Electronic medical record asynchronous structuring method and device, equipment and medium

Abstract

The invention relates to the technical field of medical data structuring, and discloses an electronic medical record asynchronous structuring method and device, equipment and a medium, and the method comprises the steps: carrying out the semantic feature fusion of multi-modal treatment data of a target patient, and obtaining the fusion treatment data; data structuring is conducted on the fused treatment data with the first granularity, and first structured data is obtained; under the condition that the deep structuring condition is met, data structuring is conducted on the first structured data with the second granularity, and second structured data is obtained; performing multi-stage clinical correctness verification on the second structured data to obtain a data verification result; and under the condition that the data verification result shows that the second structured data meets the clinical correctness requirement, obtaining the structured medical record according to the second structured data. The method has the beneficial effects that active resource allocation in the structured process is realized, the processing burden of doctors in the clinical diagnosis and treatment process is reduced, and the processing efficiency of electronic medical record structuring is improved on the basis of ensuring the accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical data structuring, and in particular to an asynchronous electronic medical record structuring method, device, equipment and medium. BACKGROUND

[0002] In the process of clinical diagnosis and treatment, the consultation data obtained by the doctor from the patient can be multi-modal data, including but not limited to the patient's self-feelings orally stated to the doctor, image or image data generated by the patient receiving medical examination, etc. In order to record the consultation of the patient in real time, the doctor usually needs to interrupt the ongoing diagnosis and treatment process, extract the key information in the consultation data of the patient to write the electronic medical record. For the consultation data in the form of multi-modal non-structured data, the doctor is difficult to directly determine the key information from it, which affects the standardization and completeness of the electronic medical record. Secondly, interrupting the diagnosis and treatment process also seriously interferes with the natural clinical workflow and reduces the efficiency of the consultation.

[0003] Therefore, it is urgent to propose a technical solution that can ensure the completeness of the electronic medical record while effectively reducing the interference of the electronic medical record generation process on the clinical operation. SUMMARY

[0004] The present application provides an asynchronous electronic medical record structuring method, device, equipment and medium, which divides the medical record structuring process into multiple stages, realizes active resource allocation in the structuring process, and segments and progressively structures the consultation data in multiple granularities, thereby reducing the processing burden of the doctor in the clinical diagnosis and treatment process and improving the processing efficiency of the electronic medical record structuring on the basis of ensuring the accuracy.

[0005] In order to achieve the above-mentioned purpose, the main technical solution adopted by the present application includes: In a first aspect, the present application provides an asynchronous electronic medical record structuring method, which comprises: obtaining multi-modal consultation data of a target patient, and performing semantic feature fusion on the multi-modal consultation data to obtain fusion consultation data of the target patient; performing data structuring on the fusion consultation data in a first granularity to obtain first structured data; performing data structuring on the first structured data in a second granularity under the condition that a deep structuring condition is met, to obtain second structured data; wherein the second granularity is smaller than the first granularity; performing multi-level clinical correctness verification on the second structured data to obtain a data verification result of the second structured data; and obtaining a structured medical record of the target patient according to the second structured data under the condition that the data verification result indicates that the second structured data meets the clinical correctness requirement.

[0006] The electronic medical record asynchronous structured method proposed in the embodiments of the present application performs multi-modal fusion on the multi-modal consultation data of a target patient to obtain fused consultation data; performs preliminary structuring on the fused consultation data in a coarse granularity to obtain first structured data; on this basis, further structures the first structured data in a fine granularity under the condition that a deep structuring condition is met to obtain second structured data; in the case that the second structured data passes verification and it is confirmed that the second structured data meets the clinical correctness requirement, the second structured data is output as structured medical records. Compared with related technologies, the present application divides the medical record structuring process into multiple stages, so that active resource allocation can be performed on each stage respectively, thereby relieving the computing power resource pressure caused by the structured process and improving the balance of computing power resource allocation. In addition, the present application progressively structures the consultation data in different granularities in each stage, which can preliminarily process in a coarse granularity in the clinical diagnosis and treatment process, reduces the time occupancy of the structured processing on the clinical diagnosis and treatment process, reduces the processing burden of the doctor, and further processes in a fine granularity without affecting the clinical diagnosis and treatment process, improves the comprehensiveness of the structured result, and further ensures the standardization and integrity of the electronic medical record.

[0007] Optionally, the data structuring of the first structured data in the second granularity to obtain second structured data comprises: performing medical entity extraction on the first structured data in the second granularity to obtain deep structured entities; performing entity matching on the deep structured entities in a structured knowledge graph to determine graph knowledge entities corresponding to the deep structured entities; performing inference enhancement on the deep structured entities according to the entity association relationship of the graph knowledge entities in the structured knowledge graph to obtain enhanced supplementary information of the deep structured entities; obtaining the second structured data according to the deep structured entities and the enhanced supplementary information.

[0008] Optionally, the method is executed by an asynchronous structuring system; whether the deep structuring condition is met is determined by: performing load detection on the asynchronous structuring system to obtain a load detection result; when the load detection result indicates that the asynchronous structuring system is in an idle state, it is determined that the deep structuring condition is met.

[0009] Optionally, the semantic feature fusion of the multi-modal consultation data to obtain the fused consultation data of the target patient comprises: performing feature extraction and unified semantic space mapping on the multi-modal consultation data respectively to obtain multi-modal data features; performing multi-modal attention fusion on the multi-modal data features to obtain fusion data features of the multi-modal medical data; performing semantic text integration on the fusion data features according to a clinical scene rule to obtain the fusion medical data.

