Structured data extraction method and system for electronic medical record

Abstract

The application discloses a structured data extraction method and system of electronic medical records, and relates to the technical field of medical information, which comprises the following steps: constructing a three-database joint retrieval system; obtaining electronic medical record texts and patient payment documents, performing structured pre-segmentation on the electronic medical record texts to obtain segmented text segments; performing forward filling and reverse filling based on the segmented text segments to generate forward filling results and reverse filling results; performing forward and reverse conflict processing on the forward filling results and the reverse filling results, adopting a high-confidence result as a main value when the forward filling result of a same segmented text segment is inconsistent with the reverse filling result, and marking the difference; comparing doctor record contents with the three-database joint retrieval system, performing conflict marking based on the comparison result, and cross- verifying the completeness of disposal items in combination with the patient payment documents; and outputting structured data, which helps to solve the problem that the prior art cannot realize medical term normalization.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and in particular to a method and system for extracting structured data from electronic medical records. Background Technology

[0002] In the process of building healthcare information systems, electronic medical records, as the core data carrier for clinical diagnosis and treatment, medical insurance settlement, scientific research analysis, and quality control, contain a large amount of unstructured free text information. Doctors often use colloquial, abbreviated, and personalized expressions when writing medical records, resulting in multiple writing formats for the same medical term. Information such as tooth position, diagnosis, treatment plan, and procedures are recorded in a mixed manner, making it difficult to directly use electronic medical records for automated retrieval, statistical analysis, and data reuse.

[0003] Current methods for structuring electronic medical records largely rely on manual annotation, simple keyword matching, or regular expression extraction, resulting in low efficiency, poor consistency, and limited coverage. Some technologies attempt to use large language models for text extraction and normalization, but these are prone to issues such as terminology drift, content illusion, and information omissions, failing to guarantee the accuracy of medical semantics. Furthermore, most existing solutions employ only a one-way extraction logic; when doctors' written content is incomplete or intermediate diagnostic and treatment levels are omitted, it is difficult to automatically fill in the missing information, leading to incomplete structured results. In addition, the treatment information in medical records lacks a mechanism for cross-verification with external documents, making it impossible to verify the authenticity and completeness of treatment items, thus affecting data credibility.

[0004] Therefore, there is an urgent need for a structured extraction method for electronic medical records that can achieve standardized and normalized medical terminology, bidirectional hierarchical completion, semantic fidelity, and verification of treatment information in conjunction with payment receipts. Summary of the Invention

[0005] In view of this, the present invention proposes a structured data extraction method and system for electronic medical records, which can achieve standardized normalization of medical terminology, bidirectional hierarchical completion, semantic fidelity, and can be combined with payment receipts for verification of treatment information in electronic medical records.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for extracting structured data from electronic medical records includes: Construct a standard graph knowledge base, a customized graph knowledge base, and a customized document knowledge base in the medical field to form a joint retrieval system of the three bases; Obtain electronic medical record text and patient payment receipts; perform structured pre-segmentation on the electronic medical record text to obtain segmented text fragments including standard terminology, original text basis, and confidence level; Based on the segmented text fragments, perform positive fill-in to generate positive fill-in results; Based on the segmented text fragments, reverse fill is performed to generate reverse fill results; The positive and negative filling results are processed for conflict. When the positive and negative filling results of the same text segment are inconsistent, the high confidence result is used as the principal value and the difference is marked. The doctor's records are compared with the three-database joint retrieval system. Conflict marking is performed based on the comparison results, and the completeness of the treatment items is cross-verified in conjunction with the patient's payment receipts. Output structured data carrying standard terminology, source text, confidence level, and conflict markers.

