Chinese electronic medical record automatic ICD coding method and system

Abstract

The application belongs to the field of data processing, and provides a Chinese electronic medical record automatic ICD coding method and system, which comprises the following steps: preprocessing discharge diagnosis to obtain standardized diagnosis; performing correlation analysis on the Chinese electronic medical record text and the standardized diagnosis to generate diagnosis-evidence pairs; mapping the standardized diagnosis to ICD coding to obtain diagnosis coding results; mining missing clinical manifestations based on the Chinese electronic medical record text and the diagnosis-evidence pairs; gradually coding the missing clinical manifestations, and obtaining hierarchical coding results if the coding is successful; generating diagnosis suggestions based on a coding name library, and coding the missing clinical manifestations with coding failure by using the diagnosis suggestions to obtain suggestion coding results; and performing verification processing on the diagnosis coding results, the hierarchical coding results and the suggestion coding results to obtain final coding results. The application expands the coverage range of diagnosis and coding, and improves the integrity of overall coding.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to an automatic ICD encoding method and system for Chinese electronic medical records. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] The International Classification of Diseases (ICD), a crucial international standard developed and maintained by the World Health Organization, systematically categorizes various diseases, symptoms, and related external causes globally into a hierarchical, standardized alphanumeric coding system. Globally, the aging population and increasing public health awareness have jointly increased the demand for healthcare services. With the widespread application of information technology tools such as Electronic Health Records (EHRs) and Hospital Information Systems (HIS), the scale of medical data continues to grow; however, the efficient management and utilization of this data still faces many challenges. Against this backdrop, the ICD, through its standardized processing of medical information, provides crucial support for the effective integration and in-depth analysis of large-scale data. It not only facilitates the interoperability of medical information across institutions, regions, and even globally, but also plays an indispensable role in various fields such as epidemiological research, healthcare quality assessment, health resource allocation, health insurance payments, and public health policy formulation.

[0004] ICD coding refers to the process of matching appropriate ICD codes to medical texts, a task traditionally performed manually by professional coders. Manual coding is not only inefficient and labor-intensive, but also struggles to maintain consistent accuracy due to the complexity of the coding system itself and coder fatigue from long hours. To overcome these challenges, automated ICD coding has gradually emerged as a promising alternative and has attracted widespread attention in the field of medical informatics. However, current automated coding methods still have several limitations.

[0005] At the language and standards level, existing research largely focuses on English datasets (such as MIMIC III, MIMIC IV, and their subsets), with relatively little research specifically targeting Chinese scenarios. Currently, the ICD-10 standard has been localized for Chinese scenarios, resulting in the "National Clinical Version 2.0 of Disease Classification and Codes." This localization makes it difficult to directly transfer models trained on English datasets to the Chinese medical context. Furthermore, Chinese medical texts differ significantly from English in terms of language style, terminology, and medical record writing standards, further increasing the difficulty of cross-language model adaptation.

[0006] At the methodological and application level, most existing autocoding research adopts a multi-label classification framework, but it still has significant shortcomings: the models are heavily dependent on training data and have poor generalization ability when faced with ICD codes that were not encountered during the training phase. More importantly, these methods are disconnected from the actual operational processes of coders. Actual coding work is more complex and flexible: coders not only need to assign appropriate ICD codes to diseases explicitly listed in the discharge diagnosis, but also need to read through the entire medical document to identify and code diseases that, although not explicitly mentioned in the diagnosis, were actually involved or implied during the diagnosis and treatment process, thereby ensuring the accuracy and completeness of the coding results. In addition, if the autocoding system can simulate the actual working logic of coders and establish an explicit association between the diagnosis and the corresponding textual evidence, it will help improve the interpretability of the system and make the coding decision-making process more transparent and traceable.

[0007] In summary, current automatic ICD encoding methods are not yet fully adapted to the standard characteristics of the Chinese medical environment, and they fail to effectively match the actual workflow of coders when processing complex medical texts, resulting in their encoding performance in Chinese clinical scenarios still needing improvement. Summary of the Invention

[0008] To address the aforementioned issues, this invention proposes an automatic ICD encoding method and system for Chinese electronic medical records. This invention fully considers the standard differences in the Chinese medical environment and takes into account the actual workflow of coders when encoding complex medical texts, achieving satisfactory encoding results in Chinese medical scenarios.

[0009] According to some embodiments, the first solution of the present invention provides an automatic ICD encoding method for Chinese electronic medical records, which adopts the following technical solution:

[0010] An automatic ICD encoding method for Chinese electronic medical records, comprising:

[0011] Based on the patient's original electronic medical records, the Chinese electronic medical record text and discharge diagnosis were obtained;

[0012] Using a pre-built automatic ICD encoding model, assign corresponding ICD codes to Chinese electronic medical record texts, including:

[0013] Preprocessing of discharge diagnoses leads to standardized diagnoses;

[0014] The association analysis between Chinese electronic medical record text and standardized diagnosis was performed to retrieve the corresponding clinical manifestations and supporting evidence for the standardized diagnosis and generate diagnosis-evidence pairs.

[0015] Based on the diagnosis-evidence pairs and a pre-built coding name library, the normalized diagnosis is mapped to the ICD coding to obtain the diagnosis coding result;

[0016] Differential analysis of Chinese electronic medical record texts and diagnosis-evidence pairs was conducted to uncover missed clinical manifestations;

[0017] Based on the coding name library, the missed clinical manifestations are encoded step by step according to the ICD hierarchical structure. If the encoding is successful, the hierarchical coding result is obtained.

[0018] Diagnostic suggestions are generated based on the coding name database. The diagnostic suggestions are then used to encode the missed clinical manifestations that failed to be coded, resulting in suggested coding results.

[0019] The diagnostic coding results, hierarchical coding results, and suggested coding results are verified to obtain the final coding results.

[0020] Furthermore, the encoded name library is stored according to five levels: chapter, group, class, subcategory, and code;

[0021] The stored content contains all the codes in the ICD-10 standard and their corresponding standard names;

[0022] Each data entry includes: chapter range code, chapter name; group range code, group name; class code, class name; suborder code, suborder name; disease code, disease name.

[0023] Furthermore, the encoded name library sets corresponding disease causal chain knowledge in each standard disease name for medical logic verification in the subsequent diagnostic matching process;

[0024] The disease causal chain includes four core medical dimensions: etiology, location, pathology, and clinical manifestations.

