Electronic medical record content quality control method and system based on large language model

Abstract

The invention provides an electronic medical record connotation quality control method and system based on a large language model, and the method comprises the steps: carrying out the modal classification of a whole-process medical record of a patient, and uniformly converting the whole-process medical record into a corresponding text modal; performing hierarchical division on the whole-process medical record of the text mode to form a medical record content tree; according to different nodes of the medical record content tree, retrieving a quality control rule set corresponding to the nodes of the medical record content tree, and retrieving a corresponding diagnosis and treatment knowledge set; for each quality control rule in the quality control rule set, if a plurality of medical record content tree nodes are needed, establishing a corresponding context set, constructing a quality control cue word list, inputting cue words into the connotation quality control big language model, judging whether a quality control problem exists in the cue words or not by the connotation quality control big language model, and proposing a modification suggestion; and checking and revising the output result to obtain a final quality control result. According to the method, key point missing, description conflicts and deep diagnosis and treatment logic contradictions in medical records can be recognized and corrected, and the accuracy and stability of quality control results are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical text error detection, and in particular relates to a method and system for quality control of electronic medical record content based on a large language model. Background Art

[0002] Electronic medical records (EMRs) are the core data repository for critical patient information, including the physician's professional diagnostic basis and conclusions, the treatment process, and the evolution of disease. They are essential data for ensuring healthcare quality and building smart healthcare systems. A key component of the healthcare quality control process, EMRs not only improve medical records and standardize the treatment process, but also provide the foundation for timely identification of healthcare quality deficiencies. They are of great significance and value in ensuring healthcare quality and safety and strengthening healthcare service oversight.

[0003] Medical record quality control primarily encompasses formal and connotation quality control. Formal quality control addresses simple missing items, while connotation quality control primarily addresses flaws in the textual content. It reviews, intervenes, and monitors aspects of medical record contextual consistency (data consistency, logical consistency, etc.), objective logic and evidence-based medical logic (diagnostic adequacy, drug interactions, etc.), the rationality of medical interviews and the standardization of documentation (terminology standardization, etc.), regulatory compliance, and the correspondence between medical orders and documentation, as well as any similarities. Currently, traditional information technology and artificial intelligence technologies offer promising solutions for formal quality control. However, due to fragmented capabilities, high model construction costs, and weak transferability, current intelligent medical record content quality control still primarily relies on manual recognition, far from fully leveraging the benefits of information and intelligent technologies in improving the efficiency of medical record content quality control. With the development of large language model technology, the application prospects of this technology in the medical field are evident. However, there are currently no mature methods and systems for electronic medical record content quality control based on large language models on the market. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for quality control of electronic medical record content based on a large language model, aiming to solve the problems of scattered quality control capabilities of medical record content, high model construction costs, and weak migration capabilities in the existing technology.

[0005] In a first aspect, a method for quality control of electronic medical record content based on a large language model comprises the following steps:

[0006] Step S1: Input the patient's full medical history from the medical history data standard interface;

[0007] Step S2: Classify the patient's full medical history into different modalities, use different preprocessing methods according to different modalities, and uniformly convert them into corresponding text modalities;

[0008] Step S3: Input the full-process medical record in text mode into the medical record content segmentation module. The medical record content segmentation module divides the medical record into different levels according to different content modules to form a medical record content tree. Different levels and nodes of the content tree correspond to different parts of the medical record.

[0009] Step S4: According to different nodes of the medical record content tree, the quality control rule set corresponding to the medical record content tree node is retrieved through the quality control rule retrieval system, and the corresponding diagnosis and treatment knowledge set is retrieved through the diagnosis and treatment knowledge retrieval system;

[0010] Step S5: For each quality control rule in the quality control rule set, if it requires multiple medical record content tree nodes, a corresponding context set is established, and a quality control prompt word list is constructed. The prompt words in the processed list are input into the connotation quality control large language model. The connotation quality control large language model determines whether there are quality control issues and proposes modification suggestions;

[0011] Step S6: After the connotation quality control model outputs quality control issues and modification suggestions, the coders and doctors review and revise the output results to obtain the final quality control results.

