Voice medical record automatic generation method and system for intelligent diagnosis and treatment scene

Abstract

The invention provides a voice medical record automatic generation method and system for an intelligent diagnosis and treatment scene, and relates to the technical field of natural language processing, and the method comprises the steps: executing audio stream collection, and obtaining audio data; carrying out noise filtering on the audio data, and recognizing a transcriptional text by adopting an automatic speech recognition model; a BERT-based NER model is adopted to recognize a fine-grained entity, standard terms are matched through an SNOMED-CT data set, disease classification codes are distributed through an ICD-10 coding system, semantic matching is conducted on the standard terms and the disease classification codes and a diagnosis and treatment entity scheme, and a key entity set is output; the method comprises the following steps: performing data standardization and structured processing on a medical record file to generate a structured key entity set, generating a preliminary medical record file based on a medical record database, and performing examination and storage processing on the preliminary medical record file to generate a final medical record file, thereby realizing automatic generation of the voice medical record oriented to an intelligent diagnosis and treatment scene, and providing a reliable medical record basis for intelligent diagnosis and treatment.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to a method and system for automatically generating voice medical records for intelligent diagnosis and treatment scenarios. Background Technology

[0002] In intelligent medical scenarios, with the continuous growth of patients' medical needs and the increasing standardization requirements of medical services, efficient and accurate completion of medical record recording has become a key link in ensuring the quality of medical care.

[0003] However, traditional medical record generation methods primarily rely on doctors manually recording and organizing patient conversations. This approach is not only inefficient and error-prone, but also struggles to accurately capture medical terminology, symptom descriptions, and diagnostic details when processing real-time audio data, especially in noisy or dialectal environments. While advancements in speech recognition and natural language processing technologies have led to the development of tools for speech-to-text and information extraction, these tools generally lack a complete medical record generation workflow. They fail to effectively integrate audio acquisition, transcription, entity recognition, standardization, automatic generation, quality control, storage, and feedback correction, resulting in inconsistent accuracy, non-standard data formats, and difficulty in seamless integration with medical standards. Furthermore, existing methods often overlook the importance of knowledge graph assistance and multi-round review when dealing with the complexities of clinical dialogues, such as semantic ambiguity and inconsistencies, further limiting the reliability and traceability of medical records.

[0004] Therefore, it is necessary to provide a method and system for automatically generating voice medical records for intelligent diagnosis and treatment scenarios to solve the above-mentioned technical problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and system for automatically generating voice medical records for intelligent medical scenarios, which solves the technical problems of low efficiency, unstable accuracy, and lack of complete process optimization in existing intelligent medical record generation scenarios.

[0006] The present invention provides a method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios, the method comprising: By performing real-time audio stream acquisition on the doctor-patient consultation dialogue, consultation audio data is obtained; The diagnostic audio data is noise filtered, and an automatic speech recognition model trained on medical terminology is used to identify the corresponding diagnostic transcript text of the diagnostic audio data. The BERT-based NER model is used to identify the fine-grained medical entities corresponding to the medical transcription text. Standard medical terms are matched to the fine-grained medical entities using the SNOMED-CT dataset. Disease classification codes are assigned to the fine-grained medical entities using the ICD-10 coding system. The standard medical terms and disease classification codes are semantically matched with the corresponding medical entity schemes to output a set of key medical entities. The set of key diagnostic and treatment entities is subjected to data standardization and data structuring to generate a structured set of key diagnostic and treatment entities. Based on a preset medical record database, preliminary medical record documents corresponding to the set of structured key diagnostic and treatment entities are automatically generated. The preliminary medical record documents are reviewed and stored to generate the final medical record documents.

[0007] Preferably, the step of filtering noise from the diagnostic audio data and using an automatic speech recognition model trained on medical terminology to recognize the corresponding diagnostic transcript text of the diagnostic audio data specifically includes: The noise reduction algorithm based on wavelet transform is used to filter noise in the medical audio data. The signal-to-noise ratio of the audio after the noise reduction process is collected. If the audio signal-to-noise ratio is lower than the preset signal-to-noise ratio threshold, the noise filtering is performed again based on the wavelet transform noise reduction algorithm; otherwise, the automatic speech recognition model is triggered. The automatic speech recognition models include the DeepSpeech model, the Google Speech-to-Text model, and the iFlytek speech recognition model; The DeepSpeech model is pre-trained using a corpus of doctor-patient dialogues to identify common medical terms in the audio data. The Google Speech-to-Text model is trained by integrating a medical-specific language pack to identify specialized medical terms in the diagnostic audio data; The iFlytek speech recognition model is trained with dialect sub-models for Chinese dialects. The dialect sub-models are used to identify dialect medical terms in the medical audio data and map the dialect medical terms to standard Mandarin medical terms through a real-time dialect-Mandarin mapping library. The general medical terminology, the specialized medical terminology, and the standard Mandarin medical terminology are compiled to generate the medical transcription text.

[0008] Preferably, after the key diagnostic entity set is output, the diagnostic dialogue between the doctor and the patient needs to be monitored in real time. If supplementary diagnostic information is identified, the necessity coefficient for updating the entity is calculated. Determine whether the set of key diagnostic entities needs dynamic association update processing, and the necessity coefficient of the entity update. The calculation formula is as follows: In the formula, This represents the candidate medical entity corresponding to the supplementary medical information. This represents the key medical entities in the set of key medical entities; This indicates the calculation of semantic similarity; This indicates the type weight of the key medical entities in the set of key medical entities; If the entity update necessity coefficient If the value exceeds the preset update threshold, it is determined that the set of key diagnostic and treatment entities needs to undergo dynamic association update processing.

[0009] Preferably, the step of performing data standardization and data structuring on the set of key diagnostic and treatment entities to generate a structured set of key diagnostic and treatment entities specifically includes: The data structuring process uses natural language processing algorithms to convert the set of key diagnostic entities into a basic key-value pair structure, and then divides the basic key-value pair structure into a multi-level diagnostic structure according to medical logic, forming a hierarchical key-value pair set. The data standardization process first dynamically selects an appropriate medical data standard based on the current hospital information system type, and then checks the format compliance of the hierarchical key-value pair set based on the appropriate medical data standard through a standard format verification engine, and corrects the format deviation of the hierarchical key-value pair set that does not comply with the format. The adapted medical data standards include health information exchange standards and rapid medical interoperability resource standards.

