Semantic enhancement auxiliary inquiry system and method based on medical knowledge graph

Abstract

The invention discloses a semantic enhancement auxiliary inquiry system and method based on a medical knowledge graph, and the system comprises a patient information receiving module, a theoretical knowledge graph storage library, an evidence-based knowledge graph storage library, a double-track diagnosis path construction module, a medical knowledge conflict judgment module, and a semantic enhancement report generation module. The double-track diagnosis path construction module inquires the received patient information in a theoretical knowledge graph storage library and an evidence-based knowledge graph storage library which are independent from each other in parallel, and a theoretical diagnosis path and an evidence-based diagnosis path are generated respectively; the medical knowledge conflict judgment module dynamically compares the two paths in real-time interaction, and performs priority judgment according to a preset medical judgment rule; finally, the semantic enhancement report generation module presents the two paths and the conflict judgment result to the user at the same time. Transparency and interpretability of the whole interrogation process are assisted, and reliable semantic enhancement decision support is provided for doctors when the doctors face complex medical knowledge conflicts.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of healthcare informatics, and in particular to a semantic enhancement assisted diagnosis system and method based on a medical knowledge graph. BACKGROUND

[0002] With the penetration of information technology in the medical field, assisted diagnosis systems have become a research hotspot, aiming to use computers to process massive medical knowledge to provide decision support for doctors' diagnosis and treatment.

[0003] Currently, the construction of an assisted diagnosis system usually relies on a medical knowledge graph. However, the existing technology mainly faces the following two technical bottlenecks when constructing and applying these knowledge graphs, which limits their clinical practicability: For example, many assisted diagnosis systems are based on standardized theoretical knowledge graphs. The data sources of such graphs are mainly authoritative medical textbooks, clinical guidelines, or traditional medical classics (for example, the TCM books "Treatise on Cold Damage" and "Synopsis of Golden Chamber"). The defects are: The update speed of theoretical knowledge is much slower than the development of clinical practice. The standard treatment in textbooks may not be the optimal choice for clinical frontline, leading to rigid system recommendations. Theoretical knowledge is a highly generalized summary of disease commonality, which is difficult to cover atypical symptoms or complex complications of patients in the real world, leading to a sharp decline in model accuracy when facing complex cases.

[0004] For another example, another type of system turns to construction based on evidence-based knowledge graphs. The data sources of such graphs are mainly massive, anonymous real-world clinical case data. The system constructs knowledge through data mining and statistical analysis (for example, the statistical correlation between symptom A and disease B is 85%). The defects are: Evidence-based knowledge graphs can only provide statistical correlations, but often lack mechanistic explanations based on pathology or TCM theory. When the system only tells the doctor that symptom A-disease B is related, but cannot explain why it is related, the doctor is difficult to fully trust the decision. Statistical results are highly dependent on the quality and coverage of training data, which may amplify data bias in specific hospitals or specific regions, leading to the conclusion not having universality.

[0005] In addition, if combined with multiple knowledge sources, when theoretical knowledge and evidence-based knowledge conflict, the system does not know how to handle it, for example, in assisted inquiry, the patient inputs a set of symptoms (such as stomach distension, yellow and greasy tongue fur), the theoretical knowledge graph (theoretical knowledge graph repository) based on the theory of classics may deduce it to a functional diagnosis (such as spleen and stomach dampness in traditional Chinese medicine), and the evidence-based knowledge graph (evidence-based knowledge graph repository) based on massive EMR statistics may find that this set of symptoms has a 90% probability of pointing to an organic diagnosis (such as Helicobacter pylori infection), and the two diagnostic paths are completely opposite in the subsequent treatment plan (for example, using heat-clearing and dampness-removing Chinese medicine or using antibiotics).

[0006] The existing technical solutions, including some published knowledge graph fusion technologies (for example, Chinese patent application No. 202511101543.7 discloses a general knowledge graph fusion system architecture), aim to try to combine multiple graphs into a static and unified knowledge base. Such a fusion-for-fusion approach tries to mask or automatically solve conflicts when building, rather than presenting conflicts when applying.

[0007] Therefore, when doctors encounter the above-mentioned real conflicts between theory and evidence in real-time clinical interaction, the existing black box system cannot provide doctors with interpretable conflict analysis and transparent decision-making assistance basis, resulting in doctors not adopting its suggestions, making it difficult for the assisted inquiry system to land in the clinic.

[0008] In summary, there is an urgent need in the art for a semantic enhancement assisted inquiry system and method based on medical knowledge graph. SUMMARY

[0009] The purpose of the present application is to solve the defects in the prior art and propose a semantic enhancement assisted inquiry system and method based on medical knowledge graph.

