Medical aid diagnosis method and system

Abstract

The application discloses a kind of medical auxiliary diagnosis method and system, including S1: receiving medical record information, through semantic analysis, key information extraction and structured processing, generate structured medical data set;S2: the medical data set is compared with the data in medical knowledge graph, rule base, generate suspected diagnosis result set;S3: the suspected diagnosis result set is checked and confidence score is scored, filter suspected diagnosis result that does not comply with compliance standard or confidence is lower than threshold, output compliance high confidence diagnosis set;S4: diagnosis recommendation list is generated based on the compliance high confidence diagnosis set, and according to the associated path of the medical knowledge graph, the medical basis source corresponding to each diagnosis recommendation is labeled.Using the present application can provide scientific diagnosis reference and complete basis traceability for doctors, improve the efficiency and quality of clinical diagnosis.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence technology, specifically a medical auxiliary diagnostic method and system. Background Technology

[0002] In current clinical diagnostic work, the traditional model faces many challenges. On the one hand, medical record data is mostly in unstructured form (such as handwritten medical records, oral records, electronic texts, etc.), containing complex information such as symptom descriptions, physical signs, past medical history, and examination results. Doctors need to spend a lot of time manually sorting out key information, and the accuracy of diagnosis is easily affected by omissions or misunderstandings. On the other hand, the medical knowledge system is vast and updated rapidly, making it difficult for doctors to fully grasp the relationships and key points of differentiation of all diseases. Especially for rare diseases or complex complications, relying solely on experience can easily lead to misdiagnosis or missed diagnosis.

[0003] Existing assisted diagnostic systems have significant shortcomings: some systems can only perform simple matching of structured data and cannot process the deep semantic information in unstructured medical records; some systems rely on fixed rule reasoning, lack flexibility, and are difficult to adapt to complex and ever-changing clinical scenarios; more importantly, systems that rely solely on large models for assisted diagnosis lack interpretability of diagnostic results, only outputting the name of the suspected disease and failing to show doctors the source of medical knowledge on which the diagnosis is based, resulting in low trust in the diagnostic results among doctors and difficulty in effectively assisting clinical decision-making. Summary of the Invention

[0004] The technical objective of this invention is to address the above-mentioned shortcomings by providing a medical auxiliary diagnostic method and system. This aims to solve the problems in existing technologies by integrating the advantages of large models, knowledge graphs, and rule bases to construct an intelligent auxiliary diagnostic system that combines accuracy, interpretability, and security. This provides doctors with scientific diagnostic references and complete evidence traceability, thereby improving the efficiency and quality of clinical diagnosis.

[0005] In a first aspect, the present invention provides a medical auxiliary diagnostic method, comprising the following steps: S1: Receive medical record information and generate a structured medical data set through semantic parsing, key information extraction and structured processing; S2: Compare the medical data set with the data in the medical knowledge graph and rule base to generate a set of suspected diagnosis results; S3: Perform compliance verification and confidence scoring on the set of suspected diagnostic results, filter out suspected diagnostic results that do not meet compliance standards or have a confidence level below the threshold, and output a set of compliant high-confidence diagnoses; S4: Generate a list of recommended diagnoses based on the compliant high-confidence diagnostic set, and label the source of medical evidence for each diagnostic recommendation according to the association path of the medical knowledge graph.

[0006] In S1, receiving medical record information includes: receiving text medical records manually entered by doctors, receiving electronic medical record documents, receiving voice-input medical records and converting them into text, and receiving medical record data in different formats.

[0007] In step S1, a structured medical data set is generated through semantic parsing, key information extraction, and structured processing. This includes: using a large model fine-tuned and trained with a medical corpus, identifying key medical entities in the medical record information through word segmentation, part-of-speech tagging, and entity recognition. These key medical entities include symptoms, signs, past medical history, and examination results. Then, through semantic role labeling and dependency parsing, the logical relationships between these medical entities are mined. These logical relationships include at least the association between medical history and symptoms. Finally, the parsed information is transformed into a structured medical data set according to a preset medical data standard. This medical data set is a standardized data set with multiple dimensions, including patient basic information, symptom list, sign data, medical history information, and examination indicators.

