Information processing device, method, and program

Abstract

To enable the generation of text that accurately describes a patient's condition in an information processing device, method, and program. [Solution] The processor obtains one or more analysis results regarding the patient's medical image, obtains diagnostic information regarding the patient's diagnosis other than the analysis results, and generates medical text about the patient based on the analysis results and diagnostic information.

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus, method, and program.

Background Art

[0002] In recent years, with the advancement of medical devices such as CT (Computed Tomography) devices and MRI (Magnetic Resonance Imaging) devices, image diagnosis using higher-quality and higher-resolution medical images has become possible. In particular, by image diagnosis using CT images, MRI images, etc., the lesion area can be accurately identified, and appropriate treatment is being carried out based on the identified results.

[0003] Also, CAD (Computer-Aided Diagnosis) using a learning model learned by deep learning or the like analyzes medical images to detect characteristics such as the shape, density, position, and size of abnormal shadows such as lesions included in the medical images. The analysis results obtained in this way are associated with inspection information such as patient name, gender, age, and the imaging device that acquired the medical image, and are stored in a database. The medical images and analysis results are transmitted to the terminal of the radiologist who reads the medical images. The radiologist reads the medical images by referring to the distributed medical images and analysis results on their own reading terminal, and creates a reading report.

[0004] On the other hand, with the increased performance of CT and MRI scanners mentioned above, the number of medical images to be interpreted is increasing. Therefore, in order to reduce the burden on radiologists in their interpretation work, various methods have been proposed to support the creation of medical documents such as interpretation reports. For example, a method has been disclosed that generates text to be included in an interpretation report based on keywords representing findings from medical images entered by the radiologist and information representing the characteristics of abnormal shadows included in the analysis results of the medical images (see Patent Document 1). In the method described in Patent Document 1, medical text to be included in an interpretation report is generated using a learning model such as a recurrent neural network that has been trained on machine learning to generate medical text from input characters representing characteristics. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 2019-153250 [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] When performing diagnostic imaging, other tests such as blood tests are often conducted in conjunction with acquiring images of the patient. Furthermore, diagnostic imaging may involve the use of multiple types of images, including not only CT images but also MRI images and others. For this reason, if only the results of medical image analysis and findings entered by the radiologist are used, as in the method described in Patent Document 1, the text generated by the learning model will not necessarily be a medical document that accurately describes the patient's condition.

[0007] This disclosure is made in light of the circumstances described above and aims to enable the generation of medical documents that accurately represent the patient's condition. [Means for solving the problem]

[0008] The information processing device according to this disclosure comprises at least one processor, the processor acquires one or more analysis results relating to a patient's medical image, We obtain diagnostic information regarding the patient's diagnosis, in addition to the analysis results. Based on the analysis results and diagnostic information, medical documents about the patient are generated.

[0009] Furthermore, in the information processing device described herein, the processor selects and discards analysis results based on diagnostic information. It may also be a system that generates medical documents that include selected analysis results.

[0010] Furthermore, in the information processing device described herein, the processor may generate medical documents including analysis results in order of priority according to the diagnostic information.

[0011] Furthermore, in the information processing device described herein, the processor may generate medical documents including diagnostic information and analysis results.

[0012] Furthermore, in the information processing device according to this disclosure, the diagnostic information may include first information that has been confirmed regarding a lesion contained in a medical image.

[0013] In this case, the first piece of information may include at least one of the following: the measurement results of the lesion, the definitive diagnosis of the lesion, and the patient's medical history.

[0014] Furthermore, in the information processing device provided for this disclosure, the diagnostic information may include a second piece of confirmed information other than information relating to lesions contained in medical images.

[0015] In this case, the second piece of information may include at least one of the purpose of the examination in which the medical image was acquired and the image conditions relating to the medical image.

[0016] In the information processing apparatus according to the present disclosure, the diagnostic information may include third information representing a judgment result by a radiologist for a medical image.

[0017] In this case, the third information may include at least one of an uncertain diagnosis result regarding the medical image, the relevance between a lesion included in the medical image and tissues other than the lesion, and a selection result of an analysis result by the radiologist.

[0018] In the information processing apparatus according to the present disclosure, the diagnostic information may include fourth information representing an examination result performed on a patient.

[0019] In this case, the fourth information may include at least one of an examination result by a diagnostic device different from the imaging device that acquires a medical image of the patient, an analysis result of a different type of medical image from the medical image, and an examination result of the patient's biological information.

[0020] The information processing method according to the present disclosure acquires one or more analysis results regarding a medical image of a patient, acquires diagnostic information regarding the diagnosis of the patient other than the analysis result, and generates a medical document regarding the patient based on the analysis result and the diagnostic information.

[0021] Note that it may be provided as a program for causing a computer to execute the information processing method of the present disclosure.

Advantages of the Invention

[0022] According to the present disclosure, a medical document with content accurately representing the patient's condition can be generated.

Brief Description of the Drawings

[0023] [Figure 1] A diagram showing an example of the schematic configuration of a medical information system to which the information processing apparatus according to the first embodiment is applied [Figure 2] A block diagram showing an example of the hardware configuration of the information processing apparatus according to the first embodiment [Figure 3] Block diagram showing an example of the functional configuration of the information processing device according to the first embodiment. [Figure 4] Figure showing an example of analysis results. [Figure 5] A schematic diagram illustrating a recurrent neural network. [Figure 6] A diagram showing examples of diagnostic information and analysis results. [Figure 7] A diagram showing examples of labels for diagnostic information. [Figure 8] A diagram showing examples of labels for diagnostic information. [Figure 9] A diagram showing examples of labels for diagnostic information. [Figure 10] A diagram showing examples of labels for diagnostic information. [Figure 11] A flowchart illustrating the process performed in the first embodiment. [Figure 12] Block diagram showing an example of the functional configuration of the information processing device according to the second embodiment. [Figure 13] A diagram showing an example of a table defining rules for selecting and discarding analytical information in the second embodiment. [Modes for carrying out the invention]

[0024] The embodiments of this disclosure will be described below with reference to the drawings. First, the configuration of the medical information system 1 to which the information processing device according to the first embodiment is applied will be described.

[0025] Figure 1 shows a schematic diagram of the medical information system 1. The medical information system 1 shown in Figure 1 is a system that, based on examination orders from physicians in clinical departments using a known ordering system, photographs the area of ​​the patient to be examined, stores the medical images obtained through the photography, allows radiologists to interpret the medical images and create interpretation reports, and allows physicians in the requesting clinical departments to view the interpretation reports and perform detailed observations of the medical images being interpreted.

