A voice recognition-based nursing document automatic generation system

Abstract

This invention discloses an automatic nursing documentation generation system based on speech recognition. The system includes a speech acquisition module, a medical-specific speech recognition engine module, a structured nursing documentation parsing module, a system integration and adaptation module, a data security and compliance management module, and a terminal interaction module. This invention achieves automated, voice-driven generation of medical and nursing documentation throughout the entire process by adapting speech acquisition to all clinical scenarios, optimizing speech recognition specifically for nursing, and automatically generating structured nursing documentation. It can be seamlessly integrated into existing hospital medical document writing systems, bedside nursing cart systems, and PDA mobile nursing terminal systems as a standalone app or functional plugin, allowing for rapid deployment without modifying the original system architecture. This solves the core pain point of tedious medical document writing for frontline clinical medical staff, which consumes more than half of the consultation time, improving nursing documentation writing efficiency by over 80% and significantly reducing the time spent on non-medical documentation by medical staff.

Description

Technical Field

[0001] This invention relates to the fields of medical information technology and intelligent voice technology, specifically to an automatic nursing document generation system based on speech recognition. Background Technology

[0002] With the continuous advancement of medical informatization, electronic medical record systems have been fully implemented in medical institutions at all levels across the country. Nursing documents, as a core component of legally binding medical documents, are an objective record of the entire clinical nursing process and a core basis for medical quality control, evidence in medical disputes, and clinical nursing research. Their writing quality and timeliness are directly related to medical safety and the quality of clinical nursing services.

[0003] Currently, clinical nursing documentation is predominantly done manually. After completing bedside nursing procedures, nurses must return to the nursing station to manually enter a large amount of documentation into the electronic medical record system. This task consumes an average of over 60% of nurses' daily working hours, and in departments with high documentation volumes such as intensive care, emergency rooms, and operating rooms, this percentage can exceed 70%. Nurses spend a significant amount of their core work time on document entry, drastically reducing the time available for direct patient-facing clinical care, patient observation, and health services, thus failing to meet core clinical needs. Manual entry cannot achieve real-time writing after bedside procedures, requiring post-procedure completion, which is highly susceptible to problems such as memory bias, data errors, and time discrepancies, posing serious risks to medical safety and potential medical disputes. Existing general-purpose speech-to-text tools are not specifically optimized for clinical nursing scenarios. They cannot accurately recognize specialized content such as nursing terminology, generic drug names, and disease diagnosis codes, with a terminology recognition accuracy rate of less than 80%. They can only transcribe plain text and cannot automatically match and generate structured nursing documents. Nurses still need to manually complete field entry, formatting, and verification, failing to fundamentally reduce the burden of paperwork. Existing solutions cannot seamlessly integrate with existing hospital medical document writing systems, bedside nursing cart systems, and PDA mobile nursing terminals, requiring large-scale modifications to the original system architecture. This results in high deployment costs and long cycles. Furthermore, the lack of a full-process compliance management mechanism for medical data makes it difficult to meet the security management requirements of sensitive medical data and hinders compliant application in clinical scenarios.

