Traditional Chinese medicine intelligent diagnosis system and method based on multi-source information fusion and knowledge graph

Abstract

The application discloses a traditional Chinese medicine intelligent dialectical system and method based on multi-source information fusion and a knowledge graph, and relates to the technical field of traditional Chinese medicine intelligent auxiliary diagnosis and treatment. The system comprises a multi-source information acquisition module, which is used for acquiring patient tongue image, face image, pulse waveform, inquiry text, voice information, previous medical record information, physical information, regional climate information and reexamination feedback information; and a multi-source feature analysis module, which is used for extracting dialectical features related to symptoms and signs. The system and method can identify original syndrome persistence, syndrome transformation, superimposed complex syndrome, stage fluctuation and syndrome relief by combining the current syndrome element combination of a patient with historical syndrome nodes, disease course stage nodes and syndrome migration edges, and improve the adaptive ability of dynamic dialectical in a continuous reexamination scenario.

Description

Technical Field

[0001] This invention relates to the field of intelligent auxiliary diagnosis and treatment in traditional Chinese medicine, and more specifically, to an intelligent diagnostic system and method for traditional Chinese medicine based on multi-source information fusion and knowledge graph. Background Technology

[0002] Traditional Chinese medicine (TCM) diagnosis requires a comprehensive assessment of tongue appearance, complexion, pulse, medical history, patient history, constitution, and regional climate information to determine the nature of the disease (cold / heat, deficiency / excess, organ location, and disease progression). Existing TCM-assisted diagnostic systems largely rely on single-source tongue appearance recognition, rule-based matching of medical history, or static knowledge base reasoning. While they can perform some symptom structuring, they lack unified constraints on the quality, completeness, temporal stability, and semantic reliability of data from different sources, easily accepting low-reliability information as valid evidence. During follow-up visits, patient symptoms may shift with medication feedback, symptom duration, disease stage, and environmental changes. Existing systems struggle to distinguish between persistent original symptoms, symptom transformation, overlapping concurrent symptoms, periodic fluctuations, and symptom relief, leading to insufficient dynamic diagnostic interpretation. Therefore, a TCM intelligent diagnostic system and method are needed that can evaluate the reliability of multi-source diagnostic information, construct evidence edges for syndrome elements, resolve conflicts, identify disease stages, and reason about symptom migration. Summary of the Invention

[0003] The purpose of this invention is to provide a TCM intelligent diagnostic system and method based on multi-source information fusion and knowledge graph. It solves the problems of existing TCM auxiliary diagnosis and treatment systems that rely on single tongue image recognition, consultation rule matching or static knowledge base reasoning. These systems lack unified constraints on the collection quality, completeness, time stability and semantic reliability of data from different sources, easily accept low-reliability information as valid evidence, and have difficulty distinguishing between the persistence of the original syndrome, syndrome transformation, superposition of concurrent syndromes, stage fluctuations and syndrome relief, resulting in insufficient dynamic diagnostic interpretation and failing to meet the needs of use.

[0004] This invention achieves the above objectives through the following technical solution: a TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph, the system comprising: The module includes a multi-source information acquisition module, a multi-source feature analysis module, a source credibility assessment module, a syndrome element evidence edge construction module, a conflict resolution module, a disease stage identification module, a syndrome migration reasoning module, a syndrome differentiation result output module, and a follow-up visit feedback update module. The multi-source information acquisition module is used to collect patient tongue images, facial color images, pulse waveforms, consultation text, voice information, past medical records, physical constitution information, regional climate information, and follow-up consultation feedback information; The multi-source feature parsing module is used to extract diagnostic features related to symptoms and signs; The source credibility assessment module is used to determine source credibility based on collection quality, completeness, stability, and semantic clarity. The evidence edge construction module is used to establish evidence edges between information sources, symptom and sign nodes and syndrome element nodes in the TCM knowledge graph, and to record source type, source credibility, collection time, symptom duration and evidence strength. The conflict resolution module is used to process contradictory evidence. The disease stage identification module is used to determine the current disease stage; The syndrome migration reasoning module is used to determine the current syndrome state based on the syndrome migration knowledge graph; The diagnostic result output module is used to output the main syndrome, concurrent syndrome, core syndrome elements, migration path and explanatory information; The follow-up visit feedback update module is used to update the evidence edge weights and syndrome migration edge weights.

[0005] Furthermore, the multi-source information acquisition module includes: Image acquisition unit, pulse acquisition unit, text acquisition unit, voice acquisition unit, medical record access unit, constitution input unit, climate access unit, and follow-up visit feedback unit; The image acquisition unit is used to acquire tongue image and facial color image and record the results of illumination, color correction and occlusion detection. The pulse acquisition unit is used to acquire pulse waveforms and record sampling frequency, effective band and noise level; The text acquisition unit is used to acquire consultation texts and previous medical record texts; The voice acquisition unit is used to acquire voice signals and generate speech-to-text transcription. Each unit records the patient identification, collection time, and source type in the corresponding data record.

[0006] Furthermore, the multi-source feature parsing module includes: Tongue analysis unit, facial complexion analysis unit, pulse analysis unit, medical history analysis unit, and voice analysis unit; The tongue image analysis unit is used to extract features such as tongue color, tongue coating thickness, tongue coating color, cracks, teeth marks, and petechiae. The facial color analysis unit is used to extract facial color features such as reddish, pale, sallow, and dull. The pulse analysis unit is used to extract pulse rate, pulse cycle, peak amplitude and waveform stability features; The consultation and analysis unit is used to extract information on cold and heat, sweating, pain, diet, sleep, bowel movements, emotions, and disease course; The speech analysis unit is used to extract features of sound intensity, speech rate, breath continuity, and speech fatigue.