[0007] Based on the above technical solution, the present invention can be further improved as follows: Optionally, the step of obtaining the electronic medical record text and the patient's payment receipt, and performing structured pre-segmentation on the electronic medical record text to obtain segmented text fragments including standard terminology, original text evidence, and confidence levels, includes: The electronic medical record text is segmented according to a preset structured template, and the electronic medical record text is divided into at least one segmented text fragment among chief complaint, diagnosis, treatment suggestion, treatment plan, treatment, and medical order; Match each segment of text with corresponding standard terminology, source text references, and confidence levels.

[0008] Optionally, the matching of standard terms, original text basis, and confidence level for each segmented text fragment includes: The segmented text fragments were subjected to terminology normalization matching using a three-database joint retrieval system to obtain standard terms. The original text corresponding to the segmented text fragments shall be used as the original text basis; Confidence scores are generated based on search relevance and semantic consistency.

[0009] Optionally, the step of performing positive fill based on the segmented text fragments to generate positive fill results includes: Starting from the diagnostic level, the process proceeds sequentially through treatment suggestions, treatment plans, and finally to the treatment level. Based on the three-database joint retrieval system and the large language model, the segmented text fragments are reordered from top to bottom to obtain positive filling results.

[0010] Optionally, the step of performing reverse fill based on the segmented text fragments to generate a reverse fill result includes: Starting from the treatment level, the upper-level diagnosis level is deduced sequentially through treatment plans and treatment recommendations, and the omitted intermediate level elements are filled in to obtain the reverse filling result.

[0011] Optionally, the conflict marker includes at least one of terminology conflict, path conflict, or hierarchy missing.

[0012] Optionally, the simultaneous cross-validation of the completeness of the processing items based on patient payment receipts includes: The treatment items extracted from the electronic medical record text will be compared with the treatment items and drug list in the patient's payment receipt; If the patient's payment receipt exists but is not recorded in the electronic medical record, a supplementary prompt will be triggered; If the electronic medical record exists but the patient's payment receipt does not reflect it, it will be marked as pending verification.

[0013] A structured data extraction system for electronic medical records, comprising: The three-database construction module is used to build a standard graph knowledge base, a customized graph knowledge base, and a customized document knowledge base in the medical field, forming a joint retrieval system of the three databases; The pre-segmentation module is used to acquire electronic medical record text and patient payment receipts, and to perform structured pre-segmentation on the electronic medical record text to obtain segmented text fragments including standard terminology, original text basis, and confidence level. The positive fill module is used to perform positive fill based on the segmented text fragments and generate positive fill results; The reverse filling module is used to perform reverse filling based on the segmented text fragments and generate reverse filling results; The conflict handling module is used to handle the positive and negative conflicts between the positive and negative filling results. When the positive and negative filling results of the same segment of text are inconsistent, the high confidence result is used as the main value and the difference is marked. The comparison module is used to compare the doctor's records with the three-database joint retrieval system, perform conflict marking based on the comparison results, and cross-verify the integrity of the treatment items in conjunction with the patient's payment receipts; The results output module is used to output structured data carrying standard terminology, source text, confidence level, and conflict markers.

[0014] An electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the method described herein.

[0015] A non-transitory computer-readable storage medium having a computer program stored thereon, the computer program implementing the steps of the method when executed by a processor.

[0016] The present invention has the following advantages: This invention presents a structured data extraction method for electronic medical records. Through a three-database joint retrieval system, it achieves precise normalization of medical terminology, avoiding terminology drift and content illusion, thus improving data accuracy. A bidirectional filling mechanism, employing both forward and reverse filling, automatically completes missing diagnostic and treatment level information in medical records, resolving text incompleteness issues. Conflict handling based on confidence levels enhances the reliability of results, and cross-verification of information using payment receipts ensures data authenticity and completeness. The final output is standardized structured data, significantly improving the efficiency of electronic medical record processing. Attached Figure Description

[0017] For illustrative and not limiting purposes, the present invention will now be described in conjunction with embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the structured data extraction method for electronic medical records according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the main components of the structured data extraction system for electronic medical records in an embodiment of the present invention; Figure 3 This is a schematic diagram of the physical structure of the electronic device provided by the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0019] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be used interchangeably where appropriate for the embodiments of the invention described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] It should be noted that, where there is no conflict, the embodiments and features of the present invention can be combined with each other. The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] Figure 1This is a flowchart illustrating the structured data extraction method for electronic medical records according to an embodiment of the present invention, as shown below. Figure 1 As shown, the structured data extraction method for electronic medical records provided in this embodiment of the invention includes the following steps S101 to S107.