[0012] Furthermore, in step S2, if the modality of the medical record is an image modality, image text recognition is performed; if the modality of the medical record is a medical imaging modality, image interpretation is performed to extract text descriptions related to diagnosis and treatment; if the modality of the medical record is a sound modality, speech text recognition is performed and converted into text, and noise cleaning is performed on the medical record data in text modality and the medical record data converted from other modalities.

[0013] Furthermore, in step S3, after the medical record data is input, the medical record content segmentation module divides the medical record into levels according to different content modules to form a medical record content tree to obtain structured data. The different levels and nodes of the content tree correspond to different parts of the medical record, including the admission record in the admission record, personal history, past history, current medical history, medical history, test report, imaging report and discharge record.

[0014] Furthermore, in step S4, the rule documents related to medical record writing are sorted out to form a medical record quality control rule base, which is used to identify missing key points and shallow description conflicts in medical records; for diagnosis and treatment guidelines, expert consensus and medical literature, a multimodal large model is used to extract relevant diagnosis and treatment knowledge to form a diagnosis and treatment knowledge base, which is used to identify and correct deep-level conflicts in medical records.

[0015] Furthermore, the input of the multimodal large model includes text and images of diagnosis and treatment guidelines, expert consensus and medical literature. The multimodal large model extracts key information from the text and obtains a natural language representation of the image modality data.

[0016] Furthermore, in step S5, the extracted medical record quality control points are compared with the connotation quality control rules to identify the missing key points and shallow description conflicts in the medical records. By designing multi-stage, multi-dimensional instruction tasks, combined with diagnosis and treatment guidelines, expert consensus and diagnosis and treatment knowledge in medical literature, the connotation quality control large language model is used to realize the identification and correction of deep-level conflicts in the rationality of medical questions, contextual logical consistency, diagnostic adequacy and drug interactions.

[0017] Furthermore, the training process of the connotation quality control large language model includes data labeling, instruction data construction and instruction fine-tuning;

[0018] Data annotation: Manual annotation is performed based on real medical record data to construct a medical record quality control dataset. The annotation content includes medical record quality control issues and the corresponding original medical record text fragments;

[0019] Instruction data construction: Collect real medical records and annotated quality control sample data, perform data cleaning and preprocessing to ensure data quality, and build a quality control instruction dataset based on these high-quality sample data;

[0020] Instruction fine-tuning: Use the constructed instruction dataset to fine-tune the connotation quality control language model, so that it can more efficiently understand and apply medical quality control instructions and provide more accurate quality control suggestions and guidance.

[0021] Furthermore, in step S6, after the actual application of the connotation quality control large language model, by collecting feedback information on the model output from doctors and relevant experts, a quality control comparison data set is constructed, and a reinforcement learning training reward model based on human feedback is used to guide the optimization of the quality control model to form a closed-loop optimization path, thereby achieving a closed-loop improvement in the authenticity, accuracy and ethical consistency of the quality control results.

[0022] In a second aspect, a large language model-based electronic medical record content quality control system is provided, which is used to implement any of the above-mentioned large language model-based electronic medical record content quality control methods, including:

[0023] The medical record content segmentation module is used to hierarchically divide several original full-process medical record texts according to different content modules of the medical record to form a medical record content tree;

[0024] The annotation module is used to manually annotate medical record texts at various stages, and properly preserve the original medical record fragments at each stage that contain the key points of medical record quality control, the corresponding key points of medical record quality control, and the corresponding quality control issues;

[0025] The quality control rule construction and retrieval module is used to process the rule documents related to medical record writing to form the connotation quality control rules, and retrieve the medical records of different content modules to obtain the quality control rule set;

[0026] The diagnosis and treatment knowledge construction and retrieval module is used to process diagnosis and treatment guidelines, expert consensus and medical literature data to form diagnosis and treatment knowledge, and retrieve diagnosis and treatment knowledge sets for different diseases;

[0027] A data construction module is used to generate an instruction training data set and construct a quality control comparison data set based on the labeled data of the labeling module, the quality control rule set obtained by the quality control rule construction and retrieval module, and the diagnosis and treatment knowledge set obtained by the diagnosis and treatment knowledge construction and retrieval module;

