Intelligent verification method and system for traditional Chinese medicine prescription quality control based on multiple national standards

Abstract

The application provides a method and system for intelligent verification of TCM medical record treatment method, prescription and drug quality control based on multiple national standards, which comprises: constructing a knowledge base fused by disease, syndrome, treatment method, pharmacopoeia, internal medicine and TCM drug quality control standard terms; obtaining medical record text and identifying disease, syndrome, treatment method, prescription and TCM drug by word segmentation; mapping non-standard expressions into standard terms based on the knowledge base, outputting structured five-tuple and inputting AI unit for multi-stage semantic analysis to obtain semantic feature vector, candidate knowledge support set and evidence chain abstract; performing two-dimensional linkage verification according to five-tuple, feature vector and knowledge base to generate field missing and logic inconsistency defect labels and push early warning; reusing verification for archived medical records, calculating completeness rate, consistency rate and total defect rate and generating a report; and storing desensitized log data into a backup array and generating hash association storage. The application can realize standardization, intelligentization and traceability of TCM medical record treatment method, prescription and drug quality control.

Description

Technical Field

[0001] This application relates to the field of medical information data processing technology, and in particular to an intelligent verification method and system for quality control of TCM medical records, theories, prescriptions, and medicines based on multiple national standards. Background Technology

[0002] With the continuous advancement of the digital quality control system for Traditional Chinese Medicine (TCM), the National Health Commission, the State Administration of Traditional Chinese Medicine, the National Medical Products Administration, and the China Association of Traditional Chinese Medicine have successively issued five independent standards: "TCM Clinical Diagnosis and Treatment Terminology Part 1: Diseases," "TCM Clinical Diagnosis and Treatment Terminology Part 2: Syndromes," "TCM Clinical Diagnosis and Treatment Terminology Part 3: Treatment Methods," the 2025 edition of the "Pharmacopoeia of the People's Republic of China," and the "Quality Control Standards for Clinical Research of Traditional Chinese Medicine." These five standards only independently define and standardize terms related to diseases, syndromes, treatment methods, and traditional Chinese medicine. The standards for diseases, syndromes, and treatment methods are fragmented, lacking officially established correlations between diseases, syndromes, and treatment methods. They also lack a comprehensive, integrated system of theory, method, prescription, and medicine that fully covers syndromes, treatments, prescriptions, and medicines. For a long time, there has been a lack of unified and authoritative standardized criteria for quality control of the complete theory, method, prescription, and medicine content in TCM medical records. Currently, TCM medical records generally suffer from three major quality control deficiencies: omissions of key fields related to diseases, syndromes, treatment methods, prescriptions, and traditional Chinese medicines; logical disconnect between theory, method, prescription, and medicine; and non-standard use of clinical terminology.

[0003] Two mainstream digital processing systems have emerged in the industry: intelligent prescription matching systems and knowledge graph-based medical record quality control systems. However, neither of these existing technologies can fully meet the quality control requirements of the entire medical record chain under the integration of four standards.

[0004] The first type of similar prior art is an automatic prescription configuration system based on digital symptom analysis disclosed in patent application CN121725975A. This system is designed for online consultations where patients input their symptoms. The software consists of a five-layer structure: a database module, a semantic analysis module, a first encoding module, a second encoding module, and a retrieval module. The database module only has two types of storage units: the first unit stores the user's colloquial symptom description text, and the second unit only presets a two-dimensional matching vector between symptoms and prescriptions. It does not incorporate a treatment terminology database and lacks a five-level chained storage structure for diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines. The semantic analysis module only uses a general word segmentation model, splitting only the core symptom words and degree modifiers, failing to convert free medical record text into standard disease, syndrome, treatment, prescription, and traditional Chinese medicine terminology. The retrieval module only outputs a prescription recommendation list and lacks functions for batch importing medical records, defect screening, and quality control indicator statistics. The existing technology has the following defects: (1) The underlying knowledge base constructs a binary correspondence between symptoms and prescriptions, lacks treatment method dimension association rules, and cannot cover the complete chain of TCM syndrome differentiation of "theory, method, prescription and medicine"; (2) It does not connect with five authoritative standards and has no standardized alignment architecture for terminology; (3) It is only for prescription recommendations in consultation and does not support batch quality control of hospital archived medical records, and cannot solve the difficulties in hospital quality control such as missing fields, inconsistency between theory and method, and non-standard terminology; (4) The knowledge base is a static preset library and has no expert consensus modeling and clinical iteration optimization mechanism.

[0005] The second category of similar prior art is a method and system for evaluating the quality of TCM medical records based on knowledge graphs, disclosed in patent application CN122025053A. This system is used for logical review of TCM medical records within hospitals, and its architecture consists of six fixed units: a medical record parsing module, a knowledge graph construction module, a consistency verification module, an interpretation and identification module, a fusion and discrimination module, and a quality judgment module. The knowledge graph construction in this scheme relies solely on automatically extracting explicit co-occurrence relationships from historical hospital medical record texts to generate association edges for syndromes, treatment methods, and prescriptions, rather than establishing a standardized comparison system based on five authoritative standards. The medical record parsing module only performs text character cleaning and keyword extraction, without designing a standardized terminology mapping subunit. The consistency verification module only performs logical verification on existing syndrome, treatment, and prescription co-occurrence relationships within the medical record, lacking an independent standardized comparison knowledge base as a unified judgment benchmark. Furthermore, the system lacks a pre-writing real-time early warning interaction unit for medical record writing, and can only conduct post-review after all medical records have been archived. The existing technology has the following defects: (1) The knowledge graph association is derived from the co-occurrence of historical medical record texts. It does not build a standardized comparison dataset based on the five standards. The graph rules of different institutions are very different and there is no unified quality control standard. (2) It does not rely on the standards of disease, syndrome and treatment to standardize medical record terminology. It also does not integrate the pharmacopoeia and Chinese medicine quality control standards to build a complete theory, method and prescription comparison table. (3) There is no multidisciplinary expert consensus modeling process. It cannot absorb the experience of clinical experts to correct the matching logic. (4) It has not established a four-layer quality control system of terminology uniqueness verification, association logic verification, clinical adaptation verification and compliance desensitization verification. (5) It only supports post-archiving quality control and does not have a real-time warning function for physician writing. (6) It cannot automatically quantify and output standardized quality control indicators such as the completeness rate of key fields and the consistency rate of theory, method and prescription. It can only output qualitative classification labels.

[0006] Furthermore, some primary care and traditional Chinese medicine hospitals still use manual medical record quality control methods, relying on management personnel to review each medical record against five standards, manually checking the completeness of entries, the reasonableness of matching, and the standardization of terminology, and manually recording defects in a logbook. This method lacks a digital analysis, matching, and statistical framework, resulting in extremely low review efficiency, high subjectivity, inconsistent standards, and the inability to implement full-scale verification.

[0007] In summary, existing intelligent prescription matching systems lack treatment method dimensions and medical record quality control capabilities; quality control systems based on medical record co-occurrence maps lack five unified standards and an expert iterative database construction mechanism; and manual quality control has extremely low levels of standardization and digitization. None of these three existing technologies possess a complete architecture that integrates five standards, relies on expert consensus modeling, iteratively generates standardized syndrome-treatment-prescription-medication comparison rules based on clinical feedback, and simultaneously considers real-time front-end alerts for physicians' writing and quantitative quality control at the back-end of archived medical records.

[0008] Furthermore, existing electronic medical record quality control technologies primarily focus on explicit issues such as field completeness, format standardization, and writing deadlines, lacking the ability to effectively identify the dialectical and therapeutic logical relationships reflected in syndromes, treatment methods, and prescriptions in traditional Chinese medicine medical records. In medical record scenarios where free text, semi-structured text, and non-standardized expressions coexist, relying solely on fixed rules, keyword matching, or shallow text comparison is insufficient to accurately identify implicit semantic relationships, terminology variations, and chain-like logical conflicts. It lacks multi-stage reasoning capabilities and traceable evidence chain generation capabilities for complex issues such as "whether the syndrome supports the treatment method," "whether the treatment method derives the prescription," and "partially reasonable but overall dialectical deviation," making it difficult to meet the requirements of large-scale, real-time, and interpretable quality control.

[0009] The patent information disclosed in this background section is only used to enhance the understanding of the prior art in this field and should not be regarded as the prior art being able to solve all the technical problems to be solved by this invention. Summary of the Invention

[0010] This application provides an intelligent verification method and system for quality control of TCM medical records based on multiple national standards. To address the problems of existing technologies struggling to stably identify complex semantic content in TCM medical records and making in-depth judgments on the chain relationships of theory, method, prescription, and medicine, this application further sets up an intelligent reasoning unit in the quality control system of theory, method, prescription, and medicine. This unit works in collaboration with the structured parsing of medical records, knowledge base matching, and dual-dimensional linkage verification unit. It performs multi-stage semantic parsing, terminology standardization and disambiguation, knowledge enhancement retrieval, reconstruction of the diagnosis and treatment logic chain, and generation of evidence interpretation on the free text of medical records. This improves the system's ability to understand complex medical record content, adapt to non-standard expressions, identify implicit dialectical conflicts, and provide interpretable output of quality control conclusions.

[0011] Furthermore, addressing the structural deficiencies of existing intelligent prescription matching systems, such as the lack of a treatment dimension and standardized terminology alignment architecture, the reliance of knowledge graph quality control systems on historical medical record co-occurrence and the absence of a unified quality control benchmark, and the low efficiency and inconsistent standards of traditional manual quality control, this application uses five authoritative standards as the underlying benchmark. It establishes a correlation between diseases, syndromes, and treatment methods through expert consensus and integrates pharmacopoeia and traditional Chinese medicine quality control standards to construct a unified judgment benchmark for the entire chain of disease, syndrome, treatment, prescription, and medicine. On this basis, a five-level chain-linked verification architecture is built to simultaneously achieve field integrity screening and two-level matching verification. It also provides an independent knowledge base editing and clinical feedback iteration mechanism, along with four layers of data quality control rules to support multi-domain experts in independently correcting matching relationships. It achieves a complete closed loop of real-time early warning and backfilling at the physician's writing front end and quantitative statistics at the end of the medical record back end. It takes into account the dual application scenarios of clinical physicians assisting in the writing of diagnostic and treatment guidelines and medical management personnel in batch quality control, and solves the common industry problems of missing medical record fields, non-standard terminology, and disconnect between theory, method, prescription, and medicine logic in a one-stop manner.

[0012] Firstly, this application provides an intelligent verification method for quality control of TCM medical records based on multiple national standards, including the following: Construct a knowledge base that integrates multiple types of standard terminology; The system acquires electronic medical record text entered by the front-end physician workstation, performs word segmentation on the electronic medical record text, identifies disease text fragments, syndrome text fragments, treatment text fragments, prescription text fragments, and traditional Chinese medicine text fragments to obtain original text entities; based on the multi-class standard terminology fusion knowledge base, it maps non-standard expressions in the original text entities to standard terms, outputting structured medical record quintuple data, which includes standardized diseases, standardized syndromes, standardized treatment methods, standardized prescriptions, and standardized traditional Chinese medicines; the structured medical record quintuple data is input into an artificial intelligence inference unit, which performs multi-stage semantic parsing on the structured medical record quintuple data to obtain semantic feature vectors, candidate knowledge support sets, and evidence chain summaries; Based on the structured medical record quintuple data, the semantic feature vector, the candidate knowledge support set, and the multi-class standard terminology fusion knowledge base, a two-dimensional theory, method, prescription, and drug linkage verification is performed to generate defect markers. Based on the defect markers and the evidence chain summary, early warning information is pushed to the front-end physician workstation. Retrieve archived medical records from the hospital's final medical record database, reuse the dual-dimensional theory, method, prescription, and drug linkage verification for each archived medical record, and record the defect information of each archived medical record; calculate the completeness rate of key fields, the consistency rate of theory, method, prescription, and drug, and the total defect rate of medical record theory and method information in batches based on the defect information of all archived medical records, and generate defect distribution ledgers and quality control analysis reports according to the department dimension and / or physician dimension and / or disease dimension; The log data is anonymized and stored in a distributed backup storage array. A unique hash code is generated for the anonymized log data, and the unique hash code is associated with and stored with the anonymized log data. The log data includes: knowledge base construction and modification logs generated during the construction and modification of the multi-category standard terminology fusion knowledge base; front-end verification interaction logs generated during the execution of the dual-dimensional theory, law, prescription, and drug linkage verification; and final medical record quality control logs generated during the final quality control of the archived medical records.

