Method and system for providing prescription information based on language models

Abstract

The method for providing prescription information based on a language model according to the present invention may include the steps of: receiving a request for a recommended exercise curriculum from a medical staff terminal logged in with a medical staff account; acquiring the exercise curriculum in response to the request; and providing the acquired exercise curriculum to the medical staff terminal.

Description

Technical Field

[0001] The present invention relates to a method and system for providing prescription information using a language model.

Background Art

[0002] Musculoskeletal diseases refer to pain or injuries occurring in the musculoskeletal system such as muscles, nerves, tendons, ligaments, bones, and surrounding tissues. Musculoskeletal diseases occur in various parts of the body such as the neck, waist, arms, and legs.

[0003] According to a report by the World Health Organization (WHO), the economic loss due to musculoskeletal diseases is shown to be the fourth highest among all diseases, and musculoskeletal diseases are chronic pain that affects not only daily life but also economic activities.

[0004] In principle, the treatment of musculoskeletal diseases should start with less invasive treatments, and conservative treatments using non-drug therapies (such as exercise therapy, education, cognitive therapy, or relaxation therapy) should be first implemented, and then drug therapy and surgical treatment should be considered sequentially.

[0005] Treatment guidelines actively recommend conservative treatment using non-drug therapies for musculoskeletal diseases, and research on methods for implementing conservative treatment using non-drug therapies for musculoskeletal diseases is actively being conducted mainly in the United States and Europe.

[0006] On the other hand, along with the rapid development of artificial intelligence (AI) technology in recent years, language models (such as ChatGPT) that can have a natural conversation with people have emerged.

[0007] Unlike conventional chatbots (Chatbot) that were manually constructed and only made limited responses, language models have the technical ability to communicate naturally like humans and provide rapid and accurate information, bringing innovation to the artificial intelligence market.

[0008] In line with this trend, there is a growing need to use language models to provide services in industries that require specialized knowledge, such as the medical industry.

[0009] Even in industries that rely on offline services, providing medical services online has become commonplace. [Overview of the Initiative] [Problems that the invention aims to solve]

[0010] This invention relates to a method and system for providing prescription information based on a language model that can search for and recommend appropriate prescription information to patients.

[0011] In particular, the present invention relates to a method and system for providing prescription information based on a language model that can retrieve and recommend prescription information based on a large-scale language model.

[0012] More specifically, the present invention relates to a method and system for providing prescription information based on a language model, which can be trained to retrieve and recommend appropriate prescription information based on the patient's condition and the physician's requirements. [Means for solving the problem]

[0013] To solve the above problems, the prescription information provision method based on the language model according to the present invention may include the steps of: receiving a prescription information recommendation request from a medical staff terminal logged in with a medical staff account; in response to the request, extracting medical staff metadata of the medical staff linked to the medical staff account and patient metadata of the patient to be treated linked to the medical staff account from a server; generating a prompt in a pre-set format using the medical staff metadata and the patient metadata; acquiring the prescription information using a large language model (LLM) that receives the prompt as input; and providing the acquired prescription information to the medical staff terminal.

[0014] Furthermore, the present invention further includes the step of transmitting the prescription information to at least one medical system in cooperation with the server in response to an approval event occurring in the medical staff account, wherein the at least one medical system may include any one of the following: an electronic health record (EHR) system, an electronic medical record (EMR) system, an ordering communication system (OCS), a clinical information system (CIS), and a personal health record (PHR) system for patients.

[0015] Furthermore, the prescription information recommendation request is received on the medical staff terminal with the page of the at least one medical system provided, and in the step of transmitting the prescription information, the prescription information may be output to the page of the at least one medical system.

[0016] Furthermore, the pre-configured prompts may include system prompts and user prompts, wherein the system prompts define the role of the server as the medical staff, and the user prompts may be configured to identify the patient to whom the prescription information is recommended and the indications for the patient.

[0017] Furthermore, the server's exercise database has exercise items for treating each indication associated with it as a group, and each of the exercise items for treating the indication has exercise difficulty level information associated with it. In the step of obtaining the prescription information, the exercise items for treating the indication may be searched from the exercise database, and the searched exercise items may be assigned to each of the different days that constitute a pre-set exercise period to prescribe an exercise program.

[0018] Furthermore, the search for the exercise items may be carried out considering at least one of the following: prescription information or prescription history from the medical staff, the patient's age, the patient's gender, and the patient's exercise preference.

[0019] Furthermore, the present invention further includes the step of obtaining a different exercise program using the prompt based on receiving a different prescription information recommendation request from the medical staff terminal, wherein in the step of obtaining the different exercise program, one of the exercise items may be excluded from the exercise program, and other exercise items that are associated with the same group as the excluded exercise item and have the same difficulty level information associated with them may be included in the exercise program.

[0020] Furthermore, the present invention further includes the steps of verifying whether the prescription information satisfies pre-set verification criteria, and if the verification results in the prescription information not satisfying the pre-set verification criteria, regenerating the prescription information based on the prompt and the language model, wherein the pre-set verification criteria may relate to at least one of the following: the degree of relevance of the patient's indication to the treatment, the difficulty level of the exercise items included in the prescription information, the patient's age, and the patient's exercise inclination.

[0021] Furthermore, the medical staff metadata may include information on at least one of the medical staff's specialty, hospital name, career history, and past prescription history, and the patient metadata may include information on at least one of the patient's age, sex, indications, surgical history, diagnosis, pain site, pain score, past exercise prescription records, and temperament.

[0022] On the other hand, the prescription information provision system based on the language model according to the present invention includes a communication unit that receives prescription information recommendation requests from a medical staff terminal logged in with a medical staff account, and a control unit that, in response to the request, extracts medical staff metadata of the medical staff linked to the medical staff account and patient metadata of the patient being treated linked to the medical staff account from a server. The control unit may use the medical staff metadata and the patient metadata to generate a prompt in a pre-set format, acquire the prescription information using a large language model (LLM) that receives the prompt as input, and provide the acquired prescription information to the medical staff terminal.

[0023] On the other hand, the program according to the present invention is a program that is executed in an electronic device according to one or more processes and stored on a computer-readable recording medium, and the program may include a command that causes the program to execute the following steps: receiving a prescription information recommendation request from a medical staff terminal logged in with a medical staff account; in response to the request, extracting medical staff metadata of a medical staff linked to the medical staff account and patient metadata of a patient linked to the medical staff account from a server; generating a prompt in a pre-set format using the medical staff metadata and the patient metadata; acquiring the prescription information using a large language model (LLM) that receives the prompt as input; and providing the acquired prescription information to the medical staff terminal. [Effects of the Invention]

[0024] As described above, the prescription information recommendation method and system based on the language model according to the present invention receives a prescription information recommendation request from a medical staff terminal logged in with a medical staff account, and in response to the request, extracts medical staff metadata of the medical staff linked to the medical staff account and patient metadata of the patient being treated linked to the medical staff account from the server, and generates a prompt in a pre-set format using the medical staff metadata and patient metadata. In this way, the present invention enables the large-scale language model to recommend appropriate prescription information from the perspective of the medical staff, taking into account the patient's condition.

[0025] Furthermore, the prescription information recommendation method and system based on the language model according to the present invention obtain the prescription information using a large language model (LLM) that receives the prompt as an input, and provide the obtained prescription information to the medical staff terminal. By inputting patient information and the requirements of medical staff into the large language model, appropriate prescription information is recommended, and the recommended prescription information can be reflected in the medical system.

