A medical record system using a large-scale language model, its operation method, and a method for creating medical records.

Abstract

This invention provides a medical record system using a large-scale language model that maximizes the performance of the large-scale language model, enabling rapid processing of clinical information and supporting accurate record-keeping and decision-making, as well as a method for operating this system and a method for creating medical records. [Solution] A medical record system using a large language model according to one embodiment includes a medical record creation device that performs a function of automatically creating medical records (Electronic Medical Records; EMRs) in conjunction with a large language model (LLM). This function preprocesses medical record information to generate structured input data, inputs the input data into the large language model to generate a medical record checklist based on the medical record information, and requests the large language model to create the medical record checklist to generate the medical record.

Description

Technical Field

[0001] The present invention relates to a medical record system using a large language model, an operation method thereof, and a medical record creation method. More specifically, the present invention relates to a medical record system using a large language model for managing information related to patient diagnosis and treatment, etc., an operation method thereof, and a medical record creation method.

Background Art

[0002] Generally, an Electronic Medical Record (EMR) is an essential document for recording patient diagnosis, treatment, etc. The process of creating a medical record is cumbersome and requires a lot of time. Therefore, problems such as creation errors and omission of important information frequently occur in the creation of medical records.

[0003] Recently, technologies for managing medical records using artificial intelligence models have been proposed. Among them, Natural Language Processing (NLP) technology has advantages for processing unstructured data such as doctors' findings, diagnoses, and prescriptions, and is thus positioned as an important technology in the field of medical records.

[0004] In particular, a Large Language Model (LLM) can automatically generate medical records on behalf of a medical team based on natural language understanding and generation capabilities. However, it is known that the performance of the large language model varies depending on the quality and format of the provided data. Therefore, in order to effectively use the large language model in a medical environment, it is important to systematically preprocess the data structure, input important information as input data, and systematically design the architecture.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

[0006] The object of the present invention is to provide a medical record system using a large-scale language model that maximizes the performance of the large-scale language model, thereby supporting the rapid processing of clinical information and the accurate creation of records and decision-making, as well as a method for operating the system and a method for creating medical records. [Means for solving the problem]

[0007] The medical record system using a large-scale language model according to the present invention includes a medical record creation device that performs a function of automatically creating medical records (Electronic Medical Records; EMRs) in conjunction with a large-scale language model (LLM). This function includes pre-processing medical record information to generate structured input data, inputting the input data into the large-scale language model to generate a medical record checklist based on the medical record information, and requesting the large-scale language model to create the medical record checklist to generate the medical record.

[0008] The input data includes input data of a first structure containing information on chronic diseases, input data of a second structure containing information on disease types, and input data of a third structure containing recent medical information and test information.

[0009] Here, the input data for the first structure is configured to include the disease name, diagnosis date, and major test results by date for the chronic disease, so as to minimize the amount of information related to the chronic disease.

[0010] Furthermore, the input data for the second structure is composed of simplified information related to acute and transient illnesses, and includes the disease name, date, and progress information for the acute and transient illnesses.

[0011] Furthermore, the input data for the third structure consists of the latest medical information and test information, along with a pre-set number of past medical information and test information.

[0012] On the other hand, in the generation of the medical record checklist, a first medical record checklist generated by the large-scale language model is output, real-time information acquired during the treatment process is input to the large-scale language model, and the large-scale language model generates and outputs a second medical record checklist that reflects the latest information.

[0013] Furthermore, the output of the first medical record checklist and the second medical record checklist provides a function to modify and add items included in the first medical record checklist and the second medical record checklist.

[0014] Furthermore, the real-time information includes at least one of the following: the conversation during the medical treatment, the medical records, and the examination records.

[0015] Furthermore, the large-scale language model is fine-tuned or subjected to transfer learning based on real-time information obtained during the generation process of the second medical record checklist.

[0016] Furthermore, if the large-scale language model generates the second medical record checklist and there is additional information that does not fall under the items of the second medical record checklist, it will separately include the additional information in the second medical record checklist.

[0017] On the other hand, in the generation of the medical record, a first medical record created by the large-scale language model is output, and the first medical record is modified or verified to generate a second medical record.

[0018] Furthermore, in the output of the first medical record, if there are items that have been omitted from creation by the large-scale language model, those missing items are highlighted.

[0019] The correction or verification of the aforementioned first medical record is then performed by the medical team.

[0020] Furthermore, the large-scale language model is fine-tuned or transfer-learned based on the second medical record.

