Sentence generation apparatus, method for generating sentence, and program

Abstract

To provide a sentence generation apparatus capable of easily creating a summary sentence of a medical record of a patient.SOLUTION: The sentence generation apparatus of the present disclosure includes: display control means for displaying icons indicating items to be included in a medical text for each type of the medical text; reception means for receiving the designation of any of the icons of items for generating a summary text; generation means for generating the summary text on the basis of electronic medical record information by inputting an instruction sentence corresponding to the designated icon to a language model optimized by machine learning; and output means for outputting the generated summary text.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a text generation device, a text generation method, and a program. [Background technology]

[0002] There are technologies to reduce the burden of preparing documents for doctors and other medical professionals.

[0003] For example, Patent Document 1 discloses a method of extracting written expressions contained in medical information, interpreting the extracted written expressions as medical facts, extracting definite diseases that have been definitively diagnosed from the medical information, and creating an explanation of the diagnostic basis for the definite diseases that have been definitively diagnosed based on the medical facts. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2015-138402 Summary of the Invention [Problem to be solved by the invention]

[0005] Doctors and other medical professionals use electronic medical record information to create medical record summaries to record the details of medical treatment for patients. However, creating medical record summaries requires time to search for the necessary text in the electronic medical record information and to summarize the text. In addition, the quality and impression of summaries vary depending on the creator.

[0006] An example of an objective of the present disclosure is to provide a text generation device or the like that can easily create a summary of a patient's medical record. [Means for solving the problem]

[0007] A text generation device in one aspect of the present disclosure includes a display control means for displaying icons indicating items to be included in the medical text for each type of medical text, a reception means for receiving a designation of one of the icons for the items for which a summary is to be generated, a generation means for generating a summary based on electronic medical record information by inputting an instruction sentence corresponding to the designated icon into a language model, and an output means for outputting the generated summary.

[0008] In one aspect of the present disclosure, a text generation method involves a computer displaying icons indicating items to be included in each type of medical text, accepting a specification of one of the icons for the items for which a summary is to be generated, and inputting an instruction sentence corresponding to the specified icon into a language model, thereby generating a summary based on electronic medical record information and outputting the generated summary.

[0009] A program in one aspect of the present disclosure causes a computer to display, for each type of medical text, icons indicating items to be included in the medical text, accept specification of one of the icons for the items for which a summary sentence is to be generated, input an instruction sentence corresponding to the specified icon into a language model, thereby generating a summary sentence based on electronic medical record information, and outputting the generated summary sentence. [Effects of the Invention]

[0010] One example of the benefits of the present disclosure is that it allows for easy creation of summaries of patient medical records. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a block diagram showing the configuration of a sentence generation device according to the present disclosure. [Figure 2] FIG. 2 is a diagram showing a hardware configuration in which the text generation device according to the present disclosure is realized by a computer device and its peripheral devices. [Figure 3] FIG. 3 is a diagram showing an example of a summary generation screen in the present disclosure. [Figure 4] FIG. 4 is a diagram showing an example of a summary generation screen in the present disclosure. [Figure 5] FIG. 5 is a flowchart showing the operation of sentence generation in the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, with reference to the drawings, embodiments of a text generation device, a text generation method, a program, and a non-transitory recording medium for recording a program according to the present disclosure will be described in detail. The present embodiments do not limit the disclosed technology.

[0013] [First embodiment] 1 is a block diagram showing the configuration of a text generation device 100 according to the present disclosure. As shown in FIG. 1, the text generation device 100 includes a display control unit 101, a receiving unit 102, a generation unit 103, and an output unit 104.

[0014] The text generation device 100 is a device for generating medical text that summarizes a patient's medical records, etc. An example of a medical text is a medical summary that summarizes the diagnosis, symptoms, findings, treatment details, nursing care details, and treatment progress, etc. Examples of summaries include discharge summaries that are created when a patient is transferred to another department or discharged from the hospital, as well as weekly summaries and monthly summaries that are created at predetermined intervals. However, the medical text of the present disclosure is not limited to these texts as long as it is a medical text that summarizes a patient's medical records, etc.

