A method, system and device for generating intelligent ward round medical records

Abstract

This invention provides an intelligent ward round medical record generation method, system, and device, relating to the field of medical text generation technology. The method includes: collecting multi-source data during ward rounds; the multi-source data includes voice data, image text, and structured data; processing the multi-source data separately to obtain input vectors; constructing a knowledge base, including medical knowledge and a hybrid retrieval strategy; training a large-scale medical record generation model; based on the RAG method, obtaining retrieval results from the knowledge base according to query information in the input vector, then concatenating these results with the query information to obtain context, and submitting this context to the large-scale medical record generation model to generate an initial medical record. This solution, through model fine-tuning, enables the general-purpose large-scale model to possess medical professional knowledge, medical record writing standards, and doctor-patient dialogue understanding capabilities, and can accurately identify medical record structure and medical terminology. It also solves the problems of knowledge lag and insufficient professionalism in large-scale models, providing real-time and authoritative medical references for medical record generation.

Description

Technical Field

[0001] This invention relates to the field of medical text generation technology, and in particular to a method, system and device for generating intelligent ward round medical records. Background Technology

[0002] Ward rounds are a core part of doctors' daily clinical work, but the heavy workload of writing medical records after ward rounds often consumes a significant amount of doctors' time, thus reducing their work efficiency. Existing electronic medical record systems mostly rely on structured templates for data entry, still requiring doctors to perform a large amount of manual work, resulting in low efficiency. In recent years, Large Language Models (LLM) have been attempted for medical text generation, but the following pain points remain: Hallucination problem: The model may fabricate non-existent diagnostic or medication information; Knowledge lag: The model cannot obtain the latest treatment guidelines in real time; Data heterogeneity: Ward rounds involve multiple sources such as voice, images, and structured data, which are difficult to process uniformly; Conflict between personalization and standardization: The generated medical records are often too template-based and do not conform to the doctor's personal habits. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent ward round medical record generation method, system, and device to solve at least one of the aforementioned technical problems existing in the prior art.

[0004] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an intelligent ward round medical record generation method, comprising the following steps: Step 1: Collect multi-source data during ward rounds; the multi-source data includes voice data, image text, and structured data.

[0005] In one feasible implementation, the voice data refers to the doctor-patient dialogue content collected through a recording device, including chief complaints, consultations, and medical orders; The image text refers to paper reports and medical records from other hospitals collected through a camera scanning device. The structured data refers to basic patient information (such as identity and demographic information, hospitalization information and medical insurance information), test results and hospital history records collected through hospital management systems (such as HIS system, LIS system or PACS system).

[0006] Step 2: Process the multi-source data separately to obtain the input vector.

[0007] In one feasible implementation, the specific processing method of step 2 includes: Step 21: Transcribe the speech data using the ASR model (Automatic Speech Recognition Model); then, using voiceprint recognition, distinguish between the doctor's speech and the patient's speech, and construct the dialogue logic. Step 22: Use OCR to perform text recognition on the image text, and then perform table structure recognition and key information extraction (KIE) to obtain structured data; Step 23: Extract medical record information from the dialogue logic and all structured data using the NER model (Entity Recognition Model) and convert it into an input vector; the medical record information includes chief complaint, symptoms, present medical history, personal history, family history, abnormal indicators, diagnostic basis, diagnostic results and treatment plan, etc.

[0008] Step 3: Build a knowledge base, including medical knowledge and a hybrid search strategy.

[0009] In one feasible implementation, the specific method for constructing the knowledge base includes adding metadata tags (such as authority level, source, and version number) to literature data such as treatment guidelines (e.g., diagnostic method textbooks, treatment guidelines, etc.), medical record writing standards, international disease classification coding libraries (e.g., ICD-10 / 11 coding libraries), and hospital-specific knowledge (e.g., characteristic treatment plans, lists of commonly used drugs, and surgical templates). Then, the data is converted into vectors through a text embedding model and stored in a vector database (e.g., Milvus database or Pinecone database) to achieve efficient and low-latency semantic retrieval.

