Method and system for constructing traditional chinese medicine dermatosis external treatment knowledge graph based on local skin lesions

Abstract

The present application belongs to the field of traditional Chinese medicine informatics and artificial intelligence technology, and relates to a method and system for constructing a knowledge graph of traditional Chinese medicine dermatosis external treatment based on local skin lesions. The method includes defining the ontology of traditional Chinese medicine dermatosis diagnosis and treatment and its association, including lesion area, diagnosis features, syndrome type, external treatment entity, contraindications and review items; self-adaptive segmentation of skin images to obtain local lesion area, extraction of visual representation and physical characteristics to generate lesion area evidence unit; constructing a diagnosis feature discrimination matrix for feature matching to generate corresponding candidate syndrome type and diagnosis evidence combination; constructing external treatment safety adaptation rules to input candidate syndrome type and evidence unit to generate safety adaptation evaluation results of external treatment entity; constructing a diagnosis and treatment semantic association chain to write into a knowledge graph database, and generating review items queued for inspection according to the abnormal triggering conditions of matching and evaluation. The medical quality control level of traditional Chinese medicine auxiliary diagnosis and treatment is comprehensively improved.

Description

Technical Field

[0001] This invention belongs to the field of traditional Chinese medicine informatics and artificial intelligence technology, and relates to a method and system for constructing a knowledge graph of external treatment of skin diseases in traditional Chinese medicine based on local skin lesions. Background Technology

[0002] Traditional Chinese medicine (TCM) dermatology relies heavily on the differentiation of syndromes based on local characteristics of skin lesions, such as scaling, exudation, and ulceration. Furthermore, external treatments are strictly constrained by the location of the lesion and the integrity of the skin barrier. While TCM knowledge graphs are widely used, existing solutions often employ coarse-grained classification or entity recognition based on entire medical records, images, or medical entities, making it difficult to separately and structurally represent the multidimensional local features of different lesion areas within the same image. Moreover, existing dual-path reasoning or priority determination schemes often revolve around trade-offs between theoretical and evidence-based approaches, failing to directly address the crucial verification issue of "safe compatibility between specific local lesion states and external treatment procedures," which is deeply dependent on TCM external treatments for dermatology. This results in current technologies being unable to achieve high-precision, safe, closed-loop verification when faced with conflicts in the differentiation of multiple lesions or the triggering of contraindications for external treatment. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned problems in existing technologies by proposing a method for constructing a knowledge graph of external treatments for skin diseases in Traditional Chinese Medicine based on local skin lesions.

[0004] The objective of this invention can be achieved through the following technical solution: a method for constructing a knowledge graph of external treatments for skin diseases in traditional Chinese medicine based on local skin lesions, comprising: Determine the ontology types for TCM dermatology diagnosis and treatment, configure ontology attributes, and define the relationships between the ontology types, wherein the ontology types include lesion areas, diagnostic features, syndrome types, external treatment entities, contraindications, and review items; Skin images are subjected to adaptive lesion segmentation to divide at least one local lesion region. Visual representation information and skin physical features of each local lesion region are extracted to generate corresponding lesion region evidence units. Construct a diagnostic feature discrimination matrix for candidate syndrome configuration, perform feature matching on the evidence units of the lesion area based on the diagnostic feature discrimination matrix, and generate candidate syndromes and diagnostic evidence combinations corresponding to the local lesion area; Construct external treatment safety adaptation rules, input the candidate syndrome type and the lesion area evidence unit into the external treatment safety adaptation rules, and generate the safety adaptation evaluation result of the external treatment method entity corresponding to the local lesion area; Based on the lesion area evidence unit, the visual diagnosis evidence combination, the candidate syndrome, the safety adaptation assessment result, and the review item, a diagnostic semantic association chain is constructed and written into the knowledge graph database. Corresponding review items are generated according to the preset abnormal triggering conditions in the feature matching and adaptation assessment process and are queued for inspection.

[0005] As an optional embodiment of the present invention, determining the ontological type of traditional Chinese medicine dermatology diagnosis and treatment includes: Set the lesion region as a first ontology type independent of the skin image; Set the contraindication to a restrictive second ontology type independent of the external treatment entity; Configure the review item as a traceable third ontology type that records dialectical conflicts, restrictive exclusion feature triggers, or external treatment adaptation anomalies.

[0006] As an optional embodiment of the present invention, adaptive lesion segmentation is performed on the skin image to divide it into at least one local lesion region, including: Detect edge boundaries, color abrupt changes, and texture differences in the skin image to determine the area of ​​the lesion to be examined; The lesion area to be examined is masked, and the geometric contour parameters and area distribution characteristics of the area are calculated. If the skin image contains multiple local lesion areas, a unique lesion area identifier is assigned to each local lesion area, and the lesion area identifier is written into the corresponding lesion area evidence unit.

[0007] As an optional embodiment of the present invention, the visual characterization information includes the color characteristics, geometric distribution characteristics, and surface state characteristics of the lesion area, which characterize the deformation of the surface material or the exudation of secretions. The physical characteristics of the skin include the physiological structural location where the lesion occurs and the state of skin barrier integrity.

[0008] As an optional embodiment of the present invention, a diagnostic feature discrimination matrix for candidate syndrome configuration is constructed, and feature matching is performed on the evidence units of the lesion area based on the diagnostic feature discrimination matrix to generate candidate syndromes and diagnostic evidence combinations corresponding to the local lesion area, including: For each of the candidate certificate types, a primary certificate feature is configured as an admission criterion, a supporting evidence feature is configured as an auxiliary reference, and an exclusion feature is configured as an exclusion criterion. The visual representation information is compared and matched with the main evidence feature, the supporting evidence feature, and the exclusion feature respectively. In response to the admission criteria that meet the main evidence characteristics and the exclusion characteristics are not triggered, the corresponding candidate evidence type is determined, and the matching main evidence characteristics and supporting evidence characteristics are written into the visual diagnosis evidence combination.

