Methods, apparatus, devices, and readable storage media for extracting clinical finding events

Abstract

The application provides a method, device, equipment and readable storage medium for extracting a clinical finding event, comprising: acquiring a disease description text, adding a string to the front of the disease description text to generate a first text; inputting the first text into a trained deep learning model for prediction and acquiring a prediction result; restoring the prediction result into a plurality of triplets, wherein each triplet comprises the start and end positions of a subject word in the disease description text and the start and end positions of a value corresponding to each category in the disease description text; and generating a clinical finding event composed of a plurality of four-tuple data according to the plurality of triplets. The problems that a corresponding case cannot be extracted according to the description of a patient or the extracted case is inaccurate are solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of artificial intelligence, and particularly relates to a method and device for extracting clinical discovery events, equipment and a readable storage medium. BACKGROUND

[0002] With the continuous development of medical informatization, especially the gradual popularization of Internet hospitals and medical big data analysis, in many application scenarios, algorithms and machines are needed to analyze and understand medical texts such as patient condition descriptions, thereby providing data support for the next diagnosis and treatment action or scientific research activity. For example, in pre-examination triage, clinical discovery events are extracted from patient condition descriptions to provide a basis for intelligent pre-examination triage; in the medical big data analysis scene, clinical discovery events are extracted from patient medical histories and medical examination reports to structure and process information of medical texts, and to provide data preparation for subsequent data analysis. However, in the prior art, there is a problem that the corresponding case cannot be extracted according to the patient's description, or the extracted case is inaccurate.

[0003] Therefore, the present application is proposed. SUMMARY

[0004] The present application discloses a method and device for extracting clinical discovery events, equipment and a readable storage medium, which aims to solve the problem that the corresponding case cannot be extracted according to the patient's description, or the extracted case is inaccurate.

[0005] The first embodiment of the present application provides a method for extracting clinical discovery events, comprising:

[0006] Obtaining a condition description text, and adding a first string to the front of the condition description text to generate a first text;

[0007] Inputting the first text into a trained deep learning model for prediction, and obtaining a prediction result;

[0008] Restoring the prediction result into a plurality of triplets, wherein each triplet comprises the start and end positions of a subject word in the condition description text, and the start and end positions of the value corresponding to each category in the condition description text;

[0009] According to the plurality of triplets, a clinical discovery event composed of a plurality of four-tuple data is generated.

[0010] Preferably, before the first text is input into the trained deep learning model for prediction and the prediction result is obtained, the deep learning model is trained, specifically:

[0011] Obtaining a condition description text in the training data, and converting the condition description text into a text id and a text vector;

[0012] According to the head and tail positions of the subject word, the subject word text vector corresponding to the head and tail of the subject word is extracted from the text vector;

[0013] The subject word text vector is conditionally normalized, and a normalized vector is generated;

[0014] The normalized vector is input into a fully connected layer and a binary classifier to predict the head and tail positions of the value corresponding to each category of the subject word.

[0015] Preferably, the disease description text in the training data is obtained, and the disease description text is converted into a text vector, specifically:

[0016] The disease description text in the training data is obtained, and the disease description text is converted into a text id of a BERT model, and the text id of the BERT model is transmitted to an encoder of the BERT;

[0017] The text vector generated by the encoder is obtained.

[0018] Preferably, the clinical finding event composed of a plurality of four tuple data is generated according to a plurality of the triplets, specifically:

[0019] The triplets of the subject word with the same head and tail positions are merged, and the values of the same category are merged to generate a four tuple data of a clinical finding event;

[0020] The first string describing the word is replaced with an empty string, and the replacement word appearing in the subject word is replaced in reverse according to the lookup table, and the virtual word in the occurrence state is replaced in reverse to generate the clinical finding event composed of a plurality of four tuple data.