[0010] Preferably, the step of automatically generating preliminary medical record documents corresponding to the set of structured key medical entities based on a preset medical record database specifically includes: Based on the treatment scenario type corresponding to the structured key treatment entity set, a treatment record template is dynamically matched from the preset treatment record database; The preliminary medical record document is generated by using an automatic medical record generation tool to perform intelligent field filling on the structured key medical entity set based on the medical record template.

[0011] Preferably, the step of using an automatic medical record generation tool to perform intelligent field filling on the structured key medical entity set based on the medical record template specifically includes: The automatic medical record generation tool performs field mapping preprocessing on the structured key diagnosis and treatment entity set, and, in conjunction with the health information exchange standard and the rapid medical interoperability resource standard, clarifies the entity field correspondence between the structured key diagnosis and treatment entities in the structured key diagnosis and treatment entity set and the fields of the diagnosis and treatment medical record template, and constructs a two-way mapping index for entity fields; If the structured key diagnostic and treatment entities in the structured key diagnostic and treatment entity set have a one-to-one correspondence with the fields of the medical record template, then the structured key diagnostic and treatment entities are directly filled into the corresponding fields of the medical record template. If the structured key diagnostic and treatment entities in the structured key diagnostic and treatment entity set have a many-to-one correspondence with the fields of the medical record template, then based on the bidirectional mapping index of the entity fields, a pre-trained semantic matching model for diagnostic and treatment fields is used to calculate the semantic correlation degree between all the structured key diagnostic and treatment entities and the corresponding fields of the medical record template, and the structured key diagnostic and treatment entity with the highest semantic correlation degree is selected to fill the corresponding field of the medical record template.

[0012] Preferably, the preliminary medical record document is reviewed and processed based on a multi-dimensional review engine, specifically including: The multi-dimensional review engine includes a logic rule engine, an AI quality control engine, and an NLP semantic inspection engine. The logic rule engine uses a medical logic rule library to perform logical verification on the set of structured key diagnostic entities in the preliminary medical record document; The AI ​​quality control engine uses a pre-trained medical large model to first extract the medical terms from the structured key medical entity set in the preliminary medical record document. Then, it matches the medical terms with the standard medical terms in the standard medical term library. By calculating the matching confidence of the medical term word vector corresponding to the medical term and the standard medical term word vector corresponding to the standard medical term, the medical terms with a matching confidence of not less than the matching confidence threshold are retained. The NLP semantic inspection engine uses a bidirectional long short-term memory network combined with an attention mechanism to perform multi-dimensional semantic verification on the preliminary medical record document based on the timeline of the diagnosis and treatment dialogue between the doctor and the patient.

[0013] An automatic voice medical record generation system for intelligent diagnosis and treatment scenarios, the system comprising: The audio acquisition module is used to acquire real-time audio streams from the doctor-patient dialogue to obtain diagnostic audio data. The text transcription module is used to filter noise from the medical audio data and to identify the medical transcription text corresponding to the medical audio data using an automatic speech recognition model trained on medical terminology. The entity output module is used to identify fine-grained medical entities corresponding to the medical transcription text using a BERT-based NER model, match standard medical terms to the fine-grained medical entities using the SNOMED-CT dataset, assign disease classification codes to the fine-grained medical entities using the ICD-10 coding system, perform semantic matching between the standard medical terms and disease classification codes and the corresponding medical entity scheme, and output a set of key medical entities. The medical record generation module is used to perform data standardization and data structuring on the set of key diagnostic and treatment entities to generate a structured set of key diagnostic and treatment entities. Based on a preset medical record database, it automatically generates preliminary medical record documents corresponding to the set of structured key diagnostic and treatment entities. The module then reviews and stores the preliminary medical record documents to generate the final medical record documents.

[0014] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor performs the steps of the automatic voice medical record generation method for intelligent diagnosis and treatment scenarios as described above.

[0015] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the automatic voice medical record generation method for intelligent diagnosis and treatment scenarios as described above.

[0016] Compared with related technologies, the automatic voice medical record generation method and system for intelligent diagnosis and treatment scenarios provided by this invention have the following beneficial effects: This invention acquires diagnostic audio data by performing real-time audio stream acquisition of doctor-patient dialogues; it filters noise from the audio data and uses an automatic speech recognition model trained on medical terminology to identify the corresponding transcribed text; it uses a BERT-based NER model to identify fine-grained diagnostic entities corresponding to the transcribed text, matches standard diagnostic terms to these entities using the SNOMED-CT dataset, assigns disease classification codes to them using the ICD-10 coding system, and performs semantic matching between the standard diagnostic terms and disease classification codes and the corresponding diagnostic entity schemes to output a set of key diagnostic entities; it then performs data standardization and structuring processing on the key diagnostic entity set to generate a structured set of key diagnostic entities, automatically generates preliminary diagnostic medical record documents corresponding to the structured set of key diagnostic entities based on a preset diagnostic medical record database, and reviews and stores these preliminary documents to generate the final diagnostic medical record document. This enables the automatic generation of voice medical records for intelligent diagnostic scenarios, providing a reliable medical record foundation for intelligent diagnosis and treatment.

[0017] This invention acquires audio data of doctor-patient dialogues in real time, repeatedly optimizes audio quality using wavelet transform denoising algorithms, and combines this with an automatic speech recognition model trained on medical terminology. This model accurately identifies both common and specialized medical terms, and processes dialectal medical terms through dialect sub-models and a real-time dialect-Mandarin mapping library, effectively ensuring the integrity and accuracy of the transcribed medical text and adapting to diverse medical communication scenarios. Based on a BERT-based NER model to identify fine-grained medical entities, this invention matches standard medical terms using the SNOMED-CT dataset and assigns disease classification codes using the ICD-10 coding system, achieving standardization of medical entities. By monitoring medical dialogues in real time, this invention dynamically identifies and updates the set of key medical entities, avoiding information omissions. The structured set of key medical entities, formed through data standardization and structuring, supports cross-system interoperability of medical data. This invention generates preliminary medical record documents by dynamically matching medical record templates based on a preset medical record database and combining intelligent field filling. The documents are then reviewed by a multi-dimensional review engine to ensure compliance with the logic, terminology, and semantics of the medical records. Attached Figure Description

[0018] Figure 1 A flowchart of a method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios provided in an embodiment of the present invention; Figure 2 This is a system block diagram of an automatic voice medical record generation system for intelligent diagnosis and treatment scenarios provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] like Figure 1 The diagram shown is a flowchart of a method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios provided by an embodiment of the present invention. Figure 1The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S4 are detailed as follows: S1 acquires diagnostic audio data by performing real-time audio stream acquisition on the doctor-patient dialogue. The real-time audio stream acquisition covers the entire dialogue and interaction between doctors and patients throughout the diagnosis and treatment process, including the chief complaint, symptom description, medical history inquiry, physical examination feedback, diagnostic explanation, and treatment suggestions. This is achieved through continuous and uninterrupted audio signal capture using audio acquisition devices deployed in the diagnosis and treatment setting. The diagnostic and treatment audio data is the raw audio signal carrying all the information from the diagnostic and treatment dialogue, obtained through real-time audio stream acquisition.