[0010] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: A semantic enhancement assisted inquiry system based on medical knowledge graph, comprising: A patient information receiving module for receiving patient information input by a user, including at least unstructured chief complaints and structured signs; A theoretical knowledge graph repository for storing theoretical diagnosis paths reflecting theoretical knowledge based on medical classics or clinical guidelines; An evidence-based knowledge graph repository for storing evidence-based diagnosis paths reflecting clinical statistical rules based on massive clinical case data; A dual-track diagnosis path construction module connected with the patient information receiving module, the theoretical knowledge graph repository and the evidence-based knowledge graph repository, for querying the patient information in the theoretical knowledge graph repository and the evidence-based knowledge graph repository in parallel, respectively generating at least one theoretical diagnosis path and one evidence-based diagnosis path matched with the patient information; A medical knowledge conflict adjudication module connected with the dual-track diagnosis path construction module, for dynamically comparing the theoretical diagnosis path and the evidence-based diagnosis path in the real-time interaction of assisted interrogation, identifying consistent nodes and conflict nodes in the two paths, and performing quantitative analysis and priority determination on the conflict nodes according to preset medical adjudication rules; A semantic enhancement report generation module connected with the medical knowledge conflict adjudication module, for generating an interpretable report containing the following contents: Display of the theoretical diagnosis path; Display of the evidence-based diagnosis path; Tips for the conflict nodes and consistent nodes, and the priority determination result output by the medical knowledge conflict adjudication module.

[0011] Further, the dual-track diagnosis path construction module further includes a three-party semantic alignment module, which is used to simultaneously map the unstructured chief complaint to the theoretical terms in the theoretical knowledge graph repository and the clinical manifestations in the evidence-based knowledge graph repository; the system further includes: a manual correction interface connected with the three-party semantic alignment module, for receiving user correction instructions for the mapping result; and a model iterative training unit, for incrementally training the model of the three-party semantic alignment module according to the correction instructions, to improve the accuracy of subsequent alignment.

[0012] Further, the evidence-based knowledge graph repository further includes symptom evolution timing information; The dual-track diagnosis path construction module further adopts a dynamic timing diagram embedding technology, for modeling the symptom evolution timing path in the evidence-based knowledge graph repository, to distinguish the influence of different symptom occurrence sequences on the diagnosis result; The system further includes a disease course development prediction module, for generating at least one future disease course evolution probability path based on the timing path model generated by the dynamic timing diagram embedding technology, in combination with the current patient information, and outputting it in the semantic enhancement report generation module.

[0013] Further, the medical knowledge conflict adjudication module further includes a multi-task provenance learning framework; The framework is used to realize the traceability of the diagnosis by performing the following parallel tasks: Task one: predicting diagnosis results according to the patient information; Task two: path tracing, output the complete theoretical diagnosis path of the diagnosis result in the theoretical knowledge graph repository; Task three: evidence-based tracing, output the complete evidence-based diagnosis path of the diagnosis result in the evidence-based knowledge graph repository; Task four: conflict quantification, calculate the numerical conflict score between the theoretical diagnosis path and the evidence-based diagnosis path.

[0014] Further, the medical knowledge conflict adjudication module preset medical adjudication rules are a rule set for the purpose of presenting conflicts, which makes the system perform the following operations: When the conclusions of the theoretical diagnosis path and the evidence-based diagnosis path are consistent, the semantic enhancement report generation module outputs a high confidence recommendation; When the conclusions of the two are in conflict, and the evidence-based strength of the evidence-based diagnosis path is greater than the preset evidence-based confidence threshold, output the evidence-based preferred recommendation, and prompt the conflict point with the theoretical path; When the conclusions of the two are in conflict, and the evidence-based strength of the evidence-based diagnosis path is less than the evidence-based confidence threshold, output the theoretical routine recommendation, and prompt the lack of strong evidence-based data support.

[0015] Further, the dual-track diagnosis path construction module adopts differentiated path search strategies for the heterogeneous characteristics of the theoretical knowledge graph repository and the evidence-based knowledge graph repository when performing queries: For the theoretical knowledge graph repository, ontology-based reasoning or graph traversal algorithm is adopted to search for theoretical derivation paths containing TCM syndromes, treatment methods, prescriptions and their compatibility relationships; For the evidence-based knowledge graph repository, a weighted graph search algorithm based on statistical correlation is adopted to search for evidence-based paths containing high confidence statistical associations between symptoms, signs, laboratory indicators, disease names, and prescriptions.

[0016] Further, the system further comprises a check suggestion optimization unit; the unit is used to automatically generate the next step check suggestion when the evidence-based priority rule or the theoretical priority rule is triggered; and the check suggestion optimization unit is also used to sort a plurality of possible check suggestions based on the cost-effectiveness data in the evidence-based knowledge graph repository, to preferentially recommend the check item with the highest diagnostic value or the most cost-effective.

