Medical field recording and analyzing system

Abstract

The invention relates to a medical field record analysis system, and the system comprises a user authentication module which is used for user login verification, authority distribution and login abnormity processing; the voice processing module is used for collecting, preprocessing, transferring and structured storage of medical field voice; the AI intelligent analysis module is used for analyzing the voice transfer text, extracting key medical information and automatically filling the standard report template with the key medical information, and is connected with the voice processing module; the data management module is used for editing and deleting the report, maintaining the version history of the report and realizing cross-device data synchronization, and is connected with the AI intelligent analysis module; the PDF export module is used for rendering the structured report into a PDF file, performing encryption export and multi-channel sharing, and is connected with the data management module; and the central control module is used for coordinating the operation of each module, monitoring the system state and processing abnormal events. The method has the beneficial effects that the accuracy of speech recognition can be improved, and the accuracy of analysis decision and the system applicability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and medical emergency technology, and particularly relates to a medical scene record analysis system. BACKGROUND

[0002] Voice recognition technology is widely used in outpatient scenarios, which can convert doctors' real-time dictation into structured text, greatly improving outpatient efficiency, quickly converting voice into text, and automatically generating medical records, greatly shortening the time of writing medical records. Doctors do not need to record while asking for medical advice, can maintain eye contact and have more meaningful interactions with patients, making patients feel more attention and improving patient satisfaction.

[0003] In the prior art, in medical scenes such as emergency and surgery, the traditional manual recording method has the problems of low efficiency and easy omission of key information, which may affect the treatment effect of patients and subsequent medical processes, and cannot meet the real-time and accuracy requirements in emergency scenarios. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a medical scene record analysis system, which can improve the accuracy of voice recognition and improve the accuracy of analysis and decision-making and the applicability of the system.

[0005] The technical scheme of the present application is as follows: a medical scene record analysis system, comprising:

[0006] A user authentication module for user login verification, permission allocation and login exception handling;

[0007] A voice processing module for collecting, preprocessing, transcribing and structuring storage of medical scene voice;

[0008] An AI intelligent analysis module for analyzing voice transcription text, extracting key medical information, automatically filling into a standardized report template, and connecting the voice processing module;

[0009] A data management module for editing, deleting and maintaining report version history, and realizing cross-device data synchronization, and connecting the AI intelligent analysis module;

[0010] A PDF export module for rendering structured reports into PDF files, encrypting and exporting, and sharing through multiple channels, and connecting the data management module;

[0011] A central control module for coordinating the operation of each module, monitoring system status and handling abnormal events, and connecting the user authentication module, voice processing module, AI intelligent analysis module, data management module and PDF export module.

[0012] As a preferred scheme of the present application, the user authentication module uses the SHA-256 algorithm to encrypt and store the password, and combines the RSA asymmetric encryption to transmit sensitive information, including:

[0013] An identity verification unit is configured to receive a user-inputted mobile phone number and password.

[0014] A permission management unit is configured to verify a user role.

[0015] An exception handling unit is configured to monitor the number of login failures, and trigger temporary locking when the number of login failures exceeds 3.

[0016] As a preferred scheme of the present application, the voice processing module includes:

[0017] A voice collection unit is configured to collect audio by calling a device microphone.

[0018] A noise reduction unit is configured to remove environmental noise by using a spectral subtraction method.

[0019] A voice-to-text unit is configured to convert voice to text based on a pre-trained model, and output structured text.

[0020] A text storage unit is configured to store the text data in a distributed database after encryption.

[0021] A translation unit is configured to translate the text.

[0022] As a preferred scheme of the present application, in the voice processing module, 300 milliseconds of environmental sound is collected before starting recording, the spectral distribution and decibel value of the noise are analyzed, a noise characteristic file is established, and parameters are adjusted according to the noise characteristic file.

[0023] In a high-noise environment, the sampling rate is increased to 48000 Hz, and a double-microphone beamforming technology is enabled to focus on human voice; in a low-noise environment, pre-emphasis filtering is enabled to enhance high-frequency human voice signals.

[0024] As a preferred scheme of the present application, in the voice processing module, after starting recording, an independent audio segment is generated every 30 seconds, each segment is named, temporarily stored in a local cache area, and the start and end time and verification information of each segment are recorded.

[0025] If recording is interrupted, the current unfinished segment less than 30 seconds is saved immediately, the interruption state is marked, and the interruption time is recorded.

