Medical treatment support device

Abstract

To provide a medical treatment support device for generating information to be utilized for treatment of a patient by processing a wide range of information representing a condition of the patient by a minimum type of artificial intelligence.SOLUTION: A medical treatment support device 1 includes a machine learning part 111 for performing machine learning relating to generation of advice by using a combination of a text representing a condition of a patient and advice to be utilized for treatment of the patient as learning data, for a large-scale language model having been pre-trained with a text of 500GB or more, an acquisition part 112 for acquiring the text representing the condition of the patient, and a generation part 116 for inputting the text into the large-scale language model and causing the large-scale language model to generate advice to be utilized for the treatment of the patient.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a medical support device. [Background technology]

[0002] Medical professionals use various information indicating the condition of patients, such as medical questionnaires, medical records, test results, test images, etc., in treating patients. In order to reduce the burden on medical professionals, there is a demand for technology that processes this information using artificial intelligence and generates information that can be used in treating patients.

[0003] Regarding generating information to be utilized in the treatment of a patient from information indicating the condition of the patient, Patent Document 1 discloses a medical information processing system including an artificial intelligence engine that processes medical information based on a database, the medical information processing system including a reception unit that receives medical information including medical images, a first classification processing unit that classifies the medical images into one of two or more categories set in advance based on the database, a selection processing unit that selects one of the two or more categories based on the medical information, and a second classification processing unit that classifies the medical images into a peculiar category when the category specified by the first classification processing unit does not match the category selected by the selection processing unit. The technology of Patent Document 1 can effectively perform processing such as learning of artificial intelligence using medical images. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2018-014059 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, the technology of Patent Document 1 is only capable of effectively performing processing such as learning of artificial intelligence using medical images, and there is room for further improvement in terms of processing a wide range of information, not limited to images such as medical questionnaires, medical records, test results, test images, etc., using a minimum type of artificial intelligence to generate information that can be used in treating patients.

[0006] The present invention has been made in light of the above circumstances, and its purpose is to process a wide range of information indicating a patient's condition using a minimum of artificial intelligence to generate information that can be used to treat the patient. [Means for solving the problem]

[0007] As a result of intensive research into solving the above problem, the inventors have found that the above object can be achieved by acquiring information indicating a patient's condition in the form of text and processing the text in a large-scale language model that has been subjected to machine learning in advance. The inventors have then completed the present invention. Specifically, the present invention provides the following.

[0008] The present invention provides a medical support device comprising: a machine learning unit that performs machine learning related to generating advice using a combination of text indicating a patient's condition and advice to be used in the treatment of the patient as training data for a large-scale language model that has been pre-trained with 500 GB or more of text; an acquisition unit that acquires text indicating the patient's condition (first text); and a generation unit that inputs the first text to the large-scale language model and causes the large-scale language model to generate advice to be used in the treatment of the patient.

[0009] In the present invention, the acquisition unit can acquire information indicating a patient's condition in the form of text. However, various types of text information, such as medical questionnaires and charts written in natural language, and texts of test values, have different writing styles. Therefore, if these types of text information are structured by an artificial intelligence that performs natural language analysis processing, such as Microsoft Azure's (registered trademark) LUIS (Language Understanding), it is considered that various types of structured information that differ for each writing style can be obtained. There is a concern that a device that handles such various types of structured information will have a complex configuration using various types of artificial intelligence that are tailored to the various types of structured information.

[0010] It is known that a large-scale language model, which is a language model pre-trained with a large amount of text data, can generate an appropriate response as an output for a natural language given as an input. The present invention uses a large-scale language model pre-trained with 500 GB (gigabytes) or more of text. It is considered that as the amount of training data increases, the performance of the language model increases along the "Broken Neural Scaling Laws" (BNSL), and the language model acquires emergent capabilities when transitioning between segments in the BNSL. Therefore, the large-scale language model acquires emergent capabilities related to identifying the meaning of text, which are acquired according to the amount of text used in pre-training. However, even if the large-scale language model acquires such emergent capabilities, it is unclear whether it will generate appropriate advice to be utilized in the treatment of a patient when text indicating the patient's condition is input.

[0011] In the medical field, inappropriate advice can affect the health and life of a patient. Therefore, the present invention performs machine learning for generating advice on such a large-scale language model using a combination of text indicating a patient's condition and advice to be used in the patient's treatment as learning data. As a result, the present invention can input the acquired text and cause the large-scale language model to generate appropriate advice to be used in the patient's treatment. Therefore, the present invention can process text of a wide range of information indicating a patient's condition using the minimum type of artificial intelligence, the large-scale language model described above, and generate information to be used in the patient's treatment. Effect of the Invention

[0012] The present invention is capable of processing a wide range of information indicative of a patient's condition using a minimal amount of artificial intelligence to generate information that can be utilized in treating the patient. [Brief description of the drawings]

[0013] [Figure 1] FIG. 1 is a block diagram showing the hardware and software configurations of a system S according to the present embodiment. [Diagram 2] FIG. 2 is a main flowchart showing an example of a preferable flow of the support process executed by the support device 1 of this embodiment. [Diagram 3] FIG. 3 is a continuation of FIG. [Figure 4] FIG. 4 is a continuation of FIG. [Diagram 5] FIG. 5 is a continuation of FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] Hereinafter, an example of an embodiment of the present invention will be described in detail with reference to the drawings.

[0015] <System S> 1 is a block diagram showing the hardware and software configurations of a system S of this embodiment. Hereinafter, an example of a preferred embodiment of the hardware and software configurations of the system S of this embodiment will be described with reference to FIG.

[0016] The system S is composed of a medical support device 1 that processes a wide range of information indicating a patient's condition using a minimum type of artificial intelligence and generates information to be used in treating the patient, and a terminal T that is configured to communicate with the support device 1 via a network N.

[0017] [Support device 1] The medical support device 1 includes a control unit 11, a storage unit 12, and a communication unit 13. There is no particular limitation on the type of the support device 1. Examples of the type include a server, a cloud server, and the like.

[0018] The support device 1 is configured to acquire a wide range of information in the form of text that indicates the condition of a patient, such as test values ​​measured by a wearable device, test images such as X-ray images and CT images, reports from emergency personnel, and clinical trial charts, and process the acquired text using a large-scale language model such as ChatGPT.The support device 1 then generates advice in the large-scale language model, such as expected illnesses, appropriate destinations for the illnesses involved in emergency transport, and the number of subjects required for the clinical trial.

[0019] The assistance device 1 is configured to execute a process of making a large-scale language model that has been pre-trained using 500 GB or more of text perform machine learning, and a process of making the large-scale language model that has been subjected to machine learning generate advice. The execution of these processes is realized, for example, via an API related to the use of the large-scale language model. The following is a more detailed configuration example of the assistance device 1.

[0020] [Control unit 11] The control unit 11 includes a central processing unit (CPU), a random access memory (RAM), and a read only memory (ROM).

[0021] The control unit 11 cooperates with the storage unit 12 and / or the communication unit 13 as necessary. The control unit 11 then realizes software components of the program of this embodiment executed by the assistance device 1, such as a machine learning unit 111, an acquisition unit 112, a conversion unit 113, an image recognition unit 114, a voice recognition unit 115, a generation unit 116, and a speech unit 117. The functions provided by each of the software components of the program of this embodiment will be described in the description of a preferred flow of the assistance process, which will be described later.