[0017] A semantic enhancement assisted inquiry method based on medical knowledge graph, comprising the following steps: Receive, through the patient information receiving module, obtain the patient information input by the user, which at least contains unstructured chief complaints and structured signs; store, pre-store theoretical diagnosis paths in a theoretical knowledge graph repository and evidence-based diagnosis paths in an evidence-based knowledge graph repository; double-track construction, through a double-track diagnosis path construction module, the patient information is queried in the theoretical knowledge graph repository and the evidence-based knowledge graph repository in parallel, and matched theoretical diagnosis paths and evidence-based diagnosis paths are generated respectively; conflict analysis, through a medical knowledge conflict resolution module, the theoretical diagnosis path and the evidence-based diagnosis path are compared in real time, the consistency nodes and the conflict nodes are identified, and the conflict nodes are quantitatively analyzed and priority determined according to a preset medical resolution rule; semantic enhancement generation, through a semantic enhancement report generation module, an auxiliary diagnosis report is generated which presents the theoretical diagnosis path, the evidence-based diagnosis path, and the priority determination results of the conflict and consistency nodes.

[0018] Further, the double-track construction step further comprises: adopting a three-way semantic alignment step, the unstructured chief complaint is mapped to the theoretical terms in the theoretical knowledge graph repository and the clinical manifestations in the evidence-based knowledge graph repository; receiving correction, through a manual correction interface, receiving user correction instructions for the mapping results; iterative training, according to the correction instructions, incrementally training the model of the three-way semantic alignment.

[0019] Further, the medical resolution rule in the conflict analysis step makes the method perform the following operations: when the conclusions of the theoretical diagnosis path and the evidence-based diagnosis path are consistent, a high confidence result is recommended in the semantic enhancement generation step; when the theoretical and evidence-based paths conflict, according to the comparison of the evidence-based strength of the evidence-based diagnosis path and the preset threshold, an evidence-based preferred or theoretical routine result is recommended in the semantic enhancement generation step, and the conflict point is explicitly prompted.

[0020] Compared with the prior art, the beneficial effects of the present application are: By setting a double-track diagnosis path construction module and a semantic enhancement report generation module, the theoretical-based deduction path and the evidence-based statistical path are presented to the doctor in parallel, and through the double-evidence display method, the reasoning process of the auxiliary inquiry is completely transparent, interpretable, and traceable, which enhances the doctor's trust in the system.

[0021] By incorporating a medical knowledge conflict adjudication module, conflicts are acknowledged and proactively presented in real-time interactions. This module quantifies and prioritizes conflicts based on medical adjudication rules (such as evidence strength assessment), providing doctors with semantically enhanced decision-making assistance in complex clinical scenarios (e.g., when theory and clinical data are inconsistent). This addresses the difficulty in reconciling theoretical lag and evidence-based limitations. Furthermore, it not only utilizes static theoretical knowledge but also models the symptom evolution sequence in the evidence-based knowledge graph repository using dynamic temporal graph embedding technology, enabling the system to distinguish key disease progression differences such as fever followed by cough versus cough followed by fever. Simultaneously, by including a disease progression prediction module, it achieves forward-looking prediction of future disease progression, significantly improving the intelligence and clinical applicability of assisted diagnosis.

[0022] By setting up a three-party semantic alignment module, a manual correction interface, and a model iteration training unit, a closed-loop feedback system was constructed. Doctors (experts) can correct the alignment errors of the system for unstructured complaints in real time. The system can learn and optimize the model immediately, thereby ensuring that the system continues to evolve in clinical use. Its understanding of medical terminology, especially the complex semantics in the field of integrated traditional Chinese and Western medicine, will become increasingly accurate. Attached Figure Description

[0023] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0024] Figure 1 A schematic diagram of the overall architecture of a semantically enhanced assisted diagnosis system according to an embodiment of the present invention; Figure 2 : A schematic diagram of the overall process of a semantic enhancement-assisted diagnosis method in an embodiment of the present invention; Figure 3 A detailed flowchart illustrating the workflow of the dual-track diagnostic path construction module in this embodiment of the invention. Figure 4 A detailed flowchart illustrating the workflow of the medical knowledge conflict adjudication module in this embodiment of the invention. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0026] Example 1: Please see Figure 1 This embodiment provides a semantically enhanced assisted diagnosis system based on medical knowledge graphs, aiming to solve the problem of conflict between theoretical and evidence-based knowledge in the field of healthcare informatics when assisting diagnosis.

[0027] like Figure 1 As shown, the system preferably includes the following modules: The patient information receiving module is the interface between the system and the user (usually a doctor or medical professional). It receives patient information input by the user. This patient information is multimodal and includes at least: Unstructured chief complaint: This refers to the patient's verbal description, such as recent heartburn, bloating, and loss of appetite.

[0028] Structured signs: These are standardized information obtained by doctors through professional examinations (inspection, auscultation and olfaction, inquiry and palpation, or physical examination), such as yellow and greasy tongue coating and wiry and rapid pulse selected through the interface.