[0021] S2, Noise filtering is performed on the diagnostic audio data, and an automatic speech recognition model trained on medical terminology is used to identify the diagnostic transcript text corresponding to the diagnostic audio data; The step of filtering noise from the medical audio data and using an automatic speech recognition model trained on medical terminology to identify the corresponding medical transcript text from the medical audio data specifically includes: The noise reduction algorithm based on wavelet transform is used to filter noise in the medical audio data. The signal-to-noise ratio of the audio after the noise reduction process is collected. If the audio signal-to-noise ratio is lower than the preset signal-to-noise ratio threshold, the noise filtering is performed again based on the wavelet transform noise reduction algorithm; otherwise, the automatic speech recognition model is triggered. The automatic speech recognition models include the DeepSpeech model, the Google Speech-to-Text model, and the iFlytek speech recognition model; The DeepSpeech model is pre-trained using a corpus of doctor-patient dialogues to identify common medical terms in the audio data. The Google Speech-to-Text model is trained by integrating a medical-specific language pack to identify specialized medical terms in the diagnostic audio data; The iFlytek speech recognition model is trained with dialect sub-models for Chinese dialects. The dialect sub-models are used to identify dialect medical terms in the medical audio data and map the dialect medical terms to standard Mandarin medical terms through a real-time dialect-Mandarin mapping library. The general medical terminology, the specialized medical terminology, and the standard Mandarin medical terminology are compiled to generate the medical transcription text.

[0022] Wavelet transform denoising algorithms are signal processing techniques used to handle noise interference in medical audio data. By performing multi-scale decomposition and reconstruction of the audio signal, they separate and remove background noise from the medical environment, such as ambient noise in the examination room and equipment operating noise, while retaining the effective medical speech signal. The audio signal-to-noise ratio (SNR) is a core indicator for measuring the quality of denoised medical audio, specifically the ratio of the intensity of the effective medical speech signal to the intensity of the remaining noise signal. A preset SNR threshold is a pre-defined audio quality judgment standard based on the accuracy requirements of medical speech recognition. When the SNR of the denoised audio reaches or exceeds this threshold, it indicates that the audio quality meets the requirements for subsequent speech recognition, triggering an automatic speech recognition model; otherwise, denoising processing needs to be performed again.

[0023] Automatic speech recognition models include the DeepSpeech model, the Google Speech-to-Text model, and the iFlytek speech recognition model. The DeepSpeech model uses real-life doctor-patient dialogues as training data, focusing on recognizing common medical terms used in clinical settings, such as basic symptom descriptions like "headache" and "fatigue." The Google Speech-to-Text model is trained by integrating medical-specific language packs, focusing on recognizing specialized medical terms used in the diagnosis and treatment process. The iFlytek speech recognition model trains dialect sub-models to address the characteristics of different regional Chinese dialects, accurately recognizing dialect medical terms in audio. It also uses a real-time dialect-to-Mandarin mapping library, a dedicated database storing the correspondence between dialect expressions and standard Mandarin medical terms, to convert dialect terms into standardized Mandarin medical terms.

[0024] General medical terminology refers to common and basic medical expressions used in clinical settings. Specialized medical terminology refers to professional medical terms applicable only to specific medical specialties or fields, such as "myocardial infarction" and "glucocorticoids." Dialectal medical terminology refers to medical terms used when patients describe their symptoms using Chinese dialects. Medical transcripts are textual data that summarizes these three types of terms according to the temporal and semantic logic of the medical dialogue, fully preserving the key information from the dialogue.

[0025] In practical applications, the noise reduction parameters of wavelet transform are dynamically adjusted according to the actual noise characteristics of the diagnosis and treatment scenario. For example, in a clinic environment with high equipment operating noise, the noise reduction effect is improved by enhancing the suppression of high-frequency signals, ensuring that the audio signal-to-noise ratio after multiple iterations consistently meets the preset signal-to-noise ratio threshold. A multi-model collaborative recognition mechanism automatically switches or combines models based on the real-time audio content. When basic expressions such as "cough" and "fever" are recognized, the DeepSpeech model trained on diagnostic dialogue corpus is prioritized to improve the recognition efficiency of common diagnostic and treatment terms. When specialized terms such as "coronary atherosclerosis" and "paclitaxel" are captured, the Google Speech-to-Text model, which integrates medical-specific language packages, is automatically triggered to ensure the recognition accuracy of specialized diagnostic and treatment terms. If dialect speech features are detected, the corresponding dialect sub-model of the iFlytek speech recognition model is immediately invoked, and terminology conversion is completed using a real-time dialect-Mandarin mapping library, avoiding information distortion caused by dialect differences.

[0026] S3, using a BERT-based NER model to identify the fine-grained medical entities corresponding to the medical transcription text, matching standard medical terms to the fine-grained medical entities using the SNOMED-CT dataset, assigning disease classification codes to the fine-grained medical entities using the ICD-10 coding system, semantically matching the standard medical terms and disease classification codes with the corresponding medical entity schemes, and outputting a set of key medical entities; Among them, the BERT-based NER model is a named entity recognition model optimized for the characteristics of medical texts. It enhances its ability to recognize entities in the medical field through pre-training with specialized medical corpora. Its core advantage lies in achieving fine-grained recognition; that is, it can not only locate entities in transcribed medical texts but also capture their attribute information. For example, from the sentence "The patient has experienced dull pain in the left upper abdomen half an hour after meals for the past week, occurring 2-3 times daily," it can accurately identify fine-grained medical entities with attribute descriptions such as "dull pain, location: left upper abdomen, onset time: half an hour after meals, frequency: 2-3 times daily, duration: for the past week," rather than simply recognizing broad expressions like "abdominal pain." Fine-grained medical entities refer to core information units with specific attributes or contextual relevance in the medical process, covering four core types: symptoms and signs, disease diagnosis, examination and test results, and treatment plans.

