Method and device for creating clinical text information time series data, method and device for visualizing and displaying clinical text information time series, and clinical text information time series visualization system

Abstract

To provide clinical text information time series data manufacturing method and device, clinical text information time series visualization display method and device as well as clinical text information time series visualization system which enable intuitive time series display to be performed.SOLUTION: A clinical text information time series data manufacturing device 1 comprises: clinical text information extraction means 5 which receives input of a clinical text created in a clinical site, extracts the disease name, inspection name, medicine name, treatment name, clinical medicine expression and time expression included in the clinical text and adds tags to them; input means 2 which inputs an annotation file of the clinical text in which the tags are added to the clinical medicine expression and time expression; time relation identification means 3 which identifies a time relation between the time expression and the clinical medicine expression for the annotation file extracted and added with the tags by the clinical text information extraction means 5 or the input annotation file; and time series data formation means 4 which turns the clinical medicine expression into time series data 14 along the time expression.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] This invention relates to a technology for converting the progression of a patient's pathology and symptoms, as well as the status of medication and examinations, recorded in clinical texts such as electronic medical records, into time-series data and visualizing it in a time-series display format. [Background technology]

[0002] Electronic medical record systems have traditionally been widely used in hospitals, clinics, and other medical settings. However, these systems have many complex functions, making it difficult to view a patient's examination and treatment history. In other words, in medical settings, most important information from the perspective of medical research and actual clinical practice is currently entered in free-form fields. To make information about the previous consultation easier to view, doctors typically copy and paste the previous description into the free-form field for the current consultation. To help doctors easily understand at a glance the progression of a patient's pathology and symptoms, as well as their test and treatment history, it is useful to visualize the progression of pathology, symptoms, medication, and tests along a chronological timeline. This approach, known as a clinical pathway, is widely used in medical settings. This primarily organizes the treatment history and schedule of hospitalized patients in a timeline format, and not only improves the efficiency of team medical care consisting of doctors, nurses, pharmacists, and others, but also, due to its intuitive ease of understanding, is used to explain treatment plans to patients. However, this type of visualization is currently performed manually.

[0003] As a technology for structuring text information entered in free-entry fields of electronic medical records, etc., an information processing system having a function for accepting free input of text information and displaying the free-entry text information and structured information based on the free-entry text information is known (see Patent Document 1). However, the information processing system of Patent Document 1 makes a person who enters free-entry text input be aware of structuring the text information while entering the information, and is not capable of visualizing the progression of a patient's lesions and symptoms or the status of medication and examination implementation in a chronological format.

[0004] Furthermore, a technology proposed for requirements of a time series visualization system from text is known as a technology for visualizing the progression of a patient's pathology and symptoms, as well as the status of medication and examinations, recorded in electronic medical records, in a time series format (see Non-Patent Document 1). However, the technology disclosed in Non-Patent Document 1 has the problem that it does not mention at all the language processing part of the natural language in the text, and merely indicates the requirements required for time series display. Furthermore, although the requirements for time series visualization are based on clinical medicine, there is also the problem that the time axis is not clearly indicated.

[0005] Additionally, a technology that combines classical language processing technology with rule-based processing to visualize time series is known (see Non-Patent Document 2). However, the technology disclosed in Non-Patent Document 2 has only been verified using one example document, and considering the complexity of linguistic expressions and time relationships, it is clear that handwritten rules alone cannot handle a wide variety of clinical texts. Furthermore, the time series display is simply a matter of inputting information into a general existing tool that is not specialized for medicine, which also presents a problem of limited practicality.

[0006] The technical foundation for time-series visualization of clinical texts such as electronic medical records is divided into a language processing part that extracts clinical medical facts from clinical texts and a time-series display part that visualizes the extracted clinical medical facts. However, as mentioned above, the technologies disclosed in Patent Document 1 and Non-Patent Documents 1 and 2 all lack feasibility and practicality. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2020-86541 [Non-patent literature]

[0008] [Non-Patent Document 1] C.Hallett, “Multi-modal presentation of medical histories”, 13th international conference on Intelligent userinterfaces, Association for Computing Machinery (IUI'08), pp.80-89, 2008. [Non-patent document 2] H. Jung et al., “Building Timelines from Narrative Clinical Records: Initial Results Based-on Deep Natural Language Understanding”, BioNLP 2011 Workshop, Association for Computational Linguistics, pp. 146-154, 2011. Summary of the Invention [Problem to be solved by the invention]

[0009] In view of this situation, the present invention aims to provide a method and device for creating clinical text information time-series data that extracts clinical medical expressions from various clinical texts such as electronic medical records using language processing, automatically analyzes time expressions specific to clinical medicine to create time-series data, and provides an intuitive time-series display. [Means for solving the problem]

[0010] In order to solve the above problem, the clinical text information time-series data creation method of the present invention comprises: The computer Extracting and tagging at least one clinical medical expression, such as a disease name, a test name, a drug name, a treatment name, and a clinical stage expression, and a time expression, from clinical text created in a clinical setting; Time expressions include absolute time expressions that uniquely correspond to a point on the time axis, and relative time expressions that do not uniquely correspond to a point on the time axis. time Relative A time relationship representing the time information between the expression and the clinical medical expression is identified, and the clinical medical expression is converted into time-series data along the time expression.

[0011] With this configuration, clinical medical expressions can be extracted and tagged using language processing from clinical texts created in clinical settings, such as electronic medical records and CT interpretation reports, and time expressions specific to clinical medicine can be automatically analyzed to create time series data and provide an intuitive time series display. To extract information from clinical texts, a trained language processing model is used, which provides clinical medical expressions in natural languages ​​such as Japanese and English, the relationships between them, as well as temporal expressions and contextual relationships, to automatically interpret clinical medical facts (such as the occurrence of lesions, medication and test status) described in the clinical texts. Here, clinical text refers to text created in clinical settings, such as electronic medical records (medical records) and X-ray and CT scan findings. Clinical medical expressions, including disease names, test names, drug names, procedure names, and clinical stage expressions, are linguistic expressions that express medical concepts. Clinical stage expressions include expressions such as admission, discharge, consultation, medication instructions, and chemotherapy. Temporal relationships are relationships between clinical medical expressions that express temporal information, such as when a certain event occurred or when a certain condition existed. Converting clinical medical expressions into time-series data using temporal relationships along the timeline of the time expression involves converting clinical medical expressions into time-series data. For time expressions specific to clinical texts (including "before surgery," "after surgery," "before hospitalization," "after discharge," and "fifth course"), clinical medical facts can be arranged in the chronological order described in the clinical text by their type (lesion, medication, test, etc.). The extraction and tagging of clinical medical expressions are performed simultaneously or intertemporally.

