 5 consists of input devices such as a mouse and keyboard, and receives various instructions from the user.

[0031] The display unit 6 is a display attached to the information processing system 1, or a mobile terminal used by medical professionals via an external server, etc.

[0032] Figure 2 shows an example of a specific configuration of the information processing system 1, or the learning device 2 and inference device 3 that constitute the information processing system 1. In this example, the information processing device 1 has a CPU 20, a GPU 21, RAM 22, ROM 23, and a storage device 24, which are connected by a system bus 25.

[0033] Below, an example of training an inference model using the learning device 2 of this embodiment will be explained with reference to Figure 3.

[0034] In step S30, the acquisition unit 21 acquires first medical information corresponding to PCCT image data obtained from the subject using a photon counting type radiation detector, and CT image data obtained from the subject using an energy integrating type radiation detector. It then transmits the data to the training data acquisition unit 22 and proceeds to the next step.

[0035] In step S31, the training data acquisition unit 22 extracts the first correct label from the acquired first medical information using known natural language processing techniques, thereby acquiring first training data consisting of CT image data and the first correct label paired together. The training data acquisition unit 22 transmits the information of the first training data to the learning unit 24 and proceeds to the next step.

[0036] In step S32, the learning unit 24 obtains information on the CT inference model from a storage unit such as the data server 4 and performs a learning process using the first training dataset.

[0037] A CT inference model is a model that takes CT image data as input and performs predetermined inferences. The tasks performed may include classification, detection, extraction, etc. The class to be classified may be an organ, a lesion site, or other region depicted in the image, or it may be a model that infers the degree of abnormality of a predetermined region from the information depicted in the image. Furthermore, the model may perform either classification or regression.

[0038] When the learning process using the first training data is completed for the inference model, the learning unit 24 sends the parameter information determined by the learning process to the data server 4 and terminates the flow.

[0039] This configuration makes it possible to efficiently generate training data for a CT inference model that performs CT inference based on medical information assigned to medical image data that is different from the input of the inference model, such as PCCT image data.

[0040] Furthermore, PCCT image data has a higher resolution and is more detailed than CT image data. In addition to integrated image data from PCCT, energy discrimination image data can be used to enable more accurate image interpretation. Therefore, PCCT image data contains more information than CT image data, and consequently, the accuracy of the generated medical information tends to be higher. Due to these characteristics, by using the first medical information for PCCT image data as the ground truth label for CT image data, the information processing system 1 can create highly accurate training data. It is also possible that feature regions that have become easier to interpret due to PCCT imaging are locally depicted on the CT image data. In such cases, training a CT inference model using CT image data with ground truth labels based on medical information obtained from PCCT image data as training data enables more accurate inference.

[0041] (Variation 1) Next, using Figure 4, we will describe a modified example 1 of the learning of the inference model by the learning device 2 of this embodiment.

[0042] In the modified example 1, when second medical information is attached to CT image data, the determination unit 23 compares the first medical information with the second medical information to determine whether or not to use it as first training data to which a first correct label based on the first medical information is attached. Specifically, even if there are omissions or annotations based on omissions in the medical information attached to the CT image data, it is possible to create highly accurate training data. Hereafter, the same figure numbers will be used for components similar to those in the above embodiment, and explanations will be omitted as appropriate.

[0043] In step 40, the acquisition unit 21 further acquires second medical information for the CT image data. The second medical information is medical information attached to the CT image data, and is either a medical report generated by a physician through image interpretation or interviews, or document information contained in a medical report generated by image processing of the CT image data. The acquisition unit 21 transmits the acquired second information to the determination unit 23 and proceeds with processing.

[0044] In step S41, the determination unit 23 compares the first medical information and the second medical information and determines whether the two pieces of information match. The degree of matching does not have to be an exact match; a score may be calculated based on the degree of matching, such as a partial match. If the first medical information and the second medical information do not match, the determination unit 23 proceeds to step S31. On the other hand, if the first medical information and the second medical information match, the process is terminated.

[0045] This configuration allows for the creation of highly accurate training data, as correct labels based on the first medical information can be assigned even if there are omissions in the second medical information. On the other hand, if the second medical information and the first medical information match, it is unnecessary to create correct labels based on the first medical information, and therefore no processing is required.

[0046] [Second Embodiment] Hereinafter, an example of the data creation flow for the information processing system 1 in this embodiment will be described using Figures 5 and 6. Note that components similar to those in the first embodiment will be given the same figure numbers, and their descriptions will be omitted as appropriate.