[0029] The theoretical knowledge graph repository is a pre-built, structured knowledge base. Its data sources primarily come from authoritative medical classics (such as the *Shanghan Lun* and *Jinkui Yaolue* in Traditional Chinese Medicine, or the *Merck Manual of Clinical Practice* and *Harrison's Principles of Internal Medicine* in modern medicine) and industry-recognized clinical guidelines. This repository stores ontological relationships based on these sources, encompassing disease-symptom-treatment connections. The system employs a description logic reasoning mechanism to generate theoretical diagnostic paths based on these ontological relationships. These theoretical diagnostic paths are specifically represented as directed graph structures containing at least symptom nodes, syndrome nodes, diagnostic nodes, and treatment plan nodes. For example, it stores (symptom node "yellow and greasy tongue coating" -> syndrome node "damp-heat in the spleen and stomach" -> diagnostic node "damp-heat accumulation in the spleen syndrome" -> treatment plan node "clearing heat and promoting diuresis").

[0030] Evidence-based knowledge graph repositories are independent knowledge bases that exist in parallel with theoretical knowledge graph repositories. Their data sources are massive, anonymized, real-world clinical case data (Electronic Medical Records, EMR) or clinical trial databases. Evidence-based knowledge graph repositories contain statistical knowledge, reflecting evidence-based diagnostic pathways that reflect correlation and probability. For example, through statistical mining of millions of medical records, it stores data such as (clinical manifestation: yellow and greasy tongue coating, high correlation (85%), clinical diagnosis: Helicobacter pylori infection).

[0031] The dual-track diagnostic path construction module maintains connections with the patient information receiving module, the theoretical knowledge graph repository, and the evidence-based knowledge graph repository. Its core function is parallel processing. Upon receiving patient information, the dual-track diagnostic path construction module does not attempt to merge the theoretical knowledge graph repository and the evidence-based knowledge graph repository. Instead, it simultaneously initiates queries in both independent repositories, generating (at least one) matching theoretical diagnostic path and (at least one) matching evidence-based diagnostic path based on the input patient information.

[0032] The medical knowledge conflict resolution module dynamically compares and presents conflicts during real-time interactive consultations. It receives two (or more) paths generated by the dual-track diagnostic path construction module and executes them. Comparison: Identify identical nodes (i.e., consistent nodes, such as both the theoretical knowledge graph repository and the evidence-based knowledge graph repository considering yellow and greasy tongue coating as a key symptom) and different nodes (i.e. conflicting nodes, such as the theoretical knowledge graph repository pointing to spleen and stomach damp-heat, while the evidence-based knowledge graph repository points to H. pylori infection).

[0033] Decision: Based on pre-defined medical decision rules (see below), perform quantitative analysis on conflicting nodes (e.g., assess the strength of evidence for the evidence-based diagnostic pathway, i.e., the number of cases supporting the pathway) and give priority determination (e.g., determine that the evidence-based diagnostic pathway should be given priority).

[0034] The semantically enhanced report generation module is responsible for presenting the analysis results to doctors in an interpretable way. It receives the analysis results from the medical knowledge conflict adjudication module and generates a semantically enhanced, non-black-box auxiliary diagnostic report. This report presents the conflict without hiding it. (a) Theoretical diagnostic paths derived from the theoretical knowledge graph repository; (b) Evidence-based diagnostic pathways derived from the evidence-based knowledge graph repository; (c) Highlighting the conflict points and consistency nodes between the two, and the priority judgment results output by the adjudication module (e.g., text prompt: Conflict warning: There is a discrepancy between the theoretical path and the evidence-based path here. Evidence-based data (10,000+ cases) suggests prioritizing the investigation of HP infection).

[0035] Example 2: A semantically enhanced diagnostic aid method Please see Figure 2 This embodiment illustrates the assisted consultation method executed by the system based on Embodiment 1. Taking a specific integrated traditional Chinese and Western medicine treatment scenario as an example, its workflow (i.e., method steps) is as follows: First, in step S1, the doctor inputs the following through the patient information receiving module: Unstructured chief complaint: The patient reports recurrent heartburn and bloating for the past month. Structured physical signs: Yellow and greasy tongue coating; Next, in step S2 (see...) Figure 3 The dual-track diagnostic pathway construction module is launched, and the built-in three-party semantic alignment module is launched, which solves the problem of aligning patient speech: mapping heartburn and bloating to theoretical terms in the theoretical knowledge graph repository (e.g., stomach fire rising, spleen and stomach qi stagnation) and clinical representations in the evidence-based knowledge graph repository (e.g., reflux, indigestion). Then, a differentiated path search is performed. In the theoretical knowledge graph repository, ontology-based reasoning (such as SPARQL query or descriptive logic reasoning) is used to search for theoretical paths of stomach fire rising and yellow and greasy tongue coating, and generate a theoretical diagnostic path: spleen and stomach damp heat. In the evidence-based knowledge graph repository, a weighted graph search based on statistical correlation was used to search for statistical pathways of reflux + yellow and greasy tongue coating, which revealed (HP infection, correlation 90%, evidence strength 15,000 cases). An evidence-based diagnostic pathway was generated: Helicobacter pylori (HP) infection; At the same time, the dynamic time-series graph embedding function is activated. If a patient has multiple medical records (e.g., first bloating, then heartburn), the time-series information in the evidence-based knowledge graph repository will be used to model the evolution of this symptom to improve the accuracy of path matching.