[0021] On the other hand, the large-scale language model predicts and suggests necessary treatment and follow-up (F / U) items for the patient based on the medical record information, thereby supporting the medical team's decision-making.

[0022] Here, the large-scale language model generates a medical record checklist that includes information for the follow-up observation.

[0023] On the other hand, the method for operating a medical record system using a large-scale language model according to the present invention is a method for operating a medical record creation device that performs the function of automatically creating medical records (Electronic Medical Records; EMRs) in conjunction with a large-scale language model (LLM), wherein the medical record information is preprocessed to generate structured input data, the input data is input to the large-scale language model to generate a medical record checklist based on the medical record information, and the large-scale language model is requested to create the medical record checklist to generate the medical record.

[0024] And the method for automatically creating a medical record according to the present invention is a method for automatically creating an Electronic Medical Record (EMR) based on a medical record creation device linked to a Large Language Model (LLM). In this method, medical record information is preprocessed to generate structured input data, the input data is input into the large language model to generate a medical record checklist based on the medical record information, and the large language model is requested to create the medical record checklist to generate a medical record.

Advantages of the Invention

[0025] The medical record system using a large language model according to the present invention, this operation method, and the medical record creation method can perform rapid inference by systematically preprocessing and simplifying the structure of input data, and have the effect of being able to create a medical record quickly and accurately with only the simplified important information.

[0026] The technical effects of the present invention as described above are not limited to the effects mentioned above, and other technical effects not mentioned can be clearly understood from the following description.

Brief Description of the Drawings

[0027] [Figure 1] It is a conceptual diagram showing a medical record system using a large language model according to this embodiment. [Figure 2] It is a block diagram showing an operation method of a medical record system using a large language model according to this embodiment. [Figure 3] It is a conceptual diagram showing an operation and a medical record creation method of a medical record system using a large language model according to this embodiment.

Modes for Carrying Out the Invention

[0028] Embodiments of the present invention will be described in detail below with reference to the attached drawings. However, these embodiments are not limited to those disclosed below and can be embodied in a variety of forms. These embodiments are provided simply to complete the disclosure of the present invention and to fully inform those in the ordinary skill of the scope of the invention. The shapes of elements in the drawings may be exaggerated for the sake of clearer explanation, and elements indicated by the same reference numeral in the drawings refer to the same element.

[0029] Figure 1 is a conceptual diagram showing a medical record system using a large-scale language model according to this embodiment.

[0030] As shown in Figure 1, the medical record system 1000 according to this embodiment is used to create medical records (Electronic Medical Records; EMRs) throughout the patient's treatment process, and can record and manage the patient's health status, diagnosis, treatment, and test results. Therefore, the medical record system 1000 includes a medical record creation device 200 that is linked with a large-scale language model 100.

[0031] The large-scale language model 100 is a model pre-trained based on the medical domain and can be built on a separate model server 110 and linked with the medical record creation device 200 via wireless communication.

[0032] Furthermore, the medical record creation device 200 is composed of an electronic device on which a medical record creation program 200a can be installed, and a variety of electronic devices including display elements such as PCs, netbooks, tablet PCs, and smartphones can be used.

[0033] The medical record creation device 200 may include a data storage unit 210, a processing unit 220, and an input unit 230.

[0034] The data storage unit 210 is configured to include memory and stores the medical record creation program 200a and medical record information 10 for a large number of patients. The medical record creation program 200a works in conjunction with the large-scale language model 100 to perform the automated creation of medical records and the decision-making support of the medical team. The medical record information is accumulated patient-specific medical record information from the past to the present, including patient personal information, visit information, diagnostic information, prescription information, and test information.

[0035] The processing unit 220 then generates and executes processes for the automatic creation of medical records and for supporting the decision-making of the medical team, based on the medical record creation program 200a.

[0036] Furthermore, the input unit 230 may include an interface device for the medical team to input information related to the automated medical record creation process and decision support process into the medical record creation device 200.

[0037] On the other hand, the operation of the medical record system 1000 using a large-scale language model and the method of creating medical records will be explained in detail below with reference to the attached diagrams.

[0038] Figure 2 is a block diagram showing the operation method of the medical record system using the large-scale language model according to this embodiment, and Figure 3 is a conceptual diagram showing the operation of the medical record system using the large-scale language model according to this embodiment and the method of creating medical records.

[0039] As shown in Figures 2 and 3, the medical record system 1000 according to this embodiment can be applied to the automatic creation of patient medical records and to supporting decision-making by medical teams.