[0013] Furthermore, a multi-category standard terminology fusion knowledge base is constructed, including: The original texts of six categories of preset terms were imported, namely, the disease standard terminology library, the syndrome standard terminology library, the treatment standard terminology library, the pharmacopoeia standard terminology library, the traditional Chinese medicine internal medicine standard terminology library, and the traditional Chinese medicine quality control standard terminology library. After text cleaning, the following sets of standard terms were formed: disease standard terminology set, syndrome standard terminology set, treatment standard terminology set, prescription standard terminology set, and traditional Chinese medicine standard terminology set. Based on the preset mapping rules of diseases, syndromes and treatments, a built-in comparison relationship of diseases, syndromes and treatments is generated; Acquire historical clinical data, and iteratively correct the disease-syndrome comparison relationship, syndrome-treatment method comparison relationship, and treatment method-prescription-medicine matching relationship based on the historical clinical data to generate a theory-treatment-prescription-medicine comparison table. The theory, method, prescription, and medicine comparison table is sequentially subjected to terminology uniqueness verification, association logic verification, clinical fit verification, and compliance desensitization and evidence preservation processing. The standardized five-level comparison dataset of diseases, syndromes, treatment methods, prescriptions, and Chinese medicines is output and stored in the database.

[0014] Furthermore, the aforementioned theory-method-formula-medicine comparison table is subjected to terminology uniqueness verification, association logic verification, clinical fit verification, and compliance desensitization and evidence preservation processing in sequence. A standardized five-level comparison dataset of diseases, syndromes, treatment methods, formulas, and traditional Chinese medicines is output and stored in the database, including: The terms in the comparison table of theory, method, prescription and medicine are checked for duplicates and duplicate terms and vague expressions are deleted to obtain a deduplicated comparison table. The chain matching relationships of diseases, syndromes, treatments, prescriptions, and Chinese medicines in the deduplicated comparison table are verified one by one to see if they conform to the TCM syndrome differentiation rules and to remove contradictory associations of pathogenesis, so as to obtain the logically verified comparison table. Based on the clinical historical data, the matching entries in the logically validated lookup table that conflict between general standards and specialist diagnosis and treatment are filtered to obtain a clinically adapted lookup table; The clinically adapted control dataset is anonymized, a unique hash value is generated and associated with the anonymized control dataset, the knowledge base construction and modification log is retained, and a standardized five-level control dataset of disease, syndrome, treatment, prescription and Chinese medicine is output and stored in the database.

[0015] Furthermore, the multi-stage semantic parsing includes: The electronic medical record text is subjected to contextual semantic segmentation, cross-sentence relationship identification, and semantic role labeling to identify key medical semantics in chief complaint, present medical history, four diagnostic information, syndrome differentiation description, treatment method description, and prescription description, and to obtain semantic feature vectors. The extracted disease, syndrome, treatment, prescription, and Chinese medicine expressions are subjected to terminology normalization and semantic disambiguation processing. Diagnostic and treatment terms that are similar in word meaning but different in semantic meaning are distinguished to obtain standardized terms after disambiguation. Based on the semantics of medical record input, similar knowledge retrieval and multi-hop association candidate expansion are performed to generate a candidate knowledge support set; It assists in reasoning about the semantic relationships between diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines, and identifies implicit conflict situations such as local statements being valid but the overall diagnostic direction deviating, similar wording but inconsistent medical semantics, and local reasonableness but inconsistent global links. Based on the identified implicit conflict situations, it generates an evidence chain summary, which includes an explanation of the cause of the problem, a hint of the conflict location, and suggestions for revision.

[0016] Furthermore, the dual-dimensional theory-method-prescription-drug linkage verification includes: Traverse the structured medical record quintuple data and determine whether there are blank or unfilled markers in the disease field, syndrome field, treatment method field, prescription field, and traditional Chinese medicine field. If they exist, generate field missing markers. Based on the multi-category standard terminology fusion knowledge base, the matching relationship between disease and syndrome, the matching relationship between syndrome and treatment method, and the matching relationship between treatment method and prescription are verified sequentially. If any matching relationship does not satisfy the corresponding relationship in the knowledge base, a logical inconsistency defect label is generated. Based on the semantic feature vector, the candidate knowledge support set, and the evidence chain summary, the logical inconsistency defect label is weighted by confidence and classified by conflict type, and an enhanced logical inconsistency defect label is output. Based on the missing field marker and / or the enhanced logical mismatch defect label, along with the evidence chain summary, a warning message including defect location and revision direction suggestions is pushed to the front-end physician workstation, and the front-end verification interaction log is retained.

[0017] Furthermore, after pushing warning information including defect location and revision direction suggestions to the front-end physician workstation based on the missing field marker and / or the enhanced logical mismatch defect label, along with the evidence chain summary, and retaining the front-end verification interaction log, the process also includes a syndrome-driven automatic backfilling and cyclic verification release step: Once a standardized disease is obtained and selected by the front-end physician workstation, based on the multi-category standard terminology fusion knowledge base, a standardized syndrome matching the standardized disease is pushed. After the standardized syndrome is selected, the corresponding standardized treatment method is automatically matched and filled into the treatment method input box. At the same time, a prescription recommendation list matching the standardized disease, standardized syndrome and standardized treatment method is pushed. When manual modifications to the automatically backfilled treatment methods or autonomous selection of syndromes and prescriptions are obtained, the matching relationship between the modified diseases and syndromes, the matching relationship between syndromes and treatment methods, and the matching relationship between treatment methods and prescriptions are reviewed again based on the knowledge base of the multi-category standard terms. If a mismatch occurs, the reason for the conflict is highlighted and the recommended entries are refreshed. Repeatedly perform field integrity checks and chain matching checks until there are no missing field markers and no logical mismatch defect labels, then allow the medical record to be saved.

[0018] Furthermore, the formula for calculating the completeness rate of the key fields is: the number of medical records without omissions divided by the total number of medical records multiplied by 100%; The formula for calculating the consistency rate of theory, treatment, prescription, and medication is: the number of cases with complete matching of disease, syndrome, treatment method, and prescription / medication divided by the total number of cases multiplied by 100%. The formula for calculating the total defect rate of the medical record information is: the number of medical records with any missing field or mismatched field divided by the total number of medical records multiplied by 100%.

[0019] Secondly, this application provides an intelligent verification system for quality control of TCM medical records based on multiple national standards, including the principles, methods, prescriptions, and medications. The system comprises: The front-end interactive hardware layer, including the clinical physician workstation and the medical record quality control management PC, is connected to the central computing hardware layer through bidirectional communication via the hospital's local area network. The central computing hardware layer includes a database server and an algorithm computing server, which are directly connected by optical fiber to achieve data interaction. The database server stores a knowledge base of multiple standard terms, a full set of medical record data, and quality control log evidence records. The algorithm computing server carries semantic parsing, matching and verification, artificial intelligence reasoning, and quantitative statistical calculation programs. A distributed backup storage array, continuously synchronized with the database server via a data bus, is used for off-site de-identified backup; The software system deployed on the central computing hardware layer includes: The multi-standard terminology fusion knowledge base construction module is used to import the original text of six types of preset terms, generate a standardized five-level comparison dataset of diseases, syndromes, treatments, prescriptions, and Chinese medicines based on preset disease syndrome treatment mapping rules and clinical historical data, and store it in the database server. The electronic medical record structured parsing module is used to perform word segmentation and standard terminology mapping on the electronic medical record text uploaded by the front-end physician workstation, and output structured medical record quintuple data. The AI ​​reasoning unit, whose input is connected to the output of the electronic medical record structured parsing module, is used to receive the structured medical record quintuple data, perform multi-stage semantic parsing, structured extraction, terminology normalization and semantic disambiguation, knowledge enhancement retrieval, diagnosis and treatment logic chain reconstruction and semantic consistency judgment, and evidence chain interpretation generation on the structured medical record quintuple data, and output semantic feature vectors, candidate knowledge support sets and evidence chain summaries; A dual-dimensional theory-method-formula-medicine linkage verification module is provided. Its first input is connected to the output of the electronic medical record structured parsing module to obtain the structured medical record five-tuple data. Its second input is connected to the output of the artificial intelligence inference unit to obtain the semantic feature vector, the candidate knowledge support set, and the evidence chain summary. Its third input is connected to the database server to obtain the standardized disease, syndrome, treatment method, formula, and traditional Chinese medicine five-level comparison dataset. This module performs field integrity verification and chain matching verification of disease, syndrome, treatment method, formula, and traditional Chinese medicine based on the structured medical record five-tuple data, the semantic feature vector, the candidate knowledge support set, the evidence chain summary, and the five-level comparison dataset. It generates defect markers and pushes warning information to the front end along with the evidence chain summary. The medical record batch quality control and quantitative statistics module is used to retrieve archived medical records from the hospital's final medical record database for batch quality control verification, calculate the completeness rate of key fields, the consistency rate of theory, method, prescription and drug, and the total defect rate of medical record theory and method information, and generate a defect distribution ledger and quality control analysis report. The end-to-end log storage and compliance module is used to de-identify log data and store it in the distributed backup storage array. It also generates a unique hash code for the de-identified log data and associates this unique hash code with the de-identified log data for storage. The log data includes: knowledge base construction and modification logs generated by the multi-standard terminology fusion knowledge base construction module during knowledge base construction and modification; front-end verification interaction logs generated by the dual-dimensional theory, law, prescription, and drug linkage verification module during verification execution; and final medical record quality control logs generated by the batch quality control and quantitative statistics module during final quality control of archived medical records.

[0020] Furthermore, the multi-standard terminology fusion knowledge base construction module includes a standard terminology import subunit, a mapping rule generation subunit, a clinical iterative optimization subunit, and a four-layer integrated quality control verification subunit connected in sequence. The standard terminology import subunit is equipped with four data source interfaces, which are respectively connected to the disease standard terminology import interface, syndrome standard terminology import interface, treatment standard terminology import interface, pharmacopoeia standard terminology import interface, traditional Chinese medicine internal medicine standard terminology import interface and traditional Chinese medicine quality control standard terminology import interface. The four-layer integrated quality control verification subunit internally integrates a terminology uniqueness verification unit, an association logic verification unit, a clinical adaptation verification unit, and a compliance de-identification and evidence storage unit. The output of the multi-standard terminology fusion knowledge base construction module is connected to the five-level control dataset storage area. The standardized disease, syndrome, treatment, prescription, and Chinese medicine five-level control dataset in the five-level control dataset storage area includes a complete set of standard disease entries, a complete set of standard syndrome entries, a complete set of standard treatment entries, an in-hospital expanded Chinese medicine entries, a disease syndrome and treatment built-in correlation control group, and a matching prescription and medicine entry group.

[0021] Furthermore, the artificial intelligence inference unit includes: The multi-stage semantic parsing subunit is used to perform contextual semantic segmentation, cross-sentence relationship identification, and semantic role labeling on the electronic medical record text, identify key medical semantics in the chief complaint, present medical history, four diagnostic information, syndrome differentiation description, treatment method description, and prescription description, and output semantic feature vectors. The structured extraction subunit is used to extract diseases, syndromes, treatments, prescriptions, Chinese medicines and their related descriptions from the electronic medical record text, generate structured objects that can be verified later, and perform completeness perception marking on the missing, weak expression and abnormal position of key diagnostic and treatment elements. The terminology normalization and semantic disambiguation subunit is used to combine the multi-class standard terminology fusion knowledge base and synonym mapping relationship to map non-standard expressions into standard terms, and to disambiguate diagnostic and treatment terms that are similar in word meaning but different in semantic meaning. The knowledge-enhanced retrieval subunit is used to vectorize knowledge items related to diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines, and to perform similar knowledge retrieval and multi-hop association candidate expansion based on the semantics of medical record input, locate evidence fragments, and generate candidate knowledge support sets. The diagnostic logic chain reconstruction and semantic consistency judgment subunit is used to assist in judging the semantic association between diseases, syndromes, treatment methods, prescriptions, and Chinese medicines, reconstruct the path of syndrome differentiation and treatment, identify implicit conflict situations such as local expressions being valid but the overall syndrome differentiation direction deviating, similar wording but inconsistent medical semantics, and local rationality but inconsistent global links, and generate conflict type labels. The evidence chain interpretation and generation subunit is used to generate a problem cause description, conflict location prompt, revision direction suggestion and evidence chain summary that can be manually reviewed based on the conflict type label, the candidate knowledge support set and the completeness awareness mark, and output the evidence chain summary to the dual-dimensional theory, law, prescription and drug linkage verification module.