[0004] In summary, existing nursing documentation technologies cannot simultaneously address multiple pain points such as low efficiency in clinical document entry, heavy workload for medical staff, poor adaptability to clinical scenarios, insufficient system integration compatibility, and lack of data security and compliance. They also fail to achieve the core goal of returning the working time of medical staff to patients. Therefore, the proposed automatic nursing documentation generation system based on speech recognition is particularly important. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a voice recognition-based automatic nursing document generation system. This system achieves automated voice-driven generation of medical and nursing documents throughout the entire process through a complete technical solution: voice acquisition adapted to all clinical scenarios, voice recognition optimized for nursing, automatic generation of structured nursing documents, seamless integration with multiple systems, full-process data compliance management, and multi-terminal interaction. It can be directly integrated as a standalone app or functional plugin into existing hospital medical document writing systems, bedside nursing cart systems, and PDA mobile nursing terminal systems without requiring modifications to the original system architecture. This completely solves the core pain point of tedious medical document writing for frontline medical staff, which consumes more than half of the consultation time. It improves nursing document writing efficiency by over 80%, significantly reduces the time spent on non-medical documentation by medical staff, truly realizing the core goal of returning the working time of medical workers to patients, while ensuring the real-time, accuracy, and compliance of medical document writing, fully adapting to the work needs of all clinical diagnosis and nursing scenarios.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an automatic nursing document generation system based on speech recognition, the system comprising the following components: The voice acquisition module is used to adapt to the audio acquisition of nursing scenarios on multiple terminals, complete the environmental noise suppression and purification processing of the target voice signal, and output a standardized audio stream. The medical-specific speech recognition engine module is connected to the speech acquisition module and is used to process standardized audio streams into text based on a nursing-specific fine-tuned speech recognition model. It also combines a nursing professional terminology dictionary to complete the error correction and normalization of the recognition results and output standardized nursing text. The nursing document structured parsing module is connected to the medical-specific speech recognition engine module. It is used to build a nursing document template library that conforms to national standards. It uses a natural language understanding model to extract entities, identify intents and match fields in standardized nursing texts to automatically generate structured nursing documents. The system interface and adaptation module communicates with the nursing document structure parsing module and is used to achieve seamless integration with the hospital's existing information systems and terminal devices based on the medical industry standard interface, so as to complete the automatic synchronization and data interaction of structured nursing documents. The data security compliance management module is connected to the other modules to realize encryption, operation tracking, access control and data desensitization of voice data and document data throughout the entire process. The terminal interaction module communicates with the other modules to provide a human-computer interaction interface for multiple terminals, enabling voice command control, real-time display of recognition results, document fine-tuning, and one-click submission.

[0007] Furthermore, the voice acquisition module supports multi-channel audio acquisition from bedside nursing carts, PDA mobile nursing terminals, nurse station workstations, smart wearable devices, and sterile voice terminals in operating rooms. The audio acquisition adopts a standardized 16kHz 16bit mono format. It has a built-in processing unit combining beamforming and adaptive filtering algorithms based on multi-microphone arrays. It preloads corresponding environmental noise models for different clinical scenarios such as wards, ICUs, operating rooms, and emergency rooms, filtering background noise such as medical equipment operation sounds and conversation sounds, to achieve purification of the target voice signal. The signal-to-noise ratio is improved by no less than 30dB. It supports keyword voice wake-up and contactless one-click start / stop acquisition functions in sterile scenarios. It supports continuous long voice acquisition and real-time audio stream slicing processing, and supports automatic separation and filtering of non-target human voices, ensuring the accuracy and adaptability of voice acquisition in all clinical scenarios.

[0008] Furthermore, the medical-specific speech recognition engine module incorporates a nursing-specific speech recognition model. The model training process employs an adaptive weighted cross-entropy loss function to optimize parameters. The expression for this loss function is: ,in This is the value of the adaptive weighted cross-entropy loss function, used for loss calculation and parameter iterative optimization during model training; This represents the total number of tokens within a single training batch. This is the index of the currently calculated token sequence; For the first The real label of each token, with a value of 1 indicating that the corresponding token is the target content, and a value of 0 indicating that it is not the target content; For the model to the first Output the predicted probability of each token; For the first The weight coefficient of nursing terminology corresponding to each token is derived from the clinical importance level of the nursing professional terminology dictionary and the frequency of occurrence in the million-level nursing corpus. The importance level is divided into 1-5 levels. Core nursing operation terms, generic drug names, and disease diagnosis terms correspond to level 5, with a weight value range of 2.5-3.0. Common descriptive words correspond to level 1, with a fixed weight of 1.0. To adaptively adjust the weights for accents, the weights are derived from the recognition accuracy of the corresponding accent corpus during model training. When the recognition accuracy of a single accent term is below 95%, the weights are increased proportionally, with a maximum value of 2.0. When the recognition accuracy is above 98%, the weights are fixed at 1.0.