[0022] [Storage section 12] The memory unit 12 is a device in which data and / or files are stored, and includes a storage unit that stores data non-temporarily using a hard disk, a semiconductor memory, a recording medium, a memory card, and the like.

[0023] The storage unit 12 may have a mechanism that enables connection via the network N to a storage device or storage system such as a network attached storage (NAS), a storage area network (SAN), cloud storage, a file server, and / or a distributed file system.

[0024] The memory unit 12 stores a wide range of information indicating the patient's condition, such as programs executed by the microcomputer, neural networks for image recognition, data for voice recognition, data for voice synthesis, medical questionnaires, test values ​​measured by wearable devices, test images such as X-ray images and CT images, reports from emergency personnel, and medical records from clinical trials.

[0025] (Large-scale language models) The large language model (LLM) in this embodiment is a type of probabilistic model used in natural language processing, and is a model for probabilistically predicting how likely a given word or sentence is to occur as natural language.

[0026] The large-scale language model may be stored in an external cloud server or the like and executed via an API (Application Programming Interface) related to the use of the large-scale language model, or may be stored in the storage unit 12. By executing the large-scale language model via an API, the support device 1 has a simpler configuration and is more cost-effective than a configuration in which the large-scale language model is stored in the storage unit 12. By storing the large-scale language model in the storage unit 12, the support device 1 can execute processing related to the large-scale language model without communicating with an external device such as a cloud server point by point. This allows the support device 1 to reduce processing delays related to communication, risks to information security, and the like.

[0027] The large-scale language model is trained on a large amount of text, at least 500 GB (gigabytes). Examples of such models include ChatGPT, GPT-3.5, and GPT-4 from OpenAI (registered trademark), and LLaMa from Meta AI (registered trademark). It is considered that as the amount of training data increases, the performance of the language model increases along the broken neural scaling law (BNSL), and the model acquires emergent capabilities when transitioning between segments in the BNSL. Therefore, the large-scale language model of this embodiment has an emergent capability for identifying the meaning of the text, which is acquired according to the amount of text used in pre-training.

[0028] The large-scale language model is preferably a model having a large number of parameters, at least 10 billion or more, so that the large-scale language model can perform processing that fully reflects the emergent capabilities obtained from a large amount of text.

[0029] (Neural network for image recognition) When machine learning of the large-scale language model is performed using examination image-related learning data described below, it is preferable that an image recognition neural network that receives an examination image as an input and outputs text of findings on the examination image is stored in the storage unit 12. In this way, the assistance device 1 can generate and provide advice to be utilized in treating a patient not only from text indicating the patient's condition but also from examination images indicating the patient's condition, simply by adding a minimum type of artificial intelligence, namely an image recognition neural network, to the artificial intelligence of the large-scale language model.

[0030] The neural network for image recognition is pre-trained using training data that associates examination images with text of findings for the examination images. The text of findings is not particularly limited and may be text of findings provided by a medical professional who is skilled in reading the examination images. In this way, the support device 1 can generate text of findings for the examination images from the examination images and input the text to the large-scale language model.

[0031] (Voice recognition data) It is preferable that the storage unit 12 stores voice recognition data used for recognizing text corresponding to the voice from the voice data. This allows the support device 1 to perform voice recognition on the voice indicating the patient's condition, and input the text obtained by the voice recognition to the large-scale language model.

[0032] (Data for voice synthesis) It is preferable that the storage unit 12 stores voice synthesis data used for synthesizing a voice reading the text from the text, so that the assistance device 1 can provide advice generated by a large-scale language model in the form of voice.

[0033] (Questionnaire) It is preferable that a medical questionnaire is stored in the storage unit 12. This allows the support device 1 to generate advice to be utilized in treating a patient with a condition related to the medical questionnaire, and to provide the advice to the patient and / or medical staff, etc. It is preferable that the medical questionnaire is an electronic questionnaire written in MML (Medical Markup Language) to facilitate data exchange with other medical institutions.

[0034] The medical questionnaire contains information such as the patient's address, name, sex, and age, the patient's main complaint (presence or absence of fever, runny nose, diarrhea, etc.), medical history (diseases that the patient had in the past but recovered from), and the date of the medical treatment related to the questionnaire. By entering the main complaint on the medical questionnaire, the support device 1 can generate advice including the disease corresponding to the main complaint, treatment methods, and lifestyle improvement measures, and provide the advice to the patient and / or medical staff, etc.

[0035] (Medical Record) It is preferable that the storage unit 12 stores medical records of clinical trials conducted to confirm the efficacy and safety of new medicines and treatments. This allows the support device 1 to generate advice indicating the number of subjects examined based on the medical records and provide the advice to medical professionals and others involved in the clinical trials.

[0036] The medical record contains, for example, (1) the address, name, sex, and age of the patient, (2) the name of the patient's illness and major symptoms, (3) the treatment method for the patient, and (4) the date of the medical procedure related to the medical record. It is preferable that the medical record is an electronic medical record written in MML (Medical Markup Language) to facilitate data exchange with other medical institutions.

[0037] The medical record preferably includes (3) examination history data, which is a history of examinations, as a treatment method for the patient. The examinations include, for example, specimen examinations performed using specimens, examinations that generate examination images (medical images), physiological measurements, genetic data, etc.

[0038] Sample tests include, for example, tests to measure AST (aspartate aminotransferase) / ALT (alanine aminotransferase), which are used as indicators of liver function, albumin, which is used to evaluate nutritional status, creatinine, which is used to evaluate the filtration function of the glomeruli in the kidneys, and occult blood in urine, which is used as an indicator of diseases such as urinary tract stones and urinary tract infections.

[0039] The examination image includes, for example, medical images captured by an X-ray device, a computed tomography device (CT device), a magnetic resonance imaging device (MRI device), a positron emission tomography device (PET device), a single photon emission computed tomography device (SPECT device), an ultrasound examination device, etc. The effect of including these various medical images in the examination image will be described in detail later in the section on the machine learning step.

[0040] The physiological measurements include, for example, measurements by a wearable device of pulse, body temperature, blood pressure, blood glucose level, number of steps walked, electrocardiogram, etc. The effect of including these various measurements in the physiological measurements will be described in detail later in the section on machine learning steps.

[0041] The genetic data includes, for example, information on the base sequence of a gene, etc. In this way, the support device 1 can generate advice to be utilized in the treatment of a patient based on the relationship between a genetic pattern and a disease, and provide the advice to a patient and / or a medical professional, etc.

[0042] It is preferable that the medical record includes (3) a standard medical treatment plan (clinical pathway) for the patient as a treatment method for the patient. In this way, the support device 1 can generate advice to be used in the treatment of the patient based on the clinical pathway and provide the advice to the patient and / or medical staff, etc.

[0043] It is preferable that the medical record includes (3) medical fees (receipts) for insured medical treatment for the patient as a treatment method for the patient. This allows the support device 1 to generate advice to be used in the treatment of the patient based on the receipts and provide the advice to the patient and / or medical staff, etc.

[0044] [Communications Section 13] The communication unit 13 is not particularly limited as long as it connects the support device 1 to the network N and enables communication with the blockchain B and the terminal T. Examples of the communication unit 13 include a wireless device compatible with a mobile phone network, a device connectable to a wireless LAN, and a network card compatible with the Ethernet standard.

[0045] [Network N] The type of the network N is not particularly limited as long as it enables communication between the support device 1 and the terminal T, etc. The type of the network N is, for example, the Internet, a mobile phone network, a wireless LAN, etc.