[0022] The intelligent verification method and system for quality control of TCM medical records, theories, methods, prescriptions, and medications based on multiple national standards provided in this application have at least the following beneficial effects: 1) This application constructs a multi-standard terminology fusion knowledge base based on five authoritative standards as the underlying benchmark. Among them, the disease standard, syndrome standard, and treatment method standard unify the terminology of diagnosis and treatment principles in medical records from the source. Addressing the issue of the three sets of standards being independent and lacking built-in connections, a comparative relationship between diseases, syndromes, and treatment methods is established through preset mapping rules and iterative correction using clinical historical data. Furthermore, the application integrates pharmacopoeia and traditional Chinese medicine quality control standards to improve the matching rules for prescriptions and drugs, ultimately forming a standardized five-level comparative dataset of diseases, syndromes, treatment methods, prescriptions, and traditional Chinese medicines, which is then stored in the database. This knowledge base serves as the sole underlying judgment benchmark for the system, eliminating the shortcomings of existing technologies such as chaotic terminology, large differences in matching rules, and the lack of a unified quality control standard, thus providing a unified judgment basis for the quality control of medical records from different institutions and specialties. Meanwhile, the aforementioned knowledge base adopts an independent editing and clinical feedback iteration mechanism, and is equipped with four layers of data quality control rules: terminology uniqueness verification, correlation logic verification, clinical adaptation verification, and compliance de-identification and evidence storage. It supports experts from multiple fields to independently modify and expand matching relationships, which can overcome the defects of deep coupling between rules and code in existing technologies and the fact that only R&D personnel can modify them, so that the knowledge base can continuously adapt to the diagnostic and treatment characteristics of different specialties.

[0023] 2) This application utilizes an electronic medical record structured parsing module to perform word segmentation and standard terminology mapping on the medical record text entered by the front-end physician workstation. This converts colloquial expressions, aliases, and non-standard custom terms in the original text entities into standard, regulated terms, outputting structured medical record five-tuple data. This enables standardized alignment of terminology in freely written medical records. Based on this, an artificial intelligence reasoning unit performs multi-stage semantic parsing on the structured medical record five-tuple data, extracting semantic feature vectors, candidate knowledge support sets, and evidence chain summaries. The reasoning results are constrained by a standard terminology system, disease-syndrome correspondence, syndrome-treatment correspondence, and treatment-prescription-medication correspondence, distinguishing it from the boundless free generation method. Through a hybrid architecture combining rule constraints, knowledge guidance, and model understanding, this application can not only identify the completeness of medical record entries but also the standardization of terminology and the self-consistency of the diagnostic chain, upgrading the ability from surface-level field inspection to the logical judgment of traditional Chinese medicine connotations.

[0024] 3) This application sets up two independent operational sub-units in parallel: a key field integrity verification and a disease, syndrome, treatment method, prescription, and Chinese herbal medicine chain matching verification, forming a two-dimensional theory-method-prescription-medicine linkage verification architecture. The integrity verification sub-unit traverses the structured medical record five-tuple data, identifies whether there are blank or unfilled fields for disease, syndrome, treatment method, prescription, and Chinese herbal medicine, and generates field missing markers; the chain matching verification sub-unit, based on a multi-class standard terminology fusion knowledge base, sequentially verifies the matching relationship between disease and syndrome, the matching relationship between syndrome and treatment method, and the matching relationship between treatment method and prescription. If any matching relationship does not meet the corresponding relationship in the knowledge base, a logical mismatch defect label is generated. The above linkage verification architecture fully covers the five-level logical chain of disease, syndrome, treatment method, prescription, and Chinese herbal medicine in traditional Chinese medicine syndrome differentiation and treatment, and can simultaneously identify four types of defects: field omissions, theory-method inconsistencies, prescription mismatches, and non-standard terminology, significantly improving the completeness and accuracy of quality control.

[0025] 4) This application includes a bidirectional data channel within the early warning interaction subunit that directly connects to the writing interface of the clinical physician's workstation. When a physician enters and selects a standardized disease, the system pushes matching standardized syndromes based on the knowledge base. After selecting a standardized syndrome, the system automatically matches and backfills the corresponding standardized treatment method into the treatment method input box, while simultaneously pushing a list of recommended prescriptions. This automatic backfilling and recommendation operation relies on a complete theory-method-prescription-medicine comparison table, ensuring from the writing source that the five core fields—disease, syndrome, treatment method, prescription, and Chinese medicine—are complete. Standardized terminology is used throughout, and logical consistency of theory-method-prescription-medicine is guaranteed through comparison relationships. When a physician manually modifies the automatically backfilled content, the system re-verifies the matching relationships. If logical discrepancies are found, the conflict reason is highlighted and the recommended entries are refreshed. Only after repeatedly performing integrity checks and chain-matching checks until there are no missing field markers and no logical discrepancy defect tags can the medical record be saved, forming a complete closed loop from writing to verification to correction to release, overcoming the deficiency of existing systems that can only perform post-evaluation after archiving.

[0026] 5) This application directly connects to the hospital's final medical record database through a batch quality control and quantitative statistics module. It performs a two-dimensional, linked verification logic for each archived medical record, automatically recording defect information. Based on this, the system batch calculates three standardized quality control indicators: the completeness rate of key fields, the consistency rate of medical records' ...

[0027] 6) This application obtains knowledge base construction and modification logs, front-end verification interaction logs, and final medical record quality control logs as log data through a full-process log storage and compliance module. After anonymizing the log data, it is stored in a distributed backup storage array. A unique hash code is generated for each anonymized log data, and this unique hash code is associated with the anonymized log data for storage. The above evidence storage operations ensure full-process traceability and ownership confirmation, meeting the compliance requirements of hospital medical record management internal audits and data intellectual property registration. Furthermore, the hardware in this application adopts a layered architecture, including a front-end interaction hardware layer, a central computing hardware layer, and a distributed backup storage array. This supports multi-terminal concurrent real-time quality control in tertiary hospitals and can also be easily deployed in smaller, grassroots institutions, balancing the flexibility of large-scale deployment with lightweight implementation. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0029] Figure 1 A flowchart of an intelligent verification method for quality control of TCM medical records based on multiple national standards is provided for embodiments of this application. Figure 2 A structural diagram of an intelligent verification system for quality control of TCM medical records based on multiple national standards is provided in this application embodiment. Figure 3 A schematic diagram of the internal structure of the multi-standard terminology fusion knowledge base construction module provided in the embodiments of this application; Figure 4 This is a schematic diagram of the internal structure of the artificial intelligence inference unit provided in an embodiment of this application; Figure 5 This is a schematic diagram of multi-source result joint verification and result aggregation provided in the embodiments of this application.

[0030] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0031] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0032] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0033] The technical solution of this application and how it solves the above-mentioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0034] Example 1: This application provides an intelligent verification method for quality control of TCM medical records based on multiple national standards, such as the principles, methods, prescriptions, and medications. Figure 1 As shown, the method includes the following steps S1-S5.

[0035] S1: Construct a knowledge base that integrates multiple types of standard terms.

[0036] In some embodiments, step S1 specifically includes steps S1-1 to S1-4.

[0037] S1-1: Import the original texts of six categories of preset terms: disease standard terminology library, syndrome standard terminology library, treatment standard terminology library, pharmacopoeia standard terminology library, traditional Chinese medicine internal medicine (new century fifth edition) standard terminology library, and traditional Chinese medicine quality control standard terminology library. After text cleaning, form disease standard terminology set, syndrome standard terminology set, treatment standard terminology set, prescription standard terminology set, and traditional Chinese medicine standard terminology set.

[0038] Specifically, the original texts of the above six categories of preset terms are: "Traditional Chinese Medicine Clinical Diagnosis and Treatment Terminology Part 1: Diseases", "Traditional Chinese Medicine Clinical Diagnosis and Treatment Terminology Part 2: Syndromes", "Traditional Chinese Medicine Clinical Diagnosis and Treatment Terminology Part 3: Treatment Methods", the 2025 edition of the "Pharmacopoeia of the People's Republic of China", "Internal Medicine of Traditional Chinese Medicine (New Century Fifth Edition)", and "Quality Control Standards for Clinical Research of Traditional Chinese Medicine". After importing and cleaning, the syndrome standard terminology set contains 1369 sets of national standard syndromes, and the treatment method standard terminology set contains 1168 sets of national standard treatment methods.

[0039] Text cleaning includes: removing invalid characters such as spaces, tabs, and line breaks from the original text; eliminating duplicate term entries; and normalizing different expressions of the same term, such as unifying "wind-heat attacking the lung syndrome" and "wind-heat attacking the lung syndrome" into the same standard expression.

[0040] S1-2: Based on the preset mapping rules of diseases, syndromes and treatments, generate built-in comparison relationships of diseases, syndromes and treatments.

[0041] The pre-defined mapping rules for diseases, syndromes, and treatments were developed in advance by a multidisciplinary team of experts, including renowned traditional Chinese medicine practitioners, pharmaceutical experts, and medical management experts. An initial mapping rule library was formed through multiple rounds of offline consensus discussions. Addressing the issue that the three systems of disease standards, syndrome standards, and treatment standards are independent and lack inherent connections, the expert team used the theory of pathogenesis differentiation as a bridge to establish logical mapping relationships between the three sets of standards.

[0042] S1-3: Obtain historical clinical data, iteratively correct the relationship between disease and syndrome, the relationship between syndrome and treatment method, and the relationship between treatment method and prescription drug based on the historical clinical data, and generate a theory-treatment-prescription-drug comparison table.

[0043] The clinical historical data was derived from archived medical records of several national key specialties, including clinical diagnosis and treatment data from proctology, neurology, cardiology, and gynecology. The iterative correction method involved comparing the co-occurrence relationships of diseases, syndromes, treatment methods, prescriptions, and traditional Chinese medicines in the clinical historical data with the current comparison table, calculating the matching hit rate and conflict rate. For conflicting items, the expert team reviewed and adjusted the comparison relationships, iterating through multiple rounds until the matching hit rate reached a preset threshold. During the implementation process, 17 pieces of feedback from clinical departments were collected, and all were incorporated into the comparison table after expert review, generating the initial comparison table of theories, methods, prescriptions, and medicines.

[0044] S1-4: Perform terminology uniqueness verification, association logic verification, clinical fit verification, and compliance desensitization and evidence storage processing on the theory, method, prescription and medicine comparison table in sequence, and output a standardized five-level comparison dataset of diseases, syndromes, treatment methods, prescriptions and Chinese medicines and store it in the database.

[0045] In some embodiments, step S1-4 specifically includes steps S1-4-1 to S1-4-4.

[0046] S1-4-1, Terminology Uniqueness Verification: Check the terms in the Theory, Method, Formula, and Drug Comparison Table for duplicates and delete duplicate terms and vague expressions to obtain a deduplicated comparison table.

[0047] The plagiarism detection uses a combination of precise string matching and semantic similarity calculation. The preset similarity threshold is 0.85. When the semantic similarity between two terms is greater than this threshold, they are marked as suspected duplicate entries and are automatically merged by the system or confirmed by experts.

[0048] S1-4-2, Association Logic Verification: Verify the chain matching relationship of diseases, syndromes, treatments, prescriptions, and Chinese medicines in the deduplicated comparison table one by one to see if it conforms to the TCM syndrome differentiation rules and eliminate contradictory associations of pathogenesis, and obtain the comparison table after logical verification.

[0049] Traditional Chinese medicine (TCM) diagnostic principles include: the pathogenesis of a syndrome should match the efficacy classification of the treatment method; for example, heat syndromes should not be treated with warming methods, and deficiency syndromes should not be treated with drastic purgatives. The target of the treatment method should correspond to the efficacy and indications of the prescription. The correlation logic verification subunit has a built-in rule library for detecting pathogenesis conflicts. If a known contradictory relationship is detected between the syndrome's pathogenesis and the treatment's efficacy, the entry is marked as suspicious and moved to the pending review queue.

