Information processing device, examination support method, and examination support program

Abstract

This information processing device comprises: a reception unit that receives input of a text describing subject's medical examination results in a natural language; an estimation unit that, on the basis of the text, estimates an examination item the subject should undergo, using a language model trained on the natural language by machine learning; and a presentation unit that presents the examination item estimated by the estimation unit. The presented examination item can be used for decision making or the like by a medical worker.

Description

Information processing device, inspection support method, and inspection support program

[0001] The present disclosure relates to an information processing device, an inspection support method, and an inspection support program.

[0002] In light of the social issue of overwork among medical professionals, attempts are being made to utilize information technology to provide medical support. For example, Patent Document 1 below discloses an invention relating to a test presentation device that selects and presents recommended tests for a test subject. More specifically, the test presentation device uses the test subject's medical record data to detect tests taken by test subjects with the same or similar medical history as the test subject, and presents these as recommended tests.

[0003] International Publication No. 2021 / 246084

[0004] Here, the medical record may contain doctor's findings and the like written in natural language. However, Patent Document 1 does not mention medical record data including natural language descriptions, and therefore does not describe how to process natural language descriptions. An exemplary objective of the present disclosure is to provide a technology that enables the presentation of appropriate test items according to medical results written in natural language.

[0005] An information processing device according to an exemplary aspect of the present disclosure includes a receiving means for receiving input of text in which the medical examination results of a subject are written in natural language, an estimating means for estimating the test items that the subject should undergo based on the text using a language model that has undergone machine learning of natural language, and a presenting means for presenting the test items estimated by the estimating means.

[0006] An examination assistance method according to an exemplary aspect of the present disclosure includes a reception process in which at least one processor receives input of text in which the medical examination results of a subject are described in natural language; an estimation process in which, based on the text, a language model that has been machine-learned to learn natural language is used to estimate the examination items that the subject should undergo; and a presentation process in which the examination items estimated by the estimation process are presented.

[0007] An examination assistance program according to an exemplary aspect of the present disclosure causes a computer to function as: a receiving means for receiving input of text in which the medical examination results of a subject are written in natural language; an estimating means for estimating the examination items that the subject should undergo based on the text using a language model that has undergone machine learning of natural language; and a presenting means for presenting the examination items estimated by the estimating means.

[0008] According to one exemplary aspect of the present disclosure, an exemplary effect is achieved in that a technology can be provided that makes it possible to present appropriate test items according to medical results described in natural language.

[0009] FIG. 1 is a block diagram showing a configuration of an information processing device according to the present disclosure. FIG. 2 is a flow diagram showing a flow of an examination support method according to the present disclosure. FIG. 3 is a diagram showing an overview of an examination support system according to the present disclosure. FIG. 4 is a block diagram showing a configuration of another information processing device according to the present disclosure. FIG. 5 is a diagram showing an example of a template used to generate a query. FIG. 6 is a flow diagram showing a flow of processing executed by the information processing device shown in FIG. 4. FIG. 7 is a block diagram showing a configuration of a computer that functions as an information processing device according to the present disclosure.

[0010] The following are examples of embodiments of the present invention. However, the present invention is not limited to the exemplary embodiments shown below, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technical means employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, embodiments obtained by appropriately omitting some of the technical means employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, the effects mentioned in the exemplary embodiments shown below are examples of effects expected in the exemplary embodiments, and do not define the scope of the present invention. In other words, embodiments that do not exhibit the effects mentioned in the exemplary embodiments shown below may also be included in the scope of the present invention.

[0011] [First Exemplary Embodiment] A first exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. This exemplary embodiment is a basic form for each of the exemplary embodiments described below. Note that the scope of application of each technical means employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise. Furthermore, each technical means shown in the drawings referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise.

[0012] (Configuration of information processing device 1) The configuration of the information processing device 1 will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the information processing device 1. As will be described in detail below, the information processing device 1 has a function of estimating the test items that a subject should undergo from text in which the subject's medical examination results are described in natural language, and presenting the estimated test items. As shown in Fig. 1, the information processing device 1 has a reception unit 101, an estimation unit 102, and a presentation unit 103.

[0013] The reception unit 101 receives input of text in which the medical results of the subject are described in a natural language. For example, the reception unit 101 may receive input of an electronic medical record in which the medical results of the subject are described in a natural language. In this case, the electronic medical record may be in a format in which the medical results are input for each input item, or in a format in which the medical results are freely described, or may include both of these formats.

[0014] Here, "medical care" refers to either an examination or treatment, or both. The medical care results may include the subject's most recent medical care results, or may include previous medical care results. Furthermore, a "subject" is a person who is the subject of a test. For example, a patient suffering from a disease or a person considering undergoing a test for the prevention or early detection of a disease may be considered a "subject."

[0015] The estimation unit 102 estimates the test items that the subject should undergo, based on the text input received by the receiving unit 101, using a language model that has been machine-learned from natural language. Here, the "language model" refers to a language model constructed by machine learning using a large amount of data to determine the arrangement of components (such as words) in a sentence expressed in natural language or the arrangement of sentences in a piece of writing. The language model may be recorded inside the information processing device 1 or may be recorded outside the information processing device 1.

[0016] For example, the language model may be one that has been machine-learned to generate answers to questions written in natural language (so-called generative AI (Artificial Intelligence)). In this case, the estimation unit 102 may generate a query asking about the test items that the subject should undergo from the text received by the receiving unit 101, and input the query to the language model. As a result, text indicating the test items that the subject should undergo is output from the language model.