[0009] As an optional implementation of the present invention, the external treatment safety adaptation rules are constructed, including: Establish multidimensional adaptation rules between the external treatment entity and the applicable lesion surface state characteristics, skin barrier integrity state, and physiological structure parts, and set the corresponding range of physicochemical operation parameters for each external treatment entity; The range of physicochemical operation parameters includes at least one of the following: drug dosage form, mode of action, duration of single treatment, treatment frequency, and standard chemotherapy cycle.

[0010] As an optional embodiment of the present invention, generating a safety adaptation assessment result for the external treatment entity corresponding to the local lesion area includes: In response to the fact that the surface condition of the lesion and the skin barrier integrity of the local lesion area meet the preset adaptation criteria of the corresponding external treatment entity, an applicable label is configured for the external treatment entity; In response to the preset safety constraints that match the physiological structural location, the surface condition of the lesion, or the skin barrier integrity state to the corresponding external treatment entity, a caution label is configured for the external treatment entity; In response to the skin barrier integrity status of the local lesion area being characterized as a damaged state, the lesion surface state characteristics conforming to preset infection characteristics, or the lesion area evidence unit matching the restrictive treatment item corresponding to the contraindication, a disabling label is configured for the external treatment entity.

[0011] As an optional embodiment of the present invention, a diagnostic semantic association chain is constructed and written into a knowledge graph database, and review items are generated and queued for inspection based on preset abnormality triggering conditions in the feature matching and adaptation evaluation process, including: Taking the lesion area evidence unit as the association starting point, directed semantic association edges pointing to the visual diagnosis evidence combination, the candidate syndrome type, the safety adaptation assessment result and the review item are generated sequentially, and the mapping association between the lesion area identifier assigned to each of the local lesion areas and the skin image is written into each of the directed semantic association edges; In response to the configuration of the main evidence feature not meeting the corresponding admission judgment condition, triggering the configuration of the exclusion feature, or the safety adaptation assessment result of the external treatment entity being either a caution label or a disable label, a corresponding review item is generated and enters the review queue for inspection.

[0012] This invention also proposes a knowledge graph construction system for external treatment of skin diseases in traditional Chinese medicine based on local skin lesions, including: The first unit is used to determine the ontology type for diagnosis and treatment of skin diseases in traditional Chinese medicine, configure ontology attributes and define the relationship between the ontology types, wherein the ontology type includes lesion area, observation characteristics, syndrome type, external treatment entity, contraindication item and review item; The second unit is used to perform adaptive lesion segmentation on the skin image to divide at least one local lesion region, extract the visual representation information and skin physical features of each local lesion region, and generate a corresponding lesion region evidence unit. The third unit is used to construct a diagnostic feature discrimination matrix for candidate syndrome configuration, and to perform feature matching on the evidence unit of the lesion area based on the diagnostic feature discrimination matrix to generate candidate syndrome and diagnostic evidence combination corresponding to the local lesion area. The fourth unit is used to construct the external treatment safety adaptation rule. The candidate syndrome and the lesion area evidence unit are input into the external treatment safety adaptation rule to generate the safety adaptation evaluation result of the external treatment method entity corresponding to the local lesion area. The fifth unit is used to construct a diagnostic semantic association chain based on the lesion area evidence unit, the visual diagnosis evidence combination, the candidate syndrome, the safety adaptation assessment result, and the review item, and to write it into the knowledge graph database. It also generates corresponding review items according to the preset abnormal triggering conditions in the feature matching and adaptation assessment process and queues them for inspection.

[0013] The present invention also provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to implement the aforementioned method for constructing a knowledge graph of external treatments for TCM skin diseases based on local skin lesions when executing executable instructions.

[0014] Compared with the prior art, the present invention has the following beneficial technical effects: (1) By treating “lesion area” as a separate ontology type, adaptive segmentation and physical and visual feature extraction are performed to transform multiple lesions in the same skin image into independent evidence units, so that the differential lesions in different parts of the same patient can be accurately mapped, avoiding feature ambiguity caused by full image averaging, and significantly improving the source expression accuracy of knowledge graph data.

[0015] (2) By constructing a feature discrimination matrix for visual diagnosis, the extracted visual representation information and skin physical characteristics are directly matched with the TCM diagnostic standard of "main syndrome-supporting evidence-exclusion". This changes the traditional diagnostic model that relies entirely on the expert's subjective experience, and generates a standardized and structured candidate syndrome type and visual diagnosis evidence combination, providing a clear and intuitive derivation path for clinical decision-making.

[0016] (3) By constructing external treatment safety adaptation rules, candidate syndrome types are cross-linked with dimensions such as the integrity of the physical barrier, physiological location, and surface condition of the lesion. Through multi-dimensional assessment, classification adaptation labels of applicable, cautious, and prohibited use are generated, thereby eliminating clinical safety risks caused by inappropriate drug dosage forms and modes of action from a mechanistic perspective.

[0017] (4) A directed semantic association chain starting from the evidence unit of the lesion area is constructed, and the mapping association between the lesion area identifier and the skin image is written, so that the knowledge graph has the ability to trace the source of diagnosis and treatment. At the same time, based on the preset abnormal conditions such as insufficient main evidence, triggering of exclusion items, or generation of caution and prohibition tags, review items are automatically generated and queued for inspection, thus constructing a highly sensitive human-machine collaborative dual inspection mechanism, ensuring that abnormal diagnosis and treatment decisions are traceable and improving the medical quality control level of TCM auxiliary diagnosis and treatment. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for constructing a knowledge graph of external treatment for TCM skin diseases based on local skin lesions according to an embodiment of the present invention; Figure 2 This is a block diagram of a knowledge graph construction system for external treatment of skin diseases in traditional Chinese medicine based on local skin lesions, according to an embodiment of the present invention. Detailed Implementation

[0019] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings to further illustrate the technical solutions of the present invention. However, the present invention is not limited to these embodiments.