[0050] S1-4-3, Clinical Fit Verification: Based on the matching entries in the reference table after filtering and verifying the clinical historical data, the general standards and specialty diagnosis and treatment conflicts are matched to obtain the clinically fitted reference table.

[0051] Specifically, the logically validated lookup table is cross-referenced with the clinical historical data of each national key specialty. The frequency and confidence of each matching relationship in different specialties are statistically analyzed. Matching entries with a frequency below the preset threshold and marked as infrequent by multiple specialties are deleted, thereby retaining matching relationships that meet general standards and are widely verified in clinical practice.

[0052] S1-4-4, Compliant Desensitization and Evidence Storage Processing: Desensitize the clinically adapted control dataset, generate a unique hash value and store it in association with the desensitized control dataset, retain the knowledge base construction and modification log, and output standardized five-level control datasets of diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines and store them in the database.

[0053] The anonymization process involves removing fields from the control dataset that could identify individual patients or medical institutions, retaining only pure medical knowledge content such as disease terminology, syndrome terminology, treatment terminology, prescription terminology, traditional Chinese medicine terminology, and their matching relationships. The hash value is generated using the SHA-256 algorithm and used for integrity verification and tamper-proof evidence preservation of the control dataset.

[0054] The final output of the standardized five-level control dataset contains 7,093 sets of syndrome-treatment relationship comparisons, as well as compliant prescription recommendations for each set of comparisons, realizing the integrated storage of five-level matching rules for diseases, syndromes, treatments, prescriptions, and Chinese medicines.

[0055] S2: Obtain the electronic medical record text entered by the front-end physician workstation, perform word segmentation on the electronic medical record text, identify disease text fragments, syndrome text fragments, treatment text fragments, prescription text fragments, and traditional Chinese medicine text fragments to obtain the original text entities; based on a multi-class standard terminology fusion knowledge base, map the non-standard expressions in the original text entities to standard terms, and output structured medical record quintuple data, which includes standardized diseases, standardized syndromes, standardized treatment methods, standardized prescriptions, and standardized traditional Chinese medicines; input the structured medical record quintuple data into the artificial intelligence inference unit, which performs multi-stage semantic parsing on the structured medical record quintuple data to obtain semantic feature vectors, candidate knowledge support sets, and evidence chain summaries.

[0056] The core innovation of this step lies in unifying the terminology of medical record diagnosis and treatment from the source through disease standards, syndrome standards, and treatment standards. It transforms the colloquial expressions, aliases, and non-standard custom terms in freely written medical records into standardized terminology, providing a standardized data foundation for subsequent two-dimensional verification.

[0057] In some embodiments, step S2 specifically includes steps S2-1 to S2-7.

[0058] S2-1: Clinicians fill out electronic medical records at workstations, and the system retrieves the medical record text in real time.

[0059] The real-time capture process involves configuring a monitoring program on the front-end clinical physician workstation to capture each input event of the physician in the electronic medical record writing interface. The newly added text content is then encapsulated into a JSON format data packet in an incremental manner and transmitted to the central computing server via the hospital's local area network.

[0060] S2-2: The word segmentation unit breaks down the diagnosis and treatment-related text fragments, distinguishing between disease, syndrome, treatment method, prescription and traditional Chinese medicine content.

[0061] The word segmentation unit uses the HanLP word segmentation tool and loads a TCM clinical dictionary to improve the accuracy of medical terminology recognition. After word segmentation, the text fragments are classified according to preset keyword rules and contextual positions. For example, the content after the "disease:" tag is classified as a disease text fragment, the content after the "syndrome:" tag is classified as a syndrome text fragment, the content after the "treatment method:" tag is classified as a treatment method text fragment, and the content in the prescription area is classified as a prescription text fragment.

[0062] S2-3: The terminology mapping unit converts non-standard expressions into national standard terms and outputs structured five-element medical record data.

[0063] The terminology mapping process is as follows: using the disease standard terminology set, syndrome standard terminology set, and treatment standard terminology set from the multi-category standard terminology fusion knowledge base as the mapping benchmark, each term in the original text entity is queried using both exact matching and fuzzy matching. If an exact match is found, the term is directly converted; if only a fuzzy match is found, the semantic similarity between the candidate standard term and the original expression is calculated, and the standard term with the highest similarity is selected as the mapping result; if none of the three methods are successful, the term is marked as an unrecognized term and awaits manual review.

[0064] S2-4: Perform contextual semantic segmentation, cross-sentence relationship identification, and semantic role labeling on electronic medical record texts to identify key medical semantics in chief complaint, present medical history, four diagnostic methods, syndrome differentiation description, treatment method description, and prescription description, and obtain semantic feature vectors.

[0065] Contextual semantic segmentation employs a TCM pre-trained language model based on the Transformer architecture. This model takes the sentence sequence of the medical record text as input and outputs a semantic vector representation of each sentence. Cross-sentence relationship identification captures the implicit connections between diagnostic and treatment elements separated by multiple sentences through an attention mechanism. Semantic role labeling categorizes text fragments into six categories according to a pre-defined medical semantic framework: chief complaint, present illness history, four diagnostic methods, syndrome differentiation, treatment method, and prescription / medication.

[0066] S2-5: Perform terminology normalization and semantic disambiguation on the extracted disease, syndrome, treatment, prescription, and Chinese medicine expressions. Distinguish between diagnostic and treatment terms that are similar in word meaning but different in semantic meaning to obtain standardized terms after disambiguation.

[0067] Semantic disambiguation specifically involves: for medical terms with multiple meanings, a comprehensive judgment is made based on the context of the sentence in which the term is located and other medical elements in adjacent sentences. For example, the semantic connotation of "heat" is different in "body heat but not rising" and "heat toxin stagnation". The disambiguation subunit selects the term with the highest matching degree from multiple candidate terms as the final semantic label by identifying the modifiers, degree words and pathogenesis descriptions in the context of the term.

[0068] S2-6: Based on the semantics of medical record input, perform similar knowledge retrieval and multi-hop association candidate expansion to generate a candidate knowledge support set.

[0069] The similarity knowledge recall process involves using the semantic feature vector output by S2-4 as the query vector and performing vector similarity retrieval in a multi-class standard terminology fusion knowledge base. This retrieves K knowledge entries that are semantically similar to the current medical record as candidates, with K pre-set to 10. The multi-hop association candidate expansion process involves further retrieving the second-order neighbor nodes in the knowledge base connected by five-level association edges (disease-syndrome-treatment-prescription-traditional Chinese medicine) to expand the scope of supplementary knowledge entries indirectly related to the current medical record. These supplementary knowledge entries collectively form the candidate knowledge support set.

[0070] S2-7: To assist in reasoning about the semantic relationships between diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines, identify implicit conflicts such as: local statements being valid but the overall diagnostic direction deviating; similar wording but inconsistent medical semantics; and local reasonableness but inconsistent global links. Based on the identified implicit conflicts, generate an evidence chain summary, which includes an explanation of the cause of the problem, hints at the conflict location, and suggestions for revision.

[0071] The auxiliary reasoning process involves inputting a quintuple of standardized disease, syndrome, treatment method, prescription, and traditional Chinese medicine (TCM) after disambiguation in S2-5 into the reasoning engine. The engine then sequentially performs single-hop consistency checks and multi-hop link consistency checks. The single-hop consistency check verifies whether there are direct matching relationships between disease and syndrome, syndrome and treatment method, and treatment method and prescription / medicine. The multi-hop link consistency check verifies whether the entire link from syndrome to prescription / medicine, through treatment method, satisfies the transitive closure constraint in the knowledge base. If any check fails, the terminology name, conflict type, and relevant knowledge base matching rules of the conflicting node are recorded as an evidence chain summary.

[0072] The artificial intelligence reasoning units consisting of S2-4 to S2-7 are constrained by the standard terminology system, the correspondence between disease syndromes, the correspondence between syndromes and treatment methods, and the correspondence between treatment methods and prescriptions. This is different from the boundless free generation method, enabling the system to identify "whether it is standardized", "whether it is self-consistent" and "whether it forms a complete diagnosis and treatment logic chain", thus upgrading the system from surface field inspection to the ability to make logical judgments based on the connotation of traditional Chinese medicine.

[0073] S3: Perform a two-dimensional joint verification of theory, method, prescription and drug based on the structured medical record five-tuple data, semantic feature vector, candidate knowledge support set and multi-class standard terminology fusion knowledge base, generate defect markers, and push early warning information to the front-end physician workstation according to the defect markers and evidence chain summary.

[0074] In some embodiments, step S3 specifically includes steps S3-1 to S3-6.

[0075] S3-1: Traverse the structured medical record quintuple data and determine whether there are blank or missing markers in the disease field, syndrome field, treatment method field, prescription field, and traditional Chinese medicine field. If they exist, generate field missing markers.

[0076] Blank or unfilled fields are marked as: fields with an empty string value, fields with a null value, fields containing only spaces or tabs, and fields with preset placeholders such as "to be filled" or "none".

[0077] S3-2: Based on a multi-class standard terminology fusion knowledge base, sequentially verify the matching relationship between disease and syndrome, the matching relationship between syndrome and treatment method, and the matching relationship between treatment method and prescription. If any matching relationship does not satisfy the corresponding relationship in the knowledge base, a logical inconsistency defect label is generated. Based on the semantic feature vector, candidate knowledge support set, and evidence chain summary, the logical inconsistency defect label is weighted by confidence and classified by conflict type, and the enhanced logical inconsistency defect label is output.

[0078] The confidence weighting is specifically implemented by calculating weights based on the similarity scores between the semantic feature vectors and each item in the candidate knowledge support set. The higher the similarity, the greater the contribution weight of the candidate item to the current judgment. The conflict type classification is specifically implemented by dividing the logical inconsistency defect labels into four subcategories according to the location of the conflict: disease-syndrome mismatch, syndrome-treatment mismatch, treatment-prescription mismatch, and syndrome-prescription transmission break. Each subcategory corresponds to a different revision strategy.

[0079] S3-3: Based on the field missing marker and / or the enhanced logical mismatch defect label, along with the evidence chain summary, push the warning prompt information, including defect location and revision direction suggestions, to the front-end physician workstation, and retain the front-end verification interaction log.

[0080] The warning notification is sent in real-time via a pop-up component on the front-end physician workstation interface. The pop-up includes a defect type icon, the specific defect location, a summary of the evidence chain explaining the reasons for the conflict, and suggested revision directions. Physicians can directly click to accept the suggestion or close the notification to continue making changes.

[0081] S3-4: After obtaining the standardized syndrome that has been entered and selected by the front-end physician workstation, based on the knowledge base of multi-category standard terminology, the standardized syndrome that matches the standardized disease is pushed. After the standardized syndrome is selected, the corresponding standardized treatment method is automatically matched and backfilled into the treatment method filling box. At the same time, a list of prescriptions and medicines that match the standardized syndrome and standardized treatment method is pushed.

[0082] The recommended prescription list is generated as follows: Standardized disease is used as the first search key, standardized syndrome as the second search key, and standardized treatment method as the third search key. All prescription entries that simultaneously meet the syndrome and treatment method matching criteria are searched within the prescription matching item group of the knowledge base. The top five prescriptions are then sorted by clinical frequency from highest to lowest and returned as the recommended list. The prescription matching item group upon which this recommended list is based has pre-included data on 4093 basic Chinese medicinal herbs, including commonly used hospital-prepared formulations, granules, and processed medicinal materials, ensuring that the recommended results simultaneously comply with pharmacopoeia standards and Chinese medicine quality control standards.

[0083] S3-5: When manual modifications to the automatically backfilled treatment methods or autonomous selection of syndromes or prescriptions are obtained, the matching relationship between the modified syndromes and treatment methods and the matching relationship between treatment methods and prescriptions is reviewed again based on the knowledge base of multiple standard terms. If a mismatch occurs, the reason for the conflict is highlighted and the recommended entries are refreshed.

[0084] The highlighting feature is as follows: the background color of the input box containing the conflicting field is marked as a warning color, and the reason for the conflict is displayed in text form below the input box. At the same time, the list of recommended items is automatically refreshed, and new recommended items that match the modified relationship are added to the sidebar of the interface.

[0085] S3-6: Repeatedly perform field integrity checks and chain matching checks until there are no missing field markers and no logical mismatch defect labels, then allow the medical record to be saved.