[0022] S101, construct a standard graph knowledge base, a customized graph knowledge base, and a customized document knowledge base in the medical field to form a joint retrieval system of the three bases.

[0023] The medical field standard diagram knowledge base contains ICD codes, standard dental terminology, and a standard hierarchical path from diagnosis to treatment recommendations to treatment plans to management. The standard graph knowledge base in the medical field can provide normalization objectives and hierarchical constraints.

[0024] The customized graph knowledge base includes medical institutions' custom treatment pathways, local terminology mappings, and normalized prototypes of historical cases. A customized graph knowledge base can cover organization-specific syntax and local specifications.

[0025] The customized document knowledge base consists of vectorized indexes of unstructured documents such as treatment guidelines and operating procedures. Customized document knowledge bases can supplement long-tail scenarios not covered by graph knowledge bases.

[0026] In the medical field, a standard graph knowledge base, a customized graph knowledge base, and a customized document knowledge base jointly define the hierarchical constraints of structured output: under the same diagnosis, the candidate set of treatment suggestions is limited; under the same treatment suggestion, the candidate set of treatment plans is limited; and so on until the doctor's order.

[0027] S102, obtain the electronic medical record text and patient payment receipt, perform structured pre-segmentation on the electronic medical record text, and obtain segmented text fragments including standard terminology, original text basis, and confidence level.

[0028] Electronic medical record text: Free text written by doctors, including chief complaint, examination findings, diagnosis, treatment, medical orders, etc. Patient payment receipts: These include a list of treatment items, a list of medications, and the amount, and are used to cross-verify the completeness of the treatment. The electronic medical record text is segmented according to a preset structured template, and the electronic medical record text is divided into at least one segmented text fragment among chief complaint, diagnosis, treatment suggestion, treatment plan, treatment, and medical order; Match each segment of text with corresponding standard terminology, source text references, and confidence levels.

[0029] The segmented text fragments were subjected to terminology normalization matching using a three-database joint retrieval system to obtain standard terms. The original text corresponding to the segmented text fragments shall be used as the original text basis; Confidence scores are generated based on search relevance and semantic consistency.

[0030] Segmented text fragments { Diagnosis: [...], Treatment recommendations: [...] Treatment plan: [...], "Disposal": [...], "Next suggestion / plan": [...], "Doctor's orders": [...] } S103, perform positive fill based on segmented text fragments, and generate positive fill results.

[0031] Starting from the diagnostic level, the process proceeds sequentially through treatment suggestions, treatment plans, and finally to the treatment level. Based on the three-database joint retrieval system and the large language model, the segmented text fragments are reordered from top to bottom to obtain positive filling results.

[0032] S104, Perform reverse filling based on segmented text fragments to generate reverse filling results.

[0033] Starting from the treatment level, the upper-level diagnosis level is deduced sequentially through treatment plans and treatment recommendations, and the omitted intermediate level elements are filled in to obtain the reverse filling result.

[0034] For each pre-segmented text fragment after structured pre-segmentation, the system simultaneously initiates parallel searches to the three-database joint retrieval system: For the standard graph knowledge base and the customized graph knowledge base in the medical field, graph path matching is used to obtain the node positions, candidate lower-level nodes, and local medical terminology and unique treatment paths in the standard treatment path; for the customized document knowledge base, vector similarity retrieval is used to supplement long-tail medical expressions not covered by the graph knowledge base. After merging the search results from the three databases, a Large Language Model (LLM) is used for re-ranking, selecting the optimal candidate that is semantically consistent with the original medical record. The re-ranking process is strictly constrained, retaining only content that semantically matches the original text, without generating or inferring information not recorded in the original text, avoiding model illusions, and ensuring the authenticity and reliability of the extracted results.