[0040] First, the large-scale language model 100 is a model that has been pre-trained in the medical domain and has been fine-tuned to support a series of operations such as generating medical record checklists, real-time monitoring and updating, automatic generation of medical records, verification of medical records, and decision support.

[0041] Therefore, before automatically creating patient medical records, the medical record creation program 200a can perform data feeding operations on the large-scale language model 100. Data feeding operations refer to the process of providing the large-scale language model 100 with data related to the automatic creation of medical records.

[0042] Here, the medical record creation program 200a can generate input data 20 by preprocessing the patient's medical record information 10 stored in the data storage unit 210 (S100). More specifically, the medical record creation program 200a can generate input data 20 by extracting date information, medical information, examination information, and disease type information from the medical record information 10.

[0043] As an example, the medical record creation program 200a extracts date information and medical information, and reconstructs the data by date. Here, medical information may include diagnosis, surgery name, problem list, and treatment plan. The medical record creation program 200a then reconstructs examination information by date. Here, examination information may include examination images, reports, examination reports, and numerical data. Furthermore, the medical record creation program 200a simplifies and reconstructs disease type information. Here, disease type information may include chronic disease information and acute and transient disease information. At this time, the medical record creation program 200a summarizes information related to chronic diseases into diagnosis and major test indicators and converts it into a time-series format, and summarizes information related to acute and transient diseases into date, diagnosis, and a brief course of events and simplifies it.

[0044] Therefore, the medical record creation program 200a can generate structured input data 20 including a first structure, a second structure, and a third structure based on the medical record information 10.

[0045] As an example, the input data 20 of the first structure is data related to chronic diseases, and is configured to include only the minimum amount of information related to chronic diseases.

[0046] Chronic diseases are illnesses that require consistent management, but they are relatively less important compared to recent conditions or acute problems that affect current medical care. Therefore, the input data 20 of the first structure can include only key information related to chronic diseases, such as the name of the chronic disease, the date of diagnosis, and the main test results by date, so that chronic disease information is minimized.

[0047] Furthermore, chronic diseases progress over a long period, resulting in a large amount of disease-related data. However, using a large amount of data as input data 20 increases the complexity of model inference and can lead to model overload. Therefore, the input data 20 of the first structure includes only key information related to chronic diseases, allowing the large-scale language model 100 to operate effectively without being constrained by unnecessary historical data.

[0048] As an example, the input data 20 of the second structure is data related to acute and transient illnesses, and includes a structure in which the related information is simplified. Acute and transient illnesses are characterized by the rapid onset of symptoms within a short period of time. Therefore, the input data 20 of the second structure can include only the acute and transient illness name, date, and progression information so that the information on acute and transient illnesses is simplified. Thus, the input data 20 of the second structure shortens the model processing time and enables effective inference by omitting unnecessary details and using only core information about the main illness.

[0049] Furthermore, acute and transient illnesses have the characteristic of having relatively simple symptoms or being treatable within a short time. Therefore, even if the input data 20 of the second structure contains only simplified information, the large-scale language model 100 can perform inference accurately and quickly.

[0050] As an example, the input data 20 of the third structure may include a specified number of recent medical and examination records. That is, the input data 20 of the third structure includes the most recent medical and examination records along with a set number of past medical and examination records. The recent medical and examination records include the most important information regarding the patient's current condition.

[0051] Therefore, the input data 20 of the third structure reflects the latest patient status and has high real-time responsiveness, and includes new diseases or treatment responses, so the large-scale language model 100 can perform reliable inference.

[0052] On the other hand, once the input data 20 of the first structure, the second structure, and the third structure are generated, the medical record creation program 200a combines the input data 20, which includes at least one of the first structure, the second structure, and the third structure, and performs data entry into the large-scale language model 100 (S200).

[0053] Therefore, the large-scale language model 100 can improve its inference speed and performance by reducing unnecessary data through minimized chronic disease information. In addition, the large-scale language model 100 has the advantage of being able to perform rapid inference by focusing on simplified acute and transient disease information, recent medical information, and test information, as the input information is simplified.

[0054] Meanwhile, once the input of input data 20 is completed, the large-scale language model 100 generates a first medical record checklist 30 containing the items necessary for creating medical records (S300). The medical record creation program 200a then outputs the first medical record checklist 30 via the medical record creation device 200. Here, the large-scale language model 100 automatically generates the first medical record checklist 30 necessary for creating medical records based on the input data 20, and can recommend additional information needed by the medical team. The medical record creation program 200a also provides the option for the medical team to modify or add items to the first medical record checklist 30 according to their concerns, the patient's condition, and circumstances.