[0026] As shown in Figure 1, the medical information system 1 is configured by connecting multiple imaging devices 2, multiple image interpretation terminals (WS) 3, medical consultation terminals (WS4), an image server 5, an image database 6, a report server 7, and a report database 8, all of which are connected to each other via a wired or wireless network 10 and capable of communicating with one another.

[0027] Each device is a computer on which an application program is installed to function as a component of the medical information system 1. The application program is distributed on recording media such as DVDs (Digital Versatile Discs) and CD-ROMs (Compact Disc Read Only Memory), and installed on the computer from that recording media. Alternatively, it is stored in a storage device of a server computer connected to network 10, or in network storage, in a state that is accessible from the outside, and downloaded and installed on the computer upon request.

[0028] The imaging device 2 is a modality that generates medical images representing the area to be diagnosed by imaging that area of ​​the patient. Specifically, this includes plain X-ray imaging devices, CT scanners, MRI scanners, and PET (Positron Emission Tomography) scanners. The medical images generated by imaging device 2 are transmitted to the image server 5 and stored in the image database 6.

[0029] The Image Interpretation WS3 is a computer used, for example, by a radiologist in the radiology department for interpreting medical images and creating interpretation reports, and it incorporates the information processing device 20 (details described later) according to the first embodiment. The Image Interpretation WS3 handles requests to view medical images from the image server 5, various image processing on medical images received from the image server 5, display of medical images, and acceptance of input of findings text related to medical images. The Image Interpretation WS3 also performs analysis processing on medical images, assists in creating interpretation reports based on the analysis results, requests registration and viewing of interpretation reports from the report server 7, and displays interpretation reports received from the report server 7. These processes are carried out by the Image Interpretation WS3 executing software programs for each process.

[0030] The Clinical WS4 is a computer used by, for example, physicians in a clinical department for detailed examination of images, viewing of image interpretation reports, and creation of electronic medical records. It consists of a processing unit, display devices such as a display, and input devices such as a keyboard and mouse. The Clinical WS4 performs the following operations: requests to view images from the image server 5, display of images received from the image server 5, requests to view image interpretation reports from the report server 7, and display of image interpretation reports received from the report server 7. These operations are performed by the Clinical WS4 executing software programs for each operation.

[0031] Image server 5 is a general-purpose computer with software programs installed that provide the functionality of a database management system (DBMS). Image server 5 also includes storage that constitutes image DB 6. This storage may be a hard disk drive connected to image server 5 via a data bus, or a disk drive connected to a NAS (Network Attached Storage) or SAN (Storage Area Network) connected to network 10. Furthermore, when image server 5 receives a request to register a medical image from imaging device 2, it formats the medical image into a database format and registers it in image DB 6.

[0032] Furthermore, in this embodiment, the image server 5 stores diagnostic information related to the patient's diagnosis. The diagnostic information will be described later.

[0033] Image DB6 stores image data and associated information of medical images acquired by imaging device 2. The associated information includes, for example, an image ID (identification) to identify individual medical images, a patient ID to identify the patient, an examination ID to identify the examination, a unique ID (UID: unique identification) assigned to each medical image, the date and time of the examination in which the medical image was generated, the type of imaging device used in the examination to acquire the medical image, patient information such as patient name, age, and gender, the examination site (imaging site), imaging information (imaging protocol, imaging sequence, imaging method, imaging conditions, use of contrast agent, etc.), and information such as the series number or acquisition number if multiple medical images were acquired in a single examination.

[0034] Furthermore, when the image server 5 receives a viewing request from the image interpretation WS3 and the medical treatment WS4 via the network 10, it searches for medical images registered in the image database 6 and sends the retrieved medical images to the requesting image interpretation WS3 and medical treatment WS4.

[0035] The report server 7 incorporates software programs that provide the functionality of a database management system to a general-purpose computer. When the report server 7 receives a request to register an image interpretation report from the image interpretation WS3, it formats the image interpretation report into a database format and registers it in the report DB8.

[0036] The report DB8 stores image interpretation reports containing findings written by radiologists using the image interpretation WS3. The image interpretation report may include, for example, the medical image being interpreted, an image ID to identify the medical image, a radiologist ID to identify the radiologist who performed the interpretation, the name of the lesion, location information of the lesion, and characteristics of the lesion.

[0037] Furthermore, when the report server 7 receives a request to view an image interpretation report from the image interpretation WS3 and medical treatment WS4 via the network 10, it searches the report DB8 for the image interpretation report registered therein and sends the retrieved report to the requesting image interpretation WS3 and medical treatment WS4.

[0038] Network 10 is a wired or wireless local area network that connects various devices within the hospital. If the image interpretation WS3 is installed in another hospital or clinic, Network 10 may be configured to connect the local area networks of each hospital via the Internet or a dedicated line.

[0039] Next, an information processing device 20 according to the first embodiment will be described. First, the hardware configuration of the information processing device 20 according to the first embodiment will be described with reference to Figure 2. As shown in Figure 2, the information processing device 20 includes a CPU (Central Processing Unit) 11, non-volatile storage 13, and memory 16 as a temporary storage area. The information processing device 20 also includes a display 14 such as a liquid crystal display, input devices 15 such as a keyboard and a pointing device such as a mouse, and a network I / F (Interface) 17 connected to a network 10. The CPU 11, storage 13, display 14, input devices 15, memory 16, and network I / F 17 are connected to a bus 18. Note that the CPU 11 is an example of a processor in this disclosure.

[0040] Storage 13 is implemented using HDD (Hard Disk Drive), SSD (Solid State Drive), flash memory, etc. Information processing programs 12 are stored in storage 13 as a storage medium. The CPU 11 reads the information processing programs 12 from storage 13, expands them into memory 16, and executes the expanded information processing programs 12.

[0041] Next, the functional configuration of the information processing device according to the first embodiment will be described. Figure 3 is a diagram showing the functional configuration of the information processing device according to the first embodiment. As shown in Figure 3, the information processing device 20 includes an information acquisition unit 21, an analysis unit 22, a text generation unit 23, and a display control unit 24. The CPU 11 executes the information processing program 12, and the CPU 11 functions as the information acquisition unit 21, the analysis unit 22, the text generation unit 23, and the display control unit 24.

[0042] The information acquisition unit 21 acquires a medical image G0, as an example of an image, from the image server 5 via the network interface 17. In the first embodiment, a CT image of the lung is used as medical image G0 as an example. The information acquisition unit 21 also acquires diagnostic information related to the diagnosis of the patient from whom medical image G0 was acquired, from the image server 5 via the network interface 17. The diagnostic information will be described later.