[0027] The SNOMED-CT dataset is a standardized dataset for clinical terminology, containing standardized terms and logical relationships between terms covering diseases, symptoms, drugs, procedures, and other areas. When used for fine-grained diagnostic entity matching, it establishes semantic associations between entities and standard terms, such as uniformly matching "myocardial infarction" and "myocardial infarction" as "myocardial infarction," eliminating inconsistencies in terminology expression in diagnostic dialogues and ensuring that the output diagnostic terms conform to generally accepted clinical standards.

[0028] The ICD-10 coding system is used for the classification and coding of diseases and related health issues. When assigning codes to fine-grained medical entities, for disease-type entities within an entity, a unique disease classification code is matched according to the classification rules of the coding system. This coding enables the structured identification of disease information.

[0029] The diagnostic and treatment entity plan is a set of entity screening and prioritization rules based on different diagnostic and treatment scenarios. For example, in an outpatient scenario, the plan prioritizes entities such as "chief complaint symptoms, present medical history, preliminary diagnosis, and prescribed medications". Semantic matching verifies whether the standard terms matched by SNOMED-CT and the disease classification codes assigned by ICD-10 are consistent with the entity plan requirements corresponding to the current diagnostic and treatment scenario, eliminating redundant and logically contradictory entities and retaining the core information that meets the plan requirements.

[0030] The set of key diagnostic and treatment entities is a structured information set formed after the above steps. The key diagnostic and treatment entities contained therein all meet the three conditions of standardized terminology, accurate coding, and compliance with scenario requirements. Specifically, it covers information such as the patient's core symptoms, clear diagnostic results, key examination indicators, and standardized treatment plans.

[0031] Through the above methods, fine-grained diagnostic entities in diagnostic transcripts are accurately identified, terminology is standardized using SNOMED-CT, disease coding is standardized using ICD-10, and key diagnostic entities that meet the needs of the scenario are selected through semantic matching, ensuring the accuracy, standardization, and applicability of the set of key diagnostic entities.

[0032] After the set of key diagnostic entities is output, the doctor-patient dialogue needs to be monitored in real time. If supplementary diagnostic information is identified, the necessity coefficient for entity updates should be calculated. Determine whether the set of key diagnostic entities needs dynamic association update processing, and the necessity coefficient of the entity update. The calculation formula is as follows: In the formula, This represents the candidate medical entity corresponding to the supplementary medical information. This represents the key medical entities in the set of key medical entities; This indicates the calculation of semantic similarity; This indicates the type weight of the key medical entities in the set of key medical entities; If the entity update necessity coefficient If the value exceeds the preset update threshold, it is determined that the set of key diagnostic and treatment entities needs to undergo dynamic association update processing.

[0033] The supplementary medical information refers to medical information captured during the monitoring process that is not included in the initial set of key medical entities, such as "past history of gastric ulcer disease" or "skin rash reaction after medication" mentioned by the patient later. Candidate medical entities are entities extracted from the supplementary medical information using a BERT-based NER model consistent with the identification of the initial fine-grained medical entities; they are potential entities to be determined whether to include in the set of key medical entities.

[0034] Semantic similarity refers to quantifying the degree of semantic association between candidate diagnostic entities and existing entities in the set of key diagnostic entities using a pre-trained medical semantic model. The type weights of key diagnostic entities are preset weight values ​​based on the degree of influence of the entity on diagnostic decisions and the integrity of medical records. For example, entities of diagnosis and core symptoms have higher weights than entities of auxiliary examinations and general signs. This is intended to prioritize entities with a greater impact on the core information of medical records when calculating the necessity of updates.

[0035] The entity update necessity coefficient is a quantitative indicator calculated by combining semantic similarity and the type weight of key diagnostic entities. Its value directly reflects the supplementary value of candidate diagnostic entities to the set of key diagnostic entities. The preset update threshold is a judgment criterion set according to the information needs of different diagnostic scenarios. For example, the preset update threshold for outpatient scenarios is relatively moderate, taking into account both information integrity and set stability.

[0036] Dynamic association update processing refers to the update operation performed on the set of key diagnostic and treatment entities when the necessity coefficient of an entity update exceeds a preset update threshold. This operation does not simply add new candidate entities, but rather establishes associations between them and existing entities, or adds entirely new core entities with no associations to the set. At the same time, it verifies the logical consistency of entities within the updated set to ensure the accuracy and standardization of the updated set of key diagnostic and treatment entities.

[0037] By employing the above method, supplementary information from the diagnostic dialogue is continuously captured after the key diagnostic entity set is output. The necessity of updates is calculated by combining semantic similarity and the weight of key diagnostic entity types, accurately determining whether dynamic updates are needed. This avoids redundant updates that could interfere with diagnostic efficiency while ensuring that the key diagnostic entity set dynamically adapts to the diagnostic process, continuously guaranteeing its integrity and timeliness.

[0038] S4. Perform data standardization and data structuring on the set of key diagnostic and treatment entities to generate a structured set of key diagnostic and treatment entities. Based on a preset medical record database, automatically generate preliminary medical record documents corresponding to the set of structured key diagnostic and treatment entities. Review and store the preliminary medical record documents to generate final medical record documents.

[0039] The step of performing data standardization and structuring processing on the set of key diagnostic and treatment entities to generate a structured set of key diagnostic and treatment entities specifically includes: The data structuring process uses natural language processing algorithms to convert the set of key diagnostic entities into a basic key-value pair structure, and then divides the basic key-value pair structure into a multi-level diagnostic structure according to medical logic, forming a hierarchical key-value pair set. The data standardization process first dynamically selects an appropriate medical data standard based on the current hospital information system type, and then checks the format compliance of the hierarchical key-value pair set based on the appropriate medical data standard through a standard format verification engine, and corrects the format deviation of the hierarchical key-value pair set that does not comply with the format. The adapted medical data standards include health information exchange standards and rapid medical interoperability resource standards.