[0010] Optionally, the multi-modal medical data includes voice text and printed text, and examination data and historical medical records of the target patient; the multi-modal data features are obtained by performing feature extraction and unified semantic space mapping on the multi-modal medical data, including: performing time sequence logic analysis on the voice text to extract sequence features of the voice text; and performing spatial layout analysis on the printed text to extract layout features of the printed text; performing data quantitative analysis on the examination data of the target patient to extract numerical features of the examination data; and performing entity extraction coding on the historical medical records of the target patient to extract structured features of the historical medical records; obtaining the multi-modal data features according to the sequence features, the layout features, the numerical features and the structured features.

[0011] Optionally, the data verification result of the second structured data is obtained by performing multi-level clinical correctness verification on the second structured data, including: performing text form verification on the second structured data to obtain a text form verification result of the second structured data; in a case where the text form verification result indicates that the second structured data meets a preset text form requirement, performing semantic logic verification on the second structured data to obtain a semantic logic verification result of the second structured data; in a case where the semantic logic verification result indicates that the second structured data meets a preset semantic logic requirement, performing medical logic verification on the second structured data to obtain a medical logic verification result of the second structured data; obtaining the data verification result according to the text form verification result, the semantic logic verification result and the medical logic verification result.

[0012] Optionally, the fusion medical data is obtained according to a multi-modal data fusion model; and the method further includes: in a case where the data verification result indicates that the second structured data does not meet the clinical correctness requirement, performing abnormal alarm on the second structured data to remind medical staff to perform abnormal correction on the second structured data; Obtain abnormal correction data of the second structured data, and perform parameter adjustment on the multi-modal data fusion model according to the abnormal correction data to optimize the multi-modal data fusion model.

[0013] In a second aspect, an electronic medical record asynchronous structuring device is provided, and the device comprises: A multi-modal data fusion module is configured to obtain multi-modal consultation data of a target patient, perform semantic feature fusion on the multi-modal consultation data, and obtain fusion consultation data of the target patient. A first data structuring module is configured to perform data structuring on the fusion consultation data at a first granularity, and obtain first structured data. A second data structuring module is configured to perform data structuring on the first structured data at a second granularity when a deep structuring condition is met, and obtain second structured data; the second granularity is smaller than the first granularity. A clinical correctness verification module is configured to perform multi-level clinical correctness verification on the second structured data, and obtain a data verification result of the second structured data; when the data verification result indicates that the second structured data meets a clinical correctness requirement, obtain a structured medical record of the target patient according to the second structured data.

[0014] In a third aspect, a computer device is provided, comprising a memory and a processor, which are communicatively connected with each other, and the memory stores computer instructions; the processor executes the computer instructions to perform the method in any of the above embodiments.

[0015] In a fourth aspect, a computer readable storage medium is provided, which stores computer instructions; the computer instructions are used to make a computer execute the method in any of the above embodiments.

[0016] In a fifth aspect, a computer program product is provided, which comprises computer instructions; the computer instructions are used to make a computer execute the method in any of the above embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0018] Figure 1 A step diagram of an electronic medical record asynchronous structuring method provided for an embodiment of the present application is provided. Figure 2 A time flow diagram of an electronic medical record asynchronous structuring method in an embodiment of the present application is provided. Figure 3 A step diagram of data structuring in a second granularity in an embodiment of the present application is provided. Figure 4 A step diagram of judging a deep structuring condition in an embodiment of the present application is provided. Figure 5 A step diagram of semantic feature fusion in an embodiment of the present application is provided. Figure 6 A step diagram of feature extraction and unified semantic space mapping in an embodiment of the present application is provided. Figure 7 A step diagram of multi-level clinical correctness verification in an embodiment of the present application is provided. Figure 8 A step diagram of second structured data verification result feedback in an embodiment of the present application is provided. Figure 9 A module diagram of an electronic medical record asynchronous structuring device provided for an embodiment of the present application is provided. Figure 10 A structural schematic diagram of a computer device provided for an embodiment of the present application is provided. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of 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 protection of the present application.

[0020] In a clinical diagnosis and treatment process, the consultation data obtained by a doctor from a patient can be multi-modal data, including but not limited to the patient's own feelings orally stated to the doctor, image or image data generated by the patient receiving medical examination, etc. In order to record the consultation situation of the patient in real time, the doctor usually needs to interrupt the ongoing diagnosis and treatment process, extract the key information in the consultation data of the patient to write an electronic medical record. For the consultation data in the form of multi-modal unstructured data, the doctor is difficult to directly determine the key information therefrom, resulting in the standardization and completeness of the medical record being affected. Secondly, interrupting the diagnosis and treatment process also seriously interferes with the natural clinical workflow, reducing the efficiency of the consultation. Therefore, it is urgent to propose a technical solution which can ensure the completeness of the electronic medical record while effectively reducing the interference of the electronic medical record generation process on the clinical operation.

[0021] Based on the above problems, the electronic medical record asynchronous structuring method, device, equipment and medium provided by the application are provided. The multi-modal consultation data of a target patient is subjected to semantic feature fusion to obtain fused consultation data. The fused consultation data is subjected to data structuring at a first granularity to obtain first structured data. The first structured data is subjected to data structuring at a second granularity under the condition that a deep structuring condition is met to obtain second structured data. The second structured data is subjected to multi-level clinical correctness verification to obtain a data verification result. In the case where the data verification result indicates that the second structured data meets the clinical correctness requirement, a structured medical record is obtained according to the second structured data.

[0022] The electronic medical record asynchronous structuring method provided by the application is used to fuse the multi-modal consultation data of a target patient to obtain fused consultation data. The fused consultation data is subjected to preliminary structuring at a coarse granularity to obtain first structured data. On this basis, the first structured data is subjected to further structuring at a fine granularity under the condition that a deep structuring condition is met to obtain second structured data. In the case where the second structured data passes the verification and it is confirmed that the second structured data meets the clinical correctness requirement, the second structured data is output as a structured medical record.