[0047] In this embodiment, the inference model of the CT to be trained is a pre-trained inference model and is stored in the data server 4.

[0048] The data server 4 stores the pre-trained pre-CT inference model 410 and the second training data 415, which is the training data used to train the pre-CT inference model, in association with each other. The training device 2 transmits the trained CT inference model 420 from the training device 2 to the data server 4. The transmitted information may be updated parameter information of the CT inference model. If the model structure differs from the pre-CT inference model 410, the structural information of the CT inference model 420 and the parameter information are stored in association with each other.

[0049] The learning flow of the learning device 2 in this embodiment will be described below with reference to Figure 6.

[0050] In step S61, the acquisition unit 21 further acquires the second training data used to train the pre-CT inference model 410 as second medical information, and after transmitting the second medical information to the determination unit 23, proceeds to the next step.

[0051] In step S62, the determination unit 61 determines whether the first correct label based on the first medical information matches the second correct label based on the second medical information. This matching may be determined by calculating the degree of matching. If the determination determines that the correct labels match, the learning flow is terminated. On the other hand, if it is determined that the correct labels are different, the information of the first correct label is sent to the training data acquisition unit 22, and processing proceeds.

[0052] With this configuration, even if there are annotation errors in the second training data 415 used to train the pre-CT inference model 410, it is possible to create highly accurate training data in which the first correct label based on the first medical information attached to the PCCT image data is used as the correct label for the CT image data.

[0053] (Modification 2) Next, Figure 7 will be used to describe a second modified example of the learning of the inference model by the learning device 2 of this embodiment. Other embodiments and configurations similar to those described above will be given the same figure numbers, and their descriptions will be omitted as appropriate.

[0054] In this modified version, the inference result for CT image data obtained by a pre-CT classifier is acquired as second medical information, and compared with the first medical information for PCCT image data. If the first medical information and the second medical information differ, a first correct label based on the first medical information is assigned to the CT image data, and training data is generated.

[0055] The learning process in this modified example 2 will be explained below using Figure 7.

[0056] In step S71, the acquisition unit 21 further acquires an inference model of the pre-CT scan, uses the acquired pre-CT inference model to perform inference processing on the CT image data, transmits the inference result as second medical information to the determination unit 23, and proceeds to the next step.

[0057] In step S72, the determination unit 23 determines whether the first medical information and the second medical information match. If the two pieces of information are different, the process proceeds to step S31. If the two pieces of information are different, the process ends.

[0058] This configuration allows for the creation of training data in which correct labels based on first medical information are assigned to CT image data that the pre-CT inference model cannot classify well. Furthermore, by training the CT inference model using this training data, highly accurate inference becomes possible.

[0059] (Variation 3) The acquisition unit 21 may further acquire the acquisition interval between the first medical information and the second medical information. The determination unit 23 determines that if the information acquisition interval is large, it will not assign the first correct label to the CT image data. Alternatively, it may determine whether or not to assign the first correct label to the CT image data based on the diagnosis and findings described in the first medical information and the information acquisition interval.

[0060] [Third Embodiment] Using the inference model generated by the learning method described in the first and second embodiments, inference processing is performed on the medical image data to be inferred.

[0061] Below, an example of the inference flow of the information processing system 1 in this embodiment will be described using Figures 1, 8, and 9.

[0062] Figure 8 schematically shows the configuration of the inference device 3 and its relationship to the data server 4. The model acquisition unit 31 acquires the CT inference model 420, which was learned in either the first or second embodiment, from the data server 4.

[0063] The inference flow of the inference device 3 in this embodiment will be explained using Figure 9 below.

[0064] In step S90, the data acquisition unit 31 acquires the medical image data to be inferred, transmits the acquired information to the inference unit 33, and proceeds to the next step.

[0065] In step S91, the model acquisition unit 32 acquires a CT inference model that has been trained using training data consisting of pairs of ground truth labels based on medical information assigned to PCCT image data obtained using a photon counting type radiation detector for a subject, and CT image data obtained using an energy integrating type radiation detector for the same subject. After transmitting the information of the inference model to the inference unit 33, the process proceeds to the next step.

[0066] In step S92, the inference unit 33 performs a predetermined inference process on the acquired medical image data to be inferred using the CT inference model, transmits the inference result to the display control unit 34, and then proceeds to the next step.

[0067] In step S93, the display control unit 34 displays the inference result on the display unit 6 and then terminates the inference flow.