[0028] A training module, used to train the large language model for connotation quality control, including instruction training and reinforcement learning based on human feedback;

[0029] The medical record quality control module is used to call the connotation quality control large language model to perform connotation quality control on the medical record text at each stage by combining the corresponding quality control rules, diagnosis and treatment knowledge and context, identify quality control issues and provide modification suggestions;

[0030] The review module is used to review and revise the output results of the connotation quality control model after the quality control issues and modification suggestions are output, and obtain the final quality control results;

[0031] The database module is used to store the data required by the system and the data generated by the system, and provides data support for the annotation module, quality control rule construction and retrieval module, diagnosis and treatment knowledge construction and retrieval module, data construction module, training module, medical record quality control module and review module.

[0032] In a third aspect, a computer-readable storage medium stores a program for a method for quality control of the content of electronic medical records based on a large language model. When the program for the method for quality control of the content of electronic medical records based on a large language model is executed by a processor, the steps of the method for quality control of the content of electronic medical records based on a large language model as described in any one of the above items are implemented.

[0033] Beneficial effects of the present invention:

[0034] The present invention is based on natural language processing and large language model technology, integrating electronic medical record parsing, structured processing, quality control point extraction, deep review and reinforcement learning optimization to construct an efficient and intelligent electronic medical record content quality control method and system. Compared with the existing technology, the present invention uses the powerful semantic understanding and reasoning capabilities of the large language model to accurately identify key quality control points such as contextual logic consistency, diagnostic adequacy, drug interactions, and terminology standardization in medical records, thereby improving the intelligence level of medical record quality control. In view of the multimodal characteristics of medical record data, the present invention supports the conversion and standardization of non-text data such as images, videos, and voice, ensuring that all medical record information is input in text mode, improving data integrity and consistency, and combining automatic parsing technology to achieve structured processing of medical records. Different from traditional manual review or rule matching methods, the present invention adopts instruction fine-tuning, and through high-quality labeled data, domain knowledge injection and semantic matching optimization, the model is adapted to different medical scenarios and improves generalization and migration capabilities. During the review stage, the present invention constructs a knowledge base based on medical record writing standards, diagnosis and treatment guidelines, expert consensus and medical literature, combined with the deep semantic analysis of the large language model, to accurately identify medical record defects and achieve comprehensive medical record quality control. In addition, the present invention introduces reinforcement learning based on human feedback, and uses feedback from doctors and experts to continuously optimize the model to ensure the accuracy and credibility of the review results and enhance the adaptability and stability of the system. Through the pre-training capability of the large language model, the present invention effectively reduces the cost of model construction and maintenance, and only requires a small amount of training data to achieve high-quality quality control, reduce data labeling and rule construction work, and improve the scalability and application value of the system. Ultimately, the present invention can not only accurately review individual medical records, but also perform batch analysis on large-scale medical record data, discover common problems in medical institutions in medical record writing and diagnosis and treatment processes, provide efficient data support for medical quality supervision, and help build an accurate, efficient, and sustainably optimized medical quality control system, and improve overall medical quality and patient safety assurance capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 It is a flow chart of the electronic medical record content quality control method of the present invention;

[0037] Figure 2 This is an example diagram of the process of using a multimodal large model to analyze the diagnosis and treatment flowchart to obtain semantic representation;

[0038] Figure 3It is a relationship diagram of various modules of the electronic medical record content quality control system of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clear and distinct, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] Example 1:

[0041] like Figure 1 As shown, this embodiment provides a method for quality control of electronic medical record content based on a large language model, including the following steps:

[0042] Step S1: Input the patient's full-process medical record data from the medical record data standard interface.

[0043] Step S2: Classify the patient's full-process medical records into different modalities, use different preprocessing methods according to different modalities, and uniformly convert them into corresponding text modalities.

[0044] Specifically, during the pre-processing of the input full-process medical records, if the modality is an image, image text recognition is performed; if the modality is a medical imaging modality, image interpretation is performed to extract text descriptions related to diagnosis and treatment; if the modality is an audio modality, speech text recognition is performed and converted to text. Noise cleaning is performed on medical records in text modality and medical records converted from other modalities. This includes format standardization, removal of irrelevant characters, and correction of image recognition errors to improve data quality.