[0010] In one feasible implementation, the knowledge base also includes a knowledge update and version management mechanism to ensure that the knowledge is up-to-date. Specifically, it includes: based on the keywords of the literature data, periodically crawling version numbers from a preset website and comparing them with the current version number of the corresponding literature data in the knowledge base; if there is a version update, downloading the new literature data from the preset website and replacing the corresponding old literature data in the knowledge base.

[0011] In one feasible implementation, the hybrid retrieval strategy includes: parsing query information (such as drug name, disease name, etc.) from the input vector, retrieving the knowledge base to obtain a first retrieval result; performing metadata filtering on the first retrieval result according to a preset authority level range (such as level one to three) to obtain a second retrieval result; sorting the second retrieval result in descending order based on its relevance to the query information, and then performing Top-K filtering to obtain the final retrieval result.

[0012] In one feasible implementation, the specific method for calculating the correlation includes: Step a1: Based on the keywords in the query information, perform BM25 scoring on all document data in the knowledge base to achieve sparse retrieval. The specific calculation formula includes: ; in, Indicates query Terms in; express The inverse document frequency, specifically calculated using the following formula: ,in, This indicates the total number of documents in the knowledge base; Indicates inclusion The number of documents; express In the One document Frequency of occurrence in; express Length (number of words); This represents the average length of all documents in the knowledge base. and All are (adjustable) parameters, and their specific values ​​can be as follows: and ; Step a2: Calculate the cosine similarity between the query information vector and each document vector in the knowledge base, thereby utilizing an efficient index structure to achieve low-latency semantically intensive retrieval. The specific expression includes: ; in, Represents a query information vector; express ; express and Cosine similarity between them; Step a3: Calculate the relevance score. The specific expression includes: ; in, express The relevance score; This represents the balancing weight parameter, used to balance the contribution of sparse and dense retrieval, and can be set to a value of 0.5. express The normalized value of the BM25 score, express The normalized value of the cosine similarity is used to eliminate dimensional differences. Indicates the first Authoritative level Bonus points; This represents the weighting parameter for the authority level bonus, used to adjust the influence of authority on the final score.

[0013] In one feasible implementation, the The specific expressions include: ; in, , and These represent preset segment values, and have This ensures that highly authoritative literature receives more additional benefits.

[0014] In one feasible implementation, the level of authority includes: the diagnostic and treatment guidelines belong to the first level of authority; the medical record writing guidelines and the International Classification of Diseases coding database belong to the second level of authority; the hospital's proprietary knowledge belongs to the third level of authority; and the remaining literature belongs to the fourth level of authority.

[0015] Step 4: Train a large model for generating medical records.

[0016] In one feasible implementation, step 4 specifically includes: Step 41: Train the basic model (e.g., Qwen3-32B) using the LoRA (low-rank adaptation) fine-tuning method to generate prompts, resulting in a large-scale medical record generation model. This allows the model to develop deep memory and structural cognition of medical facts and medical record formats. Specific training data includes: dialogue logic, structured data, and labeled data. The labeled data is a standardized medical record template to form a gold standard. Specific prompts, such as: "You are a senior attending physician. Please fill in the medical record template based on the following information, and be sure to cite the provided treatment guidelines." Step 42: Introduce the CoT (Co-Thinking Chain) into the training data to form a triple consisting of input, CoT, and final result. This forces the model to generate a reasoning process before giving the final result, thereby compelling the model to "think" and "explain" the rationality of the final result. This greatly improves the interpretability and factual accuracy of the final result and reduces the probability of the model directly "fabricating" information. For example: Input (chief complaint, abnormal indicators), generate thought chain (diagnostic logic: the chief complaint meets the standard Y of disease A, disease B is excluded, so the preliminary diagnosis is disease A), and then generate the final result (diagnosis result). Step 43: In the key data points of the final result, mark the source of information (e.g., pointing to which original multi-source data or which index of medical knowledge it comes from) in order to form a traceability mechanism and avoid model fabrication.

[0017] In one feasible implementation, the standardized medical record template is derived from a medical record rule base; the medical record rule base is used to associate the standardized medical record template with keywords (such as chief complaint) in the medical record information, and to set the rules, types and value ranges for filling in the keywords.

