A method and system for automatic medical record filling based on voice input

Abstract

This invention provides a method and system for automatically filling in medical records based on voice input, belonging to the field of medical information processing technology. Firstly, during diagnosis and treatment, voice input data is received in stages through a voice acquisition device, and real-time diagnosis and treatment stage identifiers are obtained. Based on these real-time diagnosis and treatment stage identifiers, a staged voice processing model is constructed. Semantic hierarchical processing is then performed on the voice input data to generate hierarchical semantic results, establishing a dynamic mapping relationship between these results and medical record fields, and converting the data into a filling template to generate staged medical record documents. Semantic inheritance processing is then performed on documents from different stages to correct semantic gaps, resulting in a complete medical record document, which is then uploaded to the hospital's electronic medical record system for automatic filling. This invention improves the efficiency and accuracy of medical record recording.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and more specifically, to a method and system for automatically filling in medical records based on voice input. Background Technology

[0002] In the healthcare industry, medical records are a crucial part of the medical process, detailing patient information, treatment procedures, diagnostic results, and treatment plans. Traditional methods of recording medical records rely primarily on manual handwriting or keyboard input by medical staff. This approach is not only inefficient but also prone to errors or omissions of important information due to distraction during busy consultations.

[0003] With the development of speech recognition technology, some hospitals have begun to try introducing voice-based medical record entry to improve the efficiency of medical record recording. However, existing voice-based medical record entry methods have many problems. On the one hand, existing speech processing models are often general-purpose and not optimized for different stages of diagnosis and treatment. They cannot accurately identify and extract semantic information related to the current stage of diagnosis and treatment, resulting in inaccurate and incomplete medical record content. On the other hand, in terms of medical record field mapping, existing methods usually use fixed mapping relationships and cannot dynamically adjust mapping rules according to different stages of diagnosis and treatment. This results in mismatches between the generated medical record content and the preset medical record fields, increasing the workload of subsequent manual correction. In addition, in multi-stage diagnosis and treatment, there may be semantic gaps between medical record documents from different stages. Existing methods cannot effectively correct these semantic gaps, resulting in logically incoherent generated medical record documents, affecting the quality and usability of the medical records. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, the present invention provides a method for automatically filling in medical records based on voice input, the method comprising:

[0005] During the diagnosis and treatment process, the voice acquisition device receives medical staff’s voice input data related to the medical record in stages, and simultaneously obtains the real-time diagnosis and treatment stage identifier corresponding to the diagnosis and treatment process.

[0006] Based on the real-time diagnosis and treatment stage identifier, the clinical scene feature library is retrieved, and scene-related features that match the real-time diagnosis and treatment stage identifier are extracted from the clinical scene feature library to construct a staged speech processing model.

[0007] The medical record-related voice input data is input into a staged voice processing model, and semantic hierarchical processing is performed to generate hierarchical semantic results containing surface semantic information, deep semantic information and scene-related semantic information.

[0008] A preset medical record field library is retrieved, and a dynamic mapping relationship between hierarchical semantic results and medical record fields is established. The dynamic mapping relationship is adjusted as the real-time diagnosis and treatment stage identifier changes.

[0009] Based on the dynamic mapping relationship, the hierarchical semantic results are converted into text content that meets the requirements of the corresponding medical record fields, filled into the target fields of the preset medical record template, and a staged medical record document is generated.

[0010] Semantic inheritance processing is performed on the staged medical record documents corresponding to different real-time diagnosis and treatment stages to correct the semantic gaps between fields caused by stage switching, so as to obtain complete medical record documents. The complete medical record documents are then uploaded to the hospital's electronic medical record system for structured storage, and the automatic medical record filling operation is completed.

[0011] Furthermore, the present invention also provides an automatic medical record filling system based on voice input, comprising:

[0012] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to perform the above-described voice-based medical record autofill method by executing the machine-executable instructions.

[0013] In another aspect, the present invention also provides a computer program product, the computer program product including machine-executable instructions stored in a computer-readable storage medium, wherein the processor of the voice-based medical record autofill system reads the machine-executable instructions from the computer-readable storage medium, and the processor executes the machine-executable instructions, causing the voice-based medical record autofill system to perform the above-described voice-based medical record autofill method.

[0014] Based on the above, by receiving medical record-related voice input data from medical personnel in stages during the diagnosis and treatment process, and simultaneously acquiring real-time diagnosis and treatment stage identifiers, and then retrieving a clinical scenario feature library based on the real-time diagnosis and treatment stage identifiers to construct a staged voice processing model, the voice processing can closely align with the actual needs of different diagnosis and treatment stages, effectively improving the accuracy of voice recognition and the targeting of semantic extraction. Inputting the voice input data into the staged voice processing model for semantic layering processing generates layered semantic results containing surface, deep, and scenario-related semantic information, further mining the semantic connotations within the voice data. A preset medical record field library is retrieved to establish a dynamic mapping relationship, which adjusts according to changes in the real-time diagnosis and treatment stage identifiers, ensuring that the layered semantic results can be accurately converted into text content that meets the requirements of the medical record fields, improving the accuracy and flexibility of medical record filling. Based on the dynamic mapping relationship, the system fills in the preset medical record template to generate staged medical record documents. Semantic inheritance processing is performed on documents of different stages to correct the semantic gaps between fields caused by stage switching. Finally, a logically coherent and complete medical record document is obtained and uploaded to the hospital's electronic medical record system for structured storage. This realizes the automatic, efficient and accurate filling of medical records, which greatly improves the efficiency of medical work and the quality of medical records. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the execution flow of the automatic medical record filling method based on voice input provided in an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of exemplary hardware and software components of the voice-based automatic medical record filling system provided in an embodiment of the present invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an embodiment of the automatic medical record filling method based on voice input provided by the present invention. The following is a detailed description of the automatic medical record filling method based on voice input.

[0018] Step S110: During the diagnosis and treatment process, the voice acquisition device receives the medical record-related voice input data input by the medical personnel in stages, and simultaneously obtains the real-time diagnosis and treatment stage identifier corresponding to the diagnosis and treatment process.

[0019] In this embodiment, the complete process of a patient undergoing routine medical treatment in a hospital is used as the application scenario. This scenario begins with the patient entering the examination room and proceeds sequentially through stages such as consultation, examination, diagnosis, treatment plan formulation, and medical orders. These stages will be used throughout the entire document as a unified application scenario. In this scenario, medical personnel will record relevant medical record information via voice at different stages of the treatment. The voice acquisition device needs to receive this voice data in stages and simultaneously obtain the corresponding real-time treatment stage identifiers.

[0020] Step S111: During the diagnosis and treatment process, the voice data related to medical records input by medical personnel is received in batches according to the time nodes through the time-sharing acquisition capability of the voice acquisition device. Each batch of voice data is marked with a corresponding acquisition timestamp, so that the acquisition time period can be traced through the acquisition timestamp of each batch of voice data.

[0021] In the aforementioned application scenario, once a patient enters the consultation room, the voice acquisition device initiates its time-sharing data collection function. For example, at the initial stage of the consultation, medical personnel begin asking the patient about their basic information. The voice acquisition device receives the voice data input by the medical personnel during preset time intervals, such as at regular intervals, as a batch. For each batch of voice data received, the voice acquisition device automatically assigns a timestamp to that batch, accurate enough to reflect the specific time period during which the data was collected. This timestamp allows for accurate tracking of the specific time period during the consultation process when that batch of voice data was collected, ensuring that each batch of voice data has a clear time attribution.

[0022] Step S112: Obtain the real-time treatment stage identifier synchronized with the collection timestamp through the hospital treatment management system interface. The real-time treatment stage identifier includes the consultation stage identifier, examination stage identifier, diagnosis stage identifier, treatment plan formulation stage identifier, and medical order stage identifier. Through the time synchronization mechanism, the collection timestamp and the time recorded by the hospital treatment management system are based on the same time base, so as to achieve accurate correspondence between the collection timestamp and the real-time treatment stage identifier.

[0023] In the aforementioned application scenario, after the voice acquisition device timestamps the voice input data for each batch of medical records, it can access the hospital's treatment management system interface via the hospital's internal network. The hospital's treatment management system records information for each stage of the entire treatment process. Its time system is calibrated with the voice acquisition device's time system through a dedicated time synchronization mechanism to ensure complete consistency in time accuracy, for example, both are accurate to the same time unit. Thus, when the voice acquisition device needs to obtain the real-time treatment stage identifier corresponding to the timestamp of a particular batch of voice data, it can accurately obtain the identifier corresponding to the ongoing treatment stage at that point in time through the hospital's treatment management system interface. For example, in the consultation stage, the identifier obtained would be the consultation stage identifier, thereby achieving an accurate correspondence between the acquisition timestamp and the real-time treatment stage identifier.

[0024] Step S113: Send an identifier query request containing the collection timestamp through the stage identifier query capability of the hospital diagnosis and treatment management system. After receiving the identifier query request, retrieve the diagnosis and treatment operation record corresponding to the collection timestamp. Extract the diagnosis and treatment operation type performed by the medical personnel at that time point from the diagnosis and treatment operation record. Determine the corresponding real-time diagnosis and treatment stage identifier based on the diagnosis and treatment operation type. The diagnosis and treatment operation type and the real-time diagnosis and treatment stage identifier correspond one-to-one. Feed back the determined real-time diagnosis and treatment stage identifier to the relevant voice processing process.

[0025] In the aforementioned application scenario, after acquiring the acquisition timestamp, the voice acquisition device can generate an identifier query request containing the acquisition timestamp based on the stage identifier query capability provided by the hospital's medical management system, and send this request to the hospital's medical management system. Upon receiving the request, the hospital's medical management system can immediately retrieve the record corresponding to the acquisition timestamp from its stored medical operation records. For example, if the acquisition timestamp of a batch of voice data corresponds to the time period when a medical staff member is inquiring about a patient's medical history, the retrieved medical operation record will show that the medical staff member is performing a consultation operation at this time. Then, the type of medical operation performed by the medical staff member is extracted from the medical operation record. Since the type of medical operation corresponds one-to-one with the real-time medical stage identifier, such as the consultation operation type corresponding to the consultation stage identifier, the corresponding real-time medical stage identifier can be determined as the consultation stage identifier based on the extracted consultation operation type, and this identifier is fed back to the subsequent voice processing related processes.

[0026] Step S1131: Parse the identifier query request containing the collection timestamp, determine the time retrieval range, which is the interval between the collection timestamp and the preset time period, and retrieve all medical operation record entries within the time retrieval range in the medical operation record database.

[0027] In the above application scenario, after receiving an identifier query request containing a collection timestamp, the hospital's medical management system first parses the request. The collection timestamp is extracted from the request, and then a time retrieval range is determined according to pre-set rules. This time retrieval range is typically a preset time interval before and after the collection timestamp. For example, the time retrieval range can be set to a certain time interval before and after the collection timestamp. After determining the time retrieval range, the hospital's medical management system can access the medical operation record database and retrieve all medical operation record entries whose operation times fall within the determined time retrieval range. These medical operation record entries may contain information about various medical operations performed by medical personnel during that time period.

[0028] Step S1132: Adjust the time precision of each diagnosis and treatment operation record item, compare the operation time of the record item with the collection timestamp, and filter the target record items whose difference between the operation time and the collection timestamp is less than a set threshold; if there are multiple target record items, select the record item corresponding to the diagnosis and treatment operation with the highest priority as the final target item according to the priority order of the diagnosis and treatment operations, and extract the diagnosis and treatment operation type field of the specific diagnosis and treatment operation name performed by the medical personnel from the final target item.