[0043] The analysis unit 22 derives the analysis results of the medical image G0 by analyzing the medical image G0. To this end, the analysis unit 22 has a machine learning model 22A that detects abnormal shadows such as lesions contained in the medical image G0 and determines the characteristics of the detected abnormal shadows for each of a predetermined set of characteristic items.

[0044] Examples of characteristics that can be identified regarding abnormal lung shadows include the location of the abnormal shadow, the type of attenuation (solid and ground-glass opacities), the presence or absence of spicules, the presence or absence of calcification, the presence or absence of cavities, the presence or absence of pleural invagination, the presence or absence of pleural contact, and the presence or absence of pleural infiltration. However, the examples of characteristics are not limited to these.

[0045] In the first embodiment, the learning model 22A consists of a convolutional neural network that has been machine-learned using training data, such as deep learning, to discriminate the characteristics of abnormal shadows in medical images.

[0046] The learning model 22A is constructed using machine learning, for example, by using combinations of medical images containing abnormal shadows and characteristic items representing the characteristics of the abnormal shadows as training data. When a medical image is input to the learning model 22A, it outputs a characteristic score derived for each characteristic item in the abnormal shadow contained in the medical image. The characteristic score is a score that indicates the prominence of the characteristic for each characteristic item. The characteristic score takes a value between 0 and 1, for example, and the higher the characteristic score, the more prominent that characteristic is.

[0047] For example, if the characteristic score for "presence or absence of spicules," one of the characteristic items of an abnormal shadow, is above a predetermined threshold (e.g., 0.5), the characteristic of the abnormal shadow regarding "presence or absence of spicules" is identified as "spicula present (positive)." If the characteristic score for "presence or absence of spicules" is below the threshold, the characteristic of the abnormal shadow regarding the presence or absence of spicules is identified as "spicula absent (negative)." The threshold of 0.5 used for characteristic determination is merely an example, and an appropriate value should be set for each characteristic item. In addition, if the characteristic score is near the threshold (e.g., 0.4 or higher and 0.6 or lower), it may be identified as a false positive.

[0048] Figure 4 shows an example of the analysis results derived by the analysis unit 22. As shown in Figure 4, the characteristic information identified by the analysis unit 22 includes the location of the abnormal shadow, the type of attenuation value, and characteristic items of spicules, calcification, cavities, and pleural invagination. The characteristics of each are "upper left section," "solid type," "spicules present," "calcification present," "cavities present," and "no pleural invagination." In Figure 4, a + is assigned if "present" is positive, and a - is assigned if "absent" is negative.

[0049] In addition to convolutional neural networks, any learning model such as a Support Vector Machine (SVM) can be used as the learning model 22A.

[0050] Alternatively, a separate learning model may be constructed for detecting abnormal shadows from medical image G0 and for determining the characteristics of the abnormal shadows.

[0051] The text generation unit 23 generates medical text about the patient based on the analysis results derived by the analysis unit 22 and the diagnostic information acquired by the information acquisition unit 21. In this embodiment, the text generation unit 23 consists of a learning model 23A constructed by machine learning to generate findings to be included in the radiology report as medical text from the input information. As the learning model 23A, for example, a neural network such as a recurrent neural network described in U.S. Patent No. 10181098 or No. 10268671 can be used. In this embodiment, the learning model 23A is constructed by training a recurrent neural network using supervised learning. The training data used in this case is data that associates various combinations of analysis results and various diagnostic information with various training texts that should be generated from the analysis results and diagnostic information.

[0052] Figure 5 is a schematic diagram of a recurrent neural network. As shown in Figure 5, the recurrent neural network 40 consists of an encoder 41 and a decoder 42. The encoder 41 is input with the labels of the analysis results and diagnostic information derived by the analysis unit 22. The encoder 41 is input with the 1-hot representation of the labels of the analysis results and diagnostic information. A 1-hot representation is a vector in which each label is represented by a vector in which one component is 1 and the remaining components are 0. For example, if the 1-hot representation is a vector consisting of three elements, then (1,0,0), (0,1,0), and (0,0,1) will each represent three different labels.

[0053] The encoder 41 transforms the 1-hot representation of each label using an embedding matrix to derive a vector representation of each label. Each element of the embedding matrix is ​​a learning parameter. The learning parameters are determined by machine learning of the recurrent neural network 40.

[0054] The decoder 42 is composed of multiple networks connected together, each consisting of an input layer, a hidden layer, and an output layer. Each network receives the vector representation xt output by the encoder 41 and the output ht of the preceding network as inputs. The hidden layer performs the operation shown in equation (1) below. In equation (1), Wh, Wx, and b are learning parameters, which are determined through learning. tanh is the activation function. Note that the activation function is not limited to tanh; a sigmoid function or the like may also be used.

[0055] ht=tanh(ht-1·Wh+xt·Wx+b) (1)

[0056] Here, let's assume the analysis results and diagnostic information of the training data were "left subpleural lung," "4.2 cm," "spicula+," and "tumor." "Left subpleural lung" is a term that describes the location in the lung, so a label indicating location is assigned. "4.2 cm" is the diameter, so a label indicating size is assigned. "Spicula+" is labeled to indicate that spicules are positive, and "tumor" is labeled to indicate a medically small mass. These labels are input to encoder 41, and a vector representation of each label is output.

[0057] The decoder 42 receives the output and vector representation of the previous stage as input to the input layer of each neural network, and outputs the finding statement, "A small mass with a diameter of [size] and [spicules] is observed at [location]." The text generation unit 23 generates the statement, "A 4.2 cm diameter mass with spicules is observed in the subpleural region of the left lung," by embedding the analysis results and diagnostic information into the labels included in the finding statement output by the learning model 23A.

[0058] Here, the diagnostic information used in this embodiment will be described. Diagnostic information is information related to the diagnosis of the patient other than the analysis results derived by the analysis unit 22. Specifically, diagnostic information includes confirmed information regarding lesions contained in medical images (referred to as the first information D1), confirmed information other than information regarding lesions contained in medical images (referred to as the second information D2), information representing the judgment result of the radiologist on the medical images (referred to as the third information D3), and information representing the results of examinations performed on the patient (referred to as the fourth information D4).

[0059] The first piece of confirmed information D1 regarding the lesion contained in medical image G0 includes, for example, measurement information such as the size of the lesion (length and width or area), the confirmed diagnosis of the lesion, the patient's medical history from which medical image G0 was obtained, and the treatments performed on the patient. Regarding the size of the lesion, information representing the change over time from the size of the lesion contained in medical images previously obtained for the same patient can be used as the first piece of information D1. Information representing change indicates whether the size of the lesion has increased, decreased, or remained unchanged. The confirmed diagnosis of the lesion is a diagnosis confirmed by a physician, such as whether the lesion is cancerous or a benign tumor. The patient's medical history is the history of illnesses the patient has suffered from in the past from which medical image G0 was obtained. The treatments performed on the patient include the details of the surgery performed on the patient, the type of medication used, the amount of medication, and the duration of medication.