[0018] Step 5: Based on the RAG method (retrieval / enhancement / generation), the retrieval results are obtained from the knowledge base according to the query information (such as chief complaint, symptoms, diagnosis results, etc.) in the input vector, and then concatenated with the query information to obtain the context. The context is then submitted to the medical record generation model to generate the initial medical record.

[0019] Secondly, based on the same inventive concept, this application also provides an intelligent ward round medical record generation system, including a data acquisition module, a data processing module and a result generation module; The data acquisition module is used to collect multi-source data during ward rounds; The data processing module includes an input vector unit, a knowledge base, a large model unit, and an initial medical record unit. The input vector unit is used to process multi-source data separately to obtain input vectors; The knowledge base stores medical knowledge and hybrid retrieval strategies; The large model unit is used to store medical records to generate large models; The initial medical record unit, based on the RAG method, obtains the retrieval results from the knowledge base according to the query information in the input vector, then concatenates them with the query information to obtain the context, and submits it to the large medical record generation model to generate the initial medical record; The result generation module is used to send out the initial medical records.

[0020] Thirdly, based on the same inventive concept, the present invention provides an intelligent ward round medical record generation device, including a processor, a memory, and a bus. The memory stores instructions and data that can be read by the processor. The processor is used to call the instructions and data in the memory to execute the intelligent ward round medical record generation method as described above. The bus connects the various functional components for transmitting information.

[0021] In one possible implementation, the generating device further includes smart glasses (worn by the doctor) that integrate miniature recording and photographic scanning devices. The recording device is used to capture the doctor-patient dialogue in real time; The aforementioned photo-scanning device is used to take photos and scan paper reports and medical records from other hospitals.

[0022] By adopting the above technical solution, the present invention has the following beneficial effects: This invention provides an intelligent ward round medical record generation method, system, and device. Through model fine-tuning, the general-purpose large model acquires capabilities such as medical professional knowledge, medical record writing standards, and doctor-patient dialogue understanding, and can accurately identify medical record structure and medical terminology. By constructing a knowledge base, it solves the problems of knowledge lag and insufficient professionalism of the large model, providing real-time and authoritative medical references for medical record generation. Through the RAG method, it solves the problem of illusory miswriting in the results generated by the large model, thereby avoiding medical accidents. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 A flowchart of an intelligent ward round medical record generation method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the specific processing method for step 2 provided in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the specific processing method for step 4 in an embodiment of the present invention. Figure 4 This is a diagram of an intelligent ward round medical record generation system provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0027] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0028] The present invention will be further explained below with reference to specific embodiments.

[0029] It should also be noted that the specific embodiments or implementation methods described below are a series of optimized settings listed by the present invention to further explain the specific content of the invention, and these settings can be combined or used in conjunction with each other.

[0030] Example 1: like Figure 1 As shown in the figure, the intelligent ward round medical record generation method provided in this embodiment includes the following steps: Step 1: Collect multi-source data during ward rounds; the multi-source data includes voice data, image text, and structured data.

[0031] Furthermore, the voice data refers to the doctor-patient dialogue content collected through recording equipment, including chief complaints, consultations, and medical orders; The image text refers to paper reports and medical records from other hospitals collected through a camera scanning device. The structured data refers to basic patient information, test results, and hospital history records collected through hospital management systems (such as HIS, LIS, or PACS).

[0032] Step 2: Process the multi-source data separately to obtain the input vector.

[0033] Furthermore, such as Figure 2 As shown, the specific processing method for step 2 includes: Step 21: Transcribe the speech data using the ASR model; then, using voiceprint recognition, distinguish between the doctor's speech and the patient's speech, and construct the dialogue logic. Step 22: Use OCR to perform text recognition on the image text, and then perform table structure recognition and key information extraction (KIE) to obtain structured data; Step 23: Extract medical record information from the dialogue logic and all structured data using the NER model and convert it into an input vector; the medical record information includes chief complaint, symptoms, present medical history, personal history, family history, abnormal indicators, diagnostic basis, diagnostic results and treatment plan, etc.