[0008] Prior to S2, the medical knowledge graph and rule base are constructed. The medical knowledge graph uses diseases as core nodes and associates entities including symptoms, signs, past medical history, examination indicators, treatment plans, and medical guidelines, and establishes multi-dimensional relationships, with each relationship having a weight. The rule base includes disease diagnostic criteria, differential diagnosis rules, and critical value warning rules. The rule base supports regular updates to adapt to the latest medical guidelines.

[0009] In step S2, the medical data set is compared with data in the medical knowledge graph and rule base to generate a set of suspected diagnostic results. This includes: using a large model to call the medical knowledge graph and rule base, matching the medical data set with the association relationships in the knowledge graph, and extracting a set of associated candidate diseases; then, combining the diagnostic rules in the rule base, selecting candidate diseases based on the set of associated candidate diseases, and through multi-step reasoning of the large model, analyzing the fit between the candidate diseases and medical record data to generate a set containing several suspected diagnostic results.

[0010] Prior to S3, a compliance rule base and a confidence scoring system are constructed. The compliance rule base includes general medical diagnostic standards, specialist treatment standards, and regional medical policy requirements. The compliance rule base supports manual updates and automatic synchronization, and the latest standards are imported in real time by connecting to authoritative medical databases. The confidence scoring system includes scoring indicators such as symptom matching completeness, evidence support strength, medical consensus support, and medical history fit. Each indicator is assigned a preset weight, and the confidence score of each suspected diagnosis result is calculated by weighted summation.

[0011] In step S3, compliance verification and confidence scoring are performed on the suspected diagnosis result set. Suspected diagnosis results that do not meet compliance standards or have confidence scores below the threshold are filtered out, and a set of compliant high-confidence diagnoses is output. This includes: comparing and verifying each suspected diagnosis result in the suspected diagnosis result set with the compliance rule base. The verification dimensions include whether the suspected diagnosis result meets the current medical diagnosis standard definition, whether it matches the scope of the specialist diagnosis and treatment standards of the patient's department, whether it meets the diagnostic requirements in the regional medical policy, and whether there is a diagnostic conclusion that contradicts the patient's medical history. For a suspected diagnosis result that fails any of the verifications, it is marked as "compliance abnormal" and removed. Set a confidence threshold, mark suspected diagnostic results with confidence scores below the confidence threshold as low confidence and remove them; retain suspected diagnostic results that pass compliance verification and have confidence scores above the confidence threshold to form the final set of compliant high-confidence diagnoses.

[0012] In step S4, a diagnostic recommendation list is generated based on the compliant high-confidence diagnostic set. According to the association path of the medical knowledge graph, the medical basis source corresponding to each diagnostic recommendation is labeled, including: each diagnostic recommendation in the list includes a disease name, confidence score, diagnostic priority, and treatment suggestions; based on the association path of the knowledge graph and compliance verification, the diagnostic basis and compliance basis are labeled for each diagnostic recommendation. The diagnostic basis includes associated symptoms, medical history, examination indicators, and corresponding knowledge graph associations. The compliance basis includes the medical diagnostic standards, specialist treatment standards, or regional medical policies that the diagnostic results comply with, and the confidence score details are displayed.

[0013] Secondly, the present invention provides a medical auxiliary diagnostic system, comprising: Data understanding module: Used to receive medical record information and generate a structured medical data set through semantic parsing, key information extraction and structured processing; Knowledge Decision Module: Used to compare the medical data set with data in the medical knowledge graph and rule base to generate a set of suspected diagnosis results; The security constraint module is used to perform compliance verification and confidence scoring on the suspected diagnosis result set, filter out suspected diagnosis results that do not meet compliance standards or have a confidence level below the threshold, and output a set of compliant high-confidence diagnoses. Output Explanation Module: Used to generate a list of recommended diagnoses based on the set of compliant high-confidence diagnoses, and to label the source of medical evidence for each recommended diagnosis according to the association path of the medical knowledge graph.