[0060] Regarding the generation of findings using the first information D1, diagnostic information may be directly included in the findings (for example, 10 mm in diameter, history of primary lung cancer or colorectal cancer). In this case, the diagnostic information includes labels for size, confirmed diagnosis of malignancy, and medical history as the first information D1, and a learning model 23A can be constructed by training a neural network using training data that includes training texts describing these labels.

[0061] Furthermore, for nodules with a major diameter of less than 10 mm, for example, it is difficult to evaluate internal characteristics other than calcification, so the findings report may be generated in a way that does not include negative internal findings (bronchial radiolucency, cavities, fat, etc.) in the analysis results. In this case, the diagnostic information includes, for example, a size label as the first piece of information D1, and the analysis results include labels such as "nodule, bronchial radiolucency-, cavity-". The learning model 23A can be constructed by training a recurrent neural network 40 using training data that includes training text that does not include descriptions of the labels for bronchial radiolucency and cavity.

[0062] Furthermore, if primary lung cancer is confirmed as a diagnosis, the change in size over time (increase or decrease) becomes more important than the imaging characteristics (such as the presence of spicules). For this reason, the findings report may only describe the change in size over time without including the analysis results that describe the characteristics, or it may be acceptable to omit analysis results that contradict the confirmed diagnosis (for example, findings suggesting benignity when primary lung cancer has been confirmed). In this case, the diagnostic information includes, as first information D1, a label for a confirmed malignant diagnosis such as primary lung cancer and a label for "increase," and the analysis results include labels for positive characteristic items, and among these labels, the label for a confirmed malignant diagnosis is increasing, but the description of the label for positive characteristic items is not included. The learning model 23A can be constructed by training a recurrent neural network 40 using training data that includes training text.

[0063] The second piece of confirmed information D2, other than information about lesions contained in medical image G0, is information unrelated to lesions in medical image G0. Specifically, it includes the purpose of the examination for acquiring medical image G0 and the image conditions used when acquiring medical image G0. Acquisition of medical image G0 is performed as part of an examination to determine the patient's condition, and there are various examination purposes, such as detailed examination or follow-up. In addition, when acquiring medical image G0, various conditions (window level, window width, and slice interval) are prepared in various modalities to make the target organ easier to see. Image conditions are the conditions used to generate medical image G0, such as the window level, window width, and slice interval.

[0064] Regarding the generation of findings using the second piece of information D2, the content to be included in the findings differs depending on whether the purpose of the examination is a detailed examination or follow-up observation. For example, if it is a detailed examination, the findings should include detailed information, while if it is follow-up observation, the findings should include changes in size over time (increase or decrease). In this case, for example, the learning model 23A can be constructed by training a recurrent neural network 40 using training data that includes a label indicating that the purpose of the examination is a detailed examination as the second piece of information D2 in the diagnostic information, and training data that includes training data that includes a label indicating that the purpose of the examination is follow-up observation as the second piece of information D2, and training data that includes training data that includes only the label for size change among the characteristic labels included in the analysis results.

[0065] Furthermore, when the slice thickness of the CT image is 5 mm, the pixel values ​​are averaged due to the partial volume effect, resulting in low contrast and making it difficult to evaluate internal and peripheral characteristics. If a definitive statement of findings is generated in such a case, the creator will feel uncomfortable with the statement. Therefore, the statement of findings should be generated in such a way that analysis results that are difficult to evaluate are either included or omitted depending on the image conditions, or the statement is made to lower or raise the level of confidence (changing the ending from "confirmed" to "doubtful"), or the expression is made to be vague or definitive (referring to ground-glass nodules as faint nodules, and nodules as areas of increased density). In this case, the diagnostic information should include a label indicating that the slice thickness of the CT image is 5 mm as a second piece of information D2, and should include labels for positive characteristic items among the characteristic items included in the analysis results, and the training data should include training sentences in which the labels for positive characteristic items are written with "doubtful" instead of "confirmed" at the end, thereby constructing a learning model 23A.

[0066] The third piece of information, D3, which represents the radiologist's judgment on the medical image, is information representing the radiologist's interpretation of the lesion contained in medical image G0. Specifically, it includes the unconfirmed diagnosis of medical image G0, the relationship between the lesion and non-lesion tissues, and the radiologist's selection of multiple analysis results.

[0067] Regarding the generation of findings statements using the third piece of information D3, if the third piece of information D3 is an unconfirmed diagnostic result or a relationship between a lesion and tissue other than the lesion, the findings statement should be generated so that the third piece of information D3 is directly included in the findings statement. For example, findings statements such as "suspected primary lung cancer," "suspected mediastinal invasion," or "suspected intrapulmonary metastasis" should be generated. In this case, the diagnostic information should include a label such as "suspected mediastinal invasion" as the third piece of information D3, and the learning model 23A should be constructed by training the recurrent neural network 40 using training data that includes training text describing the label "suspected mediastinal invasion."

[0068] Furthermore, regarding the radiologist's selection of multiple analysis results, the report should be generated to include only the characteristic items selected by the radiologist. For characteristic items not selected by the radiologist, they should not be included in the report, but the report may be generated in a way that lowers the confidence level, reduces their importance, or obscures their significance.

[0069] The fourth piece of information, D4, which represents the results of tests performed on the patient, includes interpretation results for medical images acquired using a different imaging device than the one used to acquire medical image G0, as well as test results other than those using images, such as blood tests.

[0070] Regarding the generation of findings statements using the fourth piece of information D4, differentiation from tuberculoma, a lung disease, is performed by combining imaging diagnostics with blood tests. Therefore, diagnostic information such as blood test results should be directly included in the findings statement. For example, if the blood test result is "QuantiFERON negative" and the information of suspected symptoms based on the blood test result is "non-tuberculous mycobacterial infection," the findings statement should be written as "Since the QuantiFERON result is negative, non-tuberculous mycobacterial infection is suspected." In this case, the diagnostic information should include, for example, the label "blood test" and the label of suspected symptoms as the fourth piece of information D4, and a learning model 23A should be constructed by training a recurrent neural network 40 using training data that includes training text describing these labels. In this case, the learning model 23A may be constructed to generate findings statements that include labels of characteristic items included in the analysis results, or it may be constructed to generate findings statements that do not include labels of characteristic items.