[0034] Step 3: Build a knowledge base, including medical knowledge and a hybrid search strategy.

[0035] Furthermore, the specific method for constructing the knowledge base includes adding metadata tags (such as authority level, source, and version number) to literature data such as treatment guidelines (e.g., diagnostic method textbooks, treatment guidelines, etc.), medical record writing standards, international disease classification coding libraries (e.g., ICD-10 / 11 coding libraries), and hospital-specific knowledge (e.g., characteristic treatment plans, lists of commonly used drugs, and surgical templates). Then, through a text embedding model, the data is converted into vectors and stored in a vector database (e.g., Milvus database or Pinecone database) to achieve efficient and low-latency semantic retrieval.

[0036] Furthermore, the knowledge base also includes a knowledge update and version management mechanism to ensure that the knowledge is up-to-date. Specifically, it includes: based on the keywords of the literature data, regularly crawling version numbers from a preset website and comparing them with the current version number of the corresponding literature data in the knowledge base; if there is a version update, downloading the new literature data from the preset website and replacing the corresponding old literature data in the knowledge base.

[0037] Furthermore, the hybrid retrieval strategy includes: parsing query information (such as drug name, disease name, etc.) from the input vector, retrieving the knowledge base to obtain a first retrieval result; performing metadata filtering on the first retrieval result according to a preset authority level range (such as level one to three) to obtain a second retrieval result; sorting the second retrieval result in descending order based on its relevance to the query information, and then performing Top-K filtering to obtain the final retrieval result.

[0038] Furthermore, the specific calculation method for the correlation includes: Step a1: Based on the keywords in the query information, perform BM25 scoring on all document data in the knowledge base to achieve sparse retrieval. The specific calculation formula includes: ; in, Indicates query Terms in; express The inverse document frequency, specifically calculated using the following formula: ,in, This indicates the total number of documents in the knowledge base; Indicates inclusion The number of documents; express In the One document Frequency of occurrence in; express Length (number of words); This represents the average length of all documents in the knowledge base. and All are (adjustable) parameters, and their specific values ​​can be as follows: and ; Step a2: Calculate the cosine similarity between the query information vector and each document vector in the knowledge base, thereby utilizing an efficient index structure to achieve low-latency semantically intensive retrieval. The specific expression includes: ; in, Represents a query information vector; express ; express and Cosine similarity between them; Step a3: Calculate the relevance score. The specific expression includes: ; in, express The relevance score; This represents the balancing weight parameter, used to balance the contribution of sparse and dense retrieval, and can be set to a value of 0.5. express The normalized value of the BM25 score, express The normalized value of the cosine similarity is used to eliminate dimensional differences. Indicates the first Authoritative level Bonus points; This represents the weighting parameter for the authority level bonus, used to adjust the influence of authority on the final score.

[0039] Furthermore, the aforementioned The specific expressions include: ; in, , and These represent preset segment values, and have This ensures that highly authoritative literature receives more additional benefits.

[0040] Furthermore, the aforementioned levels of authority include: treatment guidelines are at the first level of authority; medical record writing standards and the International Classification of Diseases coding database are at the second level of authority; proprietary knowledge of this hospital is at the third level of authority; and other literature is at the fourth level of authority.

[0041] Step 4: Train a large model for generating medical records.

[0042] Furthermore, such as Figure 3 As shown, the specific method for step 4 includes: Step 41: Train the basic model (e.g., Qwen3-32B) using the LoRA (low-rank adaptation) fine-tuning method to generate prompts, resulting in a large-scale medical record generation model. This allows the model to develop deep memory and structural cognition of medical facts and medical record formats. Specific training data includes: dialogue logic, structured data, and labeled data. The labeled data is a standardized medical record template to form a gold standard. Specific prompts, such as: "You are a senior attending physician. Please fill in the medical record template based on the following information, and be sure to cite the provided treatment guidelines." Step 42: Introduce the CoT (Co-Thinking Chain) into the training data to form a triple consisting of input, CoT, and final result. This forces the model to generate a reasoning process before giving the final result, thereby compelling the model to "think" and "explain" the rationality of the final result. This greatly improves the interpretability and factual accuracy of the final result and reduces the probability of the model directly "fabricating" information. For example: Input (chief complaint, abnormal indicators), generate thought chain (diagnostic logic: the chief complaint meets the standard Y of disease A, disease B is excluded, so the preliminary diagnosis is disease A), and then generate the final result (diagnosis result). Step 43: In the key data points of the final result, mark the source of information (e.g., pointing to which original multi-source data or which index of medical knowledge it comes from) in order to form a traceability mechanism and avoid model fabrication.