[0017] Furthermore, for example, the language model may be an estimation model that estimates the test items that the subject should undergo, by performing machine learning on the relationship between the text of the medical result written in natural language and the test items that the subject should undergo based on the medical result. In this case, the estimation unit 102 inputs the text received by the receiving unit 101 or a query generated from the text into the language model, and outputs a value indicating the test items that the subject should undergo from the language model.

[0018] The presentation unit 103 presents the test items estimated by the estimation unit 102. The target of the presentation is arbitrary, and may be, for example, a medical professional such as the subject's doctor, or the subject or a person related to the subject. Furthermore, the manner of presentation is not particularly limited. For example, the presentation unit 103 may present the test items by displaying them on a display device, by outputting them as audio from an audio output device, or by printing them out on a printer. Furthermore, the device that presents the response (for example, the display device, audio output device, or printer) may be included in the information processing device 1 or may be an external device to the information processing device 1.

[0019] (Effects of Information Processing Device 1) As described above, the information processing device 1 includes the reception unit 101 that receives input of text in which the medical examination results of the subject are described in natural language, the estimation unit 102 that estimates the test items that the subject should undergo based on this text using a language model that has undergone machine learning of natural language, and the presentation unit 103 that presents the test items estimated by the estimation unit 102. This configuration makes it possible to present appropriate test items according to the medical examination results described in natural language.

[0020] (Examination Support Program) The functions of the information processing device 1 described above can also be realized by a program. The examination support program according to this exemplary embodiment causes a computer to function as: a receiving means for receiving input of text in which the subject's medical results are written in natural language; an estimating means for estimating the test items that the subject should undergo based on the text using a language model that has undergone machine learning of natural language; and a presenting means for presenting the test items estimated by the estimating means. Therefore, the examination support program according to this exemplary embodiment has the effect of making it possible to present appropriate test items according to the medical results written in natural language.

[0021] (Flow of Inspection Support Method) The flow of the inspection support method will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the inspection support method. Note that the execution entity of each step in this inspection support method may be a processor provided in the information processing device 1, or may be a processor provided in another device. In other words, the execution entities of each step may be processors provided in different devices.

[0022] In S1 (reception process), at least one processor receives input of text in which the medical results of the subject are written in a natural language.

[0023] In S2, at least one processor uses a language model that has undergone machine learning of natural language based on the text input received in S1 to estimate the test items that the subject should undergo (estimation process).

[0024] In S3 (presentation process), at least one processor presents the test items estimated in S2.

[0025] (Effects of Examination Assistance Method) As described above, the examination assistance method according to this exemplary embodiment includes a reception process in which at least one processor receives input of text in which the medical examination results of the subject are described in natural language, an estimation process in which, based on the text, the subject is estimated to have test items to be taken by the subject using a language model that has undergone machine learning of natural language, and a presentation process in which the test items estimated by the estimation process are presented. This provides the effect of making it possible to present appropriate test items according to the medical examination results described in natural language.

[0026] [Second Exemplary Embodiment] A second exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. Note that the scope of application of each technical means employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise. Furthermore, each technical means shown in each drawing referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise.

[0027] (Configuration of Test Support System 5A) The configuration of test support system 5A will be described with reference to Fig. 3. Fig. 3 is a diagram showing an overview of test support system 5A. Test support system 5A is a system having a function of accepting input of text in which the medical examination results of a subject are described in natural language, estimating the test items that the subject should undergo from the text, and presenting the estimated test items.

[0028] As shown in the figure, the inspection support system 5A includes an information processing device 1A, a language model 2A, and a display device 3A. The language model 2A may be stored inside the information processing device 1A. If the information processing device 1A includes an output device such as a display device, the display device 3A may be omitted.

[0029] 1, the information processing device 1A has a function of estimating the test items that a subject should undergo from text in which the subject's medical examination results are described in natural language and presenting the estimated test items. When estimating the test items, a language model 2A is used.

[0030] The language model 2A is a machine-learned model that outputs answers to questions written in natural language. For example, the language model 2A may be a GPT-based model such as GPT-4 (Generative Pre-trained Transformer-4), an LLaMA-based model such as ELYZA, or a model of another series.

[0031] The display device 3A is a device that serves as a UI (User Interface) in the examination support system 5A. The display device 3A displays, for example, the examination items that the subject should undergo, thereby presenting the examination items to the subject, medical personnel, and the like. While FIG. 3 shows a stationary display device 3A, a portable display device can also be applied. For example, a portable terminal device (e.g., a tablet terminal, a smartphone, etc.) used by the subject or medical personnel can also be used as a UI device.

[0032] The examination support system 5A may have at least one of an audio output device and a printer, together with or instead of the display device 3A. Although not shown, the examination support system 5A also has an input device for inputting medical results into the information processing device 1A. The input device may be any device capable of inputting medical results, such as a keyboard, a touch panel, or a pen tablet. Medical results may also be input by voice. In this case, a voice input device such as a microphone may be provided. Alternatively, a document reading device such as a scanner may be provided to accept input of medical results written on paper. If the information processing device 1A has an input device, that input device may be used.

