An electronic medical record generation method, device, equipment and storage medium

Abstract

Embodiments of the present application provide a kind of electronic medical record generation method, device, equipment and storage medium.The method comprises: the unstructured medical record data of current patient is input to the generation type medical record model that has been constructed, obtains the first type structured medical record data and semi-structured medical record data of current patient;Semi-structured medical record data is input to the standardized medical data model that has been constructed, obtains the second type structured medical record data of current patient;Based on the structured knowledge base of medical record that has been constructed, first type structured medical record data and second type structured medical record data are handled, and the electronic medical record of current patient is generated.The embodiments of the present application can realize the automatic generation of structured electronic medical record, improve the accuracy and standardization of medical record data in electronic medical record, enhance the accurate intuitive description of electronic medical record to patient condition through medical record structured knowledge base, provide more efficient, convenient and reliable standardized diagnosis and treatment reference for doctor.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of data processing, in particular to a method and device for generating electronic medical records, and a storage medium. BACKGROUND

[0002] With the deepening popularization of medical informatization, electronic medical records have become an indispensable tool for hospitals at all levels, and have covered clinical practice comprehensively.

[0003] In today's clinical practice, electronic medical records can generally be divided into semi-structured electronic medical records and fully-structured electronic medical records. Doctors can refer to some recognized medical record filling specifications, and compile the diagnosis and treatment information of patients, such as personal information, specific symptoms, diagnosis, and treatment information, according to the format requirements of semi-structured electronic medical records or fully-structured electronic medical records, to obtain the electronic medical record of the patient.

[0004] However, the medical record data entered in the semi-structured electronic medical record mainly depends on the subjective diagnosis and treatment writing habits of the doctor, and there are many problems of similar medical record data. The medical record data entered in the fully-structured electronic medical record mainly depends on the medical knowledge and experience of the doctor, and requires the doctor to quickly select appropriate medical record options for the patient in the fully-structured electronic medical record. Limited by the medical knowledge and experience of each doctor, it is impossible to guarantee the efficient and accurate entry of medical record data in the fully-structured electronic medical record.

[0005] Therefore, the improvement and innovation of electronic medical records have become a hot issue that needs to be solved in the field of medical informatization. SUMMARY

[0006] Embodiments of the present application provide a method and device for generating electronic medical records, and a storage medium, which realize the automatic generation of structured electronic medical records, improve the accuracy and standardization of medical record data in electronic medical records, enhance the accurate and intuitive description of patient conditions by electronic medical records, and provide more efficient and reliable standardized diagnosis and treatment reference for doctors.

[0007] In a first aspect, embodiments of the present application provide a method for generating electronic medical records, which comprises:

[0008] inputting the unstructured medical record data of the current patient into the constructed generative medical record model to obtain the first type of structured medical record data and semi-structured medical record data of the current patient;

[0009] inputting the semi-structured medical record data into the constructed standardized medical data model to obtain the second type of structured medical record data of the current patient;

[0010] Based on the constructed structured medical record knowledge base, the first type of structured medical record data and the second type of structured medical record data are processed to generate the electronic medical record of the current patient.

[0011] In a second aspect, the embodiments of the present application provide an electronic medical record generation device, which comprises:

[0012] A first structured module is configured to input the unstructured medical record data of the current patient into the constructed generative medical record model to obtain the first type of structured medical record data and semi-structured medical record data of the current patient.

[0013] A second structured module is configured to input the semi-structured medical record data into the constructed standardized medical data model to obtain the second type of structured medical record data of the current patient.

[0014] An electronic medical record generation module is configured to process the first type of structured medical record data and the second type of structured medical record data based on the constructed structured medical record knowledge base to generate the electronic medical record of the current patient.

[0015] In a third aspect, the embodiments of the present application provide an electronic device, which comprises:

[0016] A processor and a memory, the memory is configured to store a computer program, and the processor is configured to invoke and run the computer program stored in the memory to execute the electronic medical record generation method provided in the first aspect of the present application.

[0017] In a fourth aspect, the embodiments of the present application provide a computer readable storage medium configured to store a computer program, and the computer program causes a computer to execute the electronic medical record generation method provided in the first aspect of the present application.

[0018] In a fifth aspect, the embodiments of the present application provide a computer program product comprising computer programs / instructions, which, when executed by a processor, implement the electronic medical record generation method provided in the first aspect of the present application.

[0019] The embodiment of the present application provides a kind of electronic medical record generation method, device, equipment and storage medium, the unstructured medical record data of current patient is input to the generation type medical record model that has been constructed, obtains the first type structured medical record data and semi-structured medical record data of current patient.The semi-structured medical record data is input to the standardized medical data model that has been constructed, obtains the second type structured medical record data of current patient.Then, based on the structured knowledge base of medical record that has been constructed, first type structured medical record data and second type structured medical record data are processed, i.e.the electronic medical record of current patient can be generated, to realize the automatic generation of structured electronic medical record, improve the accuracy and standardization of medical record data in electronic medical record, eliminate the tedious process of frequently manually entering medical record data in electronic medical record, enhance the accurate intuitive description of electronic medical record to patient condition through medical record structured knowledge base, provide more efficient, convenient and reliable standardized diagnosis and treatment reference for doctor. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 The flow chart of the electronic medical record generation method shown in the embodiment of the present application is shown.

[0022] Figure 2 The principle diagram of the generation process of the electronic medical record shown in the embodiment of the present application is shown.

[0023] Figure 3 The method flow chart of the structured processing of unstructured medical record data by the generation type medical record model shown in the embodiment of the present application is shown.

[0024] Figure 4 The method flow chart of the training process of the generation type medical record model shown in the embodiment of the present application is shown.

[0025] Figure 5 The method flow chart of the construction process of the standardized medical data model shown in the embodiment of the present application is shown.

[0026] Figure 6 The method flow chart of generating electronic medical record based on medical record structured knowledge base shown in the embodiment of the present application is shown.

[0027] Figure 7 The method flow chart of the construction process of medical record structured knowledge base shown in the embodiment of the present application is shown.

[0028] Figure 8A principle block diagram of an electronic medical record generation device is shown in an embodiment of the present application.

[0029] Figure 9 A schematic block diagram of an electronic device is shown in an embodiment of the present application. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a 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 work fall within the scope of the present application.

[0031] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product, or device.

[0032] In order to solve the problem that electronic medical records need to rely on the medical knowledge and experience of doctors and cannot guarantee efficient and accurate entry of case data into electronic medical records, an embodiment of the present application designs a scheme for automatically generating fully structured electronic medical records. For any patient's unstructured medical record data, the scheme successively passes through a constructed generative medical record model and a standardized medical data model to obtain corresponding first and second structured medical record data. Then, based on a constructed medical record structured knowledge base, the first and second structured medical record data are processed, and the electronic medical record of the patient is generated, thereby realizing automatic generation of structured electronic medical records and improving the accuracy and standardization of medical record data in electronic medical records.

[0033] Figure 1 A flowchart of an electronic medical record generation method is shown in an embodiment of the present application. With reference to Figure 1 , the method can include the following steps:

[0034] S110, inputting the unstructured medical record data of the current patient into the constructed generative medical record model to obtain first structured medical record data and semi-structured medical record data of the current patient.

[0035] Wherein, the current patient can be any patient in the medical service system that needs a doctor to diagnose and treat the patient's condition. When the current patient visits the hospital, the patient usually needs to perform various condition checks with the assistance of a doctor to obtain various condition check results of the current patient. In order to ensure the accuracy of the doctor's condition check of the patient, it is usually required that the condition check results can be unstructured text information without being structured in advance by any subjective experience, so as to ensure the readability of the patient's condition check results. Therefore, the unstructured medical record data of the current patient in the present application can be the original condition check information of the current patient from various condition checks in multiple types and multiple sources.

[0036] In the present application, since the patient visits the hospital, the doctor can check the actual condition of the patient at any time and make a new condition diagnosis and analysis. For example, the doctor issues a new medical order when checking the ward. Moreover, under the examination of the doctor, various intelligent monitoring instruments in the hospital can be used to check the condition of the patient. Or, the patient may be treated across hospitals, and the medical diagnosis information made by other hospitals for the patient needs to be obtained.

[0037] Therefore, in order to ensure the comprehensiveness of the electronic medical record, the present application can support the use of mobile terminal data entry, Internet of Things data entry, and historical medical history documents and other text materials to obtain unstructured medical record data of the patient from multiple information sources. Then, the unstructured medical record data of the current patient can be multi-modal medical record data entered by a multi-modal method.

[0038] Next, the present application can explain and describe the multi-modal entry method of the unstructured medical record data of the current patient as follows:

[0039] 1) Mobile terminal data entry

[0040] The present application supports mobile data entry of multiple devices such as mobile phone scanning, handheld personal digital assistant (PDA) scanning, tablet computer entry, and multi-media voice entry.