[0043] Furthermore, the standardized medical record template is derived from the medical record rule base; the medical record rule base is used to associate the standardized medical record template with keywords (such as chief complaint) in the medical record information, and to set the rules, types and value ranges for filling in the keywords.

[0044] Step 5: Based on the RAG method (retrieval / enhancement / generation), the retrieval results are obtained from the knowledge base according to the query information (such as chief complaint, symptoms, diagnosis results, etc.) in the input vector, and then concatenated with the query information to obtain the context. The context is then submitted to the medical record generation model to generate the initial medical record.

[0045] Example 2: like Figure 4 As shown, this embodiment provides an intelligent ward round medical record generation system, including a data acquisition module, a data processing module, and a result generation module; The data acquisition module is used to collect multi-source data during ward rounds; The data processing module includes an input vector unit, a knowledge base, a large model unit, and an initial medical record unit. The input vector unit is used to process multi-source data separately to obtain input vectors; The knowledge base stores medical knowledge and hybrid retrieval strategies; The large model unit is used to store medical records to generate large models; The initial medical record unit, based on the RAG method, obtains the retrieval results from the knowledge base according to the query information in the input vector, then concatenates them with the query information to obtain the context, and submits it to the large medical record generation model to generate the initial medical record; The result generation module is used to send out the initial medical records.

[0046] Example 3: This embodiment provides an intelligent ward round medical record generation device, including a processor, a memory, and a bus. The memory stores instructions and data that can be read by the processor. The processor is used to call the instructions and data in the memory to execute the intelligent ward round medical record generation method described above. The bus connects the various functional components for transmitting information.

[0047] Furthermore, the generating device also includes smart glasses (worn by the doctor), which integrate miniature recording and scanning devices; The recording device is used to capture the doctor-patient dialogue in real time; The aforementioned photo-scanning device is used to take photos and scan paper reports and medical records from other hospitals.

[0048] In another embodiment, this solution can also be implemented using an integrated device, which may include corresponding modules that perform one or more steps in the various embodiments described above. A module may be one or more hardware modules specifically configured to perform the corresponding step, or implemented by a processor configured to perform the corresponding step, or stored in a computer-readable medium for implementation by a processor, or implemented through some combination thereof.

[0049] The processor executes the various methods and processes described above. For example, the method implementations in this scheme can be implemented as software programs tangibly contained in a machine-readable medium, such as memory. In some implementations, part or all of the software program can be loaded and / or installed via memory and / or a communication interface. When the software program is loaded into memory and executed by the processor, one or more steps of the methods described above can be performed. Alternatively, in other implementations, the processor can be configured to execute one of the methods described above by any other suitable means (e.g., by means of firmware).

[0050] This device can be implemented using a bus architecture. A bus architecture can include any number of interconnect buses and bridges, depending on the specific application of the hardware and overall design constraints. The bus connects various circuits, including one or more processors, memory, and / or hardware modules. The bus can also connect various other circuits such as peripherals, voltage regulators, power management circuitry, external antennas, etc.

[0051] Buses can be Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, or Extended Industry Standard Component (EISA) buses, etc. Buses can be divided into address buses, data buses, control buses, etc.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart rounding medical record generation method, characterized in that, include: Step 1: Collect multi-source data during ward rounds; the multi-source data includes voice data, image text, and structured data; Step 2: Process the multi-source data separately to obtain the input vector; Step 3: Construct a knowledge base, including medical knowledge and hybrid search strategies; Step 4: Train a large-scale medical record generation model; Step 5: Based on the RAG method, retrieve the search results from the knowledge base according to the query information in the input vector, then concatenate them with the query information to obtain the context, and submit it to the medical record generation model to generate the initial medical record.