[0012] In the method for producing clinical text information time-series data of the present invention, it is preferable that the clinical medical expressions are classified and identified into the following five types 1) to 5) for the time relationship. 1) A clinical medical expression occurs or exists in a temporal expression (hereinafter referred to as a "simultaneous relationship"). 2) The clinical medical expression occurs or exists before the temporal expression and does not exist in the temporal expression (hereinafter referred to as a "prior relationship"). 3) The clinical medical expression occurs or exists after the temporal expression and does not exist in the temporal expression (hereinafter referred to as the "posterior relationship"). 4) The clinical medical expression occurs in or exists in the temporal expression and exists after the temporal expression (hereinafter referred to as the "onset relationship"). 5) The clinical medical expression occurs or exists before the temporal expression and exists until the temporal expression (hereinafter referred to as the "termination relationship").

[0013] By classifying and identifying the above five types 1) to 5), it is possible to accurately assign temporal relationship information to clinical text. In this invention, in order to create time series data specialized for clinical medical applications, we restructured the existing linguistic specifications related to time expressions and temporal relationships and designed five temporal relationship types 1) to 5). The five temporal relationship types are a necessary and sufficient configuration of temporal relationships that link temporal expressions that should be considered as targets for extraction from clinical text with clinical medical facts.

[0014] Each clinical medical expression has a state attribute attached to it, and the state attribute is given a type tag: "planned," "implemented," or "not implemented." The value of the state attribute represents the state at a specific point in time in the surrounding context, and is determined relative to time. In a simultaneous relationship, if the status attribute is "scheduled," it means that it is scheduled for that time (such as the scheduled implementation date), if the status attribute is "implemented," it means that it has been implemented at that time, and if the status attribute is "not implemented," it means that it has not been implemented at that time (implementation at that time has been postponed). Furthermore, a prior relationship means that if the status attribute is "planned," it is scheduled to be performed at some point before that point in time; if the status attribute is "performed," it means that it has been performed at some point before that point in time; and if the status attribute is "not performed," it means that it has not been performed at some point before that point in time. Furthermore, a post-event relationship means that if the state attribute is "planned," it is planned to be performed at some point after that point in time; if the state attribute is "performed," it means that it has been performed at some point in time after that point in time; and if the state attribute is "not performed," it means that it has not been performed at some point in time after that point in time. In addition, the start relationship means that if the status attribute is "planned," it is scheduled to start at that point in time; if the status attribute is "implemented," it means that it had not been implemented up until that point but had started at that point in time; and if the status attribute is "not implemented," it means that it had been implemented up until that point in time but had finished at that point in time. In addition, the termination relationship means that if the state attribute is "planned", it is scheduled to end at that point in time; if the state attribute is "implemented", it means that the activity had been implemented up until that point but had ended at that point in time; and if the state attribute is "not implemented", it means that the activity had not been implemented up until that point in time but had started at that point in time.

[0015] In the clinical text information time-series data creation method of the present invention, time expression tags are preferably classified and identified into at least six types of expressions: "date expression," "time expression," "period expression," "frequency expression," "age expression," and "clinical-specific expression." Classifying and identifying time expressions to be considered for extraction from clinical text into at least six types of expressions—date, time, period, frequency, age, and clinical-specific expressions—enables more accurate classification. In addition to the six types, a "not applicable to any" type may be added to accommodate cases that do not fit any of the above categories, making a total of seven types. The six types of time expression—date, time, period, frequency, age, and clinical-specific—and the five types of time relationships (simultaneous relationship, before relationship, after relationship, start relationship, and end relationship)—can be used to classify into 30 types (6 x 5), enabling clinical medical facts to be displayed in a chronological order not only as occurrence dates or dates of implementation, but also as "periods" that may have no beginning or end. Here, frequency refers to a set of dates or the number of times performed, such as "every three days," and clinically specific refers to expressions specific to clinical practice, such as "before hospitalization," "after treatment," or "third course."

[0016] In the method for creating clinical text information time-series data of the present invention, the time expression may include the creation date and time of the clinical text, and may be added as metadata of the clinical text. As a time expression, the creation date and time of the clinical text (Document The inclusion of Creation Time (DCT) enables the creation of more accurate time-series data. Note that DCT does not appear explicitly in the text, but is instead added as metadata to the clinical text.

[0017] In the method for creating clinical text information time-series data of the present invention, it is preferable that the tagging types of clinical medical expressions further include at least one of site expressions, change expressions, and feature expressions. By adding site expressions, change expressions, feature expressions, etc. to the tagging types of clinical medical expressions, it will be possible to create more accurate time series data. A site expression is an expression that indicates the location of a disease or lesion, and indicates a body part that cannot be identified by the disease name. Change expressions are expressions that express changes related to lesions, test values, drug values, and treatments, such as "increase in tumor mass" or "reduced to 2.5 mg." Feature expressions are modifying expressions related to the scale, value, range, and degree of lesions, symptoms, and areas, such as "small amount" or "2.5 cm x 5 cm."

[0018] The clinical text information time series visualization display method of the present invention is a method for displaying time series data of clinical medical expressions in clinical text created using any of the clinical text information time series data creation methods described above, in which a time display axis is displayed according to the time expression, and the clinical medical expressions are displayed in correspondence with the time display axis for each tag type according to their temporal relationships. This enables intuitive time series display based on the time representation specific to clinical medicine. The time axis can be either the horizontal or vertical axis in a two-dimensional display, or it can be assigned to one axis in a three-dimensional display, for example.

[0019] The clinical text information time-series data creation device of the present invention includes an input means for inputting an annotation file of clinical text that is tagged with at least one clinical medical expression, such as a disease name, an examination name, a drug name, a treatment name, and a clinical stage expression, and a time expression, extracted from the clinical text created in a clinical setting; a time relation identification means for identifying a time relation representing temporal information between the time expression and the clinical medical expression from the input annotation file; and a time series data creation means for converting the clinical medical expression into time series data according to the time expression. The time expressions include absolute time expressions that uniquely correspond to a certain point on the time axis by themselves, and relative time expressions that do not uniquely correspond, and the time relationship identifying means identifies a time relationship that represents temporal information between the relative time expression and the clinical medical expression based on the relationship with other time expressions. do.

[0020] It is preferable that the clinical text information time-series data creation device of the present invention further includes a clinical text information extraction means for inputting clinical text and extracting and tagging at least one clinical medical expression, such as a disease name, a test name, a drug name, a treatment name, and a clinical stage expression, and a time expression contained in the clinical text using a trained language processing model. Here, the language processing model can be an existing language processing model that has been trained to be able to accurately identify clinical medical expressions (disease names, test names, drug names, treatment names, clinical stage expressions) and time expressions.

