A method and system for intelligent coding review and automated data entry of medical records
By using natural language processing and human-computer collaboration, the review and entry of medical record codes are automated, solving the problems of low efficiency and insufficient accuracy in existing technologies, and achieving efficient and accurate medical record coding and data quality management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGSU CANCER HOSPITAL
- Filing Date
- 2026-04-02
- Publication Date
- 2026-06-30
Smart Images
Figure CN121961498B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a method and system for intelligent coding review and automated entry of medical records. Background Technology
[0002] The patient's medical record cover sheet is the core summary of the patient's medical treatment information. It plays a fundamental role in hospital management, medical statistics, and clinical research. It is also a key data source for DRG / DIP medical insurance payment, hospital accreditation, and public hospital performance evaluation. Its core element is the quality of the International Classification of Diseases and Surgical Procedures (ICD) codes, which directly relates to the rational allocation of medical resources, hospital economic benefits, and the scientific nature of management decisions.
[0003] The existing methods for reviewing and entering codes on the front page of medical records face the following technical problems: Over-reliance on manual operation leads to inefficiency, requiring coders to switch back and forth between medical records and the medical record management system, manually recording diagnostic information, diagnostic procedures, pathology results, and medication records, and then manually entering them into the medical record management system. This cumbersome process wastes a significant amount of time. The accuracy of coding is limited; when dealing with complex tumor diseases, especially those with multiple primary cancers, rare tumors, and novel treatment methods, it is difficult to avoid errors in proactive judgment. Furthermore, clinicians sometimes write diagnoses and describe treatment processes in medical records in a non-standard, incomplete, or ambiguous manner, increasing the difficulty for coders to make accurate judgments and leading to errors and omissions.
[0004] Furthermore, existing medical record coding is insufficient for mining the value of data, and cannot carry out deeper data quality control, statistical analysis, or provide coding knowledge services to clinicians, thus limiting the further release of the value of medical record data. Summary of the Invention
[0005] To address the technical problems of the prior art, this application provides a method and system for intelligent coding review and automated entry of medical records. It uses natural language processing to extract key information from electronic medical records, intelligently recommends ICD coding schemes based on the information subject, assists in the automated entry of the medical record front page, and improves coding accuracy and efficiency through human-machine collaboration, reduces coding error rate, and forms an efficient and accurate medical record coding work mode.
[0006] This application provides a method for intelligent coding review and automated data entry of medical records, including:
[0007] Step S10: Obtain the medical record front page data to be entered from the electronic medical record system through multi-source acquisition, including structured data of the patient's discharge diagnosis, treatment process, pathology report, medication record and cost information, and classify and preprocess the data according to the medical record type;
[0008] Step S20: Analyze the text data in the medical record homepage data based on the pre-trained natural language processing model to understand the doctor's diagnosis description and treatment process. Combine the constructed coding rule base and knowledge graph, and output a recommended coding list and abnormal prompts through rule mapping and reasoning.
[0009] Step S30: Establish a human-machine collaboration mechanism. Coders will focus on reviewing abnormal information based on abnormal prompts, correct codes based on professional judgment, and automatically enter the reviewed medical record coding data into the medical record management system.
[0010] Furthermore, multi-source acquisition methods are used to obtain medical record homepage data, and structured data is directly obtained through WebService interface and database view. For unstructured text data such as discharge diagnosis, treatment process and pathology report, computer vision methods based on natural language processing are used for information extraction and parsing.
[0011] Based on the treatment method, disease complexity, and admission type, the collected medical record homepage data were initially screened using keywords and divided into non-surgical tumor medical records, surgical treatment medical records, and difficult and severe medical records for differentiated coding.
[0012] All medical record front page data are preprocessed to identify and correct spelling errors. For missing information in pathology reports, the data is automatically completed by combining relevant pathological examination data from the diagnosis and treatment process. Non-standardized terms in the data are mapped to a standard dictionary to unify the terminology and establish a connection with ICD coding rules. Unstructured data is converted into structured fields.