[0086] Specifically, each time a physician completes a modification operation and triggers a save command, the system automatically calls back the complete verification process from S3-1 to S3-2. Only when all verification items return a passing result is the save interface unlocked and medical record data written to the database allowed. If the verification fails, the save operation is locked, and the physician must complete the correction according to the pop-up prompts before attempting to save again. In the actual implementation, this cyclical verification mechanism was used to test 1453 outpatient real-time medical records, identifying 158 cases with missing fields and 70 cases with mismatched treatment principles, methods, prescriptions, and medications. All of these issues were detected with pop-up warnings before saving, and the defects were cleared after the physician made immediate corrections.

[0087] S4: Retrieve archived medical records from the hospital's final medical record database, perform dual-dimensional theory, method, prescription, and medication linkage verification on each archived medical record, and record the defect information of each archived medical record; calculate the completeness rate of key fields, the consistency rate of theory, method, prescription, and medication, and the total defect rate of medical record theory and method information in batches based on the defect information of all archived medical records, and generate defect distribution ledgers and quality control analysis reports according to the department dimension and / or physician dimension and / or disease dimension.

[0088] The retrieval process is as follows: The system retrieves all archived medical record records within a specified time range in reverse order of the archive timestamps in the final medical record database. It iterates through each record and performs the complete verification logic in S3 on the structured medical record 5-tuple data of each record. Each defective record is associated with the medical record number and archive time for traceability.

[0089] In some embodiments, the formula for calculating the completeness rate of key fields in step S4 is: the number of cases without missing fields divided by the total number of cases multiplied by 100%; the formula for calculating the consistency rate of theory, method, prescription and medicine is: the number of cases where all diseases, syndromes, treatment methods and prescriptions are matched divided by the total number of cases multiplied by 100%; the formula for calculating the total defect rate of the theory and method information of cases is: the number of cases with any defect of missing fields or mismatched fields divided by the total number of cases multiplied by 100%.

[0090] In the specific implementation process, 53,356 TCM medical records that had already undergone manual quality control were used as test samples. The aforementioned batch quality control process was used for full automated verification. The system automatically output that the key field completeness rate was 99.55%, the consistency rate of theory, method, prescription, and medication was 99.91%, and the total defect rate of medical record theory and method information was 0.54%. Simultaneously, the system automatically outputs a stratified quality control report on defect distribution by department. Manual review of a single batch of 53,356 medical records took approximately 72 hours; after adopting the aforementioned batch quality control process, the time was reduced to 20 minutes; the defect omission rate decreased from 28.6% with manual methods to 1.1% with the system.

[0091] S5: De-identify the log data and store the de-identified log data in a distributed backup storage array; generate a unique hash code for the de-identified log data and associate the unique hash code with the de-identified log data for storage; the log data includes: knowledge base construction and modification logs generated during the construction and modification of a multi-category standard terminology fusion knowledge base, front-end verification interaction logs generated during the execution of the dual-dimensional theory, law, prescription and drug linkage verification, and final medical record quality control logs generated during the final quality control process of archived medical records.

[0092] The desensitization process specifically involves removing all information fields from the log data that can directly or indirectly identify the patient, including but not limited to name, ID number, hospital number, bed number, and contact information, retaining only the technical fields related to the quality control and verification process.

[0093] The unique hash code is generated by calculating the entire de-identified log data using the SM3 or SHA-256 algorithm. This hash code is associated with the corresponding de-identified log data in the form of key-value pairs and stored in the distributed backup storage array for subsequent data integrity verification and anti-tampering verification.

[0094] Distributed backup storage arrays are physical storage nodes independent of the central computing hardware layer. They are connected to the database server through a storage area network and support off-site disaster recovery backup.

[0095] In some embodiments, the hardware environment upon which the above steps rely is a three-tier physical architecture: the front-end interaction hardware layer includes clinical physician workstations and medical record quality control management PCs, with each terminal connected to the central server via an intra-hospital LAN switch; the central computing hardware layer includes two independent physical hosts: a database server and an algorithm computing server, which are directly connected via a 10 Gigabit fiber optic cable for high-speed data interaction; and the storage backup hardware layer is a distributed backup storage array, continuously synchronized with the database server via a storage bus. The medical record text data generated by the front-end terminals is encrypted and uploaded to the algorithm computing server. The algorithm computing server calls the multi-category standard terminology fusion knowledge base stored in the database server to perform a complete matching verification, and then pushes the warning information and quality control indicators back to the corresponding front-end terminals. All operation logs are synchronized in real time to the distributed backup storage array for permanent storage.

[0096] It should be noted that the specific implementation methods described above also apply to the lightweight stand-alone deployment mode. Small-scale primary-level TCM hospitals do not need to adopt a dual-server architecture; they can use an all-in-one machine to integrate the database and computing unit on the same physical host, enabling local lightweight deployment of the knowledge base. This supports small-scale medical record quality control in a single department, and all module logic and verification processes remain consistent with the above embodiments.

[0097] In a complete embodiment, the method provided in this application was fully verified in a real hospital environment. The implementation environment, implementation process and verification results are as follows.

[0098] Implementation Environment. In terms of hardware, this embodiment is configured with a dual-path algorithm server, a database storage server, 30 clinical physician PC terminals, and 3 medical record management terminals. Regarding data foundation, 53,356 TCM medical records that have completed manual quality control registration are used as test samples. In terms of standards and benchmarks, clinical diagnostic and treatment terminology is standardized from the source based on disease standards, syndrome standards, and treatment principles. Addressing the issue of the three sets of standards being independent and lacking built-in correlation, a consensus among five composite experts was established to build an internal correlation between syndromes and treatment principles. Furthermore, the pharmacopoeia and TCM quality control standards are introduced to improve the integrated theory-method-formula-medicine comparison table, incorporating 4,093 basic TCM herbs, including 124 commonly used hospital-made preparations, 320 formula granules, and 580 processed medicinal herbs. This ultimately forms a standardized theory-method-formula-medicine comparison table containing 1,369 national standard diseases, 2,060 standard syndromes, 1,168 standard treatment principles, 7,093 disease-syndrome comparison relationships, and 1,796 syndrome-treatment comparison relationships, along with recommended compliant formulas and medicines. This form allows physicians to fill in symptoms and automatically import standard treatment methods, ensuring consistency in matching theory, treatment, prescription, and medication across all medical records while standardizing clinical terminology and avoiding gaps in core fields.

[0099] Implementation Process. Knowledge Base Construction and Implementation Phase: Import the original texts of five standards, conduct multiple rounds of offline consensus discussions among five multidisciplinary experts, and complete the initial matching relationships of 1796 sets of syndromes, treatments, prescriptions, and drugs; collect 17 feedback opinions from clinical departments, and form the final control dataset after four layers of quality control verification; and fully retain the discussion logs and iteration modification logs.

[0100] Clinical front-end writing scenario testing phase: 1453 outpatient real-time medical records were randomly selected for writing tests. The system verified and identified 158 medical records with missing fields and 70 medical records with mismatched theory, method, prescription and medicine. All of the above 228 defective medical records triggered pop-up warnings before saving. After the physicians corrected the errors in time according to the prompts, the defects were cleared.

[0101] In the batch quality control testing phase of archived medical records: a full automated verification was performed on 53,356 existing medical records. The system automatically output that the completeness rate of key fields was 99.55%, the consistency rate of treatment principles, methods, prescriptions, and medications was 99.91%, and the overall defect rate of treatment principles and methods information in the medical records was 0.54%. The system also automatically statistically analyzed the defect distribution by department and output hierarchical quality control reports. The manual review of a single batch of 53,356 medical records took approximately 72 hours, while the time was reduced to 20 minutes after adopting this method; the defect miss rate decreased from 28.6% with manual methods to 1.1% with the system.

[0102] Data storage implementation phase: All datasets and more than 50,000 medical record verification records are uniformly de-identified and hash values ​​are generated. The entire process operation log is retained and can be directly used for data intellectual property rights confirmation and registration and hospital quality control file archiving.

[0103] Advantages of the case comparison. Compared to the comparison document CN121725975A, which only provides single-round symptom-to-prescription recommendations for consultation scenarios, this embodiment covers all medical records in the hospital, fully including five levels of diagnostic and treatment elements: disease, syndrome, treatment method, prescription, and traditional Chinese medicine. It supports dual-mode operation of pre-event warning for writing front-end and batch quality control of archived medical records afterward.

[0104] Compared to the prior art CN122025053A, which only constructs a map based on co-occurrence relationships in medical records, this embodiment uses five authoritative standards as a benchmark. Through multidisciplinary expert manual modeling and multi-specialty clinical data iteration, the matching relationships have a unified authoritative judgment basis, while providing the ability to automatically output quantitative quality control indicators.

[0105] Compared to traditional purely manual quality control methods, this embodiment significantly reduces labor costs, eliminates subjective judgment differences in manual review, and achieves standardization, digitalization, and traceability of medical record quality control.

[0106] Optional alternative implementation methods. In the lightweight stand-alone implementation method, small primary-level TCM hospitals do not need to configure a dual-server architecture. They can use an all-in-one machine to integrate the database and computing unit on the same physical host. The hardware only retains the physician's stand-alone terminal. The knowledge base is deployed locally in a lightweight manner, which can support small-scale medical record quality control in a single department. All module logic and verification processes are completely consistent with the above embodiments.

[0107] In the AI ​​model extension implementation, a traditional Chinese medicine language model can be connected to the front end of the five-level matching and verification unit. Based on the standard comparison dataset of this solution, fine-tuning and training can be performed to achieve flexible matching and assisted judgment for complex cases involving concurrent syndromes and special specialties. The underlying four-standard fusion knowledge base and dual-dimensional verification architecture remain the core foundation of this solution; the integration of the AI ​​model does not change the main technical solution.

[0108] It should be noted that the above-described implementation environment, implementation process, and verification results are only used to illustrate the feasibility and technical effects of the technical solution of this application, and do not constitute any limitation on the scope of protection of the claims. Those skilled in the art can make adaptive adjustments to the specific parameters, data sources, and hardware configurations in the above embodiments without departing from the concept of this application.

[0109] Example 2: This application provides an intelligent verification system for quality control of TCM medical records based on multiple national standards, such as the principles, methods, prescriptions, and medications. Figure 2 As shown, the system includes a front-end interaction hardware layer 100, a central computing hardware layer 200, and a storage backup hardware layer 300.

[0110] The front-end interactive hardware layer 100 includes a clinical physician workstation 110 and a medical record quality control management PC terminal 120, which are connected to the central computing hardware layer 200 through bidirectional communication via the hospital's local area network.

[0111] Specifically, the clinical physician workstation 110 is equipped with a standard monitor, keyboard, and mouse. Physicians use the clinical physician workstation 110 to input electronic medical record text and receive warning prompts and recommended items from the system. The medical record quality control management PC terminal 120 is used by quality control personnel to view quality control results and export statistical reports. The front-end terminal only uploads medical record text data in one direction and receives recommendation or defect prompts from the system in two directions. All data transmission is encrypted via the hospital's local area network.

[0112] The central computing hardware layer 200 includes a database server 210 and an algorithm computing server 220. The database server 210 and the algorithm computing server 220 are directly connected by optical fiber to achieve data interaction. The database server 210 stores a knowledge base of multiple standard terms, full medical record data and quality control log evidence records. The algorithm computing server 220 carries semantic parsing, matching verification, artificial intelligence reasoning and quantitative statistical calculation programs.

[0113] Specifically, the database server 210 and the algorithm processing server 220 are two independent physical hosts connected directly via a 10 Gigabit fiber optic cable to achieve high-speed data interaction, ensuring unimpeded data transmission during large-scale batch verification of medical records. The four standard original terminology databases, the disease syndrome treatment method prescription and drug comparison dataset, and the full medical record data stored in the database server 210 are all persistently stored. The algorithm processing server 220 uses a structured query language to call the above data in the database server 210 and write back the verification results.

[0114] The storage backup hardware layer 300 is equipped with a distributed backup storage array, which is continuously synchronized with the database server 210 via a data bus for off-site de-identification backup.

[0115] Specifically, the distributed backup storage array is a physical storage node independent of the central computing hardware layer 200. It is connected to the database server 210 through a storage area network, supporting off-site disaster recovery backup of data. The full comparison dataset and quality control verification records in the database server 210 are synchronized to the distributed backup storage array in real time, meeting the compliance requirements for data intellectual property rights preservation.