[0040] Understandably, data structuring is used to transform unstructured sets of key diagnostic and treatment entities into organized data directly applicable to medical record generation. It achieves the orderly organization of key diagnostic and treatment entity information through natural language processing algorithms. Medical logic specifically refers to the information organization logic that conforms to clinical diagnosis and treatment processes and medical record writing standards. For example, it follows a progressive logic of diagnostic and treatment information: "Patient basic information - Chief complaint - Present illness - Physical examination - Auxiliary examinations - Diagnosis - Treatment plan." Based on this, the basic key-value pair structure is divided into multi-level diagnostic and treatment structures. For instance, under the "Present illness" level, further subdivisions are made into second-level sub-levels such as "Symptom onset time," "Symptom triggering factors," and "Symptom relief methods," allowing the basic key-value pairs to be aligned according to the diagnostic and treatment logic, ultimately forming a hierarchical set of key-value pairs that clearly presents the hierarchical relationships of diagnostic and treatment information.

[0041] Data standardization eliminates data format differences in key diagnostic and treatment entity sets, ensuring compatibility with existing hospital information systems. Its core lies in dynamically selecting appropriate medical data standards. Since different hospital information systems may use different data specifications, the compatibility standards of the current hospital information system are first identified, and then the appropriate medical data standards are determined. Specifically, the health information exchange standard focuses on the format specifications for cross-system transmission of medical data, clarifying naming rules, data types, and encoding formats for data fields to ensure data identifiability across different diagnostic and treatment systems. The rapid medical interoperability resource standard emphasizes real-time data interoperability, standardizing data access interfaces and sharing protocols to support efficient data access and integration across different medical platforms.

[0042] The standard format validation engine is the core tool for performing data standardization. Based on the selected applicable medical data standard, it performs item-by-item checks on the format compliance of the hierarchical key-value pair set. The format compliance check determines whether the field naming of the hierarchical key-value pairs conforms to standard regulations, whether the data content conforms to standard terminology specifications, and whether the field completeness meets standard requirements. If a format deviation is found, it is adjusted through a format deviation correction operation, for example, correcting "symptom name-headache" to "symptom-headache," ultimately ensuring that the hierarchical key-value pair set fully complies with the applicable medical data standard.

[0043] The above method transforms the set of key diagnostic and treatment entities into a hierarchical set of key-value pairs that conform to medical logic. At the same time, it adapts to the health information exchange standard or rapid medical interoperability resource standard according to the type of hospital information system. Data deviations are eliminated through format verification and correction, so that the data has both structured organization and standardized specifications, which meets the needs of subsequent medical record generation and hospital system storage and interaction.

[0044] The automatic generation of preliminary medical record documents corresponding to the structured key medical entity set based on a preset medical record database specifically includes: Based on the treatment scenario type corresponding to the structured key treatment entity set, a treatment record template is dynamically matched from the preset treatment record database; The preliminary medical record document is generated by using an automatic medical record generation tool to perform intelligent field filling on the structured key medical entity set based on the medical record template.

[0045] Among them, the diagnosis and treatment scenario type refers to the specific diagnosis and treatment scenario category determined based on the core information in the structured key diagnosis and treatment entity set, such as internal medicine outpatient scenario, surgical emergency scenario, pediatric specialty diagnosis and treatment scenario, etc. The judgment logic combines the "disease type" and "symptom urgency" fields in the key diagnosis and treatment entity. For example, when the key diagnosis and treatment entity contains emergency symptoms such as "acute abdominal pain" or "traumatic bleeding", it is judged as an emergency scenario.

[0046] The pre-built medical record database is a database that stores medical record templates corresponding to various medical scenarios. The medical record templates contained therein conform to clinical medical record writing standards, and the field settings cover the core medical information in the corresponding scenarios. In addition, the pre-built medical record database supports updating and supplementing medical record templates according to the medical record writing habits of different hospitals and local medical standards, ensuring the practicality and adaptability of medical record templates.

[0047] Dynamic matching of medical record templates refers to selecting medical record templates that are fully compatible with the identified medical scenario from a pre-set medical record database, rather than using generic medical record templates. The matching process verifies the compatibility of the fields in the medical record template with the fields in the structured key medical entity set, ensuring that the medical record template can fully contain the structured key medical entity information.

[0048] The automatic medical record generation tool needs to be adapted to the medical scenario, integrating a field mapping preprocessing module, a semantic matching model for diagnostic and treatment fields, and a standard format adaptation component. The preliminary medical record document is an initial version of the medical record formed after intelligent field filling. It fully covers all the preset fields of the medical record template. The field content comes from a structured set of key diagnostic and treatment entities and conforms to the format requirements of the health information exchange standard and the rapid medical interoperability resource standard. It carries key information of the entire diagnosis and treatment process.

[0049] The step of using an automatic medical record generation tool to perform intelligent field filling on the structured key medical entity set based on the medical record template specifically includes: The automatic medical record generation tool performs field mapping preprocessing on the structured key diagnosis and treatment entity set, and, in conjunction with the health information exchange standard and the rapid medical interoperability resource standard, clarifies the entity field correspondence between the structured key diagnosis and treatment entities in the structured key diagnosis and treatment entity set and the fields of the diagnosis and treatment medical record template, and constructs a two-way mapping index for entity fields; If the structured key diagnostic and treatment entities in the structured key diagnostic and treatment entity set have a one-to-one correspondence with the fields of the medical record template, then the structured key diagnostic and treatment entities are directly filled into the corresponding fields of the medical record template. If the structured key diagnostic and treatment entities in the structured key diagnostic and treatment entity set have a many-to-one correspondence with the fields of the medical record template, then based on the bidirectional mapping index of the entity fields, a pre-trained semantic matching model for diagnostic and treatment fields is used to calculate the semantic correlation degree between all the structured key diagnostic and treatment entities and the corresponding fields of the medical record template, and the structured key diagnostic and treatment entity with the highest semantic correlation degree is selected to fill the corresponding field of the medical record template.

[0050] The field mapping preprocessing step combines health information exchange standards and rapid medical interoperability resource standards to clarify and confirm the relationship between structured key diagnostic and treatment entities and fields in medical record templates, ensuring that the corresponding logic conforms to medical data specifications. The entity-field correspondence refers to the matching association between structured key diagnostic and treatment entities and fields in medical record templates, specifically divided into one-to-one and many-to-one relationships. This relationship is the core basis for determining the field filling method and directly affects the accuracy and rationality of the filling results.

[0051] The entity field bidirectional mapping index is an index system built on the confirmed correspondence of entity fields. It has bidirectional query capabilities, which can quickly locate the fields of the corresponding medical record template through the structured key medical entity, and can also query the related structured key medical entity in reverse through the fields of the medical record template.