[0029] In the above application scenario, after the hospital's medical management system retrieves all medical operation record entries within the time retrieval range from the medical operation record database, it needs to process the operation time of each record entry. Due to potential minor time errors between different devices or systems, the operation time of each medical operation record entry is first adjusted for time precision to ensure it is completely consistent with the time format and precision of the collected timestamp. Then, the operation time of each record entry is compared with the collected timestamp, and the time difference between the two is calculated. Next, record entries whose difference between the operation time and the collected timestamp is less than a set threshold are selected as target record entries. If multiple target record entries exist after filtering, the hospital's medical management system can select them according to a preset priority order of medical operations. For example, within the same time retrieval range, there may be both consultation operation records and preliminary examination operation records. Based on the priority order, the consultation operation has a higher priority than the preliminary examination operation, so the record entry corresponding to the consultation operation is selected as the final target entry. Then, a field specifically recording the name of the medical operation performed by the medical personnel, i.e., the medical operation type field, is extracted from this final target entry.

[0030] Step S11321: Obtain the operation time of each diagnosis and treatment operation record entry, convert the operation time into the same time format as the collection timestamp, calculate the difference between the operation time and the collection timestamp of each diagnosis and treatment operation record entry, and record the diagnosis and treatment operation record entry corresponding to each difference.

[0031] In the above application scenario, when processing medical operation records, the hospital's medical management system first obtains the operation time recorded in each entry. Since these operation times may come from different subsystems and their time formats may differ, all operation times need to be uniformly converted to the same time format as the collection timestamp, for example, uniformly converting it to a format including year, month, day, hour, minute, and second. After conversion, for each medical operation record entry, the difference between its operation time and the collection timestamp is calculated. This difference reflects the time interval between the operation time and the collection timestamp. Then, each difference is associated with the corresponding medical operation record entry for subsequent filtering operations.

[0032] Step S11322: Compare the calculated difference with the set threshold, retain the diagnosis and treatment operation record entries whose difference is less than the set threshold, and use the retained diagnosis and treatment operation record entries as target record entries.

[0033] In the above application scenario, after calculating the difference between the operation time and the collection timestamp for each medical operation record entry, the hospital's medical management system can compare this difference with a preset threshold. This threshold is determined based on the actual situation of the medical process and the time accuracy requirements, and is used to determine whether the operation time is sufficiently close to the collection timestamp. For medical operation record entries with a difference less than the preset threshold, it indicates that their operation time is very close to the time point corresponding to the collection timestamp, and they are likely the medical operation record corresponding to that collection timestamp. Therefore, these entries are retained as target record entries.

[0034] Step S11323: If there is only one target record entry, directly use that target record entry as the final target entry; if there are multiple target record entries, retrieve the preset list of treatment operation priorities; the list of treatment operation priorities sets the priority of each treatment operation according to the order and importance of the treatment process, with the priority from high to low corresponding to different treatment operation types.

[0035] In the above application scenario, after filtering to obtain the target record entries, it is necessary to determine the final target record entry. If there is only one target record entry after filtering, then this entry is the unique medical operation record corresponding to the collection timestamp, and it is directly used as the final target entry. If there are multiple target record entries, it means that multiple medical operations are being performed or recorded before and after the time point corresponding to the collection timestamp. In this case, the hospital's medical management can retrieve a preset list of medical operation priorities. This list of medical operation priorities is preset according to the sequence of the medical process and the importance of various operations in the medical process. The list of medical operation priorities clearly specifies the priority order of different types of medical operations, arranged in descending order of priority.

[0036] Step S11324: Match the treatment operation type in each target record entry with the treatment operation priority list to determine the priority level of each target record entry.

[0037] In the above application scenario, after the hospital's medical management system retrieves the list of priority medical operations, it can match the medical operation type field in each target record entry with the medical operation types in the list one by one. For example, if the medical operation type of a target record entry is medical history taking, the priority level corresponding to the medical history taking operation is found in the priority list; if the medical operation type of another target record entry is vital sign measurement, its corresponding priority level is also found in the list. Through this matching method, the corresponding priority level for each target record entry is determined.

[0038] Step S11325: Select the target record entry with the highest priority level. If there are multiple target record entries with the same priority level, compare the difference between the operation time and the collection timestamp of the multiple target record entries with the same priority level, and select the target record entry with the smallest difference.

[0039] In the above application scenario, after determining the priority level for each target record entry, the hospital's medical management system can select the target record entry with the highest priority level. If multiple target record entries have the same priority level, then it is necessary to further compare the difference between the operation time and the collection timestamp of these entries. The smaller the difference, the closer the operation time is to the time point corresponding to the collection timestamp, and the more likely the entry is to be the accurate medical operation record corresponding to that collection timestamp. Therefore, the target record entry with the smallest difference is selected as the final target entry.

[0040] Step S11326: Determine the final selected target record entry as the final target entry, and extract the diagnosis and treatment operation type field from the final target entry.

[0041] In the above application scenario, after the screening and comparison process described above, the hospital treatment management system determines the final target item. Then, from this final target item, a treatment operation type field is extracted, which is specifically used to record the names of the specific treatment operations performed by medical personnel. The content of this treatment operation type field will be used to subsequently determine the real-time treatment stage identifier.

[0042] Step S1133: Retrieve the preset operation-stage mapping relationship. This operation-stage mapping relationship stores the correspondence between all diagnosis and treatment operation types and real-time diagnosis and treatment stage identifiers. Each diagnosis and treatment operation type uniquely corresponds to a real-time diagnosis and treatment stage identifier.

[0043] In the above application scenario, after the hospital's medical management system extracts the medical operation type field from the final target item, it can retrieve a preset operation-stage mapping table within the system. This preset operation-stage mapping table is pre-defined according to the hospital's medical standards and procedures, and it stores in detail the correspondence between all possible medical operation types and real-time medical stage identifiers. In this correspondence, each specific medical operation type uniquely corresponds to a real-time medical stage identifier; for example, the consultation operation type corresponds to the consultation stage identifier, and the examination operation type corresponds to the examination stage identifier, etc.

[0044] Step S1134: Find the entry in the operation-stage mapping relationship that matches the extracted diagnosis and treatment operation type, and obtain the real-time diagnosis and treatment stage identifier corresponding to the entry.

[0045] In the above application scenario, after the hospital treatment management system retrieves the operation-stage mapping relationship, it can compare the treatment operation types extracted from the final target item with the items in the mapping relationship one by one. When an item is found that completely matches the extracted treatment operation type, the real-time treatment stage identifier corresponding to that item is the identifier that needs to be obtained. For example, if the extracted treatment operation type is a diagnostic operation, then the item corresponding to the diagnostic operation is found in the operation-stage mapping relationship, and the real-time treatment stage identifier corresponding to that item is the diagnostic stage identifier.

[0046] Step S1135: If the obtained real-time diagnosis and treatment stage identifier belongs to the set of valid stage identifiers including consultation stage identifier, examination stage identifier, diagnosis stage identifier, treatment plan formulation stage identifier, and medical order stage identifier, then the real-time diagnosis and treatment stage identifier is used as the query result; if the obtained real-time diagnosis and treatment stage identifier does not belong to the set of valid stage identifiers including consultation stage identifier, examination stage identifier, diagnosis stage identifier, treatment plan formulation stage identifier, and medical order stage identifier, an operation type confirmation request is sent to the medical personnel terminal.

[0047] In the above application scenario, after obtaining the real-time treatment stage identifier, the hospital treatment management system can verify the validity of the identifier. The hospital treatment management system pre-defines a set of valid stage identifiers, which includes five types: consultation stage identifier, examination stage identifier, diagnosis stage identifier, treatment plan formulation stage identifier, and medical order stage identifier. If the obtained real-time treatment stage identifier belongs to one of these valid types, it is directly fed back as the query result to the relevant voice processing workflow. If the obtained real-time treatment stage identifier is not in the valid set, it indicates a possible error in operation type recognition or other anomalies. In this case, the hospital treatment management system can send an operation type confirmation request to the terminal device being used by the medical personnel via the network, requesting the medical personnel to confirm the current treatment operation type.

[0048] Step S1136: Receive the confirmed treatment operation type from the medical personnel's terminal, search for the entry that matches the confirmed treatment operation type in the operation-stage mapping relationship, and obtain the corresponding real-time treatment stage identifier.

[0049] In the above application scenario, after receiving an operation type confirmation request, the medical personnel will confirm the operation type based on the actual diagnosis and treatment procedure, and then feed the confirmed operation type back to the hospital's diagnosis and treatment management system through the terminal. Upon receiving the confirmed operation type, the hospital's diagnosis and treatment management system will again search for an entry matching the confirmed operation type in the operation-stage mapping relationship, and retrieve the corresponding real-time diagnosis and treatment stage identifier from that entry.

[0050] Step S1137: Associate and store the confirmed real-time diagnosis and treatment stage identifier with the collection timestamp, so that the corresponding real-time diagnosis and treatment stage identifier can be quickly matched by the collection timestamp when called in the future, so that each collection timestamp corresponds to a unique real-time diagnosis and treatment stage identifier.

[0051] In the above application scenario, after the hospital's treatment management system obtains the confirmed real-time treatment stage identifier, it can associate and store this identifier with the corresponding data collection timestamp. For example, a record can be created in the system's database, saving the data collection timestamp and the corresponding real-time treatment stage identifier as association fields. In subsequent processes, when the real-time treatment stage identifier needs to be obtained via the data collection timestamp, a quick query and match can be performed directly in the database, ensuring that each data collection timestamp accurately corresponds to a unique real-time treatment stage identifier and avoiding confusion.

[0052] Step S120: Based on the real-time diagnosis and treatment stage identifier, retrieve the clinical scene feature library, extract the scene-related features that match the real-time diagnosis and treatment stage identifier from the clinical scene feature library, and construct a staged speech processing model.

[0053] In the aforementioned application scenarios, once the real-time diagnosis and treatment stage identifier is obtained, the clinical scene feature library can be retrieved based on this identifier. The clinical scene feature library stores various scene feature information related to different diagnosis and treatment stages. The corresponding scene-related features can be accurately located using the real-time diagnosis and treatment stage identifier. For example, when the real-time diagnosis and treatment stage identifier is the consultation stage identifier, scene-related features related to the consultation stage are extracted from the clinical scene feature library. These features are then used to construct a staged speech processing model suitable for the consultation stage, enabling targeted processing of the speech data collected at this stage.

[0054] Step S121: Use the real-time diagnosis and treatment stage identifier as an index to retrieve the clinical scenario feature library, traverse the scenario feature data stored in the clinical scenario feature library, filter the target scenario feature data whose stage identifier label is consistent with the index, and extract the set of commonly used medical terms, semantic expression logic rules and field association priority of the diagnosis and treatment stage from the target scenario feature data. All scenario feature data contain the corresponding stage identifier label.

[0055] In the above application scenario, this embodiment uses the obtained real-time diagnosis and treatment stage identifier as an index to access the clinical scenario feature library. Each scenario feature data in the clinical scenario feature library carries its own stage identifier label, indicating the diagnosis and treatment stage to which the data belongs. This embodiment traverses all scenario feature data in the clinical scenario feature library using this index, checking one by one whether the stage identifier label of each data is consistent with the real-time diagnosis and treatment stage identifier. When a scenario feature data with a stage identifier label consistent with the index is found, it is identified as the target scenario feature data. Then, the set of medical terms frequently used in this diagnosis and treatment stage is extracted from the target scenario feature data, such as symptom description terms and past medical history terms that may be involved in the consultation stage; simultaneously, the semantic expression logic rules specific to this stage are extracted, such as the logic rules for inquiring about the development of the illness in chronological order during the consultation stage; and the association priority between various medical record fields, such as the association priority of the patient's basic information field may be higher than that of some other fields during the consultation stage.