[0015] Fig. 2 is a diagram showing an example of a hardware configuration in which the sentence generation device 100 according to the present disclosure is realized by a computer device 500 including a processor. As shown in Fig. 2, the sentence generation device 100 includes a CPU (Central Processing Unit) 501, memories such as a ROM (Read Only Memory) 502 and a RAM (Random Access Memory) 503, a storage device 505 such as a hard disk for storing a program 504, a communication I / F (Interface) 508 for network connection, and an input / output interface 511 for inputting and outputting data.

[0016] The CPU 501 runs an operating system to control the entire text generation device 100 according to the present disclosure. The CPU 501 also reads programs and data into memory from a recording medium 506 attached to a drive device 507, for example. The CPU 501 also functions as the display control unit 101, the reception unit 102, the generation unit 103, and the output unit 104 according to the present disclosure, or as part of these units, and executes processing or commands in the flowchart shown in FIG. 5, which will be described later, based on the program.

[0017] The recording medium 506 is, for example, an optical disk, a flexible disk, a magneto-optical disk, an external hard disk, or a semiconductor memory. Some of the recording media in the storage device are non-volatile storage devices, and the programs are recorded therein. The programs may also be downloaded from an external computer (not shown) connected to a communication network.

[0018] The input device 509 is realized by, for example, a mouse, a keyboard, built-in key buttons, etc., and is used for input operations. The input device 509 is not limited to a mouse, a keyboard, or built-in key buttons, and may be, for example, a touch panel. The output device 510 is realized by, for example, a display, and is used to check output.

[0019] As described above, the text generation device 100 shown in FIG. 1 is realized by the computer hardware shown in FIG. 2. However, the means for realizing each part of the text generation device 100 in FIG. 1 is not limited to the configuration described above. Furthermore, the text generation device 100 may be realized by a single physically coupled device, or may be realized by two or more physically separated devices connected by wire or wirelessly. For example, the input device 509 and the output device 510 may be connected to the computer device 500 via a network. Furthermore, the text generation device 100 shown in FIG. 1 can also be configured by cloud computing or the like.

[0020] The display control unit 101 is a means for displaying icons indicating items to be included in each type of medical text. For example, the display control unit 101 displays the icons on a display screen displaying an application program for displaying medical text. The items are headings for categorizing the contents of the medical text, such as basic patient information, surgical history, hospitalization history, injury / illness information, and prescription history. Each icon is associated with an instruction to be input to the language model. The instruction includes instructions for acquiring electronic medical record information required to create a summary for each item and generating a summary of the acquired electronic medical record information. The instruction may also include assigning a specific role to the language model, such as an internist or a surgeon. In this case, a summary tailored to the assigned role can be generated. The instruction is, for example, a prompt used as input to a large-scale language model, but is not limited to a prompt. An example of a prompt in the present disclosure is, "You are an internist. Please create this month's hospitalization record for patient XX. Include the chief complaint and treatment history in the hospitalization record."

[0021] The prompt may include generating a summary to satisfy predetermined generation conditions. Examples of the generation conditions include conditions regarding the content and format to be included in the summary, such as not including speculative content, using polite language such as "desu" and "masu," and listing in bullet points. Furthermore, the receiving unit 102 may receive a deadline for generating a summary when a discharge summary or other document must be created within a predetermined day after discharge.

[0022] The types of medical documents vary depending on the purpose, the author, and the person to whom they are addressed. For example, a nursing summary is a document written by a nurse and addressed to the medical staff at the receiving hospital when a patient is transferred or discharged from another hospital or changes ward, in order to ensure smooth continuation of nursing care at the next hospital. Other types of medical documents include documents written by doctors to medical staff, and documents written by doctors or nurses to patients.

[0023] The receiving unit 102 is a means for receiving a designation of any of the icons of the items for which a summary is to be generated. The receiving unit 102 receives the designation of the icon through an on-screen operation by the user on the screen of an application program for generating a summary. The number of icons received by the receiving unit 102 may be one or more. The receiving unit 102 may also receive an addition of an icon through an on-screen operation by the user. An instruction statement corresponding to the item is also linked to the added icon.

[0024] The generation unit 103 is a means for generating a summary based on the electronic medical record information by inputting an instruction sentence corresponding to the received icon into a language model. The generation unit 103 uses the language model to acquire text information related to the item indicated by the icon from the electronic medical record information and summarizes the acquired electronic medical record information to generate a summary. The generation unit 103 may, for example, input the instruction sentence associated with the icon into a language model stored in the storage device 505, or may input the instruction sentence associated with the icon into a language model stored in an external system. If a generation deadline is set, the generation unit 103 generates the summary so as to meet the deadline. The generation unit 103 may also generate the summary during the night when normal business operations are not being performed.