[0046] [Terminal T] There is no particular limitation on the terminal T. The terminal T includes a terminal (user terminal) used by a user such as a patient or a medical professional, and a terminal (data terminal) that provides information (data) indicating the condition of a patient.

[0047] The user terminal includes, for example, a personal computer, a laptop computer, a smartphone, a tablet terminal, etc. The user terminal is capable of receiving advice generated by the assistance device 1 and executing a process of displaying the advice, etc.

[0048] The data terminal includes a terminal similar to the user terminal, as well as a wearable device capable of measuring one or more of pulse, body temperature, blood pressure, blood sugar level, number of steps walked, electrocardiogram, etc., a terminal capable of providing test values ​​measured by a testing device and test images taken by a testing device, a terminal for an ambulance crew member, a terminal for a medical institution, a terminal capable of providing information stored in a medical database, etc. The data terminal can execute a process of providing information indicating the condition of a patient to the support device 1, etc.

[0049] In particular, the data terminal preferably includes a wristband-type wearable device, such as a smart watch. This allows the support device 1 to acquire test values ​​related to one or more of pulse, body temperature, blood pressure, blood glucose level, number of steps walked, electrocardiogram, etc., from the wearable device that can be worn at all times. This allows the support device 1 to provide more useful advice based on daily test values ​​(biological data) and their fluctuations (for example, changes in blood pressure when waking up in the morning, etc.). This also allows the support device 1 to acquire test values ​​related to one or more of pulse, body temperature, blood pressure, blood glucose level, number of steps walked, electrocardiogram, etc., from the wearable device worn by the patient during the medical interview. This allows the support device 1 to provide more useful advice based on the test values ​​(biological data) during the medical interview.

[0050] [Main Flowchart of Support Processing] Fig. 2 is a main flow chart showing an example of a preferred flow of the support processing executed by the support device 1 of this embodiment. Fig. 3 is a diagram continuing from Fig. 2. Fig. 4 is a diagram continuing from Fig. 3. Fig. 5 is a diagram continuing from Fig. 4. The following is an example of a preferred flow of the support processing executed by the support device 1 of this embodiment using Figs. 2 to 5.

[0051] [Machine learning step] First, the control unit 11 executes the machine learning unit 111 in cooperation with the storage unit 12 etc. Then, the control unit 11 performs machine learning related to the generation of advice on the above-mentioned large-scale language model using a combination of text indicating a patient's condition and advice to be utilized in treating the patient as learning data (machine learning step).

[0052] By performing the machine learning step, the assistance device 1 can fine-tune the large-scale language model so that it generates appropriate advice to be utilized in the treatment of the patient when the acquired text is input. This allows the assistance device 1 to process the text of a wide range of information indicating the patient's condition using the above-mentioned large-scale language model, which is a minimum type of artificial intelligence, and generate information to be utilized in the treatment of the patient.

[0053] The machine learning in the machine learning step includes supervised learning (e.g., a task of predicting the next word) and unsupervised learning (e.g., a task of learning vector representations of words) using the above-mentioned learning data. Fine-tuning in the machine learning step includes adjustments such as a procedure of evaluating the learned language model and adjusting parameters such as weights of the model as necessary, and a procedure of improving the performance of the model by appropriately adjusting the values ​​of hyperparameters exemplified by the number of hidden layers of the neural network, the number of units, the learning rate, the strength of regularization, etc.

[0054] (Learning data) The learning data includes, for example, learning data related to generating advice indicating a disease predicted based on test values ​​from a wearable device and / or lifestyle habits a patient with the test values ​​should have in order to maintain their health (test value-related learning data), learning data related to generating advice indicating a disease predicted from a test image and / or lifestyle habits a patient with the test image should have in order to maintain their health based on the text of the image recognition results of the test image (test image-related learning data), learning data related to generating advice indicating a transport destination appropriate for dealing with a disease predicted from a report by emergency medical personnel based on the report (emergency transport destination-related learning data), learning data related to generating advice indicating the number of subjects required for a clinical trial based on the medical records of subjects in the clinical trial (clinical trial-related learning data), learning data related to news that affects health status (news learning data), etc.

[0055] The support device 1 of this embodiment uses a large-scale language model as artificial intelligence, and can use a wide variety of large amounts of information written in natural language as learning data. As a result, the large-scale language model becomes a language model that generates useful advice that reflects machine learning using a wide variety of information. Therefore, even though the types of artificial intelligence are kept to a minimum, the support device 1 of this embodiment can generate useful advice to be used in the treatment of patients using the large-scale language model that has undergone machine learning using the above-mentioned various types of learning data.

[0056] The test value-related learning data is data that associates text of one or more types of test values ​​(e.g., pulse, body temperature, blood pressure, blood glucose level, number of steps walked, electrocardiogram, etc.) that can be measured by a wearable device with advice indicating diseases predicted based on the test values ​​and / or lifestyle habits that a patient with the test values ​​should have in order to maintain their health. The test value-related learning data preferably includes information on test values ​​accumulated in large quantities in a medical database (medical big data). This allows the support device 1 to perform machine learning on a large-scale language model to generate more useful advice.

[0057] Since the test value-related learning data includes test value-related learning data related to pulse, the large-scale language model can generate, from a test value indicating bradycardia, advice indicating the possibility of a disease such as arteriosclerosis, thyroid disorder, or side effects of medicine, and advice indicating lifestyle habits that will lead to the improvement of the disease, etc. In addition, the large-scale language model can generate, from a test value indicating tachycardia, advice indicating the possibility of a disease such as angina pectoris, myocardial infarction, arrhythmia, heart failure, pulmonary embolism, emphysema, menopausal disorder, or hyperthyroidism, and advice indicating lifestyle habits that will lead to the improvement of the disease, etc. In addition, the large-scale language model can generate, from a test value indicating arrhythmia, advice indicating the possibility of a disease such as heart disease such as coronary artery disease, heart valve disorder, heart failure, or congenital heart disease, thyroid abnormality, lung abnormality, side effects of medicine, lifestyle problems such as stress, lack of sleep, overwork, smoking, excessive intake of alcohol / caffeinated beverages, or obesity, and advice indicating lifestyle habits that will lead to the improvement of the disease, etc.

[0058] By including test value-related learning data related to body temperature, the large-scale language model can generate, from a test value indicating low body temperature, advice indicating the possibility of a disease such as a decrease in basal metabolism or autonomic nerve disorder, and advice indicating lifestyle habits that will lead to the improvement of the disease. In addition, this allows the large-scale language model to generate, from a test value indicating fever, advice indicating the possibility of a disease such as various infectious diseases such as cold, influenza, COVID-19 infection, allergies, secondary immunodeficiency syndrome, chronic pyelonephritis, heat stroke, intestinal perforation, acute prostatitis, myocardial infarction, and unstable angina, and advice indicating lifestyle habits that will lead to the improvement of the disease.

[0059] Since the test value-related learning data includes test value-related learning data related to blood pressure, the large-scale language model can generate, from a test value indicating high blood pressure, advice indicating the possibility of various diseases that cause secondary hypertension, such as renal parenchymal abnormalities, renal vascular abnormalities, endocrine system abnormalities, vascular abnormalities, brain / central nervous abnormalities (brain stem vascular compression, etc.), genetic abnormalities, and drug-induced abnormalities, as well as advice indicating lifestyle habits that will lead to the improvement of the diseases. In addition, the large-scale language model can generate, from a test value indicating low blood pressure, advice indicating the possibility of various diseases that cause secondary hypotension, such as bleeding, heart disease, malnutrition due to gastrointestinal disease, endocrine abnormalities, and malignant neoplasms, as well as essential hypotension, as well as advice indicating lifestyle habits that will lead to the improvement of the diseases.