[0020] Example 1 Based on the technical problems highlighted in the background, this embodiment proposes a method for constructing a knowledge graph of external treatments for skin diseases in Traditional Chinese Medicine (TCM) based on local skin lesions, such as... Figure 1 As shown, it includes: S1, determine the ontology type for TCM skin disease diagnosis and treatment, configure ontology attributes and define the association between each ontology type, wherein the ontology type includes lesion area, observation characteristics, syndrome type, external treatment entity, contraindications and review items; S2, perform adaptive lesion segmentation on the skin image to divide at least one local lesion region, extract the visual representation information and skin physical features of each local lesion region, and generate a corresponding lesion region evidence unit; S3, construct a diagnostic feature discrimination matrix for candidate syndrome configuration, perform feature matching on the evidence units of the lesion area based on the diagnostic feature discrimination matrix, and generate candidate syndromes and diagnostic evidence combinations corresponding to the local lesion area; S4, construct external treatment safety adaptation rules, input the candidate syndrome type and the lesion area evidence unit into the external treatment safety adaptation rules, and generate the safety adaptation evaluation result of the external treatment method entity corresponding to the local lesion area; S5. Based on the lesion area evidence unit, the visual diagnosis evidence combination, the candidate syndrome, the safety adaptation assessment result, and the review item, a diagnostic semantic association chain is constructed and written into the knowledge graph database. The corresponding review item is generated according to the preset abnormal triggering conditions in the feature matching and adaptation assessment process and queued for inspection.

[0021] The ontology types in TCM dermatology diagnosis and treatment serve as the top-level classification of core elements in the knowledge graph, forming the underlying logical system of the knowledge graph. This embodiment defines six ontology types: lesion region, diagnostic features, syndrome differentiation, external treatment entity, contraindications, and verification items. After determining the ontology types, unique ontology attributes are configured for each type, establishing the necessary foundation for structured graph data and precise retrieval. Different types of ontology focus on different information dimensions in medicine. For example, "lesion region" primarily carries physical and visual information about skin lesions, and its unique attributes need to include original image identification, lesion region identification, collection site, surface condition, and skin integrity status. "External treatment entity" mainly represents treatment intervention methods, and its unique attributes need to include physicochemical operation parameters such as dosage form, mode of action, duration of action, and frequency of administration. The necessity of configuring unique ontology attributes lies in the fact that mixing or failing to distinguish attributes will lead to information confusion between entity nodes, making it impossible to accurately map real clinical data. On the other hand, the design of the exclusive ontology attributes enables knowledge extraction to follow the map, avoids redundant useless fields, and greatly improves the information indexing efficiency of the graph when assisting decision-making.

[0022] The relationships between different ontology types also need to be defined to clarify the logical link from skin lesions to treatment and fill the top-level framework of the knowledge graph. For example, the "lesion area" is set to be associated with "inspection features", "inspection features" are used to determine "syndrome type", "lesion area" combined with "syndrome type" is mapped to "external treatment entity" and is subject to "contraindications", and "abnormal fitting results" ultimately flow to "review items".

[0023] To obtain objective local data to enrich the data source of the knowledge graph, this embodiment performs adaptive lesion segmentation on the original skin image. In actual clinical practice, patients with skin diseases may simultaneously present with multiple different skin lesions on different body parts (for example, the same patient's arm may show dry, thick scales, while the folded area may show erythematous erosion and exudation). If the entire image is directly treated as a whole for global feature extraction, the color, moisture, or skin lesion state of different lesion sites will interfere with each other, resulting in incorrect diagnostic results. To solve this problem, this embodiment performs adaptive segmentation on the image, subdividing it into local lesion regions, and assigning a unique lesion region identifier to each region (such as lesion ID-R001, R002). Visual representation information (such as shape, color, presence of exudation or scaling, etc.) and skin physical characteristics (such as whether the location is on the face or limbs, and whether the local skin integrity is damaged, etc.) are extracted from each local lesion region to form corresponding lesion region evidence units. This ensures that each local lesion area has independent and accurate medical description data, enabling the underlying data of the knowledge graph to have local traceability.

[0024] Traditional diagnostic methods rely heavily on doctors' subjective experience, making direct integration with algorithms difficult. This embodiment constructs a diagnostic feature discrimination matrix to transform the extracted visual representation information and skin physical characteristics into the judgment logic for TCM clinical diagnosis. The diagnostic feature discrimination matrix configures different feature discrimination dimensions for each candidate syndrome: "primary syndrome features" to determine the syndrome threshold, "supporting evidence features" to provide auxiliary decision support, and "exclusion features" that act as a veto. During matching, the corresponding visual representation information from the evidence unit of the lesion area is substituted into the discrimination matrix. For example, if the skin lesion surface condition is "silver-white scales, dry skin," it meets the primary syndrome threshold of "blood deficiency and wind-dryness syndrome" and does not trigger the exclusion feature "obvious exudation." In this case, the matrix output determines that the candidate syndrome for the lesion area is blood deficiency and wind-dryness syndrome, and compiles the features matching the primary and supporting evidence into a diagnostic evidence combination for the lesion area. This process transforms TCM diagnostic experience into structured and computable matrix rules, improving the objectivity of auxiliary diagnosis.

[0025] Furthermore, to prevent mistreatment or worsening of skin lesions, this embodiment also constructs a safety adaptation rule for external treatments based on the requirement to strictly adhere to the safe physical boundaries of the skin lesion site when applying traditional Chinese medicine external therapies in clinical practice. Traditional knowledge graphs often only focus on "treatment plans corresponding to syndrome types," that is, matching the syndrome with the medication, easily overlooking the safety issues of topical preparations when in direct contact with the skin. The safety adaptation rule for external treatments deeply correlates syndrome types, external treatment entities, and evidence units of the lesion area (such as the state of skin integrity, specific physiological structure, and other physical limitations). For example, although a certain local lesion area has been determined to meet the "blood deficiency and wind dryness syndrome," its physical characteristics show that the skin integrity at that location is in a "damaged" state and it is located in the "face" physiological area. The external treatment safety adaptation rule, through logical verification, determines that if a high-strength exfoliating ointment is used on a particular area, it may cause stinging or infection risks due to barrier damage and site sensitivity. Therefore, the safety adaptation assessment result of this external treatment entity is automatically marked with a "prohibited" or "use with caution" label, and a gentler, more suitable solution or redness-reducing wet compress is recommended instead. This external treatment safety adaptation rule effectively intercepts unsafe external treatment recommendations.