[0056] Step S122: If the target scene feature data has missing content, obtain relevant knowledge from the associated clinical guideline database, generate the missing set of commonly used medical terms, semantic expression logic rules or field association priorities, and supplement them into the target scene feature data so that the supplemented target scene feature data fully includes the set of commonly used medical terms, semantic expression logic rules and field association priorities.

[0057] In the aforementioned application scenarios, when extracting the set of commonly used medical terms, semantic expression logic rules, and field association priorities from the target scenario feature data, this embodiment may find that some content in the target scenario feature data is missing. For example, the set of commonly used medical terms may lack certain latest symptom terms, or the semantic expression logic rules may be incomplete. In this case, a clinical guideline database associated with the clinical scenario feature library can be accessed. This clinical guideline database contains the latest clinical diagnosis and treatment guidelines and standards. Based on the type of missing content in the target scenario feature data, a search is performed in the clinical guideline database to obtain the corresponding missing content, and this content is then added to the target scenario feature data. This ensures that the supplemented target scenario feature data fully includes the above three aspects, meeting the requirements for constructing a phased speech processing model.

[0058] Step S123: Construct the basic structure of a staged speech processing model that includes a semantic input part, a feature matching part, and a semantic output part. Import a set of commonly used medical terms into the semantic input part as the term matching benchmark for speech data. Configure the semantic expression logic rules into the feature matching part as the logical basis for semantic analysis. Set the field association priority into the semantic output part as the priority standard for semantic results and field mapping.

[0059] In the aforementioned application scenario, this embodiment begins by constructing the basic structure of a staged speech processing model. This basic structure mainly comprises three parts: a semantic input part, a feature matching part, and a semantic output part. First, a set of commonly used medical terms extracted and supplemented from the target scene feature data is imported into the semantic input part. This term set will serve as the benchmark for subsequent term matching of the speech data; the content of the speech data will be compared and identified against these terms. Next, semantic expression logic rules are configured in the feature matching part. When performing semantic analysis on the speech data, the feature matching part will determine the logical relationships between semantics based on these rules, ensuring the accuracy of the semantic analysis. Finally, field association priority is set in the semantic output part. When the semantic output part maps semantic results to medical record fields, the mapping order and importance can be determined according to the field association priority; fields with higher priority will be mapped first.

[0060] Step S124: Select historical speech processing case data whose stage identifier matches the current real-time diagnosis and treatment stage identifier, input the historical speech processing case data into the basic structure of the staged speech processing model, and optimize the model processing effect by adjusting the term matching parameters of the semantic input part, the logical analysis parameters of the feature matching part, and the priority ranking parameters of the semantic output part.

[0061] In the aforementioned application scenarios, to ensure the constructed staged speech processing model better adapts to the speech processing needs of the current real-time diagnosis and treatment stage, historical speech processing case data consistent with the current real-time diagnosis and treatment stage identifier can be selected. For example, if the current real-time diagnosis and treatment stage identifier is the examination stage identifier, historical speech processing case data from the examination stage can be selected. This historical case data is input into the basic structure of the staged speech processing model, which then processes the data. During processing, the term matching parameters of the semantic input part, such as the similarity threshold for term matching, can be continuously adjusted to improve the accuracy of term recognition; the logical analysis parameters of the feature matching part, such as the weight of logical relationship judgments, can be adjusted to optimize the logical rationality of semantic analysis; and the priority ranking parameters of the semantic output part, such as the weight values ​​of different field priorities, can be adjusted to make the mapping between semantic results and medical record fields more consistent with actual needs, thereby optimizing the overall processing effect of the model.

[0062] Step S125: When the semantic parsing accuracy of the staged speech processing model on historical speech processing case data reaches the set standard, the final staged speech processing model is determined, so that the staged speech processing model can process speech data according to the scene association features corresponding to the real-time diagnosis and treatment stage identifier.

[0063] In the aforementioned application scenarios, by continuously adjusting model parameters and testing and optimizing the staged speech processing model using historical speech processing case data, the model's semantic parsing accuracy can be evaluated. When the evaluation results show that the model's semantic parsing accuracy on historical speech processing case data reaches the system's preset standard, it indicates that the model has good processing capabilities. At this point, the model is determined as the final staged speech processing model. This staged speech processing model can accurately and efficiently process the collected medical record-related speech input data based on the scene association features corresponding to the current real-time diagnosis and treatment stage identifier.

[0064] Step S130: Input the medical record-related voice recording data into the staged voice processing model, perform semantic layering processing, and generate a layered semantic result containing surface semantic information, deep semantic information, and scene-related semantic information.

[0065] In the aforementioned application scenario, after constructing the staged speech processing model, the medical record-related speech data received in stages is input into the model. The staged speech processing model performs semantic hierarchical processing on the speech data. This process extracts different levels of semantic information from the speech data, including surface semantic information, deep semantic information, and scene-related semantic information, and finally integrates the above information to generate a hierarchical semantic result.

[0066] Step S131: Input the medical record-related voice input data into the semantic input part of the staged voice processing model, perform term matching with the set of commonly used medical terms, and extract the medical terms and expressions directly corresponding to the medical record-related voice input data as surface semantic information, so that the surface semantic information covers the direct medical expressions in the voice data.

[0067] In the above application scenario, medical record-related voice input data is first input into the semantic input part of the staged voice processing model. The semantic input part already contains a set of commonly used medical terms corresponding to the current real-time diagnosis and treatment stage identifier. The model converts the voice data into text and then matches it one by one with the terms in the set of commonly used medical terms. When a matching medical term is found, the term and its expression in the voice data can be extracted. The medical terms and expressions extracted directly from the voice data constitute the surface semantic information, ensuring that the surface semantic information can comprehensively cover the direct medical expressions in the voice data.

[0068] Step S132: Transmit the surface semantic information to the feature matching part of the staged speech processing model, perform semantic reasoning on the surface semantic information according to the semantic expression logic rules of the feature matching part, mine the diagnosis and treatment related information hidden behind the surface semantic information, and use the mined information as deep semantic information so that the deep semantic information reflects the diagnosis and treatment related content not directly reflected in the surface semantic information.

[0069] In the aforementioned application scenarios, after the surface semantic information is generated, it can be transmitted to the feature matching part of the staged speech processing model. The feature matching part is configured with the semantic expression logic rules for the current diagnosis and treatment stage. Based on these rules, the model performs in-depth semantic reasoning on the surface semantic information, analyzing the inherent connections between various expressions within the surface semantic information and uncovering diagnosis and treatment-related information that is not directly reflected in the surface semantic information. For example, if the surface semantic information mentions a patient with "cough and fever," semantic reasoning may uncover the implicit diagnosis and treatment-related information such as "possible respiratory infection." This information constitutes the deep semantic information.

[0070] Step S133: Based on the semantic expression logic rules, determine the semantic adaptation dimension corresponding to the current real-time diagnosis and treatment stage identifier, divide the deep semantic information into multiple semantic sub-modules according to the semantic adaptation dimension, match the scene attribute label corresponding to the real-time diagnosis and treatment stage identifier for each semantic sub-module, and each semantic sub-module contains only the expression content belonging to the same semantic adaptation dimension.

[0071] In the aforementioned application scenario, after generating deep semantic information, the feature matching component can determine the semantic adaptation dimension corresponding to the current real-time diagnosis and treatment stage identifier based on semantic expression logic rules. Different diagnosis and treatment stages have different semantic adaptation dimensions; for example, the semantic adaptation dimensions for the diagnosis stage may include etiology analysis dimensions, symptom judgment dimensions, etc. Then, the deep semantic information is divided according to these semantic adaptation dimensions, grouping expressions belonging to the same semantic adaptation dimension together to form multiple semantic sub-modules. Each semantic sub-module contains only content within the same semantic adaptation dimension, ensuring consistency of module content. Finally, each semantic sub-module is matched with a scene attribute tag corresponding to the current real-time diagnosis and treatment stage identifier for subsequent association with medical record fields.

[0072] Step S1331: Extract the semantic adaptation dimension description corresponding to the current real-time diagnosis and treatment stage identifier from the clinical scenario feature library. The semantic adaptation dimension description contains the core dimensions that the semantic analysis needs to cover under the corresponding stage. Determine the semantic adaptation dimension of the current stage based on the semantic adaptation dimension description.

[0073] In the aforementioned application scenario, the feature matching component of the staged speech processing model accesses a clinical scenario feature library to extract a semantic adaptation dimension description corresponding to the current real-time diagnosis and treatment stage identifier. This semantic adaptation dimension description details the core dimensions that need to be covered when performing semantic analysis at the current diagnosis and treatment stage. For example, in the treatment plan formulation stage, the semantic adaptation dimension description might indicate that core dimensions such as treatment method, medication regimen, and treatment cycle need to be included. Based on the aforementioned semantic adaptation dimension description, the feature matching component determines the specific semantic adaptation dimension for the current stage.

[0074] Step S1332: Analyze the deep semantic information sentence by sentence, identify the semantic adaptation dimension corresponding to each sentence, and classify the expressions in the deep semantic information that belong to the same dimension according to the identified semantic adaptation dimension to form multiple semantic sub-modules.

[0075] In the above application scenarios, after determining the semantic adaptation dimension, the feature matching part analyzes the deep semantic information sentence by sentence. For each statement, by analyzing its core meaning and the content involved, the semantic adaptation dimension corresponding to the statement is identified. For example, if a statement is about drug selection, its dimension related to the drug application plan is identified. Then, all statements belonging to the same semantic adaptation dimension in the deep semantic information are grouped together to form independent semantic sub-modules, each focusing on a specific semantic adaptation dimension.

[0076] Step S1333: Retrieve the preset stage-tag mapping relationship. This stage-tag mapping relationship stores the correspondence between all real-time diagnosis and treatment stage identifiers and scene attribute tags. Each real-time diagnosis and treatment stage identifier corresponds to multiple scene attribute tags, and each scene attribute tag corresponds to a specific semantic adaptation dimension.

[0077] In the aforementioned application scenarios, the feature matching component retrieves a pre-defined stage-label mapping relationship. This mapping relationship records the correspondence between all real-time diagnosis and treatment stage identifiers and scene attribute labels. Each real-time diagnosis and treatment stage identifier corresponds to multiple scene attribute labels, and each scene attribute label explicitly corresponds to a specific semantic adaptation dimension. For example, a diagnosis stage identifier might correspond to scene attribute labels such as etiology label and symptom label, where the etiology label corresponds to the semantic adaptation dimension of etiology analysis, and the symptom label corresponds to the semantic adaptation dimension of symptom judgment.

[0078] Step S1334: Extract all corresponding scene attribute tags from the stage-tag mapping relationship based on the current real-time diagnosis and treatment stage identifier, and match the extracted scene attribute tags with the semantic adaptation dimension corresponding to the semantic submodule.

[0079] In the above application scenario, the feature matching part extracts all scene attribute tags corresponding to the current real-time diagnosis and treatment stage identifier in the stage-tag mapping relationship. Then, the extracted scene attribute tags are matched with the semantic adaptation dimensions corresponding to each semantic sub-module, and it is checked whether the semantic adaptation dimension of each semantic sub-module can find the corresponding scene attribute tag.