[0009] Furthermore, the nursing document structured parsing module incorporates a natural language understanding model. This model employs a nursing-specific hierarchical adaptive attention mechanism to extract core entities. The calculation expression for this attention mechanism is as follows: ,in An adaptive attention weight output for nursing-specific levels, used for accurate extraction of core entities and field matching in nursing texts; The query matrix for the attention mechanism is derived from the word embedding vectors of the input nursing text; The key matrix for the attention mechanism is derived from the word embedding vectors of the input nursing text; The value matrix for the attention mechanism is derived from the word embedding vectors of the input nursing text; The dimension of the key matrix is ​​512, which is the same as the dimension of the word embedding vector. This is a weighted matrix for nursing field hierarchy. The weights are derived from the mandatory level of fields in the nursing documentation writing standards and the priority of clinical nursing quality control. Mandatory core fields include patient unique identifier, recording time, vital signs, nursing interventions, and doctor's order execution status, with corresponding weight values ​​ranging from 1.8 to 2.2. Non-mandatory supplementary fields have a fixed weight of 1.0. The matrix dimensions are... The output dimensions are a perfect match.

[0010] Furthermore, the structured parsing module for nursing documents includes a built-in template library that conforms to national nursing document writing standards. The template library covers general nursing records, critical patient nursing records, temperature charts, medical order execution records, intake and output records, health education records, surgical nursing records, admission nursing assessment forms, and discharge nursing guidance forms. It is also compatible with specialty nursing templates from different departments such as internal medicine, surgery, obstetrics and gynecology, pediatrics, ICU, operating room, and emergency. Through a natural language understanding model, it completes entity extraction, intent recognition, and field matching of standardized nursing texts, automatically associates the patient's unique identification information to complete basic information filling, automatically calibrates the recording time to keep it synchronized with the hospital system time, and automatically matches the standardized terminology corresponding to nursing operations, realizing one-click generation of different types of nursing documents and cross-document data linkage updates.

[0011] Furthermore, the structured parsing module for nursing documents incorporates a data verification unit. This unit connects to the hospital's medical order system and the patient's electronic medical record system, performing multi-dimensional rationality checks on the extracted structured fields. These checks include consistency between the nursing level / diet type and the medical order content, matching of vital sign values ​​with clinically reasonable ranges, logical rationality of intake and output statistics, and suitability of nursing measures to the patient's diagnosis. When discrepancies exist between the identified content and the clinically standardized medical order information, a real-time warning is issued on the terminal interface, along with annotations of the discrepancy fields and reasons. This assists nursing staff in quickly correcting the content, ensuring the accuracy and compliance of nursing documents.

[0012] Furthermore, the system integration module supports the HL7FHIR medical information interaction standard interface and is compatible with the WebServiceRESTfulAPI general interface. It seamlessly integrates with existing hospital systems such as HIS, LIS, PACS, electronic medical record, EMR nursing information, NIS bedside nursing cart, PDA mobile nursing terminal, surgical anesthesia, and intensive care information systems. It supports a plug-in deployment mode and can run as an independent APP on mobile and fixed terminals of Android, iOS, and Windows systems, or be embedded as a functional plug-in into existing nursing-related systems in the hospital without modifying the original system architecture and database structure. The deployment cycle does not exceed 7 working days. It supports real-time synchronization and bidirectional interaction of cross-system data, enabling one-time generation of document data for sharing across the entire system.

[0013] Furthermore, the data security compliance management module employs the national cryptographic SM4 algorithm to segment and encrypt the data throughout the entire process of voice acquisition, transmission, storage, recognition, and document generation, ensuring secure transmission. It tracks all operations related to voice acquisition, document generation, modification, submission, and review, recording the operator's time, content, and device terminal IP address information for full traceability. Based on the hospital's RBAC role-based access control system, it configures document operation permissions for different personnel, enabling hierarchical permission management and real-time interception of unauthorized operations. It automatically desensitizes sensitive personal information such as patient names, ID numbers, contact information, home addresses, and medical history, complying with relevant regulations such as the Personal Information Protection Law and the Electronic Medical Record Application Management Standards. It also supports automatic data backup and disaster recovery functions, ensuring the integrity and security of medical data.