Brief Description of Drawings

[0026] [Figure 1] It is a conceptual diagram for explaining the prescription information providing environment based on the language model according to the present invention. [Figure 2] It is a conceptual diagram for explaining the prescription information providing system based on the language model according to the present invention. [Figure 3] It is a conceptual diagram for explaining the large language model used in the present invention. [Figure 4] It is a conceptual diagram for explaining the large language model used in the present invention. [Figure 5a] It is a flowchart for explaining the prescription information providing method based on the language model according to the present invention. [Figure 5b] It is a flowchart for explaining the prescription information providing method based on the language model according to the present invention. [Figure 6] It is a conceptual diagram for explaining the prescription information providing method based on the language model provided by the present invention. [Figure 7a] It is a conceptual diagram for explaining the prescription information providing method based on the language model provided by the present invention. [Figure 7b] It is a conceptual diagram for explaining the prescription information providing method based on the language model provided by the present invention. [Figure 7c] It is a conceptual diagram for explaining the prescription information providing method based on the language model provided by the present invention. [Figure 7d] It is a conceptual diagram for explaining the prescription information providing method based on the language model provided by the present invention. [Figure 7e] This is a conceptual diagram illustrating the method for providing prescription information based on the language model provided in the present invention. [Figure 8] This is a conceptual diagram illustrating the method for providing prescription information based on the language model provided in the present invention. [Figure 9] This is a conceptual diagram illustrating the method for providing prescription information based on the language model provided in the present invention. [Figure 10] This is a conceptual diagram illustrating the method for providing prescription information based on the language model provided in the present invention. [Modes for carrying out the invention]

[0027] The embodiments disclosed herein will be described in detail below with reference to the accompanying drawings, but regardless of the reference numerals used in the drawings, identical or similar components will be given the same reference numerals, and redundant descriptions thereof will be omitted. The suffixes “module” and “part” used for components in the following description are added or used interchangeably for the sake of ease of writing the specification and do not have any distinguishing meaning or role in themselves. Furthermore, when describing the embodiments disclosed herein, if it is determined that a detailed description of the relevant prior art would obscure the gist of the embodiments disclosed herein, such detailed description will be omitted. In addition, the accompanying drawings are intended solely to facilitate the understanding of the embodiments disclosed herein, and the technical ideas disclosed herein should not be limited by the accompanying drawings and should be understood to include all modifications, equivalents and substitutions that fall within the concept and technical scope of the present invention.

[0028] Terms including ordinal numbers such as "1st," "2nd," etc., may be used to describe various components, but the components are not limited to those defined by these terms. These terms are used solely to distinguish one component from another.

[0029] When it is stated that one component is “connected” or “linked” to another component, it should be understood that it may be directly connected or linked to the other component, but there may also be another component between them. On the other hand, when it is stated that one component is “directly connected” or “directly linked” to another component, it should be understood that there is no other component between them.

[0030] A singular expression includes plural forms unless otherwise clearly indicated in the context.

[0031] In this application, terms such as “includes” or “having” are intended to specify the presence of features, figures, steps, actions, components, parts, or combinations thereof as described in the specification, and should be understood not to preemptively exclude the possibility of the presence or addition of one or more other features, figures, steps, actions, components, parts, or combinations thereof.

[0032] The memory unit 120 may be configured to include at least one of the following: patient DB 210, medical staff DB 220, exercise DB 230, and prescription DB 240. In Figure 2, the exercise DB 230 and prescription DB 240 are shown independently (or separately), but the exercise DB 230 and prescription DB 240 may exist as a grouped and linked DB. For example, prescription DB 240 may be configured to include exercise DB 230 (i.e., exercise DB exists within prescription DB). Also, as shown in Figure 2, it is possible for exercise DB 230 and prescription DB 240 to exist separately.

[0033] The patient database 210 may store and contain patient information (patient user information or medical information), which may include, for example, the patient's name, age, sex, surgical history, disease (present illness, past illness, etc.), indications, diagnosis, pain location, medical history, diagnosis results, treatment process, test information (e.g., X-ray images, blood test information, etc.), details of prescribed exercise curriculum, details of prescribed medications, patient's personality, allergy information, patient's lifestyle, eating habits, exercise habits, whether or not they drink alcohol and smoke, and stress level, or at least one of these.

[0034] Medical Staff DB220 may contain medical staff information (user information or medical procedure information of medical staff). For example, Medical Staff DB220 may include at least one of the following: medical staff name, age, work history, specialty, hospital of employment, and research field.

[0035] The exercise database 230 may contain multiple exercise items, the difficulty level of each exercise item, explanatory information, exercise videos, and at least some of the exercise methods, all of which are associated with one another. The exercise database 230 may also contain multiple exercise items grouped together based on indications, and each of the multiple exercise items may have associated explanatory information such as its name, exercise method, and exercise effects.

[0036] The prescription database 240 may contain a prescription history for at least one patient and one medical staff member. The prescription history may include at least one of the following: prescription date, prescribing physician, prescribing patient, prescribed medication or exercise, diagnosis, treatment process, and prescription details.

[0037] In this invention, the patient DB210, medical staff DB220, exercise DB230, and prescription DB240 can be described as database 200 or storage unit.

[0038] The communication unit 130 may be configured to transmit or receive data. The communication unit 130 can receive service requests, including user queries, from the medical staff's user terminal 20 and transmit prescription information corresponding to the user requests. The communication unit 130 may also be configured to receive information necessary for the present invention from an external server.

[0039] The communication unit 130 can support various communication methods depending on the communication standard of the device (or electronic device) performing the communication. For example, the communication unit 130 supports WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), and Bluetooth. TM The system may be configured to communicate using at least one of the following technologies: (Registered Trademark) RFID (Radio Frequency Identification), Infrared Data Association (IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wi-Fi Direct, or Wireless USB (Wireless Universal Serial Bus).

[0040] On the other hand, the control unit 140 may be configured to control the overall operation of the prescription information provision system 100 according to the present invention. Specifically, the control unit 140 can process signals, data, information, etc. that are input or output through the above-mentioned components, or provide or process appropriate information and functions to the user.

[0041] As shown in Figure 3, the control unit 140 can take information 310 extracted from the database 200 (e.g., patient's age, sex, diagnosis, surgical history, pain location, pain score, past exercises, etc.) and a service request 320 from the medical staff (e.g., "I would like a 4-week curriculum of low-difficulty exercises") as input to the Large Language Model (LLM) 142, and obtain an exercise curriculum 330 corresponding to the medical staff's request as a response from the Large Language Model 142.

[0042] More specifically, the control unit 140 can identify the intentions (requirements) of the medical staff from the user queries included in the service request and extract metadata of the patient or medical staff from at least one of the patient DB 210 and the medical staff DB 220. Furthermore, the control unit 140 can generate prompts using the extracted metadata and search for and recommend prescription information using the large-scale language model 142 that receives the prompts as input.

[0043] Furthermore, the control unit 140 can perform a series of data processing operations to search for and recommend prescription information that corresponds to the service requests of medical staff.

[0044] The control unit 140 analyzes the intentions (or requirements) of the medical staff from the service request and can analyze (understand or grasp) the intentions of the medical staff when they requested the service.

[0045] The control unit 140 can utilize at least one of various artificial intelligence technologies to analyze the intent behind a service request. For example, the artificial intelligence technologies may include at least one of the following: Machine Reading Comprehension (MRC), Paraphrasing, Abstract Summarization, Text Generation (TG), Question Generation (QG), Natural Language Understanding (NLU), OCR-NLP (Textanding), and Natural Language-Based Querying (NL2SQL).

[0046] The prompt generation unit 141 may be configured to design (or construct or generate) appropriate prompts to obtain desired results from a language model, and to perform "prompt engineering" techniques that utilize a very large-scale language model to perform natural language processing tasks.

[0047] Here, a prompt is an input value used to generate a response (or answer or result) from a language model. The prompt contains instructions or commands for the language model, which then generates a response based on these instructions.

[0048] The prompt generation unit 141 understands (analyzes) the requirements of medical staff from user queries included in service requests of medical staff through natural language processing, and can extract information necessary for prompt generation from a database (at least one of the patient DB, medical staff DB, exercise DB, and prescription DB) based on the analysis results.

[0049] The prompt generation unit 141 can generate prompts from the language model using service requests (user queries), analysis results, and extracted information to obtain prescription information corresponding to the requests of medical staff as a response.

[0050] In the present invention, the prompt may include at least one of a system prompt 141a and a user prompt 141b.

[0051] In this invention, "system prompt" is an instruction or command word for defining (assigning) the role of the prescription information provision system 100, defining the persona of the medical staff who are the prescribing entity, and commanding the language model to recommend prescription information based on the defined persona (for example, ">[INST] < <sys>It may be configured as follows: "We would like a rehabilitation specialist or orthopedic surgeon to take on the role of recommending exercise for the patient's rehabilitation." In other words, the system prompt may be configured to define the server's role as the aforementioned medical staff.

[0052] Furthermore, the "user prompt" is information that defines the target of prescription information recommendation (or prescription) by the prescription information provision system 100 (for example, "<< / sys> This may include: "A 34-year-old male with an ankle pain score of approximately 6, diagnosed with a simple ankle sprain." In other words, the user prompt may be configured to identify the patient to whom the prescribing information is recommended (or prescribed) and the indications for the patient.