[0014] The medical auxiliary diagnostic method and system of the present invention have the following advantages: (1) Improve the safety and reliability of diagnosis: Through compliance verification and confidence scoring, diagnostic results that do not comply with medical norms or lack sufficient evidence are effectively filtered out, avoiding misleading doctors' decisions, reducing medical risks, and ensuring the safety and reliability of clinical diagnosis. (2) Improve diagnostic efficiency: Through automatic parsing and structured medical record information by a large model, the time doctors spend manually sorting through data is greatly reduced; at the same time, compliant and highly confident diagnostic recommendations are quickly generated, providing doctors with direct reference and shortening the diagnostic thinking cycle, which is especially suitable for busy scenarios such as emergency rooms and outpatient clinics. (3) Improve diagnostic accuracy: By integrating the professional resources of medical knowledge graphs and rule bases, and combining the deep reasoning ability and double screening of the large model, the correlation between medical record data and diseases is fully explored, reducing misdiagnosis and missed diagnosis caused by insufficient experience or information omission, and improving the scientificity and accuracy of diagnosis. (4) Achieve interpretable and compliant traceability in diagnosis: Breaking through the "black box" problem of traditional auxiliary diagnostic systems, it not only clearly marks the source of medical evidence for diagnostic recommendations, but also displays compliance verification results and confidence score details, enhancing doctors' trust in diagnostic results, and facilitating medical quality traceability and supervision. (5) Standardize diagnostic process: Based on standardized medical knowledge graphs, rule bases and compliance rules, it ensures that the diagnostic reasoning and screening process follows unified medical standards, promotes the standardization of clinical diagnosis, and has important auxiliary value, especially for primary healthcare institutions or young doctors. (6) Adapt to complex clinical scenarios: The large model has powerful semantic understanding and reasoning capabilities, and can handle diverse unstructured medical record data; customized compliance rules and adjustable confidence thresholds adapt to the clinical needs of different departments and regions, and have broad clinical applicability. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0016] The invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 This is a flowchart of a medical auxiliary diagnostic method according to Embodiment 1 of the present invention; Figure 2 This is a logical structure block diagram of a medical auxiliary diagnostic system according to Embodiment 2 of the present invention; Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0019] It should be understood that in the description of the embodiments of the present invention, terms such as "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance, nor as indicating or implying order. In the embodiments of the present invention, "multiple" refers to two or more.

[0020] In this embodiment of the invention, "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, B existing alone, or both A and B existing simultaneously. Furthermore, in this document, the character " / " generally indicates that the preceding and following related objects have an "or" relationship. Example

[0021] like Figure 1 As shown, the medical auxiliary diagnostic method provided in this embodiment is an intelligent auxiliary diagnostic method that combines large-scale models and knowledge graphs. It constructs a four-level architecture of "data understanding – knowledge decision-making – security constraints – output interpretation" to achieve intelligent and compliant operation of the entire process from medical record information parsing to diagnostic recommendation and traceability. Leveraging the powerful semantic understanding capabilities of large-scale models, it performs in-depth analysis and structuring of unstructured medical record data. At the knowledge decision-making layer, it integrates knowledge graphs and professional rule bases, generating accurate suspected diagnostic results through large-scale model reasoning. Finally, it presents clear diagnostic recommendations and traceable explanations to doctors, providing scientific and efficient auxiliary support for clinical diagnosis, effectively reducing the risk of misdiagnosis, and improving diagnostic accuracy and standardization.

[0022] This embodiment provides a medical auxiliary diagnostic method, which may include the following steps: S1: Receives medical record information and generates a structured medical data set through semantic parsing, key information extraction, and structured processing.

[0023] It can receive medical record information, including: receiving text medical records manually entered by doctors, receiving electronic medical record documents, receiving voice input medical records and converting them into text, and can receive medical record data in different formats.

[0024] A large model, fine-tuned and trained on a medical corpus, is used to identify key medical entities in medical record data through word segmentation, part-of-speech tagging, and entity recognition. These key medical entities include symptoms, signs, past medical history, and examination results. Then, through semantic role labeling and dependency parsing, the logical relationships between medical entities are mined, including at least the association between medical history and symptoms. Finally, the parsed information is transformed into a structured medical data set according to a pre-defined medical data standard. This medical data set is a standardized data set with multiple dimensions, including basic patient information, symptom list, sign data, medical history information, and examination indicators.

[0025] This step involves data understanding. It receives medical record information (including symptoms, signs, medical history, examination results, etc.) input or uploaded by doctors, and uses the semantic understanding capabilities of large models to perform in-depth analysis, key information extraction, and structured processing to form a standardized medical data set, providing data support for subsequent diagnostic reasoning.