[0023] Compared with the related art, the medical record structuring process is divided into multiple stages in the application, so that active resource allocation can be performed on each stage respectively, thereby relieving the computing power resource pressure caused by the structuring process and improving the balance of computing power resource allocation.

[0024] In addition, the consultation data is subjected to progressive structuring at different granularities in each stage. In the clinical diagnosis and treatment process, the consultation data can be preliminarily processed at a coarse granularity to reduce the time occupancy of the structuring process on the clinical diagnosis and treatment process, thereby reducing the processing burden of the doctor. Further processing can be performed at a fine granularity without affecting the clinical diagnosis and treatment process, thereby improving the comprehensiveness of the structuring result and ensuring the standardization and integrity of the electronic medical record.

[0025] According to an embodiment of the application, an electronic medical record asynchronous structuring method is provided. It should be noted that the steps shown in the flowchart can be executed in a computer system such as a group of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0026] Referring to Figure 1 The electronic medical record asynchronous structuring method provided in the embodiment includes: S100. Obtain multi-modal consultation data of a target patient, fuse the multi-modal consultation data according to semantic features, and obtain fused consultation data of the target patient.

[0027] S200. The fused medical data is structured at the first granularity to obtain the first structured data.

[0028] S300. Under the condition of satisfying the deep structuring, the first structured data is structurated with the second granularity to obtain the second structured data; wherein the second granularity is smaller than the first granularity.

[0029] S400. Perform multi-level clinical correctness verification on the second structured data to obtain the data verification result of the second structured data; if the data verification result indicates that the second structured data meets the clinical correctness requirements, obtain the structured medical record of the target patient based on the second structured data.

[0030] Multimodal medical data can be multimodal data generated by the target patient during the medical visit, including but not limited to the patient's verbal description of their feelings to the doctor, examination data and imaging data generated by the target patient's medical examination, and the patient's historical medical records generated during previous medical visits. The above data can be in the form of text, audio, images or videos.

[0031] The first granularity can be the granularity for preliminary structuring of the target patient's fused medical data, corresponding to multiple fields in the first structured data, including but not limited to the target patient's chief complaint and present medical history. The second granularity can be the granularity for further structuring of the target patient's fused medical data, corresponding to multiple fields in the second structured data, including but not limited to the target patient's onset time, triggering factors, and accompanying symptoms. It can be understood that the second granularity is smaller than the first granularity, and the second field corresponding to the second granularity can be a subfield of the first field corresponding to the first granularity. In this case, the content of the second field can be a refined version of the first field. The second field can also differ from the first field, structuring the target patient's medical data from different perspectives to improve the comprehensiveness of the structured medical record.

[0032] Specifically, during clinical diagnosis and treatment, the patient's medical data is collected and recorded in real time to obtain multimodal patient data. This multimodal data can include the patient's verbal descriptions of their feelings to the doctor, examination data and imaging materials from medical examinations, and historical medical records from previous visits. The patient's verbal descriptions can be obtained by recording the patient's conversation with the doctor in real time using an Automatic Speech Recognition (ASR) module during clinical diagnosis and treatment, and then transcribing the recordings in real time using pre-trained acoustic and language models based on the medical scenario. The patient's examination data and imaging materials can be obtained by connecting to a Laboratory Information System (LIS) or a Picture Archiving and Communication System (PACS) to retrieve the patient's test values ​​and structured imaging conclusions from the system. The target patient's medical history can be obtained by connecting to an electronic medical record (EMR) system and extracting key information from the medical history.

[0033] Understandably, after obtaining the multimodal medical data of the target patient, the data from different modalities is scattered in the temporal dimension due to their different generation times, failing to accurately correspond to the relevant medical scenarios. Therefore, it is necessary to align the multimodal medical data in both the temporal and semantic dimensions to achieve accurate association between them. The temporal alignment process can include: for any modality of medical data, timestamping the data based on its generation time; arranging the multimodal medical data chronologically based on the timestamps to form the target patient's treatment timeline, representing the order and sequence of all medical actions during the patient's visit. The semantic alignment process can include: pre-constructing a medical knowledge graph, which covers the relationships between various medical entities, including but not limited to diseases, symptoms, test data, imaging data, and medication information; and performing semantic recognition and association analysis on the multimodal medical data based on the medical knowledge graph to obtain the medical logical relationships between them, providing a pathological logical basis for semantic feature fusion.

[0034] Furthermore, based on the temporal and medical logical relationships between multimodal medical data, semantic feature fusion is performed on the multimodal medical data. By integrating the associated medical data, multiple medical scenarios for the target patient are formed, resulting in fused medical data for the target patient. It is understandable that this fused medical data can be used to summarize any medical process of the target patient and continuously depict the evolution of the target patient's condition, thereby completely and accurately preserving the target patient's medical data and improving the efficiency and accuracy of medical communication.

[0035] Furthermore, the fused medical data is structured at a first granularity to extract key information corresponding to the first field, resulting in first structured data. In some embodiments, the process of structuring the fused medical data at the first granularity can be implemented through a first structured model. The first structured model can be a lightweight model, including a distilled Large Language Model (LLM) or an efficient sequence model, characterized by small parameter size and low computational requirements. The first structured model can be triggered in real time during clinical diagnosis and treatment to synchronously perform data structuring during the doctor's consultation with the target patient, effectively reducing the interference of the data structuring process on clinical diagnosis and treatment and improving the doctor's diagnostic and treatment efficiency.