[0007] Furthermore, the source credibility assessment module includes: Quality scoring unit, complete scoring unit, stable scoring unit, semantic scoring unit, and credibility fusion unit; The quality scoring unit generates a collection quality score based on image clarity, waveform noise, proportion of effective text content, and speech-to-text confidence. The complete scoring unit generates a completeness score based on the missing key fields. The stability scoring unit generates a stability score based on the characteristic fluctuations of the same patient at adjacent collection times. The semantic scoring unit generates a semantic clarity score based on keyword coverage, contradictory expression ratio, and invalid description ratio. The credibility fusion unit generates source credibility based on preset weights and marks low-credibility sources according to source credibility thresholds. The preset weights are determined by the verification results of historical labeled cases and the results of expert calibration.

[0008] Furthermore, the evidence element evidence edge construction module includes: Node matching unit, evidence element association unit, and evidence strength generation unit; The node matching unit is used to match the diagnostic features to the symptom and sign nodes based on feature similarity and TCM diagnostic rules. The evidence element association unit is used to establish an evidence chain that points from the information source node to the symptom node and from the symptom node to the evidence element node, based on the association rules between the symptom node and the evidence element node. The evidence strength generation unit is used to generate evidence strength based on source credibility, node matching degree, main symptom weight and symptom duration factor. The main symptom weight is determined by the chief complaint information, the doctor's key information collection information and the main symptom attribute in the diagnostic rules.

[0009] Furthermore, the conflict resolution module includes: The system includes a strength generation unit for evidence elements, a conflict relationship identification unit, and a conflict resolution unit. The evidence element support strength generation unit is used to fuse multiple evidence element edges pointing to the same evidence element node. The conflict relationship identification unit is used to identify opposite relationships, partially inconsistent relationships and concurrent relationships based on the evidence element relationship table, and to generate the conflict degree in combination with the evidence element support strength. When the degree of conflict reaches the conflict threshold, the conflict handling unit performs weakening processing on low-credibility evidence, retention processing on evidence that is consistent with the main symptom and whose duration meets the disease course condition, transformation processing on evidence that has an evolutionary relationship with historical symptoms, and marking processing on evidence that meets the concomitant conditions.

[0010] Furthermore, the disease stage identification module includes: Historical status reading unit, duration evaluation unit, follow-up visit interval evaluation unit, medication feedback evaluation unit, and stage determination unit; The stage determination unit determines whether the patient is currently in the initial stage, development stage, transformation stage, concurrent stage, recovery stage, or recurrence stage based on historical syndrome status, symptom duration, follow-up visit interval, medication feedback, symptom aggravation trend, symptom relief trend, and new symptom information, and uses the current disease stage as a constraint condition for conflict resolution and syndrome migration reasoning.

[0011] Furthermore, the symptom transfer reasoning module includes: The system includes a syndrome migration map storage unit, a current syndrome element combination generation unit, a historical syndrome comparison unit, and a syndrome status determination unit. The syndrome migration map storage unit is used to store symptom and sign nodes, syndrome element nodes, syndrome nodes, disease stage nodes, and syndrome migration edges. The syndrome migration edges record the migration direction, migration conditions, and migration weight. The historical syndrome comparison unit is used to match the current syndrome element combination with historical syndrome nodes; The syndrome status determination unit determines the current syndrome status as persistent original syndrome, syndrome transformation, superimposed syndrome, phased fluctuation, or syndrome relief based on the matching results, the current disease stage, and the migration weight.

[0012] Furthermore, the diagnostic result output module includes: The main symptom determination unit, the concurrent symptom determination unit, the evidence tracing unit, and the explanatory text generation unit; The main syndrome determination unit determines the main syndrome based on the strength of syndrome element support, the result of conflict resolution, and the syndrome status. The concurrent syndrome determination unit determines the concurrent syndrome based on the evidence marked as concurrent conditions. The evidence tracing unit outputs information sources, symptom and sign nodes, syndrome element nodes, source credibility, and collection time corresponding to the syndrome. The explanatory text generation unit generates explanatory text that includes the syndrome migration path, the basis for the disease stage, and the conflict resolution results. The follow-up visit feedback update module incrementally updates the evidence edge weights and syndrome migration edge weights based on changes in symptoms, tongue and pulse, medication feedback, and new consultation information after the follow-up visit.

[0013] A TCM intelligent diagnostic method based on multi-source information fusion and knowledge graph is applied to the aforementioned TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph. The method includes the following steps: The system collects patient tongue images, facial images, pulse waveforms, consultation texts, voice information, past medical records, constitution information, regional climate information, and follow-up visit feedback. It extracts multi-source diagnostic features and establishes relationships between patient identification, collection time, and source type. Source credibility is determined based on collection quality, completeness, stability, and semantic clarity. Diagnostic features are mapped to symptom and sign nodes and syndrome element nodes, constructing syndrome element evidence edges that record source credibility, collection time, symptom duration, and evidence strength. Evidence edges pointing to the same syndrome element node are merged, and syndrome element support strength is calculated. Contradictory or inconsistent syndrome element evidence is weakened, preserved, transformed, or labeled. The current disease stage is identified. The syndrome migration knowledge graph is invoked to determine the current syndrome state. The system generates the main syndrome, concurrent syndromes, core syndrome elements, syndrome migration path, evidence source, and interpretation information. Evidence edge weights and syndrome migration edge weights are updated based on follow-up visit feedback.