[0068] This configuration allows inference to be performed using a trained CT inference model based on highly accurate training data.

[0069] (Variation 1) Hereinafter, with reference to Figure 10, a modified example 1 of the inference flow of the information processing system 1 in this embodiment will be described.

[0070] In this modified example, the inference unit 33 determines whether the medical information attached to the CT image data matches the inference result obtained by the CT inference model. If they do not match, the display control unit 34 causes the display unit 6 to display the attached medical information and the inference result in a way that allows for comparison.

[0071] In step S1003, the inference unit 33 compares the medical information attached to the CT image data with the inference result for the CT image data. If the medical information and the inference result are different, the medical information and the inference result are sent to the display control unit 34, and the process proceeds to the next step.

[0072] In step S1004, the display control unit 34 displays the medical information and the inference result information on the display unit 6 in a way that allows for comparison.

[0073] This configuration makes it possible to determine whether the medical information added to CT image data by a physician or other medical professional differs from the inference results of the CT inference model. If they differ, it is possible to request confirmation of the differences in the medical information.

[0074] (Modification 2) Hereinafter, a modified example 2 of the inference flow of the information processing system 1 in this embodiment will be described with reference to Figure 11.

[0075] In this modified example, the inference unit 33 acquires the inference model 410 of the pre-CT and the inference model 420 of the CT generated through the learning flow of the above-described embodiment, and performs inference using both inference models.

[0076] Furthermore, if the inference results from the two inference models differ, the display control unit 34 associates the two inference results with the difference information between the two inference results and causes the display unit 6 to perform display processing.

[0077] In step S1101, the inference unit 33 obtains the pre-CT inference model 410 and the CT inference model 420 generated through the learning process described above from the data server 4, and then proceeds to the next step.

[0078] In step S1102, the inference unit 33 performs a first inference process using the pre-CT inference model 410 and a second inference process using the CT inference model 420 on the medical data to be inferred, and then proceeds to the next step.

[0079] In step S1103, the inference unit 33 determines whether the first inference result from the first inference process and the inference result from the second inference process match. If they do not match, it sends the first and second inference results, and any difference information between the inference results, to the display control unit 34, and proceeds to step S1104. If they do match, the inference flow ends.

[0080] In step S1104, the display control unit 34 displays the first and second inference results and difference information obtained from the inference unit 33 on the display unit 6.

[0081] This configuration allows the user to see information about the differences when the inference results from the pre-CT inference model and the CT inference model differ.

[0082] (Variation 3) Furthermore, the inference unit 33 may acquire pixel information that contributed to the inference result using visualization technology such as Grad-CAM, and the display control unit 34 may display the acquired pixel information on the display unit 6. Here, the configuration implemented by the inference unit 33 has been described, but this modified example may also be applied to verifying the learned inference model in the learning unit 24 of the first and second embodiments.

[0083] This configuration allows the user to understand which pixels in the CT image were examined and which pixels the CT inference model used to perform the inference process.

[0084] (Modification 4) Based on first medical information corresponding to PCCT image data obtained using a photon counting type radiation detector, a first ground truth label may be assigned to the pseudo-generated CT image data obtained from the PCCT data using the radiation detector.

[0085] This configuration makes it possible to create training data for a CT inference model from information obtained from a single scan of a subject using a photon-counting type radiation detector.

[0086] (Other examples) Furthermore, the present invention can also be realized by performing the following process: that is, supplying software (program) that realizes the functions of the above-described embodiment to a system or device via a network or various storage media, and having the computer (or CPU or MPU, etc.) of that system or device read and execute the program.

Claims

1. An acquisition unit that acquires first medical information corresponding to PCCT image data obtained on a subject using a photon counting type radiation detector, CT image data obtained on the subject using an energy integrating type radiation detector, and second medical information corresponding to the CT image data. A training data acquisition unit acquires first training data in which a first correct label based on the first medical information is assigned to the CT image data, A determination unit that determines whether or not to use the first training data for training a CT inference model that performs predetermined inferences on CT image data, by comparing the first medical information and the second medical information, A learning unit that learns the inference model of the CT using the first training data which the determination unit has determined to be used for training the inference model, An information processing system characterized by having the following features.

2. The information processing system according to claim 1, characterized in that the first medical information is medical information generated by a user with respect to the PCCT image data.

3. The information processing system according to claim 1, characterized in that the first medical information is information based on the results inferred using a PCCT inference model with respect to the PCCT image data.