[0036] Understandably, after obtaining the first structured data, doctors need to verify its content and promptly correct any errors to ensure its accuracy. The first granularity is coarse-grained; the first field can contain semi-structured text primarily in natural language, thus preserving the logic of the patient's clinical description during their visit and ensuring a complete record of the medical data. When verifying the first structured data, doctors can understand and confirm it based on natural language logic, reducing their workload.

[0037] Furthermore, after obtaining and confirming the first structured data, it is determined whether the current condition for deep structuring is met. If the condition is met, the first structured data is structured at a second granularity, extracting key information corresponding to the second field to obtain the second structured data. In some embodiments, the process of structuring the fused medical data at a second granularity can be implemented through a second structuring model. The second structuring model can be a large-scale ensemble model, including a large language model with hundreds of billions of parameters or a multimodal fusion model, etc., characterized by large parameter scale and high computing power requirements, resulting in a long structuring time. Compared with the first structuring model, the second structuring model has the ability to semantically decompose and standardize fine-grained fields. It can perform deep semantic understanding of the first structured fields according to a preset standardized template to obtain second structured data containing highly structured data, improving the comprehensiveness of the structuring results and thus ensuring the standardization and integrity of electronic medical records.

[0038] Understandably, the number of structured processes occurring simultaneously during clinical diagnosis and treatment is typically far greater than the number occurring outside of clinical procedures. However, the total computing resources of the asynchronous structured systems used to execute these processes are constant. This leads to situations where asynchronous structured systems experience computing resource constraints during clinical diagnosis and treatment, while their computing resources are idle outside of these processes. Consequently, computing resources cannot be effectively utilized, impacting the efficiency of structuring medical data. This application divides the structuring process of multimodal medical data into stages by setting a first granularity and a second granularity. The first stage uses the first granularity for data structuring, while the second stage uses the second granularity. The first and second stages can be performed at different times. The first stage, with its coarse-grained structuring, reduces the computing resource requirements of the structured process, enabling parallel execution of multiple processes during clinical diagnosis and treatment, thus improving data structuring efficiency. The second stage of fine-grained structuring demands higher computing resources. It's crucial to fully utilize the idle computing resources of the asynchronous structuring system during periods outside of clinical diagnosis and treatment to ensure the accuracy and completeness of the structured medical records. This embodiment proactively allocates resources for both the first and second stages, ensuring each stage receives the necessary computing resources while alleviating the computing resource pressure caused by the structuring process and improving the balance of computing resource allocation.

[0039] It should be noted that deep structuring conditions can be various conditions related to the clinical diagnosis and treatment process. Whether further data structuring is needed depends on the progress of the clinical diagnosis and treatment process, in order to adapt to the actual clinical process and reduce interference with clinical diagnosis and treatment. For example, deep structuring conditions could be the completion of the clinical diagnosis and treatment process or being in the stage of awaiting test results.

[0040] In some embodiments, the second-granularity data structuring process can also be performed on the fused medical data, structuring the fused medical data at the second granularity to obtain third-structured data. Multi-field data fusion is then performed on the first and third structured data to obtain a structured fusion result. It is understood that the second field may correspond to the same content as the first field, or it may correspond to different content. The third structured data and the first structured data can complement or verify each other. By fusing the first and third structured data, the information richness of the structured data can be improved while eliminating ambiguities in the first or third structured data, effectively improving the robustness and accuracy of the structured fusion result, thereby ensuring the accuracy of the structured medical record.

[0041] Furthermore, after obtaining the second structured data, multi-level clinical correctness verification is performed on it to progressively verify the second structured data across multiple dimensions, resulting in data verification results. In some embodiments, each level of the multi-level clinical correctness verification process can be implemented in different ways, including but not limited to rule verification based on rule engines, entity relationship verification based on knowledge graphs, and semantic logic verification based on large language models. It is understood that the verification results at different levels can be complementary or coordinated. The data verification result obtained by fusing the multi-level verification results improves the comprehensiveness and anomaly detection rate of the data verification results, thereby accurately identifying anomalies in the second structured data and improving the accuracy and rationality of the structured medical records.

[0042] In some embodiments, when anomalies occur in the second structured data, these anomalies are fed back to the first structured model to optimize and adjust its structured knowledge base. The optimized first structured model then restructures the fused medical records at a first granularity to obtain the fourth structured data. If the conditions for deep structuring are met, the fourth structured data is then structured at a second granularity to obtain the fifth structured data. If the fifth structured data meets clinical accuracy requirements, the target patient's structured medical record is obtained based on it. It is understood that the structured knowledge base upon which the first structured model relies for data structuring may have certain defects, failing to effectively structure some medical records, leading to ambiguity in the structured data. When anomalies occur in the second structured data, these anomalies represent the defects in the structured knowledge base. Therefore, by feeding back the anomalies in a closed loop to the first structured model for targeted optimization, the structuring capability of the first structured model is improved, enhancing the accuracy and completeness of the fourth structured data and providing an effective data foundation for the second-granularity data structuring process.

[0043] Furthermore, if the data verification results indicate that the second structured data meets clinical correctness requirements, it means that there are no anomalies in the second structured data, it conforms to clinical medical logic, and the data format meets the standardization requirements. At this point, the second structured data is output as the target patient's structured medical record to the electronic medical record system for archiving and storage.

[0044] Reference Figure 2 As shown, in some embodiments, semantic feature fusion and first-granularity data structuring can be performed in a first temporal state, while second-granularity data structuring and multi-level clinical correctness verification can be performed in a second temporal state. The first temporal state differs from the second temporal state, preceding it in the temporal progression direction. The time interval between the first and second temporal states can be determined based on the progress of the clinical diagnosis and treatment process, and can transition from the first to the second temporal state when deep structuring conditions are met, for further data structuring processing. For example, the first temporal state can be a busy time when a doctor is conducting a clinical consultation with a target patient, and the second temporal state can be an idle time when no data structuring is in progress.