[0026] Medical record information reception: Supports multiple input methods, including doctors manually entering text medical records, uploading electronic medical record documents (such as Word and PDF formats), voice input to text, etc. The system is compatible with medical record data of different formats to meet diverse clinical use scenarios.

[0027] Large-scale model semantic analysis: A large-scale model, fine-tuned and trained on a medical corpus, is used to perform deep semantic analysis on unstructured medical record information. First, natural language processing techniques such as word segmentation, part-of-speech tagging, and entity recognition are used to identify key medical entities in the medical records, including symptoms (such as "persistent cough" and "high fever"), signs (such as "blood pressure 160 / 100 mmHg" and "heart rate 95 bpm"), past medical history (such as "5-year history of hypertension" and "3-year history of diabetes"), and examination results (such as "white blood cell count 12 × 10⁻⁶"). 9 The system then uses semantic role labeling and dependency parsing to uncover logical relationships between entities, such as the symptom association in "coughing for 3 days with chest pain" and the association between medical history and current symptoms in "hypertension for 5 years, now experiencing dizziness". Finally, the parsed information is transformed into structured data according to predefined medical data standards, including a standardized data set of dimensions such as patient basic information, symptom list, physical signs data, medical history information, and examination indicators. For example, for the medical record text "Patient is a 56-year-old male with a 5-year history of hypertension, who has not taken medication regularly, and has experienced persistent dizziness and blurred vision for the past 2 days, with a blood pressure of 180 / 110 mmHg", the structured data generated after parsing is as follows: Gender: Male, Age: 56 years old, Past Medical History: Hypertension (5 years, no regular medication), Current Symptoms: Persistent dizziness (2 days), blurred vision (2 days), Physical Signs: Blood pressure 180 / 110 mmHg.

[0028] S2: Compare the medical data set with data in the medical knowledge graph and rule base to generate a set of suspected diagnosis results.

[0029] Prior to S2, a medical knowledge graph and rule base were constructed. The medical knowledge graph uses diseases as the core nodes and associates entities including symptoms, signs, past medical history, examination indicators, treatment plans, and medical guidelines, and establishes multi-dimensional relationships, with each relationship having a weight. The rule base includes disease diagnostic criteria, differential diagnosis rules, and critical value warning rules. The rule base supports regular updates to adapt to the latest medical guidelines.

[0030] The large model calls upon the medical knowledge graph and rule base to match the medical data set with the related entities in the knowledge graph, extracting a set of related candidate diseases. Then, combined with the diagnostic rules in the rule base, candidate diseases are selected based on the set of related candidate diseases. Through multi-step reasoning of the large model, the fit between the candidate diseases and the medical record data is analyzed, generating a set containing several suspected diagnostic results.

[0031] This step involves knowledge-based decision-making. The large model calls upon a pre-defined medical knowledge graph and a professional rule base to match the structured medical record data with the disease-symptom, disease-sign, disease-medical history, and disease-examination / test results relationships in the knowledge graph. Combined with the diagnostic gold standard and differential diagnosis rules in the rule base, a set of suspected diagnostic results is generated through multi-dimensional reasoning.

[0032] Prior to this step, knowledge resources need to be built. A dedicated medical knowledge graph and rule base need to be constructed. The medical knowledge graph uses diseases as core nodes, linking them to entities such as symptoms, signs, past medical history, examination indicators, treatment plans, and medical guidelines, establishing multi-dimensional relationships. For example, "hypertension" is associated with symptoms such as "dizziness," "headache," and "blurred vision"; with signs such as "elevated blood pressure"; with past medical history such as "obesity history" and "family history of hypertension"; and with examination indicators such as "blood pressure measurement" and "blood lipid test." Each relationship is assigned a weight (based on clinical incidence and correlation strength). The rule base includes disease diagnostic criteria (such as the disease diagnostic conditions in the "Internal Medicine" treatment guidelines), differential diagnosis rules (such as "dizziness accompanied by elevated blood pressure should be considered as a complication of hypertension, excluding cervical spondylosis"), and critical value warning rules (such as "blood pressure ≥180 / 120 mmHg requires vigilance for hypertensive emergencies").

