Periodontal patient data management method based on knowledge graph

Abstract

The present application relates to the field of knowledge graph, especially to a periodontal patient data management method based on knowledge graph. The content includes: acquiring periodontal patient heterogeneous data and preprocessing to generate a comprehensive feature vector; based on the periodontal patient heterogeneous data and the comprehensive feature vector, extracting medical entities and identifying entity relationships, and constructing a periodontal patient knowledge graph; introducing a dynamic multi-dimensional semantic driven relationship weighting algorithm to dynamically assign weights to the entity relationships in the knowledge graph, and obtaining dynamically adjusted entity relationship weights; based on the newly added heterogeneous data of the periodontal patient, iteratively updating the entity relationship weights to obtain the updated knowledge graph. The technical problem that the traditional periodontal patient data management method is difficult to model multiple heterogeneous data sources, cannot be semantically processed and dynamically updated, and leads to lack of intelligent support for periodontal disease diagnosis and treatment decision-making is solved.

Description

Technical Field

[0001] This invention relates to the field of knowledge graphs, and more particularly to a method for managing periodontal patient data based on knowledge graphs. Background Technology

[0002] With the increasing incidence of periodontal disease, a large amount of clinical data on periodontal patients has accumulated in the oral healthcare system. This data comes from diverse sources and is complex in type, including structured examination indicators (such as periodontal probing depth and attachment loss) as well as unstructured text records (such as electronic medical records and doctor's comments), medical imaging reports, and surgical procedure records. Existing electronic medical record systems mostly store structured data in tabular form, lacking effective parsing capabilities for unstructured information and making it difficult to achieve semantic fusion between different types of data. Furthermore, periodontal disease is characterized by chronic evolution, multiple intervention cycles, and joint assessment of multiple indicators. Traditional data management methods cannot reflect the disease progression path and individual differences, severely restricting the realization of intelligent decision-making and precise intervention. Currently, some studies have attempted to introduce knowledge graphs for medical information management, but these efforts mostly remain at the level of static knowledge organization.

[0003] In summary, traditional periodontal patient data management methods still have the problems of difficulty in uniformly modeling multi-source heterogeneous data, and the inability to perform semantic processing and dynamic updates, resulting in a lack of intelligent support for periodontal disease diagnosis and treatment decisions. Summary of the Invention

[0004] This invention provides a knowledge graph-based method for managing periodontal patient data, which addresses the technical problem that traditional periodontal patient data management methods struggle to uniformly model multi-source heterogeneous data, cannot perform semantic processing and dynamic updates, and thus lack intelligent support for periodontal disease diagnosis and treatment decisions.

[0005] The present invention provides a knowledge graph-based method for managing periodontal patient data, which specifically includes the following technical solutions: A knowledge graph-based method for managing periodontal patient data includes the following steps: S1. Obtain heterogeneous data of periodontal patients and preprocess it to generate a comprehensive feature vector; based on the heterogeneous data of periodontal patients and the comprehensive feature vector, extract medical entities and identify entity relationships to construct a knowledge graph of periodontal patients; S2. A dynamic multidimensional semantic-driven relation weighting algorithm is introduced to dynamically assign weights to entity relations in the knowledge graph, resulting in dynamically adjusted entity relation weights. Based on newly added heterogeneous data of periodontal patients, the entity relation weights are iteratively updated to obtain the updated knowledge graph.

[0006] Preferably, S1 specifically includes: The heterogeneous data of periodontal patients includes structured numerical indicators, semi-structured text descriptions, and unstructured medical images. The heterogeneous data of periodontal patients is preprocessed to obtain numerical feature vectors, text feature vectors, and image feature vectors, and the numerical feature vectors, text feature vectors, and image feature vectors are concatenated to generate a comprehensive feature vector.

[0007] Preferably, S1 specifically includes: Based on the semi-structured text descriptions in heterogeneous periodontal patient data, medical entities are extracted as nodes in the knowledge graph. Then, combined with comprehensive feature vectors, semantic relationships between medical entities are identified as edges in the knowledge graph. Semantic embedding mapping is performed on medical entities and their relationships to obtain semantic embedding vectors.

[0008] Preferably, S2 specifically includes: The dynamic multidimensional semantic-driven relationship weighting algorithm, based on the periodontal patient knowledge graph, combines the semantic similarity of medical entities, information frequency weights, and time decay factors to generate dynamically adjusted entity relationship weights.

[0009] Preferably, S2 specifically includes: The semantic similarity is obtained by calculating the inner product between the semantic embedding vectors of medical entities.

[0010] Preferably, S2 specifically includes: The information frequency weight is calculated by introducing adjustment parameters and combining the co-occurrence frequency of entity relationships and the credibility of the data source.

[0011] Preferably, S2 specifically includes: The co-occurrence count is the number of times the entity relationship appears after normalization; the data source credibility is preset according to different data sources. When the entity relationship exists in different data sources, a fusion strategy is used to calculate the credibility of the overall data source.