[0080] Step S1335: Assign a unique scene attribute label to each semantic submodule. If a semantic submodule cannot be matched with the corresponding scene attribute label, retrieve the standard label corresponding to the semantic adaptation dimension from the clinical guideline database and assign it as a supplementary scene attribute label to the semantic submodule.

[0081] In the above application scenario, for semantic submodules that can match scene attribute labels, the feature matching part assigns a unique scene attribute label to them. If a semantic submodule's semantic adaptation dimension cannot find a match in the extracted scene attribute labels, the feature matching part will access the clinical guideline database to retrieve the standard label corresponding to that semantic adaptation dimension. The clinical guideline database contains various standard medical terms and label definitions. The retrieved standard label is assigned as a supplementary scene attribute label to the semantic submodule, ensuring that each semantic submodule has a corresponding scene attribute label.

[0082] Step S1336: Assign the same scene attribute labels to all semantic sub-modules of the same semantic adaptation dimension. If there are inconsistent labels, re-execute the label assignment step.

[0083] In the above application scenario, when assigning scene attribute labels to semantic submodules, the feature matching part can ensure that all semantic submodules belonging to the same semantic adaptation dimension are assigned the same scene attribute labels. If, during the assignment process, it is found that semantic submodules of the same semantic adaptation dimension are assigned different scene attribute labels, it indicates that an error has occurred in the assignment process. In this case, the label assignment step needs to be re-executed to find the cause of the error and correct it to ensure the consistency of label assignment.

[0084] Step S1337: Determine the priority order of semantic adaptation dimensions based on the treatment focus corresponding to the real-time treatment stage identifier, sort the semantic sub-modules after adding scene attribute tags according to the priority order, combine the sorted semantic sub-modules in sequence to form complete scene-related semantic information, and adjust the expression content or sorting order of the semantic sub-modules.

[0085] In the aforementioned application scenarios, different real-time diagnosis and treatment stages correspond to different treatment priorities. Based on these priorities, the priority order of each semantic adaptation dimension can be determined. For example, the treatment priority in the treatment plan formulation stage might be medication regimen, thus the medication regimen dimension would have a higher priority. The feature matching part sorts the semantic sub-modules after adding scene attribute tags according to the determined priority order, placing the semantic sub-modules with higher priority at the beginning. Then, the semantic sub-modules are sequentially connected and combined according to the sorted order to form complete scene-related semantic information. During the combination process, if inconsistencies or unreasonable ordering of the semantic sub-module descriptions are found, the descriptions can be appropriately adjusted or reordered to ensure the logic and coherence of the scene-related semantic information.

[0086] In one example, step S1337 includes the following sub-steps:

[0087] Step S13371: Obtain the treatment focus description corresponding to the current real-time treatment stage identifier. This treatment focus description contains the content that needs to be given priority in the corresponding stage of treatment. Determine the importance of each semantic adaptation dimension based on the treatment focus description.

[0088] In the aforementioned application scenario, taking the diagnostic stage as an example, the current real-time diagnosis and treatment stage is identified as the diagnostic stage identifier. A description of the key diagnostic and treatment priorities for this stage is retrieved from the clinical scenario feature database. This description explicitly states, "Prioritize identifying the disease type and analyzing the cause, while also considering symptom correlation judgment and complication screening." Based on this description, the semantic adaptation dimensions for the diagnostic stage can be determined to include disease judgment, cause analysis, symptom correlation, and complication screening. Among these, disease judgment and cause analysis are the core priorities and have the highest importance, followed by symptom correlation, while complication screening has a relatively lower importance.

[0089] Step S13372: Assign a corresponding priority score to each semantic adaptation dimension. The higher the importance of the semantic adaptation dimension, the higher the priority score is assigned, and the lower the importance of the semantic adaptation dimension, the lower the priority score is assigned.

[0090] In the above application scenario, the priority score range is set from 1 to 10 points, and the scores are allocated according to the importance of each semantic adaptation dimension: 10 points for symptom judgment, 9 points for etiology analysis, 7 points for symptom association, and 5 points for complication screening. This score quantification clearly distinguishes the importance of each dimension, providing a quantitative basis for subsequent ranking and ensuring the objectivity of priority judgment.

[0091] Step S13373: Take the priority score of the semantic adaptation dimension corresponding to each semantic submodule as the priority score of that semantic submodule, and sort all semantic submodules in order of priority score from high to low.

[0092] In the above application scenario, the diagnostic phase has been divided into four semantic sub-modules, corresponding to four semantic adaptation dimensions: Symptom Judgment Sub-module (10 points), Etiology Analysis Sub-module (9 points), Symptom Association Sub-module (7 points), and Complication Screening Sub-module (5 points). Preliminary ranking based on scores from highest to lowest is as follows: Symptom Judgment Sub-module — Etiology Analysis Sub-module — Symptom Association Sub-module — Complication Screening Sub-module.

[0093] Step S13374: If there are semantic sub-modules with the same priority score, refer to the order in which the semantic sub-modules with the same priority score appear in the deep semantic information, and sort the semantic sub-modules with the same priority score in the order of appearance from front to back. Arrange the semantic sub-modules after the initial sorting and secondary sorting in sequence to form a semantic sub-module sequence, so that the semantic sub-modules in the semantic sub-module sequence are arranged from high to low priority and the sub-modules with the same priority are arranged in the order of appearance.

[0094] In the above application scenario, if a semantic adaptation dimension of "symptom severity assessment" is added during the diagnosis stage, with a priority score of 9 points (the same as the score for the etiology analysis dimension), a corresponding "symptom severity assessment submodule" is generated. At this point, it is necessary to retrieve deep semantic information to check the order of appearance of the etiology analysis submodule and the symptom severity assessment submodule. If the deep semantic information mentions etiology analysis first and then symptom severity assessment, the order after secondary sorting is: etiology analysis submodule — symptom severity assessment submodule. The final semantic submodule sequence is adjusted to: symptom judgment submodule — etiology analysis submodule — symptom severity assessment submodule — symptom association submodule — complication screening submodule, ensuring that submodules with the same score are ordered according to their semantic appearance logic.

[0095] Step S13375: If the semantic submodules in the semantic submodule sequence are incomplete or ambiguous, adjust the description of the semantic submodules.

[0096] In the aforementioned application scenario, an examination of the semantic submodule sequence revealed that the complication screening submodule was described as "no other obvious problems," which was vague and incomplete, failing to clarify whether the screening for complications had been completed. Based on the standard description of complication screening in the clinical guideline database, it was revised to "Preliminary screening revealed no complications related to the current condition, such as those in the lungs or heart," thus expanding the scope of the screening and providing specific conclusions, eliminating ambiguity, ensuring accurate and standardized wording, and meeting medical documentation requirements.

[0097] Step S13376: Connect the contents of each semantic submodule sequentially according to the order of the semantic submodule sequence, add grammatically correct conjunctions or transitional statements between the contents of adjacent semantic submodules, adjust the overall contents of the connected contents, and form complete scene-related semantic information.

[0098] In the above application scenarios, the contents of each sub-module are connected sequentially according to the final semantic sub-module sequence, and transition words or sentences are added between adjacent sub-modules: After the content of the symptom judgment sub-module, "Based on the symptoms and examination results, the preliminary diagnosis is acute bronchitis," add "the core cause is"; after the content of the etiology analysis sub-module, "bacterial infection causes airway mucosal inflammation," add "and it is related to the patient's recent exposure to cold leading to decreased immunity"; after the content of the symptom association sub-module, "cough and sputum symptoms are directly related to airway inflammation stimulation," add "and at the same time"; after the content of the complication screening sub-module, "after preliminary screening, no complications related to the current symptoms in the lungs, heart, or other parts of the body were found." After the connection adjustment, complete scene-related semantic information is formed: "[Disease Judgment] Based on symptoms and examination results, the preliminary diagnosis is acute bronchitis. The core cause is [Etiological Analysis] bacterial infection causing airway mucosal inflammation, which is related to the patient's recent exposure to cold leading to decreased immunity. [Symptom Association] Coughing and sputum production are directly related to airway inflammation. [Complication Screening] Preliminary screening has not found any complications related to the current condition, such as lung or heart complications." This ensures overall logical coherence, fluent expression, and clear association of scene attribute tags with corresponding content.

[0099] Step S134: If a semantic submodule cannot match the corresponding scene attribute tag, retrieve the standard tag corresponding to the semantic adaptation dimension from the clinical guideline database as a supplementary scene attribute tag and assign it to the semantic submodule. Combine the semantic submodules with the added scene attribute tags to form scene-related semantic information. The scene attribute tags are pre-bound to the real-time diagnosis and treatment stage identifier.

[0100] In the above application scenario, this step further emphasizes and supplements step S1335. When a semantic submodule cannot match a scenario attribute label, standard labels are explicitly retrieved from the clinical guideline database as a supplement, and the semantic submodules with added labels are combined to form scenario-related semantic information. At the same time, the pre-binding relationship between scenario attribute labels and real-time diagnosis and treatment stage identifiers is emphasized to ensure a close connection between scenario-related semantic information and the current diagnosis and treatment stage.

[0101] Step S135: Construct a hierarchical semantic association matrix, using the surface semantic information as the base layer data of the hierarchical semantic association matrix, the deep semantic information as the intermediate layer data of the hierarchical semantic association matrix, and the scene association semantic information as the top layer data of the hierarchical semantic association matrix. The surface semantic information is split into multiple independent semantic representation units and filled into the base layer, the deep semantic information is split into multiple reasoning result units and filled into the intermediate layer, and the scene association semantic information is split into multiple scene label units and filled into the top layer.

[0102] In the above application scenarios, to clearly demonstrate the hierarchical relationships and associations between surface semantic information, deep semantic information, and scene-related semantic information, a hierarchical semantic association matrix can be constructed. This hierarchical semantic association matrix consists of a base layer, an intermediate layer, and a top layer. First, the surface semantic information is broken down into multiple independent semantic expression units, each representing a sentence or a complete surface semantic expression, and these units are then filled into the base layer of the matrix. Next, the deep semantic information is broken down into multiple inference result units, each corresponding to a diagnostic and treatment association information derived from inference, and these units are filled into the intermediate layer of the matrix. Finally, the scene-related semantic information is broken down into multiple scene label units, each corresponding to a scene attribute label and related content, and these units are filled into the top layer of the matrix. Through this method, the three layers of semantic information present a clear hierarchical structure in the matrix.

[0103] Step S136: Compare the interrelated basic layer semantic representation units and intermediate layer reasoning result units in the hierarchical semantic association matrix, identify the correspondence of core words and determine the semantic mapping type, record the semantic mapping type and core word pairs as association nodes in the corresponding cells of the hierarchical semantic association matrix, analyze the scene binding relationship between intermediate layer data and top layer data and mark the binding nodes, and integrate the hierarchical semantic association matrix with the semantic information of each layer to form a hierarchical semantic result.