[0035] S105, perform forward and reverse conflict handling on the forward and reverse filling results. When the forward and reverse filling results of the same segment of text are inconsistent, the high confidence result is used as the principal value and the difference is marked.

[0036] The positive and negative fill results are compared item by item according to the structured field dimension. The fill result is selected based on the confidence level value, and the inconsistent field positions and contents are marked to form difference annotation information.

[0037] S106 compares the doctor's records with the three-database joint retrieval system, performs conflict marking based on the comparison results, and cross-verifies the completeness of the treatment items in conjunction with the patient's payment receipts.

[0038] The conflict markers include at least one of terminology conflict, path conflict, or hierarchy missing.

[0039] The treatment items extracted from the electronic medical record text will be compared with the treatment items and drug list in the patient's payment receipt; If the patient's payment receipt exists but is not recorded in the electronic medical record, a supplementary prompt will be triggered; If the electronic medical record exists but the patient's payment receipt does not reflect it, it will be marked as pending verification.

[0040] S107 outputs structured data carrying standard terminology, source text, confidence level, and conflict markers.

[0041] The data, after being processed through forward and reverse filling, conflict handling, conflict marking, and cross-validation, are integrated to form a complete structured data set containing standard terminology, source text, confidence level, conflict marking, execution status, and prompts, which is then output.

[0042] Example 1: Structured extraction of initial medical records from a dental clinic: The input material is: electronic medical record text (original handwritten by the doctor): The patient is a 42-year-old male. Chief complaint: pain in the lower left posterior teeth for 3 days, which worsens at night.

[0043] Examination: Tooth 36 showed deep caries on the distal occlusal surface, with positive probing and percussion, and negative mobility. There was no obvious redness or swelling of the gingiva. Periapical radiographs showed mild widening of the periapical ligament of tooth 36.

[0044] >Consider the acute exacerbation of chronic pulpitis in tooth 36, accompanied by early-stage periapical periodontitis.

[0045] Today we performed a pulpotomy to decompress the tissue, applied a devitalizing agent, and scheduled root canal preparation for next week. Doctor's orders: Avoid chewing on the affected side and attend all follow-up appointments.

[0046] Patient payment receipts: |Item|Quantity|Amount| |---|---|---| |Marrow opening surgery|1|180 yuan| |Deactivating agent sealing|1|50 yuan| |Root tip slices|1|60 yuan| LLM pre-segments the medical record text according to the schema and identifies the boundaries of each element: |schema field|segmentation result| Chief complaint: "Pain in the lower left posterior tooth for 3 days, worsening at night" |Oral Examination| "Depth caries on the distal occlusal surface of tooth 36, tenderness upon probing (+), percussion (+), mobility (-), no obvious redness or swelling of the gingiva"| |Auxiliary Examination| "Peripheral X-ray shows mild widening of the periapical membrane at 36"| |Diagnosis|"Considers acute exacerbation of chronic pulpitis in tooth 36, with early periapical periodontitis"| |Treatment| "Today, perform bone marrow decompression and administer devitalizing agent."| |Next Plan| "Schedule root canal preparation for next week"| |Doctor's advice| "Avoid chewing on the affected side and attend follow-up appointments on time"|; Note: The same schema field may appear multiple times (two instances appeared in the diagnosis of Example 1, which were handled separately).