[0026] When resuming recording, the last segment number of the previous recording is identified, and recording is resumed from the next segment number.

[0027] When identification is performed after recording is completed, all audio segments are spliced in the order of the segment number to generate a complete recording file.

[0028] As a preferred scheme of the present application, the AI intelligent analysis module comprises:

[0029] a text analysis unit, configured to perform entity recognition on the text;

[0030] an information extraction unit, configured to extract key information of the text;

[0031] a structured filling unit, configured to map the extracted information to preset report fields;

[0032] a risk assessment unit, configured to calculate a treatment risk score based on the emergency timeline data and assess the treatment risk;

[0033] a recommendation unit, configured to retrieve an optimal treatment scheme from a database based on the current emergency scene;

[0034] a feedback collection unit, configured to mark manual corrections as training samples.

[0035] As a preferred scheme of the present application, in the AI intelligent analysis module, after extracting the emergency measures, the relevant treatment guidelines are matched through the knowledge graph, the optimal treatment scheme is retrieved from the database based on the current emergency scene, the recommended results are displayed in the form of a pop-up window, and the evidence level is labeled.

[0036] As a preferred scheme of the present application, in the AI intelligent analysis module, the manual corrections of unidentified items are marked as training samples, an incremental learning algorithm is adopted, new samples are used to update the AI analysis model every week, the recognition accuracy before and after the update is compared, if the improvement is ≥5%, the new model is solidified, otherwise it is rolled back.

[0037] As a preferred scheme of the present application, the data management module comprises:

[0038] an editing unit, configured to provide a rich text editor;

[0039] a deletion unit, configured to mark a deletion time after receiving a deletion instruction;

[0040] a version control unit, configured to record the differences of each modification by using a Git-like incremental storage algorithm;

[0041] a data synchronization unit, configured to realize real-time synchronization of multi-terminal data through the MQTT protocol.

[0042] As a preferred scheme of the present application, the PDF export module comprises:

[0043] a template rendering unit, configured to map report data to a preset PDF template by using the XSL-FO technology;

[0044] a data verification unit, configured to detect unidentified fields, and if there are any, a prompt window is popped up;

[0045] Encryption unit, adding digital signatures to PDF files based on the SM2 national encryption algorithm;

[0046] The transmission unit calls the third-party interface through the HTTPS protocol to push the PDF file to the target.

[0047] The beneficial effects of the present invention are:

[0048] 1. In the present invention, the voice processing module can perform real-time multi-language translation to resolve language barriers in cross-border rescue or treatment of foreign patients; it can perform environmentally adaptive dynamic noise reduction and sampling rate adjustment, automatically optimize recording parameters based on noise differences in complex environments such as ambulances and operating rooms, and improve audio quality; it can record in segments and resume recording at breakpoints to prevent recording interruptions caused by equipment failure or misoperation, ensuring the integrity of audio data during the emergency process; it supports voice commands or gestures to trigger recording, solving the problem of doctors' hands being busy and unable to operate manually during emergency treatment, thereby improving the accuracy of voice recognition and system applicability;

[0049] 2. In this invention, the AI ​​intelligent analysis module can intelligently recommend first aid knowledge. The recommendation algorithm uses a combination of collaborative filtering and content matching to improve the accuracy of first aid decision-making. It also uses knowledge distillation technology for incremental learning to avoid catastrophic forgetting, ensure continuous optimization of the system, and improve the accuracy of analysis and decision-making and the applicability of the system.

[0050] 3. In the present invention, the data management module visualizes the dynamic timeline, and the timeline visualization supports touch-screen operation. Deviation analysis helps doctors review the emergency process and optimize subsequent operations. The PDF export module renders the structured report into a PDF file in the medical industry standard format, supports encrypted export and multi-channel sharing, and improves applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a schematic diagram of the system of the present invention;

[0052] Figure 2 It is a module diagram of the system of the present invention;

[0053] Figure 3 A user login diagram for the system of the present invention;

[0054] Figure 4 This is a voice recording diagram of the system of the present invention;

[0055] Figure 5 A report detail diagram of the system of the present invention;