2. The generation method according to claim 1, characterized in that, The voice data refers to the doctor-patient dialogue content collected through recording equipment, including chief complaints, consultations, and medical orders; The image text refers to the paper reports and medical records from other hospitals collected through a photo scanning device; The structured data refers to the patient's basic information, test results, and the hospital's historical medical records collected through the hospital management system.

3. The generation method according to claim 1, characterized in that, The specific processing method in step 2 includes: Step 21: Transcribe the speech data using the ASR model; then, using voiceprint recognition, distinguish between the doctor's speech and the patient's speech, and construct the dialogue logic. Step 22: Using the OCR method, perform text recognition on the image text, then perform table structure recognition and key information extraction to obtain structured data; Step 23: Extract medical record information from the dialogue logic and all structured data using the NER model and convert it into an input vector; the medical record information includes chief complaint, symptoms, present medical history, personal history, family history, abnormal indicators, diagnostic basis, diagnostic results and treatment plan.

4. The generation method according to claim 1, characterized in that, The specific method for constructing the knowledge base includes adding metadata tags to the literature data, then converting it into vectors through a text embedding model, and finally storing it in a vector database.

5. The generation method according to claim 4, characterized in that, The hybrid retrieval strategy includes: parsing query information from the input vector, retrieving the knowledge base to obtain a first retrieval result; filtering the first retrieval result with metadata according to a preset authority level range to obtain a second retrieval result; sorting the second retrieval result in descending order based on its relevance to the query information, and then performing Top-K filtering to obtain the final retrieval result.

6. The generation method according to claim 5, characterized in that, The specific calculation method for the correlation includes: Step a1: Based on the keywords in the query information, perform BM25 scoring on all document data in the knowledge base. The specific calculation formula includes: ； in, Indicates query Terms in; express The inverse document frequency, specifically calculated using the following formula: ,in, This indicates the total number of documents in the knowledge base; Indicates inclusion The number of documents; express In the One document Frequency of occurrence in; express Length; This represents the average length of all documents in the knowledge base. and All are parameters; Step a2: Calculate the cosine similarity between the query information vector and each document vector in the knowledge base. The specific expression includes: ； in, Represents a query information vector; express ; express and Cosine similarity between them; Step a3: Calculate the relevance score. The specific expression includes: ； in, express The relevance score; Indicates the balance weight parameters; express The normalized value of the BM25 score; express The normalized value of the cosine similarity; Indicates the first Authoritative level Bonus points; This indicates the weighting parameter for the authority level bonus.

7. The generation method according to claim 6, characterized in that, The specific expressions include: ； in, , and These represent preset segment values, and have .

8. The generation method according to claim 1, characterized in that, Step 4 specifically includes: Step 41: Train the basic model to generate prompt words using the LoRA fine-tuning method to obtain a large medical record generation model; the specific training data includes: dialogue logic, structured data, and label data; the label data is a standardized medical record template; Step 42: Introduce thought chains into the training data to form triples that sequentially include input, thought chains, and final results; Step 43: Mark the source of information in the key data points of the final result.

9. An intelligent ward round medical record generation system, characterized in that, It includes a data acquisition module, a data processing module, and a result generation module; The data acquisition module is used to collect multi-source data during ward rounds; The data processing module includes an input vector unit, a knowledge base, a large model unit, and an initial medical record unit. The input vector unit is used to process multi-source data separately to obtain input vectors; The knowledge base stores medical knowledge and hybrid retrieval strategies; The large model unit is used to store medical records to generate large models; The initial medical record unit, based on the RAG method, obtains the retrieval results from the knowledge base according to the query information in the input vector, then concatenates them with the query information to obtain the context, and submits it to the large medical record generation model to generate the initial medical record; The result generation module is used to send out the initial medical records.

10. An intelligent ward round medical record generation device, characterized in that, It includes a processor, a memory, and a bus. The memory stores instructions and data read by the processor. The processor is used to call the instructions and data in the memory to execute the generation method as described in any one of claims 1-8. The bus connects the functional components for transmitting information.

Images