[0041] The electronic medical record system developed in the medical service platform can be installed in advance on each mobile terminal to support medical staff to log in to the electronic medical record system at any time and anywhere. When the doctor checks the actual condition of each patient at any time, any mobile terminal carried by the doctor can be used to scan the patient information to extract the mobile terminal to be connected to the electronic medical record system and match the various visit data of the current patient. That is, the electronic medical record system can be entered in the mobile terminal, and under the authorization of the current patient, the current visit information and information such as the reception department of the current patient can be presented. Moreover, the access record of this time is automatically entered in the electronic medical record system.

[0042] From the above, by mobile terminal can scan the patient information, to interface to the electronic medical record system in the patient's each visit data. Then, using the scanning of the patient's two-dimensional code, shooting text, voice recording and other ways, and using optical character recognition (Optical Character Recognition, referred to as OCR) technology or speech recognition technology and so on to process, to automatically enter the electronic medical record system in the patient's constantly updated visit information, so as to support medical staff using any mobile terminal in multiple scenarios to enter the patient's visit data anytime, anywhere, as the patient's related unstructured medical record data, effectively improve the work efficiency of doctors, can avoid the error caused by handwriting record, improve the accuracy of patient visit data and the work flexibility and comfort of medical staff.

[0043] For example, when the doctor rounds, support using any mobile terminal to interface to the electronic medical record system in the current patient's visit account, and real-time input the doctor's order information issued to the current patient. Or, medical staff through multimedia voice input a patient's medical record data, etc.

[0044] Then, the mobile terminal data entry mode can also support medical staff to query and update the patient's visit information in real time. Because the patient's condition and treatment process is dynamic, this way can ensure the accuracy and timeliness of the patient's visit information. Moreover, the mobile terminal real-time updates the patient's visit information, which can ensure that the doctor can understand the patient's condition and treatment situation in time. Such as large ward round, nurses nursing process can be recorded at any time through mobile terminal equipment to enter the patient's real-time health status, which is convenient for doctors to view and adjust the clinical path in real time.

[0045] 2) Internet of Things data entry

[0046] Considering that the patient will usually use various intelligent monitoring instruments in the hospital to check the actual condition under the doctor's diagnosis, to obtain the patient's related medical record data. Among them, the intelligent detection instrument can include but not limited to blood oxygen monitor, blood pressure monitor, electrocardiogram monitor, body temperature monitor, anesthesia monitor, non-invasive blood glucose monitor, craniocerebral monitor, urine analyzer, etc.

[0047] Therefore, in order to ensure the comprehensiveness of the patient electronic medical record, the present application can open an interface for each intelligent detection instrument in the electronic medical record system, so that each intelligent monitoring instrument can access the electronic medical record system using Internet of Things technology. Therefore, when a patient uses a certain intelligent monitoring instrument to monitor the patient's condition, the intelligent monitoring instrument can upload the patient's condition monitoring data to the electronic medical record system and match the patient's identification information, thereby supporting the patient to enter relevant condition monitoring data at any time and place as relevant unstructured medical record data of the patient, reducing the labor consumption and the problems of misrecording, wrong recording, delayed recording and the like when passively recording, and improving the accuracy and timeliness of the condition monitoring data. Moreover, the intelligent monitoring instrument uploads the patient's condition monitoring data to the electronic medical record system in real time without manual transfer by medical staff, which can improve the effect and quality of patient medical diagnosis and ensure the diagnosis and treatment experience of the patient.

[0048] Taking blood oxygen monitoring as an example, the blood oxygen monitor can upload the blood oxygen monitoring data of any patient to the electronic medical record system in real time and match the patient's identification information after being connected to the electronic medical record system. The blood oxygen monitoring data can include but is not limited to blood oxygen saturation, heart rate, respiratory rate and the like.

[0049] When the intelligent monitoring instrument uploads the patient's condition monitoring data to the electronic medical record system, the electronic medical record system can also format and verify the uploaded condition monitoring data according to the pre-set data standard, so as to ensure the accuracy and integrity of the patient's condition monitoring data and the accuracy and integrity of the electronic medical record.

[0050] Therefore, the doctor can check the relevant condition monitoring data of each patient by logging in to the electronic medical record system, so as to ensure timely treatment of the patient when the condition monitoring data of a certain patient is abnormal and avoid delaying the treatment of the patient.

[0051] 3) Historical medical history literature and other text treatment input

[0052] Considering that the patient may be treated across hospitals and has historical medical record data in multiple hospitals, the present application can directly scan various types of text materials related to the patient's condition. Moreover, the OCR technology is used to analyze various types of text materials and convert them into editable digital format text content, which is then input into the electronic medical record system and matched to the identification information of the relevant patient as relevant unstructured medical record data of the patient.

[0053] The various types of text materials related to the patient's condition can include but are not limited to relevant medical record data and various medical literature when the patient is treated in an external hospital.

[0054] From the above, the present application can adopt the above-mentioned multi-modal input mode, through the seamless integration between multiple mobile terminals, intelligent monitoring instruments and other text sources and OCR technology, to automatically collect and input the unstructured medical record data of any patient, so as to provide real-time, accurate, complete and convenient patient medical record data support for the generation of subsequent electronic medical records, and improve the accuracy and real-time performance of the patient's unstructured medical record data.

[0055] In the present application, in order to facilitate doctors to issue accurate medical orders for patients according to electronic medical records, it is usually necessary to structure the unstructured medical record data of the patient, so that the patient medical record data in the electronic medical record can be arranged according to the hierarchy, have certain order and systematization, thereby enhancing the readability of the patient's electronic medical record.

[0056] In order to realize the structured processing of the unstructured medical record data, the present application can pre-construct a generative medical record model, and use a large amount of historical unstructured medical record data of patients as training samples to train the structured function of the generative medical record model, so as to realize the structured processing of the unstructured medical record data.

[0057] Since the unstructured medical record data of any patient is usually text content, the generative medical record data can be a natural language processing model based on Transform architecture, which is also one of the most advanced natural language processing models at present. For example, the generative medical record model can be a Variational Auto Encoder (VAE) model.

[0058] Then, after obtaining the unstructured medical record data of the current patient, the present application can input the unstructured medical record data into the constructed generative medical record model, so as to structure the unstructured medical record data through the generative medical record model.

[0059] And since there are various medical terms and medical terminologies in the medical service system, which cannot be all counted as training samples during the training of the generative medical record model, the generative medical record model may not be able to completely structure some medical record data. Therefore, as shown in Figure 2 When the unstructured medical record data of the current patient is structured by the generative medical record model, the first type of structured medical record data that has completed the structured processing and the semi-structured medical record data that has not completed the structured processing can be obtained.

[0060] S120, inputting the semi-structured medical record data into the constructed standardized medical data model to obtain the second type of structured medical record data of the current patient.

[0061] In order to ensure the full structured readability of the electronic medical record, since HL7 is the name of an international organization for the development of health information standards, as a standard for the transmission and exchange of medical health information, it can be used by the national interconnection medical interaction specification to speed up the development process of medical informatization. For the latest v3 standard of HL7, HL7 v3 is model-driven, based on the reference information model (RIM) in the HL7 v3 standard.

[0062] Among them, the RIM model can be an abstract data model for describing various medical data and medical terminology information in the medical information system, which almost comprehensively covers all medical information involved in the medical service field.

[0063] Therefore, by effectively processing, processing and integrating various medical information involved in the HL7 v3 RIM, the application can be converted into simple and clear structured medical vocabulary and the association relationship between various medical vocabularies. Then, according to the above various structured medical vocabularies and the association relationship between various medical vocabularies, a corresponding standardized medical data model can be constructed in advance, so that the standardized medical data model can almost comprehensively cover the structured medical vocabulary and the association relationship between various medical vocabularies in all medical information involved in the medical service field.

[0064] Then, for the semi-structured medical record data obtained after the generated medical record model processing, the application can input the current patient's version of the structured medical record data into the constructed standardized medical data model. As shown in Figure 2 The structured medical vocabulary and the association relationship between various medical vocabularies covered in the standardized medical data model are used to continue the structured processing of the semi-structured medical record data, to obtain the corresponding second type of structured medical record data, so as to realize the full structured conversion of the current patient's unstructured medical record data.

[0065] S130, based on the constructed medical record structured knowledge base, processing the first type of structured medical record data and the second type of structured medical record data to generate the electronic medical record of the current patient.

