Device and method for providing medical chat service based on generative artificial intelligence

The generative AI-based medical chat service addresses the challenge of interpreting medical records by generating clear and contextually relevant responses, enhancing patient understanding and addressing low health literacy through AI-driven explanations and persona-based interactions.

WO2024237490A1PCT designated stage expired Publication Date: 2024-11-21SEOUL NAT UNIV HOSPITAL +1
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
PCT/KR2024/004971
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2024-04-11
Filing Date
2024-04-12
Publication Date
2024-11-21

AI Technical Summary

Technical Problem

Patients face difficulties in understanding their medical records due to complex language and medical terminology, and existing natural language processing algorithms fail to effectively interpret these records, exacerbated by low health literacy and limited time for detailed explanations in clinical settings.

Method used

A generative artificial intelligence-based medical chat service that uses a device with processors and memory to generate prompts from user input, including medical records, and outputs responses using a generative AI model, incorporating features like persona selection and retrieval-augmented generation to provide clear and contextually relevant answers.

Benefits of technology

The solution provides immediate, consistent, and customized answers to patients' questions about their medical records, improving understanding and reducing the complexity of medical information through prompt engineering and AI-driven explanations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure KR2024004971_21112024_PF_FP_ABST
    Figure KR2024004971_21112024_PF_FP_ABST
Patent Text Reader

Abstract

A device and a method for providing a medical chat service based on generative artificial intelligence, according to one embodiment, are disclosed. The device for providing a medical chat service based on generative artificial intelligence, according to one embodiment, comprises: one or more processors; and a memory in which one or more programs executed by the one or more processors are stored, wherein the one or more processors generate a prompt on the basis of a user input and command, and inputs the prompt into a generative artificial intelligence model so as to output a corresponding response.
Need to check novelty before this filing date? Find Prior Art

Description

Device and method for providing a medical chat service based on generative artificial intelligence

[0001] The disclosed embodiments relate to a device and method for providing a medical chat service based on generative artificial intelligence. Specifically, the disclosed embodiments relate to a technology that uses generative artificial intelligence to provide answers to users' questions about medical records in a chat format.

[0002] [Cross-reference to related applications]

[0003] This application claims priority to Republic of Korea Provisional Patent Application No. 10-2023-0061952, filed May 12, 2023, and Republic of Korea Patent Application No. 10-2024-0048955, filed April 11, 2024, the entire contents of which are incorporated herein by reference.

[0004] [Description of the national research and development project that supported this invention]

[0005] This study was supported by [Project ID: 1711181132, Project Number: 2021R1F1A1060787, Ministry of Science and ICT, Project Management (Specialist) Organization: National Research Foundation of Korea, Research Project Name: Individual Basic Research (MSIT), Research Project Name: Development and Commercialization Verification of a Rapid Dementia Prediction Platform Using Healthcare Big Data, Project Implementing Organization: Seoul National University Bundang Hospital Medical Artificial Intelligence Center, Research Period: 20210601 ~ 20240229].

[0006] This research was supported by [Project ID: 1711160533, Project Number: 2022-0-00184-001, Ministry of Science and ICT, Project Management (Specialized) Organization: National Research Foundation of Korea, Research Project Name: Development of Core Source Technologies for Human-Centered Artificial Intelligence, Research Project Name: Research on the Development of Flexible and Evolving Artificial Intelligence Technology in Line with Increasingly Strengthened Ethical Policies, Project Implementing Organization: Korea Advanced Institute of Science and Technology, Research Period: 20220401 ~ 20261231].

[0007] The "wellcare" industry, combining healthcare and AI, is on the rise, and various medical AI solutions have been developed. However, due to regulations such as the Personal Information Protection Act, technological application is limited. Furthermore, the emergence of Chat GPT ushered in the era of massive AI. Fine-tuning massive generative AI, which understands human language, using datasets based on precise expertise is expected to yield higher performance than natural language processing algorithms trained solely on domain-specific knowledge.