[0021] The second embodiment of the present application provides a device for extracting a clinical finding event, comprising:

[0022] A disease description text acquisition unit is configured to acquire a disease description text, and add a first string to the front of the disease description text to generate a first text;

[0023] A prediction result acquisition unit is configured to input the first text into a trained deep learning model for prediction, and acquire a prediction result;

[0024] A restoration unit is configured to restore the prediction result into a plurality of triplets, wherein each triplet includes the head and tail positions of the subject word in the disease description text, and the head and tail positions of the value corresponding to each category in the disease description text;

[0025] The clinical finding event generation unit generates a clinical finding event composed of a plurality of four-tuple data according to the plurality of three-tuples.

[0026] Preferably, the prediction result acquisition unit further comprises a deep learning model training module, specifically configured to:

[0027] The disease description text in the training data is acquired, and the disease description text is converted into a text id and a text vector.

[0028] The head and tail positions of the subject word are generated according to the text id, and the subject word text vector corresponding to the head and tail of the subject word is extracted from the text vector according to the head and tail positions of the subject word.

[0029] The subject word text vector is conditionally normalized to generate a normalized vector.

[0030] The normalized vector is input into a fully connected layer and a binary classifier to respectively predict the head and tail positions of the value corresponding to each category of the subject word.

[0031] Preferably, the disease description text in the training data is acquired, and the disease description text is converted into a text vector, specifically as follows:

[0032] The disease description text in the training data is acquired, the disease description text is converted into a text id of a BERT model, and the text id of the BERT model is transmitted to an encoder of the BERT.

[0033] The text vector generated by the encoder is acquired.

[0034] Preferably, the clinical finding event generation unit is specifically configured to:

[0035] The three-tuple of the subject word with the same head and tail positions is merged, and the values of the same category are merged to generate a four-tuple data of a clinical finding event.

[0036] The description word is replaced with an empty string, and the replacement word appearing in the subject word and the virtual word in the occurrence state are replaced in reverse according to the replacement table to generate a clinical finding event composed of a plurality of four-tuple data.

[0037] The third embodiment of the present application provides a device for extracting a clinical finding event, comprising a memory and a processor, the memory stores a computer program, and the computer program can be executed by the processor to implement the method for extracting a clinical finding event according to any one of the above.

[0038] The fourth embodiment of the present application provides a computer readable storage medium storing a computer program, the computer program being executable by a processor of a device where the computer readable storage medium is located to implement a method for extracting a clinical finding event according to any one of the preceding claims.

[0039] Based on the method, device, equipment and readable storage medium for extracting a clinical finding event provided by the present application, a first text is generated by adding a string to the front of the obtained disease description text, the first text is predicted by the trained deep learning model to obtain a prediction result, the prediction result is restored into a plurality of triplets, and a clinical finding event composed of a plurality of four-tuple data is generated according to the plurality of triplets, thereby solving the problem that a corresponding case cannot be extracted according to the description of a patient or the extracted case is inaccurate. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of the method for extracting a clinical finding event provided by the first embodiment of the present application;

[0041] Figure 2 is a schematic diagram of a model algorithm provided by the present application;

[0042] Figure 3 is a module schematic diagram of the device for extracting a clinical finding event provided by the first embodiment of the present application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0044] In order to better understand the technical solutions of the present application, the embodiments of the present application will be described in detail below with reference to the drawings.

[0045] It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0046] The terms used in the embodiments of the present application are merely for the purpose of describing the specific embodiments, and are not intended to limit the present application. The singular forms "a", "an" and "the" used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0047] It should be understood that the term "and / or" as used herein merely describes an associated relationship between associated objects, and can represent three relationships, for example, A and / or B can represent three cases of A alone, A and B together, and B alone. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects.

[0048] Depending on the context, the word "if" as used herein can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if it is determined" or "if (a stated condition or event) is detected" can be interpreted as "when it is determined" or "in response to determining" or "when (a stated condition or event) is detected" or "in response to detecting (a stated condition or event)".