[0033] The large model invokes a medical knowledge graph and rule base to initiate the diagnostic reasoning process. First, it matches the structured medical record data with the relationships in the knowledge graph to extract a set of associated candidate diseases. Then, it combines diagnostic rules from the rule base to filter the candidate diseases. For example, based on the matching combination of "history of hypertension + significantly elevated blood pressure + dizziness and blurred vision", candidate diseases such as "hypertensive emergency" and "hypertensive encephalopathy" are filtered out. Next, through multi-step reasoning by the large model, it analyzes the fit between the candidate diseases and the medical record data, generating a set containing multiple suspected diagnostic results, providing a verification object for the safety constraint layer.

[0034] S3: Perform compliance verification and confidence scoring on the set of suspected diagnostic results, filter out suspected diagnostic results that do not meet compliance standards or have a confidence level below the threshold, and output a set of compliant high-confidence diagnoses.

[0035] Prior to S3, a compliance rule base and a confidence scoring system were constructed. The compliance rule base included general medical diagnostic standards, specialist treatment standards, and regional medical policy requirements. The compliance rule base supported manual updates and automatic synchronization, and the latest standards were imported in real time by connecting to authoritative medical databases. The confidence scoring system included scoring indicators such as symptom matching completeness, evidence support strength, medical consensus support, and medical history fit. Each indicator was assigned a preset weight, and the confidence score of each suspected diagnosis was calculated by weighted summation.

[0036] The suspected diagnoses in the set of suspected diagnoses are compared and verified one by one with the compliance rule base. The verification dimensions include whether the suspected diagnoses meet the current medical diagnosis standard definition of the disease, whether they match the scope of the specialist diagnosis and treatment standards of the patient's department, whether they meet the diagnostic requirements in the regional medical policy, and whether there are any diagnostic conclusions that contradict the patient's medical history. If a suspected diagnoses fails any of the verifications, they are marked as "compliance abnormal" and removed. A confidence threshold is set, and suspected diagnoses with confidence scores below the confidence threshold are marked as low confidence and removed. Suspected diagnoses that pass the compliance verification and have confidence scores above the confidence threshold are retained to form the final set of compliant high-confidence diagnoses.

[0037] This step is a safety constraint and a core safety assurance step, performing dual verification and screening on the suspected diagnostic results output by S2. On the one hand, compliance verification is carried out, comparing the results against medical diagnostic standards, departmental treatment guidelines, and regional medical policies to eliminate those that do not comply with the standards. On the other hand, confidence scoring is performed, calculating confidence scores based on dimensions such as symptom matching, evidence completeness, and support from medical consensus, filtering out results below a preset threshold to ensure the compliance and reliability of the output diagnostic recommendations.

[0038] Prior to this step, compliance verification rules are established. The system has a built-in compliance rule library covering compliance standards in three core dimensions: First, general medical diagnostic standards, including disease diagnostic standards and clinical pathway guidelines issued by the National Health Commission; second, specialty treatment standards, with customized compliance rules for different departments (such as internal medicine, surgery, and emergency medicine), such as priority rules for critical care diagnosis in emergency medicine and diagnostic standards for chronic disease complications in internal medicine; and third, regional medical policy requirements, adapting to medical quality management regulations and diagnostic standards corresponding to medical insurance treatment items in different regions. The compliance rule library supports manual updates and automatic synchronization, and can achieve real-time import of the latest standards by connecting to authoritative medical databases.

[0039] Compliance verification execution: The suspected diagnostic results generated by S2 are compared and verified against the compliance rule base one by one. Verification dimensions include: whether the diagnostic result conforms to the current disease's treatment guidelines, whether it matches the scope of treatment of the patient's department, whether it meets the diagnostic requirements in the regional medical policy, and whether there are any diagnostic conclusions that contradict the patient's medical history. For example, if the patient has no surgical history, complication diagnoses that require surgical history to support them will be removed. For suspected diagnostic results that do not meet any of the compliance standards, the system marks them as "compliance abnormal" and removes them.

[0040] A multi-dimensional confidence scoring system was constructed, with scoring indicators including: symptom completeness (the proportion of suspected disease symptoms matching the patient's symptoms out of the total number of core symptoms of the disease), evidence support strength (the degree to which objective data such as patient signs and examination results support the diagnosis, calculated based on knowledge graph association weights), medical consensus support (the degree of acceptance of the diagnosis in authoritative medical guidelines and clinical studies), and medical history fit (the correlation between the diagnosis and the patient's past medical and treatment history). Each indicator was assigned a preset weight (optimized based on clinical expert opinions and machine learning algorithms), and a confidence score (range 0-100) for each suspected diagnosis was calculated by weighted summation.