[0014] The beneficial effects of this invention are as follows: 1. By simultaneously acquiring tongue images, facial images, pulse waveforms, consultation text, voice information, past medical records, constitution information, regional climate information, and follow-up consultation feedback information through a multi-source information acquisition module, the system can comprehensively reflect the patient's current symptoms and signs, past evolution process, and external environmental influences, avoiding insufficient diagnostic evidence caused by a single data source.

[0015] 2. The source credibility assessment module scores the acquisition quality, completeness, stability, and semantic clarity, and binds the source credibility with the dialectical features, so that low-quality images, noisy pulses, incomplete medical texts, and semantically contradictory information are constrained in subsequent reasoning, thereby improving the reliability of the multi-source information fusion process.

[0016] 3. By constructing the evidence edge module, a traceable evidence chain is established between the information source, symptom and sign nodes and the evidence element nodes. The source type, source credibility, collection time, symptom duration and evidence strength are recorded, so that the formation process of the main syndrome and the concurrent syndrome can be traced and verified by doctors.

[0017] 4. The conflict resolution module weakens, preserves, transforms, and marks contradictory or inconsistent evidence elements, which can distinguish between low-credibility evidence, stable evidence, intermittent interference evidence, and mixed evidence, reducing misjudgments caused by evidence conflicts.

[0018] 5. By combining the disease course stage identification module and the syndrome migration reasoning module, the current syndrome element combination of the patient is combined with the historical syndrome nodes, disease course stage nodes and syndrome migration edges. This enables the identification of original syndrome persistence, syndrome transformation, superposition of concurrent syndromes, stage fluctuations and syndrome relief, thereby improving the adaptability of dynamic syndrome differentiation in continuous follow-up visits.

[0019] 6. The follow-up consultation feedback update module continuously updates the evidence edge weights and syndrome migration edge weights based on changes in symptoms, tongue and pulse, medication feedback, and new consultation information. This enables the system to revise the knowledge graph based on subsequent diagnosis and treatment results, improving its stability and interpretability during long-term use. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a system block diagram of the present invention; Figure 2 This is a flowchart of the source credibility assessment process for the present invention; Figure 3 This is a flowchart illustrating the conflict resolution process of the present invention. Figure 4 This is a flowchart of the overall method of the present invention. Detailed Implementation

[0021] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.

[0022] Example 1: Please see Figures 1-3 This invention provides a TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph. The system can be deployed on local servers in medical institutions, cloud-based diagnostic platforms, doctor workstations, mobile consultation terminals, or in computing devices connected to tongue image acquisition devices, facial color acquisition devices, pulse diagnosis sensors, and voice acquisition devices.

[0023] In this embodiment, the TCM intelligent syndrome differentiation system includes a multi-source information acquisition module, a multi-source feature analysis module, a source credibility assessment module, a syndrome element evidence edge construction module, a conflict resolution module, a disease stage identification module, a syndrome migration reasoning module, a syndrome differentiation result output module, and a follow-up visit feedback update module.

[0024] The multi-source information acquisition module is used to collect images of the patient's tongue, complexion, pulse waveform, consultation text, voice information, past medical records, physical condition, regional climate information, and follow-up consultation feedback. Tongue images are used to reflect the patient's tongue color, thickness of the tongue coating, tongue coating color, cracks, teeth marks, and petechiae; facial images are used to reflect the patient's complexion as reddish, pale, sallow, or dull; pulse waveforms are used to reflect pulse rate, pulse cycle, peak amplitude, and waveform stability; medical history texts are used to reflect information on chills and fever, sweating, pain, diet, sleep, bowel movements, emotions, and disease course; voice information is used to reflect voice intensity, speech rate, breath continuity, and voice fatigue; past medical history information is used to reflect historical symptoms, historical syndrome elements, past medications, follow-up visit intervals, and treatment feedback; constitution information is used to reflect the patient's constitution type, age, gender, underlying medical history, and long-term condition; regional climate information is used to reflect the temperature, humidity, season, and climate bias of the patient's location; follow-up visit feedback information is used to reflect symptom aggravation, symptom relief, symptom recurrence, medication response, and new symptoms.

[0025] In practical implementation, the multi-source information acquisition module includes an image acquisition unit, a pulse acquisition unit, a text acquisition unit, a voice acquisition unit, a medical record access unit, a constitution input unit, a climate access unit, and a follow-up visit feedback unit. The image acquisition unit acquires tongue and facial images and records lighting conditions, color correction results, and occlusion detection results; the pulse acquisition unit acquires pulse waveforms and records sampling frequency, effective bands, and noise levels; the text acquisition unit acquires consultation text and previous medical record text; the voice acquisition unit acquires voice signals and generates speech-to-text transcription; the medical record access unit accesses the patient's historical medical records; the constitution input unit receives the patient's constitution information; the climate access unit accesses the climate information of the patient's location; and the follow-up visit feedback unit acquires changes in symptoms, tongue and pulse patterns, medication feedback, and new consultation information during follow-up visits. Each unit writes the patient identifier, acquisition time, and source type into the corresponding data record for subsequent time-series tracing.