[0026] The data elements obtained above are logically reorganized to achieve closed-loop management of clinical risks. Specifically, based on lesion area evidence units, visual examination evidence combinations, candidate syndrome types, safety adaptation assessment results, and review items, a diagnostic semantic association chain containing entities and relationships at all levels is constructed. Simultaneously, during the computational reasoning process, preset abnormal triggering conditions are monitored in real time. If abnormal situations occur, such as "the proportion of matching main syndrome features does not reach the preset threshold, making it impossible to determine the candidate syndrome type," or "the exclusion feature is triggered, leading to a diagnostic conflict," or "the safety adaptation assessment result generates a caution label or a disabled label," a corresponding review item will be generated and automatically written into the pending review queue, awaiting manual review by the physician. By using the "review item" as a downstream node of the association chain, illogical or potentially risky outputs can be intercepted and reported in a timely manner, ensuring a safe closed loop between automated knowledge construction and actual diagnostic and treatment work.

[0027] Preferably, the ontological type of TCM dermatology diagnosis and treatment is determined, including: Set the lesion region as a first ontology type independent of the skin image; Set the contraindication to a restrictive second ontology type independent of the external treatment entity; Configure the review item as a traceable third ontology type that records dialectical conflicts, restrictive exclusion feature triggers, or external treatment adaptation anomalies.

[0028] In actual clinical photographs, a skin image often includes environmental information such as healthy skin and clothing in the background, and multiple skin lesions with different locations and manifestations may exist simultaneously in one image. If the entire image is treated as a single entity, the knowledge graph cannot specifically describe the subtle changes in each local skin lesion. By setting the "lesion region" as an independent primary ontology, a "containment and being contained" relationship can be established between the image and local lesions. This allows the algorithm to create nodes for a specific lesion point in the image (such as erythema lesions on the back of the left hand and erosion lesions on the right wrist), thereby providing underlying structural data support for subsequent independent diagnosis and treatment matching for different local lesion regions.

[0029] External therapies are specific treatment methods, such as wet compress therapy and fire needling therapy. Contraindications, on the other hand, are the constraints that limit their use. If contraindications are directly written as attributes and fixed within a specific therapy entity, the graph knowledge associations become bloated and difficult to maintain, because a specific clinical contraindication (e.g., "not for pregnant women" or "not for broken skin") often simultaneously constrains multiple external therapies. By setting them as independent second entities, a contraindication node can act like a "safety constraint network," flexibly acquiring association edges with multiple potentially conflicting therapies, and dynamically performing safety checks on various treatment methods.

[0030] Traditional medical knowledge graphs typically halt recommendations or display error messages when system recommendations encounter obstacles (such as insufficient scores on diagnostic indicators) or safety risks (such as matching prohibited drugs), leaving no traceable record in the patient's knowledge graph data chain. This embodiment configures the review item as a traceable third ontology type that records diagnostic conflicts, triggering of restrictive exclusion features, or abnormal external treatment adaptation. When various anomalies occur during clinical reasoning, such as unmet primary symptoms, triggered exclusion features, or ultimately resulting in caution or prohibition tags, a corresponding review item entity node is generated for this diagnosis in the knowledge graph database. This node records the specific location and cause of the anomaly, serving as a diagnostic record and log traceability, allowing clinicians to clearly trace and verify at which diagnostic or adaptation stage the conflict occurred.

[0031] Preferably, adaptive lesion segmentation is performed on the skin image to divide it into at least one local lesion region, including: Detect edge boundaries, color abrupt changes, and texture differences in the skin image to determine the area of ​​the lesion to be examined; The lesion area to be examined is masked, and the geometric contour parameters and area distribution characteristics of the area are calculated. If the skin image contains multiple local lesion areas, a unique lesion area identifier is assigned to each local lesion area, and the lesion area identifier is written into the corresponding lesion area evidence unit.

[0032] For clinically acquired skin images, the transition zone between lesions and normal skin has its own characteristics. These include identifying the lines (edge ​​boundaries) where erythema and normal skin color meet, identifying the transition points where the color changes from normal skin tone to light red or dark red (color abrupt changes), and identifying areas where the skin surface becomes rougher due to scaling or crusting (texture differences). Through feature extraction in these three dimensions, the algorithm can select areas suspected of being lesions as "lesion regions to be examined." Then, by generating a binarized or multidimensional mask, the lesion area is locked and separated from the healthy skin background using specific pixel markers. The algorithm then calculates the aspect ratio (length-to-width ratio) and area percentage of the boundary contour within this locked region. These geometric and area parameters are core quantitative data for subsequent judgment of lesion morphology (e.g., whether it is a fused patch or scattered distribution).

[0033] If a skin image contains multiple localized lesion areas, a unique lesion area identifier is assigned to each localized lesion area, and this identifier is written into the corresponding lesion area evidence unit. For example, if multiple scattered skin lesions appear simultaneously in a skin image (e.g., multiple different papules or crusts on the back of a patient's hand or wrist), each of these areas is assigned a unique lesion number (e.g., the first lesion is ID_01, the second is ID_02). Subsequently, these lesion numbers are written into the data file corresponding to each lesion, i.e., into the lesion area evidence unit. This ensures that subsequent differentiation and adaptation rules are performed independently for a specific lesion, avoiding overlap and confusion of information from multiple lesions in the same image.