[0033] 3, a question sentence a1 input into the language model 2A is displayed on the display device 3A. In the "main text" portion of this question sentence a1, information indicating the medical treatment results of the subject (the patient in this example), such as age, gender, chief complaint, and medical history, is input in natural language. These medical treatment results are input by, for example, a doctor via the input device.

[0034] Here, the information processing device 1A may receive medical results including information useful for estimating the test items that the subject should undergo, i.e., information related to the test items that the subject should undergo. For example, medical professionals and subjects who are users of the test support system 5A may input, as medical results, vital data such as body temperature, medical history, family history, lifestyle history, and doctor's findings in addition to the information described above. In particular, medical results from the initial consultation often contain a lot of information useful for estimating the test items that the subject should undergo. For this reason, it is preferable for the user to input medical results from the initial consultation.

[0035] The information processing device 1A uses the language model 2A to estimate the test items that the subject should undergo from the medical examination results input as described above, i.e., from text written in natural language, and outputs the estimated test items to the display device 3A. In the example of Fig. 3, an answer sentence a2 generated by inputting a question sentence a1 into the language model 2A is displayed on the display device 3A. The answer sentence a2 indicates test items that are considered necessary for the patient, and by referring to these test items, medical professionals such as doctors can efficiently determine what tests to recommend to the subject.

[0036] In other words, the test support system 5A can contribute to efficient decision-making by medical professionals and the like. This also enables medical professionals and the like to shorten the time required to decide on test items, which reduces the burden on medical professionals and the like. In particular, at the first consultation, a doctor needs to determine what test items a patient should undergo in order to decide on a subsequent treatment plan, and shortening the time required for such a decision reduces the burden on the doctor. The test support system 5A can also be used to determine the test items for tests that subjects voluntarily undergo, such as for health checkups.

[0037] (Configuration of Information Processing Device 1A) The configuration of information processing device 1A will be described with reference to FIG. 4. FIG. 4 is a block diagram showing the configuration of information processing device 1A. As shown in the figure, information processing device 1A includes a control unit 10A that controls each unit of information processing device 1A and a storage unit 11A that stores various data used by information processing device 1A. Information processing device 1A also includes a communication unit 12A that enables information processing device 1A to communicate with other devices, an input unit 13A that accepts various data input to information processing device 1A, and an output unit 14A that enables information processing device 1A to output various data. As shown in the figure, control unit 10A of information processing device 1A includes a reception unit 101A, an estimation unit 102A, a presentation unit 103A, and a query generation unit 104A.

[0038] The reception unit 101A receives input of text in which the medical treatment results of the subject are described in a natural language, similar to the reception unit 101 of the information processing device 1. For example, the reception unit 101A may receive input of the subject's electronic medical record.

[0039] Furthermore, for example, the reception unit 101A may accept input of text obtained by performing voice recognition on medical treatment results input by a doctor or the like. In this case, the control unit 10A may include a voice recognition unit that generates text from voice data. Furthermore, for example, the reception unit 101A may accept input of text obtained by performing character recognition on medical treatment results input by handwriting by a doctor or the like. In this case, the control unit 10A may include a character recognition unit that generates text from image data.

[0040] Similar to the estimation unit 102 of the information processing device 1, the estimation unit 102A estimates the test items that the subject should undergo based on the text received as input by the receiving unit 101A, using a language model 2A that has undergone machine learning of natural language. More specifically, the estimation unit 102A inputs a query generated by the query generation unit 104A into the language model 2A. The estimation unit 102A then estimates the test items output from the language model 2A as the test items that the subject should undergo. Note that the term "query" refers to a data generation instruction or generation command for the language model 2A. Therefore, the term "query" in the following description can be replaced with a "generation instruction" or a "generation command."

[0041] As described in the first exemplary embodiment, the language model used may generate an answer to a question written in natural language, or may estimate the test items that a subject should undergo. When the latter language model is used, learning using a large amount of training data is required to improve the accuracy of estimating the test items. On the other hand, when the former language model is used, fine-tuning the language model using such machine learning is effective for improving accuracy, but is not essential. In this exemplary embodiment, an example will be described in which a language model 2A corresponding to the former is used.

[0042] The presentation unit 103A presents the test items estimated by the estimation unit 102A, similar to the presentation unit 103 of the information processing device 1. Specifically, the presentation unit 103A presents the test items estimated by the estimation unit 102A to medical professionals, subjects, etc. by displaying the test items estimated by the estimation unit 102A on the display device 3A. As described above, the presentation mode of the test items is arbitrary, and the device used for presentation is also arbitrary.

[0043] The query generation unit 104A generates a query to be input to the language model 2A. As described above, the language model 2A has been machine-trained to output an answer to a question written in a natural language. Therefore, the query generation unit 104A uses the text received as input by the receiving unit 101A to generate a query (which can also be referred to as a question) inquiring about the test items that the subject should undergo. Note that a method for generating a query will be described later in the section "Query Generation Method."

[0044] As described above, the information processing device 1A includes a query generation unit 104A that generates a query inquiring about test items that the subject should undergo, using text received by the receiving unit 101A. The estimation unit 102A then inputs the generated query into the language model 2A, outputting the query, and estimates the test items that the subject should undergo. Therefore, in addition to the effects of the information processing device 1, the information processing device 1A can also achieve the effect of being able to estimate appropriate test items even when using a language model 2A that has not been sufficiently trained on test items. The estimation unit 102A may use the text generated by inputting the query into the language model 2A as an estimation result, or may process the text (e.g., extract some test items) to generate an estimation result.