[0071] Furthermore, when using other diagnostic equipment such as contrast-enhanced CT scanners or PET-CT scanners as diagnostic information, the diagnostic information should also be included in the findings statement. For example, if the test result is "low FDG uptake on mPET" and the suspected symptom based on the test result is "round atelectasis or organizing pneumonia," the findings statement should be written as "Low FDG uptake on mPET suggests round atelectasis or organizing pneumonia." It is also possible to include analysis results related to the diagnostic information in the findings statement, or to omit analysis results unrelated to patient information. Additionally, depending on whether the analysis results are related to the fourth piece of information D4, the findings statement may be generated in a way that includes or excludes analysis results related to the fourth piece of information D4, lowers or raises the confidence level, lowers or raises the importance level, or blurs or asserts them.

[0072] The following describes combinations of diagnostic information and analysis results, and examples of findings generated by the text generation unit 23. Figure 6 shows an example of diagnostic information and analysis results. The diagnostic information shown in Figure 6 is the first information D1. As shown in Figure 6, the first information D1 is characterized by the type of lesion being "tumor," the diameter being "maximum diameter 46 mm," the confirmed diagnosis of malignancy being "primary lung cancer," the treatment being "after Iressa treatment," and the change in size being "increased." These labels are Nodule, Diameter, Malignant, Treated, and Progress, respectively.

[0073] Furthermore, the analysis results showed that the lesion was located in the "upper left segment" of the lung, and the attenuation type was solid, with spicules, calcification, no cavities, and pleural contact. These were labeled as Segment, Solid, Spiculated+, Calcification+, Cavity-, and PleuralContact+, respectively.

[0074] The diagnostic information shown in Figure 6 includes a confirmed malignant diagnosis of primary lung cancer. In this case, the change in lesion size over time is more important than the analysis results of internal characteristics such as the presence of spicules. Therefore, a learning model 23A is constructed to generate findings that include only the change in size over time and not the analysis results representing internal characteristics, and to generate findings that do not include analysis results that contradict the confirmed diagnosis (for example, findings that suggest benignity when there is a confirmed diagnosis of primary lung cancer).

[0075] For example, a learning model 23A is constructed by training a recurrent neural network using training data that includes the labels of the diagnostic information and analysis results shown in Figure 6, and the training text, "The [Malignant] of [Segment] is [Treated]. The [Nodule] has further increased to [Diameter]." In the constructed learning model 23A, when the diagnostic information and analysis results are input, the analysis results are selected based on the diagnostic information to generate a report. Specifically, when the diagnostic information and analysis results shown in Figure 6 are input, the learning model 23A is constructed by training to output the report, "The [Malignant] of [Segment] is [Treated]. The [Nodule] has further increased to [Diameter]." The text generation unit 23 generates the following finding statement by embedding diagnostic information and analysis results into the labels of the finding statements output by the learning model 23A: "This patient has been treated with Iressa for primary lung cancer in the upper left section. The tumor has further increased in size to a maximum diameter of 46 mm."

[0076] Furthermore, as explained in the second piece of information D2, the third piece of information D3, and the fourth piece of information D4 above, in order to lower the confidence level of the analysis results in the findings statement, one should generate findings statements written in non-definitive language. Conversely, in order to increase the confidence level of the analysis results, one should generate findings statements written in definitive language. Examples of non-definitive language include "~is suspected," and examples of definitive language include "~is confirmed." Thus, in order to lower or raise the confidence level of the analysis results in the generated findings statement, a learning model 23A should be constructed by training the recurrent neural network 40 with training texts that increase the confidence level of the analysis results or training texts that decrease the confidence level, depending on the diagnostic information.

[0077] For example, as shown in Figure 7, the labels for the diagnostic information are Nodule [tumor] and Diameter [longest diameter 48 mm], and the labels for the analysis results are Segment [right lower lobe S6], Solid [solid type], Irregular Form [irregular shape], Spiculated+ [with spicula], Lobulated+ [with lobulated shape], Airbronchogram+ [with bronchial radiolucency], Cavity+ [with cavity], Calcification- [no calcification], and Pleural Contact+ [with pleural contact]. The contents in parentheses represent the specific content of each label. In this case, the findings statement generated by the text generation unit 23 is: "An irregular, solid tumor with a longest diameter of 48 mm is observed in the right lower lobe S6, in contact with the pleura. It is lobulated and has spicula. Bronchial radiolucency and a cavity are observed internally. No calcification is observed." In the generated findings, for characteristic items where the presence or absence is clear, the sentence ends with "~is acknowledged," "~is not acknowledged," or "~is accompanied."

[0078] On the other hand, as shown in Figure 8, the labels for the diagnostic information are Nodule [tumor] and Diameter [longest diameter 48 mm], and the labels for the analysis results are Segment [right lower lobe S6], Solid [solid type], Irregular Form [irregular shape], Spiculated+ [with spicula], Lobulated+ [with lobulated shape], Airbronchogram? [bronchial radiolucency unclear], Cavity? [cavity unclear], Calcification- [no calcification], and Pleural Contact+ [pleural contact present]. The characteristic items marked with "?" represent the analysis results of a false positive.

[0079] In this case, the findings statement generated by the text generation unit 23 will read: "An irregular, solid mass measuring 48 mm in length is observed in the right lower lobe S6, adjacent to the pleura. It is lobulated and accompanied by spicules. A low-attenuation area is observed internally, suggesting bronchial radiolucency and a cavity. No calcification is observed." In the generated findings statement, characteristics where the presence or absence is clear are concluded with "~is observed," "~is not observed," or "~is accompanied," while characteristics where the presence or absence is unclear, i.e., characteristics that are suspected to be false positives, are concluded with "~is suspected."

[0080] To change the confidence level of the generated sentences in this way, one-hot representations of "Airbronchogram? [Unclear bronchial luminescence]" and "Cavity? [Unclear cavity]" are defined in advance. By training a neural network using training data that includes training sentences containing the labels "Airbronchogram? [Unclear bronchial luminescence]" and "Cavity? [Unclear cavity]" and ending with "~suspected," it becomes possible to generate observation sentences with different confidence levels depending on the label.

[0081] Furthermore, by adding an expression vector corresponding to the level of confidence, i.e., the expression "suspected," to the vector representation xt of the bronchial radiolucency and cavity input to the decoder 42 shown in Figure 5, it becomes possible to generate a findings statement that ends with "suspected." Alternatively, by deriving vector representations in which the component that is 1 in the 1-hot representations of "bronchial radiolucency" and "cavity" is changed according to the level of confidence, it is also possible to generate a findings statement that ends with "suspected." In this case, for example, if the 1-hot representation of "bronchial radiolucency" is (1,0,0), the 1-hot representation should be changed to (0.5,0,0) according to the level of confidence.