[0025] Furthermore, the standardized diagnosis is mapped to ICD codes based on the diagnosis-evidence pair and the pre-built coded name library to obtain the diagnostic coding results, including:

[0026] The standardized diagnosis is matched precisely with the standard disease name in the pre-built coding name library. If the match is successful, the corresponding ICD code is assigned as the diagnosis code result.

[0027] If an exact match fails, a similarity score is obtained by weighted fusion of character similarity, semantic similarity, and diagnostic causal chain similarity of standard disease names in the standardized diagnostic and coded name library. Standard disease names with similarity scores exceeding a set value are selected to generate a candidate set.

[0028] A knowledge network is constructed based on the candidate set. Reasoning and comparison are performed based on the knowledge network and the diagnosis-evidence pair. The standard disease name and corresponding ICD code that match the patient's actual condition are selected from the candidate set as the diagnosis code result.

[0029] Furthermore, the differential analysis of Chinese electronic medical record texts and diagnosis-evidence pairs uncovers missed clinical manifestations, including:

[0030] Compare the abnormal clinical manifestations in the Chinese electronic medical record text with the evidence in the diagnosis-evidence pair;

[0031] All abnormal clinical manifestations in the Chinese electronic medical record text that cannot be covered by evidence in the diagnosis-evidence pair constitute missed clinical manifestations.

[0032] Furthermore, the method of using diagnostic suggestions to encode omitted clinical manifestations that failed to be coded, and obtaining suggested coding results, includes:

[0033] The diagnostic suggestions are precisely matched with the standard disease names in the coding name library. If successful, the corresponding ICD code is assigned as the suggested coding result.

[0034] If an exact match fails, a similarity score is obtained by weighted fusion of the character similarity, semantic similarity, and diagnostic causal chain similarity between the diagnostic suggestions and the standard disease names in the coding name library. Standard disease names with similarity scores exceeding a set value are selected to generate a set of suggested coding candidates.

[0035] A suggested coding knowledge network is constructed based on the suggested coding candidate set. Based on the suggested coding knowledge network, standard disease names and corresponding ICD codes that match the patient's actual condition are selected from the suggested coding candidate set as suggested coding results.

[0036] According to some embodiments, the second aspect of the present invention provides an automatic ICD encoding system for Chinese electronic medical records, which adopts the following technical solution:

[0037] An automatic ICD encoding system for Chinese electronic medical records, comprising:

[0038] The medical record text processing module is used to filter patients' original electronic medical records and obtain Chinese electronic medical record text and discharge diagnosis.

[0039] The automatic ICD encoding module is used to assign corresponding ICD codes to Chinese electronic medical record texts using a pre-built automatic ICD encoding model, including:

[0040] Preprocessing of discharge diagnoses leads to standardized diagnoses;

[0041] The association analysis between Chinese electronic medical record text and standardized diagnosis was performed to retrieve the corresponding clinical manifestations and supporting evidence for the standardized diagnosis and generate diagnosis-evidence pairs.

[0042] Based on the diagnosis-evidence pairs and a pre-built coding name library, the normalized diagnosis is mapped to the ICD coding to obtain the diagnosis coding result;

[0043] Differential analysis of Chinese electronic medical record texts and diagnosis-evidence pairs was conducted to uncover missed clinical manifestations;

[0044] Based on the coding name library, the missed clinical manifestations are encoded step by step according to the ICD hierarchical structure. If the encoding is successful, the hierarchical coding result is obtained.

[0045] Diagnostic suggestions are generated based on the coding name database. The diagnostic suggestions are then used to encode the missed clinical manifestations that failed to be coded, resulting in suggested coding results.

[0046] The diagnostic coding results, hierarchical coding results, and suggested coding results are verified to obtain the final coding results.

[0047] According to some embodiments, a third aspect of the present invention provides a computer-readable storage medium.

[0048] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the automatic ICD encoding method for Chinese electronic medical records as described in the first embodiment above.

[0049] According to some embodiments, a fourth aspect of the present invention provides a computer device.

[0050] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the automatic ICD encoding method for Chinese electronic medical records as described in the first embodiment above.

[0051] According to some embodiments, a fifth aspect of the present invention provides a computer program product or computer program.

[0052] A computer program product or computer program includes computer instructions stored in a computer-readable storage medium, wherein a processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the automatic ICD encoding method for Chinese electronic medical records as described in the first embodiment above.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] This invention combines a multi-agent collaborative mechanism with an external knowledge base to perform multi-level processing on Chinese electronic medical record text, achieving a high degree of matching between diagnosis and coding, and effectively improving the accuracy of automatic ICD coding. Through diagnostic preprocessing and standardization modules, it can automatically parse non-standardized, colloquial, or combined discharge diagnoses and transform them into structured and standardized diagnostic inputs, improving the operability and consistency of coding from the source. It proposes a continuous construction mechanism for the diagnosis-evidence base, which can achieve a precise correspondence between diagnosis and clinical manifestations. This not only simulates the experience accumulation of human coders to assist the coding process, but also improves the interpretability and traceability of the model for complex medical record information.

[0055] This invention employs a combination of character similarity, semantic similarity, and diagnostic causal chain similarity, enabling the model to accurately identify and reason even when faced with synonyms, abbreviations, spelling differences, and semantic variations, significantly improving the robustness of the fuzzy matching stage. A potential diagnosis mining module is included to effectively identify clinical manifestations that may be missed in medical record text, expanding the coverage of diagnosis and coding, thereby enhancing the overall coding integrity. Through a hierarchical coding mechanism, utilizing the five-layer structure of ICD—chapter-group-class-subcategory-code—for step-by-step reasoning, the coding process is ensured to be systematic and orderly, reducing the probability of "illusion" or ambiguous selections by the model. The coding remediation and suggested coding submodules enable secondary reasoning and diagnostic suggestion generation for clinical manifestations that failed to be successfully coded at the hierarchical level, significantly improving the system's fault tolerance and robustness in complex medical record scenarios. Attached Figure Description

[0056] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0057] Figure 1 This is a processing architecture diagram of an automatic ICD encoding method for Chinese electronic medical records in an embodiment of the present invention;

[0058] Figure 2 This is a flowchart of the evidence search submodule in an embodiment of the present invention;

[0059] Figure 3 This is a flowchart of the diagnostic matching submodule in an embodiment of the present invention;

[0060] Figure 4 This is a flowchart of the proposed encoding submodule in an embodiment of the present invention. Detailed Implementation

[0061] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0062] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0063] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0064] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0065] Example 1