[0008] Low health literacy among patients has recently become a growing issue. According to the Korea Institute for Health and Social Affairs' <Health and Welfare Forum>, "Health Literacy Status and Policy Implications as Seen from the Korean Medical Panel," three out of ten people report lacking health literacy. Health literacy tends to decline particularly among older adults and those with a vulnerable socioeconomic status. At university hospitals, outpatient consultations are extremely short, lasting about three minutes, making it difficult to receive detailed explanations from doctors.

[0009] Medical records are all records created when a patient visits a hospital, whether outpatient or inpatient, including surgical records, initial and follow-up records, admission / discharge records, and imaging test results. They are primarily issued for submission to insurance companies as supporting documentation or for use when transferring to another hospital. Patients can request copies of medical records directly, and they can be issued without physician approval. Recently, many hospitals have begun issuing copies of medical records online, and the number of such services is steadily increasing. Some hospitals are also launching mobile service offering copies of medical records.

[0010] Patients seek to interpret their medical records to understand their health. However, most medical records are written in a mixture of English and Korean and contain a wealth of medical terminology, making them difficult to understand. Furthermore, even when searching online, medical abbreviations are often difficult to find, making interpretation of medical records often impossible. Furthermore, existing natural language processing algorithms often fail to interpret medical records at all, demonstrating poor performance.

[0011] The disclosed embodiments are intended to provide answers to users' questions about medical records in a chat format using generative artificial intelligence.

[0012] A method for providing a generative artificial intelligence-based medical chat service according to one embodiment is provided, which is performed by a device including one or more processors; and a memory storing one or more programs executed by the one or more processors, the method comprising: generating a prompt based on a user input and a command; and inputting the prompt into a generative artificial intelligence model to output a corresponding response.

[0013] The user input may include at least one of text input and image input regarding medical records, including at least one of surgical records, test results, hospitalization records, discharge records, first visit records, re-visit records, nursing records, and prescriptions for humans or animals.

[0014] The command may include a first command, the prompt may include a first prompt, and the step of generating the prompt may include a step of generating the first prompt based on the user input and the first command, wherein the first command may be configured to request the generative artificial intelligence model to respond to the medical record included in the user input in a medically professional manner, excluding personal information included in the user input, wherein the personal information includes at least one of a patient name, a patient number, a date of birth, a hospital name, a hospital address, an attending physician, and an attending physician's phone number.

[0015] The command includes a second command, the prompt includes a second prompt, and the step of generating the prompt includes the step of generating the second prompt based on the user input and the second command, the second command being configured to assign a defined role and characteristic to each persona, the role being a doctor, a veterinarian, a nurse, a parent, a child, or a friend, and the characteristic corresponding to the role may include a professional attitude, an affectionate attitude, a sincere attitude, or a friendly attitude.

[0016] The command includes a third command, the prompt includes a third prompt, and the step of generating the prompt includes the step of generating a new user input by including a previous chat log exchanged between the user and the generative artificial intelligence model in the user input; and the step of generating the third prompt based on the new user input and the third command, wherein the third command can be configured to generate a question useful to the patient while maintaining the context of the previous chat log.

[0017] The above command includes a fourth command, the prompt includes a fourth prompt, and the step of generating the prompt includes the step of identifying whether the user input is a question; and if the user input is a question, the step of generating the fourth prompt based on the user input and the fourth command, wherein the fourth command may be configured to request the generative artificial intelligence model to output a predetermined amount of response to the question.

[0018] The fourth command is configured to request that the generative artificial intelligence model respond so that the response does not include a specific response, wherein the specific response may include a response suggesting a medical act or an act similar thereto and / or a response with duplicate content.

[0019] The method may include a step of detecting personal information from the user input; and, if the personal information is detected, a step of outputting a message and / or warning signal guiding exclusion of the personal information from the user input.

[0020] The step of generating the second prompt may include the step of displaying a menu on the display of the device from which an option of the persona can be selected; and the step of generating the second prompt based on the selected persona and the remaining user input excluding the selection of the option when a selection of the option of the persona is received.