[0045] The asynchronous structuring method for electronic medical records provided in this embodiment performs multimodal fusion on the multimodal medical data of the target patient to obtain fused medical data; performs preliminary structuring on the fused medical data with coarse granularity to obtain first structured data; based on this, and under the condition of satisfying the deep structuring conditions, further structuring on the first structured data with fine granularity to obtain second structured data; and when the second structured data passes the verification and is confirmed to meet the clinical correctness requirements, the second structured data is output as a structured medical record.

[0046] Compared with related technologies, this application divides the medical record structuring process into multiple stages, thereby enabling proactive resource allocation for each stage, alleviating the computational resource pressure caused by the structured process, and improving the balance of computational resource allocation.

[0047] Furthermore, this application progressively structures the medical data at different granularities at each stage. In the clinical diagnosis and treatment process, it can perform preliminary processing at a coarse granular level, reducing the time occupied by structured processing in the clinical diagnosis and treatment process, reducing the processing burden on doctors, and further processing at a fine granular level without affecting the clinical diagnosis and treatment process, thereby improving the comprehensiveness of the structured results and ensuring the standardization and integrity of electronic medical records.

[0048] Reference Figure 3 As shown, in one embodiment of this application, the first structured data is structuring at a second granularity to obtain second structured data, including: S310. Extract medical entities from the first structured data at the second granularity to obtain deep structured entities.

[0049] S320. Perform entity matching on deep structured entities in the structured knowledge graph to determine the graph knowledge entities corresponding to the deep structured entities.

[0050] S330. Based on the entity association relationships of the knowledge entities in the structured knowledge graph, perform reasoning enhancement on the deep structured entities to obtain enhanced supplementary information of the deep structured entities.

[0051] S340. Obtain second structured data based on deep structured entities and enhanced supplementary information.

[0052] The second structured model can include a built-in structured knowledge graph, which contains various medical entities and the relationships between them. Medical entities can include standard entities such as diseases, symptoms, drugs, tests, treatment plans, and medical terms. The relationships between medical entities can be related, causal, or hierarchical.

[0053] Specifically, based on the second field corresponding to the second granularity, medical entities are extracted from the first structured data to obtain deep structured entities. Entity matching is performed on the deep structured entities in the structured knowledge graph, and entity links are used to align the deep structured entities with the medical entities in the structured knowledge graph to determine the graph knowledge entities corresponding to the deep structured entities.

[0054] Furthermore, based on the entity relationships between knowledge entities in the structured knowledge graph, reasoning enhancement is performed on deep structured entities to identify the relationships between them and generate enhanced supplementary information for the deep structured entities. It is understandable that this enhanced supplementary information can be generated based on the implicit relationships between deep structured entities, or it can supplement the deep structured entities with additional information to improve the structured data.

[0055] For example, when the deep structured entities include amoxicillin and penicillin allergy history, the corresponding knowledge entity for amoxicillin in the structured knowledge graph can be amoxicillin (a β-lactam antibiotic), and the corresponding knowledge entity for penicillin allergy history can be penicillin-type drug allergy. In the structured knowledge graph, amoxicillin belongs to the penicillin class, and the entity association between amoxicillin and penicillin-type drug allergy indicates that amoxicillin should be used with caution in patients with penicillin allergy. In this case, the supplementary information could be that the target patient has a history of penicillin allergy and should use amoxicillin with caution. In another example, when the deep structured entities include amoxicillin and 2gtid, amoxicillin corresponds to a standard adult dose in the structured knowledge graph. The standard adult dose of amoxicillin is 0.5gtid, which is significantly less than 2gtid. In this case, the supplementary information could be that the target patient is using an overdose of amoxicillin.

[0056] Furthermore, text is generated based on the deep structured entity and enhanced supplementary information to obtain structured data corresponding to the second field, which serves as the second structured data.

[0057] Reference Figure 4 As shown, in one embodiment of this application, the method is executed through an asynchronous structured system; the condition for deep structuring is determined by the following method: S302. Perform load detection on the asynchronous structured system and obtain the load detection results.

[0058] S304. When the load detection result indicates that the asynchronous structured system is in an idle state, it is determined that the deep structured condition is met.

[0059] Specifically, the deep structuring conditions can be various conditions related to the clinical diagnosis and treatment process. Whether further data structuring is needed is determined based on the progress of the clinical diagnosis and treatment process to adapt to the actual clinical situation and reduce interference with clinical diagnosis and treatment. In this embodiment, the deep structuring condition can be that the asynchronous structuring system used to execute this method is in an idle state, with no ongoing structuring tasks.

[0060] Furthermore, based on the aforementioned deep structuring conditions, real-time load detection is performed on the asynchronous structured system to obtain the load detection results of the asynchronous structured system. When the load detection results indicate that the asynchronous structured system is in an idle state, it is determined that the deep structuring conditions are met, and the first structured data can be structurated at a second granularity using a second structuring model.

[0061] Reference Figure 5 As shown, in one embodiment of this application, semantic feature fusion is performed on multimodal medical data to obtain fused medical data of the target patient, including: S110. Perform feature extraction and unified semantic space mapping on the multimodal medical data to obtain multimodal data features.

[0062] S120. Perform multimodal attention fusion on the multimodal data features to obtain the fused data features of the multimodal medical visit data.

[0063] S130. Based on clinical scenario rules, semantic text integration is performed on the features of the fused data to obtain fused medical data.

[0064] Specifically, multimodal medical data corresponds to different modalities. Features are extracted from the medical data of each modality, and the data features of each modality are mapped to a unified semantic space to obtain multimodal data features. This allows the data features of different modalities to be compared and fused under the same benchmark. For example, the process of feature extraction and unified semantic space mapping can be implemented through a Transformer-based cross-modal encoder.