[0047] Joint search of three databases (taking the "diagnosis" field as an example): Segmentation: "Consider an acute exacerbation of chronic pulpitis in tooth 36" Standard graph knowledge base path matching: Node "Acute exacerbation of chronic pulpitis" was matched; the standard path is: Acute exacerbation of chronic pulpitis → Root canal treatment → Pulp opening and drainage / root canal preparation / root canal filling → Postoperative instructions; Customized graph knowledge base: Hit the institutional terminology mapping "acute exacerbation of chronic pulpitis ≈ acute exacerbation phase of chronic pulpitis"; Customized document knowledge base: Vector retrieval returns relevant paragraphs from the "Guidelines for the Diagnosis and Treatment of Endodontic Diseases" to confirm diagnostic criteria; Candidates: ["Acute exacerbation of chronic pulpitis", "Acute pulpitis", "Chronic pulpitis"]; LLM judgment: The original text "acute exacerbation of chronic pulpitis" and "acute exacerbation period of chronic pulpitis" are semantically consistent. The top 1 value "acute exacerbation period of chronic pulpitis" is selected with a confidence level of 0.95.

[0048] Positive result reporting (diagnosis → treatment recommendations → treatment plan): json { "diagnosis":[ {"Standard Terminology":"Acute exacerbation of chronic pulpitis","Original Text":"Consider acute exacerbation of chronic pulpitis","Definition":"Proposed diagnosis","Confidence Level":0.95}, {"Standard Terminology":"Periapical Periodontitis (Early Stage)","Original Text":"Early Stage with Periapical Periodontitis","Definition":"Suspected Diagnosis","Confidence Level":0.88} ], Treatment recommendations:[ {"Standard Terminology":"Root Canal Treatment","Source":"Knowledge Base Path Derivation","Confidence Level":0.93} ], Treatment plan:[ {"Standard Terminology":"Opening and Drainage + Root Canal Preparation + Root Canal Filling","Source":"Knowledge Base Path Derivation","Confidence Level":0.91}; ] } Fill in the blanks in reverse: The medical record clearly states "treatment" in the "treatment" field: "opening the medullary canal for decompression and sealing with a devitalizing agent." The reasoning can be deduced by working backwards from the treatment: Treatment plan: "Opening the medullary canal to decompress" → "Opening the medullary canal for drainage" → "Root canal treatment" The results of reverse filling are consistent with the results of forward filling, and the confidence level is increased accordingly. In this embodiment, the results of forward and reverse filling are consistent and there is no conflict, so they are directly merged.

[0049] Knowledge base hierarchy check: Diagnosis of "acute exacerbation of chronic pulpitis" → Treatment of "open pulp decompression": Pathway matching, no conflict; Diagnosis of "periapical periodontitis (early stage)" → Treatment of "opening the medullary canal for decompression": the pathways match and there is no conflict; Cross-validation of payment receipts: |Bill Items|Medical Record Processing|Results| |---|---|---| |Opening the bone marrow | Opening the bone marrow to relieve pressure | Matching | |Deactivating agent sealing|Sealing deactivating agent|Matching| |Peripheral apex radiograph|Peripheral apex radiograph (auxiliary examination)|Matching|; The bill and medical record match perfectly, and the integrity of the treatment has been verified.

[0050] Summary of quality inspection and final output: LLM overall quality check: The hierarchical relationship of each field is consistent, there are no semantic contradictions, and the final structured result is output. json { "diagnosis":[ {"Standard Terminology":"Acute exacerbation of chronic pulpitis","Original Text":"Consider acute exacerbation of chronic pulpitis","Definition":"Proposed diagnosis","Confidence Level":0.95}, {"Standard Terminology":"Periapical Periodontitis (Early Stage)","Original Text":"Early Stage with Periapical Periodontitis","Definition":"Suspected Diagnosis","Confidence Level":0.88} ], Treatment Recommendation: [{"Standard Terminology":"Root Canal Treatment","Confidence Level":0.93}], Treatment Plan: [{"Standard Terminology":"Opening and Drainage + Root Canal Preparation + Root Canal Filling","Confidence Level":0.91}], "Disposal":[ {"Standard Terminology":"Opening and Drainage","Original Text":"Opening and Decompression","Status":"Performed","Confidence Level":0.97}, {"Standard Terminology":"Intramedullary Medication Sealing (Devitalizing Agent)","Original Text":"Sealing Devitalizing Agent","Status":"Executed","Confidence Level":0.96} ], "Next suggestion / solution":[ {"Standard Terminology":"Root Canal Preparation","Original Text":"Schedule root canal preparation for next week","Status":"Plan","Confidence Level":0.95} ], "Doctor's orders":[ {"Content":"Avoid chewing on the affected side","Confidence level":0.99}, {"Content":"Timely follow-up visits","Confidence level":0.99} ] } Example 2: Extraction of conflicting medical records: Input materials: Electronic medical record text: >48. Redness and swelling around the crown, tenderness, and limited mouth opening. Diagnosis: Pericoronitis of the wisdom tooth.