[0066] like Figure 1 As shown, this embodiment provides an automatic ICD encoding method for Chinese electronic medical records. This embodiment uses the application of this method to a server as an example for illustration. It can be understood that this method can also be applied to a terminal, and can also be applied to a system including a terminal, a server, and a system, and can be implemented through the interaction between the terminal and the server. The server can be an independent physical server, a server cluster composed of multiple physical servers, or a distributed system. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communication, middleware services, domain name services, CDN security services, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and the server can be directly or indirectly connected through wired or wireless communication, which is not limited in this application. In this embodiment, the method includes the following steps:

[0067] Based on the patient's original electronic medical records, the Chinese electronic medical record text and discharge diagnosis were obtained;

[0068] Using a pre-built automatic ICD encoding model, assign corresponding ICD codes to Chinese electronic medical record texts, including:

[0069] Preprocessing of discharge diagnoses leads to standardized diagnoses;

[0070] The association analysis between Chinese electronic medical record text and standardized diagnosis was performed to retrieve the corresponding clinical manifestations and supporting evidence for the standardized diagnosis and generate diagnosis-evidence pairs.

[0071] Based on the diagnosis-evidence pairs and a pre-built coding name library, the normalized diagnosis is mapped to the ICD coding to obtain the diagnosis coding result;

[0072] Differential analysis of Chinese electronic medical record texts and diagnosis-evidence pairs was conducted to uncover missed clinical manifestations;

[0073] Based on the coding name library, the missed clinical manifestations are encoded step by step according to the ICD hierarchical structure. If the encoding is successful, the hierarchical coding result is obtained.

[0074] Diagnostic suggestions are generated based on the coding name database. The diagnostic suggestions are then used to encode the missed clinical manifestations that failed to be coded, resulting in suggested coding results.

[0075] The diagnostic coding results, hierarchical coding results, and suggested coding results are verified to obtain the final coding results.

[0076] like Figure 1 As shown, this embodiment discloses an automatic ICD encoding method for Chinese electronic medical records based on multi-agent collaboration and external knowledge. The method includes the following steps:

[0077] S1: Filter based on the patient's original electronic medical record to obtain the Chinese electronic medical record text and discharge diagnosis.

[0078] In this embodiment, the dataset used to construct the automatic ICD encoding model is:

[0079] First, for a patient's original electronic medical record, content highly relevant to disease diagnosis and coding is selected to construct the Chinese electronic medical record text. Specifically, information under attributes such as the patient's medical history, hospitalization history, treatment process, imaging opinions, ultrasound findings, and ultrasound impressions is extracted from the original electronic medical record and used as the main content of the Chinese electronic medical record text. Simultaneously, the corresponding ICD codes are summarized as tags for this data entry. Typically, each patient may have multiple diseases simultaneously, meaning the Chinese electronic medical record text will correspond to multiple ICD codes, including one primary diagnosis code and several other diagnostic codes.

[0080] S101: Filter out the patient's original electronic medical record, including the patient's medical history, admission status, treatment process, imaging opinions, ultrasound findings, and ultrasound images, and construct the Chinese electronic medical record text.

[0081] First, for a patient's original electronic medical record, select text content that is highly relevant to disease diagnosis and coding to construct the Chinese electronic medical record text.

[0082] For example, the following is an example of constructing Chinese electronic medical record text in the original electronic medical record:

[0083] The patient's original electronic medical record usually contains a large amount of information, which is classified according to the patient's diagnosis and treatment process, such as admission record, present illness history, etc. These categories are called attributes. However, not all attributes are related to disease coding. Therefore, the content of the patient's original electronic medical record that is most helpful to ICD coding, such as the chief complaint and five histories, admission status, diagnosis and treatment process, are selected as the Chinese electronic medical record text. Among them, the chief complaint and five histories include chief complaint, present illness history, personal history, marital and reproductive history, past medical history, and family history.

[0084] S102: Obtain the discharge diagnosis from the patient's original electronic medical record;

[0085] For example, the following is an example of obtaining a discharge diagnosis from the original electronic medical record:

[0086] In the original electronic medical records, the discharge diagnosis is issued by the clinician when the patient is discharged. It often reflects the patient's final disease diagnosis during the hospital stay. It often contains colloquial expressions, non-standard writing, or multiple diseases combined in one record. This can cause ambiguity in subsequent automatic coding and requires further parsing and standardization in subsequent modules.

[0087] S103: Assign a corresponding ICD code to each constructed Chinese electronic medical record text. The purpose of assigning ICD codes is to verify the coding effect of the automatic ICD coding model.

[0088] For example, the following is an example of assigning a corresponding ICD code to each Chinese electronic medical record text:

[0089] Typically, a single Chinese electronic medical record corresponds to multiple ICD codes, including a primary diagnosis code and several other diagnostic codes. Using each patient's unique medical record identifier, the ICD codes corresponding to the discharge diagnosis are compiled to form a complete coded label for the electronic medical record text, which serves as the standard answer for the test.

[0090] S2: Summarize all encoded names

[0091] In this embodiment, to ensure the accuracy and standardization of the automatic ICD encoding model, it is necessary to first compile all encoding names in the ICD encoding system to form a standardized encoding name library. Specifically, for an ICD-10 code, it can be divided according to a hierarchical structure, including five levels: chapter, group, class, subcategory, and code, with each level having a unique standard name.

[0092] S201: Construction of the Encoded Name Library:

[0093] First, all codes and their corresponding standard names in the ICD-10 standard were collected and summarized. Complete information was then collected at five levels: chapter, group, class, suborder, and code, ensuring the database coverage remained consistent with the official ICD-10 standard. A combination of character matching and rule parsing was used to establish hierarchical relationships between different levels, such as determining the correspondence between chapters and groups based on code prefixes and ranges. Finally, all code entries were formatted into a unified data structure, ensuring each entry contained ten items: chapter_scale (chapter range code), chapter_name (chapter name); section_scale (group range code), section_name (group name); class_code (class code), class_name (class name); subborder_code (suborder code), subborder_name (suborder name); diagnosis_code (disease code), diagnosis_name (disease name).

[0094] For example, the ten items corresponding to the code I25.103 in the code name library are: chapter_scale (I00-I99), chapter_name (circulatory system diseases), section_scale (I20-I25), section_name (ischemic heart disease), class_code (I25), class_name (chronic ischemic heart disease), subborder_code (I25.1), subborder_name (arteriosclerotic heart disease), diagnosis_code (I25.103), and diagnosis_name (coronary atherosclerotic heart disease).