[0021] The clinical text information time series visualization display device of the present invention is preferably a display terminal connected directly or via a network to the above-mentioned clinical text information time series data creation device of the present invention, and is preferably equipped with display means for displaying a time display axis according to a time expression and displaying clinical medical expressions in correspondence with the time display axis for each tag type according to a time relationship.

[0022] The clinical text information time series visualization system of the present invention comprises a first server for medical entity annotation, which inputs clinical text created in clinical settings, extracts and tags at least one clinical medical expression and time expression, including disease names, test names, drug names, treatment names, and clinical stage expressions, contained in the clinical text using a trained language processing model, and outputs an annotation file of the tagged clinical text; a second server for medical entity relationship annotation, which inputs the annotation file of the clinical text from the first server, identifies a time relationship representing temporal information between the time expression and the clinical medical expression for the annotation file, converts the clinical medical expression into time series data according to the time expression, and outputs the converted time series data; and a client terminal, connected via a network, which outputs the clinical text to the first server, inputs the time series data from the second server, displays a time display axis according to the time expression, and displays the clinical medical expression in correspondence with the time display axis for each tag type according to the time relationship. The system is characterized in that the time expressions include absolute time expressions that uniquely correspond to a certain point on a time axis by themselves and relative time expressions that do not uniquely correspond, and the second server identifies a time relationship that represents temporal information between the relative time expressions and the clinical medical expressions based on a relationship with other time expressions. do. With this configuration, it becomes possible to input time-series data into the client terminal that outputs the clinical text and display it in a visualized form.

[0023] The clinical text information time series visualization system of the present invention may be one in which the first server and the second server are integrated. By integrating the first server and the second server, a simpler system configuration can be achieved in which the first server for medical entity annotations and the second server for medical entity relationship annotations are integrated. [Effects of the Invention]

[0024] The clinical text information time-series data creation method and device of the present invention have the advantage of being able to extract information from a variety of clinical medical texts using high-performance language processing and display it in an intuitive time-series format that takes into account the time expressions specific to clinical medicine. As a result, the present invention can simultaneously achieve feasibility and practicality in both language processing and time-series display that none of the prior art has achieved. [Brief explanation of the drawings]

[0025] [Figure 1] Overview of the method for creating time-series data of clinical text information and the method for visualizing and displaying time-series data of clinical text information [Figure 2] Diagram of how to create clinical text information time series data (1) [Figure 3] Diagram of how to create clinical text information time series data (2) [Figure 4] Functional block diagram of a clinical text information time series data creation device and a clinical text information time series visualization display device according to the first embodiment. [Figure 5] Configuration diagram of the clinical text information time series visualization system of Example 1 [Figure 6] Data flow diagram of the clinical text information time series visualization system of Example 1 [Figure 7] Image of tagging in an experiment using medical records (1) [Figure 8] Image of tagging in an experiment using medical records (2) [Figure 9] Image of tagging in an experiment using medical records (3) [Figure 10] Image of tagging in an experiment using medical records (4) [Figure 11] Image of tagging in an experiment using medical records (5) [Figure 12] Image of time series visualization in an experimental example using medical records [Figure 13] Image of tagging in an experimental example using radiological findings (1) [Figure 14] Image of tagging in an experimental example using radiological findings (2) [Figure 15] Image of time series visualization in an experimental example using radiological findings [Figure 16] Functional block diagram of a clinical text information time-series data creation device according to a second embodiment [Figure 17] Functional block diagram of a clinical text information time-series data creation device according to a third embodiment [Figure 18] Functional block diagram of a clinical text information time series data creation device and a clinical text information time series visualization display device according to a fourth embodiment. [Figure 19] Configuration diagram of clinical text information time series visualization system of Example 5 [Figure 20] Data flow diagram of the clinical text information time series visualization system of Example 5 DETAILED DESCRIPTION OF THE INVENTION

[0026] An example of an embodiment of the present invention will be described in detail below with reference to the drawings. Note that the scope of the present invention is not limited to the following examples and illustrated examples, and many modifications and variations are possible.

[0027] FIG. 1 is an overview diagram of the method for creating time-series data of clinical text information and the method for visualizing and displaying time-series clinical text information of the present invention, in which (1) shows clinical text such as an electronic medical record or a radiology report, (2) shows an image after clinical medical expressions (hereinafter also referred to as medical entities) have been extracted from the clinical text and tagged (classified and identified), (3) shows an image after assigning relationships that hold between the tagged medical entities, and (4) shows an image of clinical text information visualized (displayed) in time series. As shown in Figures 1(1) and 1(2), the method for creating clinical text information time-series data of the present invention extracts and tags clinical medical expressions from clinical text 11 to create annotation files 12, and as shown in Figure 1(3), identifies temporal relationships and associates tags 13 to create a time series of clinical medical expressions. In this way, by creating time-series data of the information contained in the clinical text, the time series of the clinical text information can be visualized as in the time-series display 10 shown in Figure 1(4).

[0028] Fig. 2 is an explanatory diagram of the method for creating clinical text information time-series data of the present invention. As shown in Fig. 2, the method for creating clinical text information time-series data of the present invention extracts and tags at least one clinical medical expression, such as a disease name, a test name, a drug name, a treatment name, or a clinical stage expression, and a time expression contained in clinical text created in a clinical setting, identifies a time relationship representing temporal information between the time expression and the clinical medical expression, and converts the clinical medical expression into time-series data according to the time expression.

[0029] Temporal expressions contained in clinical texts are classified into six categories: "date," "time," "duration," "frequency," "age," and "clinical-specific," or an additional category, "none of the above," for a total of seven categories. 1) Date: Date expressions focused on the daily calendar 2) Time: Expressions that focus on a certain point in time during the day or expressions that express the indefinite present, such as "now" or "present" 3) Period: A period expression that focuses on representing the entire period rather than the two ends of the time axis. 4) Frequency: Frequency set representation focusing on multiple dates, times, and periods 5) Age: Expressions related to age 6) Clinical specific: Time expressions specific to medical care, such as "postoperative" 7) None of the above: If none of the above 1) to 6) applies

[0030] Time expression tags are not assigned to just "before" or "after," but to compound words such as "before the examination," "after surgery," "after resection," "three months later," and "five years ago." Time expression tags are not assigned to particles such as "from," "by," and "until," as in "from October 1st," "from November 11th," and "until December 31st." However, these particles are used to classify time relationships, as described below.

[0031] The identification of the temporal relationship between a time expression and a clinical medical expression is to identify and classify the temporal relationship (temporal relationship) between a clinical medical expression and a time expression in a clinical text. In the following explanation of the temporal relationship, the simultaneous relationship from a clinical medical expression (E1) to a time expression (E2) is referred to as "T AIt is shown as "(E1,E2)".