[0049] The "first / second" mentioned in the embodiments is only to distinguish similar objects, and does not represent a specific order of the objects. Understandably, the "first / second" can be interchanged in a specific order or sequence as permitted. It should be understood that the objects distinguished by "first / second" can be interchanged under appropriate circumstances to enable the embodiments described herein to be implemented in an order other than those illustrated or described herein.

[0050] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0051] The present application discloses a method, device, equipment and readable storage medium for extracting a clinical finding event, aiming to solve the problem that the corresponding case cannot be extracted according to the description of the patient or the extracted case is inaccurate.

[0052] Please refer to Figure 1 The first embodiment of the present application provides a method for extracting a clinical finding event, which can be executed by an equipment for extracting a clinical finding event (hereinafter referred to as an extraction equipment), and in particular, by one or more processors in the extraction equipment to implement the following steps:

[0053] S101, obtaining a disease description text, and adding a string in the front part of the disease description text to generate a first text;

[0054] In the embodiment, the extraction equipment can be a desktop computer, a notebook computer, a server, a workstation or other terminal with data processing and analysis capability. The extraction equipment can be installed with a corresponding operating system and application software, and the functions required by the embodiment can be realized through the combination of the operating system and the application software.

[0055] It should be noted that the disease description text can be a text input by a doctor to the extraction device during an inquiry. When the doctor inputs the text, a "[unused1]" string can be added in front of the disease description text to generate a first text.

[0056] Specifically, the first text is, for example, "[unused1] 9 days ago, the patient's symptoms of cough and sputum worsened after getting cold again. The patient coughs white and sticky sputum, which is easy to cough out, and the amount is not much. The patient has a fever, chills, and no obvious chill. The highest body temperature measured outside is 39°C. The patient has vomiting, and the vomit is stomach contents, without coffee-colored substances. The patient has no acid regurgitation, no hemoptysis, no chest pain, no left chest pressure, and no left shoulder and left arm medial radiation pain. The patient's spirit is not good, and the sleep is acceptable."

[0057] S102, input the first text into the trained deep learning model for prediction, and obtain a prediction result;

[0058] S103, restore the prediction result to a plurality of triplets, wherein each triplet includes the start and end positions of the subject word in the disease description text, and the start and end positions of the value corresponding to each category in the disease description text;

[0059] S104, according to a plurality of triplets, generate a clinical finding event composed of a plurality of four-tuple data.

[0060] Specifically, the triplets of the subject words with the same start and end positions are combined, and the values of the same category are combined to generate a four-tuple data of a clinical finding event.

[0061] Replace the subject word with an empty string, and replace the replacement word in the subject word and the virtual word in the occurrence state in reverse according to the reference table to generate a clinical finding event composed of a plurality of four-tuple data.

[0062] Convert the clinical finding event in the form of four-tuple to a description in the form of three-tuple. The specific method can be:

[0063] 1. Add the string "[unused1]" at the beginning of the text of the disease description;

[0064] 2. Mark the text of the four attributes of the clinical finding event (subject word, description word, anatomical site, and occurrence state) in the disease description text;

[0065] 3. Define the "category" attribute, which includes three values: "description word", "anatomical site", and "occurrence state".

[0066] 4. For a number of clinical finding events marked in the disease description text, a clinical finding event can be described by a four-tuple: (subject word, description word, anatomical site, occurrence state), which can be converted into 3 triplets:

[0067] (subject word, category == "description word", description word)

[0068] (subject word, category == "anatomical site", anatomical site)

[0069] (subject word, category == "occurrence state", occurrence state)

[0070] For the case where only the subject word is present and the description word is empty, the value of the description word is "[unused1]";

[0071] For the case where the subject word is not in the disease description text, the closest word in the original text is used to replace it, and a table of "subject word" to "replacement word" is collected: where,

[0072] The replacement word table of the subject word is, for example:

[0073] "chills": ["chills"];

[0074] "sleep quality": ["sleep"],

[0075] "mental state": ["spirit"];

[0076] The virtual word table of the occurrence state is, for example: "negative": ["without", "not seen", "not reported", "none", "not prompted", "denied", "not reached", "not again", "not mentioned", "not", "not"],

[0077] "uncertain": ["may be", "consider", "possible", "to be sorted out", "to be checked", "tendency", "suspected", "not excluded", "not completely excluded", "supported"].