[0065] Furthermore, a linear transformation is performed based on the multimodal data features to generate attention vectors. These attention vectors are then input into an attention mechanism, where attention scores are calculated for each vector. Weight normalization is applied to the attention scores of all attention vectors to obtain their respective attention weights. Finally, attention fusion is performed on the attention vectors based on these weights to obtain the fused data features of the multimodal medical data.

[0066] Furthermore, clinical scenario rules can be pre-defined, representing standardized rules for recording patient data during clinical diagnosis and treatment. For example, in a consultation scenario, the patient's chief complaint should concisely summarize the main symptoms and duration. Based on these clinical scenario rules, semantic text integration is performed on the fused data features to form text data describing the target patient's consultation process, outputting the target patient's fused consultation data.

[0067] Reference Figure 6 As shown in one embodiment of this application, the multimodal medical data includes voice text, printed text, examination data of the target patient, and historical medical records; feature extraction and unified semantic space mapping are performed on the multimodal medical data to obtain multimodal data features, including: S112. Perform temporal logic analysis on the speech text to extract its sequence features; perform spatial layout analysis on the printed text to extract its page layout features.

[0068] S114. Perform quantitative data analysis on the examination data of the target patient and extract the numerical features of the examination data; perform entity extraction coding on the historical medical records of the target patient and extract the structured features of the historical medical records.

[0069] S116. Obtain multimodal data features based on sequence features, layout features, numerical features, and structured features.

[0070] Specifically, multimodal medical data can include the target patient's verbal descriptions of their feelings to the doctor, examination data and imaging materials generated from the target patient's medical examinations, and historical medical records generated during the target patient's past medical visits. For the speech text describing the target patient's feelings, sequence features can be extracted from the speech text based on the temporal logic of the consultation dialogue between the doctor and the target patient. The specific process can include: cleaning and standardizing the speech text; inputting the speech text into a large language model, and using the large language model to perform semantic recognition and logical extraction on the speech text to obtain the temporal logic of the consultation dialogue in the speech text, which serves as the sequence features of the speech text.

[0071] For printed text describing the target patient's medical history, layout features can be extracted based on the spatial layout logic to represent the relationship between the medical history and key fields, as well as the importance correlation of each key field. For the target patient's examination data, quantitative analysis is performed on the examination data for each indicator, extracting numerical features from the examination data through methods such as threshold comparison or trend analysis. For the target patient's historical medical records, entity extraction can be performed on the structured codes under each standard field in the historical medical records to obtain information such as the target patient's historical diagnosis results and medication records, which serve as structured features. It is understood that the above features are collectively used as the multimodal data features of the target patient to provide a data foundation for the subsequent data structuring process.

[0072] Reference Figure 7 As shown, in one embodiment of this application, multi-level clinical correctness verification is performed on the second structured data to obtain the data verification result of the second structured data, including: S410. Perform text form validation on the second structured data to obtain the text form validation result of the second structured data.

[0073] S420. If the text form verification result indicates that the second structured data meets the preset text form requirements, perform semantic logic verification on the second structured data to obtain the semantic logic verification result of the second structured data.

[0074] S430. If the semantic logic verification result indicates that the second structured data meets the preset semantic logic requirements, perform medical logic verification on the second structured data to obtain the medical logic verification result of the second structured data.

[0075] S440. Obtain the data verification results based on the text form verification results, semantic logic verification results, and medical logic verification results.

[0076] Specifically, the second structured data undergoes text format verification to analyze whether it conforms to preset text format requirements, thus obtaining the text format verification result. In some embodiments, text format verification can be implemented through a rule engine, which corresponds to a clinical rule base. The clinical rule base stores various rules related to the preset text format requirements, including but not limited to medical record writing standards, medical operation routines, and medication guidelines. In the text format verification, the second structured data is matched one by one based on the rules in the clinical rule base to obtain the matching result between the second structured data and the clinical rule base. When the second structured data does not match the clinical rule base, it is marked as an anomaly, which serves as the text format verification result.

[0077] Furthermore, if the text format verification result indicates that the second structured data meets the preset text format requirements, semantic logic verification is performed on the second structured data to analyze whether it meets the preset semantic logic requirements, thus obtaining the semantic logic verification result of the second structured data. In some embodiments, semantic logic verification can be implemented through a large language model. The second structured data is input into the large language model, and standardized preprocessing is performed to convert the second structured data into a format suitable for processing by the large language model. The large language model performs semantic understanding on the format-converted second structured data to obtain the semantic logic in the second structured data. Consistency verification, coherence verification, and logical conflict verification are performed on the semantic logic of the second structured data, and anomaly marking is applied to second structured data with semantic logic anomalies, thus obtaining the medical logic verification result.

[0078] Further, if the semantic logic verification result indicates that the second structured data meets the preset semantic logic requirements, a medical logic verification is performed on the second structured data to obtain the medical logic verification result of the second structured data. In some embodiments, the medical logic verification can be implemented through a verification knowledge graph, which contains various medical entities and their entity relationships. Medical entities may include diseases, drugs, and symptoms, and entity relationships may include association relationships or causal relationships. The medical knowledge graph, the structured knowledge graph, and the verification knowledge graph can all be the same or different knowledge graphs. Entity extraction is performed on the second structured data to obtain the structured entities and their structured relationships. Entity matching is performed on the structured entities based on the medical entities in the verification knowledge graph to determine the standard entities corresponding to the structured entities in the verification knowledge graph and their standard relationships. The structured entities are verified and compared with the standard entities, and the structured relationships are verified and compared with the standard relationships. The system analyzes whether the second structured data meets the preset medical logic requirements, and anomalies are marked on the second structured data when any aspect of the structured entities or structured relationships is abnormal, thus obtaining the medical logic verification result.