[0014] Furthermore, the terminal interaction module has a built-in offline processing unit. In clinical scenarios with no network or a network signal strength below -85dBm, it completes the local generation and encrypted temporary storage of offline speech recognition documents. Once the network signal recovers to above -60dBm, it automatically synchronizes the document data to the corresponding hospital information system. It supports breakpoint resume function to avoid data loss caused by network interruption. It provides a human-computer interaction interface adapted to multiple terminals, realizes voice command control, real-time display of recognition results, document fine-tuning and one-click submission functions, supports voice command to complete the modification, deletion and confirmation of document content, adapts to contactless full-process operation in aseptic scenarios, supports supplementary editing modes of handwriting input and keyboard input, and adapts to the usage needs of nursing staff with different operating habits.

[0015] Furthermore, the system has been specifically adapted to the nursing characteristics of different clinical departments, including internal medicine, surgery, obstetrics and gynecology, pediatrics, ICU, operating room, emergency, and rehabilitation. For the ICU, it enables automatic generation of hourly nursing records and automatic 24-hour intake and output statistics. For the operating room, it enables contactless voice operation in a sterile environment and automatic verification of surgical instruments. For the pediatrics department, it enables linkage and matching of children's growth and development indicators with nursing records. For the obstetrics and gynecology department, it enables automatic generation and archiving of perinatal nursing records throughout the entire process. It adapts to the clinical nursing workflow and document writing requirements of different departments, achieving full coverage of clinical departments.

[0016] Compared with existing technologies, the speech recognition-based automatic nursing document generation system of the present invention has the following advantages: This invention provides a complete technical solution for the automated generation of medical and nursing documents through voice-driven processes. This solution includes voice acquisition adapted to all clinical scenarios, voice recognition optimized for nursing, automatic generation of structured nursing documents, seamless integration with multiple systems, full-process data compliance management, and multi-terminal interaction. It can be directly integrated as an independent APP or functional plug-in into existing hospital medical document writing systems, bedside nursing cart systems, and PDA mobile nursing terminal systems without modifying the original system architecture. It completely solves the core pain point of tedious medical document writing for frontline medical staff, which consumes more than half of the consultation time. It improves the efficiency of nursing document writing by more than 80%, significantly reduces the time spent by medical staff on non-medical documents, and truly achieves the core goal of returning the working time of medical staff to patients. At the same time, it ensures the real-time, accuracy, and compliance of medical document writing, and is fully adapted to the work needs of all clinical diagnosis and treatment nursing scenarios.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the overall closed-loop operation of the present invention. Figure 2 This is a flowchart of the nursing-specific speech recognition and document structuring intelligent generation process of the present invention; Figure 3 This is a flowchart illustrating the clinical scenario adaptation and end-to-end data security and compliance management process of the present invention. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0020] Example 1

[0021] This embodiment is applied to the general ICU ward of a tertiary hospital. It is designed for critically ill patients with endotracheal intubation and mechanical ventilation. It automatically generates nursing documents for the entire process of bedside nursing operations, adapts to the specific needs of ICU hourly nursing records and automatic 24-hour intake and output statistics. The execution environment is an ICU ward with background noise from medical equipment such as ventilators, monitors, and infusion pumps. The operation terminal is a bedside nursing cart PDA mobile nursing terminal.