[0034] Preferably, the visual characterization information includes the color characteristics, geometric distribution characteristics, and surface condition characteristics of the local lesion area, representing the deformation of the surface material or the presence of secretions. The physical characteristics of the skin include the physiological structural location where the lesion occurs and the state of skin barrier integrity.

[0035] In clinical applications, visual representation information and skin physical characteristics serve as morphological evidence for TCM syndrome differentiation and as the determination of the safety limits of external treatment entities, respectively. In this embodiment, visual representation information, skin physical characteristics, and the safety limits of external treatment entities jointly determine the specific attribute filling content of each entity in the knowledge graph. Visual representation information describes the intuitive visual state of the lesion and mainly includes three parts. The first part is color characteristics. For example, by extracting specific color parameters of the lesion, such as bright red, dark red, light red, or purplish-dark, it directly provides visual color evidence for clinically distinguishing TCM syndromes such as excessive pathogenic heat or deficiency of qi and blood. The second part is geometric distribution characteristics, used to quantify the overall spatial layout and edge morphology of the lesion area. For example, whether the lesion is symmetrically distributed on both sides of the limbs or scattered on one side, and whether the boundary of its transition area is clear or blurred. This geometric deformation information is an important visual dimension to assist in determining the classification of the disease. The third part is the surface state characteristics of the lesion reflecting the deformation of the substance or the exudation of secretions. Deformation of the surface texture reflects structural changes in the local skin tissue of the lesion, such as "scaling" caused by the peeling of the outermost stratum corneum, "lichenification" caused by thickening and protrusion of the skin due to chronic inflammation or long-term scratching, "scabs" formed by the hardening and coagulation of surface exudate, and "pigmentation" that deepens in the area during the recovery period. The presence of secretions emphasizes the moist, exudative state of the lesion. By calculating the reflectivity and texture of the skin surface, acute exudative states such as "erosion" caused by superficial epidermal peeling or "exudation" with obvious clear, thin mucus are marked as characteristic features, thus comprehensively outlining the surface pathological texture of the lesion.

[0036] The physical characteristics of the skin primarily describe lesions from the rigid constraints of human anatomy and barrier safety, mainly comprising two parts. The first part is the physiological structural location, specifying the anatomical location of the lesion in bone or soft tissue, such as whether the lesion occurs on the face, the extensor surfaces of the limbs, or the thinner skin folds of the trunk. The second part is the state of skin barrier integrity, specifically used to determine whether the lesion surface has been broken or ulcerated. Because different physiological sites of the body have varying absorption rates of chemical and physical drugs and different tolerances to irritants, and because the risk of secondary infection or severe stinging increases sharply once the skin barrier is damaged, incorporating the physiological structural location and skin barrier integrity as physical characteristic indicators of skin lesions into the knowledge graph provides a foundation for subsequent adaptation rules to issue risk contraindications for different external treatment methods (such as local fire needling or application of potent ointments).

[0037] Preferably, a diagnostic feature discrimination matrix for candidate syndrome configuration is constructed, and feature matching is performed on the evidence units of the lesion area based on the diagnostic feature discrimination matrix to generate candidate syndromes and diagnostic evidence combinations corresponding to the local lesion area, including: For each of the candidate certificate types, a primary certificate feature is configured as an admission criterion, a supporting evidence feature is configured as an auxiliary reference, and an exclusion feature is configured as an exclusion criterion. The visual representation information is compared and matched with the main evidence feature, the supporting evidence feature, and the exclusion feature respectively. In response to the admission criteria that meet the main evidence characteristics and the exclusion characteristics are not triggered, the corresponding candidate evidence type is determined, and the matching main evidence characteristics and supporting evidence characteristics are written into the diagnostic evidence combination.

[0038] This embodiment constructs a diagnostic feature discrimination matrix to derive precise candidate syndrome types and combinations of diagnostic evidence. Specifically, three features are configured for each possible TCM syndrome type in the atlas: primary syndrome features, supporting features, and exclusion features. These three features have different properties. The primary syndrome features serve as a prerequisite for determining a candidate syndrome type and must meet certain pre-defined conditions. Supporting features mainly play an auxiliary reference role, used to complete the chain of evidence for diagnosis. Exclusion features play a mutually exclusive and restrictive role, used to exclude inconsistent syndrome types. For example, for "damp-heat accumulation in the skin syndrome," its primary syndrome features in the matrix are configured as visual indicators with diagnostic significance such as "exudation" and "erosion." During feature matching, if the candidate syndrome type is "damp-heat accumulation in the skin syndrome," the features of "exudation" and "erosion" must be met. Supporting features are configured as auxiliary indicators such as "symmetrical distribution" and "bright red color," while exclusion features are configured as mutually exclusive indicators such as "dry skin lesions." Through multi-dimensional feature configuration, each TCM syndrome type has a clear judgment boundary in the calculation logic.

[0039] In the feature matching stage, the visual representation information extracted from the local lesion area is compared with the main syndrome features, supporting features, and exclusion features in the discrimination matrix for each syndrome type. For example, for a specific patient's skin lesion image, visual representation information such as "surface exudation," "local erosion," and "bilateral symmetrical distribution" is detected in the local lesion. These actually extracted features are cross-validated with the rules set in the "Damp-Heat Accumulation in the Skin Syndrome" matrix. Logical judgment is performed based on the comparison results. Only when the main syndrome features meet the admission criteria and the exclusion features are not triggered can a candidate syndrome be established. Continuing with the above example, since the lesion matched the main syndrome features "exudation" and "erosion," reaching the admission threshold, and did not match the exclusion feature "dry skin lesion," the candidate syndrome corresponding to the local lesion area was finally determined to be "Damp-Heat Accumulation in the Skin Syndrome." At the same time, the specific primary and supporting features that played a confirmatory role in this pairing, such as "exudation," "erosion," and "symmetrical distribution," were compiled and written into the visual evidence combination of the local lesion area as structured evidence to support the judgment result.