[0079] Understandably, conventional methods may only verify whether structured data meets preset requirements in a single dimension, such as focusing on the text format requirements of structured data or only verifying issues like data format or required fields. This embodiment improves upon the single verification method of conventional approaches by employing multiple verification methods to progressively verify the standardization of the second structured data across multiple dimensions. This achieves complementary advantages between different verification methods, effectively expanding the verification scope, improving the comprehensiveness and accuracy of verification results, and ultimately enhancing the clinical correctness and standardization of structured data, which is beneficial for the intelligent application of medical record data.

[0080] Reference Figure 8 As shown, in one embodiment of this application, the fused medical data is obtained based on a multimodal data fusion model; the method further includes: S450. If the data verification results indicate that the second structured data does not meet the clinical correctness requirements, an anomaly alert will be issued for the second structured data to remind medical staff to correct the anomaly.

[0081] S460. Obtain the anomaly correction data for the second structured data, and adjust the parameters of the multimodal data fusion model based on the anomaly correction data to optimize the multimodal data fusion model.

[0082] Specifically, the data validation results include text-based validation results, semantic logic validation results, and medical logic validation results. If an anomaly marker is found in any of these validation results, the second structured data is determined to not meet clinical correctness requirements. In this case, an anomaly alert is issued based on the content of the anomaly marker, reminding medical staff to correct the anomaly in the second structured data.

[0083] Furthermore, in response to the correction operations performed by medical staff on the second structured data, abnormal correction data for the second structured data is obtained. The abnormal correction data can be the second structured data after abnormal correction. The second structured data before and after abnormal correction are combined to obtain a supervision data pair. The supervision data pair is fed back to the multimodal data fusion model, and the multimodal data fusion model is retrained based on the supervision data pair to adjust the model parameters, improve the accuracy of the fused medical data output by the multimodal data fusion model, and thus improve the accuracy and standardization of structured data and structured medical records.

[0084] Accordingly, please refer to Figure 9 This application provides an asynchronous structuring device for electronic medical records, the device comprising: The multimodal data fusion module 910 is used to acquire multimodal medical data of the target patient, perform semantic feature fusion on the multimodal medical data, and obtain fused medical data of the target patient.

[0085] The first data structuring module 920 is used to perform data structuring on the fused medical data at a first granularity to obtain the first structured data.

[0086] The second data structuring module 930 is used to perform data structuring on the first structured data at a second granularity under the condition of satisfying the deep structuring condition, so as to obtain the second structured data; wherein the second granularity is smaller than the first granularity.

[0087] The clinical correctness verification module 940 is used to perform multi-level clinical correctness verification on the second structured data to obtain the data verification result of the second structured data; if the data verification result indicates that the second structured data meets the clinical correctness requirements, the structured medical record of the target patient is obtained based on the second structured data.

[0088] In some alternative implementations, the second data structuring module 930 includes: The medical entity extraction unit is used to extract medical entities from the first structured data at a second granularity to obtain deep structured entities.

[0089] The knowledge entity matching unit is used to perform entity matching on deep structured entities in a structured knowledge graph to determine the graph knowledge entity corresponding to the deep structured entity.

[0090] The entity reasoning enhancement unit is used to enhance the reasoning of deep structured entities based on the entity association relationships of knowledge entities in the structured knowledge graph, and obtain enhanced supplementary information of deep structured entities.

[0091] The second structured data acquisition unit is used to obtain second structured data based on deep structured entities and enhanced supplementary information.

[0092] In some optional implementations, the second data structuring module 930 further includes a conditional structuring matching unit, comprising: The system load detection subunit is used to perform load detection on the asynchronous structured system and obtain the load detection results.

[0093] The idle state judgment subunit is used to determine whether the deep structuring condition is met when the load detection result indicates that the asynchronous structured system is in an idle state.

[0094] In some optional implementations, the multimodal data fusion module 910 includes: The feature extraction and mapping unit is used to extract features and map them in a unified semantic space for multimodal medical data to obtain multimodal data features.

[0095] The multimodal attention fusion unit is used to perform multimodal attention fusion on multimodal data features to obtain fused data features of multimodal medical data.

[0096] The semantic text integration unit is used to perform semantic text integration on the features of the fused data according to clinical scenario rules to obtain fused medical data.

[0097] In some optional implementations, the feature extraction mapping unit includes: The modal feature analysis subunit is used to perform temporal logic analysis on speech text and extract the sequence features of speech text; and to perform spatial layout analysis on printed text and extract the layout features of printed text.

[0098] The patient data analysis subunit is used to perform quantitative analysis of the examination data of the target patient and extract the numerical features of the examination data; and to perform entity extraction and coding of the historical medical records of the target patient and extract the structured features of the historical medical records.

[0099] The multimodal data feature acquisition subunit is used to obtain multimodal data features based on sequence features, layout features, numerical features, and structured features.

[0100] In some alternative implementations, the clinical correctness verification module 940 includes: The text form verification unit is used to perform text form verification on the second structured data and obtain the text form verification result of the second structured data.

[0101] The semantic logic verification unit is used to perform semantic logic verification on the second structured data when the text form verification result indicates that the second structured data meets the preset text form requirements, and to obtain the semantic logic verification result of the second structured data.

[0102] The medical logic verification unit is used to perform medical logic verification on the second structured data when the semantic logic verification result indicates that the second structured data meets the preset semantic logic requirements, and to obtain the medical logic verification result of the second structured data.

[0103] The verification result acquisition unit is used to obtain data verification results based on text-based verification results, semantic logic verification results, and medical logic verification results.