[0066] Since a large amount of historical unstructured medical record data is usually structured to train the generative medical record model and the standardized medical data model during the construction of the above generative medical record model and the standardized medical data model. Then, after the training of the above generative medical record model and the standardized medical data model is completed, a large amount of historical structured medical record data can be obtained by sequentially inputting a large amount of historical unstructured medical record data into the generative medical record model and the standardized medical data model, so as to form a structured medical record corpus. The historical unstructured medical record data can include, but is not limited to, relevant medical record data entered by each patient during historical diagnosis and treatment, medical literature related to each disease, and the like.

[0067] In the present application, in order to comprehensively analyze the structured medical record data of any patient, a variety of historical structured medical record data contained in the above medical record corpus can be effectively processed, handled and integrated to analyze the commonality within the historical medical history of each disease. Thus, each structured medical record data can be regarded as a knowledge point, and a relatively complete medical record structured knowledge base can be constructed in advance according to the association relationship between a large amount of historical structured medical record data.

[0068] It can be understood that the medical record structured knowledge base can include a large amount of medical vocabulary, medical terminology and diagnosis and treatment methods and other related structured medical knowledge points. Moreover, the medical record structured knowledge base is constructed based on a large amount of historical unstructured medical record data and medical literature, and its scientificity and accuracy are greatly guaranteed.

[0069] Therefore, after obtaining the first type of structured medical record data and the second type of structured medical record data of the current patient, the present application can use each structured medical knowledge point in the above medical record structured knowledge base to compare and analyze the first type of structured medical record data and the second type of structured medical record data, so as to determine the electronic medical record standard template suitable for the current patient. Then, the specific structural attributes corresponding to the first type of structured medical record data and the second type of structured medical record data are automatically input into the electronic medical record standard template, so as to generate the electronic medical record of the current patient, without the need for the doctor to manually enter the basic and redundant medical record data, eliminating the tedious process of the doctor frequently manually entering the medical record data, avoiding misdiagnosis and other operations that can be caused by manual entry of medical record data, and improving the diagnosis and treatment probability of the patient. Moreover, the combination of the medical record structured knowledge base to generate the electronic medical record of the current patient can avoid diagnosis and treatment errors caused by personal cognition or omission of the doctor, and improve the medical safety and treatment effect of the current patient.

[0070] It should be noted that the medical record structured knowledge base in the present application can provide support for medical research. Through the storage of a large amount of unstructured medical record data after structured processing by the medical record structured knowledge base, useful information under the corresponding disease can be extracted from the medical record structured knowledge base, potential causes, disease courses and treatment methods are explored, and the progress of medical research is further promoted, so that the medical record structured knowledge base not only has great application value for clinical medical treatment, but also has an important promoting effect on medical research.

[0071] Moreover, based on the medical record structured knowledge base, the first type of structured medical record data and the second type of structured medical record data are processed, and a plurality of similar diagnosis results with high correlation degrees of the current patient can be determined. Then, in the electronic medical record of the current patient, the plurality of similar diagnosis results with high correlation degrees can be sorted and presented for reference by the doctor. The doctor can refer to the plurality of similar diagnosis results presented in the electronic medical record of the current patient to assist in differential diagnosis of the current patient, which helps the doctor to more accurately diagnose and treat the disease of the patient, improves the efficiency and quality of diagnosis and treatment, and thus realizes the diversity of auxiliary diagnosis and treatment of the patient.

[0072] The technical scheme provided by the embodiment of the present application inputs the unstructured medical record data of the current patient into the generated medical record model constructed, obtains the first type of structured medical record data and the semi-structured medical record data of the current patient. The semi-structured medical record data is input into the standardized medical data model constructed, and the second type of structured medical record data of the current patient is obtained. Then, based on the constructed medical record structured knowledge base, the first type of structured medical record data and the second type of structured medical record data are processed, that is, the electronic medical record of the current patient is generated, so as to realize the automatic generation of the structured electronic medical record, improve the accuracy and standardization of the medical record data in the electronic medical record, eliminate the tedious process of manually entering the medical record data in the electronic medical record, enhance the accurate and intuitive description of the patient's condition by the medical record structured knowledge base, and provide more efficient, convenient and reliable standardized diagnosis and treatment reference for doctors.

[0073] As an optional implementation scheme in the present application, in order to ensure the structured accuracy of the electronic medical record, the present application can explain the specific structured processing process of the unstructured medical record data by the generated medical record model.

[0074] Figure 3 The method flowchart for the generated medical record model of the embodiment of the present application to perform structured processing on the unstructured medical record data is shown in FIG. 1. Figure 3 As shown in FIG. 1, the method can include the following steps:

[0075] S310, inputting the unstructured medical record data of the current patient into the generated medical record model constructed, to obtain corresponding structured segmentation and structure categories of the structured segmentation.

[0076] Since structured data is usually composed of well-defined vocabulary information and stored and managed in highly organized tables or databases. Therefore, after the trained generative medical record model can have the ability to convert unstructured medical record data into a plurality of well-defined medical-related vocabularies and predict the category of each vocabulary, so as to realize the structured traceability representation of each vocabulary.

[0077] Therefore, in order to realize the structured description of the electronic medical record, the present application can input the unstructured medical record data of the current patient into the constructed generative medical record model, and process the unstructured medical record data by the generative medical record model to obtain each structured token after the unstructured medical record data is tokenized. Moreover, the generative medical record model can predict the category of each structured token in the medical service field, so as to mark the corresponding structure category for each structured token.

[0078] Among them, the structured token is a medical vocabulary with a well-defined medical service field, which can include but is not limited to drug names, department names, disease names, diagnosis and treatment methods, and various medical terms.

[0079] S320, the first type of structured token with a non-empty structure category and the structure category of the first type of structured token are determined as the first type of structured medical record data of the current patient.

[0080] Considering that the training sample of the generative medical record model is limited, it cannot realize the complete structured processing of the unstructured medical record data, so that part of the structured token cannot mark the accurate structure category.

[0081] Therefore, for each structured token after the structured processing of the unstructured medical record data, by judging whether the structure category of the structured token is empty, it can be determined whether the structure category of the structured token is successfully marked, so as to judge whether the structured token belongs to the completely structured data.

[0082] If the structure category of a structured token is non-empty, it means that the structured token belongs to the completely structured data. Therefore, the present application can divide the structured token into the first type of structured token, so as to obtain all the first type of structured token. Then, all the first type of structured token and the structure category of the first type of structured token are counted out, which can be determined as the first type of structured medical record data of the current patient.

[0083] S330, the second type of structured token with an empty structure category is determined as the semi-structured medical record data of the current patient.

[0084] If the structure category of a structured word is empty, it means that the structured word does not belong to the completely structured data. Therefore, the structured word can be divided into the second structured word, so as to obtain all the second structured words. Then, the second structured words are counted, and the semi-structured medical record data of the current patient can be determined.

[0085] In order to ensure the accuracy of the generative medical record model in the structured processing of unstructured medical record data, the specific training process of the generative medical record model can be described.

[0086] Figure 4 The method flowchart of the training process of the generative medical record model is shown in the embodiment of the present application.

[0087] As shown in Figure 4 , the method can include the following steps:

[0088] S410, for each category of medical corpus, input the medical corpus into the pre-constructed initial generative model to obtain the corresponding corpus word segmentation and the predicted category of the corpus word segmentation.

[0089] There are a large number of recognized basic medical terms in the medical service field. According to different categories to which the medical terms belong, the medical terms can be pre-divided into different medical corpora, so as to obtain medical corpora under different categories.

[0090] In some example schemes, the medical corpus can include but is not limited to a drug library, a SNOMED CT standard library, a department library, etc. Each medical corpus is marked with the category to which it belongs, and each medical corpus can include multiple medical terms under the category to which it belongs. For example, the drug library can include multiple drug names.

[0091] Moreover, in order to ensure the accurate training of the generative medical record model, an initial generative model can be pre-constructed based on a natural language processing model of a Transform architecture, so as to subsequently train the initial generative model with a large number of training samples to obtain the final generative medical record model.

[0092] Therefore, in order to ensure the accurate training of the generative medical record model, the medical corpus under each category can be used as a training sample of the generative medical record model, and the category to which each medical corpus belongs can be used as the category to which each medical term in the medical corpus belongs.

[0093] Then, for each medical corpus under each category, the medical corpus can be input into the pre-constructed initial generative model, in which all medical terms in the medical corpus can be combined into a complete text content. Moreover, using the generative property of the initial generative model itself, the text content can be subjected to corresponding word segmentation processing to obtain the corresponding segmented words, which can be output as the corresponding corpus segmented words of the medical corpus. For each corpus segmented word, the initial generative model can also be used to perform corresponding semantic analysis on the corpus segmented word to predict the category to which the corpus segmented word belongs, as the predicted category of the corpus segmented word in this application.