[0055] On the other hand, once the first medical record checklist 30 is generated, the medical team can perform medical procedures such as taking a medical history, physical examination, tests, diagnosis, and establishing a treatment plan (S400). Here, the input unit 230 acquires information such as the dialogue between the medical team and the patient during treatment and medical examination records in real time and provides it to the medical record creation program 200a. The medical record creation program 200a then provides the provided information to the large-scale language model 100.

[0056] Therefore, the large-scale language model 100 updates the first medical record checklist 30 based on the information provided in real time to generate a second medical record checklist 40 that reflects the latest information (S500). The medical record creation program 200a then outputs the second medical record checklist 40, which reflects the real-time information, via the medical record creation device 200. Here, the second medical record checklist 40 accurately represents the latest record, reflecting whether or not an examination was performed, the examination results, and the content of related conversations. The medical record creation program 200a also provides the medical team with the option to modify or add items to the second medical record checklist 40 according to their concerns, the patient's condition, and other conditions.

[0057] Furthermore, the large-scale language model 100 is fine-tuned or subjected to transfer learning based on information acquired in real time during the generation process of the second medical record checklist 40. Therefore, the large-scale language model 100 maintains the timeliness and accuracy of records when generating other medical record checklists in the future.

[0058] Meanwhile, once the second medical record checklist 40 is generated, the medical record creation program 200a requests the large-scale language model 100 to create the second medical record checklist 40 at the request of the medical team. The large-scale language model 100 then automatically creates each item of the second medical record checklist 40 (S600). Here, the large-scale language model 100, which has been trained using natural language processing technology, creates each item of the second medical record checklist 40 in a language that the medical team can easily understand. If there is additional information that does not fall under the items of the second medical record checklist 40, the large-scale language model 100 records the additional information in a separate area and completes the creation of the second medical record checklist 40.

[0059] Meanwhile, once the large-scale language model 100 generates the first medical record 50 from the second medical record checklist 40, the medical record creation program 200a outputs the first medical record 50 via the medical record creation device 200. If there are any items that have been missed by the large-scale language model 100, the program highlights the missing items to draw the medical team's attention.

[0060] On the other hand, when the medical record creation program 200a outputs the first medical record 50, the medical team verifies the contents of the first medical record 50 created by the large-scale language model 100 (S700). At this time, the medical team checks the contents of the first medical record 50 created by the large-scale language model 100 to verify its accuracy and appropriateness.

[0061] The medical team can then use the medical record creation program 200a to modify the contents of the first medical record 50 or input additional information to generate a second medical record 60 as needed (S800). At this time, the medical record creation program 200a provides the second medical record 60, modified by the medical team, to the model server 110. The large-scale language model 100 can then perform fine-tuning or transfer learning based on the second medical record 60, which includes feedback from the medical team. Thus, the large-scale language model 100 undergoes continuous improvement to maintain high accuracy when generating medical records in the future.

[0062] On the other hand, the medical record creation program 200a can support decision-making by the medical team (S900). For example, when there is a need to continuously monitor and manage the patient's condition, such as in follow-up (F / U), the medical record creation program 200a can provide a decision support option using the large-scale language model 100.

[0063] At this time, the medical record creation program 200a can request support for clinical decision-making from the large-scale language model 100. The large-scale language model 100 can then predict and suggest necessary treatments and observation items for the patient based on the patient's past medical history and data. The large-scale language model can then suggest new follow-up observation items, generate a medical record checklist including these items, and output it via the medical record creation program 200a. Furthermore, the large-scale language model 100 can analyze the patient's education plan and related materials to support the medical team's clinical decision-making.

[0064] On the other hand, this embodiment describes how, once the generation of the second medical record 60 is complete, the medical record creation program 200a performs decision support work. However, this is merely to illustrate this embodiment, and it should be made clear that the decision support work can proceed independently of the medical record generation work.

[0065] Therefore, the medical record system and its operation method using the large-scale language model according to the present invention enable rapid inference by systematically preprocessing and simplifying the structure of input data, and have the effect of enabling the rapid and accurate creation of medical records using only simplified important information.

[0066] The embodiment of the present invention described above and shown in the drawings should not be construed as limiting the technical idea of ​​the present invention. The scope of protection of the present invention is limited only by the matters described in the claims, and a person with ordinary skill in the art of the present invention may improve and modify the technical idea of ​​the present invention in various ways. Accordingly, such improvements and modifications will fall within the scope of protection of the present invention, insofar as they are obvious to a person with ordinary skill.