[0095] S202: Construction of the causal chain of the disease:

[0096] For each standard disease name in the coded name library, the corresponding disease causal chain knowledge is constructed using its different levels of names. This knowledge is used for medical logic verification in the subsequent diagnostic matching process. The disease causal chain includes four core medical dimensions: etiology, location, pathology, and clinical manifestations.

[0097] For example, the four dimensions of the disease causal chain for code I25.103 (coronary atherosclerotic heart disease) in the coding name library correspond to its different hierarchical names as follows: location dimension (corresponding to chapter_scale: I00-I99, chapter_name: circulatory system disease), etiology dimension (corresponding to section_scale: I20-I25, section_name: ischemic heart disease; and class_code: I25, class_name: chronic ischemic heart disease), pathology dimension (corresponding to subborder_code: I25.1, subborder_name: arteriosclerotic heart disease), and clinical manifestation dimension (corresponding to diagnosis_code: I25.103, diagnosis_name: coronary atherosclerotic heart disease). Furthermore, its disease causal chain is: circulatory system disease → ischemic heart disease → chronic ischemic heart disease → arteriosclerotic heart disease → coronary atherosclerotic heart disease.

[0098] S3: Using a pre-built automatic ICD encoding model, assign corresponding ICD codes to Chinese electronic medical record texts, including:

[0099] The automatic ICD coding model consists of a diagnostic preprocessing module, a diagnostic standardization module, a potential diagnostic mining module, a hierarchical coding module, a coding remediation module, and a coding result verification module.

[0100] S301: Diagnostic Preprocessing Module: This module uses a Diagnosis Processor to preprocess discharge diagnoses, resulting in several independent, structured, and standardized diagnoses. The specific steps are as follows:

[0101] First, the Diagnosis Processor agent receives Chinese electronic medical record text. The original discharge diagnosis As input, discharge diagnoses often contain non-standard expressions, including non-standardized descriptions influenced by doctors' writing habits, colloquial expressions, and complex representations of multiple diseases. The core task of Diagnosis Processor is to transform the original discharge diagnoses into more standardized and accurate versions. Analysis into multiple independent standardized diagnoses Each of them This represents a single, standardized diagnosis that has undergone analytical processing.

[0102] S302: Diagnostic Standardization Module: Taking standardized diagnoses and Chinese electronic medical record texts as input, the evidence search submodule performs correlation analysis between the two, retrieving clinical manifestations and supporting evidence corresponding to the standardized diagnosis from the Chinese electronic medical record text, generating diagnosis-evidence pairs. These pairs are used to confirm the reasonableness of the diagnostic matching results and can also be provided to the potential diagnosis mining module to assist in discovering missed clinical manifestations. The diagnosis-evidence pairs are continuously stored in the diagnosis-evidence database, enabling subsequent evidence retrieval for the same or similar diagnoses to utilize the existing knowledge in this database, thereby simulating the experience accumulation of human coders and improving the accuracy and stability of evidence retrieval. Finally, the diagnostic matching submodule maps the standardized diagnosis to specific diagnostic codes, thus obtaining the diagnostic coding results. The specific process is as follows: Figure 2 As shown,

[0103] S30201: Evidence Search Submodule: This module performs correlation analysis between Chinese electronic medical record text and standardized diagnoses, retrieves the corresponding clinical manifestations and supporting evidence for the standardized diagnoses, and generates diagnosis-evidence pairs. Specific steps are as follows:

[0104] The Evidence Finder agent identifies clinical evidence information in Chinese electronic medical record text that is highly relevant to the standardized diagnosis, including symptoms and abnormal signs directly related to the diagnosis in the chief complaint and present medical history.

[0105] Based on standardized diagnoses, corresponding historical diagnosis-evidence pairs are retrieved from the diagnosis-evidence database as supplementary reference information to enhance the comprehensiveness and stability of the current evidence search process. In other words, association analysis aims to obtain evidence pairs, while the retrieved evidence pairs play an auxiliary role, allowing association analysis to focus on these symptom manifestations in Chinese electronic medical record texts.

[0106] For standardized diagnosis The evidence search submodule handles Chinese electronic medical record texts. The different attribute contents are comprehensively analyzed and deeply mined. Finally, the evidence search submodule provides each standardized diagnosis with... Generate corresponding diagnosis-evidence pairs These diagnostic-evidence pairs not only reflect the actual evidence in the medical records, but also incorporate reference evidence patterns accumulated in the diagnostic-evidence database to improve the completeness and robustness of evidence searching. After the generated diagnostic-evidence pairs are used, they will be written into the diagnostic-evidence database to continuously enrich the database content and form a knowledge reserve similar to that accumulated by human coders over a long period of time, thereby improving the completeness and accuracy of coding. These diagnostic-evidence pairs not only provide rich auxiliary information for the determination of subsequent diagnostic matching results, but also help identify clinical manifestations that may be missed in Chinese electronic medical record texts through the correspondence between evidence and diagnosis, thereby improving the completeness and accuracy of diagnosis.

[0107] The prompt words for constructing the Evidence·Finder agent are designed as follows:

[0108] {"role": "system", "content": "You are a professional medical text analysis assistant, responsible for identifying evidence in medical record texts that supports a specific diagnosis."},

[0109] {"role": "user", "content": Please find the key evidence supporting this diagnosis in the following medical record text:}

[0110] Standardized diagnosis: {diagnosis['diagnosis']}; Chinese electronic medical record text: {medical_record}

[0111] Please return in JSON format. Example format:

[0112] "diagnosis": "hypertension", "evidence": ["Blood pressure measurement result: 160 / 95 mmHg, patient complains of headache"]

[0113] S30202: Diagnostic Matching Submodule: Based on diagnostic-evidence pairs and a pre-built coded name library, it maps standardized diagnoses to ICD codes to obtain diagnostic coding results; the specific steps are as follows:

[0114] (1) Precise matching: Based on the standardized diagnosis, a precise match is performed between the standard disease name in the pre-built coding name library. If successful, the corresponding ICD code is assigned as the diagnostic coding result.

[0115] First, standardize each diagnosis. Precise matching is performed in the coding name library. When a completely consistent standard disease name is found, the corresponding ICD code is directly assigned as the diagnostic coding result, thereby ensuring efficient processing of standardized diagnoses and avoiding unnecessary fuzzy calculations.