[0094] In the same way as described above, the medical corpus under each category can be sequentially input into the pre-constructed initial generative model, so as to obtain the corresponding corpus segmented words of the medical corpus under each category and the predicted category of the corpus segmented words.

[0095] S420, according to the first segmented word loss between the corpus segmented words and the medical terms in the medical corpus and the first category loss between the predicted category of the corpus segmented words and the category of the medical corpus, the initial generative model is initialized and trained.

[0096] Since the medical corpus under each category is composed of multiple medical terms under the category, and each medical term in the medical corpus under the same category belongs to the same category. Therefore, each corpus segmented word of the medical corpus under each category processed by the initial generative model should be as same as possible as each medical term in the medical corpus, and the predicted category of each corpus segmented word should be as same as possible as the category of the medical corpus, so as to ensure the accurate training of the initial generative model.

[0097] Therefore, after obtaining the corresponding corpus segmented words of the medical corpus under each category and the predicted category of the corpus segmented words, the application can compare each corpus segmented word obtained by the medical corpus under each category after the initial generative model and each medical term in the medical corpus, obtain the difference therebetween, and thus determine the corresponding first segmented word loss.

[0098] Moreover, for the medical corpus under each category, the predicted category of each corpus segmented word obtained by the medical corpus after the initial generative model and the category of the medical corpus can also be compared, the difference therebetween can be obtained, and thus the corresponding first category loss can be determined.

[0099] Then, using the first segmented word loss and the first category loss of the medical corpus under each category, the model structure and parameters in the initial generative model can be continuously adjusted accordingly to complete the initialization training of the initial generative model.

[0100] It can be understood that after the initial generative model is trained by using the medical corpus under each category, the initial generative model can initially have the ability of word segmentation, recognition and category marking of part of the basic unstructured medical record data, so as to facilitate subsequent retraining of the initial generative model by using a large amount of historical unstructured medical record data, and improve the training efficiency and accuracy of the generative medical record model.

[0101] In S430, a structured medical record label corresponding to the recorded multi-modal historical medical record data is determined, and the structured medical record label includes historical medical record words corresponding to the multi-modal historical medical record data and categories to which the historical medical record words belong.

[0102] After the initial generative model is trained by using the medical corpus under all categories, a large amount of historical unstructured medical record data recorded by patients in various ways during historical visits can be obtained as training samples of the initial generative model, so as to retrain and relearn the initial generative model after the initial training, and further improve the structured accuracy and comprehensiveness of the generative medical record model.

[0103] In order to realize accurate training of the generative medical record model, after the multi-modal historical medical record data is obtained, the doctor can perform artificial structured processing on the multi-modal historical medical record data to divide the multi-modal historical medical record data into structured words at different levels as historical medical record words in the present application, and determine the categories to which each historical medical record word belongs. Then, the present application can divide each historical medical record word of the multi-modal historical medical record data into a structure attribute value at different levels, and the category to which each historical medical record word belongs is used as a corresponding structure category, so as to obtain a structured medical record label corresponding to the multi-modal historical medical record data.

[0104] It can be understood that the structured medical record label corresponding to the multi-modal historical medical record data in the present application can be obtained by artificial structured processing by the doctor.

[0105] For example, it is assumed that the multi-modal historical medical record data of a patient is as follows: male, m years old, T: a degree, X: no sinus arrest, atrial fibrillation, premature ventricular contraction and other abnormal manifestations, etc.

[0106] Then, the structured medical record label of the multi-modal historical medical record data after artificial structured processing can be as shown in Table 1:

[0107] Table 1: Structured medical record label

[0108]

[0109]

[0110] From the above, in the structured medical record label corresponding to the multi-modal historical medical record data, part of the historical medical record vocabulary represented by the structure attribute value is not marked with the corresponding structure category, so that the category of part of the historical medical record vocabulary can be empty.

[0111] S440, inputting the multi-modal historical medical record data into the initial generative model after initialization and training to obtain corresponding multi-modal segmentation and predicted category of the multi-modal segmentation.

[0112] After obtaining the multi-modal historical medical record data, the application can input the multi-modal historical medical record data into the initial generative model after initialization and training. Since the initial generative model after initialization and training can preliminarily have the segmentation, recognition and category marking ability of part of the basic unstructured medical record data. Therefore, by the initial generative model, the multi-modal historical medical record data can be segmented to obtain corresponding multi-modal segmentation. Moreover, for each multi-modal segmentation, the initial generative model can be used to perform corresponding semantic analysis on the multi-modal segmentation to predict the category to which the multi-modal segmentation belongs, as the predicted category of the multi-modal segmentation in the application.

[0113] S450, retraining the initial generative model according to the second segmentation loss between the multi-modal segmentation and the historical medical record vocabulary and the second category loss between the predicted category of the multi-modal segmentation and the category to which the historical medical record vocabulary belongs to obtain the corresponding generative medical record model.

[0114] After obtaining each multi-modal segmentation corresponding to the multi-modal historical medical record data and the predicted category of the multi-modal segmentation, the application can compare each multi-modal segmentation and each historical medical record segmentation in the structured medical record label to obtain the difference therebetween to determine the corresponding second segmentation loss.

[0115] Moreover, the predicted category of each multi-modal segmentation can also be compared with the category to which a certain historical medical record segmentation similar to the multi-modal segmentation in the structured medical record label belongs to obtain the difference therebetween to determine the corresponding second category loss.

[0116] Then, by using the second segmentation loss and the second category loss corresponding to each multi-modal segmentation, the model structure and parameters in the initial generative model after initialization and training can be continuously adjusted accordingly to continuously retrain the initial generative model. Then, after completing the retraining of the initial generative model, the corresponding generative medical record model can be obtained.

[0117] As an exemplary approach in this application, since the initial generative model used during training of the generative medical record model is a model capable of predicting a text sequence (in this application, it can predict a sequence of medical attribute words composed of various multimodal word segments in the multimodal historical medical record data), a method based on conventional language modeling can be used during the retraining phase of the initial generative model to optimize the given unstructured medical record data, i.e., the multimodal historical medical record data in this application, through maximum likelihood estimation.

[0118] Specifically, the maximum likelihood estimate in a generative medical record model can be:

[0119]

[0120] Where k represents the size of the context window used in the generative medical record model to predict the next medical attribute word (i.e., the multimodal word segmentation in this application). That is, the generative medical record model can predict the next medical attribute word (i.e., the multimodal word segmentation in this application) based on the first k predicted medical attribute words x. i-k ,…,x i-1 To predict the next medical attribute word x i θ represents the parameters of the generative medical record model, and the likelihood function is optimized using stochastic gradient descent.

[0121] Furthermore, for any medical attribute word (i.e., the multimodal segmentation in this application), the window word sequence x' = x', consisting of the first k predicted medical attribute words, is formed by the previous predictions. -k …x -1 To put it another way:

[0122] h [0] =e x' W e +W p

[0123]

[0124] P(x) = Softmax(h) [L] W eT )

[0125] Among them, e x' W is the one-hot vector representation of a window word sequence x' of length k. e W represents the word vector matrix. p L represents the position vector matrix of the window word sequence x', and L represents the total number of Transformer layers.

[0126] Moreover, the Transformer model can be a neural network architecture for natural language processing tasks that introduces self-attention mechanisms in sequential data, which helps capture long-range dependencies.

[0127] The Softmax function can be a function for converting model output into a probability distribution, ensuring that the sum of all output probabilities is 1.

[0128] Then, in the training phase of the initial generative model, the next medical attribute word represented by the multi-modal segmented word can be predicted according to the k medical attribute words (that is, the k multi-modal segmented words) that have been predicted in the multi-modal historical medical record data, and P(x) is output to determine the probability of each medical attribute word being predicted. Then, using the above maximum likelihood estimation formula, the maximum likelihood function can be calculated to construct the corresponding loss function to continuously optimize and train the initial generative model, thereby completing the retraining of the initial generative model.

[0129] Suppose the multi-modal historical medical record data is "the patient feels a headache and a sore throat, and needs to take medicine", and the context window size k = 3. Then, the window word sequence with a length of k can be "the patient feels a headache", from which "a sore throat" can be predicted as the next medical attribute word, and the prediction result and the next actual medical record vocabulary are compared to adjust the model parameters through maximum likelihood estimation, so that the prediction result is closer to the next actual medical record vocabulary, thereby continuously training the generative medical record model.

[0130] It can be understood that through the above training process of the generative medical record model, the generative medical record model has the ability to structure the unstructured medical record data obtained in a multi-modal manner. Then, in order to make the generative medical record model more close to the actual patient diagnosis and treatment behavior, the present application can further fine-tune the generative medical record model.