[0045] (Regarding the Query Generation Method) As described above, the query generation unit 104A generates a query inquiring about the test items that the subject should undergo, using the text received as input by the receiving unit 101A. For example, the query generation unit 104A may generate a query using a predetermined template. This will be described with reference to FIG. 5. FIG. 5 is a diagram showing an example of a template used to generate a query.

[0046] FIG. 5 shows two types of templates, template b1 and template b2. Both template b1 and template b2 include a question asking about the examination items that the subject should undergo. Specifically, the question included in templates b1 and b2 is, "You are a sincere and excellent Japanese doctor. Please select the combination of examination items required for the patient in this text from the list of examination items: [...] and respond in list format."

[0047] It is not necessary to include the part "You are a sincere and excellent Japanese doctor" in this question. However, as in this example, by including in the question a written statement of the constraints and assumptions that are used when generating an answer in language model 2A, it is possible to generate an answer that conforms to those constraints and assumptions, i.e., an answer that indicates test items that are suitable for the subject.

[0048] Furthermore, the questions included in templates b1 and b2 include a list of candidate test items and instruct the user to select a test item from this list. The estimation unit 102A inputs a query generated using such a template, i.e., a query that includes a list and instructs the user to select a test item from the list, into the language model 2A. This allows the generation of an answer sentence indicating a test item selected from the test items included in the list.

[0049] In this way, the estimation unit 102A may use a list of candidate items for the test items that the subject should undergo and estimate the test items that the subject should undergo from the test items included in the list. This provides the effect of narrowing down the test items to be presented to the subject within the range of candidates shown in the list, in addition to the effect provided by the information processing device 1.

[0050] The test items to be included in the list can be determined arbitrarily. For example, a list that covers all common test items may be used. Alternatively, the query generation unit 104A may determine the test items to be included in the list based on the subject, the facility where the test will be performed, and the like. For example, the test items to be included in the list may be determined in advance for each attribute of the subject, such as age or gender. In this case, the query generation unit 104A may acquire attribute information indicating the subject's attributes and include test items corresponding to the attributes indicated in the attribute information in the list. Alternatively, for example, if the facility where the test will be performed has been determined, the user may be prompted to input test items that can be performed at that facility in advance. In this case, the query generation unit 104A may include the input test items in the list. Alternatively, for example, if there is a test item that has been determined in advance and does not need to be presented by the information processing device 1A, the user may be prompted to input the test item in advance, and the query generation unit 104A may not include the input test item in the list.

[0051] Furthermore, templates b1 and b2 include an item called {input} for inputting the "main text." The query generation unit 104A can generate a query inquiring about test items corresponding to the medical results by inputting text described in natural language, which has been received by the receiving unit 101A, into this item. The answer generated by the estimation unit 102A as input to the language model 2A, i.e., text indicating the test items that the subject should undergo, is displayed on the display device 3A under the control of the presentation unit 103A.

[0052] The difference between templates b1 and b2 is that template b2 includes examples of combinations of "main text" and corresponding "test items." By using template b2, the query generation unit 104A can generate a query including the above examples. Then, by inputting this query into the language model 2A, the estimation unit 102A can generate an answer based on the examples, i.e., text indicating test items based on the examples. This is a technique known as one-shot learning. By applying one-shot learning, it is expected that the possibility of estimating appropriate test items can be increased. Furthermore, few-shot learning, which includes multiple examples, may also be applied.

[0053] The query generation unit 104A may also generate a query including a statement commanding that necessary test items not be omitted from the output, thereby obtaining the effect of reducing the possibility that a diagnosis result will be invalid or a disease will be overlooked due to an omitted test, in addition to the effect achieved by the information processing device 1.

[0054] For example, the above-mentioned template b1 or b2 may be supplemented with a sentence such as, "Inclusion of unnecessary test items is acceptable, so please answer so as not to omit any necessary test items." By using such a template, the query generation unit 104A can generate a query including a sentence instructing that necessary test items not be omitted from the output.

[0055] The query generation unit 104A may also generate a query that includes related information related to the subject's doctor and commands the generation of a response sentence intended for the doctor. In this case, the presentation unit 103A presents the response sentence intended for the doctor, generated by the language model 2A. This provides the effect of being able to present a response sentence appropriate for the doctor in addition to the effect provided by the information processing device 1.

[0056] According to the above configuration, for example, by inputting the doctor's specialty as related information, it is possible to present a response that explains test items at a granularity corresponding to the doctor's specialty. For example, it is possible to present a response that describes test items that do not belong to the doctor's specialty.

[0057] For example, the query generation unit 104A may include related information about the attending physician in a query asking about test items corresponding to the medical results, thereby converting the query into a query that commands the generation of a response sentence for the attending physician. For example, the sentence "You are a sincere and excellent Japanese doctor" in the above-mentioned template b1 or b2 may be changed to "You are a doctor who is {related information}." By using such a template, the query generation unit 104A can generate a query that includes related information about the attending physician and asks about test items corresponding to the medical results. For example, by inputting the attending physician's specialty as related information, it is possible to present test items corresponding to that specialty.

[0058] In addition, the query generation unit 104A may generate a query that commands the generation of a response sentence for the attending physician, separate from the query that asks about test items according to the medical examination results. For example, the query generation unit 104A may generate a query that includes related information about the person in charge and commands the attending physician to explain the test items. This makes it possible to quickly recognize the contents of the test items, even when test items outside the attending physician's expertise are presented.