[0036] It should be further explained that the third-party semantic alignment module is used to map the unstructured chief complaint to both theoretical terms (including TCM syndromes and Western medicine disease names) in the theoretical knowledge graph repository and clinical representations in the evidence-based knowledge graph repository. The technical implementation includes: Step S201: Multilingual Medical BERT Encoder: Using a BioBERT model pre-trained on traditional Chinese medicine classics, Western medicine textbooks, and clinical medical records, semantic encoding is performed on unstructured chief complaints to obtain vector representations. ; Step S202: Two-branch mapping network, where: Branches of Traditional Chinese Medicine: Input a TCM terminology mapping network and output the similarity distribution with a TCM syndrome dictionary (such as spleen and stomach damp-heat, liver qi stagnation). ; Western medicine branches: Input a Western medicine terminology mapping network and output the similarity distribution with Western medicine disease dictionaries (such as gastroesophageal reflux disease and Helicobacter pylori infection). ; Step S203: Tripartite Consistency Constraint: Introduce a contrastive learning loss function to ensure semantic consistency in the representation of the same chief complaint across the three spaces of Traditional Chinese Medicine, Western Medicine, and Clinical Medicine. in, The loss function for three-way alignment aims to maximize the semantic similarity among the chief complaint, TCM terminology, and Western medicine terminology. The similarity function between two vectors is usually cosine similarity cos(θ) or normalized dot product. By taking the negative logarithm, the product of similarities is transformed into an optimizable loss (the higher the similarity, the smaller the loss), thereby solving the semantic gap between TCM functional diagnosis and Western medicine organic diagnosis.

[0037] It should also be noted that the dual-track diagnostic path construction module further employs dynamic temporal graph embedding technology based on temporal graph attention networks to model the symptom evolution temporal path in the evidence-based knowledge graph repository, specifically including the following steps: Step S211: Sequence of the patient's symptom progression Convert to time series snapshot sequence , where each snapshot Indicates time Symptom relationship diagram; Step S212: Employ a multi-head temporal attention mechanism to assign temporal attention weights to symptom nodes at different times: in: Symptom nodes The embedding representation at time t, For time windows; Step S213, based on temporal attention weights, aggregates historical symptom sequence information to generate a diagnostic path representation that considers the order in which symptoms occur: in, For a moment Attention weights.

[0038] This enables the system to distinguish the impact of different timing patterns, such as fever followed by cough and cough followed by fever, on diagnosis.

[0039] Subsequently, in step S3 (corresponding to) Figure 4 In the process, the medical knowledge conflict adjudication module receives the two paths generated in step S2: the spleen and stomach damp-heat path from the theoretical knowledge graph repository and the HP infection path from the evidence-based knowledge graph repository. This module immediately initiates real-time comparison, first identifying conflicting nodes in the diagnostic conclusions of the two paths. Then, the multi-task source tracing learning framework built into the medical knowledge conflict adjudication module executes its conflict quantification task to calculate a numerical conflict score between the two paths. Simultaneously, the system activates preset medical adjudication rules for judgment: the rules detect that the strength of evidence (15,000 cases) upon which the HP infection path relies is significantly greater than the preset evidence credibility threshold (e.g., 1,000 cases), thus triggering the evidence-based priority rule.

[0040] It should be noted that the conflict quantification task calculates a numerical conflict score between the theoretical diagnostic path and the evidence-based diagnostic path through the following steps: Step S301: Path node matching: A node matching algorithm based on semantic similarity is used to identify corresponding node pairs in two paths; Step S302: Conflict Intensity Calculation: For the identified conflicting node pairs Calculate the conflict score: ; in: The semantic distance function is calculated based on medical ontology (such as UMLS). The strength of evidence is defined as log(number of supporting cases / total number of cases); To ensure theoretical consistency, confidence scores were assigned based on an expert knowledge base. α, β, and γ are adjustable weighting coefficients that satisfy α + β + γ = 1.

[0041] Step S303: Path-level conflict scoring: The scores of all conflicting nodes are weighted and summed to obtain the path-level conflict score. .