[0131] Next, the fine-tuning process of the generative medical record model can be described:

[0132] In some implementable ways, the present application can introduce a target labeled medical record data with a diagnosis label, which can include but is not limited to the results of target diagnosis and differential diagnosis.

[0133] Then, the present application can train and update the generative medical record model according to the diagnosis optimization loss corresponding to the difference between the diagnosis result output by the generative medical record model according to the target labeled medical record data and the diagnosis label of the target labeled medical record data and the retraining loss of the generative medical record model.

[0134] That is, after the completion of the retraining of the generative medical record model, the application can obtain a series of unstructured historical medical record data of patients after completing diagnosis and treatment, and manually annotate the corresponding target diagnosis and differential diagnosis results. Then, the above unstructured historical medical record data after completing diagnosis and treatment is taken as the target annotated medical record data in the application, and the results of the target diagnosis and differential diagnosis annotated for it are taken as the diagnosis label of the target annotated medical record data.

[0135] For example, assuming that the target annotated medical record data is "patient's body temperature rises, sore throat", then the diagnosis label can be "the patient is diagnosed with tonsillitis".

[0136] For fine-tuning of the generative medical record model, the application can take it as a downstream task after training of the generative medical record model. By inputting the target annotated medical record data into the generative medical record model, the hidden output corresponding to the last token of the last layer of the generative medical record model is obtained Then, the hidden output Through a fully connected layer, the diagnosis result corresponding to the diagnosis label is predicted through the following formula.

[0137] P(y|x1…x n )=Softmax(h [L] W y )

[0138] Where x1…x n is the text content corresponding to the target annotated medical record data, h [L] is the last token predicted after the structured processing of the target annotated medical record data, and W y represents the weight of the fully connected layer.

[0139] Then, in the downstream task, the cross-entropy loss function can be used to analyze the difference between the diagnosis result output by the target annotated medical record data after the generative medical record model and the diagnosis label of the target annotated medical record data, so as to calculate the corresponding diagnosis optimization loss.

[0140] Where the cross-entropy loss function in the downstream task can be

[0141] And since the training target of the generative medical record model after fine-tuning using the diagnosis optimization loss in the downstream task is mainly to optimize the diagnosis effect on the target annotated medical record data, it emphasizes the particularity, which will inevitably cause part of the coverage or erasure of the basic medical vocabulary learned by the generative medical record model in the retraining stage, and lose some generality.

[0142] Therefore, in order to balance the generality and specificity of the structured processing of the generative medical record model, the application introduces the retraining loss of the generative medical record model in the retraining stage in addition to the diagnosis optimization loss obtained in the downstream task when fine-tuning the generative medical record model.

[0143] The application can fine-tune the generative medical record model to complete the training update of the generative medical record model by the following formula:

[0144] τ(c)=τ FT (c)+λτ PT (c)

[0145] Wherein, τ FT (c) can represent the diagnosis optimization loss obtained by using the target labeled medical record data in the downstream task, τ PT (c) can represent the retraining loss of the generative medical record model, and λ can represent the weight set in advance for the retraining loss of the generative medical record model, which is used to control the retention degree of the general knowledge learned in the retraining stage of the generative medical record model when fine-tuning.

[0146] Through the above formula, the diagnosis optimization loss obtained in the downstream task and the retraining loss of the generative medical record model can be combined to further fine-tune the generative medical record model, thereby balancing the generality and specificity of the structured processing ability of the generative medical record model.

[0147] In other implementable manners, after the completion of the training of the generative medical record model, it can be used in each medical institution to structure the unstructured medical record data of the patient.

[0148] Then, when any patient actually receives treatment in the corresponding medical institution, medical staff can log in to the electronic medical record system to manually input the unstructured medical record data related to the patient. Therefore, the application can train and update the generative medical record model according to the difference between the first structured medical record data of any patient and the structured correction result of the first structured medical record data.

[0149] That is, for the unstructured medical record data of any patient that has been manually input into the electronic medical record system, the application can input the unstructured medical record data into the generative medical record model, and the generative medical record model can structure the unstructured medical record data to obtain the first structured medical record data. Then, the first structured medical record data can be directly presented in the current interface of the electronic medical record system, such as the sidebar of the current interface, so that the medical staff can manually judge whether the first structured medical record data of the patient has data inconsistency or conversion deviation.

[0150] If the first type of structured medical record data has data inconsistency or conversion deviation, medical staff can directly edit and modify the first type of structured medical record data in the current interface of the electronic medical record system. When the mouse is moved away, the generated medical record model can be modified by default, so as to obtain the structured modification result of the first type of structured medical record data, thereby representing the real structured result.

[0151] Then, the first type of structured medical record data and the structured modification result of the first type of structured medical record data can be compared to determine the difference therebetween, and the generated medical record model can be fine-tuned using the difference. Through the cyclic iteration, the training and updating of the generated medical record model can be completed, so that the generated medical record model can further approach the actual patient diagnosis and treatment behavior.

[0152] It can be understood that the two fine-tuning methods of the generated medical record model mentioned above can be used to fine-tune the generated medical record model, or both methods can be used to fine-tune the generated medical record model in sequence. The application does not limit this.

[0153] According to one or more embodiments of the present application, in order to ensure the structured accuracy of the electronic medical record, the semi-structured medical record data of the current patient obtained through the generated medical record model can be further input into the constructed standardized medical data model to ensure the complete structured processing of the unstructured medical record data of the current patient, thereby realizing the overall structuring of the electronic medical record.

[0154] For the semi-structured medical record data of the current patient, the structured mapping result of the semi-structured medical record data can be determined by using the structured medical record mapping relationship defined in the constructed standardized medical data model, so as to obtain the second type of structured medical record data of the current patient.

[0155] It can be understood that by effectively processing, processing and integrating various medical information involved in HL7 v3 RIM, simple and clear structured medical vocabulary and the association relationship between various medical vocabularies can be converted. Therefore, for the standardized medical data model constructed based on HL7 v3 RIM, the standardized medical data model can almost comprehensively cover the structured medical vocabulary in all medical information involved in the medical service field and the association relationship between various medical vocabularies.

[0156] Therefore, by analyzing the structured medical vocabularies covered in the standardized medical data model and the correlation among the medical vocabularies, a structured medical record mapping relationship can be defined in the standardized medical data model. The structured medical record mapping relationship can include a mapping relationship between each structured medical vocabulary and each structure category set, so as to ensure comprehensive structuring of patient medical record data.

[0157] In the present application, by analyzing the semi-structured medical record data of the current patient, each medical vocabulary that has not been completely structured can be obtained. Then, by using the structured medical record mapping relationship defined in the constructed standardized medical data model, each medical vocabulary in the semi-structured medical record data is structured and mapped, and the structure category to which each medical vocabulary belongs can be obtained as the structured mapping result of the semi-structured medical record data. Then, by uniformly combining each medical vocabulary in the semi-structured medical record data and the structure category to which each medical vocabulary belongs, the second type of structured medical record data of the current patient can be obtained.

[0158] In order to ensure the accuracy of the second type of structured medical record data, the construction process of the standardized medical data model can be described in the present application. As shown in Figure 5 The standardized medical data model can be constructed by the following steps in the present application:

[0159] S510, determining medical entities in the medical standardization system, structured attributes of the medical entities, and correlation among the medical entities.

[0160] Since the RIM model in HL7 v3 can be an abstract data model for describing various medical data and medical terminology information in a medical information system, it almost comprehensively covers all medical information involved in the medical service field. Therefore, the RIM model in HL7 v3 can be used as the medical standardization system in the present application.

[0161] By analyzing the specific meanings of various medical vocabularies in the medical standardization system, various medical subjects such as medical events, diagnosis items, examination results, and test results in the medical standardization system can be determined as medical entities in the medical standardization system.

[0162] Based on the fact that the entities in HL7 v3 RIM can be specifically divided into three parts of roles, substances, and activities, the role part of the medical entities can be defined as different users such as patients, doctors, and nurses in the present application, the substance part of the medical entities can be defined as different items such as drugs and equipment, and the activities of the medical entities can be defined as diagnosis and treatment operations such as surgery, examination, and testing.

[0163] Moreover, the specific attribute information of each medical entity in the medical standardization system is determined, so as to obtain the structured attribute of the medical entity. For example, the structured attribute of a patient can be name, age, gender, etc., and the structured attribute of a drug can be the category, function, and time limit of the drug, etc.

[0164] By analyzing the structured attribute of each medical entity, the association relationship between the medical entities can be determined. For example, a patient can have multiple diagnosis results and multiple inspection results.

[0165] S520, according to the structured attribute of the medical entity and the association relationship between the medical entities, the corresponding structured medical record mapping relationship is determined to obtain the standardized medical data model.