[0059] Note that the query generation method is arbitrary and is not limited to the above example. For example, the query generation unit 104A may cause the language model 2A or another generation AI that has been trained by machine learning to generate queries to generate a query.

[0060] (Processing Flow) The processing flow executed by the information processing device 1A will be described with reference to Fig. 6. Fig. 6 is a flow diagram showing the processing flow executed by the information processing device 1A. This flow diagram includes each step of the inspection support method according to this exemplary embodiment.

[0061] In S11 (reception process), the reception unit 101A receives input of text in which the medical results of the subject are described in a natural language. For example, the reception unit 101A may receive input of an electronic medical record including text in which the medical results are described in a natural language. Furthermore, for example, the reception unit 101A may receive input of text generated by speech recognition of the voice-input medical results.

[0062] In S12, the query generation unit 104A generates a query inquiring about the test items that the subject should undergo, using the text received as input in S11. For example, the query generation unit 104A may generate a query by inputting the text received as input in S11 into the template b1 or b2 shown in FIG.

[0063] In S13 (estimation process), the estimation unit 102A estimates the test items that the subject should undergo, based on the text accepted as input in S11, using the language model 2A that has undergone machine learning of natural language. More specifically, the estimation unit 102A estimates that the test items that the subject should undergo are the test items that are output by inputting the query generated in S12 into the language model 2A (the test items indicated in the text generated by the language model 2A).

[0064] In S14 (presentation process), the presenting unit 103A presents the test items estimated in S13. For example, the presenting unit 103A may present the test items estimated in S13 by displaying text generated by the language model 2A on the display device 3A.

[0065] In S15, the receiving unit 101A determines whether it is necessary to improve the estimation accuracy of the test item. If the determination in S15 is YES, the process proceeds to S16, and if the determination in S15 is NO, the process proceeds to S19.

[0066] The determination in S15 may be made based on feedback from a medical professional, a subject, or the like regarding the test items presented in S14. For example, the reception unit 101A may determine YES in S15 when feedback indicating that the presented test items are inappropriate or feedback instructing the user to re-estimate the test items is detected. On the other hand, the reception unit 101A may determine NO in S15 when feedback indicating that the presented test items are appropriate is detected or when no feedback is detected within a predetermined time. If the determination in S15 is NO, the test items presented in S14 will be performed.

[0067] In S16, the query generation unit 104A generates a confirmation query inquiring about information that should be input to improve the accuracy of estimating the test items that the subject should undergo. This information can also be expressed as information that is lacking for estimating the test items. For example, the query generation unit 104A may generate a query such as, "What additional information should be input to improve the accuracy of estimating the test items?" Note that the method for generating the confirmation query is not particularly limited. For example, the query generation unit 104A may generate the confirmation query by inputting the subject's medical examination results into a predetermined template, or may cause the language model 2A or another generation AI that has been trained by machine learning to generate queries to generate the confirmation query.

[0068] In S17, the estimation unit 102A estimates that the information output by inputting the query generated in S16 into the language model 2A is information that should be input to improve the accuracy of estimating the test items that the subject should undergo.

[0069] In S18, the presentation unit 103A presents the information estimated in S17 and prompts the user to input the information. For example, the presentation unit 103A may cause the display device 3A to display, together with the text generated by the language model 2A, a message prompting the user to input the information indicated in the text.

[0070] In this way, the query generation unit 104A may generate a confirmation query that asks for information to be input in order to improve the accuracy of estimating the test items that the subject should undergo. In this case, the presentation unit 103A may present information obtained by inputting the generated confirmation query into the language model 2A. This provides, in addition to the effects of the information processing device 1, an effect of being able to extract necessary information from the subject, etc., and improve the accuracy of estimating the test items that the subject should undergo.

[0071] After S18 is completed, the process returns to S11, where the receiving unit 101A receives input of information for improving estimation accuracy. In S12, the query generation unit 104A uses the information received in S11 to generate a query asking about the test items that the subject should undergo. In S13, the estimation unit 102A inputs the query generated in the most recent S12 to the language model 2A to generate an answer sentence indicating the test items that the subject should undergo. In S14, the presentation unit 103A presents the test items estimated in the most recent S13.

[0072] In S19, the receiving unit 101A determines whether result information indicating the test results for the test items presented in S14 has been input. If the determination in S19 is NO, the processing in FIG. 6 ends. On the other hand, if the determination in S19 is YES, the process returns to S12, where the query generation unit 104A uses the input result information to generate a query asking about further test items that the subject should undergo. Then, in S13, the estimation unit 102A inputs the query generated in the most recent S12 to the language model 2A to generate an answer sentence indicating further test items that the subject should undergo. In S14, the presentation unit 103A presents the test items estimated in the most recent S13.

[0073] In this way, the receiving unit 101A may receive input of result information indicating the test results of the test items that the subject has undergone among the test items presented by the presenting unit 103A. In this case, the estimating unit 102A may estimate the test items that the subject should further undergo, using the language model 2A, based on the input result information. This provides, in addition to the effects of the information processing device 1, an effect of being able to present the next test items that should be performed, estimated based on the test results of the previously presented test items.