[0022] Nursing staff wear smart wearable voice terminals adapted for ICU scenarios and activate the system's voice acquisition function via the bedside nursing cart PDA mobile nursing terminal. The system's voice acquisition module enables multi-channel audio acquisition, using a standardized 16kHz 16bit mono format for audio acquisition. It calls upon the built-in processing unit that combines beamforming and adaptive filtering algorithms based on a multi-microphone array, pre-loads an ICU-specific environmental noise model, and filters background noise in real time, such as the operating sounds of medical equipment like ventilators, monitors, and infusion pumps, as well as conversations in the ward environment. This completes the purification and processing of the target voice signal, outputting a standardized audio stream. Throughout the process, the data security and compliance management module uses the national cryptographic SM4 algorithm to segmentally encrypt and securely transmit the data during voice acquisition and transmission. Simultaneously, the entire voice acquisition operation is logged, recording the operator, operation time, operating terminal, and IP address information.

[0023] The medical-specific speech recognition engine module receives a standardized audio stream and calls the built-in nursing-specific speech recognition model. The model's training process uses an adaptive weighted cross-entropy loss function to optimize parameters. The expression for this loss function is as follows: ,in The value of the adaptive weighted cross-entropy loss function; This represents the total number of tokens within a single training batch. This is the index of the currently calculated token sequence; For the first The true label of a token; For the model to the first Output the predicted probability of each token; For the first The nursing terminology weight coefficient corresponding to each token; To adaptively adjust the weights based on accent, the system performs speech-to-text processing on the standardized audio stream. Simultaneously, it incorporates a built-in nursing terminology dictionary to correct and normalize the recognition results, converting colloquial expressions into standardized nursing text that conforms to nursing standards. The data security and compliance management module simultaneously encrypts and tracks the data during the recognition process, and verifies the current nursing staff's operating permissions based on the hospital's RBAC role and permission system.

[0024] The nursing document structured parsing module receives standardized nursing text and calls a built-in natural language understanding model that employs a nursing-specific hierarchical adaptive attention mechanism. The calculation expression for this attention mechanism is as follows: ,in Adaptive attention weight output for nursing-specific levels; The query matrix for the attention mechanism; The key matrix for the attention mechanism; The value matrix for the attention mechanism; Let be the dimension of the key matrix; The system uses a hierarchical weight matrix for nursing fields to perform entity extraction, intent recognition, and field matching on standardized nursing text. It also calls upon built-in templates for critically ill patient nursing records and intake / output records that conform to national nursing documentation standards, adapts to ICU specialty nursing templates, automatically associates patient unique identifiers to complete basic information filling, automatically calibrates record time to maintain synchronization with the hospital system time, automatically matches standardized terminology corresponding to nursing operations, and simultaneously completes automatic 24-hour intake / output statistics and hourly nursing record field filling. The built-in data verification unit synchronously connects to the hospital's medical order system and the patient's electronic medical record system to perform multi-dimensional rationality verification on the extracted structured fields, including consistency verification of nursing level, diet type, and medical order content; matching verification of vital sign values ​​with clinically reasonable ranges; logical rationality verification of intake / output statistics; and adaptation verification of nursing measures to patient diagnosis. When discrepancies exist between the identified content and medical order information or clinical standards, a real-time warning is issued on the PDA terminal interface, along with annotations of the discrepancy fields and reasons, assisting nursing staff in quickly correcting the content.

[0025] The system integration and adaptation module uses the HL7FHIR medical information interaction standard interface, while also being compatible with the RESTful API general interface. It seamlessly integrates with the hospital's existing HIS system, electronic medical record (EMR) system, nursing information NIS system, intensive care information system, bedside nursing cart system, and PDA mobile nursing terminal system. It automatically synchronizes the verified structured critical patient nursing records and intake / output records to the corresponding hospital information systems, completing data interaction and archiving. In response to network signal fluctuations in the ICU ward, the offline processing unit of the terminal interaction module monitors the network signal strength in real time. When the network signal strength is below -85dBm, it automatically completes offline voice recognition, local document generation, and encrypted temporary storage. Once the network signal recovers to above -60dBm, it automatically synchronizes the document data to the corresponding hospital information system through the breakpoint resume function.