[0078] An example is given: from the disease description, the clinical finding event is mined, that is, given a patient's complaint or history of present illness, the clinical finding event is extracted, which contains four attributes: subject word, description word, anatomical site, and occurrence state; Where:

[0079] Subject word: refers to the disease name or symptoms caused by the disease in the patient's disease description;

[0080] Description word: describes the occurrence time characteristics, severity, shape, color, and other dimensions of the subject word;

[0081] Anatomical site: refers to the body part where the subject word occurs;

[0082] Occurrence status: the state described above (if necessary), indicated as "uncertain" or "negative";

[0083] Specifically, for example: 9 days ago, the patient's symptoms of cough and expectoration worsened after being chilled again, with white and sticky sputum, easy to cough out, not much; accompanied by fever, with chills, no obvious chill, the highest body temperature measured outside was 39°C; accompanied by vomiting, the vomit was stomach contents, no coffee-like material; no acid regurgitation, no hemoptysis, no chest pain, no left precordial squeezing sensation and left shoulder and left arm medial radiating pain and other discomforts. The spirit is not good, and the sleep is acceptable.

[0084]

[0085] It should be noted that for the extraction method of this clinical discovery event, a deep learning method is used for extraction, and the model needs to be trained. Therefore, the design and labeling of the training data is the first step.

[0086] From the above examples, some data characteristics in the text can be found as follows, which need to be handled in the design of training data and model design:

[0087] 1. The subject word must exist, and other attributes may not exist;

[0088] 2. In the same event, there may be multiple description words or multiple anatomical sites;

[0089] 3. The text position of the subject word and the description word may overlap;

[0090] 4. The subject word may not exist in the disease description text (for example: "mental state", "sleep quality"), and the description word and the anatomical site both appear in the disease description text;

[0091] 5. The occurrence status has only two values: "negative" and "uncertain", both of which are described by different virtual words. For example, in this example, "negative" is represented by "no".

[0092] Please refer to Figure 2 In a possible embodiment of the present application, before the first text is input into the trained deep learning model for prediction and the prediction result is obtained, the deep learning model is also trained, specifically:

[0093] Obtain the disease description text in the training data, and convert the disease description text into a text id and a text vector;

[0094] Generate the start and end positions of the subject word according to the text id, and extract the subject word text vector corresponding to the start and end of the subject word from the text vector according to the start and end positions of the subject word;

[0095] conditionally normalizing the subject word text vector, and generating a normalized vector;

[0096] inputting the normalized vector into a full connection layer and a binary classifier to respectively predict the start and end positions of the value corresponding to each category of the subject word.

[0097] In a possible embodiment of the present application, the disease description text in the training data is obtained, and the disease description text is converted into a text vector, specifically:

[0098] The disease description text in the training data is obtained, the disease description text is converted into a text id of a BERT model, and the text id of the BERT model is transmitted to an encoder of the BERT.

[0099] The text vector generated by the encoder is obtained.

[0100] It should be noted that the BERT model is a BERT pre-training language model, and in other embodiments, it can also be a language model improved based on BERT, such as RoBERTa, ALBERT, MacBERT, etc., which extracts features of the text to generate a text vector.

[0101] Referring to Figure 3 The second embodiment of the present application provides a device for extracting a clinical finding event, comprising:

[0102] A disease description text acquisition unit 201 is configured to obtain a disease description text, and add a string to the front of the disease description text to generate a first text.

[0103] A prediction result acquisition unit 202 is configured to input the first text into a trained deep learning model for prediction, and obtain a prediction result.