[0026] The multi-source feature analysis module is used to analyze the data acquired by the multi-source information acquisition module to obtain diagnostic features related to symptoms and signs. The multi-source feature analysis module includes a tongue image analysis unit, a facial complexion analysis unit, a pulse image analysis unit, a medical history analysis unit, and a speech analysis unit. The tongue image analysis unit extracts features such as tongue color, tongue coating thickness, tongue coating color, cracks, teeth marks, and petechiae; the facial complexion analysis unit extracts features such as redness, paleness, sallowness, and dullness; the pulse image analysis unit extracts features such as pulse rate, pulse cycle, peak amplitude, and waveform stability; the medical history analysis unit extracts information on chills and fever, sweating, pain, diet, sleep, bowel movements, emotions, and disease course; and the speech analysis unit extracts features such as voice intensity, speech rate, breath continuity, and speech fatigue.

[0027] In practical implementation, the multi-source feature parsing module associates the aforementioned diagnostic features according to patient identification, collection time, and source type, forming a temporal set of diagnostic features for the same patient. For image-based information, the system uses image segmentation, color correction, and feature classification models to extract tongue and facial features; for pulse waveforms, the system uses waveform denoising, period recognition, and peak detection to extract pulse features; for consultation text and speech-to-text, the system uses keyword recognition, semantic encoding, and symptom entity extraction to extract consultation features. This processing enables patient information from different sources to be converted into a unified symptom and sign expression format.

[0028] The source credibility assessment module determines source credibility based on acquisition quality, completeness, stability, and semantic clarity. This module includes a quality scoring unit, a completeness scoring unit, a stability scoring unit, a semantic scoring unit, and a credibility fusion unit. The quality scoring unit generates an acquisition quality score based on image clarity, waveform noise, the proportion of effective text content, and speech-to-text confidence. The completeness scoring unit generates a completeness score based on the presence of missing key fields. The stability scoring unit generates a stability score based on the characteristic fluctuations of the same patient at adjacent acquisition times. The semantic scoring unit generates a semantic clarity score based on keyword coverage, the proportion of contradictory expressions, and the proportion of invalid descriptions. The credibility fusion unit generates source credibility based on preset weights.

[0029] In this embodiment, any information source is denoted as the first. The credibility of information sources is determined as follows:

[0030]

[0031] in, Indicates the first The credibility of the source of the information; Indicates the quality score of the data collection; The score indicates the degree of completeness. Indicates the stability score; The score indicates the degree of semantic clarity. , , , These represent the weights of the corresponding scores. (The above...) , , , Normalized to For tongue and facial images, the acquisition quality score is determined based on image clarity, illumination uniformity, occlusion ratio, and color deviation; for pulse waveforms, the acquisition quality score is determined based on waveform noise, sampling continuity, and the proportion of effective bands; for medical history texts and speech information, the semantic clarity score is determined based on keyword coverage, invalid description ratio, transcription confidence, and contradictory expression ratio; for past medical records and follow-up feedback information, the completeness score is determined based on whether the chief complaint, symptoms, medication, feedback results, and follow-up time are complete.

[0032] The aforementioned weights are determined by the verification results of historical annotated cases and the results of expert calibration; when the image acquisition quality has a significant impact on the diagnostic results, the weights are increased. When medical records or voice information are used as the primary basis, improve... When historical trends become more critical in follow-up consultation scenarios, improving... The credibility fusion unit determines the source credibility threshold. Mark low-trust sources. When Less than When the system marks the source of the information as a low-trust source; when Not less than At that time, the system allows the information source to participate in the construction of evidence elements. The optimal value is determined based on the distinction between low-confidence samples and valid samples in the historical case validation set, with the highest possible value being the one that minimizes both the false acceptance and false exclusion rates. In the absence of historical samples, the initial settings are determined by experts based on the stability of the data collection equipment, the completeness of the medical history taking, and the clinical usage scenario.

[0033] The syndrome element evidence edge construction module is used to establish evidence edges between information sources, symptom and sign nodes, and syndrome element nodes in the TCM knowledge graph, and to record source type, source credibility, collection time, symptom duration, and evidence strength. The syndrome element evidence edge construction module includes a node matching unit, a syndrome element association unit, and an evidence strength generation unit.

[0034] The node matching unit is used to match diagnostic features to symptom and sign nodes based on feature similarity and TCM diagnostic rules. For the node from the... Dialectical features extracted from information sources The node matching unit calculates its relationship with the first node. Symptom and sign nodes Match degree:

[0035]

[0036] in, Indicating dialectical characteristics Symptoms and signs nodes The degree of matching; This indicates the semantic or feature similarity between diagnostic features and symptom / sign nodes; Indicates the result of rule matching; , These represent the fusion weights for similarity matching and rule matching, respectively. For structured features such as tongue appearance, facial color, and pulse, The determination is based on feature vector distance, color interval overlap, waveform similarity, or classification probability; for medical consultation text and speech-to-text text... Determined based on semantic vector similarity or keyword coverage. Rule matching results. The weights are determined based on pre-defined TCM diagnostic rules. and The determination rule is as follows: when the consistency between the rule hit results and expert interpretation results in the historical case sample is high, the rule hit rate is increased. When patients' expressions are colloquial, synonymous, or non-standardized, improve... .

[0037] The evidence element association unit is used to establish an evidence chain that points from information source nodes to symptom / sign nodes and from symptom / sign nodes to evidence element nodes, based on the association rules between symptom / sign nodes and evidence element nodes. The evidence strength generation unit is used to generate the evidence strength based on source credibility, node matching degree, principal symptom weight, and symptom duration factor. For any evidence element edge, its evidence strength is:

[0038] in, Indicates by the first Information sources and symptom / sign nodes Pointing to the evidence node The strength of the evidence; Indicates the credibility of the source; Indicates the degree of node matching; Indicates symptom and sign nodes The weight of the primary symptom; Factors indicating symptom duration.