[0116] (2) Fuzzy matching: If exact matching fails, a similarity score is obtained by weighted fusion of character similarity, semantic similarity, and diagnostic causal chain similarity of standard disease names in the standardized diagnostic and coded name library. Standard disease names with similarity scores exceeding a set value are selected to generate a candidate set, including:

[0117] If an exact match fails, the system enters the fuzzy matching stage. This stage incorporates a triple similarity fusion mechanism, integrating character similarity, semantic similarity, and diagnostic causal chain similarity. This mechanism collaboratively evaluates the matching degree of each name in the standardized diagnostic and coded name database from three dimensions: surface form, deep semantics, and medical logic. Character similarity is among the key similarity metrics used. Standardized diagnostics are primarily measured based on standardized edit distance. Compared with the standard disease names in the coding name library Literal differences, semantic similarity The vector representation is obtained by using the pre-trained medical language model MedBERT, which then captures the semantic proximity between different expressions. The calculation formula is as follows:

[0118]

[0119] in, Representative standardized diagnosis The vector representation of , Represents the standard disease names in the code name library. The vector representations of these vectors are obtained through the pre-trained model MedBERT;

[0120] Diagnostic causal chain similarity Based on the causal chains of various diseases in the coded name database, including four main dimensions: etiology, location, pathology, and clinical manifestations, the diagnostic causal chains are integrated into a structured medical logical narrative using the medical knowledge comprehension capabilities of a large language model. And compare it with the corresponding diagnostic evidence. Semantic comparison is performed to determine the degree of agreement between the disease and the evidence of the patient's actual condition. The calculation formula is as follows:

[0121]

[0122] in, Representative standardized diagnosis The corresponding vector representation of supporting evidence, Represents standard disease names based on the coded name library. The medical logic description derived from the causal chain of the disease The vector representations of these vectors are obtained through the pre-trained model MedBERT;

[0123] For example, the medical logic description obtained from the disease causal chain of code I25.103 (coronary atherosclerotic heart disease) in the code name library is as follows: The patient's circulatory system has lesions, the heart is ischemic, the disease course is chronic ischemic evolution, the pathological mechanism is that the arteries undergo atherosclerotic changes, and finally coronary atherosclerosis is formed, and the clinical diagnosis is coronary atherosclerotic heart disease.

[0124] Character similarity semantic similarity Similarity to causal chains in diagnosis The similarity score is obtained by weighting and fusing the results according to different weight parameters. The calculation formula is as follows:

[0125]

[0126] in, Finally, the top-ranked similarity scores were selected. Standard disease names constitute a candidate set ;

[0127] (3) Knowledge Graph Assistance and Re-ranking: A knowledge network is constructed based on the candidate set. Reasoning and comparison are performed based on the knowledge network and the diagnosis-evidence pair. The standard disease name and corresponding ICD code that match the patient's actual condition are selected from the candidate set as the diagnostic coding result, including:

[0128] Candidate set Often, disease names contain several semantically similar names, making it difficult to directly determine the optimal match. Therefore, a medical knowledge graph, CPubMedKG, is introduced to retrieve the clinical manifestations, common complications, etiologies, and related knowledge of candidate diseases, forming a knowledge network. Next, the KG-Enhanced Re-ranker agent integrates multi-source information to re-rank the data. In other words, the KG-Enhanced Re-ranker agent re-ranks the data based on the knowledge network. And the diagnosis-evidence pair, through semantic understanding, medical logical reasoning and evidence comparison, selects the standard disease name that best matches the patient's actual condition from the candidate set, and outputs its corresponding ICD code to form the diagnosis code result;

[0129] For example, the local subgraphs obtained from entity retrieval for coronary atherosclerotic heart disease in the medical knowledge graph CPubMedKG are shown in Table 1. The relation types involved include: common aliases (synonyms) of the disease, complications, clinical manifestations, differential diagnosis, and drug treatment. Based on this, the retrieved local subgraphs are standardized, entities of the same disease are merged and unified, and the relation types between different entities are preserved, thus forming a structured knowledge network that can be used for reasoning. The knowledge network obtained for the above disease is as follows: Coronary atherosclerotic heart disease is also known as coronary artery disease or CAD; common complication is diabetes; clinical manifestation is microalbuminuria; differential diagnosis is bronchiectasis; drug treatment is with Tongxinluo capsules...

[0130] Table 1. Local sub-maps obtained from searching for coronary atherosclerotic heart disease.

[0131]

[0132] The diagnostic matching submodule is used to map standardized diagnoses to specific diagnostic codes, as detailed in the following process: Figure 3 As shown, each diagnosis in the standardized diagnosis process is precisely matched against the coded name database. When a match is successful, the ICD code corresponding to the coded name is selected as the diagnosis code result. When a match fails, a fuzzy matching stage is entered, introducing a triple similarity fusion mechanism of character similarity, semantic similarity, and diagnostic causal chain similarity. This mechanism collaboratively evaluates the degree of matching between the standardized diagnosis and each name in the coded name database from three dimensions: surface form, deep semantics, and medical logic. Character similarity is mainly based on edit distance to measure spelling differences, homophones, or abbreviations, while semantic similarity is achieved through a pre-trained medical language model, MedBER. T obtains a vector representation to capture the semantic similarity between different expressions. The causal chain similarity of the diagnosis is based on the causal chain of each disease in the coding name library, including four main dimensions: etiology, location, pathology, and clinical manifestations. It is compared with the corresponding diagnosis-evidence pair to ensure the accuracy of the match from a medical logic perspective. Then, the three are combined into a similarity score according to the set weights, and several most similar items are selected to form a candidate set. Further, a medical knowledge graph is used to find the clinical manifestations, complications, and other knowledge corresponding to different diseases in the candidate set to construct a knowledge network. On this basis, a KG-Enhanced Re-ranker agent is introduced to integrate the knowledge network and the diagnosis-evidence pair to select the code that best matches the patient's actual condition as the diagnosis code result. Finally, it is passed to the coding result verification module for subsequent verification.

[0133] The cue word design for constructing the KG-Enhanced Re-ranker agent is as follows:

[0134] "role": "system", "content": "You are a professional medical coder who needs to standardize original diagnostic names into standardized disease names. You need to select the most appropriate standard name while maintaining consistency in medical concepts."

[0135] "role": "user", "content": Standardized diagnosis: {original_diagnosis}

[0136] Please analyze the following information to determine the most accurate standard disease name corresponding to the standardized diagnosis. Diagnosis-evidence pair: {evidence_json}; Candidate standard disease names and their knowledge network: {candidates_json}

[0137] Please select the most accurate standard disease name based on: 1. the correspondence between the original diagnostic name and the candidate standard name; 2. the degree of consistency between the diagnostic evidence and the clinical manifestations of the candidate disease; 3. supplementary information provided by the knowledge graph (if any); and 4. your professional medical knowledge. Prioritize standard names that are consistent with the standardized diagnosis in medical concepts.