[0082] This constructs a learning model 23A that generates sentences ending with high confidence for characteristic items where the presence or absence is clear, but generates sentences ending with low confidence for characteristic items where the presence or absence is unclear.

[0083] On the other hand, regarding importance, each item included in the diagnostic information and analysis results is assigned an order according to its importance. In this embodiment, a table defining the importance of each item in the diagnostic information and analysis results is stored in the storage 13. The text generation unit 23 refers to the table stored in the storage 13 and assigns an order according to the importance of each item in the diagnostic information and analysis results. Then, the recurrent neural network 40 is trained to construct a learning model 23A that changes the importance according to whether the characteristics are negative or positive, and further according to the diagnostic information, and generates a report containing a predetermined number of analysis results in order of importance.

[0084] For example, calcification is generally a benign characteristic. Also, negative characteristics are not as important as positive characteristics. Therefore, the learning model 23A is constructed to lower the importance of calcification and negative characteristics included in the analysis results, increase the importance of positive characteristics, and generate the findings statement using a predetermined number of highly important characteristics. Furthermore, depending on the confirmed diagnostic results included in the diagnostic information, it may be better to include negative characteristics for specific characteristics in the findings statement. In this case, the learning model 23A is constructed to generate the findings statement with increased importance even if a particular characteristic is negative, depending on the diagnostic information.

[0085] For example, as shown in Figure 9, the labels for the diagnostic information are Nodule [tumor] and Diameter [longest diameter 24 mm], and the labels for the analysis results are Segment [right lower lobe S6], Solid [solid type], Lobulated+ [lobulated], Airbronchogram- [no bronchial radiolucency], Cavity- [no cavity], Calcification- [no calcification], and PleuralContact+ [pleural contact]. Also, as shown in Figure 9, each item of the diagnostic information and analysis results is assigned a number in order of importance. In this embodiment, the learning model 23A is constructed to include the top n (e.g., 5) labels with the highest importance for the diagnostic information and analysis results.

[0086] In this case, as shown in Figure 9, when the diagnostic information and analysis results assigned importance are input to the learning model 23A, the following finding statement is output: "A [Pleural Contact+] [Diameter] [Lobulated+] [Nodule] is observed in [Segment]." The text generation unit 23 generates the following finding statement by embedding the diagnostic information and analysis results into the labels included in the finding statement output by the learning model 23A: "A lobulated nodule with a major diameter of 24 mm is observed in the right lower lobe S6, in contact with the pleura."

[0087] Furthermore, as shown in Figure 10, in addition to the diagnostic information and analysis results shown in Figure 9, suppose that HistoryOsteosarcoma (history of osteosarcoma) is added to the diagnostic information as the first piece of information D1, medical history. When there is a history of osteosarcoma, calcifications similar to benign ones may form. Therefore, when there is a history of osteosarcoma, it is necessary to ensure that the physician confirms the recurrence and metastasis of osteosarcoma contained in medical image G0 when reading the findings statement. Accordingly, when a history of osteosarcoma is included in the diagnostic information, the learning model 23A is constructed to increase the importance of the label "no calcification" when generating the findings statement. As a result, for example, in the learning model 23A, the importance of "no calcification" is changed to 4, and the importance of the characteristic items, which were 4-8 in importance in Figure 9, is changed to 5-9 when generating the findings statement. As a result, the findings statement generated by the text generation unit 23 will be: "A lobulated nodule measuring 24 mm in its longest diameter is observed in the right lower lobe S6. Osteosarcoma is present. No calcification is observed." Alternatively, the learning model 23A may be constructed to include a history of osteosarcoma in the findings statement.

[0088] The display control unit 24 displays the medical text generated by the text generation unit 23 on the display 14.

[0089] Next, the processing performed in the first embodiment will be described. Figure 11 is a flowchart showing the processing performed in the first embodiment. When an instruction to start processing is given, the information acquisition unit 21 acquires the medical image G0 and diagnostic information that are to be used to generate the findings statement (step ST1). Next, the analysis unit 22 analyzes the medical image G0 and derives the analysis results of the medical image G0 (step ST2). Then, the text generation unit 23 generates a findings statement about the patient as a medical document based on the analysis results and diagnostic information derived by the analysis unit 22 (step ST3). Furthermore, the display control unit 24 displays the medical document on the display 14 (step ST4), and the processing ends.

[0090] Thus, in this embodiment, medical documents about the patient are generated based on the analysis results and diagnostic information. Therefore, the generated medical documents reflect not only the analysis results but also the diagnostic information about the patient. As a result, according to this embodiment, it is possible to generate documents that accurately describe the patient's condition.

[0091] Furthermore, by generating medical documents that include analysis results selected based on diagnostic information, it is possible to generate medical documents that include necessary analysis results corresponding to the diagnosis and exclude unnecessary analysis results.

[0092] Furthermore, by generating medical documents containing analysis results with a priority order based on diagnostic information, it is possible to generate medical documents in which the analysis results are described with a priority order that reflects the diagnostic information.

[0093] Furthermore, by generating medical documents that include both analysis results and diagnostic information, it is possible to generate medical documents that allow reference to both analysis results and diagnostic information.

[0094] In the above embodiment, the medical document is generated by inputting diagnostic information and analysis results into the learning model 23A of the document generation unit 23, but the invention is not limited to this. The document generation unit 23 may also select analysis results according to the diagnostic information and generate the medical document using the selected analysis results. This will be described below as the second embodiment. Figure 12 shows the functional configuration of the information processing device according to the second embodiment. In Figure 12, the same reference numerals are used for components identical to those in Figure 3, and detailed explanations are omitted. As shown in Figure 12, the information processing device according to the second embodiment differs from the above embodiment in that the document generation unit 23 includes a selection unit 25 and a learning model 23B.

[0095] The selection unit 25 selects and rejects analysis results derived by the analysis unit 22 based on the diagnostic information. For this purpose, in the second embodiment, a table defining rules for selecting and rejecting analysis results according to the diagnostic information is stored in the storage 13. Figure 13 shows an example of a table defining the rules. As shown in Figure 13, table T1 defines each item of the analysis results horizontally and each item included in the diagnostic information vertically, defining whether or not to input each item of the diagnostic information and analysis results into the learning model 23B. In table T1, items of analysis results that are not selected for the diagnostic information are marked with ×, and items that are selected are marked with ○.