[0042] Next, in step S4, based on the analysis and adjudication results of step S3, the semantically enhanced report generation module finally generates an interpretable auxiliary diagnostic report and presents it to the user. This report clearly demonstrates the spleen and stomach damp-heat pathway derived from the theoretical knowledge graph repository, and the *Helicobacter pylori* infection pathway statistically derived from the evidence-based knowledge graph repository (with a note indicating support from 15,000 cases). Furthermore, the report highlights text based on the adjudication results (e.g., "Evidence-Based Preferred Recommendation") and includes a conflict warning: theoretical symptoms coexist with a high-probability evidence-based pathology (*Helicobacter pylori* infection). It suggests prioritizing the investigation of evidence-based pathways. Simultaneously with report generation, the system's built-in examination suggestion optimization unit is activated. This unit, based on the *Helicobacter pylori* infection diagnosis and ranked according to cost-effectiveness data in the evidence-based knowledge graph repository, automatically recommends the next examination item, such as the C13 / C14 breath test (high diagnostic value) and gastroscopy (higher cost).

[0043] Furthermore, this embodiment provides further optimization functions based on steps S2 and S3, corresponding to step S5: If, when reviewing the report, a doctor believes that the semantic alignment from heartburn to stomach inflammation in step S2 is inaccurate and determines that it should be liver fire invading the stomach, the doctor can modify it through the manual correction interface. This operation will be captured by the model iteration training unit, regarded as a high-value expert annotation, and used for incremental training of the model of the three-party semantic alignment module, making the system more intelligent in subsequent interactions. At the same time, based on the dynamic time series model established in step S2, the disease progression prediction module can also provide prospective assistance in parallel. For example, if the patient does not receive intervention this time, the module can extrapolate based on tens of thousands of similar disease progression data in the evidence-based knowledge graph repository and indicate in the report: Warning: Without intervention, 35% of patients may develop atrophic gastritis within 6 months, thus providing data support for the doctor's long-term decision-making.

[0044] To better understand the technical solution of this invention, the diagnostic value of timing modeling is further illustrated below with experimental examples, as detailed in Table 1: Table 1 As can be seen from the above, traditional systems that do not consider the time sequence give a vague diagnosis of "pneumonia (70%)" for both types of pneumonia; the present invention, through dynamic time sequence graph embedding technology, can accurately distinguish between viral and bacterial pneumonia based on the order of symptom appearance, improving the diagnostic accuracy by 15-18%.

[0045] Example 3 This embodiment further provides a complete case of semantic alignment and conflict resolution combining traditional Chinese and Western medicine to illustrate the comprehensive application of the system of the present invention.

[0046] It should be understood that the calculation process shown in this embodiment (e.g., the conflict score calculation in step E303) has been simplified for the purpose of clarity. In actual systems, the conflict intensity calculation formula as detailed in Embodiment 2 of this specification should be used for accurate execution.

[0047] Patient's chief complaint: Recently experiencing irritability, insomnia, rapid heartbeat, and sweaty palms. Step E301: Transform the unstructured chief complaint into a chief complaint (BERT encoding), including: Traditional Chinese medicine interpretation: Irritability and insomnia can be transformed into excessive heart fire (85%) and liver stagnation transforming into fire (72%). Western medicine's findings: Rapid heartbeat and sweaty palms can lead to hyperthyroidism (78%) and anxiety disorder (65%). Clinical manifestations: tachycardia (90%), excessive sweating (85%); Step E302: Construct a dual-track path, including: Theoretical approach (Traditional Chinese Medicine): Irritability and insomnia + tachycardia → excessive heart fire → clearing heart fire (Huanglian Ejiao Decoction); Evidence-based pathway (Western medicine): Tachycardia + Excessive sweating + Insomnia → Hyperthyroidism (supported by 15,000 cases, EvidenceStrength = log10(15,000 / 20,000) ≈ -0.125) → It is recommended to check TSH, T3, and T4. E303: Detection Conflict: Traditional Chinese Medicine's "Heart Fire" vs. Western Medicine's "Hyperthyroidism" Let the weighting coefficients be α=0.3, β=0.5, and γ=0.2. And set example parameter values: (Heart Fire, Hyperthyroidism) = 0.85 (High semantic distance); (Hyperthyroidism) = -0.125 (high evidence strength); (Heart Fire) = 0.72 (high consistency with theory); The ConflictScore formula defined in Example 2 is used for calculation: ConflictScore=α· -β· -γ· =0.3×(0.85)-0.5×(-0.125)-0.2×(0.72); ConflictScore=0.255+0.0625-0.144=0.1735.

[0048] The adjudication rule is as follows: Although there are conflicting scores, the system further examines the absolute evidence-based indicators: if the number of evidence-supported 15,000 is greater than the preset confidence threshold of 1,000 (or if EvidenceStrength-0.125 is detected, which is greater than the threshold-0.5), the system determines that the evidence-based evidence is sufficient and triggers "evidence-based selection".