[0116] The central computing hardware layer 200 is equipped with a software system 230, which includes a multi-standard terminology fusion knowledge base construction module 231, an electronic medical record structured parsing module 232, an artificial intelligence reasoning unit 233, a two-dimensional theory, law, prescription and drug linkage verification module 234, a batch quality control and quantitative statistics module for medical records 235, and a full-process log storage and compliance module 236.

[0117] The multi-standard terminology fusion knowledge base construction module 231 is used to import the original text of five types of preset terms, generate a standardized five-level control dataset of diseases, syndromes, treatments, prescriptions, and Chinese medicines based on preset syndrome and treatment mapping rules and clinical historical data, and store it in the database server 210.

[0118] Specifically, the original texts of the four categories of preset terms are: *Traditional Chinese Medicine Clinical Diagnosis and Treatment Terminology Part 1: Diseases*, *Traditional Chinese Medicine Clinical Diagnosis and Treatment Terminology Part 2: Syndromes*, *Traditional Chinese Medicine Clinical Diagnosis and Treatment Terminology Part 3: Treatment Methods*, the 2025 edition of the *Pharmacopoeia of the People's Republic of China*, *Internal Medicine of Traditional Chinese Medicine (New Century Fifth Edition)*, and *Quality Control Standards for Clinical Research of Traditional Chinese Medicine*. The multi-category standard terminology fusion knowledge base construction module 231 cleans the five categories of original standard texts to form a syndrome standard terminology set and a treatment method standard terminology set. The syndrome standard terminology set contains 1369 sets of national standard diseases and 2060 sets of standard syndromes, while the treatment method standard terminology set contains 1168 sets of standard treatment methods. Addressing the issue that the syndrome standard and treatment method standard systems are independent and lack built-in connections, the multi-category standard terminology fusion knowledge base construction module 231 generates 7093 sets of disease-syndrome correlation relationships and 1796 sets of syndrome-treatment method correlation relationships through preset mapping rules and iterative correction using clinical historical data. It also integrates the pharmacopoeia and traditional Chinese medicine quality control standards to improve the matching rules between treatment methods and prescriptions. The standardized three-level comparison dataset ultimately output by the multi-standard terminology fusion knowledge base construction module 231 realizes the integrated storage of five-level matching rules for diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines.

[0119] The electronic medical record structure parsing module 232 is used to perform word segmentation and standard terminology mapping on the electronic medical record text uploaded by the front-end physician workstation, and output structured medical record quintuple data.

[0120] Specifically, the electronic medical record structured parsing module 232 integrates the HanLP word segmentation tool and loads a TCM clinical dictionary to segment the electronic medical record text. Based on preset keyword rules and contextual positions, it divides text fragments into five categories: disease, syndrome, treatment method, prescription, and Chinese herbal medicine. The electronic medical record structured parsing module 232 uses the disease standard terminology set, syndrome standard terminology set, and treatment method standard terminology set from a multi-category standard terminology fusion knowledge base as a mapping benchmark. It performs two levels of queries—precise matching and fuzzy matching—on colloquial expressions, aliases, and non-standard custom terms in the original text entities, converting them into corresponding standard and normative terms. The final output is a structured medical record five-tuple data including standardized disease, standardized syndrome, standardized treatment method, standardized prescription, and standardized Chinese herbal medicine.

[0121] The input end of the artificial intelligence reasoning unit 233 is connected to the output end of the electronic medical record structured parsing module 232. It is used to receive structured medical record quintuple data, perform multi-stage semantic parsing, structured extraction, terminology normalization and semantic disambiguation, knowledge enhancement retrieval, diagnosis and treatment logic chain reconstruction and semantic consistency judgment, and evidence chain interpretation generation on the structured medical record quintuple data, and output semantic feature vector, candidate knowledge support set and evidence chain summary.

[0122] Specifically, the output of the AI ​​reasoning unit 233 is jointly constrained by the standard terminology system, the correspondence between disease syndromes, the correspondence between syndromes and treatment methods, and the correspondence between treatment methods and prescriptions. This distinguishes it from the unbounded free generation method, enabling the system to identify "whether it is standardized," "whether it is self-consistent," and "whether it forms a complete diagnostic and treatment logic chain," thus upgrading its ability from surface-level field inspection to the logical judgment of TCM connotation. The semantic feature vector, candidate knowledge support set, and evidence chain summary output by the AI ​​reasoning unit 233 serve as enhancement signals, and are input together with the structured medical record five-tuple data into the dual-dimensional theory-method-prescription-medicine linkage verification module 234 to assist in the confidence weighting and conflict type classification of the verification results.

[0123] The dual-dimensional theory-method-formula-medicine linkage verification module 234 connects to the output of the electronic medical record structured parsing module 232 to obtain structured medical record five-tuple data. The second input connects to the output of the artificial intelligence reasoning unit 233 to obtain semantic feature vectors, candidate knowledge support sets, and evidence chain summaries. The third input connects to the database server 210 to obtain standardized disease, syndrome, treatment method, formula, and traditional Chinese medicine five-level comparison dataset. The dual-dimensional theory-method-formula-medicine linkage verification module 234 is used to perform field integrity verification and disease, syndrome, treatment method, formula, and traditional Chinese medicine chain matching verification based on the structured medical record five-tuple data, semantic feature vectors, candidate knowledge support sets, evidence chain summaries, and five-level comparison datasets. It generates defect markers and pushes early warning information to the front end along with the evidence chain summary.

[0124] Specifically, the dual-dimensional theory-method-formula-medicine linkage verification module 234 internally integrates a key field integrity verification subunit 2341 and a disease-syndrome-treatment-formula-medicine chain matching verification subunit 2342 in parallel. The key field integrity verification subunit 2341 traverses the structured medical record five-tuple data, identifies whether there are blank or unfilled markers in the fields of disease, syndrome, treatment, formula, and Chinese medicine, and generates field missing markers; the disease-syndrome-treatment-formula-Chinese medicine chain matching verification subunit 2342 verifies the matching relationship between disease and syndrome, syndrome and treatment, and treatment and formula based on a multi-class standard terminology fusion knowledge base. If any matching relationship does not meet the correspondence in the multi-class standard terminology fusion knowledge base, a logical inconsistency defect label is generated. The logical inconsistency defect label is then weighted by confidence and classified by conflict type according to the semantic feature vector, candidate knowledge support set, and evidence chain summary, and the enhanced logical inconsistency defect label is output.

[0125] The medical record batch quality control and quantitative statistics module 235 is used to retrieve archived medical records from the hospital's final medical record database for batch quality control verification, calculate the completeness rate of key fields, the consistency rate of theory, method, prescription and drug, and the total defect rate of medical record theory and method information, and generate a defect distribution ledger and quality control analysis report.

[0126] Specifically, the batch quality control and quantitative statistics module 235 for medical records uses a dedicated read-only interface to directly connect to the hospital's final medical record database, eliminating the need for manual file import. The module automatically records defect information for each archived medical record, reusing the complete verification logic of the dual-dimensional theory, treatment, prescription, and medication linkage verification module 234. Based on this, the module calculates three standardized quality control indicators in batches: the key field completeness rate is the number of medical records with no omissions divided by the total number of medical records multiplied by 100%; the consistency rate of theory, treatment, prescription, and medication is the number of medical records with all matching diseases, syndromes, treatment methods, prescriptions, and Chinese medicines divided by the total number of medical records multiplied by 100%; and the total defect rate of the medical record's theory and treatment information is the number of medical records with any missing or mismatched fields divided by the total number of medical records multiplied by 100%. The module also automatically generates defect distribution ledgers and quality control analysis reports by department, physician, and disease, supporting online viewing and export / printing.

[0127] The end-to-end log storage and compliance module 236 is used to de-identify log data and store it in a distributed backup storage array. It also generates a unique hash code for the de-identified log data and associates the unique hash code with the de-identified log data for storage. The log data includes knowledge base construction and modification logs generated by the multi-standard terminology fusion knowledge base construction module 231 during the knowledge base construction and modification process, front-end verification interaction logs generated by the dual-dimensional theory, law, prescription and drug linkage verification module 234 during the verification process, and final medical record quality control logs generated by the batch quality control and quantitative statistics module 235 during the final quality control of archived medical records.

[0128] Specifically, the full-process log storage and compliance module 236 is equipped with multiple acquisition interfaces, which respectively connect to the multi-standard terminology fusion knowledge base construction module 231, the dual-dimensional theory, method, prescription, and drug linkage verification module 234, and the medical record batch quality control and quantitative statistics module 235. It collects operation records generated by these modules across the entire domain. The de-identification process involves deleting all information fields that can directly or indirectly identify the patient from the log data, including but not limited to name, ID number, hospital number, bed number, and contact information, retaining only the technical fields related to the quality control and verification process. A unique hash code is generated by calculating the entire de-identified log data using the SM3 or SHA-256 algorithm. This hash code is associated with the corresponding de-identified log data in key-value pairs and stored in a distributed backup storage array for subsequent data integrity verification and anti-tampering verification.

[0129] In some embodiments, such as Figure 3As shown, the multi-standard terminology fusion knowledge base construction module 231 internally comprises, sequentially along the data flow, a standard terminology import subunit 2311, a mapping rule generation subunit 2312, a clinical iterative optimization subunit 2313, and a four-layer integrated quality control and verification subunit 2314. The standard terminology import subunit 2311 has four data source interfaces, respectively connecting to the disease standard terminology import interface, the syndrome standard terminology import interface, the treatment method standard terminology import interface, the pharmacopoeia standard terminology import interface, the Traditional Chinese Medicine Internal Medicine (New Century Fifth Edition) standard terminology import interface, and the Traditional Chinese Medicine quality control standard terminology import interface. The four-layer integrated quality control and verification subunit 2314... Unit 2314 internally integrates a terminology uniqueness verification unit 23141, an association logic verification unit 23142, a clinical adaptation verification unit 23143, and a compliance desensitization and evidence storage unit 23144. The output of the multi-category standard terminology fusion knowledge base construction module 231 is connected to the five-level control dataset storage area. The standardized disease-syndrome-treatment-prescription-traditional Chinese medicine five-level control dataset in the three-level control dataset storage area includes a complete set of standard disease entries, a complete set of standard syndrome entries, a complete set of standard treatment entries, an in-hospital extended traditional Chinese medicine entry set, a disease syndrome and treatment built-in association control group, and a matching prescription entry group.

[0130] Specifically, the standard terminology import subunit 2311 imports five types of original texts—disease standard syndrome standard, treatment method standard, pharmacopoeia standard, traditional Chinese medicine internal medicine standard, and traditional Chinese medicine quality control standard—through four data source interfaces. The mapping rule generation subunit 2312 generates initial control relationships based on predefined syndrome-treatment mapping rules from a multidisciplinary expert team. The clinical iterative optimization subunit 2313 acquires historical clinical data from multiple national key specialties, compares the co-occurrence relationships of diseases, syndromes, treatment methods, prescriptions, and traditional Chinese medicines in the historical clinical data with the current control table, calculates the matching hit rate and conflict rate, and adjusts the control relationships after triggering an expert review process for conflicting entries, iterating through multiple rounds until the matching hit rate reaches a preset threshold. In the four-layer integrated quality control verification subunit 2314, the terminology uniqueness verification unit 23141 uses precise string matching plus semantic similarity calculation to check for duplicate terms and delete duplicate terms and ambiguous expressions; the association logic verification unit 23142 verifies the chain matching relationship one by one to see if it conforms to the TCM syndrome differentiation rules and eliminates the pathogenesis contradiction association. The pathogenesis attribute of the syndrome should match the efficacy classification of the treatment method, and the target of the treatment method should correspond to the efficacy and indication of the prescription; the clinical fit verification unit 23143 filters the matching entries that conflict between general standards and specialist diagnosis and treatment based on clinical historical data; the compliance desensitization and evidence storage unit 23144 generates a unique hash value after desensitizing the control dataset and stores it in association with the desensitized control dataset.

[0131] In some embodiments, such as Figure 4As shown, the artificial intelligence reasoning unit 233 consists of a multi-stage semantic parsing subunit 2331, a structured extraction subunit 2332, a terminology normalization and semantic disambiguation subunit 2333, a knowledge enhancement retrieval subunit 2334, a diagnosis logic chain reconstruction and semantic consistency judgment subunit 2335, and an evidence chain interpretation and generation subunit 2336, which are sequentially connected along the data flow direction.