[0053] This invention provides medical services based on a Large Language Model (LLM), enabling medical staff to recommend various prescription information (e.g., pharmaceuticals, medical devices, medical tests, treatments, exercise curricula, exercise programs, exercise plans, cognitive therapy, dietary therapy, counseling, etc.) to provide appropriate prescriptions to patients.

[0054] The present invention may be configured to be pluggable into medical systems using electronic health records (EHRs) and electronic medical records (EMRs).

[0055] Furthermore, the present invention can provide a service that uses at least one of electronic health records and electronic medical records to search for prescription targets for treating a patient's indications and recommend prescription information including the searched information to medical staff.

[0056] For the sake of explanation, this invention will describe, as an example, the recommendation of prescription information to medical staff for patients requiring exercise therapy. Musculoskeletal disorders are examples of indications for patients requiring exercise therapy. In this invention, based on a language model, exercise items for treating musculoskeletal disorders can be searched, and an exercise curriculum (or exercise program or exercise plan) including the searched exercise items can be recommended to medical staff as a prescription target.

[0057] In the present invention, an exercise item means an exercise action or type of exercise, and an exercise curriculum may include at least one exercise item and the method of performing the exercise item (for example, an exercise schedule, exercise sequence, number of repetitions, exercise duration, etc.).

[0058] Furthermore, the "prescription information provision system based on a language model" according to the present invention may be referred to as the "prescription information provision system," and the system may be referred to as the "server" or "platform."

[0059] The following describes in more detail the prescription information provision method and system based on the language model according to the present invention, along with the attached drawings. Figure 1 is a conceptual diagram illustrating the exercise curriculum provision environment based on the language model according to the present invention. Figure 2 is a conceptual diagram illustrating the prescription information provision system based on the language model according to the present invention. Figures 3 and 4 are conceptual diagrams illustrating the large-scale language model used in the present invention. Figures 5a and 5b are flowcharts illustrating the prescription information provision method based on the language model according to the present invention. Figures 6, 7a, 7b, 7c, 7d, 7e, 8, 9, and 10 are conceptual diagrams illustrating the prescription information provision method based on the language model provided in the present invention.

[0060] In this invention, a large-scale language model can be used to recommend prescription information in response to a medical service request. For example, as shown in Figure 1, the present invention can receive a medical service request 1 from a medical staff member regarding the prescription of an exercise curriculum (e.g., "Please recommend a 3-week exercise curriculum"). In this invention, a large-scale language model can be used to recommend an exercise curriculum 2 suitable for prescription to the patient in response to the medical service request. As another example, the present invention can receive a prescription medical service request from a medical staff member regarding the prescription of a drug (e.g., "Please prescribe a drug to treat the cold symptoms of a 1-year-old girl"), and in response to the request, a large-scale language model can be used to recommend prescription information including a drug to be prescribed to the patient.

[0061] Medical staff can request prescription recommendations by entering the minimum necessary information for a prescription. In this invention, various information for recommending prescriptions in response to a service request is extracted from a medical information server, prompts are generated using the extracted information to be input into a large-scale language model, and prescription information can be searched and recommended using the prompts as input to a pre-trained language model.

[0062] Here, the medical information server may include various medical information necessary for medical procedures or generated by the medical institution, such as electronic health records (EHRs) and electronic medical records (EMRs). Furthermore, based on such medical information, the medical information server can provide medical services such as diagnosing, prescribing, treating, and managing the patient's health condition.

[0063] The present invention can provide services in conjunction with a program (software, etc.) of the prescription information provision system 100 or other medical services. When collaborating with other medical services, the present invention can provide services by recommending prescription information via communication with other medical services, or by providing additional (or extended) functions of the prescription information provision system 100 to other medical services (for example, as a plug-in).

[0064] Such medical services may be provided by a healthcare system. The at least one healthcare system may include any one of the following: an electronic health record (EHR) system, an electronic medical record (EMR) system, an ordering communication system (OCS), a clinical information system (CIS), and a personal health record (PHR) system for patients.

[0065] For example, as shown in Figure 1, the present invention allows for the provision of an icon 3 (or GUI: Graphical User Interface) linked to the prescription information provision system 100 while other medical services are being provided on the medical staff's user terminal. The icon 3 may be provided to the user terminal so as to overlap with a screen 10 (for example, a "prescription search screen") provided by the other medical service. In the present invention, in response to the selection of the icon 3, a page 4 (or screen) linked to the prescription information provision system 100 can be provided to the medical staff terminal. In this case, the present invention allows for the overlaying of page 4 of the prescription information provision system 100 onto a region of the screen 10 of the other medical service, and for the reception of a service request 1 requesting prescription information through page 4.

[0066] In this invention, a "service request (or query)" includes the requirements of medical staff, and through analysis of the service request, it is possible to identify which patients require which prescriptions. In Figure 1, "I would like a 3-week exercise curriculum recommended" includes the type of prescription (exercise curriculum) and the duration of the prescription (or the duration of the 3-week exercise curriculum), and although not shown in the figure, "I would like an exercise curriculum recommended for a patient who dislikes repeating the same movements" includes the patient's dislike (dislike of repeating the same movements). In this invention, a large-scale language model can be used to search for, recommend, and provide prescription information that conforms to the requirements of medical staff.

[0067] On the other hand, as shown in Figure 2, the prescription information provision system 100 according to the present invention may be configured to include at least one of the following: an input unit 110, a storage unit 120, a communication unit 130, and a control unit 140 that performs a series of controls related to the provision of prescription information based on a large-scale language model (LLM) 142.

[0068] The input unit 110 may be configured to receive service requests (prescription information requests or queries) from medical staff U. In this invention, service requests may be entered in various formats such as text, voice, video, and images, and the input unit 110 can perform its functions by utilizing the hardware configuration of the medical staff U's user terminal 20. For example, user requests may be entered into the input unit 110 via at least one of the keyboard, mouse, touchscreen, or microphone of the user terminal 20.

[0069] In the present invention, the user terminal is also called an electronic device and may include mobile phones, smartphones, laptop computers, digital broadcasting terminals, PDAs (personal digital assistants), PMPs (portable multimedia players), navigation systems, slate PCs, tablet PCs, ultrabooks, wearable devices (such as smartwatches, smart glasses, and HMDs (head-mounted displays)).

[0070] Medical staff member U can use the prescription information recommendation service based on the large-scale language model provided in this invention using the user terminal 20 that they possess. Medical staff member U can prescribe the prescription information recommended in this invention to the patient, and the patient can perform exercise therapy based on the prescription information prescribed via the user terminal that they possess.

[0071] In this case, user terminal 20 can be understood as a terminal to which medical staff U's user account is logged, or a terminal to which a patient's user account is logged. Therefore, in this invention, "medical staff" and "medical staff account," and "patient" and "patient account" may be used interchangeably. Furthermore, in this invention, a medical staff user terminal (or a user terminal to which a medical staff account is logged) may be referred to as a "medical staff terminal," and a patient user terminal (or a user terminal to which a patient account is logged) may be referred to as a "patient terminal."

[0072] The storage unit 120 can store various information necessary for the prescription information provision system 100 according to the present invention. The storage unit 120 may be provided within the prescription information provision system 100 according to the present invention, or it may be provided on an external server 200 (or a database on an external server). In the present invention, it is sufficient for the storage unit 120 to be a space in which the information necessary for searching and recommending prescription information based on the large-scale language model 142 is stored, and it can be understood that there are no restrictions on physical space. Therefore, in the present invention, the storage unit 120 may be described interchangeably with database (DB) and memory.

[0073] On the other hand, the large-scale language model 142 may include at least one of the prescription information generation unit 142a, the prescription information verification unit 142b, and the learning unit 142c. In this case, the prescription information generation unit 142a and the prescription information verification unit 142b may be named differently depending on the type of prescription information (e.g., exercise or medication) for which a recommendation has been requested by the medical staff. For example, if there is a request for a recommendation of an exercise curriculum, the prescription information generation unit 142a and the prescription information verification unit 142b may also be called the exercise curriculum generation unit and the exercise curriculum verification unit.

[0074] The prescription information generation unit 142a may be configured to take the prompt generated by the prompt generation unit 141 as input to the large-scale language model 142 and to obtain prescription information corresponding to the service request as a response from the large-scale language model 142. The prescription information verification unit 142b may also be configured to verify whether the prescription information generated by the prescription information generation unit 142a corresponds to the request of the medical staff.