[0096] In Table T1, the analysis results specify the attenuation value (solid and ground-glass type), margins (presence or absence of spicules), internal characteristics (presence or absence of calcification, presence or absence of cavities), and periphery (presence or absence of pleural invagination). In the second embodiment, the diagnostic information used is the same as in Figure 5, including the segment, diameter, definitive diagnosis of malignancy, treatment content, and size changes. However, in the columns of Table T1, only the diameter and definitive diagnosis of malignancy are specified for simplicity of explanation. Furthermore, the diameter is divided into less than 5 mm (<5 mm) and 5 mm or more but less than 10 mm (<10 mm).

[0097] In Table T1, small abnormal shadows become difficult to examine in detail due to the partial volume effect when the CT image slice interval is 5 mm. Therefore, from the attenuation values, margins, and internal characteristics included in the analysis results, the presence or absence of cavities and surrounding area items are removed. However, since calcification is highly visible in the image due to its high brightness, the item indicating the presence of calcification is retained in the analysis results. Also, for abnormal shadows with a diameter of 5 mm or more but less than 10 mm, it is difficult to confirm the internal characteristics of the abnormal shadow, so negative internal characteristics are removed from the analysis results. However, for calcification among the internal characteristics, both positive and negative cases are retained in the analysis results. Furthermore, when a definitive diagnosis of malignancy is included, the change in lesion size over time is more important than the analysis results of internal characteristics such as the presence of spicules. Therefore, attenuation values, margins, internal characteristics, and surrounding area items are removed from the analysis results.

[0098] In addition, if there is an overlap in the decision of whether or not to include the diameter and the definitive diagnosis of malignancy in the analysis results in Table T1, the decision to remove from the analysis results will take precedence, and the analysis results will be selected accordingly. Therefore, if the diameter is 5 mm or more and less than 10 mm, and the definitive diagnosis of malignancy is included in the diagnostic information, the selection unit 25 will refer to Table T1 and essentially select the analysis results according to the definitive diagnosis of malignancy.

[0099] Furthermore, the learning model 23B in the second embodiment is constructed by training a neural network, such as a recurrent neural network, using training data that associates combinations of various analysis results and various diagnostic information after selection with training texts to be generated from the analysis results and diagnostic information, similar to the learning model 23A described above. In the first embodiment, the learning model 23A generates observation texts by selecting from the input analysis results, but in the second embodiment, the analysis results input to the learning model 23B have already been selected. Therefore, the learning model 23B generates observation texts using the input analysis results and diagnostic information.

[0100] For example, in the second embodiment, when the information acquisition unit 21 acquires the diagnostic information and analysis results shown in Figure 5, the selection unit 25 refers to table T1 and selects the analysis results. Since the diagnostic information shown in Figure 5 includes a definitive diagnosis of malignancy, the selection unit 25 removes the attenuation value, margin, internal characteristics, and surrounding items from the analysis results and inputs the diagnostic information and the selected analysis results into the learning model 23B. Specifically, the selection unit 25 inputs Nodule, the label for the tumor, Diameter, the label for the maximum diameter of 46 mm, Segment, the label for the upper left section, Malignant, the label for primary lung cancer, Treated, the label for after Iressa treatment, and Progress, the label for growth, into the learning model 23B. The learning model 23B then outputs the finding statement, "[Segment] is [Malignant] and [Treated]. [Nodule] has further increased to [Diameter]." The text generation unit 23 generates the following finding statement by embedding diagnostic information and analysis results into the labels included in the finding statement output by the learning model 23A: "This patient has been treated with Iressa for primary lung cancer in the upper left segment. The tumor has further increased in size to a maximum diameter of 46 mm."

[0101] In the second embodiment described above, the analysis results of the analysis unit 22 may be the characteristic score itself for each characteristic item. In this case, the selection unit 25 compares the characteristic score with a threshold to determine whether the characteristic item is positive or negative, but the threshold may be changed according to the diagnostic information. For example, if the diagnostic information and analysis results are as shown in Figure 10, the threshold for determining calcification may be reduced to determine that calcification is present. In this case, if table T1 is derived so that "calcification present" is recorded in the analysis results, a findings statement including "calcification present" will be generated when there is a history of osteosarcoma. Therefore, radiologists who read the findings statement will place more importance on calcification when interpreting medical images.

[0102] Furthermore, while the above embodiments apply the technology of this disclosure to generating findings to be included in a medical imaging report, it is not limited to this. For example, the technology of this disclosure may also be applied to create medical documents other than imaging reports, such as electronic medical records and diagnostic reports, as well as other documents containing strings related to images.

[0103] Furthermore, while the above embodiments use medical image G0 with the lungs as the target of diagnosis for various processing, the target of diagnosis is not limited to the lungs. In addition to the lungs, any part of the human body such as the heart, liver, brain, and limbs can be used as the target of diagnosis.

[0104] Furthermore, in each of the above embodiments, the processing of the analysis unit 22 in the information processing device 20 contained within the image interpretation WS3 may be performed by an external device, such as another analysis server connected to the network 10. In this case, the external device acquires the medical image G0 from the image server 5 and derives the analysis results by analyzing the medical image G0. The information processing device 20 then generates a report using the analysis results derived by the external device.

[0105] Furthermore, in the above embodiment, the hardware structure of the Processing Unit, which executes various processes such as the information acquisition unit 21, the analysis unit 22, the text generation unit 23, the display control unit 24, and the selection unit 25, can be the various processors shown below. As mentioned above, these various processors include a CPU, which is a general-purpose processor that executes software (programs) and functions as various processing units, as well as a Programmable Logic Device (PLD), which is a processor whose circuit configuration can be changed after manufacturing, such as an FPGA (Field Programmable Gate Array), and a dedicated electrical circuit, which is a processor with a circuit configuration specifically designed to execute a particular process, such as an ASIC (Application Specific Integrated Circuit).

[0106] A single processing unit may be composed of one of these various processors, or it may be composed of a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs or a combination of a CPU and an FPGA). Alternatively, multiple processing units may be composed of a single processor. Examples of composing multiple processing units with a single processor include, firstly, a configuration in which one or more CPUs and software are combined to form a single processor, as is typical of computers such as client and server systems, and this processor functions as multiple processing units. Secondly, a configuration using a processor that realizes the functions of the entire system, including multiple processing units, on a single IC (Integrated Circuit) chip, as is typical of a System on a Chip (SoC). Thus, various processing units are configured as hardware structures using one or more of the above-mentioned various processors.

[0107] Furthermore, the hardware structure of these various processors can more specifically utilize electrical circuits (Circuitry) that combine circuit elements such as semiconductor elements.