[0074] [Regarding the Use of a Language Model for Estimating Test Items to be Taken by a Subject] As described above, the estimation unit 102A may estimate the test items to be taken by a subject using a language model for estimating the test items to be taken by a subject, the language model being obtained by machine learning the relationship between the text of the medical result written in natural language and the test items to be taken by the subject of the medical result. This provides the effect of being able to estimate appropriate test items based on the learning results, in addition to the effect provided by the information processing device 1.

[0075] As the language model described above, for example, Bidirectional Encoder Representations from Transformers (BERT) can be applied. In this case, a language model can be used that has been pre-trained using a corpus containing a large amount of natural language text, and then fine-tuned using training data in which correct test items are labeled for the text of medical results.

[0076] The query generation unit 104A may generate a query according to the language model used. For example, when a BERT model is used, the query generation unit 104A outputs {input} in template b1 or b2 shown in FIG. 5 . In this case, the estimation unit 102A inputs the query into the BERT model, thereby outputting a prediction result of the required test item. This prediction result is a numerical value indicating the likelihood that each test item learned by fine tuning should be performed. Therefore, when such a BERT model is used, the estimation unit 102A estimates that test items for which a numerical value higher than a predetermined threshold is output are test items that should be performed.

[0077] When training such a language model, it is preferable to perform machine learning so as to minimize the possibility that necessary test items will be omitted from the output. For example, if all necessary test items are included in the estimation result, the machine-learned language model may be used, assuming that the estimation result is correct even if some unnecessary test items are included. Furthermore, when estimating test items from the output of the language model, the possibility that necessary test items will be omitted from the output can be reduced by setting the above-mentioned threshold value low.

[0078] The processing flow when using a language model to estimate the test items that a subject should undergo is generally similar to the example in Figure 6. However, steps S15 to S18 are not performed. Also, step S19 may be omitted or may be performed. When performing step S19, a language model that has been machine-learned to determine the relationship between the test results and the next test items that should be taken based on those test results can be used to estimate and present the next test items that should be taken based on the input test results.

[0079] Here, the presentation unit 103A does not necessarily need to present all of the test items estimated by the estimation unit 102A. For example, if there are test items that have been decided to be performed in advance, it is sufficient to input such test items. This allows the presentation unit 103A to present the remaining test items estimated by the estimation unit 102A, excluding those that have been decided to be performed. Furthermore, at least one of test items that cannot be performed at the facility where the test is to be performed and test items that can be performed may be input in advance. This allows the presentation unit 103A to present the test items that can be performed among the test items estimated by the estimation unit 102A.

[0080] The presentation unit 103A may also present supplemental information regarding the test items according to the specialty of the doctor in charge of the subject. In this case, for example, the test items may be classified by specialty in advance, and the doctor's specialty may be input. This allows the presentation unit 103A to identify test items that are outside the specialty of the doctor in charge of the subject from among the test items estimated by the estimation unit 102A, and present supplemental information regarding those test items. Note that the supplemental information may be prepared in advance for each test item, or may be detected by a search or the like.

[0081] [Modifications] The execution entity of each process described in each of the exemplary embodiments described above is arbitrary and is not limited to the above examples. In other words, the functions of the information processing devices 1 and 1A can be realized by multiple devices (which can also be called processors) that can communicate with each other. For example, each process described in the flow charts of FIGS. 2 and 6 can be shared and executed by multiple processors. In other words, the execution entity of the inspection support method in each of the above embodiments may be one processor or multiple processors.

[0082] [Example of Software Implementation] Some or all of the functions of the information processing device 1, 1A may be implemented by hardware such as an integrated circuit (IC chip), or may be implemented by software.

[0083] In the latter case, the information processing device 1 or 1A is realized by, for example, a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter referred to as computer C) is shown in Fig. 7. Fig. 7 is a block diagram showing the hardware configuration of computer C that functions as information processing device 1 or 1A.

[0084] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program (inspection support program) P for operating the computer C as the information processing device 1 or 1A. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing each function of the information processing device 1 or 1A.

[0085] The processor C1 may be, for example, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0086] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may also include a communication interface for transmitting and receiving data to and from other devices. The computer C may also include an input / output interface for connecting input / output devices such as a keyboard, a mouse, a display, and a printer.

[0087] The program P can also be recorded on a non-transitory, tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can acquire the program P via such a recording medium M. The program P can also be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also acquire the program P via such a transmission medium.

[0088] [Appendix 1] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0089] (Appendix A1) An information processing device comprising: a receiving means for receiving input of text in which the medical examination results of a subject are written in natural language; an estimation means for estimating the test items that the subject should undergo based on the text using a language model that has been machine-learned to understand natural language; and a presentation means for presenting the test items estimated by the estimation means.

[0090] (Appendix A2) The information processing device according to Appendix A1, further comprising: a query generation means for generating a query inquiring about test items that the subject should undergo using the text; and the estimation means for estimating that the test items that the subject should undergo are test items that are output by inputting the query into the language model.

[0091] (Supplementary Note A3) The information processing device according to Supplementary Note A2, wherein the query generating means generates the query including a statement instructing that necessary test items not be omitted from the output.

[0092] (Appendix A4) The information processing device according to Appendix A2 or A3, wherein the query generation means generates a confirmation query that asks about information that should be input to improve estimation accuracy of test items that the subject should undergo, and the presentation means presents information obtained by inputting the confirmation query into the language model.