[0104] A restoring unit 203 is configured to restore the prediction result into a plurality of triplets, wherein each of the triplets comprises the start and end positions of a subject word in the disease description text, and the start and end positions of a value corresponding to each category in the disease description text.

[0105] A clinical finding event generation unit 204 is configured to generate a clinical finding event composed of a plurality of four-tuple data according to the plurality of triplets.

[0106] Preferably, the prediction result acquisition unit further comprises a deep learning model training module, specifically configured to:

[0107] The disease description text in the training data is obtained, and the disease description text is converted into a text id and a text vector.

[0108] According to the first and last positions of the subject word, a subject word text vector corresponding to the first and last positions of the subject word is extracted from the text vector;

[0109] The subject word text vector is conditionally normalized to generate a normalized vector;

[0110] The normalized vector is input into a fully connected layer and a binary classifier to predict the first and last positions of the value corresponding to each category of the subject word.

[0111] Preferably, the disease description text in the training data is obtained, and the disease description text is converted into a text vector, specifically:

[0112] The disease description text in the training data is obtained, the disease description text is converted into a text id of a BERT model, and the text id of the BERT model is input into an encoder of the BERT;

[0113] The text vector generated by the encoder is obtained.

[0114] Preferably, the clinical finding event generation unit is specifically configured to:

[0115] The triplets of subject words with the same first and last positions are merged, and the values of the same category are merged to generate a four-tuple data of a clinical finding event;

[0116] The description word "[unused1]" is replaced with an empty string, and the replacement word appearing in the subject word and the virtual word in the occurrence state are replaced in reverse according to the lookup table to generate a clinical finding event composed of multiple four-tuple data.

[0117] The third embodiment of the present application provides a device for extracting a clinical finding event, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement the method for extracting a clinical finding event according to any one of the above embodiments.

[0118] The fourth embodiment of the present application provides a computer readable storage medium storing a computer program, and the computer program can be executed by a processor of a device where the computer readable storage medium is located to implement the method for extracting a clinical finding event according to any one of the above claims.

[0119] Based on the method, device, equipment and readable storage medium for extracting a clinical finding event provided by the application, a first text is generated by adding a string to the front of the acquired disease description text, the first text is predicted by a trained deep learning model to obtain a prediction result, the prediction result is restored into a plurality of triplets, and a clinical finding event composed of a plurality of four-tuple data is generated according to the plurality of triplets, thereby solving the problems that a corresponding case cannot be extracted according to the description of a patient or the extracted case is inaccurate.

[0120] Exemplarily, the computer program in the third and fourth embodiments of the application can be divided into one or more modules stored in the memory and executed by the processor to complete the application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the device for extracting a clinical finding event. For example, the device in the second embodiment of the application.

[0121] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the method for extracting a clinical finding event, and is connected with all parts of the method for extracting a clinical finding event through various interfaces and lines.

[0122] The memory can be configured to store the computer programs and / or modules, and the processor realizes various functions of the method for extracting a clinical finding event by running or executing the computer programs and / or modules stored in the memory and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, a text conversion function, etc.), and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, text message data, etc.), and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0123] The modules implemented can be stored in a computer readable storage medium if they are implemented in the form of software function units and sold or used as independent products. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of the above-mentioned various method embodiments when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0124] It should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate units can or can not be physically separate, and the units shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. In addition, the connection relationship between the modules in the apparatus embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0125] The above description is only a preferred embodiment of the present application, but the protection scope of the present application is not limited thereto, and any changes or replacements within the technical range disclosed by the present application can be easily thought of by those skilled in the art, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of extracting clinical finding events, characterized by, The method comprises the following steps: obtaining a disease description text, and adding a first string to the front of the disease description text to generate a first text; inputting the first text into a trained deep learning model for prediction, and obtaining a prediction result; restoring the prediction result into a plurality of triplets, wherein each triplet comprises a head-tail position of a subject word in the disease description text and a head-tail position of a value corresponding to each category in the disease description text; generating a clinical finding event composed of a plurality of four-tuple data according to the plurality of triplets, specifically: merging triplets of subject words with the same head-tail position, and merging values of the same category to generate a four-tuple data of a clinical finding event; replacing the description word with the first string with an empty string, and replacing the replacement word in the subject word and the virtual word in the occurrence state in reverse according to a reference table to generate a clinical finding event composed of a plurality of four-tuple data.