[0104] In the aforementioned application scenario, after constructing the hierarchical semantic association matrix, a sentence-by-sentence comparison can be performed between the basic-layer semantic representation units and the intermediate-layer inference result units in each row of the matrix. During the comparison process, the correspondence between the core words is identified. For example, "fever" in the basic layer may correspond to "abnormal body temperature" in the intermediate layer. Based on this correspondence, the semantic mapping type is determined, such as synonymous mapping or causal mapping. These semantic mapping types and core word pairs are recorded in the corresponding cells of the matrix, serving as association nodes. Simultaneously, the scene binding relationship between the intermediate-layer inference result units and the top-layer scene label units is analyzed. For example, if a certain inference result unit belongs to the content of the medication plan scene label unit, the aforementioned binding node is marked in the matrix. Finally, the hierarchical semantic association matrix is ​​integrated with the semantic information of each layer to form a complete hierarchical semantic result containing the semantic information of each layer and their association relationships.

[0105] Step S140: Retrieve the preset medical record field library and establish a dynamic mapping relationship between hierarchical semantic results and medical record fields. The dynamic mapping relationship is adjusted as the real-time diagnosis and treatment stage identifier changes.

[0106] In the above application scenario, after generating the hierarchical semantic results, these results need to be associated with medical record fields to convert the semantic information into medical record text content. Therefore, a preset medical record field library is retrieved, which contains relevant information for various medical record fields. Then, based on the current real-time treatment stage identifier, a dynamic mapping relationship is established between the hierarchical semantic results and medical record fields. This mapping relationship adjusts accordingly as the real-time treatment stage identifier changes to meet the needs of medical record completion at different stages.

[0107] Step S141: Retrieve the preset medical record field library, and filter candidate medical record fields in the preset medical record field library whose stage adaptation range includes the current real-time diagnosis and treatment stage identifier according to the current real-time diagnosis and treatment stage identifier. Extract the semantic reception standard of each candidate medical record field and the scene-related semantic information in the hierarchical semantic results to calculate the adaptation degree. The preset medical record field library contains the field identifier, field function description, stage adaptation range and semantic reception standard of all medical record fields.

[0108] In the above application scenario, after retrieving the preset medical record field library, this embodiment can filter the medical record fields in the library according to the current real-time diagnosis and treatment stage identifier. Each medical record field in the preset medical record field library has its own stage adaptation range. Only medical record fields whose stage adaptation range includes the current real-time diagnosis and treatment stage identifier will be selected as candidate medical record fields. For example, if the current real-time diagnosis and treatment stage identifier is the medical order stage identifier, then medical record fields whose stage adaptation range includes the medical order stage identifier, such as medication medical order fields and follow-up medical order fields, will be filtered out. Next, the semantic reception standard for each candidate medical record field is extracted. This semantic reception standard specifies the types and formats of semantic information that the field can receive. Then, the above semantic reception standard is compared with the scene-related semantic information in the hierarchical semantic results to calculate the degree of matching between the scene-related semantic information and the candidate medical record field.

[0109] Step S142: Assign corresponding hierarchical semantic result content to each candidate medical record field according to the fit degree calculation result, record the correspondence between the candidate medical record field and the hierarchical semantic result content and the fit degree value to form an initial mapping relationship, and re-select candidate medical record fields, calculate fit degree and assign content when the real-time diagnosis and treatment stage identifier is switched, and update the initial mapping relationship to form a dynamic mapping relationship.

[0110] In the above application scenario, after calculating the fit degree between each candidate medical record field and the semantic information associated with the scenario, corresponding hierarchical semantic result content can be assigned to each candidate medical record field according to the fit degree value. The higher the fit degree value, the higher the degree of matching between the candidate medical record field and a certain part of the hierarchical semantic result content, and that part of the content is assigned to the candidate medical record field. The correspondence between the candidate medical record fields and the assigned hierarchical semantic result content, as well as the fit degree value, are recorded to form an initial mapping relationship. When the real-time diagnosis and treatment stage identifier changes, such as from the consultation stage to the examination stage, the candidate medical record fields can be re-selected according to the new real-time diagnosis and treatment stage identifier, the fit degree can be recalculated and the content can be assigned, and the initial mapping relationship can be updated, thus forming a dynamic mapping relationship that adjusts with the change of diagnosis and treatment stage.

[0111] Step S150: Based on the dynamic mapping relationship, convert the hierarchical semantic results into text content that meets the requirements of the corresponding medical record fields, fill in the target fields of the preset medical record template, and generate a staged medical record document.

[0112] In the aforementioned application scenarios, with the dynamic mapping relationship established, this embodiment can convert the hierarchical semantic results into text content that conforms to the requirements of the corresponding medical record fields. For example, if the dynamic mapping relationship specifies that a certain candidate medical record field corresponds to a certain part of the scenario-related semantic information in the hierarchical semantic results, this embodiment will convert that part of the semantic information into text that conforms to the format and content requirements of that medical record field. Then, the converted text content is filled into the corresponding target field in the preset medical record template to complete the medical record filling for that stage and generate a staged medical record document, such as a medical record document for the consultation stage or a medical record document for the examination stage.

[0113] Step S151: Traverse the dynamic mapping relationship to obtain the hierarchical semantic result content corresponding to each candidate medical record field, perform text conversion processing on the hierarchical semantic result content, convert the surface semantic information into the direct expression text required by the field, convert the deep semantic information into the reasoning supplementary text required by the field, and convert the scene association semantic information into the scene attribute annotation text required by the field.

[0114] In the above application scenarios, the dynamic mapping relationship can be traversed to obtain the hierarchical semantic result content corresponding to each candidate medical record field. The obtained hierarchical semantic result content can then undergo text conversion processing. Specifically, the surface semantic information is directly converted into direct text that conforms to the requirements of the corresponding medical record field, such as converting "patient reports cough for three days" into a standardized expression required by the medical record field. Deep semantic information is converted into inferential supplementary text to supplement the implicit meaning behind the surface semantic information, such as adding "possibly a respiratory infection" as inferential supplementary text to the corresponding field. Scenario-related semantic information is converted into scenario attribute-annotated text, annotating the scenario attribute to which this part of the content belongs, such as adding "[medication plan]" before the text to make the text content conform to the requirements of the medical record field.

[0115] Step S152: Obtain the field content requirements for each candidate medical record field from the preset medical record field library to determine the expression style, expression elements and expression format of the field content, and perform lexical analysis on the semantic information related to the voice content of each candidate medical record field to identify core expression words and auxiliary expression words.

[0116] In the above application scenario, before performing text conversion processing, this embodiment can obtain the field content requirements for each candidate medical record field from a preset medical record field library. These field content requirements specify in detail the expression style of the field content, such as formal or concise; expression elements, such as required information items; and expression format, such as paragraph structure and punctuation usage. Simultaneously, lexical analysis is performed on the semantic information related to the audio content corresponding to each candidate medical record field, decomposing the semantic information into individual words, and then identifying the core expression words and auxiliary expression words. Core expression words are those that express key information, while auxiliary expression words are those that modify or supplement the core words.

[0117] Step S153: Based on the field content requirements of the candidate medical record fields, convert the colloquial core expressions into standardized medical terms, filter auxiliary expressions, retain the parts relevant to the field content requirements and remove irrelevant parts, and combine the converted core expressions and the filtered auxiliary expressions according to grammatical rules to form a preliminary text expression.

[0118] In the above application scenario, after identifying the core and auxiliary vocabulary, the core vocabulary is processed according to the content requirements of the candidate medical record fields. If the core vocabulary is colloquial, it is converted into standardized medical terminology to ensure professionalism and standardization. For auxiliary vocabulary, it is filtered according to the field content requirements, retaining those relevant to the field content and removing irrelevant parts to avoid redundant information. Then, the converted core vocabulary and the filtered auxiliary vocabulary are combined according to grammatical rules to form a preliminary text description that basically meets the content requirements of the medical record fields.

[0119] Step S154: Analyze the frequency change features by referring to the speech intonation related semantic information corresponding to the candidate medical record fields, adjust the tone intensity of the preliminary text expression according to the frequency change features, analyze the pause interval distribution by referring to the speech pause related semantic information, adjust the sentence separation and logical pause position of the preliminary text expression according to the pause interval distribution, and integrate the adjusted preliminary text expression with the scene attribute annotation text to form complete text content.

[0120] In the aforementioned application scenarios, to ensure the generated text representation better matches the tone and logic of medical personnel during voice input, semantic information related to speech intonation corresponding to the candidate medical record fields can be referenced. The frequency variation characteristics of speech intonation are analyzed; for example, a rise in intonation in certain parts may indicate emphasis. Based on these characteristics, the tone intensity of the initial text representation is adjusted, using appropriate intonation for parts requiring emphasis. Simultaneously, semantic information related to speech pauses is referenced to analyze the distribution of pause intervals in the speech. The sentence separations and logical pause positions in the initial text representation are adjusted according to the pause positions and durations, such as using commas or periods to separate sentences at pauses, ensuring the reading rhythm and logic of the text are consistent with the voice input. Finally, the adjusted initial text representation is integrated with the scenario attribute annotation text to form complete text content that meets the requirements of the medical record fields.

[0121] Step S155: Retrieve the preset medical record template, locate the corresponding target field in the preset medical record template according to the field identifier of the candidate medical record field, fill the target field with the complete text content, repeat the conversion, integration and filling steps until all candidate medical record fields corresponding to the current real-time diagnosis and treatment stage identifier are filled, and generate the staged medical record document corresponding to the real-time diagnosis and treatment stage identifier.

[0122] In the above application scenario, a preset medical record template is retrieved, which contains the standard layout and format of various medical record fields. Based on the field identifier of each candidate medical record field, the corresponding target field location is accurately located in the preset medical record template. Then, the integrated complete text content is filled into the target field location. Following the same method, all candidate medical record fields corresponding to the current real-time diagnosis and treatment stage identifier are sequentially subjected to text conversion, integration, and filling operations until all candidate medical record fields are filled. At this point, the preset medical record template becomes a staged medical record document containing complete information about the current diagnosis and treatment stage.

[0123] Step S160: Perform semantic inheritance processing on the staged medical record documents corresponding to different real-time diagnosis and treatment stage identifiers, correct the semantic gaps between fields caused by stage switching, obtain complete medical record documents, and upload the complete medical record documents to the hospital's electronic medical record system for structured storage to complete the automatic medical record filling operation.

[0124] In the aforementioned application scenario, as the diagnosis and treatment process progresses, multiple staged medical record documents corresponding to different real-time diagnosis and treatment stages can be generated. These documents may experience semantic gaps due to stage switching; for example, some symptoms mentioned in the consultation stage may not have corresponding examination results in the examination stage's medical record document. Therefore, this embodiment requires semantic inheritance processing of the aforementioned staged medical record documents, analyzing the semantic relationships between fields in medical record documents from different stages, correcting the semantic gaps caused by stage switching, and ensuring that the medical record information from each stage can be coherently and consistently connected, ultimately resulting in a complete medical record document. Then, this complete medical record document is uploaded to the hospital's electronic medical record system and stored according to the system's structured storage requirements, completing the automatic filling operation of the entire medical record.

[0125] Step S161: After the real-time diagnosis and treatment stage identifier is switched and a new staged medical record document is generated, words that appear more frequently than the set value and belong to the corresponding stage are selected from all field contents of the previous staged medical record document as key medical terms. Statements containing judgmental and conclusive statements are identified from the field contents and the core conclusion part is extracted as the diagnosis and treatment conclusion statement.

[0126] In the above application scenario, when the treatment stage changes, such as switching from the examination stage to the diagnosis stage and generating a staged medical record document for the diagnosis stage, the content of all fields in the staged medical record document of the previous stage, i.e., the examination stage, can be filtered. Words that appear more frequently than a set value and belong to the set of commonly used medical terms in the examination stage are selected and identified as key medical terms. These terms are usually the core information in the previous stage's medical record. Simultaneously, statements containing judgmental or conclusive statements, such as "the examination results show a shadow in the lungs," are identified from the field content of the examination stage's medical record document. The core conclusion portion of these statements, such as "a shadow in the lungs," is then extracted as the diagnostic conclusion.