[0021] The generative artificial intelligence model is trained by training data labeled with evaluation indices, and the training data can be constructed by labeling an evaluation indices for at least one of preference, accuracy, and possibility of dispute recorded by a medical professional for a predicted value output by the generative artificial intelligence model for input data including medical records.

[0022] The step of outputting the above response may include a step of outputting a complementary response corresponding to the prompt using a retrieval-augmented generation technique (RAG) from a vector database in which medical information is constructed as a corpus; and a step of providing a final response based on the response output using the generative artificial intelligence model and the complementary response output using the augmented generation technique.

[0023] A device for providing a generative artificial intelligence-based medical chat service according to one embodiment comprises: one or more processors; and a memory storing one or more programs executed by the one or more processors, wherein the one or more processors: generate a prompt based on a user input and a command, input the prompt into a generative artificial intelligence model, and output a corresponding response.

[0024] The disclosed embodiments utilize generative artificial intelligence to provide immediate answers to users' questions about their medical records.

[0025] The disclosed embodiments can provide consistency in answers by more explicitly rephrasing questions about a user's medical records through prompt engineering.

[0026] The disclosed embodiments can provide customized answers by applying a user-specified persona to the answer through prompt engineering.

[0027] FIG. 1 is a diagram illustrating a system that provides a generative artificial intelligence-based medical chat service according to one embodiment.

[0028] FIG. 2 is a block diagram illustrating a device that provides a generative artificial intelligence-based medical chat service according to one embodiment.

[0029] FIG. 3 is a diagram illustrating an example of a device providing a medical chat service based on generative artificial intelligence providing a screen for user input, according to one embodiment.

[0030] FIG. 4 is a diagram illustrating an example of a device providing a generative artificial intelligence-based medical chat service providing a warning signal to a user input, according to one embodiment.

[0031] FIG. 5 is a diagram illustrating an example in which, according to one embodiment, a device providing a medical chat service based on generative artificial intelligence applies a persona to provide an answer.

[0032] FIG. 6 is a diagram illustrating an example in which a device providing a generative artificial intelligence-based medical chat service proposes a recommended question in response to a user's question, according to one embodiment.

[0033] FIG. 7 is a diagram illustrating an example of a device providing a generative artificial intelligence-based medical chat service sending a chat record to an external device, according to one embodiment.

[0034] FIG. 8 is a diagram showing an example of a device providing a generative artificial intelligence-based medical chat service according to one embodiment, explaining the results of a comparison with a previous input.

[0035] FIG. 9 is a flowchart illustrating a method for providing a generative artificial intelligence-based medical chat service according to one embodiment.

[0036] Hereinafter, specific embodiments of one embodiment will be described with reference to the drawings. The following detailed description is provided to facilitate a comprehensive understanding of the invention described herein. However, this is merely an example and the invention is not limited thereto.

[0037] In describing certain embodiments, detailed descriptions of known technologies related to the present invention will be omitted if they are deemed to unnecessarily obscure the gist of the embodiments. Furthermore, numbers (e.g., "first," "second," etc.) used in the description of the embodiments are merely identifiers used to distinguish one component from another.

[0038] The terms described below are terms defined in consideration of their functions in the present invention, and may vary depending on the intention or custom of the user or operator. Therefore, their definitions should be made based on the contents throughout this specification. The terminology used in the detailed description is only for the purpose of describing one embodiment and should never be limited. Unless clearly used otherwise, the singular form includes the plural form. In this description, expressions such as "comprises" or "having" are intended to indicate certain components, numbers, steps, operations, elements, parts or combinations thereof, and should not be construed to exclude the presence or possibility of one or more other components, numbers, steps, operations, elements, parts or combinations thereof other than those described.

[0039] Furthermore, the embodiments described herein may have aspects that are entirely hardware, partially hardware and partially software, or entirely software. As used herein, the terms "unit," "apparatus," "module," "device," "server," or "system" refer to hardware, a combination of hardware and software, or a computer-related entity such as software. For example, a unit, apparatus, module, device, server, or system may refer to hardware that constitutes part or all of a platform and / or software such as an application for operating the hardware. As a specific example, a unit, apparatus, module, device, server, or system may be implemented by a processor.