[0013] Furthermore, step S20 involves a complete intelligent coding review process, integrating natural language processing (NLP), rule engines, and knowledge graphs. It is divided into five stages: feature extraction, NLP deep understanding, rule mapping, intelligent reasoning, and result generation. By simulating the logical thinking of coders, it achieves intelligent analysis of doctors' diagnostic descriptions and treatment processes. Combined with specialized coding rule bases and knowledge graphs, it ultimately outputs accurate ICD coding recommendations and anomaly alerts.
[0014] This application also provides a medical record intelligent coding review and automated data entry system, including:
[0015] Data acquisition module: Used to acquire the front page data of the medical record to be entered from the electronic medical record system through multi-source acquisition methods, including structured data of the patient's discharge diagnosis, treatment process, pathology report, medication record and cost information, and classify and preprocess the data according to the medical record type;
[0016] Intelligent Analysis Module: Used to analyze text data in medical record homepage data based on pre-trained natural language processing models, understand doctors' diagnostic descriptions and treatment processes, and output recommended coding lists and anomaly prompts through rule mapping and reasoning, combined with the constructed coding rule base and knowledge graph.
[0017] Medical record entry module: This module is used to establish a human-machine collaboration mechanism. Coders review abnormal information based on anomaly prompts, correct codes based on professional judgment, and automatically enter the reviewed medical record coding data into the medical record management system.
[0018] This application discloses the following technical effects:
[0019] This application provides a method and system for intelligent coding review and automated data entry of medical records. The intelligent coding process is embedded into electronic medical record systems and medical record management systems, guiding coders to view AI-recommended codes and their basis in a familiar work environment, enabling one-click adoption and modification, thus optimizing the user experience. The proposed human-machine collaboration mode reduces the average review time and medical record entry time, significantly improving the overall efficiency of medical record coding and entry, and increasing coding accuracy. Furthermore, the intelligent algorithm embedded in the method generates anomaly reports when it detects non-standard clinical writing causing coding difficulties, promoting the monitoring and feedback of clinical medical record writing quality and improving data quality from the source. Through a coder feedback mechanism, the coding rule base and algorithm model are continuously iterated, forming a complete quality closed-loop management system. Attached Figure Description
[0020] Figure 1 This is a flowchart illustrating a method for intelligent coding review and automated entry of medical records, provided in an embodiment of this application.
[0021] Figure 2 This is a flowchart illustrating the medical record coding review process provided in this application embodiment.
[0022] Figure 3 This is a schematic diagram of the structure of a medical record intelligent coding review and automated entry system provided in an embodiment of this application. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0024] Example 1: This application provides a method for intelligent coding review and automated entry of medical records, such as... Figure 1 As shown, the method includes:
[0025] Step S10: Obtain the front page data of the medical record to be entered from the electronic medical record system through multi-source acquisition, including structured data of the patient's discharge diagnosis, treatment process, pathology report, medication record and cost information, and classify and preprocess the data according to the medical record type.
[0026] In this embodiment, key data fields obtained from the electronic medical record system are identified before data collection, covering information throughout the entire patient diagnosis and treatment process:
[0027] Discharge diagnosis: A textual description and preliminary coding of the patient's primary diagnosis of symptoms and other diagnoses;
[0028] Treatment process: Record key events such as patient examinations, treatments, and discharge after admission;
[0029] Pathological information: Diagnostic results, tumor grade, immunohistochemical markers, etc. in the patient's pathology report;
[0030] Medication records: drug name, dosage, time of administration and type (e.g., targeted or immunotherapy drugs);
[0031] Cost information: Comprehensive medical services, diagnostic services, treatment services, and Western medicine and traditional Chinese medicine, etc.
[0032] In collaboration with the hospital's information department, we applied for access to the electronic medical record system and database, configured the WebService API and database views, and ensured secure data transmission through intranet isolation, meeting medical data privacy requirements, and verifying the stability of the data interface and the compatibility of data formats.
[0033] Data from the medical record front page was acquired using a multi-source acquisition method. Structured data was directly obtained through WebService interfaces and database views. For unstructured text data such as discharge diagnoses, treatment processes, and pathology reports, computer vision methods based on natural language processing were used for information extraction and parsing.