[0075] The learning unit 142c may be configured to train the large-scale language model 142 so that prompts corresponding to service requests and prescription information based on those prompts are generated as responses from the large-scale language model 142.

[0076] As shown in Figure 4(b), the learning unit 142c can configure the training data 400 so that appropriate prescription information is recommended to a specific patient based on the role assigned to the prescription information provision system 100. In this case, the learning unit 142c can refer to at least one of the patient DB 210, medical staff DB 220, exercise DB 230, and prescription DB 240 to configure the training data 400 so that the know-how for prescribing prescription information to patients is reflected based on the actual prescription history of medical staff. The learning unit 142c can use such training data 400 to train a large-scale language model (LLM) 142. The training data 400 may consist of sample datasets 410, 420, and 430 in a pre-configured format (data structure).

[0077] The format of sample datasets 410, 420, and 430 can be predefined to include data corresponding to at least one category (attribute or field). As shown in Figure 4(a), the categories may include at least some of the following: i) "System Role Definition" category C1, ii) "Recommended Information" category C2, iii) "Recommended Target" category C3, and iv) "Recommended Results" category C4.

[0078] The learning unit 142c can train the large-scale language model 142 to derive at least one of a prompt corresponding to a service request and a response based on the prompt as results, through training on sample datasets 410, 420, and 430.

[0079] For example, sample datasets 410 and 420 include: i) Data values ​​411 and 421 corresponding to category C1 of "System Role Definition": "We would like a rehabilitation specialist or orthopedic surgeon to take on the role of recommending exercise for the patient's rehabilitation," and "We would like a rehabilitation specialist who has worked at OO Hospital for 10 years and is in charge of the geriatric rehabilitation field to take on the role."; ii) Data values ​​412 and 422 corresponding to category C2 of "Information to be Recommended": "A 34-year-old male, ankle pain score of approximately 6, diagnosed as a simple ankle sprain," and "A 71-year-old female, lumbar spine..." The individual has a history of herniated disc surgery and is reluctant to exercise for more than 20 minutes. They have average flexibility for their age group. iii) Data values ​​413 and 423 corresponding to category C3 of "Recommended Target" are "We would like a 3-week exercise curriculum recommended" and "We would like a 10-week exercise curriculum recommended," and iv) Data values ​​414 and 424 corresponding to category C4 of "Recommended Outcomes" are "[ / INST]["Ankle raises", ...]" and "We recommend the following exercise curriculum: 10 waist bends ~~~~" may be included.

[0080] In this invention, for the sake of convenience of explanation, the prompt generation unit 141, prescription information generation unit 142a, prescription information verification unit 142b, and learning unit 142c are not specifically distinguished, but are all referred to as "control unit 140" and LLM 142.

[0081] On the other hand, the Large-Scale Language Model (LLM) 142 may be configured to generate prompts corresponding to service requests and exercise curricula based on those prompts as responses, based on learning from the training data 400. The recommendation information provision system 100 based on the language model according to the present invention will be described in more detail below with the attached drawings. As described above, the present invention can provide various types of prescription information (e.g., pharmaceuticals, exercise, dietary therapy, counseling, cognitive therapy, etc.), and below, for the sake of explanation, the method of recommending an "exercise curriculum (or exercise program)" as prescription information will be described as an example.

[0082] In this invention, it is possible to perform the process of receiving prescription information recommendation requests from a medical staff terminal logged in with a medical staff account (see S510, Figure 5a).

[0083] The control unit 140 can receive service requests, including user queries entered by medical staff, from user terminals 20 logged in with a medical staff account.

[0084] As described above, in the present invention, service requests including user queries can be received directly from medical staff via the prescription information provision system 100, or service requests can be received from medical staff via other medical services linked with the prescription information provision system 100.

[0085] For example, the control unit 140 can provide a service page offered by the prescription information provision system 100 to the user terminal of a medical staff member and receive service requests entered by the medical staff member on the service page. As another example, as shown in Figure 1, when the functions of the prescription information provision system 100 are executed while other medical services 10 (for example, a screen for another medical service) are being provided on the user terminal of a medical staff member, the control unit 140 can provide an input field 5 on the user terminal where service request information can be entered.

[0086] In this invention, a "service request" may include a user query from a medical staff member. For example, as shown in Figure 6, a service request 610 may include a user query regarding exercise curriculum recommendations, such as "I would like a 3-week exercise curriculum recommendation." A service request may consist of various data types, such as text, audio, video, images, and signals.

[0087] In the present invention, in response to the above request, the process of extracting medical staff metadata of medical staff linked to the medical staff account and patient metadata of patients to be treated linked to the medical staff account can be performed from the server (see S520, Figure 5a).

[0088] When a service request is received, the control unit 140 can identify the medical staff who will prescribe the exercise curriculum and the patient to whom the exercise curriculum will be prescribed. The control unit 140 can generate a system prompt that assigns the role of the specific medical staff to the prescription information provision system 100, and a user prompt that identifies (explains) the patient to whom the exercise curriculum will be prescribed.

[0089] The control unit 140 can take system prompts and user prompts as inputs to the large-scale language model 142 and receive an exercise curriculum recommended for a patient described as a specific medical staff member as a response.

[0090] In the present invention, the role of the prescription information provision system 100 may correspond to actual medical staff or to virtual medical staff.

[0091] The control unit 140 can identify the role (or persona) of the prescription information provision system based on at least one of the user queries included in the medical staff account and service request.

[0092] The control unit 140 can identify the role of the prescription information provision system 100 based on which medical staff's user account the service request was received from. For example, based on the fact that the service request was received from physician A's user account, the control unit 140 can identify physician A as the role of the prescription information provision system 100.

[0093] Furthermore, the control unit 140 can identify the role of the prescription information provision system 100 through analysis of user queries included in the service request. For example, the control unit 140 can extract information that can identify medical staff, such as the name, specialty, and background of medical staff, from the user query, and identify the role of the prescription information provision system 100 based on the extracted information.

[0094] Furthermore, the control unit 140 can, firstly, identify the role of the prescription information provision system 100 based on the medical staff's user account, and secondly, change the role of the prescription information provision system that was first identified based on a user query.

[0095] More specifically, the control unit 140 can refer to the medical staff DB 220 to identify the medical staff member corresponding to the user account and who requested the service. The control unit 140 can compare the information of the medical staff member who requested the service (e.g., name, gender, age, work history, specialty, hospital, etc.) with the medical staff member corresponding to the user query.

[0096] If the comparison reveals that the medical staff requesting the service and the medical staff responding to the user query are the same, the control unit 140 can identify the medical staff requesting the service as the role of the prescription information provision system 100. For example, suppose a service request is received from a rehabilitation specialist A, including a user query stating, "As a rehabilitation specialist, I would like you to recommend an exercise curriculum." Based on the fact that the specialist A requesting the service and the medical staff included in the user query are both "rehabilitation specialists," the control unit 140 can identify specialist A as the role of the prescription information provision system 100.

[0097] In contrast, if the comparison reveals that the medical staff requesting the service and the medical staff responding to the user query are different, the control unit 140 can identify the role of the prescription information provision system 100 based on the user query. For example, suppose a service request is received from rehabilitation specialist A stating, "I would like you to recommend an exercise curriculum as an orthopedic specialist with 30 years of experience." The control unit 140 can identify the role of the prescription information provision system 100 not as rehabilitation specialist A, but as a hypothetical "orthopedic specialist with 30 years of experience."

[0098] Once the role of the prescription information provision system 100 is identified, the control unit 140 can extract medical staff metadata 620 from the medical staff DB 220, as shown in Figure 6, in order to generate system prompts to be input into the large-scale language model 142.

[0099] In this invention, "medical staff meta-information" is information used to realize the role of the prescription information provision system 100, and can be understood as information about a specific medical staff member. For example, the medical staff meta-information may include at least one of the following: the medical staff member's specialty, hospital name, career history, age, gender, hospital where they work, and research field.