[0040] Definition of terms

[0041] In this specification, user input may refer to information and / or requests provided by a user to interact with a generative AI model. In other words, user input may refer to the initial input provided by the user to the generative AI model prior to prompt generation.

[0042] In this specification, a command may refer to additional instructions incorporated into a prompt that enable a generative AI model to respond in a desired direction. In other words, a command may refer to additional instructions included in a prompt that enable a generative AI model to output desired information in a desired direction or in an optimized manner based on user input.

[0043] Description of Exemplary Embodiments of the Invention

[0044] FIG. 1 is a drawing for explaining a system (10) that provides a medical chat service based on generative artificial intelligence according to one embodiment.

[0045] Referring to FIG. 1, a system (10) providing a generative artificial intelligence-based medical chat service according to one embodiment may include a server (100) and a device (200).

[0046] The server (100) may provide a chat service to a device (200) using a generative artificial intelligence model through a communication network, and may represent a server (100) computer that manages the chat service as a whole.

[0047] The server (100) may receive a prompt containing input data and a command from the device (200). The input data may include at least one of text and images related to medical records. The server (100) may use a generative artificial intelligence model to provide the device (200) with an explanation corresponding to the analysis request for the input data. The server (100) may provide the explanation based on a chat format.

[0048] To facilitate understanding in interactions between the server (100) and the user, the device (200) may request the server (100) to process a prompt that adds a command to the input data. At this time, the device (200) may include a command in the prompt requesting a more concise explanation of the user input to facilitate understanding for the user. As another example, the device (200) may include a command in the prompt requesting the application of a persona to assign a role for the generative artificial intelligence model to play in the chat.

[0049] At this time, the device (200) can be implemented as various electronic devices such as a mobile phone, a tablet PC, a digital camera, a camcorder, a laptop computer, a tablet PC, a desktop, an e-book terminal, a digital broadcasting terminal, a PDA (Personal Digital Assistants), a PMP (Portable Multimedia Player), a navigation device, an MP3 player, a wearable device, etc.

[0050] Below, FIG. 2 describes a method for generating a prompt that enables a device (200) to more effectively interact with a generative artificial intelligence model and user input.

[0051] FIG. 2 is a block diagram illustrating a device (200) that provides a generative artificial intelligence-based medical chat service according to one embodiment.

[0052] Referring to FIG. 2, the device (200) includes a processor (210) and a memory (220).

[0053] The processor (210) generates prompts based on user input and commands.

[0054] Here, the user input may specifically include, as first input data, analysis target data for the medical record, and as second input data, an analysis request for the analysis target data.

[0055] For example, the first input data may be a medical record diagnosed by a healthcare professional, such as an image created by photographing the printed medical record. In another example, the first input data may include text typed by the patient, either written down or extracted from the issued medical record.

[0056] The second input data may include a user query requesting a medical analysis of the first input data from the generative AI model. The second input data may exclude any medical analysis requests that could be interpreted as medical practice.

[0057] For example, the second input data may include a request for a description of the first input data and / or an inquiry regarding possible progression of the current health condition and / or guidance for preventing and managing a pre-existing condition based on the first input data.

[0058] At this time, the medical record may include at least one of a surgical record for a person or animal, a test result, an admission record, a discharge record, an initial visit record, a re-visit record, a nursing record, and a prescription.

[0059] The command may include additional instructions requested of the generative AI model to facilitate effective communication between the user and the generative AI model. As a specific example, the command may include supplementary instructions for user input to enable the model to process potential user requests.

[0060] That is, the processor (210) can generate a prompt to be input into the generative artificial intelligence model by adding a command to the user input.

[0061] For example, the processor (210) can generate a prompt to be input into the generative artificial intelligence model based on user input and a first command.