[0049] Step E304: Generate a semantic enhancement report and diagnostic suggestions, including: Evidence-based recommendation: Hyperthyroidism (supporting evidence: 15,000+ cases); Conflict warning: Traditional Chinese medicine theory suggests "excessive heart fire". It is recommended to combine the results of thyroid function tests to determine whether to use traditional Chinese medicine therapy to clear heart fire. Next steps: TSH / T3 / T4 (high diagnostic value, moderate cost); Disease course prediction: If TSH < 0.5 and T4 > 12, there is an 85% probability of Graves' disease, which requires long-term treatment.

[0050] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A semantic enhancement auxiliary diagnosis system based on a medical knowledge graph, characterized in that, The system comprises: a patient information receiving module for receiving user-inputted patient information containing at least unstructured chief complaints and structured signs; a theoretical knowledge graph repository for storing ontological association relationships including disease-symptom-treatment constructed based on medical classics or clinical guidelines, and generating theoretical diagnosis paths using a description logic reasoning mechanism, wherein the theoretical diagnosis paths comprise at least a directed graph structure of symptom nodes, syndrome nodes, diagnosis nodes and treatment scheme nodes; an evidence-based knowledge graph repository for storing evidence-based diagnosis paths reflecting clinical statistical rules based on massive clinical case data; a dual-track diagnosis path construction module connected with the patient information receiving module, the theoretical knowledge graph repository and the evidence-based knowledge graph repository, for querying the patient information in the theoretical knowledge graph repository and the evidence-based knowledge graph repository in parallel to generate at least one theoretical diagnosis path and one evidence-based diagnosis path matching the patient information respectively; a medical knowledge conflict resolution module connected with the dual-track diagnosis path construction module, for dynamically comparing the theoretical diagnosis path and the evidence-based diagnosis path in real-time interaction of assisted interrogation to identify consistent nodes and conflict nodes in the two paths, and performing quantitative analysis and priority judgment on the conflict nodes according to preset medical resolution rules; a semantic enhancement report generation module connected with the medical knowledge conflict resolution module, for generating an interpretable report containing the following contents: display of the theoretical diagnosis path; display of the evidence-based diagnosis path; prompting for the conflict nodes and consistent nodes, and priority judgment results output by the medical knowledge conflict resolution module.

2. The system of claim 1, wherein: The dual-track diagnosis path construction module further comprises a three-party semantic alignment module for simultaneously mapping the unstructured chief complaints to theoretical terms in the theoretical knowledge graph repository and clinical manifestations in the evidence-based knowledge graph repository, wherein the three-party semantic alignment module realizes the mapping through the following steps: Step S201 uses a BioBERT model pre-trained on Chinese medicine classics, Western medicine textbooks and clinical medical records to perform semantic coding on the unstructured chief complaints to obtain a chief complaint vector; Step S202 provides two branch mapping networks, including a Chinese medicine branch for inputting the chief complaint vector into a Chinese medicine term mapping network to output a Chinese medicine syndrome similarity distribution, and a Western medicine branch for inputting the chief complaint vector into a Western medicine term mapping network to output a Western medicine disease similarity distribution; Step S203 introduces a contrastive learning loss function as a three-party consistency constraint to ensure that the representation of the same chief complaint in the Chinese medicine, Western medicine and clinical three spaces maintains semantic consistency, thereby obtaining the mapping result; the system further comprises: a manual correction interface connected with the three-party semantic alignment module for receiving user correction instructions for the mapping result; and a model iterative training unit for incrementally training the model of the three-party semantic alignment module according to the correction instructions to improve the accuracy of subsequent alignment.

3. The system of claim 1, wherein: The evidence-based knowledge graph repository further includes symptom evolution timing information; the dual-track diagnosis path construction module further adopts a dynamic timing graph embedding technology based on a timing graph attention network for modeling the symptom evolution timing path in the evidence-based knowledge graph repository, and the technology specifically includes: Step S301 converts the symptom evolution sequence of the patient into a timing graph snapshot sequence; Step S302 adopts a multi-head timing attention mechanism to give timing attention weights to symptom nodes at different times; Step S303 aggregates historical symptom sequence information based on the timing attention weights to generate a diagnosis path representation considering the order of symptom occurrence; The system further includes a disease course development prediction module for generating at least one future disease course evolution probability path based on the timing path model generated by the dynamic timing graph embedding technology, combining current patient information, and outputting it in the semantic enhancement report generation module.