[0052] A one-to-one entity field correspondence means that a single structured key medical entity has a unique match with only one field in the medical record template. Under this type of correspondence, no additional filtering is required, and the accuracy of the field content can be guaranteed by directly filling in the field. For example, the patient's age in the structured key medical entity corresponds only to the age field in the medical record template.

[0053] A many-to-one entity field correspondence refers to a situation where multiple structured key medical entities are matched with a single field in the medical record template. In this case, further filtering is required through semantic matching to avoid redundancy or deviation in field content. For example, the main symptom field in the medical record template may correspond to multiple entities such as cough, sore throat, and fever in the structured key medical entities.

[0054] The pre-trained semantic matching model for medical fields is a semantic computing model optimized for medical scenarios. Pre-trained with a large corpus of medical texts, it possesses the ability to accurately identify the semantic associations between medical terms and fields in medical record templates, effectively quantifying the degree of fit between structured key medical entities and fields in medical record templates. Semantic relevance is a quantitative result calculated by this model, reflecting the degree of semantic fit between structured key medical entities and corresponding fields in medical record templates. A higher relevance indicates that the entity better meets the medical information requirements carried by that field, serving as the core criterion for selecting and filling content in a many-to-one correspondence relationship.

[0055] By relying on field mapping preprocessing and bidirectional mapping indexing, and combining health information exchange standards and rapid medical interoperability resource standards, a correspondence between structured key diagnostic and treatment entities and fields in medical record templates is established. Adaptive filling methods are adopted for different correspondences, and semantic association degree filtering is used to ensure the filling accuracy in many-to-one scenarios, ensuring that intelligent field filling conforms to the specifications and fits the requirements of medical record templates.

[0056] The preliminary medical record documents are reviewed and processed using a multi-dimensional review engine, specifically including: The multi-dimensional review engine includes a logic rule engine, an AI quality control engine, and an NLP semantic inspection engine. The logic rule engine uses a medical logic rule library to perform logical verification on the set of structured key diagnostic entities in the preliminary medical record document; The AI ​​quality control engine uses a pre-trained medical large model to first extract the medical terms from the structured key medical entity set in the preliminary medical record document. Then, it matches the medical terms with the standard medical terms in the standard medical term library. By calculating the matching confidence of the medical term word vector corresponding to the medical term and the standard medical term word vector corresponding to the standard medical term, the medical terms with a matching confidence of not less than the matching confidence threshold are retained. The NLP semantic inspection engine uses a bidirectional long short-term memory network combined with an attention mechanism to perform multi-dimensional semantic verification on the preliminary medical record document based on the timeline of the diagnosis and treatment dialogue between the doctor and the patient.

[0057] Among them, the multi-dimensional review engine is a multi-dimensional and three-dimensional review engine built for preliminary medical records. By integrating the synergistic effects of the logic rule engine, AI quality control engine and NLP semantic inspection engine, it conducts reviews from three core dimensions: logical consistency, terminology standardization and semantic coherence, comprehensively covering the key requirements of medical record review and ensuring the compliance and accuracy of preliminary medical records.

[0058] The logic rule engine is a review component focused on the logical consistency of medical record information. It relies on a medical logic rule library, a set of rules built upon clinical practice guidelines, medical common sense, and medical record writing standards. This library covers rules relating diagnoses and symptoms, matching treatment plans to diagnoses, and supporting rules for examination results and diagnoses. During logical verification, the logic rule engine calls upon rules from the medical logic rule library to verify the logical relationships between structured key diagnostic entities within the preliminary medical record document, identifying and marking logical contradictions, missing logical elements, and other issues.

[0059] The AI ​​quality control engine is a review component that ensures the standardization of medical record terminology. Its pre-trained medical model, trained on a large-scale medical text corpus, possesses the ability to accurately extract and process medical terminology. The standard medical terminology database is a collection of standardized medical terms conforming to the SNOMED-CT dataset and ICD-10 coding system, including core medical terms such as disease names, symptoms and signs, drug names, and examination items. The medical terminology matching confidence score is a quantitative indicator derived by calculating the similarity between the word vectors of medical terms extracted from preliminary medical record documents and the word vectors of standard medical terms, reflecting the degree of semantic matching between the two. The matching confidence score threshold is a preset judgment standard based on the accuracy requirements of clinical terminology usage. It is used to filter medical terms, retaining those with a confidence score not lower than the threshold and eliminating non-standard medical terms with insufficient confidence, ensuring that medical record terminology conforms to industry standards.

[0060] The NLP semantic inspection engine is a review component that ensures the semantic coherence and information integrity of medical records. It employs a bidirectional long short-term memory network combined with an attention mechanism to effectively capture the semantic relationships within the text context. Simultaneously, the attention mechanism focuses on key information in the medical record, improving the accuracy of semantic analysis. The consultation dialogue timeline is a time sequence generated based on real-time records of the doctor-patient consultation, marking the order in which each piece of information occurs. During multi-dimensional semantic verification, the NLP semantic inspection engine, combined with the consultation dialogue timeline, verifies the semantic coherence, temporal logic, and information integrity of the information within the initial medical record document, ensuring a high degree of consistency between the medical record content and the actual treatment process, without semantic gaps or information deviations.

[0061] In practical applications, the review focus of the multi-dimensional review engine is dynamically adjusted according to the characteristics of hospital departments. For internal medicine scenarios, the logic rule engine strengthens the verification of the correlation between symptoms and diagnoses. For example, the diagnosis of diabetes requires matching typical symptoms such as polydipsia and polyuria, while the AI ​​quality control engine prioritizes the verification of endocrinology-specific diagnostic and treatment terminology. For surgical scenarios, the logic rule engine focuses on verifying the suitability of the surgical plan with the preoperative diagnosis, while the NLP semantic inspection engine focuses on the semantic coherence of the surgical site and surgical procedure descriptions.

[0062] Meanwhile, the AI ​​quality control engine flexibly sets the matching confidence threshold based on hospital quality control standards. For example, top-tier hospitals can raise the matching confidence threshold to strictly standardize terminology usage. The NLP semantic inspection engine combines the timeline of the diagnosis and treatment dialogue to verify the temporal logic of information in the preliminary medical record documents, such as whether the order of symptom occurrence is consistent with the description of the present medical history, avoiding semantic deviations caused by the order of field filling. The three work together to form a comprehensive review loop.