[0166] After determining the medical entities in the medical standardization system, the structured attribute of the medical entity, and the association relationship between the medical entities, the application can determine the corresponding structured medical vocabulary and the category of each structured medical vocabulary according to the structured attribute of each medical entity and the association relationship between the medical entities, so as to define a structured medical record mapping relationship to obtain the corresponding standardized medical data model.

[0167] It can be understood that due to certain constraints and limitations in the actual implementation of some medical entities. For example, taking a surgery information as an example, the format of the surgery time, the value range of the surgery site, and other limitation conditions need to be defined. Therefore, the application will also define the corresponding constraints and limitations for the corresponding medical entity in the standardized medical data model.

[0168] Then, using the message parser and generator in the HL7 v3 RIM to analyze the above definition information, the corresponding standardized medical data model can be constructed, so as to use the parser and generator in the standardized medical data model to parse different elements and attributes in the HL7 message composed of unstructured medical record data, and to convert them into corresponding structured medical record data.

[0169] As an exemplary scheme in the application, the standardized medical data model in the application can include three kinds of surgery information data model, drug treatment data model, and health record data model.

[0170] Next, the application can introduce the above three kinds of data models respectively:

[0171] 1) Surgery information data model

[0172] The present application can use the HL7 v3 RIM to define elements such as surgery, surgeon, surgery process, etc. as corresponding medical entities, and define corresponding structured attributes for each medical entity. Specifically, the present application can define that a surgery process is composed of multiple steps, each step having corresponding attributes such as operation and result. At the same time, the present application can define the relationship between medical entities represented by different elements, such as the relationship between a surgeon and a surgery process. In implementing the surgery information data model, the present application can use the Subscription Profile Repository (SPR) in the HL7 v3 RIM to represent actual surgery information, including surgery ID, surgery time, surgery site, etc. In this message, elements and attributes defined in the HL7 v3 RIM are used to represent different surgery information.

[0173] 2) Drug treatment data model

[0174] In the field of drug treatment information, the present application can use the HL7 v3 RIM to define elements such as drug, dose, administration route, etc. as corresponding medical entities, and define corresponding structured attributes for each medical entity. Specifically, the present application can define that a drug contains multiple components, each component having corresponding attributes such as dose and administration route. At the same time, the present application can define the relationship between medical entities represented by different elements, such as the relationship between a drug and a treatment plan.

[0175] 3) Health record data model

[0176] In the field of health record information, the present application can use the HL7 v3 RIM to define elements such as patient, diagnosis, examination / test result, etc. as corresponding medical entities, and define corresponding structured attributes for each medical entity. Specifically, the present application can define that a patient includes basic information and past medical history, a diagnosis includes attributes such as disease name and disease type, and an examination / test result includes attributes such as examination / test item, time, and result. At the same time, the present application can define the relationship between a patient and a diagnosis, and the relationship between a patient and an examination / test.

[0177] For the three standardized medical data models constructed as described above, after the unstructured medical record data of the current patient is input into the generative medical record model to obtain the first type of structured medical record data and semi-structured medical record data of the current patient, the present application can further input the semi-structured medical record data into the three standardized medical data models to structure map each medical vocabulary in the semi-structured medical record data, so as to obtain the structure category to which each medical vocabulary belongs, thereby obtaining the second type of structured medical record data of the current patient.

[0178] According to one or more embodiments in the present application, the structured accuracy of the electronic medical record is ensured, and the present application can illustrate the specific processing process of the structured medical record data of the patient by the medical record structured knowledge base.

[0179] Figure 6 A method flowchart for generating an electronic medical record based on a medical record structured knowledge base is shown in the embodiments of the present application. As shown in Figure 6 the method can include the following steps:

[0180] S610, based on the constructed medical record structured knowledge base, normalizing and comparing the first type of structured medical record data and the second type of structured medical record data to determine the disease category and the standard clinical pathway of the current patient.

[0181] Since the medical record structured knowledge base can include a large number of medical vocabulary, medical terminology, and related structured medical knowledge points of diagnosis and treatment methods. Moreover, the medical record structured knowledge base is constructed based on a large number of historical unstructured medical record data and medical literature, and its scientificity and accuracy are greatly guaranteed.

[0182] Then, after obtaining the first type of structured medical record data and the second type of structured medical record data of the current patient, the present application can perform a normalized one-to-one comparison between the structured medical knowledge points in the medical record structured knowledge base and the first type of structured medical record data and the second type of structured medical record data to determine each structured medical knowledge point related to the first type of structured medical record data and the second type of structured medical record data. Then, by comprehensively analyzing each structured medical knowledge point related to the first type of structured medical record data and the second type of structured medical record data, the disease category and the standard clinical pathway of the current patient can be determined.

[0183] The standard clinical pathway can explicitly represent a set of standardized treatment modes and treatment procedures involved in the diagnosis and treatment of the disease category of the current patient, and can include but not limited to the treatment schedule, various diagnostic examinations, treatment target approach, treatment plan and prognosis target required in the diagnosis and treatment process of the disease category of the current patient.

[0184] S620, generating the electronic medical record of the current patient according to the electronic medical record template under the disease category and the standard clinical pathway.

[0185] Since the diagnosis items required to be performed by the patient under the same disease category are roughly the same, the medical record data required to be recorded under the same disease category is also roughly the same. Therefore, in the medical record structured knowledge base, by performing corresponding statistical analysis on the structured medical knowledge points related to each disease category, an appropriate electronic medical record template can be constructed for each disease category. Therefore, according to the disease category of the current patient, the present application can determine the electronic medical record template under the disease category in the medical record structured knowledge base.

[0186] Then, the application can automatically input the specific structure attributes corresponding to the first type of structured medical record data and the second type of structured medical record data of the current patient into the electronic medical record template under the corresponding disease category as the measured item information of the current patient under the corresponding disease category.

[0187] Moreover, in combination with the related diagnosis and treatment content provided by the above-mentioned standard clinical pathway, it can be automatically judged whether the current patient has missing examination and inspection data, and further increase the corresponding missing measurement item information in the above-mentioned electronic medical record template, so as to timely prompt the doctor whether to issue relevant medical orders, so that the current patient meets the scientific and standard clinical pathway, thereby avoiding the diagnosis and treatment deviation caused by incomplete medical record data, and improving the treatment effect and satisfaction of the patient. In combination with the related diagnosis and treatment content provided by the above-mentioned standard clinical pathway, it can also analyze whether the current patient has relatively similar associated diagnosis and treatment items, and further increase the corresponding associated diagnosis and treatment item information in the above-mentioned electronic medical record template to present to the doctor for reference and differential diagnosis, which helps the doctor to more accurately diagnose and treat the patient's disease, and improves the diagnosis and treatment efficiency and quality.

[0188] Therefore, through the automatic filling of the electronic medical record template, the electronic medical record of the current patient can be generated, and the electronic medical record can at least include the measured item information, the missing measurement item information and the associated diagnosis and treatment item information of the current patient under the corresponding disease category. Then, through the overlapping interactive "diagnosis and treatment" of artificial intelligence and human diagnosis, the application can provide the maximum diagnosis and treatment assistance for the current patient, and also improve the corresponding diagnosis and treatment efficiency, which can guarantee the professionalism of the doctor and also play the advantages of the current artificial intelligence, bringing new ideas and methods for the development of modern medicine.

[0189] In order to ensure the overall accuracy of the electronic medical record, the application can explain the specific construction process of the medical record structured knowledge base. As shown in Figure 7 The application can construct the medical record structured knowledge base through the following steps:

[0190] 710, inputting the multi-modal special disease medical record data under any disease category into the generated medical record model to obtain the first type of structured special disease medical record data and the semi-structured special disease medical record data under the disease category.

[0191] Since the diagnosis items required to be performed by the patient under the same disease category are roughly the same, the medical record data required to be recorded under the same disease category is also roughly the same. Therefore, in order to ensure the accuracy of the electronic medical record, the application can construct a suitable medical record structured knowledge base under each disease category.

[0192] Thus, after the generative medical record model and the standardized medical data model are completed, for each disease, the application can pre-acquire a large amount of historical unstructured medical record data of patients under the disease in a multi-modal manner as multi-modal special disease medical record data under the disease in the application.

[0193] Then, the multi-modal special disease medical record data under any disease is input into the constructed generative medical record model, and the multi-modal special disease medical record data is subjected to corresponding structured processing by the generative medical record model, so as to obtain the first type of structured special disease medical record data and the semi-structured special disease medical record data under the disease.

[0194] S720, the semi-structured special disease medical record data is input into the constructed standardized medical data model to obtain the second type of structured special disease medical record data under the disease.