[0039] The symptom duration factor was determined as follows:

[0040] in, Indicates symptom and sign nodes The duration of the corresponding symptoms; Indicates the current stage of the disease. Reference duration below. Main symptom weight. The weighting is determined by the chief complaint information, the key information collected by the doctor, and the principal symptom attribute in the diagnostic rules. If the symptom / sign node belongs to the chief complaint symptom, the key symptoms collected by the doctor, or the principal symptom in the diagnostic rules, it is assigned a higher weight; if it belongs to an accompanying symptom, a mild symptom, or an occasional symptom, it is assigned a lower weight. (Refer to duration.) It is determined based on the type of disease, stage of disease, and historical case statistics; for acute onset diseases... Set to a shorter time; for chronic, recurrent disease courses, Set to a longer time.

[0041] The conflict resolution module handles contradictory or inconsistent evidence. It includes an evidence support strength generation unit, a conflict relationship identification unit, and a conflict processing unit. The evidence support strength generation unit merges multiple evidence edges pointing to the same evidence node. For the same evidence node... The strength of its supporting evidence is:

[0042] in, Indicates the element node The strength of the evidence supporting the evidence; Indicates pointing to the evidence node The set of evidence elements and edges; Represents a set Middle The strength of evidence for each element of evidence. Using this method, when multiple sources support the same element of evidence, the strength of the element's support increases with the number and strength of the evidence; when only a single weak piece of evidence exists, the strength of the element's support remains at a low level.

[0043] The conflict relationship identification unit is used to identify opposite relationships, partially inconsistent relationships, and conjoint relationships based on the evidence element relationship table, and to generate the conflict degree by combining the evidence element support strength. Two evidence element nodes. and The degree of conflict between them is:

[0044] in, Indicates the element node With evidence node The degree of conflict between them; Indicates the element node With evidence node The coefficient of conflict relationship between them; , These represent the evidence support strength of the two evidence nodes, respectively. Conflict relationship coefficient. Determined by the relationships between syndrome elements in the TCM knowledge graph; when two syndrome elements have an opposite relationship. Take the higher value; when there is a partial inconsistency between the two evidence elements. Take the median value; when two symptom elements can coexist and may constitute a concurrent syndrome. Take the lower value.

[0045] The conflict resolution unit will handle conflicts when the conflict level reaches the conflict threshold. At that time, evidence with low credibility is weakened, evidence consistent with the main symptom and whose duration meets the disease course condition is retained, evidence with an evolutionary relationship with historical symptoms is transformed, and evidence that meets the concurrent conditions is marked. Based on a statistical analysis of historical cases where experts determined there was a conflict of evidence, values ​​were selected that could distinguish between stable, mixed evidence and contradictory, interfering evidence.

[0046] The disease progression stage identification module is used to determine the patient's current disease progression stage. This module includes a historical status reading unit, a duration assessment unit, a follow-up visit interval assessment unit, a medication feedback assessment unit, and a stage determination unit. The historical status reading unit reads the patient's historical symptom status; the duration assessment unit assesses the duration of symptoms; the follow-up visit interval assessment unit assesses the time interval between adjacent visits; the medication feedback assessment unit assesses changes in symptoms after medication; and the stage determination unit, based on historical symptom status, symptom duration, follow-up visit interval, medication feedback, symptom exacerbation trend, symptom relief trend, and new symptom information, determines whether the patient is currently in the initial stage, development stage, transformation stage, concurrent stage, recovery stage, or recurring stage.

[0047] In practice, the stage determination unit calculates the stage score of the patient for each stage of the disease, and determines the stage with the highest stage score as the current stage of the disease:

[0048]

[0049] in, This indicates that the patient belongs to the first category. Stage scoring for each stage of the disease course; Indicates the historical symptom state and the first The degree of matching between different stages of the disease; Indicates the duration of symptoms and the first The degree of matching between different stages of the disease; Indicates medication feedback and the first The degree of matching between different stages of the disease; This indicates worsening symptoms, symptom relief, recurrence of symptoms, or new symptoms related to the first [symptom]. The degree of matching between different stages of the disease; , , , Weighting of disease course stages in the scoring; This indicates the current stage of the disease. For newly diagnosed patients, the weighting of symptom duration and symptom change trends is increased; for returning patients, the weighting of historical symptom status and medication feedback is increased; for patients with chronic diseases, the weighting of historical symptom status and the interval between returning visits is increased. The current stage of the disease serves as a constraint for conflict resolution and symptom transfer reasoning.

[0050] The syndrome migration reasoning module is used to determine the current syndrome state based on the syndrome migration knowledge graph. The module includes a syndrome migration graph storage unit, a current syndrome element combination generation unit, a historical syndrome comparison unit, and a syndrome state determination unit. The syndrome migration graph storage unit stores symptom and sign nodes, syndrome element nodes, syndrome nodes, disease stage nodes, and syndrome migration edges; the syndrome migration edges record the migration direction, migration conditions, and migration weights. The current syndrome element combination generation unit generates the current syndrome element combination based on the candidate syndrome element set; the historical syndrome comparison unit matches the current syndrome element combination with historical syndrome nodes; and the syndrome state determination unit determines the current syndrome state—whether it is a persistent original syndrome, syndrome transformation, superimposed concurrent syndromes, staged fluctuation, or syndrome relief—based on the matching results, the current disease stage, and the migration weights.