[0034] For electronic medical record systems that support standard data interfaces, the HL7FHIR interface is configured based on WebService API permissions to enable data exchange between the medical record system and the medical record management system through interface calls; for systems without data interfaces, the medical record front page data to be entered is directly queried through the database view, and an incremental synchronization mechanism is set up to filter based on timestamps and collect only new data.
[0035] For medical record systems that cannot perform automatic database queries, data is obtained by simulating the coder's operation steps, robotic process automation tools are deployed, login credentials and navigation paths of the electronic medical record system are configured, the medical record query interface is simulated, and a computer vision model is trained to identify interface elements based on the principles of natural language processing, thereby extracting key information from unstructured data.
[0036] Based on treatment methods, disease complexity, and admission type, the collected medical record front page data were initially screened using keywords and categorized into non-surgical tumor cases, surgical treatment cases, and complex and severe cases for differentiated coding. The data was preprocessed to identify and correct spelling errors, and for missing information in pathology reports, automatic completion was performed using relevant pathological examination data from the treatment process. Non-standardized terms in the data were mapped to a standard dictionary to standardize terminology expression and establish associations with ICD coding rules. Unstructured data was converted into structured fields.
[0037] Step S20: Analyze the text data in the medical record homepage data based on the pre-trained natural language processing model to understand the doctor's diagnosis description and treatment process. Combined with the constructed coding rule base and knowledge graph, output a recommended coding list and anomaly prompts through rule mapping and reasoning.
[0038] In this embodiment, this step integrates natural language processing (NLP), rule engine and knowledge graph, and is divided into five stages: feature extraction, NLP deep understanding, rule mapping, intelligent reasoning and result generation.
[0039] In the feature extraction stage, a pre-trained entity recognition model is used to extract medical entities from the pre-processed medical record homepage data, including disease entities, treatment entities, and time entities, summarizing the diagnosis and treatment process; the encoder in the entity recognition model generates character vectors for the medical record homepage data, captures contextual semantics, and uses a bidirectional long short-term memory module to model the contextual relationships between data fields and identify entity boundaries; and establishes associations between the identified entities, describing the correspondence between drugs, treatment methods, and diseases.
[0040] In the NLP deep understanding phase, the domain-adaptive BERT model is used to perform deep semantic analysis on the text data of the medical record front page and the identified medical entities:
[0041] The NLP model first identifies the key points of a text paragraph, using diagnostic descriptions for disease confirmation and treatment processes to determine treatment details. It then analyzes the contextual logic of data and entities, focusing on the purpose of the current treatment rather than the patient's medical history when identifying "non-first-time admissions" for cancer patients. Secondly, based on the correspondence between medical entities and the contextual relationships of the text data, the NLP model categorizes treatment methods, identifies keywords related to metastasis and recurrence, and determines disease staging. Finally, through contradiction detection, the NLP model automatically identifies logical conflicts in the text and marks incorrect mappings between diagnostic descriptions and medication information as abnormal and requiring further review.
[0042] In the rule mapping stage, the NLP analysis results are matched with professional rules, and accurate mapping is achieved through knowledge graph queries to ensure coding accuracy.
[0043] The professional rules are selected from the core rule base and auxiliary rule base for establishing medical record coding:
[0044] The core rule base refers to the collection of mandatory and general coding rules stored in national standards, industry standards and clinical treatment guidelines for disease diagnosis and surgical procedure classification. It includes at least the basic coding principles and main classification axes of ICD-10 and ICD-9-CM-3.
[0045] The auxiliary rule base covers multi-dimensional rules, including a pathological diagnosis database, a drug name classification database, a cost information database, as well as the hospital's preferred coding for specific disease combinations, localized mapping rules for fuzzy diagnostic terms, and commonly used coding combinations set to improve coding efficiency.
[0046] First, based on disease diagnosis keywords, a basic coding framework is matched from the core rule base. Based on ICD coding rules, it is determined whether the tumor patient is admitted for the first time, the treatment method is identified, and the pathological characteristics are determined. Then, based on the pathological characteristics, applicable optimization rules are called from the auxiliary rule base to refine and correct the basic coding, resulting in accurate professional rules.