[0060] By including test value-related learning data related to blood glucose levels, the large-scale language model can generate, from a test value indicating hyperglycemia, advice indicating the possibility of a disease such as diabetes and advice indicating lifestyle habits that will lead to the improvement of the disease, etc. In addition, this allows the large-scale language model to generate, from a test value indicating hypoglycemia, advice indicating the possibility of a disease such as a drug-induced abnormality caused by a blood glucose-lowering agent or the like, or a chronic kidney disease, and advice indicating lifestyle habits that will lead to the improvement of the disease, etc.

[0061] The examination image-related learning data is data that associates text generated by image recognition processing of one or more types of examination images that can be captured by an X-ray device, a computed tomography device (CT device), a magnetic resonance imaging device (MRI device), a positron emission tomography device (PET device), a single photon emission computed tomography device (SPECT device), an ultrasound examination device, etc., with advice indicating a disease predicted based on the examination image and / or a lifestyle habit required for a patient of the examination value to maintain health. The examination image-related learning data preferably includes information on examination images and findings (medical big data) accumulated in large quantities in a medical database. This allows the support device 1 to perform machine learning on a large-scale language model to generate more useful advice.

[0062] Because the examination image-related learning data includes examination images related to X-ray imaging equipment, the large-scale language model can generate advice indicating the possibility of diseases such as bone diseases such as fractures and osteoporosis, dental diseases, lung cancer, pneumonia, tuberculosis, pulmonary lesions such as pleural effusion and pneumothorax, and gastrointestinal diseases, as well as advice indicating lifestyle habits that will lead to the improvement of the diseases.

[0063] Since the examination image-related learning data includes examination images from a CT scanner using X-rays, the large-scale language model can generate advice indicating the possibility of diseases such as lung diseases such as pulmonary alveolar proteinosis, adenocarcinoma, bronchioloalveolar carcinoma, pulmonary edema, pulmonary hemorrhage, and pneumonia, and vascular diseases such as aortic aneurysm and aortic dissection, as well as advice indicating lifestyle habits that will lead to the improvement of the diseases.

[0064] Because the examination image-related learning data includes examination images related to MRI equipment, the large-scale language model can generate advice indicating the possibility of diseases such as brain-related diseases, such as cerebrovascular disease (stroke, brain tumor, etc.), brain malformation, brain trauma, and degenerative brain diseases; spine and spinal cord diseases, such as spine and spinal cord diseases in the cervical, thoracic, and lumbar regions (herniated disc, cervical spondylosis, spinal canal stenosis, spinal tumor, etc.); limb diseases, such as osteosarcoma, metastatic bone tumor, rotator cuff tear (shoulder), femoral head necrosis, anterior cruciate ligament rupture, and bone contusion (knee); diseases of internal organs, such as liver cancer, gallbladder cancer, biliary tract cancer, pancreatic cancer, uterine cancer, and ovarian cancer; and prostate diseases, such as prostate cancer, as well as advice indicating lifestyle habits that will lead to the improvement of such diseases.

[0065] Since the examination image-related learning data includes examination images related to a PET device, the large-scale language model can generate advice indicating the possibility of a disease related to brain function, such as cancer, as well as advice indicating lifestyle habits that can lead to the improvement of the disease.

[0066] Since the test image-related learning data includes test images related to a SPECT device, the large-scale language model can generate advice indicating the possibility of having a disease such as cerebrovascular disease, heart disease, or cancer, as well as advice indicating lifestyle habits that can lead to the improvement of the disease.

[0067] Since the examination image-related learning data includes examination images related to ultrasound examination equipment, the large-scale language model can predict various diseases including heart diseases such as valvular heart disease, cardiac hypertrophy, intracardiac thrombus, pericarditis, and myocardial infarction; liver diseases such as fatty liver, liver cysts, liver hemangiomas, hepatitis, liver cirrhosis, liver cancer, and metastatic liver cancer; diseases related to the gallbladder and bile duct such as gallstones, common bile duct stones, gallbladder polyps, cholecystitis, gallbladder cancer, and bile duct cancer; diseases related to the kidneys and ureters such as kidney stones, ureteral stones, renal cysts, hydronephrosis, kidney cancer, and ureteral cancer; splenic diseases such as splenomegaly, splenic cysts, and metastatic splenic tumors; pancreatitis, pancreatic the like; gastrointestinal diseases such as gastric or duodenal ulcers, intestinal obstruction, enteritis, appendicitis, stomach cancer, colon cancer; bladder diseases such as cystitis, bladder stones, bladder cancer; diseases relating to the uterus and ovaries such as uterine fibroids, uterine cancer, ovarian cysts, ovarian cancer; prostate diseases such as prostate hyperplasia and prostate cancer; diseases relating to the abdominal aorta such as aortic thrombosis, abdominal aortic aneurysm, aortic dissection; and blood vessel diseases such as arteriosclerosis, varicose veins, venous thrombosis. The system can also generate advice indicating the possibility of having a disease such as a lifestyle change that will lead to improvement of the disease.

[0068] The emergency transport destination-related learning data is data that associates the text of a report provided by an ambulance crew member indicating the condition of the transport target (patient) with a disease predicted based on the report and advice indicating a transport destination suitable for treating the disease. By including the emergency transport destination-related learning data in the learning data, the support device 1 can predict various diseases from the report text and generate advice indicating a transport destination suitable for treating the disease.

[0069] The clinical trial-related learning data is data that associates the text of the medical record of a clinical trial subject with advice indicating the number of subjects required for the clinical trial. The data includes, for example, information on whether or not the administered drug is a placebo in order to handle data such as a double-blind method. By including the clinical trial-related learning data in the learning data, the support device 1 can predict the validity of the clinical trial from the text of the medical record and generate advice indicating the number of subjects required for the clinical trial. It is preferable that the clinical trial-related learning data includes information on the medical record related to the clinical trial accumulated in a large amount in a medical database (medical big data). This allows the support device 1 to perform machine learning on a large-scale language model to generate advice including the number of subjects calculated with higher accuracy. In addition, the support device 1 can perform machine learning on a large-scale language model to generate advice including the number of subjects based on the chemical similarity (for example, the relationship between optical isomers) between an existing drug and a drug to be the subject of a clinical trial.

[0070] The machine learning step preferably includes a procedure for executing machine learning at a fixed time of day, etc., on learning data (news learning data) in which text indicating that the news affects the health condition is associated with text related to various news that affects the health condition, such as recent and current weather information, infectious disease infection information, etc. It is known that the risk of disease increases with increasing temperature differences. In addition, the infection status of infectious diseases such as seasonal influenza affects the risk of disease related to the infectious disease. In this way, various news can be important information in analyzing the disease of a patient.

[0071] Therefore, by including a procedure in which the machine learning step executes machine learning related to news learning data at a fixed time of day, etc., the support device 1 can generate and provide advice based on news that affects the health condition. An example of generating such advice is, for example, generating advice indicating that there is a high possibility of Meniere's disease when news that the atmospheric pressure is low is learned by machine learning and blood pressure is low.

[0072] The support process preferably includes a series of processes (steps S1 to S2) for obtaining a medical questionnaire. This allows the support device 1 to generate information to be used in the treatment of the patient based on the medical questionnaire obtained by questioning the patient, among a wide range of information indicating the patient's condition. The medical questionnaire may include information written by the patient himself / herself, information written by a medical professional through questioning the patient, or both of these.