[0062] Here, the first command may be configured to request the generative artificial intelligence model to medically respond to the medical record, excluding any personal information contained in the user input.

[0063] As a specific example, the first command might include a command such as, "Please respond to all questions except the hospital name, address, doctor's name, and phone number. Please provide answers to the questions."

[0064] Meanwhile, personal information may include at least one of the patient name, patient number, date of birth, hospital name, hospital address, attending physician, and attending physician's phone number, but this is exemplary and is not limited to the examples described above.

[0065] That is, the processor (210) can generate a prompt that complements what the user could potentially request (e.g., a request for personal information security, a request for compliance with personal information processing regulations, a request for focusing on key content) that did not appear in the user input.

[0066] As another example, the processor (210) may generate a prompt to input into the generative artificial intelligence model based on user input and a second command.

[0067] Here, the second command can be configured to ask the generative artificial intelligence model to respond based on the roles and characteristics defined for each persona.

[0068] At this time, the role may be a doctor, a veterinarian, a nurse, a parent, a child, or a friend, and the characteristics corresponding to the role may include a professional attitude, an affectionate attitude, a sincere attitude, or a friendly attitude.

[0069] As a concrete example, a second command might include something like, "Please respond professionally yet kindly, as if a doctor were speaking to a patient."

[0070] As another example, the processor (210) may generate prompts to input into the generative artificial intelligence model based on new user input and a third command.

[0071] Here, new user input may mean input data including chat history exchanged between previous user input and corresponding generative artificial intelligence model.

[0072] Here, the third command may be configured to request the generative artificial intelligence model to generate questions that would be useful to the patient, while maintaining the context of the previous chat history.

[0073] Here, useful questions may include questions the user might be curious about in relation to the request contained in the user input. In other words, useful questions may include questions the user would have asked if they had known about the user input.

[0074] For example, a useful question might include a question that helps the user understand the topic at hand, as opposed to a question included in the user input. Another example might include a question that helps resolve a question included in the user input.

[0075] As another example, the processor (210) may generate a prompt to input into the generative artificial intelligence model based on user input and the fourth command.

[0076] Here, the processor (210) can generate a fourth prompt based on the user input and the fourth command if the user input includes a question.

[0077] At this time, the fourth command may be configured to request the generative artificial intelligence model to respond to a question contained in the user input in a predefined amount.

[0078] The processor (210) may request that the response be, for example, one line long, to help the generative artificial intelligence model provide a clear and concise explanation to the user.

[0079] Additionally, the fourth command may be configured to request that the generative AI model not include specific responses in its responses. These specific responses may include responses suggesting medical procedures or similar procedures and / or responses containing redundant content. This allows the processor (210) to prevent the generative AI model from generating redundant responses. Furthermore, specific responses may be excluded to prevent the response from being interpreted as a medical procedure or similar procedure.

[0080] In addition, the processor (210) can provide additional supplementary answers by using a database in which vectors embedded with information in the medical domain are stored, as well as by using a generative artificial intelligence model.

[0081] At this time, the processor (210) can produce a supplementary answer through a retrieval-augmented generation technique (RAG) from the database.

[0082] That is, the processor (210) can provide a final response with improved medical accuracy based on a complementary response calculated using a database and a response calculated using a generative artificial intelligence model.

[0083] It is desirable for generative artificial intelligence models to be trained using labeled training data with evaluation metrics.

[0084] At this time, the learning data can be constructed by labeling an evaluation index for at least one of preference, accuracy, and possibility of dispute recorded by a medical professional for the predicted value output by the generative artificial intelligence model for sample data containing medical records.

[0085] For example, training data could be constructed such that a single medical record is labeled with a total of 18 labels by measuring the preference, accuracy, and likelihood of disagreement for one description of the medical record and five responses to questions posed about the medical record.

[0086] Here, preference can be assigned a class proportional to the order in which medical professionals deem the response most appropriate for the explanation. Accuracy can be assessed as a three-class index, assessing whether a response is medically valid, invalid, or contains other opinions. Controversy can be assessed on a scale of 0 to 10. Controversy can be assigned a high score to responses that lack consensus among medical professionals and contain diverse perspectives.