[0108] The following are additional notes to this disclosure. (Additional note 1) Equipped with at least one processor, The aforementioned processor, Obtain one or more analysis results regarding the patient's medical images, Obtain diagnostic information regarding the patient's diagnosis other than the analysis results mentioned above, An information processing device that generates medical documents relating to the patient based on the analysis results and the diagnostic information. (Additional note 2) The processor selects the analysis results based on the diagnostic information, The information processing device described in Appendix 1 generates medical documents that include the selected analysis results. (Additional note 3) The processor is an information processing device according to appendix 1 or 2 that generates the medical document including the analysis results in order of priority according to the diagnostic information. (Additional note 4) The processor is an information processing device according to any one of the appendix items 1 to 3 that generates the medical document including the diagnostic information and the analysis results. (Additional note 5) The diagnostic information is an information processing device according to any one of the appendix items 1 to 4, which includes first information that has been confirmed regarding the lesion contained in the medical image. (Additional note 6) The information processing device according to Appendix 5, wherein the first information includes at least one of the measurement results of the lesion, the definitive diagnosis results for the lesion, and the patient's medical history. (Additional note 7) The information processing device according to any one of the appendix items 1 to 6, wherein the diagnostic information includes a second, confirmed piece of information other than information relating to lesions contained in the medical image. (Additional note 8) The information processing device according to Appendix 7, wherein the second information includes at least one of the purpose of the examination in which the medical image was acquired and image conditions relating to the medical image. (Additional note 9) The diagnostic information is an information processing device according to any one of the appendix items 1 to 8, which includes a third piece of information representing the judgment result of a radiologist on the medical image. (Additional note 10) The information processing device according to Appendix 9, wherein the third information includes at least one of the undetermined diagnostic result relating to the medical image, the relationship between the lesion contained in the medical image and other tissues other than the lesion, and the selection result of the analysis by the radiologist. (Additional note 11) The information processing device described in any one of the appendix items 1 to 10, wherein the diagnostic information includes a fourth piece of information representing the results of tests performed on the patient. (Additional note 12) The information processing device according to Appendix 11, wherein the fourth information includes at least one of the following: the results of an examination performed on a diagnostic device different from the imaging device used to acquire the medical image of the patient; the results of an analysis of a medical image of a different type than the medical image; and the results of an examination of the patient's biological information. (Additional note 13) Obtain one or more analysis results regarding the patient's medical images, Obtain diagnostic information regarding the patient's diagnosis other than the analysis results mentioned above, An information processing method for generating medical documents relating to the patient based on the analysis results and the diagnostic information. (Additional note 14) Procedures for obtaining one or more analysis results regarding a patient's medical images, A procedure for obtaining diagnostic information related to the diagnosis of the patient other than the aforementioned analysis results, An information processing program that causes a computer to execute a procedure for generating medical documents concerning the patient based on the analysis results and the diagnostic information. [Explanation of Symbols]

[0109] 1. Medical Information System 2. Imaging device 3 Image Interpretation Workshop 4. Medical Workshop 5 Image Server 6 Image Database 7. Report Server 8 Report Database 10 Networks 11 CPU 12. Information Processing Programs 13 Storage 14 displays 15 Input section 16 memory 17 Network Interface 18 bus 20 Information Processing Devices 21 Information Acquisition Department 22 Analysis Department 22A Learning Model 23 Sentence generation section 23A, 23B Learning Models 24 Display Control Unit 25 Selection Section T1 Table

Claims

1. Equipped with at least one processor, The aforementioned processor, Obtain one or more analysis results regarding the patient's medical images, Diagnostic information is obtained that includes information representing changes in the patient other than the analysis results mentioned above. An information processing device that takes the analysis results and diagnostic information including information representing the changes as input, and uses a learning model trained to generate medical documents about the patient, to generate medical documents about the patient based on the analysis results and diagnostic information including information representing the changes.

2. The information processing device according to claim 1, wherein the information representing the change is information representing the change over time from the size of a lesion contained in medical images previously acquired for the same patient.

3. The information processing apparatus according to claim 2, wherein the information representing the change is information representing whether the size of the lesion has increased, decreased, or remained unchanged.

4. The size of the lesion is determined to be at least one of the length and width or area of ​​the lesion, as described in claim 2 or 3 of the information processing apparatus.

5. The information processing device according to any one of claims 1 to 4, wherein the processor generates the findings statement described in the image interpretation report as the medical document.

6. The processor selects the analysis results based on the diagnostic information, The information processing device according to any one of claims 1 to 5, which generates medical documents including the selected analysis results.

7. The information processing apparatus according to any one of claims 1 to 6, wherein the processor generates the medical document including the analysis results in order of priority according to the diagnostic information.

8. The information processing apparatus according to any one of claims 1 to 7, wherein the processor generates the medical document including the diagnostic information and the analysis results.

9. The information processing device according to any one of claims 1 to 8, wherein the diagnostic information includes at least one of a definitive diagnosis result for a lesion contained in the medical image and the patient's medical history.

10. The information processing apparatus according to any one of claims 1 to 9, wherein the diagnostic information includes a second confirmed piece of information other than information relating to a lesion contained in the medical image.

11. The information processing apparatus according to claim 10, wherein the second information includes at least one of the purpose of the examination in which the medical image was acquired and image conditions relating to the medical image.

12. The information processing device according to any one of items 1 to 11, wherein the diagnostic information includes a third piece of information representing the result of a radiologist's judgment on the medical image.

13. The information processing device according to claim 12, wherein the third information includes at least one of an undetermined diagnostic result relating to the medical image, the relationship between a lesion included in the medical image and other tissues other than the lesion, and the selection result of the analysis by the radiologist.

14. The information processing device according to any one of items 1 to 13, wherein the diagnostic information includes a fourth piece of information representing the results of tests performed on the patient.

15. The information processing apparatus according to claim 14, wherein the fourth information includes at least one of the following: the results of an examination performed on a diagnostic device different from the imaging device used to acquire the medical image of the patient; the results of an analysis of a medical image of a different type than the medical image; and the results of an examination of the patient's biological information.

16. The computer obtains one or more analysis results regarding the patient's medical images. Diagnostic information is obtained that includes information representing changes in the patient other than the analysis results mentioned above. An information processing method for generating medical documents about a patient, using a learning model trained to generate medical documents about the patient, with the aforementioned analysis results and diagnostic information including information representing the changes as input.

17. A procedure for obtaining one or more analysis results regarding a patient's medical images, A procedure for obtaining diagnostic information that includes information representing changes in the patient other than the aforementioned analysis results, An information processing program that causes a computer to execute a procedure for generating a medical document about the patient based on the analysis results and diagnostic information including information representing the changes, using a learning model trained to generate a medical document about the patient.

Images