[0063] The preliminary medical records are reviewed and processed using a multi-dimensional review engine. The review results are then integrated to generate a medical quality feedback report, which is sent to the doctor for review. Once the doctor approves the report, the preliminary medical records are stored in a pre-defined medical record database.

[0064] In addition, a user interface allows doctors to edit preliminary medical records and patients to submit feedback and corrections. A version control mechanism is used to record changes to the preliminary medical records and generate the final medical records.

[0065] like Figure 2 The diagram shown is a system block diagram of an automatic voice medical record generation system for intelligent diagnosis and treatment scenarios provided by an embodiment of the present invention. The system includes: The audio acquisition module is used to acquire real-time audio streams from the doctor-patient dialogue to obtain diagnostic audio data. The text transcription module is used to filter noise from the medical audio data and to identify the medical transcription text corresponding to the medical audio data using an automatic speech recognition model trained on medical terminology. The entity output module is used to identify fine-grained medical entities corresponding to the medical transcription text using a BERT-based NER model, match standard medical terms to the fine-grained medical entities using the SNOMED-CT dataset, assign disease classification codes to the fine-grained medical entities using the ICD-10 coding system, perform semantic matching between the standard medical terms and disease classification codes and the corresponding medical entity scheme, and output a set of key medical entities. The medical record generation module is used to perform data standardization and data structuring on the set of key diagnostic and treatment entities to generate a structured set of key diagnostic and treatment entities. Based on a preset medical record database, it automatically generates preliminary medical record documents corresponding to the set of structured key diagnostic and treatment entities. The module then reviews and stores the preliminary medical record documents to generate the final medical record documents.

[0066] Figure 2 The apparatus of the illustrated embodiment can be used to perform corresponding actions. Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0067] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor performs the steps of the automatic voice medical record generation method for intelligent diagnosis and treatment scenarios as described above.

[0068] like Figure 3 The diagram shown is a hardware structure schematic of an electronic device according to an embodiment of the present invention. The electronic device 30 includes: a processor 31, a memory 32, and a computer program; wherein... The memory 32 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0069] Processor 31 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0070] Alternatively, the memory 32 can be either standalone or integrated with the processor 31.

[0071] When the memory 32 is a device independent of the processor 31, the device may further include: Bus 33 is used to connect the memory 32 and the processor 31.

[0072] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the automatic voice medical record generation method for intelligent diagnosis and treatment scenarios as described above.

[0073] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can be located in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0074] The present invention also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the device to implement the methods provided in the various embodiments described above.

[0075] In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0076] Through the above embodiments, this invention acquires diagnostic audio data by performing real-time audio stream acquisition of the doctor-patient dialogue; filters noise from the diagnostic audio data; uses an automatic speech recognition model trained on medical terminology to identify the corresponding diagnostic transcript text; uses a BERT-based NER model to identify fine-grained diagnostic entities corresponding to the diagnostic transcript text; matches standard diagnostic terms to the fine-grained diagnostic entities using the SNOMED-CT dataset; assigns disease classification codes to the fine-grained diagnostic entities using the ICD-10 coding system; performs semantic matching between the standard diagnostic terms and disease classification codes and the corresponding diagnostic entity schemes; outputs a set of key diagnostic entities; performs data standardization and structuring processing on the set of key diagnostic entities to generate a structured set of key diagnostic entities; automatically generates preliminary diagnostic medical record documents corresponding to the structured set of key diagnostic entities based on a preset diagnostic medical record database; and reviews and stores the preliminary diagnostic medical record documents to generate the final diagnostic medical record document. This enables the automatic generation of voice medical records for intelligent diagnostic scenarios, providing a reliable medical record foundation for intelligent diagnostics.

[0077] This invention acquires audio data of doctor-patient dialogues in real time, repeatedly optimizes audio quality using wavelet transform denoising algorithms, and combines this with an automatic speech recognition model trained on medical terminology. This model accurately identifies both common and specialized medical terms, and processes dialectal medical terms through dialect sub-models and a real-time dialect-Mandarin mapping library, effectively ensuring the integrity and accuracy of the transcribed medical text and adapting to diverse medical communication scenarios. Based on a BERT-based NER model to identify fine-grained medical entities, this invention matches standard medical terms using the SNOMED-CT dataset and assigns disease classification codes using the ICD-10 coding system, achieving standardization of medical entities. By monitoring medical dialogues in real time, this invention dynamically identifies and updates the set of key medical entities, avoiding information omissions. The structured set of key medical entities, formed through data standardization and structuring, supports cross-system interoperability of medical data. This invention generates preliminary medical record documents by dynamically matching medical record templates based on a preset medical record database and combining intelligent field filling. The documents are then reviewed by a multi-dimensional review engine to ensure compliance with the logic, terminology, and semantics of the medical records.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios, characterized in that, The method includes: By performing real-time audio stream acquisition on the doctor-patient consultation dialogue, consultation audio data is obtained; The diagnostic audio data is noise filtered, and an automatic speech recognition model trained on medical terminology is used to identify the corresponding diagnostic transcript text of the diagnostic audio data. The BERT-based NER model is used to identify the fine-grained medical entities corresponding to the medical transcription text. Standard medical terms are matched to the fine-grained medical entities using the SNOMED-CT dataset. Disease classification codes are assigned to the fine-grained medical entities using the ICD-10 coding system. The standard medical terms and disease classification codes are semantically matched with the corresponding medical entity schemes to output a set of key medical entities. The set of key diagnostic and treatment entities is subjected to data standardization and data structuring to generate a structured set of key diagnostic and treatment entities. Based on a preset medical record database, preliminary medical record documents corresponding to the set of structured key diagnostic and treatment entities are automatically generated. The preliminary medical record documents are reviewed and stored to generate the final medical record documents.

2. The method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios according to claim 1, characterized in that, The step of filtering noise from the medical audio data and using an automatic speech recognition model trained on medical terminology to identify the corresponding medical transcript text from the medical audio data specifically includes: The noise reduction algorithm based on wavelet transform is used to filter noise in the medical audio data. The signal-to-noise ratio of the audio after the noise reduction process is collected. If the audio signal-to-noise ratio is lower than the preset signal-to-noise ratio threshold, the noise filtering is performed again based on the wavelet transform noise reduction algorithm; otherwise, the automatic speech recognition model is triggered. The automatic speech recognition models include the DeepSpeech model, the Google Speech-to-Text model, and the iFlytek speech recognition model; The DeepSpeech model is pre-trained using a corpus of doctor-patient dialogues to identify common medical terms in the audio data. The Google Speech-to-Text model is trained by integrating a medical-specific language pack to identify specialized medical terms in the diagnostic audio data; The iFlytek speech recognition model is trained with dialect sub-models for Chinese dialects. The dialect sub-models are used to identify dialect medical terms in the medical audio data and map the dialect medical terms to standard Mandarin medical terms through a real-time dialect-Mandarin mapping library. The general medical terminology, the specialized medical terminology, and the standard Mandarin medical terminology are compiled to generate the medical transcription text.