[0087] Meanwhile, accuracy and arguability are preferably evaluated based on the highest preference for explanations and responses output in a three-option multiple-choice format.

[0088] The memory (220) stores instructions executed by the processor (210).

[0089] FIG. 3 is a drawing showing an example of a device (200) that provides a medical chat service based on generative artificial intelligence, according to one embodiment, providing a screen for user input.

[0090] The processor (210) can transmit user input received through an application linked to the server (100) to the server (100).

[0091] The processor (210) may receive user input through the application first UI displayed on the display of the device (200). For example, the processor (210) may receive user input obtained through a text input field as the first UI. The processor (210) may receive user input through an image upload icon as the first UI.

[0092] When user input is completed, the processor (210) may send an analysis request or transmission request to the server (100) according to the user's request. For example, when the application second UI displayed on the display of the device (200) is touched by the user, the processor (210) may request the server (100) to analyze the medical records uploaded through the user input or request transmission to another external device (200).

[0093] FIG. 4 is a diagram showing an example in which a device (200) providing a generative artificial intelligence-based medical chat service provides a warning signal to a user input according to one embodiment.

[0094] The processor (210) may output a warning signal if the uploaded image contains inappropriate information.

[0095] As in the medical record image shown on the left side of Fig. 4, if there is personal information including at least one of name, ID, age, gender, and date of birth in the uploaded image, the processor (210) may, if it recognizes this, output a message as a warning signal to guide the exclusion of the personal information.

[0096] On the other hand, the processor (210) can output a completion signal if the image does not contain inappropriate information.

[0097] As in the medical record image shown on the right side of Fig. 4, when there is no personal information in the uploaded image, the processor (210) can provide information that the image upload was successful by outputting a completion signal.

[0098] At this time, the warning signal and / or guidance signal is described as being visually output on the display of the device (200), but this is exemplary, and may be output visually, audibly, or tactilely, and is not necessarily limited to the examples described above.

[0099] FIG. 5 is a diagram showing an example in which a device (200) providing a medical chat service based on generative artificial intelligence provides an answer by applying a persona according to one embodiment.

[0100] The processor (210) can provide a response to user input transmitted to the server (100) through an application linked to the server (100).

[0101] In FIG. 5, the processor (210) provides a response by applying the 'child' persona. At this time, the processor (210) provides a prompt requesting the user to provide a response using the 'child' persona previously selected by the user or the 'child' persona set in the setting environment, thereby providing a chat message in a 'true' tone.

[0102] At this time, the processor (210) may display a third UI for setting the persona on the display of the device (200) to select the persona to be applied to the next answer. At this time, the role and character of the persona may be guided on the display to help the user easily change the persona.

[0103] Meanwhile, when the third UI is touched by the user, the processor (210) can generate the next prompt by adding a command to the new user input to have the generative artificial intelligence model provide an answer by applying the newly selected persona.

[0104] FIG. 6 is a diagram showing an example in which a device (200) providing a generative artificial intelligence-based medical chat service proposes a recommended question in response to a user's question, according to one embodiment.

[0105] In FIG. 6, the processor (210) suggests useful questions to the user. At this time, the processor (210) may suggest useful questions to the user based on the conversation the user had with the generative artificial intelligence model.

[0106] Specifically, the processor (210) can recommend useful questions to the user while maintaining the context of the previous conversation the user had with the generative artificial intelligence model. In this example, the user asks a question about "radiorefractory," and the processor (210) can then induce new questions to inform the user of "where in the body the radiorefractory is located" and "whether the condition of the radiorefractory is serious."

[0107] That is, the processor (210) can provide the next answer by maintaining the topic, volume, scope, background information, purpose, goal, relevance, the user's prior knowledge, experience, emotional context, cultural context, etc. of the conversation the user had with the generative artificial intelligence model.