[0032] In the present invention, a temporal relationship refers to a relationship between a clinical medical expression and a time expression that expresses temporal information, such as when a certain event (clinical medical expression) occurred or when a certain state (clinical medical expression) existed, and clinical medical expressions are classified and identified into the following five types 1) to 5). 1) Simultaneous relationship: A clinical medical expression occurs or exists in a temporal expression. 2) Prior relationship: The clinical medical expression occurs or exists before the temporal expression and does not exist in the temporal expression. 3) Post-relationship: The clinical medical expression occurs or exists later than the temporal expression, but does not exist in the temporal expression. 4) Onset relationship: The clinical medical expression occurs in or exists in the temporal expression and exists after the temporal expression. 5) Termination relationship: The clinical medical expression occurs or exists before the temporal expression and exists until the temporal expression.

[0033] Here, the time expression refers to the time expression tag attached to the text in the clinical text, but also includes the concept of document creation time (DCT). Generally, the DCT does not appear explicitly in the text, but is given as metadata of the text. Below, we will explain the five types of time relationships using example sentences.

[0034] (Regarding simultaneous relations) Time relationship T A(E1, E2) is a temporal relationship that indicates that clinical expression E1 occurred at the time corresponding to temporal expression E2. The start and end points of clinical expression E1 do not need to exactly coincide with temporal expression E2. If the exact start and end points are unknown but are "roughly around temporal expression E2," this also applies to cases where clinical expression E1 can be interpreted as being contained within the time corresponding to temporal expression E2. However, when the temporal relationship between the start and end points of clinical expression E1 and temporal expression E2 is clearly stated in the clinical text, as in cases 1) to 4) below, the relevant temporal relationship (before, after, start, or end) takes precedence over the simultaneous relationship. 1) If the end time of the clinical medical expression E1 is before the end time of the time expression E2, the prior relationship takes precedence. 2) If the start time of the clinical medical expression E1 is later than the time expression E2, the posterior relationship takes precedence. 3) If the start time of the clinical medical expression E1 is the time expression E2, the start relationship takes precedence. 4) If the end time of the clinical medical expression E1 is the time expression E2, the end relationship takes precedence.

[0035] [Table 1]

[0036] In Example 1 above, Test A (= Test Name) was performed on Year / Month / Day (= Time Expression: Date), and a simultaneous relationship is assigned between Test A and the date. In Example 2 above, on Year / Month / Day (= Time Expression: Date), the patient was diagnosed (= Treated) with Symptom B (= Disease Name) and was subsequently hospitalized (= Clinical Expression), and a simultaneous relationship is assigned between the diagnosis, hospitalization, and the date. In Example 3 above, the progress of Symptom C is described, but there is no time expression in the text. In this case, the document creation date and time (DCT: Date / Time) can be used as a time expression, and the follow-up observation (= Treatment) occurred on the document creation date and time. In Example 4 above, Drug D (= Drug Name) was prescribed (= Treated) for two weeks (= Time Expression: Period), and a simultaneous relationship is assigned between Drug D and the period. In example sentence No. 5 above, the last X-ray (date: year / month / date) indicates that the X-ray (= examination name) was performed on year / month / date (= time expression: date), and a simultaneity relationship is established between the examination name and the date.

[0037] As in Example 2 above, the simultaneous relationship applies to many cases, and the temporal sequence and duration are not always clear. In Example 2 above, "Symptom B" may have existed before "Year / Month / Day," and may have continued without being cured. In this case, the simultaneous relationship is also established when the start and end points are unclear and only the fact that it existed on "Year / Month / Day" is clear from the clinical text. Furthermore, as in Example 4, the use of "Drug D" may have occurred multiple times over a two-week period. In other words, although the simultaneous relationship positions clinical expression E1 within temporal expression E2 on the timeline, it does not necessarily represent a single event.

[0038] (Regarding prior relationships) The pre-existence relationship is assigned when the clinical medical expression E1 occurs before the time expression E2 on the time axis, that is, when the end point of the clinical medical expression E1 is before the start point of the time expression E2. In example sentence No. 6 below, hospitalization on a certain date is a simultaneous relationship as described above, but outpatient treatment (=clinical expression) occurred before that date, and a prior relationship is assigned between the clinical expression and the date. Also, in example sentence No. 7 below, symptom A's improvement (=symptom) occurred before that date (=date), and a prior relationship is assigned between the symptom and the date.

[0039] [Table 2]

[0040] (Regarding post-facto relationships) The posterior relationship is assigned when the clinical medical expression E1 occurs after the time expression E2 on the time axis, that is, when the start time of the clinical medical expression E1 is after the end time of the time expression E2. In the example sentence No. 8 below, the fever (= Symptom 1) continued from year / month / date, and the return to normal (= Symptom 2) occurred after year / month / date, so a posterior relationship is assigned between Symptom 2 and the date. In addition, a beginning relationship, which will be described later, is assigned between Symptom 1 and the date.

[0041] [Table 3]

[0042] (Regarding the initiation relationship) The start relation is assigned when the clinical expression E1 is an event that began at the time expression E2 on the time axis, i.e., when the start point of the clinical expression E1 is the time expression E2. In particular, expressions such as "start" and "from" are clues. In the example sentence No. 9 below, the administration of drug C began on year / month / date, so the administration of drug C (= drug name) was an event that began on year / month / date, and a start relationship is assigned between drug C and the date. Note that in the example sentence No. 8 above, fever (symptom 1) was an event that began on the date, and as mentioned above, a start relationship is assigned between fever and the date.

[0043] [Table 4]

[0044] (Regarding termination relationships) The end relation is assigned when the clinical expression E1 is an event that ends at the time expression E2 on the time axis, i.e., when the end point of the clinical expression E1 is the time expression E2. In particular, expressions such as "end" and "by" are clues. In example sentence No. 10 below, Test D was performed on a certain date (= date) (there is a contemporaneous relationship between Test D and the date), and as a result, home oxygen therapy (= treatment) was no longer necessary and ended. Therefore, home oxygen therapy (= treatment) is an event that ended on a certain date, and an end relationship is assigned between the date and the home oxygen therapy. In example sentence No. 11 below, the base of the big toe has been swollen (= symptom) for the past two weeks (= period) on a certain date (= date), and since the symptom has existed for two weeks prior to that date, an end relationship is assigned between the symptom and the date. In example sentence No. 12 below, there has been no change in the shadow on the MRI image (= symptom) for the past two months (= period), and although a specific date is not specified, the date is determined from the DCT scan, and since there has been no change for two months prior to the date, an end relationship is assigned between the symptom and the date.