[0073] [Step S1: Determine whether to obtain a medical questionnaire] The control unit 11 executes the acquisition unit 112 in cooperation with the storage unit 12 and the communication unit 13. Then, the control unit 11 executes a process of determining whether to acquire a medical questionnaire by the acquisition unit 112 (step S1, medical questionnaire acquisition determination step). If it is determined that the medical questionnaire should be acquired, the control unit 11 shifts the process to step S2. If it is determined that the medical questionnaire should not be acquired, the control unit 11 shifts the process to step S3.

[0074] [Step S2: Get the text of the questionnaire] The control unit 11 executes a process of acquiring the text of the medical questionnaire determined to be acquired in step S1 by the acquisition unit 112 (step S2, medical questionnaire acquisition execution step). The control unit 11 stores the acquired medical questionnaire text in the memory unit 12, and moves the process to step S3. The medical questionnaire acquisition execution step includes, for example, a procedure of acquiring the medical questionnaire text from a medical database, a procedure of acquiring text obtained by character recognition of a handwritten medical questionnaire, a procedure of acquiring the medical questionnaire text from a terminal T into which a patient or the like has input the medical questionnaire text, and the like.

[0075] The medical questionnaire acquisition execution step preferably includes a procedure of conducting a medical questionnaire for a patient through a dialogue via voice using voice synthesis and voice recognition, and creating a medical questionnaire based on a response from the patient. This allows the support device 1 to provide the patient with a medical questionnaire by a so-called virtual doctor expressed by synthetic voice. At this time, it is preferable that the support device 1 further provides an image (avatar) of a person corresponding to the voice. This allows the support device 1 to provide a more realistic virtual doctor expressed by computer graphics (CG) and synthetic voice, and reassure the patient.

[0076] The support process preferably includes a series of processes for acquiring test values ​​(steps S3 to S5). This allows the support device 1 to generate information to be used in the treatment of the patient based on test values ​​acquired from a wearable device or the like, among a wide range of information indicating the patient's condition.

[0077] [Step S3: Determine whether to obtain the test value] Control unit 11 executes conversion unit 113 in cooperation with memory unit 12 and communication unit 13. Control unit 11 then executes processing to determine whether to acquire a test value from terminal T, which is a wearable device, using conversion unit 113 (step S3, test value acquisition determination step). If it is determined that a test value should be acquired, control unit 11 transfers the processing to step S4. If it is determined that a test value should not be acquired, control unit 11 transfers the processing to step S5. The test value acquisition determination step includes a procedure of determining that a test value should be acquired if it is a predetermined timing, a procedure of determining that a test value should be acquired when transmitted from terminal T, which is a wearable device, and the like.

[0078] [Step S4: Obtain test values] Control unit 11 executes a process of acquiring the test values ​​determined to be acquired in step S3 from terminal T, which is a wearable device, by conversion unit 113 (step S4, test value acquisition execution step). Control unit 11 moves the process to step S5. The test value acquisition execution step includes, for example, a procedure of acquiring one or more of pulse, body temperature, blood pressure, blood glucose level, number of steps walked, and electrocardiogram from terminal T, which is a wearable device.

[0079] The test value acquisition execution step includes a procedure of acquiring one or more test values, such as pulse, body temperature, blood pressure, blood glucose level, number of steps walked, and electrocardiogram, measured by a wearable device from the terminal T, which is the wearable device. This allows the support device 1 to acquire, as test values, daily biological data that is useful information for identifying a disease of the patient from a wearable device that the patient normally wears. This also allows the support device 1 to acquire, as test values, daily biological data that is useful information for identifying a disease of the patient from a wearable device worn by the patient during a medical interview.

[0080] The test value acquisition execution step preferably stores time-series information of the acquired test values ​​in the storage unit 12. This allows the support device 1 to generate advice using the time-series information by inputting the time-series information into a large-scale language model. This also allows the support device 1 to provide the time-series information to medical professionals, medical scientists, etc., and use it as material for disease analysis and research. When providing the time-series information, the support device 1 preferably provides the information with news such as weather information attached. This allows the support device 1 to provide material with higher usability with news attached.

[0081] In the test value acquisition execution step, the acquired test values ​​and / or time series information, etc. are preferably stored in the storage unit 12 in association with the geographical location of the patient. This enables the support device 1 to perform public health information analysis, exemplified by generating an infection status of an infectious disease by region, for example, and provide the analysis results. This enables the support device 1 to support infectious disease countermeasures, etc.

[0082] The test value acquisition execution step preferably stores the acquired test values ​​and / or time-series information, etc. in a cloud storage or cloud database that is a component of the storage unit 12. This allows the support device 1 to store a large amount of test values ​​and / or time-series information that would be difficult to store and provide using only a storage device configured inside the support device 1, and to provide the information as useful information for medical care.

[0083] [Step S5: Convert the inspection value to text] The control unit 11 executes a process of converting the test values ​​acquired in step S4 into a text format by the conversion unit 113 (step S5, conversion step). The control unit 11 stores the converted text of the medical questionnaire (second text) in the memory unit 12, and moves the process to step S6. This enables the support device 1 to input the test values ​​as text into a large-scale language model and generate advice to be used in the treatment of the patient, without using a special artificial intelligence that analyzes the test values.

[0084] [Step S6: Determine whether to acquire an inspection image] Control unit 11 executes image recognition unit 114 in cooperation with memory unit 12 and communication unit 13. Control unit 11 then executes a process of determining whether to acquire an inspection image from the inspection device using conversion unit 113 (step S6, inspection image acquisition determination step). If it is determined that an inspection image should be acquired, control unit 11 transfers the process to step S7. If it is not determined that an inspection image should be acquired, control unit 11 transfers the process to step S9. The inspection image acquisition determination step includes a procedure for determining that an inspection image should be acquired when it is transmitted from the inspection device, and the like.

[0085] [Step S7: Obtain inspection image] The control unit 11 executes a process of acquiring the inspection image determined to be acquired in step S6 from the inspection device by the image recognition unit 114 (step S7, inspection image acquisition execution step). The control unit 11 moves the process to step S8.

[0086] The step of acquiring and executing an examination image includes, for example, a procedure for acquiring an examination image from one or more of various medical devices capable of generating examination images, such as an X-ray device, a computed tomography device (CT device), a magnetic resonance imaging device (MRI device), a positron emission tomography device (PET device), a single photon emission computed tomography device (SPECT device), and an ultrasound examination device.

[0087] Since the test image acquisition execution step includes a procedure for acquiring a test image from an X-ray imaging device, the support device 1 can generate advice indicating the possibility of a disease such as bone diseases such as fractures and osteoporosis, dental diseases, lung cancer, pneumonia, tuberculosis, pulmonary lesions such as pleural effusion and pneumothorax, and gastrointestinal diseases, as well as advice indicating lifestyle habits that will lead to the improvement of the disease, from text obtained by image recognition of the test image obtained by the X-ray imaging device.

[0088] Since the test image acquisition execution step includes a procedure for acquiring a test image from a CT device, the support device 1 can generate advice indicating the possibility of a disease such as lung diseases such as pulmonary alveolar proteinosis, adenocarcinoma, bronchioloalveolar carcinoma, pulmonary edema, pulmonary hemorrhage, pneumonia, etc., or vascular diseases such as aortic aneurysm and aortic dissection, from text obtained by image recognition of the test image obtained by the CT device, and advice indicating lifestyle habits that will lead to the improvement of the disease.