3. The method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios according to claim 1, characterized in that, After the set of key diagnostic entities is output, the doctor-patient dialogue needs to be monitored in real time. If supplementary diagnostic information is identified, the necessity coefficient for entity updates should be calculated. Determine whether the set of key diagnostic entities needs dynamic association update processing, and the necessity coefficient of the entity update. The calculation formula is as follows: In the formula, This represents the candidate medical entity corresponding to the supplementary medical information. This represents the key medical entities in the set of key medical entities; This indicates the calculation of semantic similarity; This indicates the type weight of the key medical entities in the set of key medical entities; If the entity update necessity coefficient If the value exceeds the preset update threshold, it is determined that the set of key diagnostic and treatment entities needs to undergo dynamic association update processing.

4. The method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios according to claim 1, characterized in that, The step of performing data standardization and structuring processing on the set of key diagnostic and treatment entities to generate a structured set of key diagnostic and treatment entities specifically includes: The data structuring process uses natural language processing algorithms to convert the set of key diagnostic entities into a basic key-value pair structure, and then divides the basic key-value pair structure into a multi-level diagnostic structure according to medical logic, forming a hierarchical key-value pair set. The data standardization process first dynamically selects an appropriate medical data standard based on the current hospital information system type, and then checks the format compliance of the hierarchical key-value pair set based on the appropriate medical data standard through a standard format verification engine, and corrects the format deviation of the hierarchical key-value pair set that does not comply with the format. The adapted medical data standards include health information exchange standards and rapid medical interoperability resource standards.

5. The method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios according to claim 1, characterized in that, The automatic generation of preliminary medical record documents corresponding to the structured key medical entity set based on a preset medical record database specifically includes: Based on the treatment scenario type corresponding to the structured key treatment entity set, a treatment record template is dynamically matched from the preset treatment record database; The preliminary medical record document is generated by using an automatic medical record generation tool to perform intelligent field filling on the structured key medical entity set based on the medical record template.

6. The method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios according to claim 5, characterized in that, The step of using an automatic medical record generation tool to perform intelligent field filling on the structured key medical entity set based on the medical record template specifically includes: The automatic medical record generation tool performs field mapping preprocessing on the structured key diagnosis and treatment entity set, and, in conjunction with the health information exchange standard and the rapid medical interoperability resource standard, clarifies the entity field correspondence between the structured key diagnosis and treatment entities in the structured key diagnosis and treatment entity set and the fields of the diagnosis and treatment medical record template, and constructs a two-way mapping index for entity fields; If the structured key diagnostic and treatment entities in the structured key diagnostic and treatment entity set have a one-to-one correspondence with the fields of the medical record template, then the structured key diagnostic and treatment entities are directly filled into the corresponding fields of the medical record template. If the structured key diagnostic and treatment entities in the structured key diagnostic and treatment entity set have a many-to-one correspondence with the fields of the medical record template, then based on the bidirectional mapping index of the entity fields, a pre-trained semantic matching model for diagnostic and treatment fields is used to calculate the semantic correlation degree between all the structured key diagnostic and treatment entities and the corresponding fields of the medical record template, and the structured key diagnostic and treatment entity with the highest semantic correlation degree is selected to fill the corresponding field of the medical record template.

7. The method for automatically generating voice medical records for intelligent diagnosis and treatment scenarios according to claim 1, characterized in that, The preliminary medical record documents are reviewed and processed using a multi-dimensional review engine. include: The multi-dimensional review engine includes a logic rule engine, an AI quality control engine, and an NLP semantic inspection engine. The logic rule engine uses a medical logic rule library to perform logical verification on the set of structured key diagnostic entities in the preliminary medical record document; The AI ​​quality control engine uses a pre-trained medical large model to first extract the medical terms from the structured key medical entity set in the preliminary medical record document. Then, it matches the medical terms with the standard medical terms in the standard medical term library. By calculating the matching confidence of the medical term word vector corresponding to the medical term and the standard medical term word vector corresponding to the standard medical term, the medical terms with a matching confidence of not less than the matching confidence threshold are retained. The NLP semantic inspection engine uses a bidirectional long short-term memory network combined with an attention mechanism to perform multi-dimensional semantic verification on the preliminary medical record document based on the timeline of the diagnosis and treatment dialogue between the doctor and the patient.

8. An automatic voice medical record generation system for intelligent diagnosis and treatment scenarios, applied to the automatic voice medical record generation method for intelligent diagnosis and treatment scenarios as described in any one of claims 1-7, characterized in that, The system includes: The audio acquisition module is used to acquire real-time audio streams from the doctor-patient dialogue to obtain diagnostic audio data. The text transcription module is used to filter noise from the medical audio data and to identify the medical transcription text corresponding to the medical audio data using an automatic speech recognition model trained on medical terminology. The entity output module is used to identify fine-grained medical entities corresponding to the medical transcription text using a BERT-based NER model, match standard medical terms to the fine-grained medical entities using the SNOMED-CT dataset, assign disease classification codes to the fine-grained medical entities using the ICD-10 coding system, perform semantic matching between the standard medical terms and disease classification codes and the corresponding medical entity scheme, and output a set of key medical entities. The medical record generation module is used to perform data standardization and data structuring on the set of key diagnostic and treatment entities to generate a structured set of key diagnostic and treatment entities. Based on a preset medical record database, it automatically generates preliminary medical record documents corresponding to the set of structured key diagnostic and treatment entities. The module then reviews and stores the preliminary medical record documents to generate the final medical record documents.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor runs the computer program stored in the memory, the processor performs the steps of the automatic voice medical record generation method for intelligent diagnosis and treatment scenarios as described in any one of claims 1-7.

10. A readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it is used to implement the steps of the automatic voice medical record generation method for intelligent diagnosis and treatment scenarios as described in any one of claims 1-7.

Images