[0138] Please return in the following format: matched_disease: "Selected standard disease name"

[0139] S303: Potential Diagnosis Mining Module: Performs differential analysis on Chinese electronic medical record text and diagnosis-evidence pairs to uncover missed clinical manifestations, including:

[0140] Compare the abnormal clinical manifestations in the Chinese electronic medical record text with the evidence in the diagnosis-evidence pair;

[0141] All abnormal clinical manifestations in the Chinese electronic medical record text that cannot be covered by evidence in the diagnosis-evidence pair constitute missed clinical manifestations.

[0142] Using Chinese electronic medical record text and diagnosis-evidence pairs as input, a Residual Evidence Analyzer agent performs differential analysis on the content of the Chinese electronic medical record text to identify and uncover potential missed clinical manifestations. The specific steps are as follows:

[0143] Residual Evidence Analyzer agent receives Chinese electronic medical record text and diagnostic evidence The intelligent agent compares the content of each attribute in the Chinese electronic medical record text item by item to determine the existing evidence set. Whether the evidence set can cover all abnormal clinical manifestations appearing in the text is determined. If some abnormal clinical manifestations are found to be not covered by the existing evidence set, they are considered to have independent diagnostic and coding value. All abnormal clinical manifestations not covered by the existing evidence set are summarized to obtain the omitted clinical manifestations. ;

[0144] S304: Hierarchical Coding Module: Based on the coding name library, it progressively codes missed clinical manifestations according to the ICD hierarchical structure. If the coding is successful, the hierarchical coding result is obtained, including:

[0145] Using a coding name library, missed clinical manifestations are mapped layer by layer, from chapter to group, class, subcategory, and finally to disease code, gradually narrowing the selection range to achieve hierarchical ICD coding. If a missed clinical manifestation can be successfully coded, the coding result is output as a hierarchical coding result. If no matching entry is found at a certain level or ambiguity occurs, the missed clinical manifestation is passed to the subsequent coding recovery module for further processing. The hierarchical coding module receives the missed clinical manifestation, calls the hierarchical coding agent, and, with the support of the coding name library, reasones step by step according to the hierarchical structure of the ICD to generate the corresponding coding result. When the coding is successful, the hierarchical coding result is passed to the coding result verification module for subsequent verification. When the coding fails, the missed clinical manifestation is transferred to the coding recovery module.

[0146] S30401: Based on the omission of clinical manifestations, diagnosis-evidence pairs and coded name library, a set number of chapter levels are initially screened out as chapter diagnostic hypotheses using a hierarchical coding agent. Further analysis is performed on the chapter diagnostic hypotheses to obtain the matching chapter levels.

[0147] S30402: Based on the omission of clinical manifestations, diagnosis-evidence pairs and matching chapter levels, a set number of group levels are initially screened out as group diagnostic hypotheses using a hierarchical coding agent. Further analysis is performed on the group diagnostic hypotheses to obtain the matching group levels.

[0148] S30403: Based on the omission of clinical manifestations, diagnosis-evidence pairs and matching group levels, a set number of class levels are initially screened out as class diagnostic hypotheses using a hierarchical coding agent. Further analysis is performed on the class diagnostic hypotheses to obtain the matching class levels.

[0149] S30404: Based on the omission of clinical manifestations, diagnosis-evidence pairs and matching class levels, a set number of sub-levels are initially screened out as sub-level diagnostic hypotheses using a hierarchical coding agent. Further analysis is performed on the sub-level diagnostic hypotheses to obtain the matching sub-levels.

[0150] S30405: Based on the omission of clinical manifestations, diagnosis-evidence pairs and matching subcategory levels, a hierarchical coding agent is used to initially screen out a set number of standard disease names as name diagnosis hypotheses. Further analysis is performed on the name diagnosis hypotheses to obtain matching standard disease names, and then the corresponding ICD codes are obtained as hierarchical coding results.

[0151] If a clinical manifestation is missed, or if no matching entry is found at a certain level, or if there is ambiguity, the coding fails.

[0152] Introducing the encoded name library as an external knowledge base into the hierarchical encoding module serves two purposes: firstly, it alleviates the "illusion" phenomenon that may occur in large model agents during inference, and secondly, it reduces ambiguity in the model selection process. The specific steps are as follows:

[0153] For missed clinical manifestations First, the process enters the chapter level. At this point, all chapter names in the coding name library are presented as candidate options to the Hierarchical ICD Coder agent for selection. The Hierarchical ICD Coder agent first formulates 3-5 preliminary chapter-level diagnostic hypotheses based on existing evidence and medical record content. Based on these hypotheses, the Hierarchical ICD Coder agent first selects the matching chapter level. After the matching chapter level is determined, the Hierarchical ICD... The Coder intelligent system enters the group-level selection phase. Similar to the chapter-level selection, all groups under the matched chapter level are used as candidates. The diagnostic hypothesis mechanism is still used to initially screen the candidate groups, forming group diagnostic hypotheses at 3-5 group levels. This step simulates the step-by-step confirmation and reasoning process of coders on medical records in the real world. After completing the selection of chapter and group levels, the system continues to select class level, subcategory level, and finally disease code based on the selection of the previous level. If the omitted clinical manifestation can be successfully coded, the coding result is output as the hierarchical coding result. If no matching entry is found at a certain level or there is ambiguity, the omitted clinical manifestation is passed to the subsequent coding recovery module for further processing.

[0154] This step limits the candidate range by introducing different names as candidates at different levels, until the final standard disease name and its corresponding ICD code are determined.

[0155] S305: Construct a coding remediation module: Generate diagnostic suggestions based on a coding name library, use these suggestions to code omitted clinical manifestations that failed to be coded, and obtain suggested coding results, including:

[0156] Using the omitted clinical manifestations of coding failure as input, diagnostic suggestions are obtained through the coding remediation agent (Rescuer agent), and then the suggested coding results are obtained through the suggested coding submodule;

[0157] S30501: The Rescuer agent analyzes the missed clinical manifestations of coding failures to obtain diagnostic suggestions. The specific steps are as follows:

[0158] For missed clinical manifestations of coding failure The Rescuer agent integrates clinical manifestations to provide diagnostic suggestions. At the same time, the diagnostic suggestion is passed to the suggestion coding submodule for further processing;

[0159] S30502: Recommended Encoding Submodule: Utilize the encoding name library to provide diagnostic recommendations. The mapping to ICD encoding and the output of suggested encoding results are as follows:

[0160] (1) Exact matching: Based on the diagnostic suggestions, an exact match is performed between the standard disease names in the coding name library. If successful, the corresponding ICD code is assigned as the suggested coding result.