[0108] FIG. 7 is a diagram showing an example of a device (200) providing a generative artificial intelligence-based medical chat service sending a chat record to an external device (200) according to one embodiment.

[0109] The processor (210) can transmit chat records to an external device (200) through an application.

[0110] First, the processor (210) can display a list of chat records that can be transmitted to an external device (200) on the display of the device (200). Thereafter, the processor (210) can transmit the chat records selected by the user to be transmitted externally to the external device (200). When data transmission is completed, the processor (210) can display a confirmation message indicating that the transmission was successful or an error message indicating that the transmission failed on the display.

[0111] FIG. 8 is a diagram showing an example of a device (200) providing a medical chat service based on generative artificial intelligence, according to one embodiment, explaining the results of a comparison with a previous input.

[0112] The processor (210) may provide the server (100) with a query that compares newly uploaded medical records with previously uploaded medical records to process a user query requesting a comparison. At this time, the processor (210) may request the server (100) to compare records with a desired date by storing the uploaded image and the upload date together with metadata.

[0113] The processor (210) may provide the server (100) with a command requesting a change in the size or condition of the diseased area as an item for comparison in response to the user query. Meanwhile, the processor (210) may provide the server (100) with a prompt that includes a command requesting additional recommendations in addition to the user query.

[0114] FIG. 9 is a flowchart illustrating a method for providing a generative artificial intelligence-based medical chat service according to one embodiment.

[0115] The method of FIG. 9 can be performed by a device (200) that provides a medical chat service based on a generative artificial intelligence model described in FIG. 2.

[0116] First, a device (200) providing a medical chat service based on an artificial intelligence model generates a prompt (910) based on user input and predefined commands.

[0117] Thereafter, a device (200) providing a medical chat service based on an artificial intelligence model inputs a prompt into a generative artificial intelligence model and outputs a corresponding response (920).

[0118] Meanwhile, in FIG. 1, the device (200) is described as providing a chat service by communicating with the generative artificial intelligence model of the server (100), but, unlike the description, the device (200) may also provide a chat service on its own through the generative artificial intelligence model built into the device (200).

[0119] The above-described method 9 has been described with reference to the sequence shown in the drawings. For purposes of illustration, the method has been depicted and described as a series of blocks; however, the present invention is not limited to the sequence of the blocks, and some blocks may occur in a different order or concurrently with other blocks than depicted and described herein, and various other branches, flow paths, and sequences of blocks may be implemented that achieve the same or similar results. Furthermore, not all depicted blocks may be required to implement the method described herein.

[0120] Meanwhile, embodiments of the present invention may include a program for performing the methods described herein on a computer, and a computer-readable recording medium including the program. The computer-readable recording medium may include program commands, local data files, local data structures, etc., alone or in combination. The medium may be specially designed and configured for the present invention, or may be one commonly used in the field of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and hardware devices specially configured to store and execute program commands such as ROMs, RAMs, and flash memories. Examples of the program may include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc.

[0121] While representative embodiments of the present invention have been described in detail above, those skilled in the art will appreciate that various modifications to the above-described embodiments are possible without departing from the scope of the present invention. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined not only by the claims set forth below but also by equivalents thereof.

[0122] A device and method for providing a medical chat service based on generative artificial intelligence according to one embodiment provides prompt engineering to help a user effectively use a generative artificial intelligence model with medical information as a domain, and is applicable to the digital medical industry.

Claims

1. One or more processors; and A method for providing a generative artificial intelligence-based medical chat service performed by a device including a memory storing one or more programs executed by one or more processors, generating a prompt based on user input and commands; and A method for providing a medical chat service based on generative artificial intelligence, comprising the step of inputting the above prompt into a generative artificial intelligence model and outputting a corresponding response.

2. In paragraph 1, A method for providing a generative artificial intelligence-based medical chat service, wherein the user input includes at least one of text input and image input regarding medical records including at least one of surgical records, test results, hospitalization records, discharge records, first-visit records, re-visit records, nursing records, and prescriptions for humans or animals.