[0051] Treatment: Remove tube 48. Doctor's orders: Oral anti-inflammatory medication for 3 days.

[0052] Patient payment receipts: |Item|Quantity|Amount| |---|---|---| |Pericoronal flushing|1|30 yuan| Oral antibiotics | 1 | 45 yuan | Standard diagram knowledge base path: Acute pericoronitis of the wisdom tooth → Acute phase treatment recommendations: anti-inflammatory drainage, extraction is contraindicated during the acute phase → Treatment plan: pericoronal irrigation and medication → extraction at a later date after inflammation is controlled. Conflict identification: The doctor's medical record states "extraction of tooth 48", but the standard path in the knowledge base requires that extraction is contraindicated in the acute phase and that pericoronal irrigation should be performed first to control inflammation.

[0053] Conflict resolution results: json { "Disposal":[ { Original text: "Removed 48", "Standard Terminology Candidate": "Tooth Extraction (48)", "Conflict marker": "Path conflict", "Conflict Explanation": "Knowledge Base Path: In the acute phase of pericoronitis, anti-inflammatory drainage should be performed first; tooth extraction is not recommended during the acute phase." Confidence level: 0.41 "Processing": "Pending manual confirmation" } ] } Cross-validation of payment receipts: The bill includes "pericoronal irrigation," but the medical record does not record pericoronal irrigation, triggering a supplementary prompt: [Additional Note] The billing item "pericoronal flushing" could not be found in the medical record processing. Please check if it was omitted.

[0054] Final output (including conflict markers): json { Diagnosis: [{"Standard Terminology":"Acute Pericoronitis","Original Text":"Pericoronitis","Definition":"Confirmed Case","Confidence Level":0.96}], Treatment Recommendation: [{"Standard Terminology":"Anti-inflammatory drainage, elective tooth extraction","Source":"Knowledge Base Path Derivation","Confidence Level":0.92}], Treatment plan:[ {"Standard Terminology":"Pericoronal irrigation and medication application","Source":"Bill Supplement Notice","Status":"Pending Confirmation"}, {"Standard Terminology":"Tooth Extraction (48)","Original Text":"Extraction 48","Conflict Marker":"Path Conflict","Confidence Level":0.41,"Status":"Pending Human Confirmation"} ], "Doctor's Order":[{"Content":"Oral anti-inflammatory medication for 3 days","Confidence Level":0.98}] } Example 3: The doctor only writes the treatment and fills in the upper-level elements in reverse. Input material: Electronic medical record text (extremely simplified writing): >11. Grade III mobility, tenderness upon percussion, periodontal pocket 8mm.

[0055] Remove 11 and make an impression.

[0056] Positive Filing: If the medical record does not have a clear diagnosis field, positive filing derives diagnostic candidates from the "chief complaint / examination": Search results: "Grade III mobility + 8mm periodontal pocket" → Candidate diagnosis: "Severe periodontitis (11)", confidence level 0.87 Fill in the blanks in reverse: Inverse reasoning from the handling of "Removal 11": - Tooth extraction → Corresponding treatment plan node: "Tooth extraction + Restoration" -Repair Plan → Corresponding Treatment Recommendation: "Post-extraction restoration (implant / fixed bridge / removable denture)" Treatment recommendations → Corresponding diagnostic candidates: "Severe periodontitis / Dental trauma / Root fracture" LLM combined the examination information (Grade III mobility, 8mm periodontal pocket) and reordered it to confirm the diagnosis as "severe periodontitis (11)", which is consistent with the positive filling.