4. The information processing system according to claim 1, characterized in that the CT inference model is an inference model pre-trained using second training data based on the second medical information.

5. The information processing system according to claim 4, characterized in that the learning unit further trains the inference model of the CT using the first training data acquired by the training data acquisition unit.

6. The information processing system according to claim 1, characterized in that the second medical information is medical information generated by a user with respect to the CT image data.

7. The information processing system according to claim 1, characterized in that the second medical information is information based on the results inferred using the pre-trained CT inference model.

8. The information processing system according to claim 1, characterized in that the determination unit determines that the first training data will not be used for training the CT inference model when the first correct label based on the first medical information matches the second correct label based on the second medical information.

9. The information processing system according to claim 1, characterized in that the determination unit determines that the first training data should be used to train the CT inference model when the first correct label based on the first medical information and the second correct label based on the second medical information are different.

10. The information processing system according to claim 1, characterized in that the determination unit determines not to use the first training data for training the CT inference model when the first correct label based on the first medical information matches the second correct label based on the second medical information, which is a correct label used for pre-training the CT inference model.

11. The information processing system according to claim 1, characterized in that the determination unit determines that the first training data should be used for additional training of the pre-trained CT inference model when the first correct label based on the first medical information and the second correct label based on the second medical information, which is a correct label used for pre-training of the CT inference model, are different.

12. A data acquisition unit for acquiring medical image data to be inferred, A model acquisition unit that acquires the CT inference model generated by the learning unit, An inference unit performs predetermined inference processing on the medical image data to be inferred using the learned CT inference model, The information processing system according to claim 1, further comprising:

13. The information processing system according to claim 12, further comprising a display control unit that displays information based on the inference results of the inference unit on a display unit.

14. The inference unit further acquires the pixel information that contributed to the predetermined inference, The information processing system according to claim 13, characterized in that the display control unit displays the pixel information in association with the medical image data to be inferred.

15. The information processing system according to claim 1, wherein the radiation detector is different from the energy-integrating type radiation detector and the photon-counting type radiation detector.

16. An inference target data acquisition unit that acquires medical image data to be inferred, A model acquisition unit acquires a CT inference model that has been trained using training data consisting of a first medical information-based correct label assigned to PCCT image data obtained for a subject using a photon counting type radiation detector, and CT image data obtained for the same subject using an energy integrating type radiation detector. An inference unit that performs predetermined inference on the medical image data to be inferred using the CT inference model, A display control unit that displays the inference results, An information processing device having a training data which is determined to be used for training the CT inference model by comparing the first medical information with the second medical information corresponding to the CT image data.

17. An acquisition unit that acquires first medical information corresponding to PCCT image data obtained using a photon counting type radiation detector on a subject, CT image data obtained using an energy integrating type radiation detector on the subject, and second medical information corresponding to the CT image data. A training data acquisition unit acquires first training data in which a first correct label based on the first medical information is assigned to the CT image data, A learning unit that, based on a comparison of the first medical information and the second medical information, determines that the first training data should be used to train a CT inference model that performs predetermined inferences on CT image data, and trains the CT inference model using the first training data, An information processing system characterized by having the following features.

18. Acquisition step of acquiring first medical information corresponding to PCCT image data obtained using a photon counting type radiation detector for a subject, CT image data obtained using an energy integrating type radiation detector for the subject, and second medical information corresponding to the CT image data, A training data acquisition step involves acquiring first training data in which a first correct label based on the first medical information is assigned to the CT image data, A determination step of determining whether or not to use the first training data for training the inference model by comparing the first medical information and the second medical information, A learning step in which a CT inference model is trained to perform predetermined inferences on CT image data using the first training data which was determined in the determination step to be used for training the CT inference model, An information processing method characterized by having the following features.

19. A step of acquiring data to be inferred, which involves acquiring medical image data to be inferred, A model acquisition step to acquire a CT inference model trained using training data consisting of a first medical information-based ground truth label assigned to PCCT image data obtained for a subject using a photon counting type radiation detector, and CT image data obtained for the same subject using an energy integrating type radiation detector. An inference step in which predetermined inference is performed on the medical image data to be inferred using the CT inference model, A display step to show the inference results, An information processing method comprising the above, wherein the training data is data determined to be used for training the CT inference model by comparing the first medical information with the second medical information corresponding to the CT image data.

20. A program for a computer to execute the information processing method described in claim 18 or claim 19.

Images