[0089] Since the test image acquisition execution step includes a procedure for acquiring a test image from an MRI device, the support device 1 can generate, from text obtained by image recognition of the test image obtained by the MRI device, advice indicating the possibility of a disease such as brain-related diseases such as cerebrovascular diseases (stroke, brain tumor, etc.), brain malformations, brain trauma, and degenerative brain diseases; spine and spinal cord diseases such as spine and spinal cord diseases in the cervical, thoracic, and lumbar regions (herniated disc, cervical spondylosis, spinal canal stenosis, spinal tumor, etc.); limb-related diseases such as osteosarcoma, metastatic bone tumor, rotator cuff tear (shoulder), femoral head necrosis, anterior cruciate ligament rupture, and bone contusion (knee) in the limbs; diseases of internal organs such as liver cancer, gallbladder cancer, biliary tract cancer, pancreatic cancer, uterine cancer, and ovarian cancer; and prostate-related diseases such as prostate cancer, as well as advice indicating lifestyle habits that will lead to the improvement of the disease.

[0090] Since the test image acquisition execution step includes a procedure for acquiring a test image from a PET device, the support device 1 can generate advice indicating the possibility of a disease related to brain function, such as cancer, and advice indicating lifestyle habits that will lead to the improvement of the disease, from text obtained by image recognition of the test image acquired by the PET device.

[0091] Since the test image acquisition execution step includes a procedure for acquiring a test image from a SPECT device, the support device 1 can generate advice indicating the possibility of having a disease such as cerebrovascular disease, heart disease, cancer, etc., and advice indicating lifestyle habits that can lead to the improvement of the disease, from text obtained by image recognition of the test image acquired by the SPECT device.

[0092] Since the examination image acquisition execution step includes a procedure for acquiring an examination image from an ultrasound examination device, the support device 1 can detect, from text obtained by performing image recognition on the examination image obtained by the ultrasound examination device, various types of diseases, including heart diseases such as valvular heart disease, cardiac hypertrophy, intracardiac thrombus, pericarditis, and myocardial infarction; liver diseases such as fatty liver, liver cyst, liver hemangioma, hepatitis, liver cirrhosis, liver cancer, and metastatic liver cancer; diseases related to the gallbladder and bile duct such as gallstones, common bile duct stones, gallbladder polyps, cholecystitis, gallbladder cancer, and bile duct cancer; diseases related to the kidneys and ureters such as kidney stones, ureteral stones, renal cysts, hydronephrosis, kidney cancer, and ureteral cancer; splenomegaly, splenic the like; spleen diseases such as pancreatitis, pancreatic cysts, pancreatic cancer, etc.; gastrointestinal diseases such as gastric and duodenal ulcers, intestinal obstruction, enteritis, appendicitis, stomach cancer, large intestine cancer, etc.; bladder diseases such as cystitis, bladder stones, bladder cancer, etc.; diseases relating to the uterus and ovaries such as uterine fibroids, uterine cancer, ovarian cysts, ovarian cancer, etc.; prostate diseases such as prostate hyperplasia and prostate cancer, diseases relating to the abdominal aorta such as aortic thrombosis, abdominal aortic aneurysm, aortic dissection, etc.; and blood vessel diseases such as arteriosclerosis, varicose veins, venous thrombosis, etc., as well as advice indicating lifestyle habits that will lead to improvement of the disease.

[0093] [Step S8: Image recognition of inspection image] The control unit 11 executes a process of performing image recognition on the examination image acquired in step S7 using the image recognition unit 114, and generating text (third text) indicating the result of the image recognition (step S8, image recognition step). The control unit 11 stores the third text in the memory unit 12, and moves the process to step S9. This allows the assistance device 1 to generate advice to be utilized in the treatment of patients, with a configuration in which the minimum type of artificial intelligence, namely artificial intelligence performing image recognition, is added to the large-scale language model.

[0094] [Step S9: Determine whether to acquire audio] The control unit 11 executes the voice recognition unit 115 in cooperation with the storage unit 12 and the communication unit 13. Then, the control unit 11 executes a process of determining whether to acquire the patient's voice (patient voice) from the terminal T or the like by the conversion unit 113 (step S9, voice acquisition determination step). If it is determined that the voice is to be acquired, the control unit 11 transfers the process to step S10. If it is not determined that the voice is to be acquired, the control unit 11 transfers the process to step S11. The voice acquisition determination step includes a procedure of determining that the voice is to be acquired when it is determined that the voice is provided from the terminal T used by the patient, and the like.

[0095] [Step S10: Obtaining text generated by speech recognition] The control unit 11 executes a process of acquiring the voice determined to be acquired in step S9 from the terminal T or the like by the voice recognition unit 115, and acquiring text (fourth text) obtained by voice recognition of the voice (step S10, voice acquisition execution step). The control unit 11 stores the fourth text in the storage unit 12, and moves the process to step S11.

[0096] The voice acquisition execution step includes a procedure for acquiring voice from a terminal T that makes a video call with a medical worker, a procedure for acquiring voice from a terminal T that communicates with the support device 1, a procedure for acquiring voice collected by a microphone installed at the medical site, a procedure for acquiring voice from a telephone that makes a voice call with a medical worker, etc.

[0097] Since the voice acquisition execution step includes a procedure for acquiring voice from the terminal T that performs a video call with a medical worker, the support device 1 can generate advice to be utilized in treating a patient based on a conversation when the medical worker video-examines the patient, for example, and provide the advice to the medical worker, etc. In this way, the support device 1 can support the medical worker who video-examines a patient at a distance and reduce the burden on the medical worker.

[0098] Since the voice acquisition execution step includes a procedure for acquiring voice from the terminal T that communicates with the support device 1, the support device 1 can provide advice to be utilized in treating a patient, for example, based on a main complaint of the patient that is spoken by the patient before the patient comes to the hospital. In this way, the support device 1 can support the patient before the patient comes to the hospital and reduce the burden on the patient.

[0099] Since the voice acquisition execution step includes a procedure for acquiring voice collected by a microphone installed at the medical site, the support device 1 can generate advice to be utilized in the treatment of a patient based on the voice of a medical professional performing a medical procedure such as examination, testing, or surgery on a patient, and provide the advice to the medical professional, etc. In this way, the support device 1 can support the medical professional performing the medical procedure and reduce the burden on the medical professional.

[0100] Since the voice acquisition execution step includes a procedure for acquiring voice from a telephone for a voice call with a medical worker, the support device 1 can generate advice to be utilized in the treatment of a patient based on a telephone conversation between the patient and the medical worker before the patient comes to the hospital, and provide the advice to the medical worker, etc. This allows the support device 1 to support the patient before the patient comes to the hospital and reduce the burden on the patient. Also, this allows the support device 1 to support the medical worker who performs medical treatment by providing advice indicating a disease, etc. predicted before the patient comes to the hospital, and reduce the burden on the medical worker.

[0101] [Step S11: Determine whether to obtain a report from an ambulance crew member] The control unit 11 executes the acquisition unit 112 in cooperation with the memory unit 12 and the communication unit 13. Then, the control unit 11 executes a process of determining whether to acquire a report from an ambulance technician via the terminal T or the like by the acquisition unit 112 (step S11, emergency text acquisition determination step). If it is determined that the report should be acquired, the control unit 11 transfers the process to step S12. If it is not determined that the report should be acquired, the control unit 11 transfers the process to step S13. The emergency text acquisition determination step includes a procedure of determining that the report should be acquired when it is determined that the report has been provided from the terminal T used by the ambulance technician, and the like.