[0047] The knowledge graph centers on diseases and links nodes related to etiology, location, treatment methods, drug name classification, and cost information to form a networked knowledge system. Each node feature contains corresponding coding rules. Inputting text data analysis results from NLP, the knowledge graph outputs associated codes and treatment rules, and verifies the logical consistency between the NLP analysis results and the knowledge graph. When errors in the correspondence between diseases and drugs are detected, a warning is sent.
[0048] In the intelligent reasoning stage, the decision-making process of the coder is simulated. Coding suggestions are generated through multi-dimensional reasoning. Based on the rule priority mechanism, the coding rules are sorted according to their weights. The NLP analysis results are input to identify key entities. The rules are queried according to the logical relationship between entities to map the entities to the corresponding codes. When multiple rules conflict, a voting mechanism is activated.
[0049] When performing anomaly detection, anomaly code E001 is triggered for rule violations, such as when the treatment method in the diagnosis process does not specify the treatment site; data integrity is checked, and when key fields of medical record data are found to be missing, they are marked as "to be completed"; a confidence score is calculated for each inference result, and when the score is lower than the threshold, it is marked as a key review target.
[0050] The results generation stage outputs a list of coding results for the medical record front page data, including code, name, and confidence level, sorted in descending order of confidence level, with coding results having a confidence level higher than 0.95 recommended; anomaly detection results are divided into data quality anomalies and logical anomalies, and the corresponding anomaly causes and handling suggestions are identified and fed back to the coders.
[0051] Figure 2 The complete workflow for automating medical record data processing based on the above five stages is presented, encompassing the entire process from data reading and rule judgment to result output:
[0052] At the start of the process, variables are first cleared to ensure a clean processing environment. Basic configuration data such as template lists and pathology categories are read to provide a basis for subsequent judgments. At the same time, a list of hospital admission numbers is obtained to determine the range of medical records that need to be processed.
[0053] Next, the process enters the file processing phase, where each medical record file in the list undergoes the same processing:
[0054] For the current file, open the Excel file and read the medical record data list. Query the number of admissions, admission date, and other information for the medical record. First, determine if the discharge diagnosis exists in the medical record: if it does not exist, delete the file directly and end the processing of the file; if the discharge diagnosis exists, check if the number of hospitalizations is normal: if it is not normal, perform a conversion operation on the number of hospitalizations to standardize the data.
[0055] Then, extract the text content of discharge diagnosis and treatment process from the medical record: process the treatment process, take the last sentence, extract the treatment methods such as radiotherapy and chemotherapy, and perform key condition judgments to expand the branch processing:
[0056] Special Medical Record Judgment Branch: Determine if it is a special medical record diagnosis: If yes, set the abnormal reason, add this case to the abnormal data reason list, and try to extract the detailed content of the pathological diagnosis: If the extraction is successful (pathological diagnosis is normal), save the discharge diagnosis record; if the extraction fails (pathological diagnosis is abnormal), record the abnormality.
[0057] Routine processing of non-special medical records: If it is not a special medical record, determine whether it is the first admission:
[0058] If it is the first admission, the detailed pathological diagnosis is extracted to determine whether it is normal: if the pathological diagnosis is normal, a discharge diagnosis record is generated and saved; if it is abnormal, it is handled as an abnormal pathological diagnosis.
[0059] If this is not the first admission, the discharge diagnosis record will be generated and saved directly.
[0060] Finally, after all files have been processed in a loop, the process enters the final stage; the relevant data for special medical records is saved, the number of abnormal table rows is retrieved, and it is checked whether any abnormal records have accumulated during the entire processing. The number of abnormal table rows is then assessed.
[0061] If the value is 0 (no abnormality), then save all the final discharge diagnosis data directly;
[0062] If the value is not 0 (there is an anomaly), save the abnormal data first, and then save the normal discharge diagnosis data.
[0063] Step S30: Establish a human-machine collaboration mechanism. Coders will focus on reviewing abnormal information based on abnormal prompts, correct codes based on professional judgment, and automatically enter the reviewed medical record coding data into the medical record management system.