[0045] Step S3: Input the full-process medical record in text mode into the medical record content segmentation module. The medical record content segmentation module divides the medical record into different levels according to different content modules to form a medical record content tree. Different levels and nodes of the content tree correspond to different parts of the medical record.

[0046] Specifically, after inputting textual medical record data, the medical record content segmentation module performs hierarchical segmentation according to the different content modules of the medical record, forming a medical record content tree and obtaining structured data. This medical record content segmentation module can be composed of various methods, such as regular expressions, or directly using relevant models for segmentation. The different levels and nodes of the content tree correspond to different parts of the medical record, including the admission record, personal history, past medical history, present medical history, medical history, laboratory reports, imaging reports, and discharge records.

[0047] Step S4: According to different nodes of the medical record content tree, the quality control rule set corresponding to the medical record content tree node is retrieved through the quality control rule retrieval system, and the corresponding diagnosis and treatment knowledge set is retrieved through the diagnosis and treatment knowledge retrieval system.

[0048] Specifically, in order to construct an internal quality control rule base, we first organize the rule documents related to medical record writing, such as the "Basic Specifications for Medical Record Writing and Management", to form a set of medical record quality control rules, which are used to identify missing key points and shallow description conflicts in medical records; for diagnosis and treatment guidelines, expert consensus and medical literature, etc., multimodal large models are used to extract relevant diagnosis and treatment knowledge, thereby constructing a diagnosis and treatment knowledge base for identifying and correcting deep-level conflicts in medical records.

[0049] A further technical solution of the present invention is: the input of the multimodal large model includes text and images of diagnosis and treatment guidelines, expert consensus and medical literature, and the multimodal large model extracts key information from the text and simultaneously obtains natural language representation of image modality data such as flowcharts.

[0050] like Figure 2 As shown, in some specific embodiments, taking the identification of the diagnosis and treatment path from the diagnosis and treatment flowchart to obtain a semantic representation as an example, the input part includes instructions for guiding the multimodal large model to analyze the diagnosis and treatment flowchart to obtain a semantic representation and the corresponding diagnosis and treatment flowchart, and the output part is only composed of the natural language representation of the flowchart, and the specific form is as follows:

[0051] Input: "This is a diagnosis and treatment flowchart for {disease name}. Please generate the corresponding natural language description. \n{diagnosis and treatment flowchart image}"

[0052] Output: "{natural language representation of the flowchart}".

[0053] The multimodal large model processes the multimodal information in diagnosis and treatment guidelines, expert consensus and medical literature to obtain the diagnosis and treatment knowledge contained therein, and then constructs a diagnosis and treatment knowledge base to provide support for subsequent in-depth quality control of medical record content.

[0054] Step S5: For each quality control rule in the quality control rule set, if it requires multiple medical record content tree nodes, a corresponding context set is established, and a quality control prompt word list is constructed. The prompt words in the processed list are input into the connotation quality control large language model. The connotation quality control large language model determines whether there are quality control problems and makes modification suggestions.

[0055] Specifically, the extracted medical record quality control points are compared with the connotation quality control rules to identify missing key points and shallow description conflicts in the medical records. By designing multi-stage, multi-dimensional instruction tasks, combined with diagnosis and treatment guidelines, expert consensus, medical literature and other diagnosis and treatment knowledge, a large language model is used to realize the identification and correction of deep-level conflicts such as the rationality of medical questions, contextual logical consistency, diagnostic adequacy, and drug interactions.

[0056] Further optimize the technical solution. The training process of the connotation quality control large language model includes data annotation, instruction data construction, and instruction fine-tuning.

[0057] Data annotation: Based on real medical record data, manual annotation is carried out to construct a medical record quality control data set. The annotation content includes medical record quality control problems and the corresponding original text fragments of the medical record.