[0026] Nursing staff can view recognition results and automatically generated nursing documents through the human-computer interaction interface of the PDA mobile nursing terminal. They can modify, delete, and confirm the document content through voice commands, adapting to the contactless full-process operation in the ICU aseptic operation scenario. It also supports supplementary editing modes for handwriting input and keyboard input. After fine-tuning the document, it can be submitted for review with one click. The data security and compliance management module keeps a complete record of the entire process of document generation, modification, submission, and review. It automatically desensitizes sensitive personal information such as patient name, ID number, and medical history information. Based on the RBAC role-based access control system, it configures the document viewing and approval permissions of corresponding reviewers, and finally completes the automatic generation and compliant archiving of nursing documents for critically ill ICU patients.

[0027] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A nursing document automatic generation system based on speech recognition, characterized in that, The system includes the following components: Voice acquisition module: used to acquire audio in nursing scenarios that are compatible with multiple terminals, complete environmental noise suppression and purification of target voice signals, and output standardized audio streams; Medical-specific speech recognition engine module: It communicates with the speech acquisition module and is used to process standardized audio streams into text based on a nursing-specific fine-tuned speech recognition model. It also combines a nursing professional terminology dictionary to complete the error correction and normalization of the recognition results and output standardized nursing text. Nursing document structured parsing module: It communicates and connects with the medical-specific speech recognition engine module to build a nursing document template library that conforms to national standards. It uses a natural language understanding model to extract entities, recognize intent and match fields in standardized nursing texts to automatically generate structured nursing documents. System integration and adaptation module: It communicates with the nursing document structure parsing module to achieve seamless integration with the hospital's existing information systems and terminal devices based on the medical industry standard interface, and completes the automatic synchronization and data interaction of structured nursing documents; Data security compliance management module: It communicates and connects with the other modules to realize encryption, operation tracking, access control and data desensitization of voice data and document data throughout the entire process; Terminal interaction module: It communicates and connects with the other modules to provide a human-computer interaction interface for multiple terminals, enabling voice command control, real-time display of recognition results, document fine-tuning, and one-click submission.

2. The automatic nursing document generation system based on speech recognition according to claim 1, characterized in that, The voice acquisition module supports multi-channel audio acquisition from bedside nursing carts, PDA mobile nursing terminals, nurse station workstations, smart wearable devices, and sterile voice terminals in operating rooms. The audio acquisition adopts a standardized 16kHz 16bit mono format and has a built-in processing unit that combines beamforming algorithms and adaptive filtering algorithms based on multi-microphone arrays. It preloads corresponding environmental noise models for different clinical scenarios such as wards, ICUs, operating rooms, and emergency rooms to filter background noise such as medical equipment operation sounds and conversation sounds.

3. The automatic nursing document generation system based on speech recognition according to claim 1, characterized in that, The medical-specific speech recognition engine module incorporates a nursing-specific speech recognition model. The model training process employs an adaptive weighted cross-entropy loss function to optimize parameters. The expression for this loss function is: ,in The value of the adaptive weighted cross-entropy loss function; This represents the total number of tokens within a single training batch. This is the index of the currently calculated token sequence; For the first The true label of a token; For the model to the first Output the predicted probability of each token; For the first The nursing terminology weight coefficient corresponding to each token; The weights are adjusted adaptively for accents.

4. The automatic nursing document generation system based on speech recognition according to claim 1, characterized in that, The structured parsing module for nursing documents incorporates a natural language understanding model. This model employs a nursing-specific hierarchical adaptive attention mechanism to extract core entities. The calculation expression for this attention mechanism is as follows: ,in Adaptive attention weight output for nursing-specific levels; The query matrix for the attention mechanism; The key matrix for the attention mechanism; The value matrix for the attention mechanism; Let be the dimension of the key matrix; This is the weight matrix for the nursing field hierarchy.