[0093] (Appendix A5) An information processing device according to any one of Appendices A2 to A4, wherein the query generation means generates a query including relevant information related to the subject's doctor and commands the generation of an answer sentence for the doctor, and the presentation means presents the answer sentence for the doctor generated by the language model.

[0094] (Appendix A6) The information processing device according to Appendix A1, wherein the estimation means estimates the test items that the subject should undergo using a language model that estimates the test items that the subject should undergo, the language model having undergone machine learning of the relationship between the text of the medical results written in natural language and the test items that the subject of the medical results should undergo.

[0095] (Appendix A7) An information processing device described in any of Appendices A1 to A6, wherein the accepting means accepts input of result information indicating the test results of the test items that the subject has taken, among the test items presented by the presenting means, and the estimating means uses the language model based on the result information to estimate further test items that the subject should take.

[0096] (Appendix A8) The information processing device according to any one of Appendices A1 to A7, wherein the estimation means uses a list of candidate items for the test items that the subject should undergo and estimates the test items that the subject should undergo from among the test items included in the list.

[0097] (Appendix B1) An examination assistance method including: a reception process in which at least one processor receives input of text in which the medical examination results of a subject are described in natural language; an estimation process in which, based on the text, a language model that has been machine-learned to understand natural language is used to estimate the examination items that the subject should undergo; and a presentation process in which the examination items estimated by the estimation process are presented.

[0098] (Appendix B2) The test support method described in Appendix B1, further comprising a query generation process in which the at least one processor uses the text to generate a query inquiring about the test items that the subject should undergo, and in the estimation process, the at least one processor estimates that the test items that the subject should undergo are the test items that are output by inputting the query into the language model.

[0099] (Supplementary Note B3) The test support method according to Supplementary Note B2, wherein in the query generation process, the at least one processor generates the query including a statement instructing that required test items not be omitted from the output.

[0100] (Appendix B4) The test assistance method described in Appendix B2 or B3, wherein the at least one processor generates a confirmation query asking about information that should be input to improve estimation accuracy of test items that the subject should undergo, and presents information obtained by inputting the confirmation query into the language model.

[0101] (Appendix B5) The examination assistance method described in any of Appendices B2 to B4, wherein the at least one processor generates a query including relevant information related to the subject's doctor and instructs the doctor to generate an answer sentence for the doctor, and presents the answer sentence for the doctor generated by the language model.

[0102] (Appendix B6) The examination support method described in Appendix B1, wherein in the estimation process, the at least one processor estimates the examination items that the subject should undergo using a language model that estimates the examination items that the subject should undergo, the language model having undergone machine learning of the relationship between the text of the medical results written in natural language and the examination items that the subject should undergo based on the medical results.

[0103] (Appendix B7) An examination support method described in any of Appendices B1 to B6, wherein the at least one processor accepts input of result information indicating the test results of the test items that the subject has taken, among the test items presented in the presentation process, and uses the language model based on the result information to estimate further test items that the subject should take.

[0104] (Appendix B8) An examination support method described in any of Appendices B1 to B7, wherein in the estimation process, the at least one processor uses a list of candidate items for the examination items that the subject should undergo, and estimates the examination items that the subject should undergo from among the examination items included in the list.

[0105] (Appendix C1) An examination assistance program that causes a computer to function as: a receiving means that receives input of text in which the medical examination results of a subject are written in natural language; an estimating means that estimates the examination items that the subject should undergo based on the text using a language model that has been machine-learned to understand natural language; and a presenting means that presents the examination items estimated by the estimating means.

[0106] (Appendix C2) The test assistance program described in Appendix C1, wherein the computer is caused to function as query generation means that uses the text to generate a query inquiring about the test items that the subject should undergo, and the estimation means estimates that the test items that are output by inputting the query into the language model are the test items that the subject should undergo.

[0107] (Supplementary Note C3) The test assistance program according to Supplementary Note C2, wherein the query generating means generates the query including a statement instructing that necessary test items not be omitted from the output.

[0108] (Appendix C4) The test assistance program described in Appendix C2 or C3, wherein the query generation means generates a confirmation query that asks for information to be input to improve the estimation accuracy of the test items that the subject should undergo, and the presentation means presents information obtained by inputting the confirmation query into the language model.

[0109] (Appendix C5) The examination assistance program described in any of Appendices C2 to C4, wherein the query generation means generates a query that includes relevant information related to the subject's doctor and commands the doctor to generate an answer sentence, and the presentation means presents the answer sentence for the doctor, generated by the language model.

[0110] (Appendix C6) The examination assistance program described in Appendix C1, wherein the estimation means estimates the examination items that the subject should undergo using a language model that estimates the examination items that the subject should undergo, the language model having undergone machine learning to learn the relationship between the text of the medical results written in natural language and the examination items that the subject should undergo based on the medical results.

[0111] (Appendix C7) A test assistance program described in any of Appendices C1 to C6, wherein the receiving means receives input of result information indicating the test results of the test items that the subject has taken, among the test items presented by the presentation means, and the estimation means uses the language model based on the result information to estimate the test items that the subject should further take.

[0112] (Appendix C8) The examination assistance program described in any of Appendices C1 to C7, wherein the estimation means uses a list of candidate examination items that the subject should undergo and estimates the examination items that the subject should undergo from the examination items included in the list.