[0051] In practice, the system compares the current combination of syndrome elements with the patient's historical syndrome nodes, and calculates the migration score of candidate syndromes based on the current stage of the disease and the syndrome migration edge. Simultaneously, after the patient's follow-up visit, the system updates the evidence edge weights and syndrome migration edge weights based on the follow-up results.

[0052]

[0053] in, This indicates that the patient has historical symptom nodes. Migration to candidate symptom node Transfer score; Indicates candidate syndrome nodes The overall support strength of the corresponding core evidence combination; Indicates the current stage of the disease. Below, historical symptom nodes Migration to candidate symptom node The edge weights of the syndrome migration; This indicates that the follow-up consultation feedback is relevant to the candidate syndrome nodes. The level of support; This represents the adjustment coefficient for follow-up visit feedback; This represents the updated evidence edge weight or symptom migration edge weight; This indicates the edge weight of evidence or the edge weight of symptom migration before the update; This indicates that the follow-up visit has confirmed the results. This indicates the update step size.

[0054] Symptom transition edge weight Initially established based on TCM diagnostic knowledge, clinical pathways, and expert experience, the system is updated statistically during operation based on changes in patient symptoms between adjacent visits. Follow-up visit feedback adjustment coefficient. The update step size is determined based on the completeness and reliability of the follow-up visit feedback. A higher value is used when the follow-up visit feedback data is complete and closely related to the current syndrome diagnosis; a lower value is used when the follow-up visit feedback is missing, the source has low reliability, or conflicts with evidence from other sources. The determination is based on the reliability of the follow-up visit feedback source, the completeness of the follow-up visit sample, and the doctor's confirmation result; when the reliability of the follow-up visit feedback source is high, the follow-up visit sample is complete, and the doctor's confirmation result is clear, the determination is improved. When follow-up consultation feedback is incomplete, semantically unclear, or conflicts with evidence from other sources, the risk is reduced. .

[0055] When the current combination of syndrome elements is highly consistent with the historical syndrome nodes, and the syndrome migration edge does not show obvious transformation conditions, the syndrome state determination unit determines it as the original syndrome continuing; when the current core syndrome element combination has shifted from the syndrome element corresponding to the historical syndrome to the syndrome element corresponding to another candidate syndrome, it is determined as syndrome transformation; when multiple syndrome-corresponding syndrome elements meet the output conditions and do not constitute a strong conflict relationship with each other, it is determined as concurrent syndrome superposition; when the current evidence mainly comes from low-confidence sources or short-term symptom changes, it is determined as phased fluctuation; when the support strength of the main syndrome elements decreases and the follow-up visit feedback shows symptom relief, it is determined as syndrome relief.

[0056] The diagnostic results output module is used to output the primary syndrome, concurrent syndromes, core syndrome elements, migration path, and explanatory information. This module includes a primary syndrome determination unit, a concurrent syndrome determination unit, an evidence tracing unit, and an explanatory text generation unit. The primary syndrome determination unit determines the primary syndrome based on the strength of syndrome element support, conflict resolution results, and syndrome status. The concurrent syndrome determination unit determines concurrent syndromes based on evidence marked as concurrent conditions. The evidence tracing unit outputs the information source, symptom and sign nodes, syndrome element nodes, source credibility, and collection time supporting the corresponding syndrome. The explanatory text generation unit generates explanatory text containing the syndrome migration path, disease stage basis, and conflict resolution results. Through these output methods, doctors can trace the data source, evidence edges, syndrome element combinations, and graph reasoning path upon which each diagnostic result is based.

[0057] The follow-up visit feedback update module is used to update the weights of evidence edges and syndrome transfer edges. During a patient's follow-up visit, this module acquires information on changes in symptoms, tongue and pulse patterns, medication feedback, and new consultation details, and writes this information back to the TCM knowledge graph and the syndrome transfer knowledge graph. If the follow-up visit results show consistency between the previous diagnosis and the patient's subsequent symptom changes and medication feedback, the weight of the corresponding syndrome element evidence edge or syndrome transfer edge is increased. If the follow-up visit results show inconsistency between the previous diagnosis and the patient's subsequent changes, the weight of the corresponding syndrome element evidence edge or syndrome transfer edge is decreased, or it is marked as interfering evidence. Through this approach, the system can avoid excessive impact of a single abnormal feedback on the knowledge graph and can gradually increase the weight of the corresponding evidence edge or syndrome transfer edge when multiple follow-up visits show consistent feedback.

[0058] Example 2: Please see Figure 4 This invention provides a TCM intelligent diagnosis method based on multi-source information fusion and knowledge graph, which is applied to the TCM intelligent diagnosis system in Example 1. The method includes the following steps.

[0059] Collect patient tongue images, facial color images, pulse waveforms, consultation texts, voice information, past medical records, physical constitution information, regional climate information, and follow-up visit feedback information, and write the patient identification, collection time, and source type into the corresponding data record.

[0060] Extracting multi-source diagnostic features. The system extracts features such as tongue color, tongue coating thickness, tongue coating color, cracks, teeth marks, and petechiae from tongue images; features such as redness, paleness, sallowness, and dullness from facial images; features such as pulse rate, pulse cycle, peak amplitude, and waveform stability from pulse waveforms; information on cold and heat, sweating, pain, diet, sleep, bowel movements, emotions, and disease course from medical records; and features such as sound intensity, speech rate, breath continuity, and speech fatigue from speech information. It also establishes associations between patient identification, collection time, and source type.