[0161] First, make diagnostic recommendations A precise match is performed in the coding name library. When a completely identical standard disease name is found, the corresponding ICD code is directly assigned as the suggested coding result.

[0162] (2) Fuzzy matching: If exact matching fails, a similarity score is obtained by weighted fusion of the character similarity, semantic similarity, and diagnostic causal chain similarity between the diagnostic suggestions and the standard disease names in the coding name library. Standard disease names with similarity scores exceeding a set value are selected to generate a set of suggested coding candidates, including:

[0163] If an exact match fails, the system enters the fuzzy matching stage. This stage incorporates a triple similarity fusion mechanism, which integrates character similarity, semantic similarity, and diagnostic causal chain similarity. This mechanism collaboratively evaluates similarity from three dimensions: surface form, deep semantics, and medical logic. Character similarity is among the key similarities assessed. Diagnostic recommendations are primarily based on standardized edit distance. Compared with the standard disease names in the coding name library Literal differences, semantic similarity The vector representation is obtained by using the pre-trained medical language model MedBERT, which then captures the semantic proximity between different expressions. The calculation formula is as follows:

[0164]

[0165] in, Representative diagnostic recommendations The vector representation of , Represents the standard disease names in the code name library. The vector representations of these vectors are generated by the pre-trained model MedBERT;

[0166] Diagnostic causal chain similarity Based on the causal chains of various diseases in the encoded name database, and leveraging the medical knowledge comprehension capabilities of a large language model, this diagnostic causal chain is integrated into a structured medical logical narrative. and in conjunction with diagnostic recommendations Corresponding missed clinical manifestations Semantic comparison is performed to determine the degree of agreement between the disease and the evidence of the patient's actual condition. The calculation formula is as follows:

[0167]

[0168] in, Representative omitted clinical manifestations The vector representation of , Represents standard disease names based on the coded name library. The medical logic description derived from the causal chain of the disease The vector representations of these vectors are obtained through the pre-trained model MedBERT;

[0169] Character similarity semantic similarity Similarity to causal chains in diagnosis The similarity score is obtained by weighting and fusing the results according to different weight parameters. The calculation formula is as follows:

[0170]

[0171] in, Finally, the top-ranked similarity scores were selected. Standard disease names constitute a suggested coding candidate set ;

[0172] (3) Knowledge Graph Assistance and Re-ranking: A suggestion coding knowledge network is constructed based on the suggestion coding candidate set. Based on the suggestion coding knowledge network, standard disease names and corresponding ICD codes that match the patient's actual condition are selected from the suggestion coding candidate set as the suggestion coding results, including:

[0173] Using the medical knowledge graph CPubMedKG, clinical manifestations, common complications, etiologies, and related knowledge of each candidate option in the suggestion coding candidate set are retrieved to form a suggestion coding knowledge network. Next, the re-ranking agent (KG-EnhancedRe-ranker agent) integrates multi-source information to re-rank the data. In other words, the KG-EnhancedRe-ranker agent selects the standard disease name that best matches the patient's actual condition from the candidate set of suggested codes based on the suggested coding knowledge network, and outputs its corresponding ICD code to form the suggested coding result.

[0174] The suggested specific process for encoding submodules is as follows: Figure 3 As shown, the diagnostic suggestions are first matched one by one with the code name library. If a match is successful, the corresponding ICD code is directly output as the suggested encoding result. When a match fails, the character similarity, semantic similarity, and diagnostic causal chain similarity between the diagnostic suggestions and each name in the code name library are calculated. The three are weighted to obtain a similarity score. Several of the most similar code names are selected to form a candidate set of suggested codes. Furthermore, a knowledge network is constructed with the help of a medical knowledge graph. The KG-Enhanced Re-ranker agent integrates the knowledge network to select the code that best matches the patient's actual condition as the suggested encoding result. Finally, it is passed to the encoding result verification module for subsequent verification.

[0175] S306: Encoding Result Verification Module: Verifies the diagnostic encoding results, hierarchical encoding results, and suggested encoding results to obtain the final encoding results.

[0176] Using diagnostic coding results, hierarchical coding results, and suggested coding results as input, the Codes Validator identifies potential duplication or redundancy among codes and ensures that all codes conform to the ICD coding standard, ultimately yielding the final coding result.

[0177] The model proposed in this invention has achieved better results than other methods on the Chinese electronic medical record dataset. The comparison of experimental results is shown in Table 2.

[0178] Table 2 Comparison of Experimental Results

[0179]

[0180] Example 2

[0181] This embodiment provides an automatic ICD encoding system for Chinese electronic medical records, including:

[0182] The medical record text processing module is used to filter patients' original electronic medical records and obtain Chinese electronic medical record text and discharge diagnosis.

[0183] The automatic ICD encoding module is used to assign corresponding ICD codes to Chinese electronic medical record texts using a pre-built automatic ICD encoding model, including:

[0184] Preprocessing of discharge diagnoses leads to standardized diagnoses;

[0185] The association analysis between Chinese electronic medical record text and standardized diagnosis was performed to retrieve the corresponding clinical manifestations and supporting evidence for the standardized diagnosis and generate diagnosis-evidence pairs.

[0186] Based on the diagnosis-evidence pairs and a pre-built coding name library, the normalized diagnosis is mapped to the ICD coding to obtain the diagnosis coding result;

[0187] Differential analysis of Chinese electronic medical record texts and diagnosis-evidence pairs was conducted to uncover missed clinical manifestations;

[0188] Based on the coding name library, the missed clinical manifestations are encoded step by step according to the ICD hierarchical structure. If the encoding is successful, the hierarchical coding result is obtained.

[0189] Diagnostic suggestions are generated based on the coding name database. The diagnostic suggestions are then used to encode the missed clinical manifestations that failed to be coded, resulting in suggested coding results.

[0190] The diagnostic coding results, hierarchical coding results, and suggested coding results are verified to obtain the final coding results.

[0191] The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1 above. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0192] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0193] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0194] Example 3

[0195] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the automatic ICD encoding method for Chinese electronic medical records as described in Embodiment 1 above.

[0196] Example 4

[0197] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the automatic ICD encoding method for Chinese electronic medical records as described in Embodiment 1 above.