[0056] In the figure: 1, user authentication module; 101, identity verification unit; 102, authority management unit; 103, exception handling unit; 2, voice processing module; 201, voice collection unit; 202, noise reduction unit; 203, voice-to-text unit; 204, text storage unit; 205, translation unit; 3, AI intelligent analysis module; 301, text analysis unit; 302, information extraction unit; 303, structured filling unit; 304, risk assessment unit; 305, recommendation unit; 306, feedback collection unit; 4, data management module; 401, editing unit; 402, deletion unit; 403, version control unit; 404, data synchronization unit; 405, timeline rendering unit; 5, PDF export module; 501, template rendering unit; 502, data verification unit; 503, encryption unit; 504, transmission unit; 6, central control module; 601, instruction scheduling unit; 602, state monitoring unit; 603, fault handling unit. DETAILED DESCRIPTION

[0057] To make the technical problems solved by the present application, the technical solutions adopted and the technical effects achieved more clear, the technical solutions of the embodiments of the present application will be further described in detail below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0058] Embodiment one:

[0059] As shown in Figures 1 to 5 , a medical scene record analysis system comprises:

[0060] A user authentication module 1 is used for user login verification, permission assignment and login exception handling, and is connected to a central control module 6.

[0061] A voice processing module 2 is used for collection, preprocessing, transcription and structured storage of medical scene voice, and is connected to an AI intelligent analysis module 3 and the central control module 6.

[0062] The AI intelligent analysis module 3 is used for analyzing voice transcription text, extracting key medical information, automatically filling into a standardized report template, and real-time assessment of emergency risk, and is connected to the voice processing module 2, a data management module 4 and the central control module 6.

[0063] The data management module 4 is used for editing, deleting and maintaining report version history, and realizing cross-device data synchronization, and is connected to the AI intelligent analysis module 3 and the central control module 6.

[0064] PDF export module 5 for rendering structured reports as PDF files, encrypted export and multi-channel sharing, connecting data management module 4 and central control module 6;

[0065] Central control module 6 for coordinating the operation of each module, monitoring system status and handling abnormal events, connecting user authentication module 1, voice processing module 2, AI intelligent analysis module 3, data management module 4 and PDF export module 5.

[0066] In this embodiment, the user authentication module 1 is responsible for user login verification, permission assignment and login exception handling, ensuring system access security. The user authentication module 1 consists of an identity verification unit 101, a permission management unit 102 and an exception handling unit 103, adopting a distributed identity authentication architecture. It is connected to the central control module 6 through the TLS1.3 encryption communication protocol for bidirectional connection, receiving login requests and returning verification results.

[0067] The identity verification unit 101 receives the user's input mobile phone number and password; the permission management unit 102 verifies the user's role, which includes doctors and administrators; the exception handling unit 103 monitors the number of login failures, triggering temporary lockout if it exceeds 3 times. The password is encrypted and stored using the SHA-256 algorithm, and sensitive information is transmitted using RSA asymmetric encryption to prevent data leakage.

[0068] In this embodiment, the voice processing module 2 is used to realize the collection, preprocessing, transcription and structured storage of medical field voice. The voice processing module 2 contains a voice collection unit 201, a noise reduction unit 202, a voice-to-text unit 203 and a text storage unit 204. The voice processing module 2 receives user recording instructions through the central control module 6 and transmits the converted text to the AI intelligent analysis module 3.

[0069] The voice collection unit 201 calls the device microphone to collect audio at a sampling rate of 44.1 kHz; the noise reduction unit 202 uses spectral subtraction to remove environmental noise; the voice-to-text unit 203 uses a Transformer-based pre-training model in the medical field to convert voice to text, outputting structured text; the text storage unit 204 stores the encrypted text data in a distributed database. The text data is encrypted using the AES-256 encryption algorithm, and the distributed database uses MongoDB. The medical field pre-training model is optimized through transfer learning, with a professional term recognition accuracy of over 98%; noise suppression uses adaptive threshold adjustment, suitable for complex environments such as ambulances and operating rooms.

[0070] The voice processing module 2 realizes real-time multilingual translation, solves the language barrier in cross-country rescue or foreign patient treatment. The voice collection unit 201 synchronously collects multilingual voice; the voice-to-text unit 203 calls a language detection model, such as an LSTM-based language classifier, to identify the current language type during transcription; the translation unit 205 uses a medical field customized Transformer translation model to translate non-Chinese text into Chinese or Chinese text into non-Chinese in real time, with a translation delay of <1s; the text storage unit 204 simultaneously saves the original text and the translated text, supporting switching between views.