[0045] [Table 5]

[0046] There are two types of time expressions: absolute time expressions that correspond uniquely to a point on the timeline by themselves, and relative time expressions that do not. An absolute time expression is a specific date such as "October 1, 2021." DCT is also an absolute time expression. On the other hand, relative time expressions are expressions such as "last time" or "after surgery," which cannot determine an exact date and time by themselves. Relative time expressions cannot determine their position on the timeline by themselves, so they must be given a temporal relationship and linked to an absolute time expression. When an appropriate absolute expression does not exist, it is sometimes possible to partially express the temporal relationship by assigning as many temporal relationships between relative time expressions as possible. Among relative temporal expressions, expressions specific to clinical practice include "postoperative," "postprocedure," and "postresection." For example, "postoperative" literally refers to the period after the end of surgery, but here it is interpreted as "a certain period after surgery." In other words, it does not refer to the long period from the end of surgery to the present, but rather to the limited period during which the surgery had an impact. When assigning a temporal relationship, the context determines whether the period falls within the period during which the effects of "postoperative" are continuing, and then one of the temporal relationships to assign (simultaneous, before, after, start, or end) is selected accordingly.

[0047] The aforementioned status attribute is assigned to medical entities such as test names and procedures (diagnosis, treatment, etc.). The value of the status attribute represents the status at a specific point in time in the surrounding context, and is determined relative to time. Therefore, even within the same sentence, a test that is "scheduled" at one point may be "performed" at a later point. When annotating temporal relationships between medical entities with state attributes, it is necessary to determine the type of temporal relationship based on the value of the state attribute so that it is consistent with the order relationship on the actual time axis. Tables 6 to 8 below show examples of clinical medical expressions, showing temporal interpretations for tests, procedures, and medicines, for combinations of three types of state attributes (scheduled, performed, not performed) and five types of time relationships.

[0048] [Table 6]

[0049] [Table 7]

[0050] [Table 8]

[0051] Figure 3 is an explanatory diagram of a method for creating clinical text information time-series data. Time-series display 10a in Figure 3 shows an example of time-series display 10. As described above, clinical expressions and time expressions are extracted and tagged, the time relationship representing the temporal information between the time expressions and the clinical expressions is identified, and the clinical expressions are converted into time-series data along the time expressions. By doing so, absolute time expressions that uniquely correspond to a certain point on the time axis ("September 17th," "October 3rd") and relative time expressions that do not ("after," "thereafter") are converted into time series as in time-series display 10a, and a schematic diagram of the clinical expressions is displayed in correspondence with the time sequence (time axis) of the time expressions. [Example]

[0052] (Clinical text information time series data creation device and clinical text information time series visualization display device) FIG. 4 shows a functional block diagram of the clinical text information time-series data generating device and the clinical text information time-series visualization display device of the first embodiment. As shown in FIG. 4, the clinical text information time-series data creation device 1 of the first embodiment includes a clinical text information extraction means 5, an input means 2, a time relation identification means 3, and a time-series data creation means 4. The clinical text information extraction means 5 inputs clinical text created in clinical settings, and extracts and tags at least one clinical medical expression and time expression, including disease name, test name, drug name, treatment name, and clinical stage expression, contained in the clinical text using a trained language processing model. The input means 2 is used to input annotation files of clinical texts tagged with clinical medical expressions and time expressions. Therefore, the clinical text information time-series data creation device of the first embodiment is configured such that clinical texts created in clinical settings can be input using the clinical text information extraction means 5 to be converted into time-series data 14, and that annotation files can be input using the input means 2 to be converted into time-series data 14. The temporal relationship identifying means 3 identifies the temporal relationship between the time expressions and the clinical medical expressions for the annotation file extracted and tagged by the clinical text information extracting means 5 or the input annotation file. The time-series data generating means 4 converts clinical medical expressions into time-series data 14 along a time representation.

[0053] Moreover, in the first embodiment, the clinical text information time-series visualization display device 60 is further provided. The clinical text information time-series visualization display device 60 is provided with a display means 6. The display means 6 is a display terminal connected to the clinical text information time-series data creation device 1 directly or via a network, and displays a time display axis according to a time expression for the time-series data 14 generated by the clinical text information time-series data creation device 1, and displays clinical medical expressions corresponding to the time display axis for each tag type according to a time relationship. In the time series display, the time points (including relative time expressions in addition to absolute time expressions) described in the clinical text are arranged in chronological order on the horizontal axis, and the types of extracted clinical medical expressions are arranged on the vertical axis. Each type of clinical medical expression is then placed in the corresponding period on the horizontal axis, thereby displaying the clinical medical facts in the clinical text in chronological order.

[0054] (Clinical text information time series visualization system) Fig. 5 shows a configuration diagram of the clinical text information time series visualization system of Example 1. As shown in Fig. 5, in the clinical text information time series visualization system 101, a server 7 and a client terminal 8 are connected to a network 9, enabling data transmission and reception. The server 7 inputs clinical text created in a clinical setting, extracts and tags at least one clinical medical expression and time expression, including disease name, test name, drug name, treatment name, and clinical stage expression, contained in the clinical text using a trained language processing model, and creates an annotation file of the tagged clinical text.The server 7 identifies, for the annotation file, the time relationship representing the temporal information between the time expression and the clinical medical expression, converts the clinical medical expression into time-series data according to the time expression, and outputs the converted time-series data. The client terminal 8 outputs clinical text to the server 7, inputs time series data from the server 7, displays a time display axis according to the time expression, and displays clinical medical expressions in correspondence with the time display axis for each tag type according to the time relationship. That is, unlike the clinical text information time-series visualization system 102 of Example 5 described later, the clinical text information time-series visualization system 101 has a configuration in which a first server for medical entity annotation and a second server for medical entity relationship annotation are integrated, thereby enabling the system to have a simple configuration. Although a notebook PC is shown as the client terminal 8 here, a wide range of terminals can be used, including other PCs such as desktop PCs, tablet terminals, and smartphones.

[0055] Fig. 6 shows a data flow diagram of the clinical text information time series visualization system of Example 1. As shown in Fig. 6, the client terminal 8 outputs clinical text to the server 7 (step S01). The clinical text is input to the server 7 (step S02). In the server 7, clinical medical expressions and time expressions are extracted (step S03), tagged, and an annotation file of the tagged clinical text is generated (step S04). The server 7 identifies a time relationship representing temporal information between the time expressions and the clinical medical expressions for the annotation file of the tagged clinical text (step S05). The server 7 converts the clinical medical expressions into time-series data according to the time expressions (step S06), and outputs the converted time-series data (step S07). The client terminal 8 receives the time series data from the server 7 (step S08) and displays the data in time series (step S09).