[0057] Final output: json { "Diagnosis":[{"Standard Terminology":"Severe Periodontitis (11)","Source":"Forward + Backward Bidirectional Derivation","Confidence Level":0.87}], Treatment Recommendation: [{"Standard Terminology":"Post-extraction Restoration","Source":"Reverse Fill Derivation","Confidence Level":0.85}], Treatment Plan: [{"Standard Terminology":"Extraction + Restoration Plan Pending","Source":"Reverse Completion Derivation","Confidence Level":0.83}], "Disposal":[ {"Standard Terminology":"Tooth Extraction (11)","Original Text":"Extraction 11","Status":"Performed","Confidence Level":0.98}, {"Standard Terminology":"Impression Making","Original Text":"Impression Making","Status":"Executed","Confidence Level":0.97} ], "Next Recommendation / Plan":[{"Standard Terminology":"Restoration Plan (Implant / Fixed Bridge / Removable Denture)","Status":"Pending Doctor's Confirmation"}] } Figure 2 This is a schematic diagram illustrating the main components of the structured data extraction system for electronic medical records according to an embodiment of the present invention. Figure 2As shown, the structured data extraction system 1 for electronic medical records provided in this embodiment of the invention includes a three-database construction module 10, a pre-segmentation module 20, a forward filling module 30, a reverse filling module 40, a conflict handling module 50, a comparison module 60, and a result output module 70.

[0058] The three-database construction module 10 is used to build a standard graph knowledge base, a customized graph knowledge base, and a customized document knowledge base in the medical field, forming a joint retrieval system of the three databases; The pre-segmentation module 20 is used to acquire electronic medical record text and patient payment receipts, perform structured pre-segmentation on the electronic medical record text, and obtain segmented text fragments including standard terminology, original text basis, and confidence level. The forward filling module 30 is used to perform forward filling based on the segmented text fragments and generate a forward filling result; The reverse filling module 40 is used to perform reverse filling based on the segmented text fragments and generate a reverse filling result; The conflict handling module 50 is used to perform forward and reverse conflict handling on the forward filling results and the reverse filling results. When the forward filling results and reverse filling results of the same segment of text are inconsistent, the high confidence result is used as the main value and the difference is marked. The comparison module 60 is used to compare the doctor's record content with the three-database joint retrieval system, perform conflict marking based on the comparison results, and cross-verify the integrity of the treatment items in conjunction with the patient's payment receipt; The results output module 70 is used to output structured data carrying standard terminology, source text, confidence level, and conflict markers.

[0059] Figure 3 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as... Figure 3 As shown, the electronic device 80 includes: a processor 801, a memory 802, and a bus 803; The processor 801 and the memory 802 communicate with each other via the bus 803. The processor 801 is used to call program instructions in the memory 802 to execute the methods provided in the above-described method embodiments, and to execute the methods provided in the embodiments of the present invention.

[0060] This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions, which cause a computer to execute the method provided in this embodiment of the invention.

[0061] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various storage media capable of storing program code, such as ROM, RAM, magnetic disk, or optical disk.