2. The method of extracting clinical finding events of claim 1, wherein, Before the step of inputting the first text into the trained deep learning model for prediction and obtaining the prediction result, the method further comprises training the deep learning model, specifically: obtaining a disease description text in training data, and converting the disease description text into a text id and a text vector; generating a head-tail position of a subject word according to the text id, and extracting a subject word text vector corresponding to the head and tail of the subject word from the text vector according to the head-tail position of the subject word; performing conditional normalization on the subject word text vector to generate a normalized vector; inputting the normalized vector into a fully connected layer connected with a binary classifier to predict the head-tail position of the value corresponding to each category of the subject word.

3. The method of extracting clinical finding events of claim 2, wherein, The step of obtaining a disease description text in training data and converting the disease description text into a text id and a text vector comprises: obtaining a disease description text in training data, converting the disease description text into a text id of a BERT model, and inputting the text id of the BERT model into an encoder of the BERT; obtaining a text vector generated by the encoder.

4. An apparatus for extracting clinical finding events, characterized by: The method comprises the following steps: obtaining a disease description text, and adding a first string to the front of the disease description text to generate a first text; inputting the first text into a trained deep learning model for prediction, and obtaining a prediction result; restoring the prediction result into a plurality of triplets, wherein each triplet comprises a head-tail position of a subject word in the disease description text and a head-tail position of a value corresponding to each category in the disease description text; generating a clinical finding event composed of a plurality of four-tuple data according to the plurality of triplets, specifically: merging triplets of subject words with the same head-tail position, and merging values of the same category to generate a four-tuple data of a clinical finding event; replacing the description word with the first string with an empty string, and replacing the replacement word in the subject word and the virtual word in the occurrence state in reverse according to a reference table to generate a clinical finding event composed of a plurality of four-tuple data. ​ The replacement of the described word for the first string to an empty string and the replacement of the replacement word appearing in the subject word in the reverse direction according to the table and the replacement of the dummy word in the occurrence state in the reverse direction are generated by the clinical discovery event generated by the plurality of four tuple data.

5. The apparatus for extracting clinical finding events of claim 4, wherein, The prediction result acquisition unit further includes a module for training a deep learning model, specifically configured to: Obtain the disease description text in the training data, and convert the disease description text into a text id and a text vector; Generate the head and tail positions of the subject word according to the text id, and extract the subject word text vector corresponding to the head and tail of the subject word from the text vector according to the head and tail positions of the subject word; Conditionally normalize the subject word text vector and generate a normalized vector; The normalized vector is transmitted to a fully connected layer and a binary classifier to predict the head and tail positions of the value corresponding to each category of the subject word.

6. The apparatus for extracting clinical finding events of claim 5, wherein, The disease description text in the training data is obtained, and the disease description text is converted into a text vector, specifically: Obtain the disease description text in the training data, convert the disease description text into a text id of the BERT model, and transmit the text id of the BERT model to the encoder of BERT; Obtain the text vector generated by the encoder.

7. An apparatus for extracting clinical finding events, the apparatus comprising: A memory and a processor are included, the memory stores a computer program, and the computer program can be executed by the processor to implement the method for extracting a clinical discovery event according to any one of claims 1 to 3.

8. A computer-readable storage medium, characterized in that, A computer program is stored, and the computer program can be executed by the processor of the device where the computer readable storage medium is located to implement the method for extracting a clinical discovery event according to any one of claims 1 to 3.

Images