[0056] In Example 1, we describe an example in which the clinical text information time-series data creation method and visualization display method of the present invention were applied to medical record text corresponding to the free-text portion of an electronic medical record or radiological findings. In both experiments, a pre-trained machine learning language model, BERT (Bidirectional Encoder Representations from Transformers), was used to identify and tag clinical medical expressions, including disease names, drug names, test names, procedure names, and clinical stage expressions, and time expressions from the input clinical text. The temporal relationships (five types) between the tagged clinical medical expressions and the time expressions were identified, converted into time-series data, and displayed in a time series. BERT is a natural language processing model published in a paper by Jacob Devlin et al. at Google in 2018. It uses AI (artificial intelligence) technology to process words used in human language (natural language) by creating distributed representations that replace them with high-dimensional vectors. Table 9 below shows the learning parameters of the machine learning language model BERT.

[0057] [Table 9]

[0058] (Example of experiment using medical records) Figures 7 to 11 are conceptual diagrams of tagging in an experiment using medical records, with (1) showing the clinical text and (2) showing the tagged text file. Note that Figures 7 to 11 are a single clinical text as a whole, but are shown divided for ease of explanation. As shown in (1) of Figures 7 to 11, the clinical text is written in free text. As shown in (2) of Figure 7, in the tagged text file, the phrase "this time" is tagged with a time expression (date). Furthermore, the phrases "advanced gastric cancer" and "S-1 + CDDP therapy" are each tagged with a clinical medical expression. In the tagged text file shown in Figure 8(2), the entry "39 years old" is tagged with a time expression (age). The entries "February 2008," "August," and "September" are tagged with a time expression (date). The entries "health checkup" and "abnormal findings" are also tagged with clinical medical expressions.

[0059] Here, the time relationship between the time expression "February 2008" (date) and the clinical medical expressions "health checkup" (examination) and "abnormal findings" (disease) is identified as a "simultaneous relationship." The temporal relationship between the time expression "August" (date) and the clinical medical expression "waist circumference" (examination) is identified as a "simultaneous relationship." The temporal relationship between the time expression "August" (date) and the clinical medical expressions "abdominal fullness" (illness) and "loss of appetite" (illness) is identified as a "starting relationship" because "since August." The temporal relationship between the time expression "September" (date) and the clinical medical expression "nausea" (illness) is identified as an "onset relationship." The temporal relationship between the time expression "September" (date) and the clinical medical expression "visit a nearby doctor" (clinical specific) is identified as a "simultaneity relationship."

[0060] In the tagged text file shown in Figure 9(2), the statements "September 17th of the same year," "After that," and "October 3rd" are tagged with time expressions (date). The statement "2 weeks" is tagged with time expressions (period), and the statement "at the beginning of treatment" is tagged with time expressions (clinical specific). In addition, "S-1" and "80 mg / m 2 " and other statements are also tagged with clinical medical expressions.

[0061] Here, the temporal relationship between the time expression "September 17th of the same year" (date) and the clinical medical expression "5-HT3 receptor antagonist" (administration) is identified as a "simultaneous relationship," and the temporal relationship between the clinical medical expressions "S-1" (administration), (first line) "CDDP" (administration), and (third line) "CDDP" (administration) is identified as a "start relationship." The temporal relationship between the time expression "after" (date) and the clinical medicine expression "Indisetron" (medication) is identified as an "initiation relationship," and the temporal relationship between the time expression "medication instructions" (clinical) and the clinical medicine expression "simultaneous relationship." The time relationship between the time expression "October 3rd" (date) and the clinical medical expression "total parenteral nutrition (TPN)" (treatment) is identified as a "start relationship."

[0062] In the tagged text file shown in Figure 10(2), the phrase "afterwards" is tagged with a time expression (date), and the phrase "after the sixth course" is tagged with a time expression (clinical specific). Furthermore, phrases such as "oral nutritional supplement" and "TPN" are also tagged with clinical medical expressions.

[0063] Here, the time relationship between the time expression "October 3rd" (date) shown in Figure 9(2) and the clinical medical expressions "oral nutritional supplement" (medication), "nutrient supplement" (medication), and "TPN" (treatment) shown in Figure 10(2) is identified as a "start relationship." The time relationship between the time expression "then" (date) and the clinical medical expressions "nutritional agent" (medication) and "main lesion" (disease) is identified as a "simultaneous relationship." The temporal relationship between the time expression "after the sixth course of treatment" (clinical specific) and the clinical medical expression "fever, bone marrow suppression" (illness) is identified as a "simultaneous relationship," and the temporal relationship between the time expression "chemotherapy" (clinical) and the clinical medical expression "onset relationship" is identified as a "start relationship."

[0064] In the tagged text file shown in Figure 11(2), the phrase "this time" is tagged with a time expression (date). Also, phrases such as "S-1+CDDP therapy" and "medication instructions" are tagged with clinical medical expressions.

[0065] Figure 12 is an image of time series visualization in an example experiment using medical records, showing the time series data visualized and displayed on a display (not shown). In the figure, "+" indicates positive (positive findings of disease), "-" indicates negative (negative findings of disease), and check marks indicate that tests or medications have been performed. As shown in Figure 12, clinical medical expressions are organized and displayed in an easy-to-understand manner, categorized into categories such as "Examination," "Treatment," "Disease," and "Clinical." At the top, time expressions such as "February 2008" and "August" are displayed in chronological order from left to right. For example, in the clinical text shown in Figure 8(1), it is simply written in free text that "No abnormal findings were found in the medical checkup in February 2008." However, in Figure 12, it is clear at a glance that a "medical checkup" was conducted as an examination in "February 2008" and that there were "no (-)" abnormal findings as an "illness."

[0066] Regarding time expressions such as "after" and "thereafter," understanding the temporal relationship from free-text clinical text requires detailed reading of the clinical text. However, in Figure 12, it is easy to see that the description "after" shown in Figure 9(1) means after "September 17th of the same year" and before "October 3rd," and it is possible to confirm at a glance when the "medication instructions" were given. Furthermore, time expressions such as "August" and "September" alone do not indicate which "August" or "September" refers to, but by using the method for creating and visualizing clinical text information time series data of the present invention, it is possible to accurately convert the time relationship with other time expressions into time series data and visualize it. The intervals between the time expressions as shown in FIG. 12 may be adjusted to match the actual time intervals, or may be adjusted to match the amount of clinical medical description.