[0040] Preferably, the external treatment safety adaptation rules are constructed, including: Establish multidimensional adaptation rules between the external treatment entity and the applicable lesion surface state characteristics, skin barrier integrity state, and physiological structure parts, and set the corresponding range of physicochemical operation parameters for each external treatment entity; The range of physicochemical operation parameters includes at least one of the following: drug dosage form, mode of action, duration of single treatment, treatment frequency, and standard chemotherapy cycle.

[0041] After identifying the candidate syndrome type, it's not enough to directly determine which topical treatment to use. Instead, a multi-dimensional safety filtering process is needed, considering the lesion's surface characteristics, skin barrier integrity, and physiological location. For example, after identifying "damp-heat accumulation in the skin syndrome," relevant topical treatment methods (such as "traditional Chinese medicine wet compress therapy") are retrieved. At this point, data from three levels are compared according to the matching rules: First, the lesion's surface characteristics: due to the "exudation" and "erosion" of the lesion, this type of wet lesion is very suitable for liquid wet compresses, but not for ointments; second, the skin barrier integrity: if "skin is damaged (eroded)," according to the rules, highly irritating ethanol lotions or occlusive therapies will be filtered out to prevent severe pain or worsening of the skin lesions; third, the physiological location: if the lesion is located on delicate areas such as the "face," the rules will restrict the use of agents that easily stain or contain peeling ingredients. Through cross-checking of multi-dimensional matching rules, safety risks from inappropriate topical treatments are effectively avoided.

[0042] Furthermore, to ensure the safe and standardized execution of the selected external therapies, specific physicochemical operational parameter ranges need to be configured for each individual external therapy. Simply put, these parameters not only specify "what medication to use" but also "how to use it." Taking the "traditional Chinese medicine wet compress therapy" for facial exudative erosion lesions as an example, the specific physicochemical operational parameters are as follows: Regarding the drug dosage form, it is limited to "solutions," prohibiting the use of powders or ointments; regarding the mode of treatment, it is specified as "cold wet compress," specifically using multiple layers of gauze to absorb the medicinal solution and apply it to the affected area, rather than "heat fumigation" or "occlusive application"; regarding the duration of a single treatment, it is set at "15 to 20 minutes each time" to prevent excessive time from causing local skin softening and necrosis; regarding the frequency of treatment, it is set at "twice daily"; and regarding the standardized treatment cycle, it is limited to "continuous use for 3 to 5 days." If this cycle is exceeded, a reassessment will be required to prevent skin allergies. These parameters, as specific physical boundaries, ensure the safety and operability of external therapies in clinical applications.

[0043] Preferably, generating the safety adaptation assessment results for the external treatment entity corresponding to the local lesion area includes: In response to the fact that the surface condition of the lesion and the skin barrier integrity of the local lesion area meet the preset adaptation criteria of the corresponding external treatment entity, an applicable label is configured for the external treatment entity; In response to the preset safety constraints that match the physiological structural location, the surface condition of the lesion, or the skin barrier integrity state to the corresponding external treatment entity, a caution label is configured for the external treatment entity; In response to the skin barrier integrity status of the local lesion area being characterized as a damaged state, the lesion surface state characteristics conforming to preset infection characteristics, or the lesion area evidence unit matching the restrictive treatment item corresponding to the contraindication, a disabling label is configured for the external treatment entity.

[0044] When configuring an "applicable" label, the surface characteristics of the lesion area and the integrity of the skin barrier must simultaneously meet the baseline conditions set by the external treatment method to ensure a high degree of match between the treatment and the physical state of the lesion. For example, continuing with the previous example of "damp-heat accumulation in the skin syndrome," the extracted lesion surface exhibits "exudation and erosion," and the skin barrier is only superficially damaged. For "traditional Chinese medicine wet compress therapy," its main physical mechanism relies on the evaporation and capillary action of the medicinal solution to astringe the wound with fluid exudation. The surface and barrier state of this lesion perfectly meet the baseline requirements of wet compress therapy; therefore, an "applicable" label is configured for traditional Chinese medicine wet compress.

[0045] When configuring the "Use with Caution" label, a single-item judgment logic is adopted. If any one of the three elements—the physiological structure of the lesion, the surface condition of the lesion, or the integrity of the skin barrier—triggers a risk warning, the treatment will be flagged. Continuing the example above, if the treatment being evaluated is "Traditional Chinese Medicine fumigation therapy," although fumigation has the effect of promoting blood circulation and clearing the meridians, the lesion's physiological structure is located on the face. The warm steam can easily cause reflexive dilation of facial capillaries, thus aggravating local erythema and swelling. This situation matches the safety constraints of fumigation therapy on the face; therefore, a "Use with Caution" label is configured for fumigation therapy, indicating that temperature and time need to be controlled during clinical operation.

[0046] When configuring the "disabled" label, an absolute exclusion judgment logic is adopted. If the skin barrier integrity of the lesion shows deep damage, signs of infection, or if the data within the evidence unit of the lesion area directly triggers the contraindications for external treatment, a "one-vote veto" will be implemented. For example, if the patient's facial lesion barrier integrity deteriorates to deep "ulceration," or if the lesion surface condition shows yellowish-white purulent discharge, it meets the preset bacterial infection characteristics, directly triggering the restrictive contraindications for the treatment on infected and damaged areas. To prevent the aggravation of skin lesions or the spread of pathogens, invasive and closed therapies such as traditional Chinese medicine occlusion or fire needling are assigned a "disabled" label to eliminate potential clinical safety hazards.