[0041] Results filtering mechanism: Set a confidence threshold (default threshold 80 points, which can be customized by doctors according to clinical scenarios, with an adjustment range of 60-90 points). The system marks suspected diagnostic results with confidence scores below the threshold as "low confidence" and removes them; retain diagnostic results that pass compliance verification and have confidence scores above the threshold to form the final set of compliant high-confidence diagnoses, which are then passed to the next step.

[0042] S4: Generate a list of recommended diagnoses based on a set of compliant, high-confidence diagnoses, and label the source of medical evidence for each diagnostic recommendation according to the association path of the medical knowledge graph.

[0043] Each diagnostic recommendation in the list includes the disease name, confidence score, diagnostic priority, and treatment suggestions. Based on the association path of the knowledge graph and compliance verification, each diagnostic recommendation is labeled with the diagnostic basis and compliance basis. The diagnostic basis includes the associated symptoms, medical history, examination indicators and corresponding knowledge graph associations. The compliance basis includes the medical diagnostic standards, specialist treatment standards or regional medical policies that the diagnostic results comply with, and displays the confidence score details.

[0044] This step is for output interpretation. Based on the diagnostic results after S3 filtering, a specific list of diagnostic recommendations is generated. At the same time, according to the association path of the knowledge graph, the medical basis source corresponding to each diagnostic recommendation is clearly marked, including related symptoms / signs, relevant medical guidelines, disease diagnosis and treatment standards, etc., so as to realize the interpretability of the diagnostic results and help doctors understand the diagnostic logic.

[0045] Diagnostic recommendation generation: Based on the set of compliant high-confidence diagnoses filtered by S3, each specific diagnostic recommendation is generated. The recommendation includes the disease name, confidence score, diagnosis priority (ranked based on the matching degree between confidence score and compliance), and brief treatment suggestions (such as "It is recommended to further improve the head CT examination" and "Immediately give antihypertensive drug treatment"). The core diagnostic information is clearly presented, making it convenient for doctors to view quickly.

[0046] Explainable source annotation: Based on the association paths of the knowledge graph and the verification results of S3, a complete explanatory source is provided for each diagnosis. This includes annotation of the diagnostic basis, such as associated symptoms / signs, medical history, examination indicators, and corresponding knowledge graph relationships; and annotation of compliance basis, explaining the specific medical standards, specialty guidelines, or regional policies that the diagnosis conforms to. Detailed confidence scores are also displayed, clearly showing the scores for each indicator. The diagnostic reasoning logic and compliance verification process are visualized, allowing doctors to intuitively understand the reliable source of the diagnostic results.

[0047] This embodiment is based on intelligent assisted diagnosis technology using large models and knowledge graphs. It is applicable to clinical diagnosis scenarios in medical institutions at all levels, including multiple departments such as internal medicine, surgery, and emergency medicine. It aims to assist doctors in integrating medical record information, mining medical correlations, generating diagnostic references, and optimizing the clinical diagnosis process through intelligent means. Example

[0048] like Figure 2 As shown, the medical auxiliary diagnostic system provided in this embodiment is a system corresponding to the medical auxiliary diagnostic method in Embodiment 1, and includes: Data Understanding Module: This module receives medical record information and generates a structured medical data set through semantic parsing, key information extraction, and structured processing.

[0049] This module includes a medical record information receiving unit, which supports multiple input methods such as text input, document upload, and speech-to-text, and is compatible with medical record data of different formats; a semantic parsing unit, which uses a large model fine-tuned from a medical corpus to realize semantic parsing functions such as medical entity recognition and logical relationship mining; and a structured processing unit, which transforms the parsed key medical information into standardized structured data according to preset specifications.

[0050] Knowledge Decision Module: This module compares the medical data set with data in the medical knowledge graph and rule base to generate a set of suspected diagnostic results.

[0051] This module includes a knowledge resource storage unit, which stores a pre-set medical knowledge graph and a professional rule base. The medical knowledge graph contains the relationships and weights between diseases and entities such as symptoms, signs, and medical history, while the rule base contains diagnostic criteria and differential diagnosis rules. The large model reasoning unit calls upon knowledge resources, matches structured data with the knowledge graph, and performs multi-dimensional reasoning in conjunction with the rule base. The suspected diagnosis generation unit generates a set of suspected diagnosis results containing multiple candidate diseases.