[0067] Table 10 below shows the performance values ​​of the clinical text information time-series data creation method of the present invention when using medical records or radiological findings. Expression recognition is the average recognition rate for a total of 10 types: nine clinical medical expressions (disease name, test name, drug name, procedure name, clinical stage expression, site expression, change expression, feature expression, and pending expression) and temporal expressions (including date, time, period, frequency, age, clinically specific, and seven types that do not apply to any of the above). Temporal relationship recognition shows the recognition rate and average recognition rate for each of five types (simultaneous relationship, before relationship, after relationship, start relationship, and end relationship). As shown in Table 10 below, when medical records were used, by using the method for creating clinical text information time-series data of the present invention, an average F1 value of 0.86 or more was achieved for 10 categories of clinical medical expressions in the recognition of clinical medical expressions, and an F1 value of approximately 0.7 or more was achieved in the recognition of simultaneous relationships, which are the main time relationships, and it was found that the method is practical for use with a variety of clinical texts. Therefore, by using the method for creating clinical text information time-series data and the method for visualizing and displaying the same according to the present invention, accurate and prompt diagnosis and treatment are possible.

[0068] [Table 10]

[0069] (Example of experiment using radiological findings) Next, we will explain an example in which the clinical text information time-series data creation method and visualization display method of the present invention are used for radiographic findings text. Figures 13 and 14 are tagging images for an experimental example using radiographic findings, with (1) showing the clinical text and (2) showing the tagged text file. Note that Figures 13 and 14 are a single clinical text as a whole, but are shown divided for ease of explanation. As shown in Figures 13(1) and 14(1), in this experimental example, the clinical text is written in free text. In the tagged text file shown in Figure 13(2), the phrases "one year ago," "last time" (on the third line), and "last time" (on the fourth line) are tagged with time expressions (dates). Furthermore, phrases such as "CT" and "left upper segment S1+2" are also tagged with clinical expressions. In contrast, the tagged text file shown in Figure 14(2) does not have time expressions, but phrases such as "mediastinum, including the right paratracheal region" and "small lymph nodes" are tagged with clinical expressions.

[0070] Here, the time relationship between the time expression "one year ago" (date) shown in FIG. 13(2) and the clinical medical expression "CT" (examination) is identified as a "simultaneous relationship" because of "one year ago." The temporal relationship between the DCT (document creation date and time) and the clinical medical expressions of "partsolidGGN" (site), "ground-glass opacity" (site), "bulla" (site), "nodule" (site), "linear trabecular opacity" (site), "small lymph node" (site), "enlargement" (site), and "GGN" (site) is identified as "concurrent relationship." In addition, the temporal relationship between DCT (document creation date and time) and the clinical medical expressions of "tumorous lesions such as adenocarcinoma in situ (AIS) and microinvasive adenocarcinoma (MIA)" (disease), "old inflammatory changes" (disease), (Figure 13(2) line 8) "hiatal hernia" (disease), "pleural effusion" (disease), "splenomegaly" (disease), "AIS and MIA" (disease), and (Figure 14(2) line 6) "hiatal hernia" (disease) is identified as a "concurrent relationship."

[0071] Figure 15 is a time-series visualization image of an experimental example using radiological findings, showing the time-series data visualized on a display. As shown in Figure 15, the data is categorized into areas related to location ("Left upper segment S1+2," "Both lung apexes," "Right upper and middle lobes of the lung and left lower lobe," "Right paratracheal and other mediastinum," and "Left upper segment of the lung"), as well as "Examination" and "Disease," and the clinical medical expressions are organized and displayed in an easy-to-understand manner. At the top, time representations of "1 year ago" and "DCT" (date and time of creation) are displayed in chronological order. In the figure, "+" indicates positive (positive findings for disease), "-" indicates negative (negative findings for disease), and "?" indicates suspicion, and check marks indicate that tests or medication have been performed. For example, in the clinical text shown in Figure 13(1), it is simply written in free text, "Compared with the CT from one year ago." However, in Figure 15, "DCT" (date and time of creation) is displayed approximately in the center of the chronological display, with "one year ago" displayed to the left. Furthermore, as an "examination" item, "CT" is displayed below the "one year ago" display, making it clear at a glance that a comparison is being made with the CT from one year ago.

[0072] Furthermore, it is possible to see at a glance the various findings in the "DCT" (document creation date and time), for example, that "small nodules" are found in the "right upper and middle lobes and left lower lobe of the lung," and that "linear trabecular shadows" are found on the "left side" of the "right upper and middle lobes and left lower lobe of the lung." Furthermore, in the clinical text shown in Figure 13(1), the passage that reads, "A partsolid GGN measuring 18 mm in long axis in S1+2 of the left upper segment of the lung has a reduced internal solid portion compared to the previous examination," is illustrated as shown in Figure 15. Although not shown here, a gradation or other display method may be used to visualize the gradual reduction in the findings.

[0073] As shown in Table 10 above, when radiological findings were used, it was found that by using the method for creating clinical text information time-series data of the present invention, the recognition performance of clinical medical expressions and time relationships from radiological findings text exceeded an average F1 value of 0.9. [Example]

[0074] Fig. 16 shows a functional block diagram of a clinical text information time-series data creation device of Example 2. As shown in Fig. 16, a clinical text information time-series data creation device 1a includes an input means 2, a time relation identification means 3, and a time-series data creation means 4. The input means 2 inputs an annotation file 12 of clinical text tagged with clinical medical expressions and time expressions. The time relationship identification means 3 identifies the time relationship between the time expressions and the clinical medical expressions for the input annotation file 12. The time series data generation means 4 generates time series data 14 from the clinical medical expressions according to the time expressions. [Example]

[0075] FIG. 17 shows a functional block diagram of the clinical text information time-series data generating device. 17, the clinical text information time-series data creation device 1 includes a clinical text information extraction means 5, an input means 2, a time relation identification means 3, and a time-series data creation means 4. The configuration of the clinical text information time-series data creation device 1 is the same as that of Example 1. Unlike Example 1, Example 3 does not include a clinical text information time-series visualization display device 60. [Example]

[0076] FIG. 18 shows a functional block diagram of a clinical text information time-series data generating device and a clinical text information time-series visualization display device according to the fourth embodiment. As shown in FIG. 18, the clinical text information time-series data creating device 1a of the fourth embodiment has the same configuration as the clinical text information time-series data creating device 1a of the second embodiment. However, the fourth embodiment further includes a clinical text information time-series visualization display device 60, which includes a display means 6. The configuration of the clinical text information time-series visualization display device 60 is the same as that of the first embodiment. As described above, the clinical text information time-series data generating device and the clinical text information time-series visualizing and displaying device of the present invention can employ a variety of configurations. [Example]

[0077] Fig. 19 shows a configuration diagram of a clinical text information time series visualization system of Example 5. As shown in Fig. 19, in a clinical text information time series visualization system 102, a first server 7a, a second server 7b, and a client terminal 8 are connected to a network 9, enabling data transmission and reception. The first server 7a is a server for medical entity annotation that inputs clinical text created in clinical settings, extracts and tags at least one clinical medical expression and time expression, including disease name, test name, drug name, treatment name, and clinical stage expression, contained in the clinical text using a trained language processing model, and outputs an annotation file of the tagged clinical text. The second server 7b is a server for annotating relationships between medical entities that inputs annotation files of clinical text from the first server 7a, identifies temporal relationships representing temporal information between time expressions and clinical medical expressions in the annotation files, converts the clinical medical expressions into time-series data according to the time expressions, and outputs the converted time-series data. The client terminal 8 outputs clinical text to the first server 7a, inputs time series data from the second server 7b, displays a time display axis according to the time expression, and displays clinical medical expressions corresponding to the time display axis for each tag type according to the time relationship.