[0061] The system determines source credibility based on acquisition quality, completeness, stability, and semantic clarity, and marks low-credibility sources according to a source credibility threshold. The system then binds source credibility to corresponding dialectical features, enabling subsequent graph reasoning to distinguish the reliability of different information sources.

[0062] The system maps diagnostic features to symptom and sign nodes and syndrome element nodes, constructing syndrome element evidence edges that record source credibility, collection time, symptom duration, and evidence strength. The system generates evidence strength based on source credibility, node matching degree, principal symptom weight, and symptom duration factors, and merges multiple evidence edges pointing to the same syndrome element node to calculate syndrome element support strength.

[0063] The system weakens, preserves, transforms, or labels contradictory or inconsistent evidence. Evidence of low credibility and inconsistent with high-credibility sources is weakened; evidence consistent with the main symptom and whose duration meets the disease course criteria is preserved; evidence showing an evolutionary relationship with historical symptoms is transformed; and evidence meeting the criteria for concurrent symptoms is labeled.

[0064] The system identifies the patient's current disease stage. Based on historical symptom status, symptom duration, follow-up visit intervals, medication feedback, symptom exacerbation trends, symptom relief trends, and new symptom information, the system determines whether the patient is currently in the initial stage, development stage, transformation stage, concurrent stage, recovery stage, or recurring stage.

[0065] Furthermore, the system invokes the syndrome migration knowledge graph to determine the current syndrome status. The system matches the current combination of syndrome elements with historical syndrome nodes, and combines the current disease stage, syndrome migration edge weights, and the degree of support from follow-up visits to determine whether the current syndrome status is persistent, syndrome transformation, superimposed syndromes, phased fluctuation, or syndrome relief.

[0066] Finally, the system generates primary syndrome, secondary syndrome, core syndrome elements, syndrome migration path, evidence source, and interpretation information, and updates the evidence edge weights and syndrome migration edge weights based on follow-up consultation feedback. Through these methods, the system can establish a traceable evidence chain among multi-source diagnostic information, perform structured processing when evidence conflicts occur, and continuously revise the knowledge graph and syndrome migration knowledge graph in follow-up consultation scenarios, thereby improving the accuracy, stability, and interpretability of TCM intelligent diagnostic results.

[0067] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph, characterized in that, The system includes: The module includes a multi-source information acquisition module, a multi-source feature analysis module, a source credibility assessment module, a syndrome element evidence edge construction module, a conflict resolution module, a disease stage identification module, a syndrome migration reasoning module, a syndrome differentiation result output module, and a follow-up visit feedback update module. The multi-source information acquisition module is used to collect patient tongue images, facial color images, pulse waveforms, consultation text, voice information, past medical records, physical constitution information, regional climate information, and follow-up consultation feedback information; The multi-source feature parsing module is used to extract diagnostic features related to symptoms and signs; The source credibility assessment module is used to determine source credibility based on collection quality, completeness, stability, and semantic clarity. The evidence edge construction module is used to establish evidence edges between information sources, symptom and sign nodes and syndrome element nodes in the TCM knowledge graph, and to record source type, source credibility, collection time, symptom duration and evidence strength. The conflict resolution module is used to process contradictory evidence. The disease stage identification module is used to determine the current disease stage; The syndrome migration reasoning module is used to determine the current syndrome state based on the syndrome migration knowledge graph; The diagnostic result output module is used to output the main syndrome, concurrent syndrome, core syndrome elements, migration path and explanatory information; The follow-up visit feedback update module is used to update the evidence edge weights and syndrome migration edge weights.

2. The TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph as described in claim 1, characterized in that, The multi-source information acquisition module includes: Image acquisition unit, pulse acquisition unit, text acquisition unit, voice acquisition unit, medical record access unit, constitution input unit, climate access unit, and follow-up visit feedback unit; The image acquisition unit is used to acquire tongue image and facial color image and record the results of illumination, color correction and occlusion detection. The pulse acquisition unit is used to acquire pulse waveforms and record sampling frequency, effective band and noise level; The text acquisition unit is used to acquire consultation texts and previous medical record texts; The voice acquisition unit is used to acquire voice signals and generate speech-to-text transcription. Each unit records the patient identification, collection time, and source type in the corresponding data record.

3. The TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph as described in claim 2, characterized in that, The multi-source feature parsing module includes: Tongue analysis unit, facial complexion analysis unit, pulse analysis unit, medical history analysis unit, and voice analysis unit; The tongue image analysis unit is used to extract features such as tongue color, tongue coating thickness, tongue coating color, cracks, teeth marks, and petechiae. The facial color analysis unit is used to extract facial color features such as reddish, pale, sallow, and dull. The pulse analysis unit is used to extract pulse rate, pulse cycle, peak amplitude and waveform stability features; The consultation and analysis unit is used to extract information on cold and heat, sweating, pain, diet, sleep, bowel movements, emotions, and disease course; The speech analysis unit is used to extract features of sound intensity, speech rate, breath continuity, and speech fatigue.