[0058] Instruction data construction: Collect real medical records and annotated quality control sample data, and perform data cleaning and preprocessing to ensure data quality. Based on these high-quality samples, relevant quality control rules, and medical treatment knowledge, a quality control instruction data set is constructed. The instruction data is divided into single-module medical record connotation quality control data and multi-module medical record connotation quality control data. The specific form is as follows:

[0059] Single-module medical record connotation quality control instruction template: "You are an excellent medical record quality control expert, and you are familiar with various types of electronic medical records and related quality control requirements.\nNext, you will be given a {medical record text type} text in a medical record, a quality control rule, and relevant medical treatment knowledge. Please judge whether there are any problems according to this rule and medical treatment knowledge, and output in json format.\n<Quality control rule>\n{Quality control rule}\n< / Quality control rule>\n\n<Medical treatment knowledge>\n{Medical treatment knowledge}\n< / Medical treatment knowledge>\n\n<{medical record text type} text>\n{Medical record text}\n<{medical record text type} text>".

[0060] Multi-module medical record connotation quality control instruction template: "You are an excellent medical record quality control expert, and you are familiar with various types of electronic medical records and related quality control requirements.\nNext, you will be given the {medical record text type list} medical record texts in a medical record and a quality control rule. Please judge whether there are any problems according to this rule.\n<Quality control rule>\n{Quality control rule}\n< / Quality control rule>\n\n<Medical treatment knowledge>\n{Medical treatment knowledge}\n< / Medical treatment knowledge>\n\n<{medical record text type1} text>\n{Medical record text1}\n< / {medical record text type1} text>\n…\n<{medical record text type n} text>\n{Medical record text n}\n< / {medical record text type n} text>".

[0061] Among them, {medical record text type list} is the medical record text type involved in this {quality control rule}. Through such a design, the medical record connotation quality control is split into a single-label classification problem for a specific rule. Each time the medical record connotation quality control large model only needs to judge whether there is a problem with a certain rule and how to modify it, which greatly reduces the task difficulty.

[0062] Instruction fine-tuning: Use the constructed instruction dataset to fine-tune the connotation quality control large language model, so that it can more efficiently understand and apply medical quality control instructions, identify quality control issues more accurately, and provide modification suggestions.

[0063] Step S6: After the connotation quality control model outputs quality control issues and modification suggestions, the coders and doctors review and revise the output results to obtain the final quality control results.

[0064] Specifically, after the actual application of the connotation quality control large language model, by collecting feedback information on the model output from doctors and relevant experts, a quality control comparison data set is constructed, and a reinforcement learning training reward model based on human feedback is used to guide the optimization of the quality control model to form a closed-loop optimization path, thereby achieving a closed-loop improvement in the authenticity, accuracy and ethical consistency of the quality control results.

[0065] Specifically, the quality control comparison dataset includes instructions for guiding the large language model to output quality control results and modification suggestions, which are constructed in the same way as the instructions in step S5, followed by a list that includes the quality control results and modification suggestions given by the doctor and the quality control results and modification suggestions output by the large language model. The specific format is as follows:

[0066] “{Input quality control instructions}\n[{Quality control results and modification suggestions given by the doctor}{Quality control results and modification suggestions output by the large language model}]”.

[0067] Example 2:

[0068] like Figure 3 As shown, this embodiment provides an electronic medical record content quality control system based on a large language model, which is used to implement an electronic medical record content quality control method based on a large language model, including:

[0069] The medical record content segmentation module is used to hierarchically divide several original full-process medical record texts according to different content modules of the medical records to form a medical record content tree.

[0070] The annotation module is used to manually annotate medical records at various stages and properly preserve the annotated medical record content quality control issues and modification suggestions along with the corresponding original medical record content;

[0071] The quality control rule construction and retrieval module is used to process the rule documents related to medical record writing to form the connotation quality control rules, and retrieve the medical records of different content modules to obtain the quality control rule set;

[0072] The diagnosis and treatment knowledge construction and retrieval module is used to process diagnosis and treatment guidelines, expert consensus, medical literature and other data to form diagnosis and treatment knowledge, and retrieve diagnosis and treatment knowledge sets for different diseases;