[0025] The language model performs processing according to a prompt corresponding to the received icon, for example, and outputs a summary. The language model may be a large-scale language model trained with a large amount of text data, or a trained model obtained by transfer learning of the large-scale language model. For example, GPT-2 (Generative Pre-Training-2), GPT-3, or GPT-4 may be used as the large-scale language model. Furthermore, T5 (Text-to-Text Transfer Transformer), BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly optimized BERT approach), or ELECTRA (Efficiently Learning an Encoder that Classifies Token Replacements Accurately) may be used as the large-scale language model.

[0026] The generation unit 103 uses a language model to execute a process of generating a summary, for example, using the following procedure. That is, the generation unit 103 extracts the patient's electronic medical record information required for generating the summary from a server device or the like of a medical institution, and provides this information to the language model together with a prompt to generate the summary. The generation unit 103 may also provide the patient's ID or the like as a prompt to the language model, thereby enabling reference to the patient's electronic medical record information and generating the summary.

[0027] The language model may generate a summary using a template sentence prepared in advance. Template sentences are prepared, for example, for each type of medical text and stored, for example, in the storage device 505. The generation unit 103 generates a summary by providing the template sentence and electronic medical record information to the language model. The template sentence is data containing text from the summary sentence, excluding patient-specific information. Here, the specific information is information specific to each patient, including personal information such as name, date of birth, and electronic medical record information of the patient. By generating a summary using the template sentence, it is possible to generate a summary in a format that is standardized by doctors or medical institutions.

[0028] The output unit 104 is a means for outputting the generated summary to a display device such as a display. For example, the output unit 104 outputs the summary on the screen of an application program for creating the summary. The output unit 104 may also transmit the generated summary to a terminal device used by the user, or may output the summary in response to a trigger such as the user's operation to generate a summary of the target patient.

[0029] The output unit 104 may display annotations for technical terms included in the summary. Technical terms may be stored, for example, in the storage device 505 together with their annotations, and the output unit 104 may compare the technical terms with the terms stored in the storage device 505 and display the annotations associated with these terms. This allows the recipient to understand the content of the summary even when the summary is generated for medical institution staff other than doctors who do not have knowledge of technical terms.

[0030] FIG. 3 is an example of a summary generation screen according to the present disclosure. The example in FIG. 3 is a screen for creating a nursing summary created by a nurse. As shown in FIG. 3, the screen includes icons 31 indicating items to be included in the medical text and a display 32 including a button for selecting whether or not to generate a summary for this patient. In the example in FIG. 3, the receiving unit 102 receives designation of one of the icons 31 for items for which a summary is to be generated on the summary generation screen. As shown in FIG. 3, the designated icon 31 may be highlighted by being displayed in bold or in a different color.

[0031] The output unit 104 may also display electronic medical record information that is the basis for the information contained in the summary. The information contained in the summary includes the patient's symptom progression, test results, treatment progression, prescribed medication information, and the doctor's findings on the patient, and is the electronic medical record information used by the generation unit 103 when generating the summary. The output unit 104 may, for example, display a link to the cited electronic medical record information contained in the summary, making it possible to output the electronic medical record information. The output unit 104 may also output information that identifies the cited electronic medical record information. The information that identifies the electronic medical record information is, for example, information such as the date and time of input of the electronic medical record information or the person who input it.

[0032] FIG. 4 is an example of a summary generation screen according to the present disclosure. The example of FIG. 4 shows a screen on which a summary 41 is displayed for each item, and the summary of the received item "Symptom Description" is displayed. Also, in the example of FIG. 4, by selecting icon 42, the user can edit the summary on the screen. Also, in the example of FIG. 4, the basis 43 of the electronic medical record information is displayed. The output example of the summary in FIG. 4 is an example of the present disclosure, and the summary may be output by other methods.

[0033] The operation of the sentence generation device 100 configured as above will be described with reference to the flowchart of FIG.

[0034] 5 is a flowchart showing an outline of the operation of the sentence generation device 100 according to the present disclosure. The processing according to this flowchart may be executed based on program control by the processor described above. The processing according to this flowchart is triggered by a user's operation to create a summary.