[0104] In some optional implementations, the clinical correctness verification module 940 further includes: The anomaly alarm correction unit is used to issue an anomaly alarm for the second structured data when the data verification results indicate that the second structured data does not meet the clinical correctness requirements, so as to remind medical staff to correct the anomaly in the second structured data.

[0105] The parameter adjustment and optimization unit is used to obtain anomaly correction data for the second structured data, and adjust the parameters of the multimodal data fusion model based on the anomaly correction data to optimize the multimodal data fusion model.

[0106] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0107] In this embodiment, the asynchronous structured electronic medical record device is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0108] Please see Figure 10 , Figure 10 This is a schematic diagram of a computer device according to an embodiment of this application. As shown in the figure, the computer device includes one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components communicate with each other using different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 10 Take a processor 10 as an example.

[0109] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0110] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0111] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0112] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0113] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0114] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0115] This application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any embodiment of this application.

[0116] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

[0117] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0118] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0119] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0120] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0121] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0122] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0123] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0124] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0125] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0126] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. An electronic medical record asynchronous structured method, characterized in that, The method comprises: acquiring multi-modal consultation data of a target patient, performing semantic feature fusion on the multi-modal consultation data to obtain fused consultation data of the target patient; performing data structuring on the fused consultation data at a first granularity to obtain first structured data; performing data structuring on the first structured data at a second granularity to obtain second structured data under the condition that a deep structuring condition is met, wherein the second granularity is smaller than the first granularity; performing multi-level clinical correctness verification on the second structured data to obtain a data verification result of the second structured data; and obtaining a structured medical record of the target patient according to the second structured data under the condition that the data verification result indicates that the second structured data meets the clinical correctness requirement.

2. The method of claim 1, wherein, The performing data structuring on the first structured data at a second granularity to obtain second structured data comprises: performing medical entity extraction on the first structured data at the second granularity to obtain deep structured entities; performing entity matching on the deep structured entities in a structured knowledge graph to determine graph knowledge entities corresponding to the deep structured entities; performing inference enhancement on the deep structured entities according to entity association relationships of the graph knowledge entities in the structured knowledge graph to obtain enhanced supplementary information of the deep structured entities; obtaining the second structured data according to the deep structured entities and the enhanced supplementary information.

3. The method of claim 1, wherein, The method is performed by an asynchronous structuring system; and whether the deep structuring condition is met is determined by the following method: performing load detection on the asynchronous structuring system to obtain a load detection result; determining that the deep structuring condition is met when the load detection result indicates that the asynchronous structuring system is in an idle state.

4. The method of claim 1, wherein, The performing semantic feature fusion on the multi-modal consultation data to obtain the fused consultation data of the target patient comprises: performing feature extraction and uniform semantic space mapping on the multi-modal consultation data respectively to obtain multi-modal data features; performing multi-modal attention fusion on the multi-modal data features to obtain fused data features of the multi-modal consultation data; performing semantic text integration on the fused data features according to clinical scene rules to obtain the fused consultation data.

5. The method of claim 4, wherein, The multi-modal consultation data comprises voice text and printed text, examination data and historical medical records of the target patient; and the performing feature extraction and uniform semantic space mapping on the multi-modal consultation data respectively to obtain multi-modal data features comprises: performing time sequence logic analysis on the voice text to extract sequence features of the voice text; and performing spatial layout analysis on the printed text to extract layout features of the printed text; performing data quantitative analysis on the examination data of the target patient to extract numerical features of the examination data; and performing entity extraction coding on the historical medical records of the target patient to extract structured features of the historical medical records; obtaining the multi-modal data features according to the sequence features, the layout features, the numerical features and the structured features.

6. The method of claim 1, wherein, The multi-level clinical correctness verification on the second structured data comprises: performing text form verification on the second structured data to obtain a text form verification result of the second structured data; in a case where the text form verification result indicates that the second structured data meets preset text form requirements, performing semantic logic verification on the second structured data to obtain a semantic logic verification result of the second structured data; in a case where the semantic logic verification result indicates that the second structured data meets preset semantic logic requirements, performing medical logic verification on the second structured data to obtain a medical logic verification result of the second structured data; the data verification result is obtained according to the text form verification result, the semantic logic verification result and the medical logic verification result.

7. The method according to any one of claims 1 to 6, characterized in that, The fusion consultation data is obtained according to a multi-modal data fusion model; the method further comprises: in a case where the data verification result indicates that the second structured data does not meet clinical correctness requirements, performing abnormality warning on the second structured data to remind medical staff to perform abnormality correction on the second structured data; obtaining abnormality correction data of the second structured data, and performing parameter adjustment on the multi-modal data fusion model according to the abnormality correction data to optimize the multi-modal data fusion model.

8. An electronic medical record asynchronous structuring apparatus characterized by comprising: The device comprises: a multi-modal data fusion module, configured to obtain multi-modal consultation data of a target patient, perform semantic feature fusion on the multi-modal consultation data, and obtain fusion consultation data of the target patient; a first data structuring module, configured to perform data structuring on the fusion consultation data at a first granularity to obtain first structured data; a second data structuring module, configured to perform data structuring on the first structured data at a second granularity in a case where a deep structuring condition is met to obtain second structured data; the second granularity is smaller than the first granularity; a clinical correctness verification module, configured to perform multi-level clinical correctness verification on the second structured data to obtain a data verification result of the second structured data; in a case where the data verification result indicates that the second structured data meets clinical correctness requirements, obtaining a structured medical record of the target patient according to the second structured data.

9. A computer device, comprising: comprise: a memory and a processor, which are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the method in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, and the computer instructions are used to make a computer execute the method in any one of claims 1 to 7.

Images