4. The TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph as described in claim 1, characterized in that, The source credibility assessment module includes: Quality scoring unit, complete scoring unit, stable scoring unit, semantic scoring unit, and credibility fusion unit; The quality scoring unit generates a collection quality score based on image clarity, waveform noise, proportion of effective text content, and speech-to-text confidence. The complete scoring unit generates a completeness score based on the missing key fields. The stability scoring unit generates a stability score based on the characteristic fluctuations of the same patient at adjacent collection times. The semantic scoring unit generates a semantic clarity score based on keyword coverage, contradictory expression ratio, and invalid description ratio. The credibility fusion unit generates source credibility based on preset weights and marks low-credibility sources according to source credibility thresholds. The preset weights are determined by the verification results of historical labeled cases and the results of expert calibration.

5. The TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph as described in claim 4, characterized in that, The evidence element evidence edge construction module includes: Node matching unit, evidence element association unit, and evidence strength generation unit; The node matching unit is used to match the diagnostic features to the symptom and sign nodes based on feature similarity and TCM diagnostic rules. The evidence element association unit is used to establish an evidence chain that points from the information source node to the symptom node and from the symptom node to the evidence element node, based on the association rules between the symptom node and the evidence element node. The evidence strength generation unit is used to generate evidence strength based on source credibility, node matching degree, main symptom weight and symptom duration factor. The main symptom weight is determined by the chief complaint information, the doctor's key information collection information and the main symptom attribute in the diagnostic rules.

6. The TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph as described in claim 5, characterized in that, The conflict resolution module includes: The system includes a strength generation unit for evidence elements, a conflict relationship identification unit, and a conflict resolution unit. The evidence element support strength generation unit is used to fuse multiple evidence element edges pointing to the same evidence element node. The conflict relationship identification unit is used to identify opposite relationships, partially inconsistent relationships and concurrent relationships based on the evidence element relationship table, and to generate the conflict degree in combination with the evidence element support strength. When the degree of conflict reaches the conflict threshold, the conflict handling unit performs weakening processing on low-credibility evidence, retention processing on evidence that is consistent with the main symptom and whose duration meets the disease course condition, transformation processing on evidence that has an evolutionary relationship with historical symptoms, and marking processing on evidence that meets the concomitant conditions.

7. The TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph as described in claim 1, characterized in that, The disease stage identification module includes: Historical status reading unit, duration evaluation unit, follow-up visit interval evaluation unit, medication feedback evaluation unit, and stage determination unit; The stage determination unit determines whether the patient is currently in the initial stage, development stage, transformation stage, concurrent stage, recovery stage, or recurrence stage based on historical syndrome status, symptom duration, follow-up visit interval, medication feedback, symptom aggravation trend, symptom relief trend, and new symptom information, and uses the current disease stage as a constraint condition for conflict resolution and syndrome migration reasoning.

8. The TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph as described in claim 7, characterized in that, The symptom transfer reasoning module includes: The system includes a syndrome migration map storage unit, a current syndrome element combination generation unit, a historical syndrome comparison unit, and a syndrome status determination unit. The syndrome migration map storage unit is used to store symptom and sign nodes, syndrome element nodes, syndrome nodes, disease stage nodes, and syndrome migration edges. The syndrome migration edges record the migration direction, migration conditions, and migration weight. The historical syndrome comparison unit is used to match the current syndrome element combination with historical syndrome nodes; The syndrome status determination unit determines the current syndrome status as persistent original syndrome, syndrome transformation, superimposed syndrome, phased fluctuation, or syndrome relief based on the matching results, the current disease stage, and the migration weight.

9. The TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph as described in claim 8, characterized in that, The diagnostic result output module includes: The main symptom determination unit, the concurrent symptom determination unit, the evidence tracing unit, and the explanatory text generation unit; The main syndrome determination unit determines the main syndrome based on the strength of syndrome element support, the result of conflict resolution, and the syndrome status. The concurrent syndrome determination unit determines the concurrent syndrome based on the evidence marked as concurrent conditions. The evidence tracing unit outputs information sources, symptom and sign nodes, syndrome element nodes, source credibility, and collection time corresponding to the syndrome. The explanatory text generation unit generates explanatory text that includes the syndrome migration path, the basis for the disease stage, and the conflict resolution results. The follow-up visit feedback update module incrementally updates the evidence edge weights and syndrome migration edge weights based on changes in symptoms, tongue and pulse, medication feedback, and new consultation information after the follow-up visit.

10. A TCM intelligent diagnostic method based on multi-source information fusion and knowledge graph, characterized in that, The method applied to the TCM intelligent diagnostic system based on multi-source information fusion and knowledge graph as described in any one of claims 1 to 9 includes the following steps: The system collects patient tongue images, facial images, pulse waveforms, consultation texts, voice information, past medical records, constitution information, regional climate information, and follow-up visit feedback. It extracts multi-source diagnostic features and establishes relationships between patient identification, collection time, and source type. Source credibility is determined based on collection quality, completeness, stability, and semantic clarity. Diagnostic features are mapped to symptom and sign nodes and syndrome element nodes, constructing syndrome element evidence edges that record source credibility, collection time, symptom duration, and evidence strength. Evidence edges pointing to the same syndrome element node are merged, and syndrome element support strength is calculated. Contradictory or inconsistent syndrome element evidence is weakened, preserved, transformed, or labeled. The current disease stage is identified. The syndrome migration knowledge graph is invoked to determine the current syndrome state. The system generates the main syndrome, concurrent syndromes, core syndrome elements, syndrome migration path, evidence source, and interpretation information. Evidence edge weights and syndrome migration edge weights are updated based on follow-up visit feedback.

Images