[0132] The multi-stage semantic parsing subunit 2331 is used to perform contextual semantic segmentation, cross-sentence relationship identification, and semantic role labeling on electronic medical record text. It identifies key medical semantics in chief complaint, present medical history, four diagnostic methods, syndrome differentiation description, treatment method description, and prescription description, and outputs semantic feature vectors.

[0133] Specifically, contextual semantic segmentation employs a TCM pre-trained language model based on the Transformer architecture. This model takes the sentence sequence of the medical record text as input and outputs a semantic vector representation of each sentence. Cross-sentence relationship identification captures the implicit connections between diagnostic and treatment elements separated by multiple sentences through an attention mechanism. Semantic role labeling categorizes text fragments into six categories according to a pre-defined medical semantic framework: chief complaint, present illness history, four diagnostic methods, syndrome differentiation, treatment method, and prescription / medication.

[0134] The structured extraction subunit 2332 is used to extract diseases, syndromes, treatments, prescriptions, traditional Chinese medicines and their related descriptions from electronic medical record text, generate structured objects that can be verified later, and perform completeness perception marking on the missing, weak expression and abnormal position of key diagnostic and treatment elements.

[0135] Specifically, the structured extraction subunit 2332, based on the output of the multi-stage semantic parsing subunit 2331, uses a combination of rule-based template matching and model-based sequence labeling to locate explicit expressions and implicit related expressions describing diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines in the medical record text. For missing, weakly expressed, and abnormal locations of key diagnostic and treatment elements, the structured extraction subunit 2332 generates completeness-aware markers and passes them to the evidence chain interpretation generation subunit 2336.

[0136] The terminology normalization and semantic disambiguation subunit 2333 is used to combine multiple standard terminology fusion knowledge bases and synonym mapping relationships to map non-standard expressions into standard terms, and to disambiguate diagnostic and treatment terms that are similar in word meaning but different in semantic meaning.

[0137] Specifically, for diagnostic and treatment terms with multiple meanings, the term unification and semantic disambiguation subunit 2333 makes a comprehensive judgment based on the context of the sentence in which the term is located and other diagnostic and treatment elements in adjacent sentences. By identifying the modifiers, degree words and pathogenesis descriptions in the preceding and following text of the term, the term with the highest matching degree is selected as the final semantic label from multiple candidate terms.

[0138] The knowledge-enhanced retrieval subunit 2334 is used to vectorize knowledge items related to diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines, and to perform similar knowledge retrieval and multi-hop association candidate expansion based on the semantics of medical record input, locate evidence fragments, and generate a candidate knowledge support set.

[0139] Specifically, the knowledge-enhanced retrieval subunit 2334 uses the semantic feature vector output by the multi-stage semantic parsing subunit 2331 as the query vector, performs vector similarity retrieval in a multi-class standard terminology fusion knowledge base, and recalls K knowledge entries that are semantically similar to the current medical record as candidates, with K preset to 10. For each recalled candidate knowledge entry, the knowledge-enhanced retrieval subunit 2334 further retrieves its second-order neighbor nodes in the multi-class standard terminology fusion knowledge base connected by five-level association edges of disease-syndrome-treatment-prescription-traditional Chinese medicine, expanding to supplementary knowledge entries indirectly related to the current medical record, which together constitute the candidate knowledge support set.

[0140] The diagnostic logic chain reconstruction and semantic consistency judgment subunit 2335 is used to assist in judging the semantic association between diseases, syndromes, treatment methods, prescriptions, and Chinese medicines, reconstruct the path of syndrome differentiation and treatment, identify implicit conflict situations such as local expressions being valid but the overall syndrome differentiation direction deviating, similar wording but inconsistent medical semantics, and local rationality but inconsistent global links, and generate conflict type labels.

[0141] Specifically, the diagnostic logic chain reconstruction and semantic consistency judgment subunit 2335 inputs the standardized disease, syndrome, treatment method, prescription, and traditional Chinese medicine (TCM) after terminology normalization and semantic disambiguation by the subunit 2333 into the inference engine. The inference engine then sequentially performs single-hop consistency checks and multi-hop link consistency checks. The single-hop consistency check verifies whether there is a direct matching relationship between the disease and syndrome, between the syndrome and treatment method, and between the treatment method and prescription. The multi-hop link consistency check verifies whether the entire link from the disease to the syndrome, through the treatment method, to the prescription satisfies the transitive closure constraint in the multi-class standard terminology fusion knowledge base. If any check fails, the diagnostic logic chain reconstruction and semantic consistency judgment subunit 2335 generates a conflict type label.

[0142] The evidence chain interpretation and generation subunit 2336 is used to generate a problem cause description, conflict location prompt, revision direction suggestion and evidence chain summary that can be manually reviewed based on the conflict type label, candidate knowledge support set and completeness awareness mark, and output the evidence chain summary to the dual-dimensional theory, law, prescription and drug linkage verification module 234.

[0143] Specifically, the evidence chain interpretation generation subunit 2336 aggregates the conflict type label, the knowledge base rule entries matched in the candidate knowledge support set, and the completeness awareness mark output by the structured extraction subunit 2332. It generates a natural language description of the cause of the problem according to the preset interpretation template, marks the specific location of the conflict in the medical record text, and generates revision direction suggestions based on the corresponding standard matching relationship in the multi-class standard terminology fusion knowledge base. Finally, it outputs the evidence chain summary that can be reviewed and audited by quality control personnel to the result aggregation and scoring unit 2343 of the dual-dimensional theory, law, prescription and drug linkage verification module 234.

[0144] In some embodiments, such as Figure 5 As shown, the dual-dimensional theory, method, prescription, and drug linkage verification module 234 also includes a result aggregation and scoring unit 2343. The input of the result aggregation and scoring unit 2343 is connected to the output of the key field integrity verification subunit 2341, the disease syndrome treatment method prescription drug chain matching verification subunit 2342, and the artificial intelligence reasoning unit 233, respectively. It is used to perform multi-source fusion of rule hit results, knowledge matching results, and model reasoning results, and to perform quality control conclusion judgment, risk level output, and generation of evidence chain and revision suggestions based on the fusion results.

[0145] Specifically, the input data for the result aggregation and scoring unit 2343 includes three sources: the first input is the field missing markers output by the key field integrity verification subunit 2341, specifically including three types of integrity defects: missing chief complaint, missing treatment method, and missing treatment record; the second input is the logical inconsistency defect labels and enhanced logical inconsistency defect labels output by the disease syndrome treatment method prescription traditional Chinese medicine chain matching verification subunit 2342, specifically including non-standard terms, alias mapping and semantic disambiguation results, as well as the disease-syndrome-treatment method-prescription-traditional Chinese medicine chain matching results and the results of conflict identification of the relationship between medicinal ingredients; the third input is the semantic feature vector, candidate knowledge support set, and evidence chain summary output by the artificial intelligence reasoning unit 233. The above three inputs correspond to three types of heterogeneous data: rule hitting results, knowledge matching results, and model reasoning results.

[0146] The results aggregation and scoring unit 2343 incorporates a weighted fusion engine that comprehensively scores the three inputs according to preset weights. The base weight for rule-matching results is 0.5, for knowledge-matching results it is 0.3, and for model inference results it is 0.2. These weights can be dynamically adjusted based on clinical feedback. The fused comprehensive score is used to determine the execution risk level and prioritize issues: a comprehensive score higher than 0.8 is marked as high risk, between 0.5 and 0.8 as medium risk, and lower than 0.5 as low risk.

[0147] Based on the fusion results, the result aggregation and scoring unit 2343 outputs a quality control conclusion, which includes three levels: pass, conditionally pass, and review. When all verification items pass and the overall score is higher than 0.9, the result aggregation and scoring unit 2343 outputs "pass"; when there is a field missing marker or the overall score is between 0.7 and 0.9, the result aggregation and scoring unit 2343 outputs "conditionally pass" and attaches a list of specific items that need to be rectified; when there is a logical inconsistency defect label or the overall score is lower than 0.7, the result aggregation and scoring unit 2343 outputs "review," at which point the system automatically pushes the medical record to the manual review queue for final confirmation by quality control experts.

[0148] The results aggregation and scoring unit 2343 also generates risk level labels, evidence chains, and revision suggestions based on the fusion results. Risk level labels are divided into three levels: high, medium, and low. High risk corresponds to situations where there is a break in the transmission of syndrome-prescription / medication or incompatibilities in drug combinations; medium risk corresponds to situations where there is a mismatch between disease and syndrome, syndrome and treatment method, or treatment method and prescription / medication; and low risk corresponds to situations where only non-standard terminology is used. The evidence chain and revision suggestions include three parts: problem description, conflict location, and suggested adjustment direction. The problem description comes from the evidence chain summary generated by the AI ​​reasoning unit 233; the conflict location indicates the specific location of the defect in the medical record text; and the suggested adjustment direction generates corresponding revision guidelines based on standard matching relationships in a multi-class standard terminology fusion knowledge base.

[0149] The results aggregation and scoring unit 2343 also integrates an early warning interaction function, which is directly connected to the writing interface of the clinical physician workstation 110 via a two-way data channel. The quality control conclusions, risk level labels, evidence chains, and revision suggestions output by the results aggregation and scoring unit 2343 are pushed to the clinical physician workstation 110 via the early warning interaction function, allowing physicians to view and correct them online in a timely manner. For medical records outputting a "review" level, the system simultaneously marks the medical record as awaiting manual review and stores it in the database server 210, allowing quality control personnel to make final confirmation on the medical record quality control management PC terminal 120.

[0150] It should be noted that the above-mentioned weight thresholds and scoring levels are merely exemplary parameters in this embodiment. In practical applications, they can be adaptively adjusted according to the quality control requirements and specialty treatment characteristics of different medical institutions, without departing from the scope of protection of this application.

[0151] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A smart verification method for quality control of TCM medical records based on multiple national standards, characterized in that: The method includes: Construct a knowledge base that integrates multiple types of standard terminology; The system acquires electronic medical record text entered by the front-end physician workstation, performs word segmentation on the electronic medical record text, identifies disease text fragments, syndrome text fragments, treatment text fragments, prescription text fragments, and traditional Chinese medicine text fragments to obtain original text entities; based on the multi-class standard terminology fusion knowledge base, it maps non-standard expressions in the original text entities to standard terms, outputting structured medical record quintuple data, which includes standardized diseases, standardized syndromes, standardized treatment methods, standardized prescriptions, and standardized traditional Chinese medicines; the structured medical record quintuple data is input into an artificial intelligence inference unit, which performs multi-stage semantic parsing on the structured medical record quintuple data to obtain semantic feature vectors, candidate knowledge support sets, and evidence chain summaries; Based on the structured medical record quintuple data, the semantic feature vector, the candidate knowledge support set, and the multi-class standard terminology fusion knowledge base, a two-dimensional theory, method, prescription, and drug linkage verification is performed to generate defect markers. Based on the defect markers and the evidence chain summary, early warning information is pushed to the front-end physician workstation. Retrieve archived medical records from the hospital's final medical record database, reuse the dual-dimensional theory, method, prescription, and drug linkage verification for each archived medical record, and record the defect information of each archived medical record; calculate the completeness rate of key fields, the consistency rate of theory, method, prescription, and drug, and the total defect rate of medical record theory and method information in batches based on the defect information of all archived medical records, and generate defect distribution ledgers and quality control analysis reports according to the department dimension and / or physician dimension and / or disease dimension; The log data is anonymized and stored in a distributed backup storage array. A unique hash code is generated for the anonymized log data, and the unique hash code is associated with and stored with the anonymized log data. The log data includes: knowledge base construction and modification logs generated during the construction and modification of the multi-category standard terminology fusion knowledge base; front-end verification interaction logs generated during the execution of the dual-dimensional theory, law, prescription, and drug linkage verification; and final medical record quality control logs generated during the final quality control of the archived medical records.