[0078] Fig. 20 shows a data flow diagram of the clinical text information time series visualization system of Example 5. As shown in Fig. 20, the client terminal 8 outputs clinical text to the first server 7a (step S101). The clinical text is input to the first server 7a (step S102). In the first server 7a, clinical medical expressions and time expressions are extracted (step S103) and tagged (step S104). An annotation file of the tagged clinical text is output to the second server 7b (step S105). The annotation file of the tagged clinical text is input to the second server 7b (step S106). The second server 7b identifies a time relationship representing temporal information between the time expression and the clinical medical expression for the annotation file (step S107). The second server 7b converts the clinical medical expression into time-series data according to the time expression (step S108) and outputs the converted time-series data (step S109). The client terminal 8 receives the time series data from the second server 7b (step S110) and displays the data in time series (step S111). [Industrial Applicability]

[0079] The present invention is useful for electronic medical record systems and the like. [Explanation of symbols]

[0080] 1,1a Clinical text information time series data creation device 2. Input Methods 3. Temporal relationship identification means 4. Time series data generation method 5. Clinical text information extraction method 6 Display means 7 Server 7a First Server 7b Second Server 8. Client Terminal 9 Network 10,10a Time series display 11 Clinical Texts 12 Annotation files 13 Relationships between tags 14 Time Series Data 60 Clinical text information time series visualization display device 101,102 Clinical Text Information Time Series Visualization System

Claims

1. A method for creating clinical text information time series data, characterized in that a computer extracts and tags at least one clinical medical expression, such as a disease name, a test name, a drug name, a treatment name, or a clinical stage expression, and a time expression contained in clinical text created in a clinical setting, wherein the time expressions include absolute time expressions that correspond uniquely to a certain point on a time axis on their own, and relative time expressions that do not correspond uniquely, and a time relationship that represents temporal information between the relative time expression and the clinical medical expression based on its relationship with other time expressions, and converts the clinical medical expression into time series data along the time expression.

2. The time relationship is such that the clinical medical representation: 1) occurring or present in said time representation; 2) occurs or exists before the time expression and is not present in the time expression; 3) occurs or exists after the time expression and is not present in the time expression; 4) occurs or exists in the time expression and exists after the time expression; 5) occurring or existing before the time expression and existing until the time expression; 2. The method for creating clinical text information time-series data according to claim 1, wherein the clinical text information is classified and identified into five types:

3. 3. The method for creating clinical text information time-series data according to claim 1, wherein the time expressions are classified and identified into at least six types of expressions: date, time, period, frequency, age, and clinically specific expressions.

4. A method for creating clinical text information time series data according to any one of claims 1 to 3, characterized in that the time expression includes the creation date and time of the clinical text, and the information is added as metadata of the clinical text.

5. 5. The method for creating clinical text information time-series data according to claim 1, wherein the tagging types of the clinical medical expressions further include at least one of site expressions, change expressions, and feature expressions.

6. A method for displaying time-series data of clinical medical expressions in a clinical text created using the method for creating clinical text information time-series data according to any one of claims 1 to 5, comprising: displaying a time display axis according to the time representation; and displaying the clinical medical expressions for each tag type in accordance with the time relationship in correspondence with the time display axis. A method for visualizing and displaying clinical text information in a time series.

7. an input means for inputting an annotation file of clinical text, which is generated in a clinical setting and which is tagged with at least one clinical medical expression, such as a disease name, a test name, a drug name, a treatment name, and a clinical stage expression, and a time expression; a time relationship identification means for identifying a time relationship representing temporal information between the time expression and the clinical medical expression for the input annotation file; a time-series data generating means for generating time-series data from the clinical medical expression along the time expression; Equipped with The time expressions include absolute time expressions that uniquely correspond to a certain point on the time axis by themselves, and relative time expressions that do not uniquely correspond, and the time relationship identification means identifies the time relationship that represents the temporal information between the relative time expressions and the clinical medical expressions based on their relationship with other time expressions.

8. 8. The clinical text information time-series data creation device according to claim 7, further comprising a clinical text information extraction means for inputting the clinical text and extracting and tagging at least one clinical medical expression, including a disease name, a test name, a drug name, a treatment name, and a clinical stage expression, and a time expression contained in the clinical text, using a trained language processing model.

9. A display terminal connected directly or via a network to the clinical text information time-series data creation device according to claim 7 or 8, A clinical text information time series visualization display device comprising: a display means for displaying a time display axis in accordance with the time expression, and displaying the clinical medical expressions in correspondence with the time display axis for each tag type in accordance with the time relationship.

10. a first medical entity annotation server that receives input of clinical text created in a clinical setting, extracts and tags at least one clinical medical expression, such as a disease name, a test name, a drug name, a treatment name, and a clinical stage expression, and a time expression, contained in the clinical text using a trained language processing model, and outputs an annotation file of the tagged clinical text; a second server for medical entity relationship annotation that receives the annotation file of the clinical text from the first server, identifies a time relationship representing temporal information between the time expression and the clinical medical expression for the annotation file, converts the clinical medical expression into time-series data according to the time expression, and outputs the converted time-series data; a client terminal that outputs the clinical text to the first server, receives the time series data from the second server, displays a time display axis in accordance with the time expression, and displays the clinical medical expression in correspondence with the time display axis for each tag type in accordance with the time relationship, and is connected to a network; The time expressions include absolute time expressions that uniquely correspond to a certain point on a time axis by themselves, and relative time expressions that do not uniquely correspond, and the second server identifies a time relationship that represents temporal information between the relative time expressions and the clinical medical expressions based on their relationship with other time expressions.

11. The clinical text information time series visualization system according to claim 10 , wherein the first server and the second server are integrated.

Images