[0100] When the control unit 140 identifies a medical staff member who requested a service as the role for the prescription information provision system 100, it can extract medical staff information associated with that medical staff member as metadata from the medical staff database 220. Even if the medical staff member who requested the service has not entered specific information about themselves, the control unit 140 can extract medical staff metadata from the medical staff database 220, thereby enabling the prescription information provision system 100 to perform the role of that medical staff member. For example, a rehabilitation specialist A can enter a user query such as "I would like you to recommend an exercise curriculum" without providing specific information about themselves, and the control unit 140 can extract medical staff metadata about specialist A from the medical staff database and assign the role of specialist A to the prescription information provision system 100.

[0101] Furthermore, the control unit 140 can extract information as metadata from the medical staff database 220 about other medical staff corresponding to a specific medical staff member identified as having a role in the prescription information provision system 100. For example, the control unit 140 can extract information as metadata from the medical staff database 220 about other medical staff corresponding to "a rehabilitation specialist like specialist A".

[0102] Furthermore, if the control unit 140 identifies a virtual medical staff member as a role in the prescription information provision system 100, it can extract medical staff information corresponding to the virtual medical staff member from the medical staff DB 220 as metadata. For example, if a virtual "orthopedic surgeon with 30 years of experience" is identified as a role in the prescription information provision system 100, the control unit 140 can refer to the exercise DB 230 and extract various metadata linked to the virtual "orthopedic surgeon with 30 years of experience".

[0103] On the other hand, when the control unit 140 receives a service request from the medical staff account, it can identify which patient among the patients associated with the medical staff account the service request pertains to.

[0104] The control unit 140 can identify which patient the service request is for based on which patient's prescription process the service request was received in. For example, suppose a service request is received while a prescription page linked to patient B is provided on the medical staff's user terminal. Based on the fact that the service request is received, the control unit 140 can identify that it is a service request for the specific patient.

[0105] In this case, the control unit 140 can receive patient information such as the patient's name and personal details along with the service request.

[0106] Furthermore, the control unit 140 can also identify which patient the service request is for by analyzing the user query included in the service request. The user query may contain information that identifies a patient (for example, "Please recommend an exercise curriculum for patient Kim Cheol-su"), and the control unit 140 can identify the patient to whom the exercise curriculum will be prescribed from the user query. As another example, the user query may contain information that identifies the patient's indication or patient tendencies (for example, "Please recommend an exercise curriculum for a patient with an ankle sprain"), and the control unit 140 can identify the patient to whom the exercise curriculum will be prescribed from the user query.

[0107] When a patient to whom an exercise curriculum is prescribed is identified, the control unit 140 can extract patient metadata 630 from the patient database 210, as shown in Figure 6, in order to generate user prompts to be entered into the large-scale language model 142.

[0108] In the present invention, "patient meta-information" refers to information used to describe a patient to whom an exercise curriculum is prescribed, and may include, for example, at least one of the following: patient preferences (e.g., exercise preferences, personality traits, lifestyle), surgical history, age, sex, exercise prescription record, diagnosis, pain location, pain score, and past exercises.

[0109] In the present invention, the process of generating a prompt in a pre-set format can be performed using the medical staff metadata and the patient metadata (see S530, Figure 5a).

[0110] The control unit 140 can generate prompts for deriving motor curriculum recommendation responses from a large-scale language model using at least a portion of the user queries and extracted metadata included in the service request.

[0111] As described above, the prompt in the present invention may include at least one of a system prompt and a user prompt.

[0112] In this invention, "system prompt" is an instruction or command word for defining (assigning) the role of the prescription information provision system 100, defining the persona of the medical staff who prescribe the exercise curriculum, and commanding the language model to recommend an exercise curriculum based on the defined persona (for example, ">[INST] < <sys>It may be configured as follows: "We would like a rehabilitation specialist or orthopedic surgeon to take on the role of recommending exercise for the patient's rehabilitation." The role of the server may also be defined as the aforementioned medical staff.

[0113] Furthermore, the "user prompt" defines the patient profile and patient medical information (for example, "<< / sys> This may include: "A 34-year-old male with an ankle pain score of approximately 6, diagnosed with a simple ankle sprain." In other words, the user prompt may be configured to identify the patient for whom the exercise curriculum is recommended and the indications for the patient.

[0114] The control unit 140 can generate prompts in a pre-configured format to be input to a large language model, including at least one of a system prompt and a user prompt. The control unit 140 can utilize the pre-configured format prompts and the large language model to assign a specific medical staff role to the prescription information provision system 100, and obtain an exercise curriculum as a response from the large language model to which the specific medical staff role has been assigned, for treating an indication for a specific patient.

[0115] In the present invention, the pre-configured format of the prompt may be configured to include at least one category (attribute or field) and data values ​​in the category, so as to correspond to the training data that the large-scale language model learns (see reference numeral "400" in Figure 4(a)).

[0116] As shown in Figure 6, the categories may include at least some of the following: i) "System Role Definition" category C1, ii) "Recommended Information" category C2, and iii) "Recommended Target" category C3.

[0117] The control unit 140 can generate a prompt 640 by arranging the categories C1, C2, C3 and the data values ​​641, 642, 643 corresponding to the categories according to a pre-set format.

[0118] Here, the "System Role Definition" category C1 and its corresponding data value 641 are commands or instructions for defining the role of the prescription information provision system 100, and can also be understood as "system prompts" in this invention. Furthermore, the "Recommended Information" category C2 and its corresponding data value 642 are information describing patients for whom an exercise curriculum is prescribed (or recommended), and can also be understood as "user prompts" in this invention.

[0119] The control unit 140 sequentially arranges a specific category and the data values ​​corresponding to that specific category, enabling the large-scale language model 142 to recognize which data value corresponds to which category. For example, as shown in Figure 6, the control unit 140 can generate a prompt 640 by sequentially arranging the data value 641 "We want a rehabilitation specialist or orthopedic surgeon to take on the role of recommending exercise for the patient's rehabilitation" in the "system role definition" category C1. The large-scale language model 142, upon receiving the prompt 640, can recognize that the data value 641 "We want a rehabilitation specialist or orthopedic surgeon to take on the role of recommending exercise for the patient's rehabilitation" belongs to the "system role definition" category C1.

[0120] Furthermore, the control unit 140 can generate data values ​​641, 642, and 643 corresponding to categories C1, C2, and C3 using keywords associated with those categories. For example, data value 641 corresponding to category C1, "System Role Definition," may include the keyword "Role 641a," and data value 642 corresponding to category C2, "Recommended Target Information," may include the name of the keyword indication 642a (for example, "Ankle Sprain"). Also, data value 643 corresponding to category C3, "Recommended Target," may include the keyword "Recommendation 643a."

[0121] On the other hand, in the pre-configured prompt 640, the "recommended target" category C3 and the corresponding data value 643 may include a script corresponding to the recommended target. The control unit 140 can generate the data value 643 corresponding to the recommended target category using the user query included in the service request 610 (for example, "I would like a 3-week exercise curriculum recommended").

[0122] The control unit 140 analyzes the intent behind a service request by analyzing user queries, and can analyze (understand or grasp) the intentions behind the service request made by medical staff. The control unit 140 can utilize at least one of various artificial intelligence technologies to analyze the intent behind a service request. For example, the artificial intelligence technologies may include at least one of the following: Machine Reading Comprehension (MRC), Paraphrasing, Abstract Summarization, Text Generation (TG), Question Generation (QG), Natural Language Understanding (NLU), OCR-NLP (Textanding), and Natural Language-Based Queries (NL2SQL).

[0123] In this invention, the process of acquiring the exercise curriculum can be performed using a large language model (LLM) that receives the prompt as input (see S540, Figure 5a). Furthermore, in this invention, the process of providing the acquired prescription information to the medical staff terminal can be performed (see S550, Figure 5a).

[0124] The large-scale language model 142 can be understood as a language model trained to generate motor curricula that conform to the patient's condition and the requirements of medical staff, based on the training data 400.

[0125] The control unit 140 processes prompts, including system prompts and user prompts, as input to a large-scale language model, thereby assigning specific medical staff roles to the prescription information provision system 100 and generating exercise curricula for patients as those specific medical staff.

[0126] For example, as shown in Figure 6, the control unit 140 can assign a role to the prescription information provision system 100 by processing a data value 641 corresponding to the "system role definition" category C1 (for example, "We want a rehabilitation or orthopedic specialist to take on the role of recommending exercise for the patient's rehabilitation") as input to the large language model. Furthermore, the control unit 140 can identify a patient to whom an exercise curriculum is prescribed based on a data value 642 corresponding to the "information to be recommended" category C2 (for example, a 34-year-old male, ankle pain score of approximately 6, diagnosis of simple ankle sprain), and input a prompt to the large language model 142 so that an exercise curriculum based on the data value 643 corresponding to the "recommended target" category C3 is recommended for the specific patient.