5. The automatic nursing document generation system based on speech recognition according to claim 1, characterized in that, The structured parsing module for nursing documents includes a built-in template library that conforms to national nursing document writing standards. The template library covers general nursing records, critical patient nursing records, temperature charts, medical order execution records, intake and output records, health education records, surgical nursing records, admission nursing assessment forms, and discharge nursing instructions. It is also compatible with specialized nursing templates from different departments such as internal medicine, surgery, obstetrics and gynecology, pediatrics, ICU, operating room, and emergency. Through a natural language understanding model, it completes entity extraction, intent recognition, and field matching of standardized nursing texts, automatically associates the patient's unique identification information to complete basic information filling, automatically calibrates the recording time to keep it synchronized with the hospital system time, and automatically matches the standardized terminology corresponding to nursing operations.

6. The automatic nursing document generation system based on speech recognition according to claim 1, characterized in that, The structured parsing module for nursing documents has a built-in data verification unit. This unit connects to the hospital's medical order system and the patient's electronic medical record system to perform multi-dimensional rationality verification on the extracted structured fields. This includes verifying the consistency between the nursing level, diet type, and medical order content; verifying the matching of vital sign values ​​with the clinically reasonable range; verifying the logical rationality of intake and output statistics; and verifying the suitability of nursing measures with the patient's diagnosis. When there is a deviation between the identified content and the clinical norms of the medical order information, a warning is issued in real time on the terminal interface, and the deviation field and the reason for the deviation are marked to assist nursing staff in quickly correcting the content.

7. The automatic nursing document generation system based on speech recognition according to claim 1, characterized in that, The system integration module supports the HL7FHIR medical information interaction standard interface and is also compatible with the WebServiceRESTfulAPI general interface. It seamlessly integrates with the hospital's existing HIS system, LIS system, PACS system, electronic medical record system, EMR nursing information system, NIS bedside nursing cart system, PDA mobile nursing terminal system, surgical anesthesia system, and intensive care information system. It supports a plug-in deployment mode and can run as an independent APP on Android, iOS, and Windows systems on mobile and fixed terminals, or be embedded as a functional plug-in into the hospital's existing nursing-related systems.

8. The automatic nursing document generation system based on speech recognition according to claim 1, characterized in that, The data security and compliance management module uses the national cryptographic SM4 algorithm to segment and encrypt the data throughout the entire process of voice acquisition, transmission, storage, recognition, and document generation. It leaves a complete record of all voice acquisition, document generation, modification, submission, and review operations, including the operator's operation time, operation content, and device terminal IP address information. Based on the hospital's RBAC role-based access control system, it configures document operation permissions for different positions and automatically desensitizes sensitive personal information such as patient names, ID numbers, contact information, home addresses, and medical history.

9. The automatic nursing document generation system based on speech recognition according to claim 1, characterized in that, The terminal interaction module has a built-in offline processing unit. In clinical scenarios with no network or a network signal strength below -85dBm, it completes the local generation and encrypted temporary storage of offline speech recognition documents. Once the network signal recovers to above -60dBm, it automatically synchronizes the document data to the corresponding hospital information system. It supports breakpoint resume function, provides a human-computer interaction interface adapted to multiple terminals, realizes voice command control, real-time display of recognition results, document fine-tuning and one-click submission functions, supports voice command to complete the modification, deletion and confirmation of document content, adapts to contactless full-process operation in aseptic scenarios, supports supplementary editing modes of handwriting input and keyboard input, and adapts to the usage needs of nursing staff with different operating habits.

10. The automatic nursing document generation system based on speech recognition according to claim 1, characterized in that, This system is specifically adapted to the nursing characteristics of different clinical departments, including internal medicine, surgery, obstetrics and gynecology, pediatrics, ICU, operating room, emergency, and rehabilitation. For the ICU, it enables automatic generation of hourly nursing records and automatic 24-hour intake and output statistics. For the operating room, it enables contactless voice operation in a sterile environment and automatic verification of surgical instruments. For the pediatrics department, it enables linkage and matching of children's growth and development indicators with nursing records. For the obstetrics and gynecology department, it enables the automatic generation and archiving of perinatal nursing records throughout the entire process, adapting to the clinical nursing workflow and documentation requirements of different departments.

Images