[0071] The voice processing module 2 performs dynamic noise reduction and sampling rate adjustment according to the environment, automatically optimizes the recording parameters for the noise differences in ambulances, operating rooms, and other complex environments, and improves the audio quality. The system automatically collects 300 milliseconds of environmental sound before starting recording, analyzes the frequency spectrum distribution and decibel value of the noise, and establishes a noise characteristic file, such as low-frequency mechanical noise and high-frequency mixed human voice. Dynamically adjust the recording parameters: according to the noise characteristic file, automatically adjust the system parameters: for high-noise scenes, increase the sampling rate to 48000Hz, and enable double-microphone beamforming technology to focus on human voice. For low-noise scenes, automatically turn on pre-emphasis filtering to enhance high-frequency human voice signals.

[0072] The voice processing module 2 performs segmented recording and breakpoint recording to prevent recording interruptions caused by device failures or misoperations and ensure the integrity of the audio data during the emergency process. Audio segmentation is stored in real time: after starting recording, the system automatically generates an independent audio segment every 30 seconds, names each segment, temporarily stores it in the local cache area, and records the start and end times and verification information of each segment. The system monitors abnormalities in real time: if an interruption occurs, it immediately saves the unfinished segment that is less than 30 seconds, marks it as "interrupted" and records the interruption time. When the recording is resumed, the system automatically identifies the last segment number of the previous recording and continues recording from the next segment number, ensuring that the segment numbers are continuous and avoiding data duplication or omission. After clicking "End and Identify", the system automatically splices all audio segments in sequence according to the segment number to generate a complete recording file. After splicing, the system verifies the information to confirm that there is no data damage. If there is an anomaly, it automatically re-splices the abnormal segment.

[0073] The voice processing module 2 supports voice instruction or gesture triggered recording, solving the problem that doctors cannot manually operate due to being busy with hands in emergency treatment. The recording is awakened by voice instruction, and the system background continuously listens to the preset wake-up word "emergency recording". The wake-up delay is less than or equal to 300 milliseconds. After recognizing the wake-up word, a prompt sound is played at 500 Hz for 200 milliseconds to feedback "recording has been started". At the same time, recording is automatically started. When the recording is finished, the voice instruction "stop recording" is supported, and the system automatically executes the "end and identify" process. The recording is triggered by gesture recognition. The front camera is enabled to recognize the preset gesture in real time. For example, the palm is opened to start recording, and the fist is clenched to end recording. A lightweight gesture recognition model is used. When the recognition confidence is greater than or equal to 80%, the corresponding operation is triggered, and the operation result is fed back through the interface animation.

[0074] In the present embodiment, the AI intelligent analysis module 3 is used to analyze the voice transcription text, extract key medical information, automatically fill in the standardized report template, and real-time assess the emergency risk. The AI intelligent analysis module 3 is composed of a text analysis unit 301, an information extraction unit 302, a structured filling unit 303, and a risk assessment unit 304. The AI intelligent analysis module 3 receives the text data of the voice processing module 2, transmits the filled report to the data management module 4, and interacts with the central control module 6 to realize risk warning.

[0075] The text analysis unit 301 uses the BIO tagging method (Begin-Inside-Outside) to perform entity recognition on the text, such as "Wang Li (name), 25 years old (age)". The information extraction unit 302 extracts key information such as trauma events and vital signs through a rule engine. The rule engine is a semantic matching based on a medical ontology library. The structured filling unit 303 maps the extracted information to the preset report fields, and marks the unrecognizable information as "unrecognized". The risk assessment unit 304 calculates the treatment risk score based on the emergency timeline data using a Logistic regression model.

[0076] The AI intelligent analysis module 3 intelligently recommends emergency knowledge. The recommendation algorithm uses a combination of collaborative filtering and content matching, with a response time of less than 200 milliseconds, improving the accuracy of emergency decision-making. After extracting emergency measures, the information extraction unit 302 matches related treatment guidelines through a knowledge graph. The recommendation unit 305 retrieves the optimal treatment plan from the database based on the current emergency scene. The recommendation results are displayed in the form of a pop-up window, and the evidence level is labeled. The central control module 6 records the viewing and adoption of the recommended content.

[0077] The AI intelligent analysis module 3 adopts a knowledge distillation technology for incremental learning to avoid catastrophic forgetting, and the model updating takes less than 1 hour, ensuring continuous optimization of the system. The feedback collection unit 306 will mark the manual correction of the "unidentified" items by the doctor when editing the report as a training sample; an incremental learning algorithm is used to update the AI analysis model with new samples every week without the need to retrain the entire model; compare the recognition accuracy before and after updating, if the improvement is ≥5%, then the new model is solidified, otherwise it is rolled back; the central control module 6 records the model iteration log and traces the training data and effect of each version of the model.