[0113] (Appendix D1) An information processing device comprising at least one processor, the at least one processor performing a reception process for receiving input of text in which the medical examination results of a subject are described in natural language, an estimation process for estimating the test items that the subject should undergo based on the text using a language model that has been machine-learned to learn natural language, and a presentation process for presenting the test items estimated by the estimation process.

[0114] The information processing device may further include a memory, and the memory may store an inspection support program for causing the at least one processor to execute each of the processes.

[0115] (Appendix D2) The information processing device described in Appendix D1, wherein the at least one processor executes a query generation process using the text to generate a query inquiring about the test items that the subject should undergo, and in the estimation process, the at least one processor estimates that the test items that the subject should undergo are the test items that are output by inputting the query into the language model.

[0116] (Supplementary Note D3) The information processing device according to Supplementary Note D2, wherein in the query generation process, the at least one processor generates the query including a statement instructing that required test items not be omitted from the output.

[0117] (Appendix D4) The information processing device described in Appendix D2 or D3, wherein the at least one processor generates a confirmation query asking about information that should be input to improve estimation accuracy of test items that the subject should undergo, and presents information obtained by inputting the confirmation query into the language model.

[0118] (Appendix D5) An information processing device described in any of Appendices D2 to D4, wherein the at least one processor generates a query including relevant information related to the subject's doctor and instructs the doctor to generate an answer sentence for the doctor, and presents the answer sentence for the doctor generated by the language model.

[0119] (Appendix D6) The information processing device described in Appendix D1, wherein in the estimation process, the at least one processor estimates the test items that the subject should undergo using a language model that estimates the test items that the subject should undergo, the language model having undergone machine learning to determine the relationship between the text of the medical results written in natural language and the test items that the subject of the medical results should undergo.

[0120] (Appendix D7) An information processing device described in any of Appendices D1 to D6, wherein the at least one processor accepts input of result information indicating the test results of the test items that the subject has taken, among the test items presented by the presentation process, and uses the language model based on the result information to estimate the further test items that the subject should take.

[0121] (Appendix D8) An information processing device described in any of Appendices D1 to D7, wherein in the estimation process, the at least one processor uses a list of candidate items for the test items that the subject should undergo and estimates the test items that the subject should undergo from among the test items included in the list.

[0122] (Appendix E) A non-transitory recording medium having recorded thereon a test assistance program that causes a computer to function as an information processing device, the test assistance program causing the computer to execute a reception process that receives input of text in which the subject's medical examination results are written in natural language, an estimation process that estimates the test items that the subject should undergo based on the text using a language model that has been machine-learned to understand natural language, and a presentation process that presents the test items estimated by the estimation process.

[0123] REFERENCE SIGNS LIST 1 Information processing device 101 Reception unit (reception means) 102 Estimation unit (estimation means) 103 Presentation unit (presentation means) 1A Information processing device 101A Reception unit (reception means) 102A Estimation unit (estimation means) 103A Presentation unit (presentation means) 104A Query generation unit (query generation means) 2A Language model

Claims

1. An information processing apparatus comprising: a reception means for receiving an input of text in which the medical treatment result of a subject is described in natural language; an estimation means for estimating, using a language model obtained by machine learning of natural language, the examination items that the subject should undergo based on the text; and a presentation means for presenting the examination items estimated by the estimation means.

2. The information processing apparatus according to claim 1, further comprising a query generation means for generating a query asking about the examination items that the subject should undergo using the text, wherein the estimation means estimates, as the examination items that the subject should undergo, the examination items output by inputting the query to the language model.

3. The information processing apparatus according to claim 2, wherein the query generation means generates the query including a sentence instructing to prevent necessary examination items from being omitted from the output.

4. The information processing apparatus according to claim 2 or 3, wherein the query generation means generates a confirmation query asking for information to be input in order to improve the estimation accuracy of the examination items that the subject should undergo, and the presentation means presents the information obtained by inputting the confirmation query to the language model.

5. The information processing apparatus according to claim 2 or 3, wherein the query generation means generates a query including related information related to the subject's attending doctor and instructing to generate a response sentence for the attending doctor, and the presentation means presents the response sentence for the attending doctor generated by the language model.

6. The information processing apparatus according to claim 1, wherein the estimation means estimates the examination items that the subject should undergo using a language model that has learned the relationship between the text of the medical treatment result described in natural language and the examination items that the subject corresponding to the medical treatment result should undergo.

7. The information processing apparatus according to any one of claims 1, 2, and 6, wherein the reception means receives an input of result information indicating the examination results of the examination items that the subject has undergone among the examination items presented by the presentation means, and the estimation means estimates, based on the result information, using the language model, the examination items that the subject should further undergo.

8. The information processing apparatus according to any one of claims 1, 2, and 6, wherein the estimation means estimates the examination items that the subject should undergo from among the examination items included in the list using a list of items that are candidates for the examination items that the subject should undergo.

9. An examination support method, comprising: a reception process in which at least one processor receives an input of text in which a medical treatment result of a subject is described in natural language; an estimation process in which, based on the text, a language model obtained by machine learning natural language is used to estimate examination items to be received by the subject; and a presentation process in which the examination items estimated in the estimation process are presented.

10. An examination support program that causes a computer to function as: reception means for receiving an input of text in which a medical treatment result of a subject is described in natural language; estimation means for estimating examination items to be received by the subject using a language model obtained by machine learning natural language based on the text; and presentation means for presenting the examination items estimated by the estimation means.

Images