[0035] As shown in FIG. 5, first, the display control unit 101 displays icons indicating items to be included in the medical text for each type of medical text (step S101). Next, the receiving unit 102 receives a designation of one of the icons of the items for which a summary is to be generated (step S102). The generating unit 103 generates a summary based on the electronic medical record information by inputting an instruction sentence corresponding to the designated icon into the language model (step S103). Finally, the output unit 104 outputs the generated summary (step S104). With this, the text generation device 100 completes the text generation operation.

[0036] In the text generation device 100, the generation unit 103 generates a summary based on electronic medical record information by inputting an instruction sentence corresponding to a specified icon into a language model. This allows a medical professional, who is a user, to generate a summary by performing an operation for generating a summary. This makes it easy to create a summary of a patient's medical record.

[0037] Modifications of the first embodiment described above will be described. In the present disclosure, the modifications can be implemented in any combination.

[0038] [Modification 1 of the first embodiment] In the first embodiment, the text generation device 100 may generate a summary using a sample text of a summary prepared in advance. The sample text is a sample of a summary used in the past for the same item, prepared for each doctor or each medical department, for example, and stored in the storage device 505. The reception unit 102 receives an instruction including, as a generation condition, that a summary be generated using the sample text. The generation unit 103 generates a summary by inputting the instruction and the sample text into a language model. By generating a summary using the sample text, it is possible to generate a summary in a format that is familiar to users such as doctors.

[0039] [Modification 2 of the First Embodiment] The receiving unit 102 may further receive a response regarding whether or not to adopt the output summary. For example, the receiving unit 102 receives a response regarding whether or not to adopt the output summary from the user on the screen on which the summary is output. Furthermore, when the receiving unit 102 receives a response regarding not adopting the output summary, the receiving unit 102 may further receive a response regarding generating a summary with different generation conditions. In this case, the receiving unit 102 may receive at least one of the following conditions for regenerating the summary: the length of the sentence, the degree of summarization, and electronic medical record information to be included in the summary. The length of the sentence is indicated, for example, by the number of characters or the number of lines. The degree of summarization indicates, for example, the extent to which the sentence is summarized relative to the length of the original sentence. When the receiving unit 102 receives a response regarding regeneration of the summary, the receiving unit 102 updates the prompt by including the received generation conditions in the prompt. Then, the generating unit 103 inputs the updated prompt into the language model to regenerate the summary.

[0040] In this modification, the receiving unit 102 receives a request for whether or not to adopt the output summary. This allows the user to confirm the content of the summary. If the receiving unit 102 receives a request not to adopt the output summary, it also receives a request to generate a summary under different generation conditions. This allows a summary that reflects the user's intention to revise the summary can be regenerated.

[0041] [Modification 3 of the First Embodiment] In each embodiment of the present disclosure, the generation unit 103 may generate summaries using a plurality of different trained models, and the output unit 104 may output the summaries generated by each of the trained models. In this case, the reception unit 102 may receive a request as to which model the summaries generated by should be adopted. As the plurality of trained models, for example, a large-scale language model that is publicly available as open source and a language model that uses a proprietary engine may be used. In this case, the user can select the better summary.

[0042] Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

[0043] For example, although multiple operations are described in a sequential order in the form of a flowchart, the order of description does not limit the order in which the multiple operations are performed. Therefore, when implementing each embodiment, the order of the multiple operations can be changed within the scope that does not affect the content.

[0044] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes.

[0045] (Appendix 1) a display control means for displaying icons indicating items to be included in the medical text for each type of medical text; a receiving means for receiving a designation of any of the icons of the items for which a summary is to be generated; a generation means for generating the summary based on electronic medical record information by inputting an instruction sentence corresponding to a designated icon into a language model; and an output means for outputting the generated summary.

[0046] (Appendix 2) The text generation device described in Appendix 1, wherein the generation means uses the language model to obtain text information related to the items indicated by each icon from electronic medical record information, and summarizes the obtained electronic medical record information to generate the summary sentence.

[0047] (Appendix 3) 2. The text generation device according to claim 1, wherein the output means displays the generated summary in an editable manner.

[0048] (Appendix 4) The text generation device according to claim 1, wherein the output means displays electronic medical record information that is the basis for the information contained in the summary.