[0127] The control unit 140 can utilize the large-scale language model 142 and the prompts to generate an appropriate exercise curriculum that takes into account the patient's indications, acting as a specific medical staff member.

[0128] As described above, the "exercise curriculum" in the present invention may include at least one exercise item and the method of performing the exercise item (e.g., exercise schedule, exercise sequence, number of repetitions, exercise duration, etc.). Furthermore, "exercise item" may mean an exercise movement or a type of exercise.

[0129] The control unit 140 can search the exercise DB 230 for exercise items necessary to treat the indications of a patient for whom an exercise curriculum is prescribed. In this case, based on the role of the prescription information provision system 100 included in the prompt, the control unit 140 can search for at least one exercise item necessary to treat the patient's indications, taking into account the patient's indications, temperament, age, sex, etc., as described in the prompt.

[0130] The exercise database 230 of the present invention may contain multiple exercise items, the difficulty level of each exercise item, explanatory information, exercise videos, and at least a portion of the exercise methods, all associated with one another. The exercise database 230 may also contain multiple exercise items grouped together based on indications, and each of the multiple exercise items may have associated explanatory information such as its name, exercise method, and exercise effects.

[0131] The control unit 140 can generate an exercise curriculum for the patient using the retrieved exercise items. In this case as well, based on the role of the prescription information provision system 100 and the patient's description, the control unit 140 can generate an exercise curriculum that includes the exercise sequence, exercise time, number of exercise days, number of exercises, exercise method, etc., of the multiple exercise items in order to treat the patient's indications.

[0132] The large-scale language model 142 can search for exercise items from all the exercise items included in the exercise DB 230 that reflect the patient's condition, exercise prescription status, medical staff requirements, etc., based on the training data 400, and generate an appropriate exercise curriculum from the retrieved exercise items. More specifically, the large-scale language model may be trained to generate an exercise curriculum over a pre-set prescription period to treat a patient indication identified based on a user prompt, as a role defined based on a system prompt.

[0133] The large-scale language model 142 can refer to the exercise DB 230 to retrieve at least one exercise item and generate an exercise curriculum by assigning the exercise item to each of several different days that constitute a pre-set prescription period.

[0134] The control unit 140 can identify the prescription period from the service request. If the service request (for example, "I would like a 3-week exercise curriculum prescribed") includes a prescription period, the control unit 140 can identify that period as the prescription period for the exercise curriculum. Furthermore, based on the service request and the patient's metadata identified from the patient database 210, the control unit 140 can identify an appropriate prescription period for prescribing the exercise curriculum to the patient.

[0135] As shown in Figure 6, the control unit 140 can obtain a target dataset 650 from the large-scale language model 142, in which a prompt and the result (exercise curriculum) based on the prompt are included as a pair. More specifically, the target dataset 650 may include i) a data value 651 corresponding to the "system role definition" category C1, ii) a data value 652 corresponding to the "recommended information" category C2, iii) a data value 653 corresponding to the "recommended target" category C3, and iv) a data value 654 corresponding to the "recommended result" category C4. Here, the data value 654 corresponding to the "recommended result" category C4 (for example, "[ / INST]["Ankle raise", ...]") may include information about the prompt-based exercise curriculum.

[0136] In the present invention, obtaining a prompt-based motor curriculum (or response) from a large-scale language model can be understood as obtaining a target dataset 650, or a data value 654 corresponding to the "recommended result" category C4 included in the target dataset 650.

[0137] On the other hand, in the present invention, the exercise DB230 may have exercise items for treating each indication associated with it as a group, and each of the exercise items for treating the indication may have exercise difficulty level information associated with it.

[0138] The control unit 140 can control the large-scale language model 142 to search for exercise items from the exercise DB 230 to treat the patient's indications based on prompts, and to generate an exercise curriculum by assigning the searched exercise items to different days or weeks that constitute a preset exercise period.

[0139] In this case, the control unit 140 can control the large-scale language model to search for exercise items based on prompts, further considering at least one of the medical staff's exercise curriculum prescription history, the patient's age, the patient's gender, and the patient's exercise inclination, and then assign the searched exercise items to generate an exercise curriculum.

[0140] On the other hand, the control unit 140 can transmit prescription information to at least one medical system linked with the server in response to an approval event occurring in the medical staff account for the exercise curriculum.

[0141] As described above, the medical system in the present invention may include any one of the following: an electronic health record (EHR) system, an electronic medical record (EMR) system, an electronic prescription system (OCS) (Ordering Communication System), a clinical information system (CIS), and a personal health record (PHR) system for patients.

[0142] The control unit 140 can receive the medical staff terminal with the page of the at least one medical system provided. In this case, the control unit 140 can transmit the prescription information to the medical system so that the prescription information is output to the page of the at least one medical system.

[0143] More specifically, the control unit 140 can generate prescription recommendation information (also referred to as a report) corresponding to the medical staff's service request information, based on the response obtained from the large-scale language model. The control unit 140 can also provide the prescription recommendation information to the medical staff's user terminal.

[0144] As described above, the prescription information provision system 100 according to the present invention can provide various types of prescription information (e.g., pharmaceuticals, medical devices, medical tests, treatments, exercise curricula, exercise programs, exercise plans, cognitive therapy, dietary therapy, counseling, etc.) as recommendation information. For example, as shown in Figure 7a, based on receiving a request for a pharmaceutical prescription from medical staff, the control unit 140 can generate a prompt 700a for recommending pharmaceutical prescription information. The control unit can also use the prompt 700a as input to the large-scale language model 142 to provide medical staff with prescription information 700, including appropriate pharmaceuticals to be prescribed to the patient. As another example, as shown in Figure 7b, based on receiving a request for an exercise curriculum prescription from medical staff, the control unit 140 can generate a prompt 710a for recommending exercise curriculum prescription information. The control unit 140 can also use the prompt 710a as input to the large-scale language model 142 to provide medical staff with prescription information 710, including an exercise curriculum to be prescribed to the patient.

[0145] On the other hand, the control unit 140 can generate prescription information in various formats (types, formats, attributes). For example, the control unit 140 can provide exercise curriculum recommendation information as at least one of the following: prescription type (see drawing reference numeral "700" in Figure 7a and drawing reference numeral "710" in Figure 7b), other medical service type (see drawing reference numeral "720" in Figure 7c), medical service type of prescription information provision system 100 (see drawing reference numeral "730" in Figure 7d), and medical service type provided to the patient (for example, application type installed on the patient terminal, see drawing reference numeral "740" in Figure 7e).

[0146] The control unit 140 can determine what type of recommendation information to provide the exercise curriculum as, based on the user query included in the service request ("Please provide the exercise curriculum as a prescription-format report").

[0147] Furthermore, the control unit 140 can identify the type of recommended information for the exercise curriculum based on the medical services being provided (or performed) to the medical staff terminal (for example, other medical services, medical services of the prescription information provision system 100, etc.). In other words, the control unit 140 can provide the exercise curriculum in the same format (or chart) as the service page currently being provided to the medical staff.

[0148] Once the control unit 140 identifies what type of recommendation information the exercise curriculum should be provided to medical staff, it can generate prompts to retrieve the exercise curriculum corresponding to the identified recommendation information type from the large-scale language model 142.

[0149] The control unit 140 can generate prompts that include information about the recommended information type of exercise curriculum. The control unit 140 can also control a large-scale language model so that an exercise curriculum of the recommended information type identified based on the prompt is retrieved.

[0150] For example, suppose a service request includes a user query that specifies the recommended information type as "prescription." The control unit 140 shown in Figure 7b may include a command 710b in the prompt 710a input to the large language model that specifies an exercise curriculum of the "prescription" type (e.g., "Please provide the exercise curriculum as a prescription-formatted report").

[0151] As another example, suppose that other medical services (e.g., EHR medical services or EMR medical services) are provided at the medical staff terminal. As shown in Figure 7c, the control unit 140 may include in the prompt 720a input to the large language model an instruction 720b that indicates an exercise curriculum of the "other medical services" type (e.g., "Please input into the EMR system chart to show the exercise curriculum.").