Claims

1. This includes a medical record creation device that performs the function of automatically creating medical records (Electronic Medical Records; EMRs) in conjunction with a Large Language Model (LLM), The aforementioned function is, Preprocess medical record information to generate structured input data, The input data is input into the large-scale language model to generate a medical record checklist based on the medical record information. A medical record system using a large-scale language model, characterized by generating medical records by requesting the large-scale language model to create the medical record checklist.

2. The aforementioned input data is Input data of a first structure containing information on chronic diseases, Input data for a second structure containing information on disease types, A medical record system using a large-scale language model according to claim 1, characterized by including input data of a third structure, which includes recent medical information and test information.

3. The input data for the first structure is: A medical record system using a large-scale language model according to claim 2, characterized in that it includes the disease name, date of diagnosis, and major test results by date for the chronic disease, so as to include the minimum amount of information related to the chronic disease.

4. The input data for the second structure described above is: A medical record system using a large-scale language model according to claim 2, characterized in that it is composed of simplified information related to acute and transient illnesses, and includes disease name, date, and progress information for the said acute and transient illnesses.

5. The input data for the third structure described above is: A medical record system using a large-scale language model according to claim 2, characterized in that it includes the latest medical information and test information along with a predetermined number of past medical information and test information.

6. In generating the aforementioned medical record checklist, The first medical record checklist generated by the aforementioned large-scale language model is output, A medical record system using a large-scale language model according to claim 1, characterized in that real-time information acquired during the medical treatment process is input to the large-scale language model, and the large-scale language model generates and outputs a second medical record checklist that reflects the latest information.

7. In the output of the first medical record checklist and the second medical record checklist, A medical record system using a large-scale language model according to claim 6, characterized in that it provides a function to modify and add items included in the first medical record checklist and the second medical record checklist.

8. The aforementioned real-time information is, A medical record system using a large-scale language model according to claim 6, characterized in that it includes at least one of the above: dialogue during medical treatment, medical records, and examination records.

9. The aforementioned large-scale language model, A medical record system using a large-scale language model according to claim 6, characterized in that it is fine-tuned or transferred learning based on real-time information obtained during the generation process of the second medical record checklist.

10. The aforementioned large-scale language model, A medical record system using a large-scale language model according to claim 6, characterized in that, when generating the second medical record checklist, if there is additional information that does not fall under the items of the second medical record checklist, the additional information is separately recorded in the second medical record checklist.

11. In generating the aforementioned medical records, Output the first medical record created by the aforementioned large-scale language model. A medical record system using a large-scale language model according to claim 1, characterized in that the first medical record is modified or verified to generate a second medical record.

12. In the output of the first medical record described above, A medical record system using a large-scale language model according to claim 11, characterized in that if there are items that have been omitted from the creation by the large-scale language model, the missing items are highlighted.

13. The correction or verification of the first medical record is, A medical record system using a large-scale language model according to claim 11, characterized in that it is performed by a medical team.

14. The aforementioned large-scale language model, A medical record system using a large-scale language model according to claim 13, characterized in that it is fine-tuned or transfer-learned based on the second medical record.

15. The aforementioned large-scale language model, A medical record system using a large-scale language model according to claim 1, characterized in that it predicts and proposes necessary treatment and follow-up (F / U) items for a patient based on the aforementioned medical record information, thereby supporting the decision-making of the medical team.

16. The aforementioned large-scale language model, A medical record system using a large-scale language model according to claim 15, characterized in that it generates a medical record checklist that includes information for the aforementioned follow-up observations.

17. In a method for operating a medical record creation device that performs the function of automatically creating medical records (Electronic Medical Record; EMR) in conjunction with a Large Language Model (LLM), Preprocess medical record information to generate structured input data, The input data is input into the large-scale language model to generate a medical record checklist based on the medical record information. A method for operating a medical record system using a large-scale language model, characterized by generating medical records by requesting the large-scale language model to create the medical record checklist.

18. In a method for automatically creating medical records (Electronic Medical Records; EMRs) based on a medical record creation device linked to a Large Language Model (LLM), Preprocess medical record information to generate structured input data, The input data is input into the large-scale language model to generate a medical record checklist based on the medical record information. A method for automatically generating medical records, characterized by requesting the creation of the medical record checklist from the large-scale language model and generating medical records.

Images