[0052] Safety constraint module: used to perform compliance verification and confidence scoring on the suspected diagnosis result set, filter out suspected diagnosis results that do not meet compliance standards or have a confidence level below the threshold, and output a compliant high-confidence diagnosis set.

[0053] This module includes a compliance rule base unit, which stores compliance standards such as general medical diagnostic guidelines, specialist treatment guidelines, and regional medical policies, and supports updates and customization; a compliance verification unit, which compares suspected diagnostic results with the compliance rule base and removes diagnostic results that do not meet the compliance standards; a confidence score unit, which calculates the confidence score of each suspected diagnostic result based on indicators such as symptom matching completeness, evidence support strength, medical consensus support, and medical history fit; and a result filtering unit, which sets a confidence threshold (which can be customized) to remove diagnostic results with confidence scores below the threshold and outputs a set of compliant high-confidence diagnoses.

[0054] Output Explanation Module: Used to generate a list of recommended diagnoses based on a set of compliant, high-confidence diagnoses, and to label the source of medical evidence for each diagnostic recommendation according to the association path of the medical knowledge graph.

[0055] This module includes a diagnostic recommendation generation unit, which outputs the disease name, confidence score, diagnostic priority, and brief treatment suggestions for each diagnostic recommendation based on a set of compliant high-confidence diagnoses; and an interpretable source annotation unit, which annotates the medical basis, compliance basis, and confidence score details corresponding to the diagnostic recommendation based on the verification results of the knowledge graph association path and security constraint module, and displays the diagnostic reasoning logic.

[0056] This embodiment achieves four core objectives through a four-level architecture: First, it accurately parses unstructured medical record information, extracts key medical elements, and transforms them into structured data; second, it integrates medical knowledge graphs and rule bases to generate reliable suspected diagnostic results through large-scale model reasoning; third, it performs compliance verification and confidence scoring on suspected results through security constraints, filtering out results with compliance risks or insufficient confidence, ensuring the safety and reliability of diagnostic recommendations; and fourth, it provides clear diagnostic recommendations and traceable sources of explanation to assist doctors in making accurate decisions, reduce the risk of misdiagnosis, and improve the standardization and safety of clinical diagnosis.

[0057] The medical auxiliary diagnostic method and system according to the present invention have been described above by way of example with reference to the accompanying drawings. However, those skilled in the art should understand that various modifications can be made to the medical auxiliary diagnostic method and system proposed in the present invention without departing from the scope of the invention. Therefore, the scope of protection of the present invention should be determined by the content of the appended claims.

Claims

1. A medical auxiliary diagnostic method, characterized in that, Includes the following steps: S1: Receive medical record information and generate a structured medical data set through semantic parsing, key information extraction and structured processing; S2: Compare the medical data set with the data in the medical knowledge graph and rule base to generate a set of suspected diagnosis results; S3: Perform compliance verification and confidence scoring on the set of suspected diagnostic results, filter out suspected diagnostic results that do not meet compliance standards or have a confidence level below the threshold, and output a set of compliant high-confidence diagnoses; S4: Generate a list of recommended diagnoses based on the compliant high-confidence diagnostic set, and label the source of medical evidence for each diagnostic recommendation according to the association path of the medical knowledge graph.

2. The medical auxiliary diagnostic method according to claim 1, characterized in that, In S1, receiving medical record information includes: receiving text medical records manually entered by doctors, receiving electronic medical record documents, receiving voice-input medical records and converting them into text, and receiving medical record data in different formats.

3. The medical auxiliary diagnostic method according to claim 1, characterized in that, In step S1, a structured medical data set is generated through semantic parsing, key information extraction, and structured processing. This includes: using a large model fine-tuned and trained with a medical corpus, identifying key medical entities in the medical record information through word segmentation, part-of-speech tagging, and entity recognition. These key medical entities include symptoms, signs, past medical history, and examination results. Then, through semantic role labeling and dependency parsing, the logical relationships between these medical entities are mined. These logical relationships include at least the association between medical history and symptoms. Finally, the parsed information is transformed into a structured medical data set according to a preset medical data standard. This medical data set is a standardized data set with multiple dimensions, including patient basic information, symptom list, sign data, medical history information, and examination indicators.