[0064] In this embodiment, the coder first reviews the list of anomalies marked by the system, processes them in order of priority, reviews relevant medical records for in-depth analysis of each anomaly, and uses the system's built-in coding query tool to verify difficult coding rules.
[0065] When correcting codes, clinical knowledge, coding experience, and historical cases are combined, and the rationality of disease development patterns and treatment logic is considered. Special cases not explicitly covered by the rules are handled, and the coding handling method is determined by comparing with similar medical cases. The coder adopts, modifies, or rejects the codes intelligently recommended in step S20, records the reasons and basis for the modification, and forms a complete review process.
[0066] Example 2: The intelligent medical record coding review and automated entry system provided in this embodiment of the invention can execute the intelligent medical record coding review and automated entry method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method, such as... Figure 3 As shown, it includes the following modules:
[0067] Data acquisition module: Used to acquire the front page data of the medical record to be entered from the electronic medical record system through multi-source acquisition methods, including structured data of the patient's discharge diagnosis, treatment process, pathology report, medication record and cost information, and classify and preprocess the data according to the medical record type;
[0068] Intelligent Analysis Module: Used to analyze text data in medical record homepage data based on pre-trained natural language processing models, understand doctors' diagnostic descriptions and treatment processes, and output recommended coding lists and anomaly prompts through rule mapping and reasoning, combined with the constructed coding rule base and knowledge graph.
[0069] Medical record entry module: This module is used to establish a human-machine collaboration mechanism. Coders review abnormal information based on anomaly prompts, correct codes based on professional judgment, and automatically enter the reviewed medical record coding data into the medical record management system.
[0070] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.
[0071] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
Claims
1. A method for intelligent coding review and automated entry of medical records, characterized in that, The method includes: Step S10: Obtain the medical record front page data to be entered from the electronic medical record system through multi-source acquisition, including structured data of the patient's discharge diagnosis, treatment process, pathology report, medication record and cost information, and classify and preprocess the data according to the medical record type; Step S20: Analyze the text data in the medical record homepage data based on the pre-trained natural language processing model to understand the doctor's diagnosis description and treatment process. Combine the constructed coding rule base and knowledge graph, and output a recommended coding list and abnormal prompts through rule mapping and reasoning. Step S20 includes five stages: feature extraction, NLP deep understanding, rule mapping, intelligent reasoning, and result generation. It integrates Natural Language Processing (NLP), rule engine, and knowledge graph. By simulating the logical thinking of a coder, it achieves intelligent analysis of doctors' diagnostic descriptions and treatment processes. Combined with a specialized coding rule base and knowledge graph, it outputs accurate ICD coding recommendations and anomaly alerts. The feature extraction process uses a pre-trained entity recognition model to extract medical entities from the pre-processed medical record homepage data, including disease entities, treatment entities, and time entities, summarizing the diagnosis and treatment process. The encoder in the entity recognition model generates character vectors for the medical record homepage data, capturing contextual semantics, and uses a bidirectional long short-term memory module to model the contextual relationships between data fields and identify entity boundaries. It also establishes associations between the identified entities, describing the correspondence between drugs, treatment methods, and diseases. The aforementioned NLP deep understanding uses a domain-adaptive BERT model to perform deep semantic analysis on the text data of the medical record front page and the identified medical entities: First, the NLP model is used to determine the key points of medical record text paragraphs, using diagnostic descriptions for disease confirmation and treatment processes for determining treatment details; and it analyzes the contextual logic of data and entities. Secondly, the NLP model classifies treatment methods and identifies metastasis and recurrence keywords to determine disease stage based on the correspondence between medical entities and the contextual relationship of medical record text data. Finally, the NLP model automatically detects logical conflicts in the text through contradiction detection, and marks erroneous mappings between diagnostic descriptions and medication information as abnormal and requiring further review. The rule mapping matches NLP analysis results with professional rules, achieving precise mapping through knowledge graph queries to ensure coding accuracy. The professional rules are selected from the core and auxiliary rule bases of medical record coding. The core rule base refers to the collection of mandatory and general coding rules stored in national standards, industry standards and clinical treatment guidelines for disease diagnosis and surgical procedure classification. It includes at least the basic coding principles and main classification axes of ICD-10 and ICD-9-CM-3. The auxiliary rule base covers multi-dimensional rules, including a pathological diagnosis database, a drug name classification database, a cost information database, as well as the hospital's preferred coding for specific disease combinations, localized mapping rules for fuzzy diagnostic terms, and commonly used coding combinations set to improve coding efficiency. First, based on disease diagnosis keywords, a basic coding framework is matched from the core rule base. Based on ICD coding rules, it is determined whether the cancer patient is a first-time hospital admission, the treatment method is identified, and the pathological characteristics are determined. Subsequently, based on the pathological characteristics, applicable optimization rules are called from the auxiliary rule base to refine and correct the basic coding, resulting in precise professional rules. Step S30: Establish a human-machine collaboration mechanism. Coders will focus on reviewing abnormal information based on abnormal prompts, correct codes based on professional judgment, and automatically enter the reviewed medical record coding data into the medical record management system.