[0049] (Appendix 5) The receiving means further receives an instruction including, as a generation condition, that the summary sentence be generated using a template sentence prepared in advance; The text generation device described in Appendix 1, wherein the generation means inputs the instruction sentence and the template text into a language model and generates the summary text by applying electronic medical record information to the template text.

[0050] (Appendix 6) The receiving means further receives an instruction including, as a generation condition, that the summary be generated using a sample sentence; 2. The text generation device according to claim 1, wherein the generation means generates the summary by inputting the instruction sentence and the sample sentence into a language model.

[0051] (Appendix 7) 2. The text generation device according to claim 1, wherein the accepting means accepts an indication of whether or not the output summary sentence should be adopted.

[0052] (Appendix 8) the generating means generates each of the summaries using a plurality of different language models; the output means outputs the generated summaries, 2. The text generation device according to claim 1, wherein the accepting means accepts whether a text generated by a language model is to be adopted.

[0053] (Appendix 9) The computer For each type of medical text, an icon indicating an item to be included in the medical text is displayed; Accept the designation of one of the icons of the items for which a summary is to be generated, generating the summary sentence based on the electronic medical record information by inputting the instruction sentence corresponding to the designated icon into a language model; and outputting the generated summary.

[0054] (Appendix 10) For each type of medical text, an icon indicating an item to be included in the medical text is displayed; Accept the designation of one of the icons of the items for which a summary is to be generated, generating the summary sentence based on the electronic medical record information by inputting the instruction sentence corresponding to the designated icon into a language model; and outputting the generated summary.

[0055] Some or all of the configurations described in Supplements 2 to 8 that are dependent on Supplement 1 above may also be dependent on Supplement 9 and Supplement 10 in the same dependent relationship as Supplements 2 to 8. Not limited to Supplements 1, 9, and 10, some or all of the configurations described as Supplements may also be dependent on various hardware, software, various recording devices for recording software, or systems, within the scope of each of the above-mentioned embodiments. [Explanation of symbols]

[0056] 100 Sentence generator 101 Display control unit 102 Reception 103 Generation part 104 Output section 500 Computer equipment 501 CPU 502 ROM 503 RAM 504 Program 505 Storage device 506 Recording Media 507 Drive Device 508 Communication Interface 511 Input / Output Interface 512 Bus

Claims

1. a display control means for displaying icons indicating items to be included in the medical text for each type of medical text; a receiving means for receiving a designation of any of the icons of the items for which a summary is to be generated; a generation means for generating the summary based on electronic medical record information by inputting an instruction sentence corresponding to a designated icon into a language model; and an output means for outputting the generated summary.

2. The text generation device according to claim 1, wherein the generation means uses the language model to obtain text information related to the items indicated by each icon from electronic medical record information, and summarizes the obtained electronic medical record information to generate the summary sentence.

3. 2. The text generation device according to claim 1, wherein said output means displays the generated summary in an editable manner.

4. The text generation device according to claim 1 , wherein the output means displays electronic medical record information that is the basis for the information contained in the summary.

5. The receiving means further receives an instruction including, as a generation condition, that the summary sentence be generated using a template sentence prepared in advance; The text generation device according to claim 1 , wherein the generation means inputs the instruction sentence and the template text into a language model and generates the summary text by applying electronic medical record information to the template text.

6. The receiving means further receives an instruction including, as a generation condition, that the summary be generated using a sample sentence; The text generation device according to claim 1 , wherein the generation means generates the summary by inputting the instruction sentence and the sample sentence into a language model.

7. 2. The text generation device according to claim 1, wherein said accepting means accepts an indication of whether or not the output summary sentence is to be adopted.

8. the generating means generates each of the summaries using a plurality of different language models; the output means outputs the generated summaries, 2. The text generation device according to claim 1, wherein said accepting means accepts a request as to which language model the text generated by is to be adopted.

9. The computer For each type of medical text, an icon indicating an item to be included in the medical text is displayed; Accept the designation of one of the icons of the items for which a summary is to be generated, generating the summary sentence based on the electronic medical record information by inputting the instruction sentence corresponding to the designated icon into a language model; and outputting the generated summary.

10. For each type of medical text, an icon indicating an item to be included in the medical text is displayed; Accept the designation of one of the icons of the items for which a summary is to be generated, generating the summary sentence based on the electronic medical record information by inputting the instruction sentence corresponding to the designated icon into a language model; and outputting the generated summary.

Images