[0152] As another example, let's assume that the medical services of the prescription information provision system 100 are provided on the medical staff terminal. As shown in Figure 7d, the control unit 140 may include in the prompt 730a input to the large language model an instruction 730b that specifies an exercise curriculum of the type "medical services of the prescription information provision system 100" (for example, "Please show the exercise curriculum in the medical staff's digital therapeutic prescription.").

[0153] Furthermore, the control unit 140 can control whether the exercise recommendation curriculum obtained from the large-scale language model is provided on the patient's terminal, rather than on the medical staff's terminal. As shown in Figure 7e, the control unit 140 can generate a prompt 740a that includes a command 740b (e.g., "Please show me in the patient application") to cause the patient's terminal to provide the exercise curriculum.

[0154] The control unit 140 can acquire various types of motor curricula based on prompts input to the large-scale language model. As described above, the control unit 140 can acquire information about motor curricula or a target dataset containing motor curricula from the large-scale language model.

[0155] Such exercise curriculum or target datasets may contain different data depending on the type of recommended information the exercise curriculum is to be provided as. Based on the data obtained from the large-scale language model, the control unit 140 can recommend the exercise curriculum as various types to the medical staff terminal.

[0156] For example, as shown in Figure 7b, the control unit 140 can provide the exercise curriculum 711 as a prescription type 710. Also, as shown in Figure 7c, the control unit 140 can cooperate with other medical services (e.g., EHR medical services or EMR medical services) to output the exercise curriculum 721 as another medical service type 720 (e.g., a patient prescription information chart). Furthermore, as shown in Figure 7d, the control unit 140 can recommend the exercise curriculum 731 as a format for a medical service type 730 (e.g., a digital therapeutic agent) provided by the prescription information provision system 100. In addition, as shown in Figure 7e, the control unit 140 can control the output of the exercise curriculum 741 to the patient terminal as an application type 740 installed on the patient terminal.

[0157] On the other hand, the control unit 140 can verify the prescription information in order to ensure the reliability of the prescription information.

[0158] The control unit 140 can verify whether the exercise program meets pre-set verification criteria. If the verification results in the program not meeting the pre-set verification criteria, the control unit 140 may regenerate prescription information based on the prompt and the language model. The pre-set verification criteria may be related to at least one of the following: the degree of relevance of the patient's indication to the treatment, the difficulty level of the exercise items included in the prescription information, the patient's age, and the patient's exercise inclination.

[0159] More specifically, as shown in Figure 8, the control unit 140 can generate prescription information using a large-scale language model (S810). The control unit 140 can also verify the generated prescription information.

[0160] The control unit 140 can verify whether the prescription information generated from the large-scale language model corresponds to the prescription information for the treatment of the patient's disease, and can determine whether such verification has been completed (S820).

[0161] Based on the determination that the verification of prescription information is complete, the control unit 140 may provide the prescription information to at least one of the medical staff and the patient (S821, S830). In this case, the control unit 140 may provide the verified prescription information to the patient account (or patient terminal) based on the completion of confirmation of the verified prescription information from the medical staff account (or receipt of medical staff verification information).

[0162] In contrast, if the control unit 140 determines, based on the verification results, that the prescription information does not correspond to the treatment of the patient's indication, it can generate new prescription information using a large-scale language model (S822, S840). The control unit 140 can verify the different prescription information and, based on the verification results, decide whether or not to provide the different prescription information to at least one of the medical staff and the patient. In this case, the control unit 140 can receive a prescription information regeneration request from the medical staff account (S822), and based on the received regeneration request, it may repeat the above-described series of processes so that prescription information different from the previous prescription information is generated.

[0163] There can be various methods for verifying prescription information. The control unit 140 can refer to a database (at least one of the patient DB 210, medical staff DB 220, exercise DB 230, and prescription DB 240) to confirm the actual treatment progress of a patient based on the prescription information. Based on the confirmation that the treatment progress of the patient's indication has improved according to the prescription information, the control unit 140 can verify that the prescription information is suitable for the treatment of the patient's indication. On the other hand, based on the confirmation that the treatment progress of the patient's indication has worsened according to the prescription information, the control unit 140 can verify that the exercise curriculum is not suitable for the treatment of the patient's indication.

[0164] On the other hand, the validation of the exercise curriculum may be performed by medical staff. As shown in Figure 9(a), the control unit 140 can provide the exercise curriculum 911 to a specific area 910 on the medical staff's user terminal. The control unit 140 can also provide around the exercise curriculum 911 a function icon 912 for receiving requests for prescribing the exercise curriculum (e.g., "prescribe as is") and a function icon 913 for receiving requests for different exercise curricula (e.g., "recommend a different exercise curriculum").

[0165] The control unit 140 can prescribe an exercise curriculum 911 to a patient based on the selection of a function icon 912 linked to a prescription request from the medical staff's user terminal.

[0166] On the other hand, the control unit 140 can generate an exercise curriculum different from the exercise curriculum 911 using prompts, based on the selection of a function icon 913 corresponding to a request for a different exercise curriculum from the medical staff's user terminal.

[0167] Based on receiving a request for recommendation of a different exercise curriculum from the medical staff terminal, the control unit 140 can acquire an exercise program different from the aforementioned exercise program using prompts and a large-scale language model 142.

[0168] In this case, the control unit 140 can obtain a different exercise curriculum by excluding one of the exercise items from the exercise program and updating the exercise curriculum by including other exercise items in the exercise DB 230 that are associated with the same group as the excluded exercise item and have the same difficulty level information associated with them.

[0169] More specifically, the control unit 140 can obtain a different exercise curriculum from the large-scale language model 142 by using the prompts used to generate the exercise curriculum 911 as input to the large-scale language model. The large-scale language model can generate an exercise curriculum different from the previously generated exercise curriculum when the same prompt is input. In this case, the control unit 140 can regenerate an exercise curriculum in which at least one of the exercise items, exercise order, exercise time, number of exercises, or exercise method included in the previously generated exercise curriculum is different.

[0170] Furthermore, the control unit 140 can receive a request from medical staff to change (substitute or replace) at least one exercise item 911a from among the multiple exercise items included in the exercise curriculum 911. As shown in Figure 9(a), when a specific exercise item 911a is selected from among the multiple exercise items included in the exercise curriculum 911, and a function icon 913 linked to a different exercise curriculum request is selected, the control unit 140 can receive a change request for the specific exercise item 911a from the medical staff's user terminal. In response to the change request, the control unit 140 can change the specific exercise item 911a to another exercise item 911b. On the other hand, although Figure 9(a) shows multiple exercise items included in the exercise curriculum 911 as all being the same, it goes without saying that the multiple exercise items may be different from each other.

[0171] As shown in Figure 9(b), the control unit 140 can generate a prompt 920 that commands (or instructs) a change to a specific exercise item 911a in response to a request to change that specific exercise item 911a. More specifically, the control unit 140 can modify the prompt 920 in a pre-set format so that the data value 921 of the "recommended item" category C3 is specified as "I would like a different exercise curriculum recommended. I would like to change the exercise of raising heels while holding onto a chair to another exercise," thereby identifying the exercise item 921a to be changed.

[0172] The control unit 140 can use the modified prompt 920 as input to the large-scale language model to obtain a different exercise curriculum in which a specific exercise item has been modified. The control unit 140 can search the exercise DB 230 for other exercise items that are intended to treat the same indication as the exercise item to be modified and have the same difficulty level, and can control the large-scale language model to regenerate the exercise curriculum to include the other exercise items found in place of the exercise item to be modified.

[0173] As shown in Figure 9(c), the exercise DB 230 may have an indication 931 and a difficulty level 933 associated with each exercise item 932. For example, the indication "ankle sprain 931a" may be associated with the exercise items heel raises while holding onto a chair 932a, ankle movements while standing 932b, and passive ankle dorsiflexion 932c, and each of the exercise items 932a to 932c may be associated with difficulty level information 933a to 933c. The control unit 140 can refer to the exercise DB 230 and control the large language model 142 to generate a new exercise item that includes other exercise items 932b (e.g., ankle movements while standing) that target the same indication 931a (ankle sprain) as the exercise item 932a ("heel raises while holding onto a chair") and have the same difficulty level 933a, 933b (e.g., "difficulty level 1").

[0174] On the other hand, the control unit 140 can generate prompts using various information and use these prompts to obtain an exercise curriculum to prescribe to the patient from a large-scale language model.