4. The medical auxiliary diagnostic method according to claim 1, characterized in that, Prior to S2, the medical knowledge graph and rule base are constructed. The medical knowledge graph uses diseases as core nodes and associates entities including symptoms, signs, past medical history, examination indicators, treatment plans, and medical guidelines, and establishes multi-dimensional relationships, with each relationship having a weight. The rule base includes disease diagnostic criteria, differential diagnosis rules, and critical value warning rules. The rule base supports regular updates to adapt to the latest medical guidelines.

5. The medical auxiliary diagnostic method according to claim 4, characterized in that, In step S2, the medical data set is compared with data in the medical knowledge graph and rule base to generate a set of suspected diagnostic results. This includes: using a large model to call the medical knowledge graph and rule base, matching the medical data set with the association relationships in the knowledge graph, and extracting a set of associated candidate diseases; then, combining the diagnostic rules in the rule base, selecting candidate diseases based on the set of associated candidate diseases, and through multi-step reasoning of the large model, analyzing the fit between the candidate diseases and medical record data to generate a set containing several suspected diagnostic results.

6. The medical auxiliary diagnostic method according to claim 1, characterized in that, Prior to S3, a compliance rule base and a confidence scoring system are constructed. The compliance rule base includes general medical diagnostic standards, specialist treatment standards, and regional medical policy requirements. The compliance rule base supports manual updates and automatic synchronization, and the latest standards are imported in real time by connecting to authoritative medical databases. The confidence scoring system includes scoring indicators such as symptom matching completeness, evidence support strength, medical consensus support, and medical history fit. Each indicator is assigned a preset weight, and the confidence score of each suspected diagnosis result is calculated by weighted summation.

7. The medical auxiliary diagnostic method according to claim 6, characterized in that, In step S3, compliance verification and confidence scoring are performed on the suspected diagnosis result set. Suspected diagnosis results that do not meet compliance standards or have confidence scores below the threshold are filtered out, and a set of compliant high-confidence diagnoses is output. This includes: comparing and verifying each suspected diagnosis result in the suspected diagnosis result set with the compliance rule base. The verification dimensions include whether the suspected diagnosis result meets the current medical diagnosis standard definition, whether it matches the scope of the specialist diagnosis and treatment standards of the patient's department, whether it meets the diagnostic requirements in the regional medical policy, and whether there is a diagnostic conclusion that contradicts the patient's medical history. For a suspected diagnosis result that fails any of the verifications, it is marked as "compliance abnormal" and removed. Set a confidence threshold, mark suspected diagnostic results with confidence scores below the confidence threshold as low confidence and remove them; retain suspected diagnostic results that pass compliance verification and have confidence scores above the confidence threshold to form the final set of compliant high-confidence diagnoses.

8. The medical auxiliary diagnostic method according to claim 1, characterized in that, In step S4, a diagnostic recommendation list is generated based on the compliant high-confidence diagnostic set. According to the association path of the medical knowledge graph, the medical basis source corresponding to each diagnostic recommendation is labeled, including: each diagnostic recommendation in the list includes a disease name, confidence score, diagnostic priority, and treatment suggestions; based on the association path of the knowledge graph and compliance verification, the diagnostic basis and compliance basis are labeled for each diagnostic recommendation. The diagnostic basis includes associated symptoms, medical history, examination indicators, and corresponding knowledge graph associations. The compliance basis includes the medical diagnostic standards, specialist treatment standards, or regional medical policies that the diagnostic results comply with, and the confidence score details are displayed.

9. A medical auxiliary diagnostic system, characterized in that, include: Data understanding module: Used to receive medical record information and generate a structured medical data set through semantic parsing, key information extraction and structured processing; Knowledge Decision Module: Used to compare the medical data set with data in the medical knowledge graph and rule base to generate a set of suspected diagnosis results; The security constraint module is used to perform compliance verification and confidence scoring on the suspected diagnosis result set, filter out suspected diagnosis results that do not meet compliance standards or have a confidence level below the threshold, and output a set of compliant high-confidence diagnoses. Output Explanation Module: Used to generate a list of recommended diagnoses based on the set of compliant high-confidence diagnoses, and to label the source of medical evidence for each recommended diagnosis according to the association path of the medical knowledge graph.

Images