[0127] Step S162: Analyze the reference relationships, causal relationships, and supplementary relationships between different fields in the medical record document of the previous stage to form field association logic, and integrate key medical terms, diagnosis and treatment conclusions and field association logic to generate the core semantic elements of the previous stage.

[0128] In the aforementioned application scenarios, the relationships between different fields in the previous stage of medical record documents can be analyzed. This involves identifying referencing relationships between fields (e.g., one field references examination data from another); causal relationships (e.g., one symptom field causes another diagnosis); and supplementary relationships (e.g., one field supplements information from another). These relationships are analyzed to form field association logic. Then, the selected key medical terms, extracted diagnostic and treatment conclusions, and the analyzed field association logic are integrated to generate core semantic elements that represent the core information of the previous stage of the medical record.

[0129] Step S163: Perform semantic association analysis on the core semantic elements and the text content of all fields in the current staged medical record document. Identify fields in the current staged medical record document that have contradictory or missing associations with the core semantic elements. Retrieve the hierarchical semantic result content corresponding to the fields that have contradictory or missing associations with the core semantic elements in the dynamic mapping relationship. Combine the core semantic elements to re-convert and integrate the text to generate the corrected text content. Replace the original content with the corrected text content. Repeat the semantic association analysis and correction steps until all staged medical record documents have completed semantic inheritance processing. Integrate all corrected staged medical record documents to obtain the complete medical record document.

[0130] In the above application scenario, after generating the core semantic elements of the previous stage, semantic association analysis can be performed between these elements and the text content of all fields in the current staged medical record document. By comparing the core semantic elements with the current field text content, fields in the current staged medical record document that contradict the core semantic elements are identified. For example, the core semantic elements may state "the patient has no history of allergies," while the current field text content states "the patient is allergic to penicillin." Fields with missing associations are also identified, such as a test result mentioned in the core semantic elements, but no corresponding diagnostic analysis field in the current staged medical record document. For these fields, the corresponding hierarchical semantic results from the dynamic mapping relationship can be retrieved, and combined with the core semantic elements of the previous stage for text transformation and integration to generate corrected text content. The corrected text content replaces the corresponding original content in the current staged medical record document, eliminating semantic contradictions and gaps. This semantic association analysis and correction process is repeated for all staged medical record documents. Finally, all corrected staged medical record documents are integrated to form a complete and coherent medical record document.

[0131] Step S1631: Extract the field names and field text content of all filled fields in the current stage of the medical record document, construct a list of field content, and extract the key medical terms from the core semantic elements one by one to form a list of key terms.

[0132] In the above application scenario, this embodiment first obtains all fields in the current stage of the medical record document that have been filled with content, extracts the field names and corresponding field text content of these fields, and organizes them into a field content list for easy subsequent analysis. At the same time, key medical terms in the core semantic elements are extracted one by one to form a key term list, which includes the core medical terms from the previous stage of the medical record.

[0133] Step S1632: Traverse the text content of each field in the field content list, break down each field text content into multiple lexical units, compare each lexical unit of the field text content with the terms in the key term list one by one, and record the matching lexical units.

[0134] In the above application scenario, the text content of each field in the field content list can be traversed. For each field text content, it is broken down into multiple independent lexical units, each of which can be a single word or phrase. Then, these lexical units are compared one by one with the terms in the key term list to check for matching lexical units. If matching lexical units are found, it indicates that the current field text content is related to the key medical terms in the core semantic elements, and these matching lexical units are recorded.

[0135] Step S1633: Count the number of consistent lexical units in the text content of each field, calculate the proportion of consistent lexical units to the total number of terms in the key term list, mark fields with a proportion exceeding a set threshold as lexical consistent fields, and mark fields with a proportion below the set threshold as lexical missing fields.

[0136] In the above application scenario, the number of consistent lexical units recorded in the text content of each field is counted. Then, the proportion of this number to the total number of terms in the keyword list is calculated. The calculated proportion is compared with a set threshold. If the proportion exceeds the set threshold, it means that the text content of the field contains a sufficient number of key medical terms and is closely related to the vocabulary of the core semantic elements, and is marked as a lexical consistent field. If the proportion is lower than the set threshold, it means that the text content of the field is missing a lot of key medical terms, and is marked as a lexical missing field.

[0137] Step S1634: Extract the diagnosis and treatment conclusion statements from the core semantic elements to form a list of conclusion statements. Traverse the text content of each field in the field content list and identify statements that contain judgmental and conclusive statements to form a list of field conclusions.

[0138] In the above application scenario, all diagnostic and treatment conclusion statements are extracted from the core semantic elements and organized into a conclusion statement list. Then, the text content of each field in the field content list is traversed to identify statements containing judgmental or conclusive statements, such as "diagnosed as pneumonia" or "surgery is recommended," and these statements are organized into a field conclusion list.

[0139] Step S1635: Compare each statement in the conclusion statement list with each statement in the field conclusion list sentence by sentence, analyze the logical relationship between the two. If the statements have the same meaning or there is a causal support relationship, mark them as logically consistent. If there is a logical conflict, mark them as contradictory fields. If there is no relationship, mark them as disconnected fields.

[0140] In the above application scenario, this embodiment compares each diagnosis and treatment conclusion in the conclusion statement list with each statement in the field conclusion list sentence by sentence. The logical relationship between the two is analyzed. If the two statements have the same meaning or a causal relationship exists (e.g., "lung infection" in the conclusion statement list and "diagnosed as pneumonia" in the field conclusion list), the field is marked as logically consistent. If the two statements conflict logically (e.g., "no history of hypertension" in the conclusion statement list and "has hypertension" in the field conclusion list), the field is marked as contradictory. If there is no correlation between the two statements (e.g., "high blood sugar" in the conclusion statement list and "normal vision" in the field conclusion list), the field is marked as disjointed.

[0141] Step S1636: Analyze the field association logic in the core semantic elements and the field association relationship in the current stage of the medical record document. If the current field association relationship is completely consistent with the field association logic of the previous stage or there is a reasonable extension, it is marked as logically compatible. If there is a logical break or rule inconsistency, it is marked as a logical deviation field.

[0142] In the above application scenarios, we can analyze the field association logic of the previous stage contained in the core semantic elements, and simultaneously analyze the field association relationships between the fields in the current stage of the medical record document. Comparing the two, if the current field association relationship is completely consistent with the previous stage's field association logic, or if it is a reasonable extension of that logic (e.g., the previous stage's field association logic was "symptom A leads to examination B," and the current stage's field association relationship is "the result of examination B supports diagnosis C"), this is a reasonable extension and is marked as logically compatible. If the current field association relationship has a logical break from the previous stage's field association logic (e.g., the previous stage mentioned symptom A, but the current stage has no related examination or diagnosis field association); or if there is a rule inconsistency, such as violating conventional medical logical relationships, it is marked as a logically biased field.

[0143] Step S1637: Integrate the marking results of the missing vocabulary field, contradictory field, disjointed field, and logical deviation field, record the previous stage core semantic element content, current field text content, and specific differences of each of the missing vocabulary field, contradictory field, disjointed field, and logical deviation field, and determine the fields that are contradictory or missing to the core semantic elements based on the marking results.

[0144] In the above application scenarios, the relevant information of the marked missing vocabulary fields, contradictory fields, disjointed fields, and logically biased fields can be integrated. For each type of field, the content of the corresponding core semantic element in the previous stage, the text content of the current field, and the specific manifestations of the differences between the two are recorded. For example, for contradictory fields, the description of the diagnosis and treatment conclusion in the core semantic element in the previous stage, the concluding statement of the current field, and the specific content of the contradiction between the two are recorded. Based on the above integrated marking results, it is clearly determined which fields are contradictory or missing from the core semantic element.

[0145] For example, step S16371: Create a tagging result integration table, which includes columns such as field name, tag type, core semantic element content of the previous stage, current field text content, and specific differences.

[0146] In the above application scenario, a tagging result integration table can be created to organize information on various tagged fields. This tagging result integration table contains multiple columns: a field name column to record the name of the tagged field; a tag type column to record the tag type of the field, such as missing words, logical contradictions, etc.; a previous stage core semantic element content column to record the previous stage core semantic element content related to this field; a current field text content column to record the current text content of this field; and a difference details column to record the specific differences between the previous stage core semantic element content and the current field text content.

[0147] Step S16372: Fill the field name marked as a missing vocabulary field into the field name column of the tagging result integration table, fill in the missing vocabulary field in the tagging type column, fill in the list of key terms corresponding to the missing vocabulary field in the front-end core semantic element content column, fill in the text content of the missing vocabulary field in the current field text content column, and fill in the missing key term name of the missing vocabulary field in the difference specific manifestation column.

[0148] In the above application scenario, for fields marked as missing words, their field names can be entered into the field name column of the tagging result integration table. Enter "missing words" in the tagging type column. Enter the content of the key term list corresponding to the missing words field in the previous stage core semantic element content column; that is, the key medical terms related to this field in the previous stage core semantic elements. Enter the current text content of the missing words field in the current field text content column. Enter the name of the missing key term in the specific difference column, clearly indicating which key terms are not reflected in the current field text content.

[0149] Step S16373: Fill in the field name column of the labeling result integration table with the field name marked as contradictory field, fill in the logical contradiction in the labeling type column, fill in the diagnosis and treatment conclusion statement corresponding to the contradictory field in the front core semantic element content column, fill in the conclusive statement of the contradictory field in the current field text content column, and fill in the specific content of the contradiction in the contradictory field in the specific manifestation of difference column.

[0150] In the above application scenario, for fields marked as contradictory fields, their field names should be entered into the Field Name column of the Marking Result Integration Table. Enter "Logical Contradiction" in the Marking Type column. Enter the diagnosis and treatment conclusion statement from the previous stage's core semantic elements in the Previous Stage Core Semantic Element Content column, corresponding to the contradictory field. Enter the concluding statement from the contradictory field in the Current Field Text Content column. Describe the specific content of the contradiction in the Specific Differences column, such as "The previous stage conclusion was no allergy, while the current field states an allergy to penicillin."

[0151] Step S16374: Enter the field name of the field marked as a disconnected field into the field name column of the mark result integration table, enter "logical disconnect" in the mark type column, enter the diagnosis and treatment conclusion statement corresponding to the disconnected field in the front core semantic element content column, enter the conclusive statement of the disconnected field in the current field text content column, and enter the specific explanation of the lack of correlation between the disconnected fields in the difference specific manifestation column.

[0152] In the above application scenario, for fields marked as disconnected fields, enter their field names in the Field Name column of the Marking Results Integration Table. Enter "Logical Disconnection" in the Marking Type column. Enter the diagnosis and treatment conclusion statement from the previous stage's core semantic elements in the Front-End Core Semantic Element Content column. Enter the concluding statement from the disconnected field in the Current Field Text Content column. Explain in the Specific Differences column the specific situations where the concluding statement of this field is unrelated to the diagnosis and treatment conclusion statement from the previous stage, such as "The previous stage conclusion was abnormal blood sugar, while the current field statement is normal vision examination; the two are unrelated."