[0047] Preferably, a diagnostic semantic association chain is constructed and written into a knowledge graph database, and review items are generated and queued for inspection based on preset abnormality triggering conditions in the feature matching and adaptation evaluation process, including: Taking the lesion area evidence unit as the association starting point, directed semantic association edges pointing to the visual diagnosis evidence combination, the candidate syndrome type, the safety adaptation assessment result and the review item are generated sequentially, and the mapping association between the lesion area identifier assigned to each of the local lesion areas and the skin image is written into each of the directed semantic association edges; In response to the configuration of the main evidence feature not meeting the corresponding admission judgment condition, triggering the configuration of the exclusion feature, or the safety adaptation assessment result of the external treatment entity being either a caution label or a disable label, a corresponding review item is generated and enters the review queue for inspection.

[0048] This embodiment establishes the traceability of the entire diagnostic chain by building a logical network in the database. Specifically, starting with the initially extracted lesion area evidence unit, the logical deduction proceeds sequentially, generating unidirectional arrows (i.e., directed semantic association edges) pointing to the combination of visual evidence, candidate syndrome types, safety adaptation assessment results, and finally the review item, thus forming a clear and closed-loop medical reasoning network. To establish a factual binding relationship between these logical deductions and the acquired images, the mapping association between the lesion area identifier assigned to each local lesion area and the skin image is directly written into these directed semantic association edges, avoiding the misinterpretation of a local lesion diagnosis as a diagnosis of the entire skin image. Taking the aforementioned facial "damp-heat skin syndrome" as an example, in the atlas database, starting from the "lesion R001 evidence unit" node, an arrow is drawn pointing to the "exudation, erosion, symmetry" visual evidence combination node, then another arrow points to the "damp-heat skin syndrome" candidate syndrome type node, subsequently passing through the assessment results and finally pointing to the review item node. The relationship features of these arrows all include the "mapping association between the lesion region identifier R001 and the facial skin image". Subsequent tracing by doctors or system analysis can directly use the directed edges in the atlas to pinpoint the specific location of the skin lesion R001 in the facial skin image with a single click.

[0049] Furthermore, during the feature comparison process, if any of the following three situations are detected: the number of main certificate features corresponding to the candidate certificate type does not meet the admission conditions, the exclusion feature is triggered, or the security assessment results in a "use with caution" or "disabled" label, the current inference judgment will be defined as abnormal, a corresponding review item will be generated immediately, and the review item will be pushed to the review queue to wait for detection.

[0050] In the case of "damp-heat accumulation in the skin syndrome," because the local lesion R001 is located on the face, the external treatment matching rule analysis process judged "traditional Chinese medicine fumigation therapy" as "use with caution." At the same time, because there was "erosion" on the lesion surface, "traditional Chinese medicine occlusion" was marked as "prohibited." When the safety assessment results of "use with caution" and "prohibited" meet the preset abnormal judgment conditions, the system will automatically assemble the relevant lesion area identifier R001, the two external treatment entities of fumigation and occlusion, and the reasons for failure to pass the safety assessment into a corresponding review item, and directly push it into the backend pending review queue, waiting for the doctor to manually review and sign it.

[0051] Example 2

[0052] Based on the principles described in Example 1, a knowledge graph construction system 100 for external treatment of TCM skin diseases based on local skin lesions is also proposed, such as... Figure 2 As shown, it includes: Unit 110 is used to determine the ontology type of TCM skin disease diagnosis and treatment, configure ontology attributes and define the association between the ontology types, wherein the ontology type includes lesion area, inspection characteristics, syndrome type, external treatment entity, contraindication item and review item; The second unit 120 is used to perform adaptive lesion segmentation on the skin image to divide at least one local lesion region, extract the visual representation information and skin physical features of each local lesion region, and generate a corresponding lesion region evidence unit. The third unit 130 is used to construct a diagnostic feature discrimination matrix for candidate syndrome configuration, and to perform feature matching on the evidence unit of the lesion area based on the diagnostic feature discrimination matrix to generate candidate syndrome and diagnostic evidence combination corresponding to the local lesion area. The fourth unit 140 is used to construct the external treatment safety adaptation rule. The candidate syndrome and the lesion area evidence unit are input into the external treatment safety adaptation rule to generate the safety adaptation evaluation result of the external treatment method entity corresponding to the local lesion area. The fifth unit 150 is used to construct a diagnostic semantic association chain based on the lesion area evidence unit, the visual diagnosis evidence combination, the candidate syndrome, the safety adaptation assessment result, and the review item, and to write it into the knowledge graph database. It also generates corresponding review items according to the preset abnormal triggering conditions in the feature matching and adaptation assessment process and queues them for inspection.

[0053] Example 3

[0054] Furthermore, an electronic device is proposed, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to implement, when executing executable instructions, a method for constructing a knowledge graph of external treatments for skin diseases based on local skin lesions in traditional Chinese medicine, as described in Embodiment 1.

[0055] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0056] Furthermore, it should be noted that the use of terms such as "first," "second," and "a" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. The terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two elements or the interaction between two elements, unless otherwise explicitly specified. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0057] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0058] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for constructing a knowledge graph of external treatments for skin diseases in Traditional Chinese Medicine based on local skin lesions, characterized in that, include: Determine the ontology types for the diagnosis and treatment of skin diseases in traditional Chinese medicine, configure ontology attributes and define the relationships between the ontology types, wherein the ontology types include lesion areas, diagnostic features, syndrome types, external treatment entities, contraindications and review items; Skin images are subjected to adaptive lesion segmentation to divide at least one local lesion region. Visual representation information and skin physical features of each local lesion region are extracted to generate corresponding lesion region evidence units. Construct a diagnostic feature discrimination matrix for candidate syndrome configuration, perform feature matching on the evidence units of the lesion area based on the diagnostic feature discrimination matrix, and generate candidate syndromes and diagnostic evidence combinations corresponding to the local lesion area; Construct external treatment safety adaptation rules, input the candidate syndrome type and the lesion area evidence unit into the external treatment safety adaptation rules, and generate the safety adaptation evaluation result of the external treatment method entity corresponding to the local lesion area; Based on the lesion area evidence unit, the visual diagnosis evidence combination, the candidate syndrome, the safety adaptation assessment result, and the review item, a diagnostic semantic association chain is constructed and written into the knowledge graph database. Corresponding review items are generated according to the preset abnormal triggering conditions in the feature matching and adaptation assessment process and are queued for inspection.