[0102] [Step S12: Get emergency text] The control unit 11 executes a process of acquiring the text (emergency text, fifth text) related to the report determined to be acquired in step S12 from the terminal T used by the ambulance crew member by the voice recognition unit 115 (step S12, emergency text acquisition execution step). The control unit 11 stores the fifth text in the storage unit 12 and moves the process to step S11. This allows the support device 1 to acquire the fifth text, which is the text reported by the ambulance crew member.

[0103] The emergency text acquisition execution step includes a procedure for acquiring a fifth text from a terminal T such as a tablet terminal used by an emergency medical technician, and a procedure for acquiring a fifth text obtained by voice recognition of a voice collected by a microphone installed at the emergency medical site.

[0104] The emergency text acquisition execution step allows the support device 1 to generate advice indicating the patient's destination based on the report from the emergency medical personnel, thereby supporting the emergency medical personnel. In this way, the support device 1 can reduce the burden on the emergency medical personnel by supporting the pruning of the destination, which requires specialized knowledge.

[0105] [Step S13: Determine whether to obtain medical records] The control unit 11 executes the acquisition unit 112 in cooperation with the memory unit 12 and the communication unit 13. Then, the control unit 11 executes a process of determining whether to acquire the medical records of the subjects of the clinical trial by the acquisition unit 112 (step S13, medical record acquisition determination step). If it is determined that the medical records should be acquired, the control unit 11 transfers the process to step S14. If it is determined that the medical records should not be acquired, the control unit 11 transfers the process to step S15. The medical record acquisition determination step includes a procedure of determining that the medical records should be acquired when it is determined that all the medical records at a given stage defined in the clinical trial are complete, and the like.

[0106] [Step S14: Get the medical record text] The control unit 11 executes a process of acquiring the text (sixth text) of the medical record determined to be acquired in step S13 by the acquisition unit 112 (step S14, medical record acquisition execution step). The control unit 11 stores the sixth text in the storage unit 12 and moves the process to step S14. This allows the support device 1 to acquire the sixth text related to the medical record of the subject of the clinical trial.

[0107] The medical record acquisition execution step includes, for example, a procedure for acquiring the medical record text from a medical database, a procedure for acquiring text obtained by character recognition of a handwritten medical record, and a procedure for acquiring the medical record text from a terminal T where a medical professional has entered the medical record text.

[0108] [Step S15: Determine whether to generate advice] The control unit 11 executes the acquisition unit 112 in cooperation with the storage unit 12 and the communication unit 13. Then, the control unit 11 executes a process of acquiring texts (first texts) indicating the patient's condition, such as the text of the medical questionnaire in step S2, the second text in step S5, the third text in step S8, the fourth text in step S10, the fifth text in step S12, and the sixth text in step S14, by the acquisition unit 112, and determining whether to generate advice to be utilized in the treatment of the patient related to the text (step S15, generation determination step). If it is determined that advice is to be generated, the control unit 11 acquires texts (first texts) indicating the patient's condition, such as the text of the medical questionnaire in step S2, the second text in step S5, the third text in step S8, the fourth text in step S10, the fifth text in step S12, and the sixth text in step S14, by the acquisition unit 112 from the storage unit 12, and moves the process to step S16. If it is not determined that the image is to be generated, the control unit 11 shifts the process to step S1, and repeats the processes from step S1 to step S17.

[0109] [Step S16: Generate advice] The control unit 11 executes the generation unit 116 in cooperation with the storage unit 12 and the communication unit 13. Then, the control unit 11 executes a process in which the generation unit 116 inputs the text acquired in step S15 into a large-scale language model and causes the large-scale language model to generate advice to be utilized in treating a patient related to the text (step S16, generation step). The control unit 11 stores the generated advice in the storage unit 12 and moves the process to step S17.

[0110] If the text obtained in step S15 includes the text of the medical questionnaire of step S2, it is preferable that the generation step includes a procedure of inputting the text of the medical questionnaire, accompanied by a prompt to generate advice to be used in the treatment of the patient based on the medical questionnaire, into a large-scale language model, and causing the large-scale language model to generate advice to be used in the treatment of the patient related to the medical questionnaire text.

[0111] If the text obtained in step S15 includes the second text of step S5, the generation step preferably includes a procedure of inputting the second text, accompanied by a prompt instructing the large-scale language model to generate advice indicating the patient's disease based on the test values, into a large-scale language model, and causing the large-scale language model to generate advice indicating the disease predicted from the test values.

[0112] If the text obtained in step S15 includes the third text of step S8, the generation step preferably includes a procedure of inputting the third text, accompanied by a prompt to generate advice indicating the patient's disease based on the image recognition results of the examination image, into a large-scale language model, and causing the large-scale language model to generate advice indicating the disease suspected from the examination image.

[0113] If the text obtained in step S15 includes the fourth text of step S10, the generation step preferably includes a procedure of inputting the fourth text, accompanied by a prompt instructing the large-scale language model to generate advice indicating the patient's illness based on the utterances of the patient and / or medical professional, into a large-scale language model, and causing the large-scale language model to generate advice indicating the illness assumed from the fourth text.

[0114] If the text obtained in step S15 includes the fifth text of step S12, the generation step preferably includes a procedure of inputting the fifth text, accompanied by a prompt instructing the large-scale language model to generate advice indicating a patient's transport destination based on a report from the ambulance personnel, into a large-scale language model, and causing the large-scale language model to generate advice indicating a transport destination appropriate for the patient's illness (the illness related to transport) assumed from the report related to the fifth text.

[0115] If the text obtained in step S15 includes the sixth text of step S14, the generation step preferably includes a procedure of inputting the sixth text, accompanied by a prompt to generate advice indicating the number of subjects required for the clinical trial based on the medical records of the clinical trial subjects, into a large-scale language model, and causing the large-scale language model to generate advice indicating the number of subjects required for the clinical trial based on the sixth text.

[0116] The text acquired in step S15 preferably includes a plurality of types of text. In this way, the generating step can input the plurality of types of text into a large-scale language model and generate advice based on co-occurrences between the plurality of types of text. The generation of advice based on co-occurrences refers to, for example, generating advice indicating the possibility of an infectious disease that causes fever, which is not inconsistent with the test value, among diseases assumed from the chief complaint of "febrile" in the medical questionnaire and the test value text, based on the test value obtained from the wearable device and indicating a possible infectious disease. The combination of types of text that cause co-occurrences is not limited to the combination of the text of the medical questionnaire and the test value text, and may be, for example, a combination of the text of the medical questionnaire and the test image text that can narrow down diseases that are not inconsistent with both the medical questionnaire and the test image, a combination of the text of the test value and the test image text that can narrow down diseases that are not inconsistent with both the test value and the test image, a combination of the text of the medical questionnaire and the report text that can narrow down diseases that are not inconsistent with both the medical questionnaire and the report from the ambulance staff, and the like.

[0117] The large-scale language model generates text by probabilistically predicting the next word for a given text, repeatedly outputting the word with the highest probability, and connecting these words. Therefore, by inputting multiple types of text to the large-scale language model, the large-scale language model can determine the word with the highest probability taking into account each of the multiple types of text. This allows the large-scale language model to generate advice that takes into account co-occurrences between multiple types of text. Therefore, the assistance device 1 can generate more useful advice that takes into account co-occurrences and provide it to medical professionals and / or patients.