[0078] In this embodiment, the data management module 4 supports user editing and deleting operations on the report, maintains the report version history, and realizes cross-device data synchronization. The data management module 4 includes an editing unit 401, a deletion unit 402, a version control unit 403, and a data synchronization unit 404. The data management module 4 interacts with the AI intelligent analysis module 3 to obtain the initial report, receives user operation instructions through the central control module 6, and synchronizes data to the cloud server.

[0079] The editing unit 401 provides a rich text editor to support modification of report fields, and the modification record is stored in the version log in real time; after receiving the deletion instruction, the deletion unit 402 marks the deletion time instead of physically deleting it, and supports data recovery within 30 days; the version control unit 403 uses a Git-like incremental storage algorithm to record the differences of each modification, such as "2025-07-1014:30 modify age from '25' to '26'"; the data synchronization unit 404 realizes real-time synchronization of multi-terminal data through the MQTT protocol to ensure data consistency. The editing operation adopts an optimistic lock mechanism to prevent concurrent conflicts; deletion recovery is based on timestamp indexing to achieve efficient data retrieval; cross-device synchronization uses incremental transmission, only transmitting the modified part to reduce bandwidth occupancy.

[0080] The data management module 4 visualizes the dynamic timeline, the timeline visualization supports touch screen operation, and the deviation analysis helps doctors review the emergency process and optimize subsequent operations.

[0081] The timeline rendering unit 405 sorts the emergency steps by timestamp, such as "00:01 tourniquet use" and "00:05 blood pressure measurement"; uses D3.js to draw an interactive timeline, supports zooming and clicking to view details; compares the actual emergency timeline with the standard process to calculate the deviation rate; if the deviation rate ≥30%, it will be automatically marked and prompted.

[0082] The calculation expression of the deviation rate is as follows:

[0083]

[0084] In the above formula, σ is the deviation rate, t is the actual time, and t0 is the standard time.

[0085] In this embodiment, the PDF export module 5 is used to render the structured report into a PDF file in the medical industry standard format, supporting encrypted export and multi-channel sharing. The PDF export module 5 is composed of a template rendering unit 501, a data verification unit 502, an encryption unit 503, and a transmission unit 504. The PDF export module 5 receives the final report data of the data management module 4, and calls the third-party interface to realize file transmission through the central control module 6.

[0086] The template rendering unit 501 uses XSL-FO technology to map the report data to the preset PDF template, supporting dynamic adjustment of table layout; the data verification unit 502 detects unrecognized fields, and if there are any, a prompt window is popped up; the encryption unit 503 adds a digital signature to the PDF file based on the SM2 national encryption algorithm, including user identity information and a timestamp, to prevent tampering; the transmission unit 504 calls the third-party interface through the HTTPS protocol to push the PDF file to the designated contact or hospital system.

[0087] In this embodiment, the central control module 6 is used to coordinate the operation of each module, monitor the system status and handle abnormal events. The central control module 6, as the core hub of the system, contains an instruction scheduling unit 601, a status monitoring unit 602, and a fault handling unit 603. The central control module 6 communicates with all other modules bidirectionally, and realizes asynchronous transmission of instructions through a message queue.

[0088] The instruction scheduling unit 601 receives user operation instructions, such as "start recording", and distributes them to the corresponding modules after analysis; the status monitoring unit 602 collects real-time running parameters of each module and visualizes them through a dashboard; the fault handling unit 603 adopts a fuse mechanism, automatically switching to a backup module and triggering an alarm when a module fails, and notifying the administrator through a short message. The instruction scheduling adopts a priority queue, for example, the priority of instructions in emergency scenes is higher than that of ordinary queries; the abnormality detection adopts a sliding window algorithm to monitor the response time of the module, and triggers the fuse when the threshold is exceeded.