[0073] A data construction module is used to generate an instruction training data set and construct a quality control comparison data set based on the labeled data of the labeling module, the rule set obtained by the quality control rule construction and retrieval module, and the diagnosis and treatment knowledge set obtained by the diagnosis and treatment knowledge construction and retrieval module;

[0074] A training module, used to train the large language model for connotation quality control, including instruction training and reinforcement learning based on human feedback;

[0075] The medical record quality control module is used to call the connotation quality control large language model to perform connotation quality control on the medical record text at each stage by combining the corresponding quality control rules, diagnosis and treatment knowledge and context, identify quality control issues and provide modification suggestions;

[0076] The review module is used to review and revise the output results of the connotation quality control model after the quality control issues and modification suggestions are output, and obtain the final quality control results.

[0077] The database module is used to store the data required by the system and the data generated by the system, and provides data support for the annotation module, quality control rule construction and retrieval module, diagnosis and treatment knowledge construction and retrieval module, data construction module, training module, medical record quality control module and review module.

[0078] In the system, the medical record content segmentation module and the database module are used to implement the operations of step S1, step S2 and step S3; the labeling module, quality control rule construction and retrieval module, diagnosis and treatment knowledge construction and retrieval module, data construction module, training module, medical record quality control module and database module are used to implement the operations of step S4 and step S5; the review module, data construction module, training module and database module are used to implement the operation of step S6.

[0079] The present invention also provides a computer-readable storage medium, on which is stored a program for an electronic medical record content quality control method based on a large language model. When the program for an electronic medical record content quality control method based on a large language model is executed by a processor, the steps of the electronic medical record content quality control method based on a large language model as described in any one of the above items are implemented.

[0080] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for quality control of electronic medical records based on a large language model, characterized by: The following steps are involved: Step S1: Input the patient's full medical history from the medical history data standard interface; Step S2: Classify the patient's full medical history into different modalities, use different preprocessing methods according to different modalities, and uniformly convert them into corresponding text modalities; Step S3: Input the full-process medical record in text mode into the medical record content segmentation module. The medical record content segmentation module divides the medical record into different levels according to different content modules to form a medical record content tree. Different levels and nodes of the content tree correspond to different parts of the medical record. Step S4: According to different nodes of the medical record content tree, the quality control rule set corresponding to the medical record content tree node is retrieved through the quality control rule retrieval system, and the corresponding diagnosis and treatment knowledge set is retrieved through the diagnosis and treatment knowledge retrieval system; Step S5: For each quality control rule in the quality control rule set, if it requires multiple medical record content tree nodes, a corresponding context set is established, and a quality control prompt word list is constructed. The prompt words in the processed list are input into the connotation quality control large language model. The connotation quality control large language model determines whether there are quality control issues and proposes modification suggestions; Step S6: After the connotation quality control model outputs quality control issues and modification suggestions, the coders and doctors review and revise the output results to obtain the final quality control results.

2. The method for quality control of electronic medical records based on a large model according to claim 1, characterized in that: In step S2, if the modality of the medical record is an image modality, image text recognition is performed; if the modality of the medical record is a medical imaging modality, image interpretation is performed to extract text descriptions related to diagnosis and treatment; if the modality of the medical record is a sound modality, speech text recognition is performed and converted into text, and noise cleaning is performed on the medical record data in text modality and the medical record data converted from other modalities.

3. The method for quality control of electronic medical record content based on a large language model according to claim 1, characterized in that: In step S3, after the medical record data is input, the medical record content segmentation module divides the medical record into different levels according to the different content modules to form a medical record content tree and obtain structured data. The different levels and nodes of the content tree correspond to different parts of the medical record, including the admission record, personal history, past history, current medical history, medical history, test report, imaging report and discharge record in the admission record.

4. The method for quality control of electronic medical record content based on a large language model according to claim 1, characterized in that: In step S4, the relevant rule documents for medical record writing are organized to form a medical record quality control rule base, which is used to identify missing key points and shallow description conflicts in medical records; for diagnosis and treatment guidelines, expert consensus and medical literature, a multimodal large model is used to extract relevant diagnosis and treatment knowledge to form a diagnosis and treatment knowledge base, which is used to identify and correct deep-level conflicts in medical records.