[0195] For the semi-structured special disease medical record data under any disease, the application can continue to input the semi-structured special disease medical record data into the constructed standardized medical data model, and the semi-structured special disease medical record data is subjected to structured processing again by the standardized medical data model, so as to obtain the second type of structured special disease medical record data under the disease.

[0196] S730, according to the structured level in the first type of structured special disease medical record data and the second type of structured special disease medical record data, a medical record structured knowledge base under the disease is generated.

[0197] By structured analysis of the first type of structured special disease medical record data and the second type of structured special disease medical record data under any disease, the structured level of the medical vocabulary in each structured medical record data and the structure category to which each medical vocabulary belongs can be determined, thereby forming a target organization structure under the disease.

[0198] Taking the multi-modal special disease medical record data under a certain disease as an example, the multi-modal special disease medical record data is "male, age 45, chief complaint of right abdominal pain for 3 days, fever for 1 day", the target organization structure under the disease can be obtained as follows:

[0199] [Gender: male], [age: 45 years old], [chief complaint: [pain: [location: right abdomen], [period: 3 days]], [fever: [period: 1 day]]].

[0200] Then, according to the relevant structured medical knowledge points under each structured level in the target organizational structure of any disease, a Mysql database can be used to create a relevant table structure to store the structured medical record data represented by each structured medical knowledge point in the target organizational structure of the disease, and a corresponding specified field is defined for each structured medical record data, thereby generating a medical record structured knowledge base under the disease, so as to subsequently utilize the medical record structured knowledge base under any disease to more accurately, normatively and standardize record and manage the structured medical record data of each patient.

[0201] Figure 8 A principle block diagram of an electronic medical record generation device is shown in an embodiment of the present application. As shown in the figure, the device 800 can include: Figure 8

[0202] A first structured module 810 is configured to input the unstructured medical record data of a current patient into a constructed generative medical record model to obtain first type structured medical record data and semi-structured medical record data of the current patient.

[0203] A second structured module 820 is configured to input the semi-structured medical record data into a constructed standardized medical data model to obtain second type structured medical record data of the current patient.

[0204] An electronic medical record generation module 830 is configured to process the first type structured medical record data and the second type structured medical record data based on a constructed medical record structured knowledge base to generate an electronic medical record of the current patient.

[0205] In some implementable manners, the first structured module 810 can be specifically configured to:

[0206] input the unstructured medical record data of the current patient into the constructed generative medical record model to obtain corresponding structured words and structure categories of the structured words;

[0207] determine the first type structured words with non-empty structure categories and the structure categories of the first type structured words as the first type structured medical record data of the current patient;

[0208] determine the second type structured words with empty structure categories as the semi-structured medical record data of the current patient.

[0209] In some implementable manners, the generative medical record model is obtained by a generative model training module. The generative model training module can be configured to:

[0210] ​For each category of medical corpus, the medical corpus is input into a pre-constructed initial generative model to obtain corresponding corpus word segmentation and a predicted category of the corpus word segmentation;

[0211] According to a first word segmentation loss between the corpus word segmentation and medical vocabulary in the medical corpus and a first category loss between the predicted category of the corpus word segmentation and the category to which the medical corpus belongs, the initial generative model is initialized and trained;

[0212] The structured medical record label corresponding to the input multi-modal historical medical record data is determined, and the structured medical record label includes historical medical record vocabulary corresponding to the multi-modal historical medical record data and a category to which the historical medical record vocabulary belongs.

[0213] The multi-modal historical medical record data is input into the initial generative model after the initialization and training to obtain corresponding multi-modal word segmentation and a predicted category of the multi-modal word segmentation.

[0214] According to a second word segmentation loss between the multi-modal word segmentation and the historical medical record vocabulary and a second category loss between the predicted category of the multi-modal word segmentation and the category to which the historical medical record vocabulary belongs, the initial generative model is retrained to obtain a corresponding generative medical record model.

[0215] In some implementable manners, the electronic medical record generation apparatus 800 can further include a generative model updating module. The generative model updating module can be used to:

[0216] According to a diagnosis optimization loss corresponding to a difference between a diagnosis result output by the target labeled medical record data after passing through the generative medical record model and a diagnosis label of the target labeled medical record data and a retraining loss of the generative medical record model, the generative medical record model is trained and updated; and / or,

[0217] According to a difference between the first type of structured medical record data of any patient and a structured correction result of the first type of structured medical record data, the generative medical record model is trained and updated.

[0218] In some implementable manners, the second structured module 820 can be specifically used to:

[0219] Using the structured medical record mapping relationship defined in the constructed standardized medical data model, a structured mapping result of the semi-structured medical record data is determined to obtain the second type of structured medical record data of the current patient.

[0220] In some implementable manners, the standardized medical data model is obtained by a data model construction module. The data model construction module can be used to:

[0221] determining medical entities in a medical standardization system, structured attributes of the medical entities, and association relationships between the medical entities;

[0222] determining corresponding structured medical record mapping relationships according to the structured attributes of the medical entities and the association relationships between the medical entities, to obtain the standardized medical data model.

[0223] In some implementations, the electronic medical record generation module 830 can be specifically configured to:

[0224] based on the constructed medical record structured knowledge base, normalizing and comparing the first type of structured medical record data and the second type of structured medical record data to determine the disease category and the standard clinical pathway to which the current patient belongs;

[0225] generating the electronic medical record of the current patient according to the electronic medical record template under the disease category and the standard clinical pathway.

[0226] In some implementations, the medical record structured knowledge base is obtained by a knowledge base construction module. The knowledge base construction module can be configured to:

[0227] inputting multi-modal disease-specific medical record data under any disease category into a constructed generative medical record model to obtain first type of structured disease-specific medical record data and semi-structured disease-specific medical record data under the disease category;

[0228] inputting the semi-structured disease-specific medical record data into a constructed standardized medical data model to obtain second type of structured disease-specific medical record data under the disease category;

[0229] generating a medical record structured knowledge base under the disease category according to the structured levels in the first type of structured disease-specific medical record data and the second type of structured disease-specific medical record data.

[0230] In some implementations, the electronic medical record of the current patient at least includes measured item information, missed item information, and associated diagnosis and treatment item information of the current patient under the disease category.

[0231] In some implementations, the unstructured medical record data of the current patient is multi-modal medical record data entered in a multi-modal manner.

[0232] In the embodiments of the present application, the unstructured medical record data of the current patient is input into the generated medical record model, to obtain the first type of structured medical record data and semi-structured medical record data of the current patient. The semi-structured medical record data is input into the standardized medical data model, to obtain the second type of structured medical record data of the current patient. Then, based on the constructed medical record structured knowledge base, the first type of structured medical record data and the second type of structured medical record data are processed, that is, the electronic medical record of the current patient is generated, so as to realize the automatic generation of the structured electronic medical record, improve the accuracy and standardization of the medical record data in the electronic medical record, eliminate the tedious process of manually entering the medical record data in the electronic medical record, enhance the accurate and intuitive description of the patient's condition by the medical record structured knowledge base, and provide more efficient, convenient and reliable standardized diagnosis and treatment reference for doctors.

[0233] It should be understood that the device embodiments and the method embodiments can correspond to each other, and similar descriptions can be referred to the method embodiments. To avoid repetition, details are not described here. Specifically, Figure 8 The device 800 shown can perform any of the method embodiments of the present application, and the foregoing and other operations and / or functions of each module in the device 800 are respectively used to implement the corresponding processes in each method in the embodiments of the present application. For the sake of brevity, details are not described here.

[0234] The device 800 of the embodiments of the present application is described above in combination with the drawings from the perspective of functional modules. It should be understood that the functional modules can be realized by hardware, or by instructions in the form of software, or by a combination of hardware and software modules. Specifically, each step of the method embodiments in the embodiments of the present application can be completed by integrated logic circuits and / or software instructions in the processor. The steps of the method disclosed in the embodiments of the present application can be directly embodied as hardware code processing for execution, or executed by a combination of hardware and software modules in the code processing. Alternatively, the software module can be located in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and combines the hardware to complete the steps in the above method embodiments.

[0235] Figure 9 The electronic device shown in the embodiments of the present application is a schematic block diagram.

[0236] As Figure 9 shown, the electronic device 900 can include:

[0237] The memory 910 is configured to store a computer program and transmit the program code to the processor 920. In other words, the processor 920 can call and run the computer program from the memory 910 to implement the method in the embodiments of the present application.

[0238] For example, the processor 920 can be configured to execute the method embodiments described above according to the instructions in the computer program.

[0239] In some embodiments of the present application, the processor 920 can include but is not limited to:

[0240] A general processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and the like.