[0118] As described above, the support device 1 of the present embodiment can obtain various information indicating the patient's condition as text, input the obtained text to a large-scale language model, and have the large-scale language model generate appropriate advice to be used in the treatment of the patient. Therefore, the support device 1 can process the text of a wide range of information indicating the patient's condition using the minimum type of artificial intelligence, the large-scale language model, and generate information to be used in the treatment of the patient.

[0119] [Step S17: Providing advice] The control unit 11 executes the speech unit 117 in cooperation with the storage unit 12 and the communication unit 13. Then, the control unit 11 executes a process of providing the advice generated in step S16 to the terminal T or the like by the acquisition unit 112 (step S17, speech step). The control unit 11 shifts the process to step S1 and repeats the processes from step S1 to step S17. In the speech step, the advice may be provided to the terminal T or the like in the form of text, or a voice synthesized from the advice using a voice synthesis neural network or the like may be provided to the terminal T or the like. By providing the advice in the form of text, the support device 1 can provide the advice while suppressing the consumption of the bandwidth of the network N. By providing the advice in the form of voice, the support device 1 can provide the advice to a patient with a medical condition that makes it difficult to use a device that displays text, such as the terminal T.

[0120] <Usage example> The following is an example of how the support device 1 of this embodiment is used.

[0121] [Advice on the medical questionnaire] A medical professional interviews a patient and provides the text of the interview to the assistance device 1 via terminal T. The assistance device 1 receives the text of the interview (text of the medical questionnaire). The assistance device 1 inputs the text of the medical questionnaire together with a prompt into a large-scale language model, and causes the large-scale language model to generate advice to be utilized in treating the patient in the condition indicated by the medical questionnaire. The assistance device 1 provides the generated advice to a medical professional or the like to help with examinations, consideration of necessary tests, and the like.

[0122] [Advice on test results] A patient wears a wearable device such as a smart watch, and provides test values ​​indicating the patient's biological data measured by the wearable device to the assistance device 1. The assistance device 1 receives the test values. The assistance device 1 inputs the text of the test values ​​together with a prompt into a large-scale language model, and causes the large-scale language model to generate advice indicating the disease of the patient whose condition is indicated by the test values. The assistance device 1 provides the generated advice to medical professionals, etc., to assist with medical examinations, consideration of necessary tests, etc.

[0123] [Advice on examination images] A medical professional performs an examination on a patient, and provides the examination image obtained by the examination to the assistance device 1 via the terminal T. The assistance device 1 receives the examination image. The assistance device 1 inputs text obtained by image recognition of the examination image together with a prompt into a large-scale language model, and causes the large-scale language model to generate advice indicating the illness of the patient in the state shown by the examination image. The assistance device 1 provides the generated advice to the medical professional, etc., to assist in the examination, consideration of necessary examinations, etc.

[0124] [Advice for video consultations] A medical professional makes a video call with a patient living far away, and provides audio data, such as the audio and / or video of the video call, to the assistance device 1 via terminal T. The assistance device 1 receives the audio data. The assistance device 1 inputs text obtained by speech recognition of the audio data into a large-scale language model together with a prompt, and causes the large-scale language model to generate advice indicating the illness of the patient in the state indicated by the audio data. The assistance device 1 provides the generated advice to a medical professional, etc., to assist with medical examinations, consideration of necessary tests, etc.

[0125] [Emergency medical advice] The ambulance staff provides a report on the patient to the support device 1 via the terminal T. The support device 1 receives the report. The support device 1 inputs the text of the report together with a prompt into the large-scale language model, and causes the large-scale language model to generate advice indicating an appropriate transport destination according to the illness of the patient related to the report, first aid according to the illness, etc. The support device 1 provides the generated advice to the ambulance staff, etc., to help them in deciding on a transport destination, implementing first aid, etc. The support device 1 can also input information related to the patient's medical history, such as the patient's medical questionnaire and chart, into the large-scale language model together with a prompt in addition to the text of the report, and cause the large-scale language model to generate advice indicating an appropriate transport destination according to the illness of the patient related to the report, first aid according to the illness, etc.

[0126] [Advice for clinical trials] A medical professional involved in a clinical trial provides a medical record relating to a subject to the support device 1 via a terminal T. The support device 1 receives the medical record. The support device 1 inputs the text of the medical record together with a prompt into a large-scale language model, and causes the large-scale language model to generate advice indicating the number of subjects considered based on the medical record. The support device 1 provides the generated advice to a medical professional involved in a clinical trial to help with the consideration of the number of subjects, etc.

[0127] It should be noted that within the scope of the concept of the present invention, a person skilled in the art may come up with various modifications and alterations. It is understood that such modifications and alterations fall within the scope of the present invention. For example, even if a person skilled in the art appropriately adds, deletes, or modifies the design of the above-described embodiment, or adds, omits, or modifies the conditions of a process, these modifications and alterations are included in the scope of the present invention as long as they include the gist of the present invention. [Explanation of symbols]

[0128] S System 1 Support equipment 11 Control section 111 Machine Learning Department 112 Acquisition Department 113 Conversion unit 114 Image Recognition Unit 115 Voice Recognition Unit 116 Generation part 117 Speech section 12 Storage section 13. Communications Department N Network T-Terminal

Claims

1. a machine learning unit that performs machine learning related to the generation of advice using a combination of text indicating a patient's condition and advice to be used in treating the patient as training data for a large-scale language model that has been pre-trained using 500 GB or more of text; an acquisition unit that acquires a text (first text) indicating a patient's condition; a generator for inputting the first text into the large-scale language model and causing the large-scale language model to generate advice to be utilized in treating the patient; A medical support device comprising:

2. A conversion unit that converts the test value acquired from the wearable device worn by the patient into text (second text), The machine learning unit performs machine learning using a combination of text indicating a test value and advice indicating a disease predicted from the test value as learning data; The acquisition unit acquires the second text, The generation unit inputs the second text and generates advice indicating a disease suspected from the test value. The support device according to claim 1 .

3. An image recognition unit that recognizes an examination image related to the patient, The image recognition unit generates a third text indicating a result of the image recognition; the machine learning unit performs machine learning using a combination of text indicating a result of image recognition for the examination image and advice indicating a disease predicted from the examination image as learning data; The acquisition unit acquires the third text, The generation unit receives the third text and generates advice indicating a disease suspected from the examination image. The support device according to claim 1 .

4. a speech unit that speaks the advice by speech synthesis, The acquisition unit acquires a fourth text obtained by performing speech recognition on a patient voice uttered by the patient, The generation unit inputs the fourth text. The support device according to claim 1 .

5. The machine learning unit performs machine learning using a combination of a report from an ambulance crew member and advice indicating a transportation destination suitable for dealing with an illness assumed based on the report as learning data, The acquisition unit acquires a text (fifth text) reported by an ambulance crew member transporting the patient, The generation unit inputs the fifth text and generates advice indicating a destination suitable for the illness related to the transport. The support device according to claim 1 .

6. The machine learning unit performs machine learning using a combination of the medical records of subjects of a clinical trial and the number of subjects required for the clinical trial as learning data, The acquisition unit acquires texts (sixth texts) of the plurality of medical records, The generation unit inputs the sixth text and generates advice indicating the number of subjects. generating a recommendation indicating the number of subjects required for the clinical trial associated with the medical record; The support device according to claim 1 .

Images