[0089] Embodiment two:

[0090] An AI analysis method for medical field records, comprising the following steps:

[0091] S1, the user inputs the mobile phone number and password for login verification;

[0092] S2, record the voice and perform recognition, convert the audio to text;

[0093] S3, extract the key data of the text, fill the key data into the report template, calculate the emergency risk score, and if it is high risk, trigger the sound and light alarm;

[0094] In step S1, the user inputs a mobile phone number and a password for login verification, including the following steps:

[0095] S101, the user inputs a mobile phone number and a password, and submits a login request;

[0096] S102, the user authentication module 1 verifies the identity, and if the verification fails, prompts "mobile phone number or password error";

[0097] S103, after verification, the central control module 6 loads the user permission configuration, and enters the system homepage. In step S2, the voice is recorded and recognized, and the audio is converted into text, including the following steps:

[0098] S201, the user clicks "start recording", and the central control module 6 sends a recording instruction to the voice processing module 2;

[0099] S202, the voice collection unit 201 starts recording, and the noise reduction unit 202 processes the audio signal in real time;

[0100] S203, the user clicks "end and identify", and the voice processing module 2 calls the medical ASR model to convert the audio into text;

[0101] S204, the text storage unit 204 stores the encrypted text in the database, and returns the processing result to the central control module 6.

[0102] In step S3, the key data of the text is extracted, the key data is filled into the report template, the first aid risk score is calculated, and if it is high risk, the sound and light alarm is triggered, including the following steps:

[0103] S301, the AI intelligent analysis module 3 receives the text data, and the text analysis unit 301 performs entity recognition;

[0104] S302, the information extraction unit 302 extracts key data such as patient information and trauma events;

[0105] S303, the structured filling unit 303 fills the key data into the report template, and marks the unrecognizable items as "unrecognized";

[0106] S304, the risk assessment unit 304 calculates the first aid risk score, and if the score is greater than or equal to 80 points (high risk), the central control module 6 triggers the sound and light alarm.

[0107] The AI analysis method for medical scene recording of the embodiment further includes: S4, exporting the report.

[0108] In step S4, the report is exported, including the following steps:

[0109] S401, the user clicks "export PDF", and the PDF export module 5 checks the integrity of the report;

[0110] S402: If there are any unidentified items, a confirmation window will pop up, and the user can continue exporting after confirmation.

[0111] S403, the template rendering unit 501 generates a PDF file and adds a digital signature;

[0112] S404 , the transmission unit 504 calls a third-party interface via the HTTPS protocol to send the PDF to the designated recipient, and records a transmission log at the same time.

[0113] In the present invention, the voice processing module 2 is capable of real-time multilingual translation, resolving language barriers in cross-border rescue or treatment of foreign patients; it can perform environmentally adaptive dynamic noise reduction and sampling rate adjustment, automatically optimizing recording parameters based on noise differences in complex environments such as ambulances and operating rooms, and improving audio quality; it can record in segments and resume recording at breakpoints, preventing recording interruptions caused by equipment failure or misoperation, and ensuring the integrity of the audio data during the emergency process; it supports voice commands or gestures to trigger recording, solving the problem of doctors' busy hands unable to operate manually during emergency treatment, and improving the accuracy of voice recognition and system applicability; the AI ​​intelligent analysis module 3 can intelligently recommend first aid knowledge, and the recommendation algorithm uses a combination of collaborative filtering and content matching to improve the accuracy of first aid decisions; it uses knowledge distillation technology for incremental learning to avoid catastrophic forgetting, ensure continuous optimization of the system, and improve the accuracy of analytical decisions and system applicability; the data management module 4 visualizes the dynamic timeline, and the timeline visualization supports touch-screen operation. Deviation analysis helps doctors review the emergency process and optimize subsequent operations. The PDF export module 5 renders the structured report into a PDF file in the medical industry standard format, supports encrypted export and multi-channel sharing, and improves applicability. Through modular design, the entire medical field record process is intelligent. Creative expansion of functions solves industry pain points such as cross-language communication, data mutual recognition, and model self-optimization, significantly improving emergency treatment efficiency and data reliability. It is suitable for various medical scenarios such as ambulances and operating rooms.

[0114] Although preferred embodiments of the present invention have been described, further changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are understood. It is apparent that various changes and modifications may be made to the present invention by those skilled in the art without departing from the spirit and scope of the present invention. Thus, the present invention is intended to encompass such changes and modifications as fall within the scope of equivalents of the present invention.