2. The method according to claim 1, characterized in that, Determine the ontological type of TCM dermatology diagnosis and treatment, including: Set the lesion region as a first ontology type independent of the skin image; Set the contraindication to a restrictive second ontology type independent of the external treatment entity; Configure the review item as a traceable third ontology type that records dialectical conflicts, restrictive exclusion feature triggers, or external treatment adaptation anomalies.

3. The method according to claim 1, characterized in that, Adaptive lesion segmentation is performed on skin images to delineate at least one local lesion region, including: Detect edge boundaries, color abrupt changes, and texture differences in the skin image to determine the area of ​​the lesion to be examined; The lesion area to be examined is masked, and the geometric contour parameters and area distribution characteristics of the area are calculated. If the skin image contains multiple local lesion areas, a unique lesion area identifier is assigned to each local lesion area, and the lesion area identifier is written into the corresponding lesion area evidence unit.

4. The method according to claim 1, characterized in that, The visual representation information includes the color characteristics, geometric distribution characteristics, and surface condition characteristics of the local lesion area, which represent the deformation of the surface material or the presence of secretions. The physical characteristics of the skin include the physiological structural location where the lesion occurs and the state of skin barrier integrity.

5. The method according to claim 1, characterized in that, Construct a diagnostic feature discrimination matrix for candidate syndrome configurations, perform feature matching on the evidence units of the lesion region based on the diagnostic feature discrimination matrix, and generate candidate syndromes and diagnostic evidence combinations corresponding to the local lesion region, including: For each of the candidate certificate types, a primary certificate feature is configured as an admission criterion, a supporting evidence feature is configured as an auxiliary reference, and an exclusion feature is configured as an exclusion criterion. The visual representation information is compared and matched with the main evidence feature, the supporting evidence feature, and the exclusion feature respectively. In response to the admission criteria that meet the main evidence characteristics and the exclusion characteristics are not triggered, the corresponding candidate evidence type is determined, and the matching main evidence characteristics and supporting evidence characteristics are written into the diagnostic evidence combination.

6. The method according to claim 4, characterized in that, Constructing external treatment safety adaptation rules, including: Establish multidimensional adaptation rules between the external treatment entity and the applicable lesion surface state characteristics, skin barrier integrity state, and physiological structure location, and set the corresponding range of physicochemical operation parameters for each external treatment entity; The range of physicochemical operation parameters includes at least one of the following: drug dosage form, mode of action, duration of single treatment, treatment frequency, and standard chemotherapy cycle.

7. The method according to claim 6, characterized in that, The safety adaptation assessment results for the external treatment entity corresponding to the local lesion area are generated, including: In response to the fact that the surface condition characteristics of the lesion in the local lesion area and the skin barrier integrity state meet the preset adaptation benchmark conditions of the corresponding external treatment entity, an applicable label is configured for the external treatment entity; In response to the preset safety constraints that match the physiological structural location, the surface condition of the lesion, or the skin barrier integrity state to the corresponding external treatment entity, a caution label is configured for the external treatment entity; In response to the skin barrier integrity status of the local lesion area being characterized as a damaged state, the lesion surface state characteristics conforming to preset infection characteristics, or the lesion area evidence unit matching the restrictive treatment item corresponding to the contraindication, a disabling label is configured for the external treatment entity.

8. The method according to claim 7, characterized in that, Constructing a semantic association chain for diagnosis and treatment and writing it into a knowledge graph database, and generating review items to be queued for inspection based on preset abnormality triggering conditions in the feature matching and adaptation evaluation process, including: Taking the lesion area evidence unit as the association starting point, directed semantic association edges pointing to the visual diagnosis evidence combination, the candidate syndrome, the safety adaptation assessment result and the review item are generated sequentially, and the mapping association between the lesion area identifier assigned to each of the local lesion areas and the skin image is written into each of the directed semantic association edges; In response to the configuration of the main evidence feature not meeting the corresponding admission judgment condition, triggering the configuration of the exclusion feature, or the safety adaptation assessment result of the external treatment entity being either a caution label or a disable label, a corresponding review item is generated and enters the review queue for inspection.

9. A knowledge graph construction system for external treatment of skin diseases in Traditional Chinese Medicine based on local skin lesions, characterized in that, include: The first unit is used to determine the ontology type for diagnosis and treatment of skin diseases in traditional Chinese medicine, configure ontology attributes and define the relationship between the ontology types, wherein the ontology type includes lesion area, observation characteristics, syndrome type, external treatment entity, contraindication item and review item; The second unit is used to perform adaptive lesion segmentation on the skin image to divide at least one local lesion region, extract the visual representation information and skin physical features of each local lesion region, and generate a corresponding lesion region evidence unit. The third unit is used to construct a diagnostic feature discrimination matrix for candidate syndrome configuration, and to perform feature matching on the evidence unit of the lesion area based on the diagnostic feature discrimination matrix to generate candidate syndrome and diagnostic evidence combination corresponding to the local lesion area. The fourth unit is used to construct the external treatment safety adaptation rule. The candidate syndrome and the lesion area evidence unit are input into the external treatment safety adaptation rule to generate the safety adaptation evaluation result of the external treatment method entity corresponding to the local lesion area. The fifth unit is used to construct a diagnostic semantic association chain based on the lesion area evidence unit, the visual diagnosis evidence combination, the candidate syndrome, the safety adaptation assessment result, and the review item, and to write it into the knowledge graph database. It also generates corresponding review items according to the preset abnormal triggering conditions in the feature matching and adaptation assessment process and queues them for inspection.

10. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to implement the method for constructing a knowledge graph of external treatment of traditional Chinese medicine skin diseases based on local skin lesions as described in any one of claims 1-8 when executing executable instructions.

Images