[0153] Step S16375: Fill in the field name column of the tagging result integration table with the field name marked as logical deviation field, fill in logical deviation in the tagging type column, fill in the field association logic content corresponding to the logical deviation field in the front-end core semantic element content column, fill in the relationship description of the logical deviation field in the current field text content column, and fill in the specific situation of logical deviation in the logical deviation field in the difference specific manifestation column.

[0154] In the above application scenario, for fields marked as logical deviation fields, their field names should be entered into the field name column of the marking result integration table. Enter "Logical Deviation" in the marking type column. Enter the field association logic from the previous stage's core semantic elements in the front-end core semantic element content column for this logical deviation field. Enter the description of the field association relationship in the current field text content column for this logical deviation field. Describe the specific circumstances of the logical deviation in the difference details column, such as "The previous stage field association logic was symptom A leading to examination B, while the current field association relationship is described as examination B leading to symptom A; the logic is reversed."

[0155] Step S16376: If there are missing rows in the tagging result integration table, complete them so that the information in each tagging field reflects the association with the core semantic elements.

[0156] In the above application scenarios, the tagging result integration table can be checked to see if any columns in the row records are missing. If any missing content is found, such as a column specifying the differences in a certain field not being filled in, it can be supplemented in a timely manner to ensure that the information in each tagging field fully reflects its association with the core semantic elements without any omissions.

[0157] Step S16377: Based on the tag types in the tag results integration table, filter out fields with tag types of missing words, logical contradictions, logical disconnects, and logical deviations. These fields with tag types of missing words, logical contradictions, logical disconnects, and logical deviations are considered as fields that are contradictory or missing to the core semantic elements.

[0158] In the above application scenario, the fields can be filtered based on the tag type column in the tagging results integration table. Fields with tag types of missing words, logical contradictions, logical gaps, and logical deviations will be filtered out. These filtered fields are those that contradict or are missing from the core semantic elements and require further correction.

[0159] Step S16378: Organize the selected field names that are contradictory or missing from the core semantic elements into a field list, and indicate the tag type and specific differences of each field that is contradictory or missing from the core semantic elements in the field list.

[0160] In the above application scenario, the names of fields that contradict or are missing from the core semantic elements are collected and compiled into a field list. This list indicates the corresponding tag type for each field, such as missing words, logical contradictions, etc., as well as the specific manifestations of the differences. This allows subsequent correction work to clearly understand the problems that need to be corrected for each field.

[0161] Step S1638: Retrieve the hierarchical semantic results corresponding to fields in the dynamic mapping relationship that have contradictions or missing associations with the core semantic elements, and combine the key medical terms, diagnosis and treatment conclusions and field association logic in the core semantic elements to redetermine the expression style and elements of the text conversion.

[0162] In the above application scenario, this embodiment retrieves the hierarchical semantic result content corresponding to the fields with contradictory or missing associations in the dynamic mapping relationship based on the field list. Then, combining the key medical terms, diagnosis and treatment conclusions, and field association logic in the core semantic elements, it redetermines the expression style to be used when converting the above hierarchical semantic result content into text, such as whether it needs to be more formal or detailed; and the expression elements that should be included, such as the key terms that must be mentioned and the logical relationships that need to be reflected, in order to prepare for generating the corrected text content.

[0163] Step S164: Upload the complete medical record document to the hospital's electronic medical record system for structured storage and complete the automatic medical record filling operation.

[0164] In the above application scenario, once the complete medical record document is obtained, it can be uploaded to the hospital's electronic medical record system via the hospital's internal network. Upon receiving the complete medical record document, the hospital's electronic medical record system processes the document according to its internal structured storage requirements, storing each field of information in the document into the corresponding database table to achieve structured storage, facilitating subsequent queries, statistics, and management operations. At this point, the entire automatic medical record filling operation is complete.

[0165] In the entire application scenario described above, the automatic medical record filling process involves the collection and processing of sensitive data such as patient medical record information. To protect patient privacy and prevent data leakage, this embodiment employs various privacy protection and anti-leakage technologies. For example, during data transmission, an encrypted transmission protocol is used to encrypt transmitted voice and text data, ensuring that the data is not stolen or tampered with during transmission. Regarding data storage, the stored medical record data undergoes anonymization processing, removing or replacing sensitive personal information of patients, such as names and ID numbers. Simultaneously, strict access control is implemented for the database, allowing only authorized medical personnel to access the relevant medical record data. During data processing, operations involving privacy data are logged for auditing and tracking purposes, ensuring the security of sensitive patient privacy data.

[0166] In one exemplary embodiment, a voice-based automatic medical record filling system is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2As shown, the voice-input-based automatic medical record filling system includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a voice-input-based automatic medical record filling method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the casing of a voice-based automatic medical record filling system, or an external keyboard, touchpad, or mouse, etc.

[0167] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for automatically filling in medical records based on voice input, characterized in that, The method includes: During the diagnosis and treatment process, the voice acquisition device receives medical staff’s voice input data related to the medical record in stages, and simultaneously obtains the real-time diagnosis and treatment stage identifier corresponding to the diagnosis and treatment process. Based on the real-time diagnosis and treatment stage identifier, the clinical scene feature library is retrieved, and scene-related features that match the real-time diagnosis and treatment stage identifier are extracted from the clinical scene feature library to construct a staged speech processing model. The medical record-related voice input data is input into a staged voice processing model, and semantic hierarchical processing is performed to generate hierarchical semantic results containing surface semantic information, deep semantic information and scene-related semantic information. A preset medical record field library is retrieved, and a dynamic mapping relationship between hierarchical semantic results and medical record fields is established. The dynamic mapping relationship is adjusted as the real-time diagnosis and treatment stage identifier changes. Based on the dynamic mapping relationship, the hierarchical semantic results are converted into text content that meets the requirements of the corresponding medical record fields, filled into the target fields of the preset medical record template, and a staged medical record document is generated. Semantic inheritance processing is performed on the staged medical record documents corresponding to different real-time diagnosis and treatment stages to correct the semantic gaps between fields caused by stage switching, so as to obtain complete medical record documents. The complete medical record documents are then uploaded to the hospital's electronic medical record system for structured storage, and the automatic medical record filling operation is completed.

2. The method for automatically filling medical records based on voice input according to claim 1, characterized in that, The process of receiving medical record-related voice input data from medical personnel in stages through a voice acquisition device during the diagnosis and treatment process, and simultaneously obtaining the real-time diagnosis and treatment stage identifier corresponding to the process, includes: During the diagnosis and treatment process, the voice acquisition device receives medical staff’s voice input data related to medical records in batches according to time nodes, and marks each batch of voice data with a corresponding collection timestamp, so that the collection period can be traced through the collection timestamp of each batch of voice data. The real-time diagnosis and treatment stage identifiers synchronized with the collection timestamp are obtained through the hospital diagnosis and treatment management system interface. The real-time diagnosis and treatment stage identifiers include the consultation stage identifier, examination stage identifier, diagnosis stage identifier, treatment plan formulation stage identifier, and medical order stage identifier. Through the time synchronization mechanism, the collection timestamp and the time recorded by the hospital diagnosis and treatment management system are based on the same time base, realizing the correspondence between the collection timestamp and the real-time diagnosis and treatment stage identifier. The hospital's diagnosis and treatment management system sends a query request for an identifier containing a collection timestamp through its stage identifier query capability. After receiving the identifier query request, the system retrieves the diagnosis and treatment operation record corresponding to the collection timestamp, extracts the type of diagnosis and treatment operation performed by the medical personnel at the time point from the diagnosis and treatment operation record, determines the corresponding real-time diagnosis and treatment stage identifier based on the diagnosis and treatment operation type, and ensures a one-to-one correspondence between the diagnosis and treatment operation type and the real-time diagnosis and treatment stage identifier. The determined real-time diagnosis and treatment stage identifier is then fed back to the relevant voice processing process. The step of retrieving a clinical scene feature library based on a real-time diagnosis and treatment stage identifier, extracting scene-related features matching the real-time diagnosis and treatment stage identifier from the clinical scene feature library, and constructing a staged speech processing model includes: The real-time diagnosis and treatment stage identifier is used as an index to retrieve the clinical scenario feature library. The scenario feature data stored in the clinical scenario feature library is traversed, and the target scenario feature data whose stage identifier label is consistent with the index is selected. The set of commonly used medical terms, semantic expression logic rules and field association priority of diagnosis and treatment stages are extracted from the target scenario feature data. All scenario feature data contain the corresponding stage identifier label. If the target scenario feature data has missing content, relevant knowledge is obtained from the associated clinical guideline database to generate the missing set of commonly used medical terms, semantic expression logic rules or field association priorities, and these are added to the target scenario feature data so that the supplemented target scenario feature data fully includes the set of commonly used medical terms, semantic expression logic rules and field association priorities. The basic structure of a staged speech processing model is constructed, which includes a semantic input part, a feature matching part, and a semantic output part. A set of commonly used medical terms is imported into the semantic input part as the term matching benchmark for speech data. Semantic expression logic rules are configured in the feature matching part as the logical basis for semantic analysis. Field association priority is set in the semantic output part as the priority standard for semantic results and field mapping. Historical speech processing case data with stage identifiers consistent with the current real-time diagnosis and treatment stage identifiers are selected and input into the basic structure of the staged speech processing model. The model processing effect is optimized by adjusting the term matching parameters of the semantic input part, the logical analysis parameters of the feature matching part, and the priority ranking parameters of the semantic output part. When the semantic parsing accuracy of the staged speech processing model on historical speech processing case data reaches the set standard, the final staged speech processing model is determined, so that the staged speech processing model can process speech data according to the scene association features corresponding to the real-time diagnosis and treatment stage identifier.

3. The method for automatically filling medical records based on voice input according to claim 1, characterized in that, The process involves inputting medical record-related speech data into a staged speech processing model, performing semantic layering processing, and generating a layered semantic result containing surface semantic information, deep semantic information, and scene-related semantic information, including: The medical record-related voice input data is input into the semantic input part of the staged voice processing model, and term matching is performed with the set of commonly used medical terms. The medical terms and expressions directly corresponding to the medical record-related voice input data are extracted as surface semantic information, so that the surface semantic information covers the direct medical expressions in the voice data. The surface semantic information is transmitted to the feature matching part of the staged speech processing model. Based on the semantic expression logic rules of the feature matching part, semantic reasoning is performed on the surface semantic information to mine the diagnosis and treatment related information hidden behind the surface semantic information. The mined information is used as deep semantic information so that the deep semantic information reflects the diagnosis and treatment related content that is not directly reflected in the surface semantic information. Based on the semantic expression logic rules, the semantic adaptation dimension corresponding to the current real-time diagnosis and treatment stage identifier is determined. The deep semantic information is divided into multiple semantic sub-modules according to the semantic adaptation dimension. Each semantic sub-module is matched with the scene attribute tag corresponding to the real-time diagnosis and treatment stage identifier. Each semantic sub-module contains only the expression content belonging to the same semantic adaptation dimension. If a semantic submodule cannot be matched with the corresponding scene attribute tag, the standard tag corresponding to the semantic adaptation dimension is retrieved from the clinical guideline database and assigned as a supplementary scene attribute tag to the semantic submodule. The semantic submodules with added scene attribute tags are combined to form scene-related semantic information. The scene attribute tags are pre-bound to the real-time diagnosis and treatment stage identifier. A hierarchical semantic association matrix is ​​constructed, with the surface semantic information as the base layer data, the deep semantic information as the intermediate layer data, and the scene association semantic information as the top layer data. The surface semantic information is split into multiple independent semantic representation units and filled into the base layer, the deep semantic information is split into multiple reasoning result units and filled into the intermediate layer, and the scene association semantic information is split into multiple scene label units and filled into the top layer. The basic layer semantic representation units and intermediate layer reasoning result units that are interconnected in the hierarchical semantic association matrix are compared to identify the correspondence of core words and determine the semantic mapping type. The semantic mapping type and core word pairs are recorded as association nodes in the corresponding cells of the hierarchical semantic association matrix. The scene binding relationship between intermediate layer data and top layer data is analyzed and the binding nodes are marked. The hierarchical semantic association matrix and semantic information of each layer are integrated to form a hierarchical semantic result. The process of retrieving a preset medical record field library and establishing a dynamic mapping relationship between hierarchical semantic results and medical record fields, wherein the dynamic mapping relationship is adjusted according to changes in the real-time diagnosis and treatment stage identifier, includes: The preset medical record field library is retrieved, and candidate medical record fields in the preset medical record field library whose stage adaptation range includes the current real-time diagnosis and treatment stage identifier are selected according to the current real-time diagnosis and treatment stage identifier. The semantic reception standard of each candidate medical record field is extracted and the scene-related semantic information in the hierarchical semantic results is used to calculate the adaptation degree. The preset medical record field library contains the field identifier, field function description, stage adaptation range and semantic reception standard of all medical record fields. Based on the fit calculation results, each candidate medical record field is assigned a corresponding hierarchical semantic result content. The correspondence between the candidate medical record fields and the hierarchical semantic result content, as well as the fit value, are recorded to form an initial mapping relationship. When the real-time diagnosis and treatment stage identifier is switched, the candidate medical record fields are re-selected, the fit is calculated and the content is assigned, and the initial mapping relationship is updated to form a dynamic mapping relationship.