2. The method for intelligent coding review and automated entry of medical records as described in claim 1, characterized in that, In step S10, the multi-source acquisition method directly obtains structured data through WebService interface and database view. For unstructured text data such as discharge diagnosis, treatment process and pathology report, information extraction and parsing are performed using computer vision methods based on natural language processing.
3. The method for intelligent coding review and automated entry of medical records as described in claim 1, characterized in that, In step S10, the classification and preprocessing process includes: Based on the treatment method, disease complexity, and admission type, the collected medical record homepage data were initially screened using keywords and divided into non-surgical tumor medical records, surgical treatment medical records, and difficult and severe medical records for differentiated coding. It identifies and corrects spelling errors, and automatically completes missing information in pathology reports by combining relevant pathology examination data from the diagnosis and treatment process. Non-standardized terms appearing in the data are mapped to a standard dictionary to unify the terminology and establish a connection with ICD encoding rules; for unstructured data, it is converted into structured fields.
4. The method for intelligent coding review and automated entry of medical records as described in claim 1, characterized in that, The intelligent reasoning simulates the coder's decision-making process, generates coding suggestions through multi-dimensional reasoning, sorts coding rules according to weight based on a rule priority mechanism, inputs NLP analysis results, identifies key entities, queries rules according to the logical relationships between entities, maps entities to corresponding codes, and activates a voting mechanism when multiple rules conflict. When performing anomaly detection, if a rule violation occurs and the treatment method in the diagnosis process does not specify the treatment site, an anomaly code E001 is triggered; when data integrity is checked, if a key field in the medical record data is found to be missing, it is marked as "to be completed"; when a confidence score is calculated for each inference result, if the score is lower than the threshold, it is marked as a key review target.
5. The method for intelligent coding review and automated entry of medical records as described in claim 1, characterized in that, The generated results output a list of coding results for the medical record front page data, including code, name, and confidence level, sorted in descending order of confidence level, with recommended coding results having a confidence level higher than 0.95; the anomaly detection results are divided into data quality anomalies and logical anomalies, the causes of the anomalies and handling suggestions are identified, and feedback is given to the coders.
6. A medical record intelligent coding review and automated data entry system, characterized in that, The system is used to implement the intelligent coding review and automated data entry method for medical records as described in any one of claims 1-5. The system includes: Data acquisition module: Used to acquire the front page data of the medical record to be entered from the electronic medical record system through multi-source acquisition methods, including structured data of the patient's discharge diagnosis, treatment process, pathology report, medication record and cost information, and classify and preprocess the data according to the medical record type; Intelligent Analysis Module: Used to analyze text data in medical record homepage data based on pre-trained natural language processing models, understand doctors' diagnostic descriptions and treatment processes, and output recommended coding lists and anomaly prompts through rule mapping and reasoning, combined with the constructed coding rule base and knowledge graph. Medical record entry module: This module is used to establish a human-machine collaboration mechanism. Coders review abnormal information based on anomaly prompts, correct codes based on professional judgment, and automatically enter the reviewed medical record coding data into the medical record management system.