3. In paragraph 1, The above command includes the first command, The above prompt includes the first prompt, The steps to create the above prompt are: A step of generating the first prompt based on the user input and the first command, A method for providing a medical chat service based on generative artificial intelligence, wherein the first command is configured to request the generative artificial intelligence model to respond to the medical records included in the user input in a medically professional manner, excluding personal information included in the user input, wherein the personal information includes at least one of patient name, patient number, date of birth, hospital name, hospital address, attending physician, and attending physician phone number.

4. In paragraph 1, The above command includes a second command, The above prompt includes a second prompt, The steps to create the above prompt are: A step of generating the second prompt based on the user input and the second command, A method for providing a generative artificial intelligence-based medical chat service, wherein the second command is configured to assign a defined role and characteristic to each persona, wherein the role is a doctor, a veterinarian, a nurse, a parent, a child, or a friend, and the characteristic corresponding to the role includes a professional attitude, an affectionate attitude, a sincere attitude, or a friendly attitude.

5. In paragraph 1, The above command includes the third command, The above prompt includes a third prompt, The steps to create the above prompt are: A step of generating a new user input by including a previous chat history exchanged between the user and the generative artificial intelligence model in the above user input; generating the third prompt based on the new user input and the third command; The third command is a method for providing a generative artificial intelligence-based medical chat service configured to maintain the context of the previous chat history and generate useful questions for the patient.

6. In paragraph 1, The above command includes the fourth command, The above prompt includes the fourth prompt, The steps to create the above prompt are: a step of identifying whether the user input is a question; and If the user input is a question, the step of generating the fourth prompt based on the user input and the fourth command, A method for providing a medical chat service based on generative artificial intelligence, wherein the fourth command is configured to request the generative artificial intelligence model to output a response of a predetermined amount in response to the question.

7. In paragraph 6, The above fourth command is, The above generative artificial intelligence model is configured to request that a specific response not be included in the above response, A method for providing a generative artificial intelligence-based medical chat service, wherein the above specific response includes a response suggesting a medical act or an act equivalent thereto and / or a response with duplicate content.

8. In paragraph 1, The above method, A step of detecting personal information from the above user input; and A method for providing a generative artificial intelligence-based medical chat service, comprising the step of outputting a message and / or warning signal guiding the user to exclude the personal information from the user input when the personal information is detected.

9. In paragraph 4, The step of generating the second prompt is: A step of displaying a menu on the display of the device for selecting options of the persona; and A method for providing a generative artificial intelligence-based medical chat service, comprising the step of generating the second prompt based on the selected persona and the remaining user input excluding the selection for the option, upon receiving a selection for the option of the persona.

10. In paragraph 1, The above generative artificial intelligence model is trained by labeled learning data with evaluation indicators, A method for providing a generative artificial intelligence-based medical chat service, wherein the above learning data is constructed by labeling an evaluation index for at least one of preference, accuracy, and possibility of dispute recorded by a medical professional for a predicted value output by a generative artificial intelligence model for input data including medical records.

11. In paragraph 1, The step of outputting the above response is: A step of outputting a complementary response corresponding to the prompt using a retrieval-augmented generation technique (RAG) in a vector database where medical information is constructed as a corpus; and A method for providing a medical chat service based on generative artificial intelligence, comprising a step of providing a final response based on a response output using the generative artificial intelligence model and a supplementary response output using the augmented generation technique.

12. One or more processors; and A device providing a generative artificial intelligence-based medical chat service including a memory storing one or more programs executed by the one or more processors, One or more of the above processors: Generate prompts based on user input and commands, A device providing a medical chat service based on generative artificial intelligence, which inputs the above prompt into a generative artificial intelligence model and outputs a corresponding response.

Citation Information

Patent Citations

  • Intelligent triage and medical question answering method and device based on GPT-2 model

    CN113889249A

  • User interface management method and system

    KR1020150040245A

  • Computer-implemented system and method for creating generative medicines for dementia and for generating tailored medical recipes for mental health disorders

    WO2022003049A1