2. The method according to claim 1, characterized in that, Construct a multi-category standard terminology fusion knowledge base, including: The original texts of six categories of preset terms were imported, namely, the disease standard terminology library, the syndrome standard terminology library, the treatment standard terminology library, the pharmacopoeia standard terminology library, the internal medicine of traditional Chinese medicine and the quality control standard terminology library of traditional Chinese medicine. After text cleaning, the disease standard terminology set, syndrome standard terminology set, treatment standard terminology set, prescription standard terminology set and traditional Chinese medicine standard terminology set were formed. Based on the preset mapping rules of diseases, syndromes and treatments, a built-in comparison relationship of diseases, syndromes and treatments is generated; Acquire historical clinical data, and iteratively correct the disease-syndrome comparison relationship, syndrome-treatment method comparison relationship, and treatment method-prescription-medicine matching relationship based on the historical clinical data to generate a theory-treatment-prescription-medicine comparison table. The theory, method, prescription, and medicine comparison table is sequentially subjected to terminology uniqueness verification, association logic verification, clinical fit verification, and compliance desensitization and evidence preservation processing. The standardized five-level comparison dataset of diseases, syndromes, treatment methods, prescriptions, and Chinese medicines is output and stored in the database.

3. The method according to claim 2, characterized in that, The aforementioned theory, method, prescription, and medicine comparison table is sequentially subjected to terminology uniqueness verification, association logic verification, clinical fit verification, and compliance desensitization and evidence preservation processing. A standardized five-level comparison dataset of diseases, syndromes, treatment methods, prescriptions, and traditional Chinese medicines is output and stored in the database, including: The terms in the comparison table of theory, method, prescription and medicine are checked for duplicates and duplicate terms and vague expressions are deleted to obtain a deduplicated comparison table. The chain matching relationships of diseases, syndromes, treatments, prescriptions, and Chinese medicines in the deduplicated comparison table are verified one by one to see if they conform to the TCM syndrome differentiation rules and to remove contradictory associations of pathogenesis, so as to obtain the logically verified comparison table. Based on the clinical historical data, the matching entries in the logically validated lookup table that conflict between general standards and specialist diagnosis and treatment are filtered to obtain a clinically adapted lookup table; The clinically adapted control dataset is anonymized, a unique hash value is generated and associated with the anonymized control dataset, the knowledge base construction and modification log is retained, and a standardized five-level control dataset of disease, syndrome, treatment, prescription and Chinese medicine is output and stored in the database.

4. The method according to claim 1, characterized in that, The multi-stage semantic parsing includes: The electronic medical record text is subjected to contextual semantic segmentation, cross-sentence relationship identification, and semantic role labeling to identify key medical semantics in chief complaint, present medical history, four diagnostic information, syndrome differentiation description, treatment method description, and prescription description, and to obtain semantic feature vectors. The extracted disease, syndrome, treatment, prescription, and Chinese medicine expressions are subjected to terminology normalization and semantic disambiguation processing. Diagnostic and treatment terms that are similar in word meaning but different in semantic meaning are distinguished to obtain standardized terms after disambiguation. Based on the semantics of medical record input, similar knowledge retrieval and multi-hop association candidate expansion are performed to generate a candidate knowledge support set; It assists in reasoning about the semantic relationships between diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines, and identifies implicit conflict situations such as local statements being valid but the overall diagnostic direction deviating, similar wording but inconsistent medical semantics, and local reasonableness but inconsistent global links. Based on the identified implicit conflict situations, it generates an evidence chain summary, which includes an explanation of the cause of the problem, a hint of the conflict location, and suggestions for revision.

5. The method according to claim 1, characterized in that, The dual-dimensional linkage verification of theory, method, prescription, and medicine includes: Traverse the structured medical record quintuple data and determine whether there are blank or unfilled markers in the disease field, syndrome field, treatment method field, prescription field, and traditional Chinese medicine field. If they exist, generate field missing markers. Based on the multi-category standard terminology fusion knowledge base, the matching relationship between disease and syndrome, the matching relationship between syndrome and treatment method, and the matching relationship between treatment method and prescription are verified sequentially. If any matching relationship does not satisfy the corresponding relationship in the knowledge base, a logical inconsistency defect label is generated. Based on the semantic feature vector, the candidate knowledge support set, and the evidence chain summary, the logical inconsistency defect label is weighted by confidence and classified by conflict type, and an enhanced logical inconsistency defect label is output. Based on the missing field marker and / or the enhanced logical mismatch defect label, along with the evidence chain summary, a warning message including defect location and revision direction suggestions is pushed to the front-end physician workstation, and the front-end verification interaction log is retained.

6. The method according to claim 5, characterized in that, After pushing warning information including defect location and revision direction suggestions to the front-end physician workstation based on the missing field marker and / or the enhanced logical mismatch defect label, along with the evidence chain summary, and retaining the front-end verification interaction log, the process also includes a syndrome-driven automatic backfilling and cyclic verification release step: Once a standardized disease is obtained and selected by the front-end physician workstation, based on the multi-category standard terminology fusion knowledge base, a standardized syndrome matching the standardized disease is pushed. After the standardized syndrome is selected, the corresponding standardized treatment method is automatically matched and filled into the treatment method input box. At the same time, a prescription recommendation list matching the standardized disease, standardized syndrome and standardized treatment method is pushed. When manual modifications to the automatically backfilled treatment methods or autonomous selection of syndromes and prescriptions are obtained, the matching relationship between the modified diseases and syndromes, the matching relationship between syndromes and treatment methods, and the matching relationship between treatment methods and prescriptions are reviewed again based on the knowledge base of the multi-category standard terms. If a mismatch occurs, the reason for the conflict is highlighted and the recommended entries are refreshed. Repeatedly perform field integrity checks and chain matching checks until there are no missing field markers and no logical mismatch defect labels, then allow the medical record to be saved.

7. The method according to claim 1, characterized in that, The formula for calculating the completeness rate of the key fields is: the number of medical records without omissions divided by the total number of medical records multiplied by 100%; The formula for calculating the consistency rate of theory, treatment, prescription, and medication is: the number of cases with complete matching of disease, syndrome, treatment method, and prescription / medication divided by the total number of cases multiplied by 100%. The formula for calculating the total defect rate of the medical record information is: the number of medical records with any missing field or mismatched field divided by the total number of medical records multiplied by 100%.

8. A smart verification system for quality control of TCM medical records based on multiple national standards, characterized in that: The system includes: The front-end interactive hardware layer, including the clinical physician workstation and the medical record quality control management PC, is connected to the central computing hardware layer through bidirectional communication via the hospital's local area network. The central computing hardware layer includes a database server and an algorithm computing server, which are directly connected by optical fiber to achieve data interaction. The database server stores a knowledge base of multiple standard terms, a full set of medical record data, and quality control log evidence records. The algorithm computing server carries semantic parsing, matching and verification, artificial intelligence reasoning, and quantitative statistical calculation programs. A distributed backup storage array, continuously synchronized with the database server via a data bus, is used for off-site de-identified backup; The software system deployed on the central computing hardware layer includes: The multi-category standard terminology fusion knowledge base construction module is used to import the original text of six types of preset terms, generate a standardized five-level comparison dataset of diseases, syndromes, treatments, prescriptions, and Chinese medicines based on preset disease, syndrome, treatment, and treatment rules and clinical historical data, and store it in the database server. The electronic medical record structured parsing module is used to perform word segmentation and standard terminology mapping on the electronic medical record text uploaded by the front-end physician workstation, and output structured medical record quintuple data. The AI ​​reasoning unit, whose input is connected to the output of the electronic medical record structured parsing module, is used to receive the structured medical record quintuple data, perform multi-stage semantic parsing, structured extraction, terminology normalization and semantic disambiguation, knowledge enhancement retrieval, diagnosis and treatment logic chain reconstruction and semantic consistency judgment, and evidence chain interpretation generation on the structured medical record quintuple data, and output semantic feature vectors, candidate knowledge support sets and evidence chain summaries; A dual-dimensional theory-method-formula-medicine linkage verification module is provided. Its first input is connected to the output of the electronic medical record structured parsing module to obtain the structured medical record five-tuple data. Its second input is connected to the output of the artificial intelligence inference unit to obtain the semantic feature vector, the candidate knowledge support set, and the evidence chain summary. Its third input is connected to the database server to obtain the standardized disease, syndrome, treatment method, formula, and traditional Chinese medicine five-level comparison dataset. This module performs field integrity verification and chain matching verification of disease, syndrome, treatment method, formula, and traditional Chinese medicine based on the structured medical record five-tuple data, the semantic feature vector, the candidate knowledge support set, the evidence chain summary, and the five-level comparison dataset. It generates defect markers and pushes warning information to the front end along with the evidence chain summary. The medical record batch quality control and quantitative statistics module is used to retrieve archived medical records from the hospital's final medical record database for batch quality control verification, calculate the completeness rate of key fields, the consistency rate of theory, method, prescription and drug, and the total defect rate of medical record theory and method information, and generate a defect distribution ledger and quality control analysis report. The end-to-end log storage and compliance module is used to de-identify log data and store it in the distributed backup storage array. It also generates a unique hash code for the de-identified log data and associates this unique hash code with the de-identified log data for storage. The log data includes: knowledge base construction and modification logs generated by the multi-standard terminology fusion knowledge base construction module during knowledge base construction and modification; front-end verification interaction logs generated by the dual-dimensional theory, law, prescription, and drug linkage verification module during verification execution; and final medical record quality control logs generated by the batch quality control and quantitative statistics module during final quality control of archived medical records.

9. The system according to claim 8, characterized in that, The multi-standard terminology fusion knowledge base construction module includes a standard terminology import subunit, a mapping rule generation subunit, a clinical iterative optimization subunit, and a four-layer integrated quality control and verification subunit connected in sequence. The standard terminology import subunit is equipped with four data source interfaces, which are respectively connected to the disease standard terminology import interface, syndrome standard terminology import interface, treatment standard terminology import interface, pharmacopoeia standard terminology import interface, traditional Chinese medicine internal medicine standard terminology import interface and traditional Chinese medicine quality control standard terminology import interface. The four-layer integrated quality control verification subunit internally integrates a terminology uniqueness verification unit, an association logic verification unit, a clinical adaptation verification unit, and a compliance de-identification and evidence storage unit. The output of the multi-standard terminology fusion knowledge base construction module is connected to the five-level control dataset storage area. The standardized disease, syndrome, treatment, prescription, and Chinese medicine five-level control dataset in the five-level control dataset storage area includes a complete set of standard disease entries, a complete set of standard syndrome entries, a complete set of standard treatment entries, an in-hospital extended Chinese medicine entry set, a built-in associated control group for diseases, syndromes and treatments, and a matching prescription and medicine entry group.

10. The system according to claim 8, characterized in that, The artificial intelligence reasoning unit includes: The multi-stage semantic parsing subunit is used to perform contextual semantic segmentation, cross-sentence relationship identification, and semantic role labeling on the electronic medical record text, identify key medical semantics in the chief complaint, present medical history, four diagnostic information, syndrome differentiation description, treatment method description, and prescription description, and output semantic feature vectors. The structured extraction subunit is used to extract diseases, syndromes, treatments, prescriptions, Chinese medicines and their related descriptions from the electronic medical record text, generate structured objects that can be verified later, and perform completeness perception marking on the missing, weak expression and abnormal position of key diagnostic and treatment elements. The terminology normalization and semantic disambiguation subunit is used to combine the multi-class standard terminology fusion knowledge base and synonym mapping relationship to map non-standard expressions into standard terms, and to disambiguate diagnostic and treatment terms that are similar in word meaning but different in semantic meaning. The knowledge-enhanced retrieval subunit is used to vectorize knowledge items related to diseases, syndromes, treatments, prescriptions, and traditional Chinese medicines, and to perform similar knowledge retrieval and multi-hop association candidate expansion based on the semantics of medical record input, locate evidence fragments, and generate candidate knowledge support sets. The diagnostic logic chain reconstruction and semantic consistency judgment subunit is used to assist in judging the semantic association between diseases, syndromes, treatment methods, prescriptions, and Chinese medicines, reconstruct the path of syndrome differentiation and treatment, identify implicit conflict situations such as local expressions being valid but the overall syndrome differentiation direction deviating, similar wording but inconsistent medical semantics, and local rationality but inconsistent global links, and generate conflict type labels. The evidence chain interpretation and generation subunit is used to generate a problem cause description, conflict location prompt, revision direction suggestion and evidence chain summary that can be manually reviewed based on the conflict type label, the candidate knowledge support set and the completeness awareness mark, and output the evidence chain summary to the dual-dimensional theory, law, prescription and drug linkage verification module.

Images