4. The method for automatically filling medical records based on voice input according to claim 1, characterized in that, Based on the dynamic mapping relationship, the hierarchical semantic results are converted into text content that meets the requirements of the corresponding medical record fields, filled into the target fields of the preset medical record template, and a staged medical record document is generated, including: The dynamic mapping relationship is traversed to obtain the hierarchical semantic result content corresponding to each candidate medical record field. The hierarchical semantic result content is then processed by text conversion, converting the surface semantic information into the direct expression text required by the field, the deep semantic information into the reasoning supplementary text required by the field, and the scene association semantic information into the scene attribute annotation text required by the field. The field content requirements for each candidate medical record field are obtained from the preset medical record field library to determine the expression style, expression elements and expression format of the field content. The semantic information related to the voice content of each candidate medical record field is analyzed by vocabulary to identify core expression words and auxiliary expression words. Based on the field content requirements of the candidate medical record fields, the colloquial core expressions are converted into standard medical terms. The auxiliary expressions are then filtered, retaining the parts relevant to the field content requirements and removing the irrelevant parts. The converted core expressions and the filtered auxiliary expressions are combined according to grammatical rules to form a preliminary text expression. The frequency change characteristics are analyzed by referring to the semantic information related to the speech intonation of the candidate medical record fields. The tone intensity of the preliminary text expression is adjusted according to the frequency change characteristics. The distribution of pause intervals is analyzed by referring to the semantic information related to speech pauses. The sentence separation and logical pause positions of the preliminary text expression are adjusted according to the distribution of pause intervals. The adjusted preliminary text expression and the scene attribute annotation text are integrated to form complete text content. Retrieve a preset medical record template, locate the corresponding target field in the preset medical record template according to the field identifier of the candidate medical record field, fill the target field with the complete text content, repeat the conversion, integration and filling steps until all candidate medical record fields corresponding to the current real-time diagnosis and treatment stage identifier are filled, and generate the staged medical record document corresponding to the real-time diagnosis and treatment stage identifier. The semantic inheritance processing of staged medical record documents corresponding to different real-time diagnosis and treatment stage identifiers corrects semantic gaps between fields caused by stage switching, resulting in a complete medical record document, including: When the real-time diagnosis and treatment stage identifier is switched and a new staged medical record document is generated, words that appear more frequently than the set value and belong to the corresponding stage are selected from all fields of the previous staged medical record document as key medical terms. Statements containing judgment and conclusion are identified from the field content and the core conclusion part is extracted as the diagnosis and treatment conclusion statement. Analyze the reference relationships, causal relationships and supplementary relationships between different fields in the previous stage of medical record documents to form field association logic, and integrate key medical terms, diagnosis and treatment conclusions and field association logic to generate the core semantic elements of the previous stage; Semantic association analysis is performed between the core semantic elements and the text content of all fields in the current staged medical record document. Fields in the current staged medical record document that have contradictory or missing associations with the core semantic elements are identified. The hierarchical semantic results corresponding to the fields with contradictory or missing associations with the core semantic elements in the dynamic mapping relationship are retrieved. The text is then re-converted and integrated with the core semantic elements to generate corrected text content. The corrected text content replaces the original content. The semantic association analysis and correction steps are repeated until all staged medical record documents have completed semantic inheritance processing. All corrected staged medical record documents are then integrated to obtain a complete medical record document.

5. The method for automatically filling in medical records based on voice input according to claim 2, characterized in that, The process involves sending a query request containing a collection timestamp through the hospital's treatment management system's stage identifier query capability. Upon receiving the query request, the system retrieves the treatment operation record corresponding to the collection timestamp, extracts the type of treatment operation performed by the medical personnel at that time point from the treatment operation record, and determines the corresponding real-time treatment stage identifier based on the treatment operation type. This includes: Parse the identifier query request containing the collection timestamp to determine the time retrieval range. The time retrieval range is the interval between the collection timestamp and the preset time. Retrieve all medical operation record entries within the time retrieval range in the medical operation record database. For each medical procedure record, the time precision is adjusted. The operation time of the record is compared with the collection timestamp. Target record entries with a difference of less than a set threshold are selected. If there are multiple target record entries, the record entry corresponding to the highest priority medical procedure is selected as the final target entry. The medical procedure type field, which records the name of the specific medical procedure performed by the medical personnel, is extracted from the final target entry. Retrieve the preset operation-stage mapping relationship, which stores the correspondence between all diagnosis and treatment operation types and real-time diagnosis and treatment stage identifiers. Each diagnosis and treatment operation type uniquely corresponds to one real-time diagnosis and treatment stage identifier. In the operation-stage mapping relationship, find the entry that matches the extracted diagnosis and treatment operation type, and obtain the real-time diagnosis and treatment stage identifier corresponding to the entry; If the obtained real-time diagnosis and treatment stage identifier belongs to the set of valid stage identifiers that includes the consultation stage identifier, examination stage identifier, diagnosis stage identifier, treatment plan formulation stage identifier, and medical order stage identifier, then the real-time diagnosis and treatment stage identifier is used as the query result; if the obtained real-time diagnosis and treatment stage identifier does not belong to the set of valid stage identifiers that includes the consultation stage identifier, examination stage identifier, diagnosis stage identifier, treatment plan formulation stage identifier, and medical order stage identifier, then an operation type confirmation request is sent to the medical personnel terminal. After receiving confirmation of the diagnosis and treatment operation type from the medical personnel's terminal, the system searches for an entry that matches the confirmed diagnosis and treatment operation type in the operation-stage mapping relationship and obtains the corresponding real-time diagnosis and treatment stage identifier. The confirmed real-time diagnosis and treatment stage identifier is associated with the collection timestamp and stored together, so that the corresponding real-time diagnosis and treatment stage identifier can be quickly matched by the collection timestamp when called in subsequent times, so that each collection timestamp corresponds to a unique real-time diagnosis and treatment stage identifier.

6. The method for automatically filling medical records based on voice input according to claim 4, characterized in that, The step involves performing semantic association analysis between the core semantic elements and the text content of all fields in the current staged medical record document. This identifies fields in the current staged medical record document that have contradictory or missing associations with the core semantic elements. The step also retrieves the hierarchical semantic results corresponding to these fields in the dynamic mapping relationship, including: Extract the field names and text content of all filled fields in the current stage of the medical record document, construct a list of field content, and extract the key medical terms from the core semantic elements to form a list of key terms. Iterate through the text content of each field in the field content list, break down each field text content into multiple lexical units, compare each lexical unit of the field text content with the terms in the key term list one by one, and record the matching lexical units. The number of consistent lexical units in the text content of each field is counted, and the proportion of consistent lexical units to the total number of terms in the key term list is calculated. Fields with a proportion exceeding a set threshold are marked as lexical consistent fields, and fields with a proportion below the set threshold are marked as lexical missing fields. Extract the diagnosis and treatment conclusion statements from the core semantic elements to form a list of conclusion statements; traverse the text content of each field in the field content list and identify statements that contain judgmental and conclusive statements to form a list of field conclusions. Each statement in the conclusion statement list is compared with each statement in the field conclusion list sentence by sentence to analyze the logical relationship between the two. If the statements have the same meaning or there is a causal support relationship, they are marked as logically consistent. If there is a logical conflict, they are marked as contradictory fields. If there is no relationship, they are marked as disconnected fields. Analyze the field association logic in the core semantic elements and the field association relationship in the current stage of the medical record document. If the current field association relationship is completely consistent with the field association logic of the previous stage or there is a reasonable extension, it is marked as logically compatible. If there is a logical break or rule inconsistency, it is marked as a logical deviation field. The labeling results of missing vocabulary fields, contradictory fields, disjointed fields, and logical deviation fields are integrated. The core semantic element content of each of the missing vocabulary fields, contradictory fields, disjointed fields, and logical deviation fields in the previous stage, the current field text content, and the specific manifestations of differences are recorded. Based on the labeling results, fields that are contradictory or missing from the core semantic elements are identified. Retrieve the hierarchical semantic results corresponding to fields in the dynamic mapping relationship that have contradictions or missing associations with the core semantic elements, and combine them with the key medical terms, diagnosis and treatment conclusions and field association logic in the core semantic elements to redetermine the expression style and elements of text conversion.

7. The method for automatically filling medical records based on voice input according to claim 5, characterized in that, The process involves adjusting the time precision of each diagnostic and treatment operation record entry, comparing the operation time of the record entry with the collection timestamp, and filtering out target record entries where the difference between the operation time and the collection timestamp is less than a set threshold. If multiple target record entries exist, the record entry corresponding to the highest priority treatment operation is selected as the final target entry, including: Obtain the operation time for each medical operation record entry, convert the operation time to the same time format as the collection timestamp, calculate the difference between the operation time and the collection timestamp for each medical operation record entry, and record the medical operation record entry corresponding to each difference; The calculated difference is compared with a set threshold. The medical record entries with a difference less than the set threshold are retained, and the retained medical record entries are used as target record entries. If there is only one target record entry, that target record entry is directly used as the final target entry; if there are multiple target record entries, a preset list of treatment operation priorities is retrieved; the priority list of treatment operation priorities sets the priority of each treatment operation according to the order and importance of the treatment process, with the priority from high to low corresponding to different treatment operation types. Match the treatment operation type in each target record entry with the treatment operation priority list to determine the priority level of each target record entry; Select the target record entry with the highest priority level. If there are multiple target record entries with the same priority level, compare the difference between the operation time and the collection timestamp of the multiple target record entries with the same priority level, and select the target record entry with the smallest difference. The final selected target record entry is determined as the final target entry, and the diagnosis and treatment operation type field is extracted from the final target entry.

8. A medical record auto-filling system based on voice input, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the voice-based medical record autofill method according to any one of claims 1 to 7 by executing the machine-executable instructions.

Images