[0241] In some embodiments of the present application, the memory 910 includes but is not limited to:

[0242] A volatile memory and / or a non-volatile memory. The non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically EPROM (EEPROM) or a flash memory. The volatile memory can be a random access memory (RAM) used as an external cache. By way of example but not limitation, many forms of RAM can be used, such as a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDR SDRAM), an enhanced SDRAM (ESDRAM), a synch link DRAM (SLDRAM) and a direct Rambus RAM (DR RAM).

[0243] In some embodiments of the present application, the computer program can be divided into one or more modules, which are stored in the memory 910 and executed by the processor 920 to complete the method provided by the present 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 electronic device.

[0244] As shown in Figure 9 The electronic device can further include:

[0245] The transceiver 930 can be connected to the processor 920 or the memory 910.

[0246] The processor 920 can control the transceiver 930 to communicate with other devices, specifically, can send information or data to other devices, or receive information or data sent by other devices. The transceiver 930 can include a transmitter and a receiver. The transceiver 930 can further include an antenna, and the number of antennas can be one or more.

[0247] It should be understood that various components in the electronic device are connected through a bus system, wherein the bus system includes a data bus, a power supply bus, a control bus and a state signal bus in addition to a data bus.

[0248] The present application also provides a computer storage medium, which stores a computer program, and the computer program makes the computer execute the method of the above-mentioned method embodiment when executed by the computer. Or, the present application embodiment also provides a computer program product containing instructions, which makes the computer execute the method of the above-mentioned method embodiment when executed by the computer.

[0249] When implemented in software, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media can be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired computer program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, or a twisted pair, as examples, then the coaxial cable, fiber optic cable, or twisted pair are included in the definition of medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), and Blu-Ray® disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0250] In one embodiment, the techniques described herein can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the software can be executed in a computer system, which can include one or more computers. The software can be stored on one or more computer readable media, such as a magnetic disk, optical disk, or solid state memory. The computer readable media can be distributed among one or more computer systems.

[0251] In several embodiments provided in the present application, it should be understood that the disclosed system, device, and method can be implemented in other ways. For example, the above-described device embodiments are merely illustrative, and the division of the modules is merely a logical function division. In actual implementation, another division manner can be used, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be omitted or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed modules can be indirect coupling or communication connection through some interfaces, devices, or modules, and can be electrical, mechanical, or other forms.

[0252] The modules illustrated as separate components may or may not be physically separate, and the components illustrated as modules may or may not be physical modules, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of the present application. For example, the functional modules in various embodiments of the present application can be integrated in one processing module, or each module can exist physically separately, or two or more modules can be integrated in one module.

[0253] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, 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 generating an electronic medical record, characterized by, include: The unstructured medical record data of the current patient is input into the constructed generative medical record model to obtain the first type of structured medical record data and semi-structured medical record data of the current patient. The semi-structured medical record data is input into the constructed standardized medical data model to obtain the second type of structured medical record data of the current patient; Based on the established structured medical record knowledge base, the first type of structured medical record data and the second type of structured medical record data are processed to generate the electronic medical record of the current patient. The generative medical record model is trained through the following steps: For each category of medical corpus, the medical corpus is input into a pre-built initial generative model to obtain the corresponding corpus word segmentation and the predicted category of the corpus word segmentation; The initial generative model is initialized and trained based on the first segmentation loss between the segmented words in the corpus and the medical vocabulary in the medical corpus, and the first category loss between the predicted category of the segmented words in the corpus and the category to which the medical corpus belongs. Determine the structured medical record tags corresponding to the entered multimodal historical medical record data. The structured medical record tags include the historical medical record vocabulary corresponding to the multimodal historical medical record data and the category to which the historical medical record vocabulary belongs. The multimodal historical medical record data is input into the initial generative model after initialization training to obtain the corresponding multimodal word segmentation and the predicted category of the multimodal word segmentation; The initial generative model is retrained based on the second segmentation loss between the multimodal segmentation and the historical medical record vocabulary, and the second category loss between the predicted category of the multimodal segmentation and the category to which the historical medical record vocabulary belongs, to obtain the corresponding generative medical record model.

2. The method of claim 1, wherein, The process of inputting the current patient's unstructured medical record data into the constructed generative medical record model to obtain the current patient's first type of structured medical record data and semi-structured medical record data includes: The unstructured medical record data of the current patient is input into the constructed generative medical record model to obtain the corresponding structured word segmentation and the structure category of the structured word segmentation; The first type of structured word segmentation with a non-empty structure category and the structure category of the first type of structured word segmentation are determined as the first type of structured medical record data of the current patient; The second type of structured word segmentation with an empty structure category is identified as the current patient's semi-structured medical record data.

3. The method of claim 1, wherein, The method further includes: Based on the diagnostic optimization loss corresponding to the difference between the diagnostic results output by the generative medical record model after processing the target-annotated medical record data and the diagnostic labels of the target-annotated medical record data, and the retraining loss of the generative medical record model, the generative medical record model is trained and updated; and / or, The generative medical record model is trained and updated based on the difference between the first type of structured medical record data of any patient and the structured correction results of the first type of structured medical record data.

4. The method of claim 1, wherein, The step of inputting the semi-structured medical record data into the constructed standardized medical data model to obtain the second type of structured medical record data of the current patient includes: The structured mapping relationship defined in the constructed standardized medical data model is used to determine a structured mapping result of the semi-structured medical record data, so as to obtain the second type of structured medical record data of the current patient.

5. The method according to claim 1 or 4, characterized in that, The standardized medical data model is obtained through the following steps: determining medical entities in a medical standardization system, structured attributes of the medical entities, and association relationships between the medical entities; determining corresponding structured medical record mapping relationships according to the structured attributes of the medical entities and the association relationships between the medical entities, so as to obtain the standardized medical data model.

6. The method of claim 1, wherein, The first type of structured medical record data and the second type of structured medical record data are processed based on the constructed medical record structured knowledge base to generate the electronic medical record of the current patient, including: based on the constructed medical record structured knowledge base, the first type of structured medical record data and the second type of structured medical record data are normalized and compared to determine the disease category to which the current patient belongs and the standard clinical pathway; generating the electronic medical record of the current patient according to the electronic medical record template under the disease category and the standard clinical pathway.

7. The method according to claim 1 or 6, characterized in that, The medical record structured knowledge base is obtained through the following steps: inputting multi-modal disease-specific medical record data under any disease category into a constructed generative medical record model to obtain the first type of structured disease-specific medical record data and the semi-structured disease-specific medical record data under the disease category; inputting the semi-structured disease-specific medical record data into a constructed standardized medical data model to obtain the second type of structured disease-specific medical record data under the disease category; generating a medical record structured knowledge base under the disease category according to the structured levels in the first type of structured disease-specific medical record data and the second type of structured disease-specific medical record data.

8. The method of claim 6, wherein, The electronic medical record of the current patient at least includes measured item information, missed item information and associated diagnosis and treatment item information of the current patient under the disease category.

9. The method of claim 1, wherein, The unstructured medical record data of the current patient is multi-modal medical record data entered through a multi-modal mode.

10. An electronic medical record generating apparatus, characterized by comprising: including: a first structured module for inputting unstructured medical record data of a current patient into a constructed generative medical record model to obtain first type of structured medical record data and semi-structured medical record data of the current patient; a second structured module for inputting the semi-structured medical record data into a constructed standardized medical data model to obtain second type of structured medical record data of the current patient; an electronic medical record generation module for processing the first type of structured medical record data and the second type of structured medical record data based on a constructed medical record structured knowledge base to generate an electronic medical record of the current patient; wherein the generative medical record model is obtained through a generative model training module, and the generative model training module is used to: for each type of medical corpus, input the medical corpus into a pre-constructed initial generative model to obtain corresponding corpus word segmentation and predicted categories of the corpus word segmentation; According to a first segmentation loss between the corpus segmentation and medical vocabulary in the medical corpus and a first category loss between a predicted category of the corpus segmentation and a category to which the medical corpus belongs, the initial generative model is initialized and trained; determining a structured medical record label corresponding to the input multi-modal historical medical record data, the structured medical record label including historical medical record vocabulary corresponding to the multi-modal historical medical record data and a category to which the historical medical record vocabulary belongs; inputting the multi-modal historical medical record data into the initialized and trained initial generative model to obtain corresponding multi-modal segmentation and a predicted category of the multi-modal segmentation; According to a second segmentation loss between the multi-modal segmentation and the historical medical record vocabulary and a second category loss between a predicted category of the multi-modal segmentation and a category to which the historical medical record vocabulary belongs, the initial generative model is retrained to obtain a corresponding generative medical record model.

11. An electronic device, comprising: comprising: a processor and a memory, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to execute the electronic medical record generation method of any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, a computer program for storing, the computer program enabling a computer to execute the electronic medical record generation method of any one of claims 1-9.

Images