[0175] As described above, the control unit 140 can identify the role of the prescription information provision system 100 using medical staff metadata, and can identify patients to whom an exercise curriculum is prescribed (or recommended) using patient profiles and patient medical information. Furthermore, the control unit 140 can utilize various information necessary for recommending the optimal exercise curriculum, such as prescription history issued by medical staff in the past, feedback information received from medical staff, and patient preference information.

[0176] The control unit 140 can generate prompts in a pre-set format using various information. In the present invention, the prompts may include at least some of the following: i) "System Role Definition" category C1, ii) "Recommended Information" category C2, and iii) "Recommended Target" category C3. The control unit 140 can configure the multiple categories C1, C2, and C3 using various information. For example, the control unit 140 can configure the data values ​​for the "Recommended Information" category C2 using patient preference information. Furthermore, the control unit 140 can generate prompts that further include new categories other than categories C1, C2, and C3, and data values ​​corresponding to the new categories.

[0177] For example, as shown in Figure 10, the control unit 140 can extract past prescription history information 1010 (e.g., prescription details, prescription reason, post-prescription follow-up information, etc.) of the medical staff requesting the service from at least one of the medical staff DB 220 and prescription DB 240. The control unit 140 can also use the past prescription history information 1010 of the medical staff requesting the service to configure data values ​​1020 corresponding to the "system prescription history" category C5. In this case, the control unit 140 can also use the past prescription history information 1010 of the medical staff to configure data values ​​corresponding to the "system role definition" category C1.

[0178] As another example, the control unit 140 can extract patient trait information 1030 from at least one of the patient database 210 and user queries, and use the extracted patient trait information to construct a data value 1040 for the "recommended information" category C2 (for example, "This patient dislikes repeating the same action multiple times").

[0179] In this way, the control unit 140 can generate prompts 1050 using various information, and use these prompts 1050 as input to a large-scale language model to obtain a customized exercise curriculum for the patient. As described above, the control unit 140 can also obtain a target dataset 1060 consisting of multiple categories and data values ​​corresponding to those categories.

[0180] As described above, the exercise curriculum recommendation method and system based on the language model according to the present invention receives a request for exercise curriculum recommendation from a medical staff terminal logged in with a medical staff account, and in response to the request, extracts medical staff metadata of the medical staff linked to the medical staff account and patient metadata of the patient being treated linked to the medical staff account from the server, and generates a prompt in a pre-set format using the medical staff metadata and patient metadata. In this way, the present invention enables a large-scale language model to recommend an appropriate exercise curriculum from the perspective of a medical staff member, taking into account the patient's condition.

[0181] Furthermore, the exercise curriculum recommendation method and system based on a language model according to the present invention uses a large language model (LLM) that receives the prompt as input to acquire the exercise curriculum and provides the acquired exercise curriculum to the medical staff terminal. As a result, patient information and medical staff requirements are input into the large language model, an appropriate exercise curriculum is recommended, and the recommended exercise curriculum can be reflected in the medical system.

[0182] On the other hand, computer-readable media include all types of recording devices that store data readable by a computer system. Examples of computer-readable media include HDDs (Hard Disk Drives), SSDs (Solid State Disks), SSDs (Silicon Disk Drives), ROMs, RAMs, CD-ROMs, magnetic tapes, floppy disks, and optical data storage devices.

[0183] Furthermore, the computer-readable medium may include storage, and may be a server or cloud storage accessible by electronic devices via communication. In this case, the computer can download the curriculum according to the present invention from the server or cloud storage via wired or wireless communication.

[0184] Furthermore, in the present invention, the computer described above is an electronic device equipped with a processor, i.e., a CPU (Central Processing Unit), and its type is not particularly limited.

[0185] On the other hand, the above detailed description should not be interpreted restrictively in any respect and should be considered illustrative. The scope of the invention should be determined by a reasonable interpretation of the appended claims, and any modifications within the scope of the equivalents of the invention are included within the scope of the invention.

Claims

1. Steps include receiving prescription information recommendation requests from medical staff terminals logged in with a medical staff account, and In response to the aforementioned request, the steps include extracting from the server medical staff metadata of the medical staff linked to the medical staff account, and patient metadata of the patients being treated linked to the medical staff account, The steps include generating a prompt in a pre-set format using the medical staff metadata and the patient metadata, The steps include: obtaining the prescription information using a Large Language Model (LLM) that receives the aforementioned prompt as input; A method for providing prescription information based on a language model, characterized by comprising the step of providing the acquired prescription information to the medical staff terminal.

2. The process further includes the step of transmitting the prescription information to at least one healthcare system that is linked to the server in response to an approval event occurring in the medical staff account for the prescription information, The aforementioned at least one medical system is A method for providing prescription information based on a language model according to claim 1, characterized in that it includes one of the following: an Electronic Health Record (EHR) system, an Electronic Medical Record (EMR) system, an Ordering Communication System (OCS), a Clinical Information System (CIS), and a Personal Health Record (PHR) system for patients.

3. The aforementioned request for prescription information recommendations is, The medical staff terminal receives the page of at least one medical system, In the step of transmitting the prescription information, A method for providing prescription information based on a language model according to claim 2, characterized in that the prescription information is output to the page of at least one medical system.

4. The aforementioned pre-configured prompts include system prompts and user prompts. The system prompt defines the role of the server as the medical staff, The method for providing prescription information based on a language model according to claim 1, characterized in that the user prompt is configured to identify the patient for whom the prescription information is recommended and the indications for the patient.

5. The motion database of the aforementioned server contains: For each indication, the exercise items for treating the said indication are associated as a group. Each of the exercise items for treating the aforementioned indications has associated information regarding the difficulty level of the exercise. In the step of obtaining the aforementioned prescription information, Search the aforementioned exercise database for exercise items to treat the aforementioned indications, A method for providing prescription information based on a language model according to claim 4, characterized in that the searched exercise items are assigned to each of the different days that constitute a predetermined exercise period, thereby prescribing an exercise program.

6. The method for providing prescription information based on a language model according to claim 5, characterized in that the search for the exercise item is performed considering at least one of the following: prescription information or prescription history from the medical staff, the patient's age, the patient's gender, and the patient's exercise inclination.

7. The step further includes, based on receiving a request for a different prescription information recommendation from the medical staff terminal, obtaining prescription information different from the prescription information using the prompt, In the step of obtaining the aforementioned different prescription information, A method for providing prescription information based on a language model according to claim 5, characterized in that one of the aforementioned exercise items is excluded from the exercise program, and other exercise items that are associated with the same group as the excluded exercise item and have the same difficulty level information associated with them are included in the prescription information.

8. A step of verifying whether the aforementioned prescription information meets pre-set verification criteria, If the verification results do not meet the pre-set verification criteria, the process further includes the step of regenerating the prescription information based on the prompt and the language model, The aforementioned pre-set verification criteria are: A method for providing prescription information based on a language model according to claim 1, characterized in that it is related to at least one of the following: the degree of relevance to the treatment of the patient's indication, the difficulty level of the exercise items included in the prescription information, the patient's age, and the patient's exercise inclination.

9. The communications department receives prescription information recommendation requests from medical staff terminals logged in with a medical staff account, The system includes a control unit that, in response to the aforementioned request, extracts medical staff metadata of medical staff linked to the medical staff account from the server, and patient metadata of patients being treated linked to the medical staff account, The control unit, Using the aforementioned medical staff metadata and patient metadata, a prompt in a pre-set format is generated. The prescription information is obtained using a Large Language Model (LLM) that receives the aforementioned prompt as input. A language model-based prescription information provision system characterized by providing the acquired prescription information to the medical staff terminal.

10. A program that is executed in accordance with one or more processes in an electronic device and stored on a computer-readable recording medium, The aforementioned program, Steps include receiving prescription information recommendation requests from medical staff terminals logged in with a medical staff account, and In response to the aforementioned request, the steps include extracting from the server medical staff metadata of the medical staff linked to the medical staff account, and patient metadata of the patients being treated linked to the medical staff account, The steps include generating a prompt in a pre-set format using the medical staff metadata and the patient metadata, The steps include: obtaining the prescription information using a Large Language Model (LLM) that receives the aforementioned prompt as input; A program stored on a computer-readable recording medium, characterized by including a command to perform the step of providing the acquired prescription information to the medical staff terminal.

Images