4. The system of claim 1, wherein: The medical knowledge conflict adjudication module further includes a multi-task provenance learning framework; The framework is used to realize the explainability provenance of diagnosis by performing the following parallel tasks: Task 1: Predict the diagnosis result according to the patient information; Task 2: Path provenance, output the complete theoretical diagnosis path of the diagnosis result in the theoretical knowledge graph repository; Task 3: Evidence-based provenance, output the complete evidence-based diagnosis path of the diagnosis result in the evidence-based knowledge graph repository; Task 4: Conflict quantification, calculate the numerical conflict score between the theoretical diagnosis path and the evidence-based diagnosis path by the following steps: Step S401 adopts a node matching algorithm based on semantic similarity to identify corresponding node pairs in the two paths; Step S402, for the identified conflict node pairs, calculates the conflict score based on their semantic distance, evidence strength, and theoretical consistency through a preset weighting formula; Step S403, the scores of all conflict nodes are weighted and summed to obtain the path-level conflict score.

5. The system of claim 1, wherein: The preset medical adjudication rules in the medical knowledge conflict adjudication module are a rule set for presenting conflicts, which makes the system perform the following operations: When the conclusions of the theoretical diagnosis path and the evidence-based diagnosis path are consistent, the semantic enhancement report generation module outputs a high-confidence recommendation; When the conclusions of the two paths conflict, and the evidence strength of the evidence-based diagnosis path is greater than the preset evidence credibility threshold, output the evidence-based preferred recommendation, and prompt the conflict point with the theoretical path; When the conclusions of the two paths conflict, and the evidence strength of the evidence-based diagnosis path is less than the evidence credibility threshold, output the theoretical routine recommendation, and prompt the lack of strong evidence data support.

6. The system of claim 1, wherein: The dual-track diagnosis path construction module adopts differentiated path search strategies for the heterogeneous characteristics of the theoretical knowledge graph repository and the evidence-based knowledge graph repository when performing queries: For the theoretical knowledge graph repository, adopt ontology-based reasoning or graph traversal algorithm to search the theoretical derivation path containing TCM syndromes, treatment methods, prescriptions and their compatibility relationships; For the evidence-based knowledge graph repository, a weighted graph search algorithm based on statistical correlation is used to search for evidence-based paths containing high-confidence statistical correlations between symptoms, signs, laboratory indicators, disease names, and prescriptions.

7. The system of claim 5, wherein: The system further includes an examination recommendation optimization unit for automatically generating a next-step examination recommendation when an evidence-based priority rule or a theory-based priority rule is triggered; and the examination recommendation optimization unit is also used to sort a plurality of possible examination recommendations based on cost-effectiveness data in the evidence-based knowledge graph repository to preferentially recommend an examination item with the highest diagnostic value or the best cost-effectiveness. 8.A semantic enhancement assisted inquiry method based on a medical knowledge graph, applied to the system of claim 1, characterized in that, The method comprises the following steps: Receiving, through a patient information receiving module, obtaining patient information input by a user, the patient information at least including unstructured chief complaints and structured signs; Storing, pre-storing theoretical diagnosis paths in a theoretical knowledge graph repository and storing evidence-based diagnosis paths in an evidence-based knowledge graph repository; Dual-track construction, through a dual-track diagnosis path construction module, querying the patient information in the theoretical knowledge graph repository and the evidence-based knowledge graph repository in parallel to respectively generate matched theoretical diagnosis paths and evidence-based diagnosis paths; Conflict analysis, which further comprises: Step S801 node alignment: using a medical ontology-based semantic similarity algorithm, each pair of nodes in the theoretical diagnosis path and the evidence-based diagnosis path is matched to identify node pairs with a semantic similarity greater than a preset threshold as aligned nodes; Step S802 conflict detection: for each aligned node pair, detecting whether the successor nodes are consistent, if the semantic similarity of the successor nodes is less than a preset threshold, marking the node pair as a conflictive node pair; Step S803 conflict quantification and determination: for each conflictive node pair, calculating a conflict score and determining a priority of the conflictive node pair according to a preset medical decision rule; Semantic enhancement generation, through a semantic enhancement report generation module, generating an auxiliary diagnosis report presenting the theoretical diagnosis path, the evidence-based diagnosis path, and the priority determination results of the conflictive and consistent nodes.

9. The method of claim 8, wherein: The dual-track construction step further comprises: Using a three-way semantic alignment step, the unstructured chief complaint is simultaneously mapped to a theoretical term in the theoretical knowledge graph repository and a clinical representation in the evidence-based knowledge graph repository; Receiving correction, through a manual correction interface, receiving a correction instruction of the user on the mapping result; Iterative training, according to the correction instruction, incrementally training the model of the three-way semantic alignment.

10. The method of claim 8, wherein: The medical decision rule in the conflict analysis step causes the method to perform the following operations: When the conclusions of the theoretical diagnosis path and the evidence-based diagnosis path are consistent, a high-confidence result is recommended in the semantic enhancement generation step; When the theoretical and evidence-based paths conflict, according to a comparison of the evidence strength of the evidence-based diagnosis path with a preset threshold, an evidence-based preferred or theoretical routine result is recommended in the semantic enhancement generation step, and the conflict point is explicitly prompted.

Images