5. The method for quality control of electronic medical record content based on a large language model according to claim 4 is characterized in that: The input of the multimodal large model includes text and images of diagnosis and treatment guidelines, expert consensus and medical literature. The multimodal large model extracts key information from the text and obtains the natural language representation of the image modality data.

6. The method for quality control of electronic medical record content based on a large language model according to claim 1, characterized in that: In step S5, the extracted medical record quality control points are compared with the connotation quality control rules to identify the missing key points and shallow description conflicts in the medical records. By designing multi-stage, multi-dimensional instruction tasks, combined with diagnosis and treatment guidelines, expert consensus and diagnosis and treatment knowledge in medical literature, the connotation quality control large language model is used to realize the identification and correction of deep-level conflicts in the rationality of medical questions, contextual logical consistency, diagnostic adequacy and drug interactions.

7. The method for quality control of electronic medical record content based on a large language model according to claim 6, characterized in that: The training process of the connotation quality control large language model includes data labeling, instruction data construction and instruction fine-tuning; Data annotation: Manual annotation is performed based on real medical record data to construct a medical record quality control dataset. The annotation content includes medical record quality control issues and the corresponding original medical record text fragments; Instruction data construction: Collect real medical records and annotated quality control sample data, perform data cleaning and preprocessing to ensure data quality, and build a quality control instruction dataset based on these high-quality sample data; Instruction fine-tuning: Use the constructed instruction dataset to fine-tune the connotation quality control language model, so that it can more efficiently understand and apply medical quality control instructions and provide more accurate quality control suggestions and guidance.

8. The method for quality control of electronic medical record content based on a large language model according to claim 1 or 7, characterized in that: In step S6, after the actual application of the connotation quality control large language model, by collecting feedback information on the model output from doctors and relevant experts, a quality control comparison dataset is constructed, and a reinforcement learning training reward model based on human feedback is used to guide the optimization of the quality control model to form a closed-loop optimization path, thereby achieving a closed-loop improvement in the authenticity, accuracy and ethical consistency of the quality control results.

9. An electronic medical record content quality control system based on a large language model, characterized by: A method for implementing the electronic medical record content quality control method based on a large language model as described in any one of claims 1 to 8, comprising: The medical record content segmentation module is used to hierarchically divide several original full-process medical record texts according to different content modules of the medical record to form a medical record content tree; The annotation module is used to manually annotate medical record texts at various stages, and properly preserve the original medical record fragments at each stage that contain the key points of medical record quality control, the corresponding key points of medical record quality control, and the corresponding quality control issues; The quality control rule construction and retrieval module is used to process the rule documents related to medical record writing to form the connotation quality control rules, and retrieve the medical records of different content modules to obtain the quality control rule set; The diagnosis and treatment knowledge construction and retrieval module is used to process diagnosis and treatment guidelines, expert consensus and medical literature data to form diagnosis and treatment knowledge, and retrieve diagnosis and treatment knowledge sets for different diseases; A data construction module is used to generate an instruction training data set and construct a quality control comparison data set based on the labeled data of the labeling module, the quality control rule set obtained by the quality control rule construction and retrieval module, and the diagnosis and treatment knowledge set obtained by the diagnosis and treatment knowledge construction and retrieval module; A training module, used to train the large language model for connotation quality control, including instruction training and reinforcement learning based on human feedback; The medical record quality control module is used to call the connotation quality control large language model to perform connotation quality control on the medical record text at each stage by combining the corresponding quality control rules, diagnosis and treatment knowledge and context, identify quality control issues and provide modification suggestions; The review module is used to review and revise the output results of the connotation quality control model after the quality control issues and modification suggestions are output, and obtain the final quality control results; The database module is used to store the data required by the system and the data generated by the system, and provides data support for the annotation module, quality control rule construction and retrieval module, diagnosis and treatment knowledge construction and retrieval module, data construction module, training module, medical record quality control module and review module.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program for the electronic medical record content quality control method based on a large language model. When the program for the electronic medical record content quality control method based on a large language model is executed by the processor, the steps of the electronic medical record content quality control method based on a large language model as described in any one of claims 1 to 8 are implemented.

Images