[0062] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for extracting structured data from electronic medical records, characterized in that, include: Construct a standard graph knowledge base, a customized graph knowledge base, and a customized document knowledge base in the medical field to form a joint retrieval system of the three bases; Obtain electronic medical record text and patient payment receipts; perform structured pre-segmentation on the electronic medical record text to obtain segmented text fragments including standard terminology, original text basis, and confidence level; Based on the segmented text fragments, perform positive fill-in to generate positive fill-in results; Based on the segmented text fragments, reverse fill is performed to generate reverse fill results; The positive and negative filling results are processed for conflict. When the positive and negative filling results of the same text segment are inconsistent, the high confidence result is used as the principal value and the difference is marked. The doctor's records are compared with the three-database joint retrieval system. Conflict marking is performed based on the comparison results, and the completeness of the treatment items is cross-verified in conjunction with the patient's payment receipts. Output structured data carrying standard terminology, source text, confidence level, and conflict markers.

2. The method for extracting structured data from electronic medical records according to claim 1, characterized in that, The process of acquiring electronic medical record text and patient payment receipts, and performing structured pre-segmentation on the electronic medical record text to obtain segmented text fragments including standard terminology, source material, and confidence levels, includes: The electronic medical record text is segmented according to a preset structured template, and the electronic medical record text is divided into at least one segmented text fragment among chief complaint, diagnosis, treatment suggestion, treatment plan, treatment, and medical order; Match each segment of text with corresponding standard terminology, source text references, and confidence levels.

3. The method for extracting structured data from electronic medical records according to claim 2, characterized in that, The standard terminology, source text basis, and confidence level for matching each segmented text fragment include: The segmented text fragments were subjected to terminology normalization matching using a three-database joint retrieval system to obtain standard terms. The original text corresponding to the segmented text fragments shall be used as the original text basis; Confidence scores are generated based on search relevance and semantic consistency.

4. The method for extracting structured data from electronic medical records according to claim 1, characterized in that, The step of performing positive fill based on the segmented text fragments to generate positive fill results includes: Starting from the diagnostic level, the process proceeds sequentially through treatment suggestions, treatment plans, and finally to the treatment level. Based on the three-database joint retrieval system and the large language model, the segmented text fragments are reordered from top to bottom to obtain positive filling results.

5. The method for extracting structured data from electronic medical records according to claim 1, characterized in that, The process of performing reverse fill based on the segmented text fragments to generate reverse fill results includes: Starting from the treatment level, the upper-level diagnosis level is deduced sequentially through treatment plans and treatment recommendations, and the omitted intermediate level elements are filled in to obtain the reverse filling result.

6. The method for extracting structured data from electronic medical records according to claim 1, characterized in that, The conflict markers include at least one of terminology conflict, path conflict, or hierarchy missing.

7. The method for extracting structured data from electronic medical records according to claim 1, characterized in that, The simultaneous cross-validation of the completeness of the processing items based on patient payment receipts includes: The treatment items extracted from the electronic medical record text will be compared with the treatment items and drug list in the patient's payment receipt; If the patient's payment receipt exists but is not recorded in the electronic medical record, a supplementary prompt will be triggered; If the electronic medical record exists but the patient's payment receipt does not reflect it, it will be marked as pending verification.

8. A structured data extraction system for electronic medical records, characterized in that, include: The three-database construction module is used to build a standard graph knowledge base, a customized graph knowledge base, and a customized document knowledge base in the medical field, forming a joint retrieval system of the three databases; The pre-segmentation module is used to acquire electronic medical record text and patient payment receipts, and to perform structured pre-segmentation on the electronic medical record text to obtain segmented text fragments including standard terminology, original text basis, and confidence level. The positive fill module is used to perform positive fill based on the segmented text fragments and generate positive fill results; The reverse filling module is used to perform reverse filling based on the segmented text fragments and generate reverse filling results; The conflict handling module is used to handle the positive and negative conflicts between the positive and negative filling results. When the positive and negative filling results of the same segment of text are inconsistent, the high confidence result is used as the main value and the difference is marked. The comparison module is used to compare the doctor's records with the three-database joint retrieval system, perform conflict marking based on the comparison results, and cross-verify the integrity of the treatment items in conjunction with the patient's payment receipts; The results output module is used to output structured data carrying standard terminology, source text, confidence level, and conflict markers.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images