Claims

1. A medical field record analysis system, characterized in that: include: User authentication module, used for user login verification, authority allocation and login exception handling; Speech processing module, used for the collection, preprocessing, transcription and structured storage of speech at medical sites; AI intelligent analysis module, used to parse speech transcription text, extract key medical information, automatically fill in standardized report templates, and connect to the speech processing module; The data management module is used to edit and delete reports, maintain report version history, synchronize data across devices, and connect to the AI ​​intelligent analysis module; PDF export module, used to render structured reports into PDF files, export encrypted files and share them across multiple channels, and connect to the data management module; The central control module is used to coordinate the operation of each module, monitor the system status and handle abnormal events, and connect the user authentication module, voice processing module, AI intelligent analysis module, data management module and PDF export module.

2. A medical on-site record analysis system according to claim 1, characterized in that: The user authentication module uses the SHA-256 algorithm to encrypt and store passwords, combined with RSA asymmetric encryption to transmit sensitive information, including: Authentication unit, used to receive the mobile phone number and password entered by the user; Permission management unit, used to verify user roles; The exception handling unit is used to monitor the number of login failures and trigger a temporary lock if the number exceeds 3.

3. A medical on-site record analysis system according to claim 1, characterized in that: The speech processing module includes: The voice collection unit is used to call the device microphone to collect audio; Noise reduction unit, which uses spectral subtraction to remove ambient noise; The speech-to-text unit converts speech to text based on a pre-trained model and outputs structured text; The text storage unit is used to encrypt the text data and store it in the distributed database; the translation unit is used to translate the text.

4. A medical on-site record analysis system according to claim 3, characterized in that: In the voice processing module, before starting recording, 300 milliseconds of ambient sound is collected, the noise spectrum distribution and decibel value are analyzed, a noise feature file is created, and parameters are adjusted based on the noise feature file: In high-noise scenarios, the sampling rate is increased to 48,000 Hz, and dual-microphone beamforming technology is enabled to focus on the human voice; in low-noise scenarios, pre-emphasis filtering is turned on to enhance high-frequency human voice signals.

5. A medical on-site record analysis system according to claim 3, characterized in that: In the voice processing module, after recording is started, an independent audio segment is generated every 30 seconds. Each segment is named and temporarily stored in the local cache. The start and end times and verification information of each segment are recorded. If the recording is interrupted, the unfinished segment of less than 30 seconds is immediately saved, the interruption status is marked, and the interruption time is recorded. When resuming recording, the last segment number of the previous recording is identified and recording is continued from the next segment number; when identification is performed after the recording is completed, all audio segments are spliced ​​in the order of segment numbers to generate a complete recording file.

6. A medical on-site record analysis system according to claim 1, characterized in that: AI intelligent analysis modules include: Text parsing unit, used to perform entity recognition on text; An information extraction unit, used to extract key information from text; Structured filler cells for mapping extracted information to pre-set report fields; The risk assessment unit calculates the treatment risk score based on the emergency timeline data and assesses the treatment risk; Recommendation unit, which retrieves the optimal treatment plan from the database based on the current emergency scenario; The feedback collection unit is used to mark the manually corrected items as training samples.

7. A medical on-site record analysis system according to claim 6, characterized in that: In the AI ​​intelligent analysis module, after extracting first aid measures, relevant treatment guidelines are matched through the knowledge graph. Based on the current first aid scenario, the optimal treatment plan is retrieved from the database, and the recommended results are displayed in a pop-up window, with the evidence level marked.

8. A medical on-site record analysis system according to claim 6, characterized in that: In the AI ​​intelligent analysis module, manual corrections of unrecognized items are marked as training samples. An incremental learning algorithm is used to update the AI ​​analysis model with new samples every week. The recognition accuracy before and after the update is compared. If the improvement is ≥5%, the new model is solidified, otherwise it is rolled back.

9. The medical on-site record analysis system according to claim 1, characterized in that: The data management module includes: Editing unit, used to provide a rich text editor; A deletion unit, used to mark the deletion time after receiving the deletion instruction; The version control unit uses a Git-like incremental storage algorithm to record the differences of each modification; The data synchronization unit is used to achieve real-time synchronization of multi-terminal data through the MQTT protocol.

10. The medical on-site record analysis system according to claim 1, characterized in that: PDF export modules include: Template rendering unit, which uses XSL-FO technology to map report data to preset PDF templates; Data verification unit, used to detect unrecognized fields and pop up a prompt window if any exist; Encryption unit, which adds digital signatures to PDF files based on the SM2 national encryption algorithm; The transmission unit calls the third-party interface through the HTTPS protocol to push the PDF file to the target.

Images