[0198] Example 5

[0199] This embodiment provides a computer program product or computer program, including computer instructions stored in a computer-readable storage medium. The processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the automatic ICD encoding method for Chinese electronic medical records described in Embodiment 1 above.

[0200] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0201] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0202] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0203] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0204] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0205] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. An automatic ICD encoding method for Chinese electronic medical records, characterized in that, include: Based on the patient's original electronic medical records, the Chinese electronic medical record text and discharge diagnosis were obtained; Using a pre-built automatic ICD encoding model, assign corresponding ICD codes to Chinese electronic medical record texts, including: Preprocessing of discharge diagnoses leads to standardized diagnoses; The association analysis between Chinese electronic medical record text and standardized diagnosis was performed to retrieve the corresponding clinical manifestations and supporting evidence for the standardized diagnosis and generate diagnosis-evidence pairs. Based on diagnosis-evidence pairs and a pre-built coded name library, standardized diagnoses are mapped to ICD codes to obtain diagnostic coding results. This includes: performing precise matching between the standardized diagnosis and standard disease names in the pre-built coded name library; if successful, assigning the corresponding ICD code as the diagnostic coding result; if precise matching fails, obtaining a similarity score by weighted fusion of character similarity, semantic similarity, and diagnostic causal chain similarity between the standardized diagnosis and standard disease names in the coded name library; selecting standard disease names with similarity scores exceeding a set value to generate a candidate set; constructing a knowledge network based on the candidate set; and performing reasoning comparison based on the knowledge network and diagnosis-evidence pairs to select standard disease names and corresponding ICD codes that match the patient's actual condition from the coded name library as the diagnostic coding result. Differential analysis of Chinese electronic medical record texts and diagnosis-evidence pairs was conducted to uncover missed clinical manifestations; Based on the coding name library, the missed clinical manifestations are encoded step by step according to the ICD hierarchical structure. If the encoding is successful, the hierarchical coding result is obtained. Diagnostic suggestions are generated based on the coding name database. The diagnostic suggestions are then used to encode the missed clinical manifestations that failed to be coded, resulting in suggested coding results. The diagnostic coding results, hierarchical coding results, and suggested coding results are verified to obtain the final coding results.

2. The automatic ICD encoding method for Chinese electronic medical records as described in claim 1, characterized in that, The encoded name library is stored in five levels: chapter, group, class, subcategory, and code. The stored content contains all the codes in the ICD-10 standard and their corresponding standard names; Each data entry includes: chapter range code, chapter name; group range code, group name; class code, class name; suborder code, suborder name; disease code, disease name.

3. The automatic ICD encoding method for Chinese electronic medical records as described in claim 2, characterized in that, The encoded name library sets corresponding disease causal chain knowledge in each standard disease name for medical logic verification in the subsequent diagnostic matching process; The disease causal chain includes four core medical dimensions: etiology, location, pathology, and clinical manifestations.

4. The automatic ICD encoding method for Chinese electronic medical records as described in claim 1, characterized in that, The differential analysis of Chinese electronic medical record texts and diagnosis-evidence pairs uncovered missed clinical manifestations, including: Compare the abnormal clinical manifestations in the Chinese electronic medical record text with the evidence in the diagnosis-evidence pair; All abnormal clinical manifestations in the Chinese electronic medical record text that cannot be covered by evidence in the diagnosis-evidence pair constitute missed clinical manifestations.

5. The automatic ICD encoding method for Chinese electronic medical records as described in claim 1, characterized in that, The method of using diagnostic suggestions to encode omitted clinical manifestations that failed to be coded, and obtaining suggested coding results, includes: The diagnostic suggestions are precisely matched with the standard disease names in the coding name library. If successful, the corresponding ICD code is assigned as the suggested coding result. If an exact match fails, a similarity score is obtained by weighted fusion of the character similarity, semantic similarity, and diagnostic causal chain similarity between the diagnostic suggestions and the standard disease names in the coding name library. Standard disease names with similarity scores exceeding a set value are selected to generate a set of suggested coding candidates. A suggested coding knowledge network is constructed based on the suggested coding candidate set. Based on the suggested coding knowledge network, standard disease names and corresponding ICD codes that match the patient's actual condition are selected from the suggested coding candidate set as suggested coding results.

6. An automatic ICD encoding system for Chinese electronic medical records, characterized in that, include: The medical record text processing module is used to filter patients' original electronic medical records and obtain Chinese electronic medical record text and discharge diagnosis. The automatic ICD encoding module is used to assign corresponding ICD codes to Chinese electronic medical record texts using a pre-built automatic ICD encoding model, including: Preprocessing of discharge diagnoses leads to standardized diagnoses; The association analysis between Chinese electronic medical record text and standardized diagnosis was performed to retrieve the corresponding clinical manifestations and supporting evidence for the standardized diagnosis and generate diagnosis-evidence pairs. Based on diagnosis-evidence pairs and a pre-built coded name library, standardized diagnoses are mapped to ICD codes to obtain diagnostic coding results. This includes: performing precise matching between the standardized diagnosis and standard disease names in the pre-built coded name library; if successful, assigning the corresponding ICD code as the diagnostic coding result; if precise matching fails, obtaining a similarity score by weighted fusion of character similarity, semantic similarity, and diagnostic causal chain similarity between the standardized diagnosis and standard disease names in the coded name library; selecting standard disease names with similarity scores exceeding a set value to generate a candidate set; constructing a knowledge network based on the candidate set; and performing reasoning comparison based on the knowledge network and diagnosis-evidence pairs to select standard disease names and corresponding ICD codes that match the patient's actual condition from the coded name library as the diagnostic coding result. Differential analysis of Chinese electronic medical record texts and diagnosis-evidence pairs was conducted to uncover missed clinical manifestations; Based on the coding name library, the missed clinical manifestations are encoded step by step according to the ICD hierarchical structure. If the encoding is successful, the hierarchical coding result is obtained. Diagnostic suggestions are generated based on the coding name database. The diagnostic suggestions are then used to encode the missed clinical manifestations that failed to be coded, resulting in suggested coding results. The diagnostic coding results, hierarchical coding results, and suggested coding results are verified to obtain the final coding results.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the automatic ICD encoding method for Chinese electronic medical records as described in any one of claims 1-5.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the automatic ICD encoding method for Chinese electronic medical records as described in any one of claims 1-5.

9. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the automatic ICD encoding method for Chinese electronic medical records as described in any one of claims 1-5.

Images