[0012] Preferably, S2 specifically includes: The time decay factor dynamically adjusts the entity relationship weights by introducing a time decay parameter and combining it with the time difference between the extraction of two medical entities from heterogeneous periodontal patient data.

[0013] Preferably, S2 specifically includes: The newly added heterogeneous data of periodontal patients are processed according to the above steps to obtain new entity relation weights; based on the new entity relation weights, a knowledge graph dynamic evolution algorithm is introduced to iteratively update the historical entity relation weights in the knowledge graph.

[0014] Preferably, S2 specifically includes: In the implementation of the knowledge graph dynamic evolution algorithm, a balance coefficient is introduced to fuse the weights of historical entity relationships and the weights of new entity relationships, thereby obtaining the updated entity relationship weights and generating the updated knowledge graph.

[0015] The beneficial effects of the technical solution of the present invention are: 1. Achieve efficient fusion and structured representation of heterogeneous data from periodontal patients: By using methods such as structured normalization, text semantic encoding, and image feature extraction, break down the barriers between numerical, text, and image data, construct a unified comprehensive feature vector, and provide multimodal support for knowledge graph construction.

[0016] 2. Construct a high-quality medical knowledge graph to improve the clinical usability of data: Utilize pre-trained medical named entity recognition and relationship extraction models, combined with existing medical knowledge bases, to achieve the automatic generation of structured medical knowledge graphs, providing knowledge support for disease tracing, disease understanding, and precision diagnosis and treatment.

[0017] 3. Introduce a dynamic multidimensional semantic-driven mechanism to enhance the credibility and timeliness of entity relationship weights: By comprehensively considering semantic similarity, co-occurrence frequency, data source credibility, and time decay factor, the strength of relationships between medical entities is quantified, making the entity relationship weights of the knowledge graph interpretable and dynamically adaptable.

[0018] 4. Supports dynamic evolution and updates of knowledge graphs to reflect the latest clinical status: By integrating historical entity relation weights and new entity relation weights through a linear weighted update mechanism, the knowledge graph can iteratively evolve as patient status and medical data change, thereby enhancing the clinical adaptability and intelligent decision support capabilities of the knowledge graph. Attached Figure Description

[0019] Figure 1 This is a flowchart of a knowledge graph-based periodontal patient data management method according to the present invention. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] The following description, in conjunction with the accompanying drawings, details a specific scheme for a knowledge graph-based periodontal patient data management method provided by this invention.

[0023] See attached document Figure 1 The diagram illustrates a flowchart of a knowledge graph-based periodontal patient data management method according to an embodiment of the present invention, which includes the following steps: S1. Obtain heterogeneous data of periodontal patients and preprocess it to generate a comprehensive feature vector; based on the heterogeneous data of periodontal patients and the comprehensive feature vector, extract medical entities and identify entity relationships to construct a knowledge graph of periodontal patients; First, heterogeneous data of periodontal patients are obtained from the medical information system, including structured numerical indicators, semi-structured text descriptions, and unstructured medical images. The structured numerical indicators include periodontal pocket depth, attachment loss, plaque index, and bleeding index. The semi-structured text descriptions include symptoms and diagnostic opinions in doctors' medical records. The unstructured medical images include oral X-rays and cone-beam computed tomography (CBCT) scans. Because periodontal patients have heterogeneous data from different sources and with inconsistent units of measurement, it is necessary to preprocess the heterogeneous data of periodontal patients to obtain numerical feature vectors, text feature vectors and image feature vectors respectively. The numerical feature vectors, text feature vectors and image feature vectors are then concatenated to generate a comprehensive feature vector that fully represents the status of a single patient. The specific implementation process of the preprocessing is as follows: For structured numerical indicators, statistical normalization is used. Specifically, the historical population mean of each structured numerical indicator (from an existing database) is subtracted from the individual structured numerical indicator value of each patient, and then divided by the standard deviation of the structured numerical indicator value to obtain the numerical feature. All numerical features of the patient are concatenated to obtain the numerical feature vector. For semi-structured text descriptions, semantic encoding is performed using a pre-trained language model in the medical field (such as Chinese Medical BERT) to map the semi-structured text descriptions to a high-dimensional semantic vector space to obtain the text feature vector. For unstructured medical images, convolutional neural networks (CNNs) are used to extract image features and output image feature vectors. The methods used in the preprocessing are all well-known to those skilled in the art and will not be elaborated here. Based on the semi-structured text descriptions in heterogeneous periodontal patient data, a pre-trained medical named entity recognition (NER) model is used to extract medical entities as nodes in the knowledge graph. Based on medical knowledge bases (such as UMLS or hospital databases) and multimodal features (i.e., numerical feature vectors, text feature vectors, and image feature vectors) in the comprehensive feature vector, a relation extraction model (such as a BERT-based relation classifier) ​​is used to identify entity relations as edges in the knowledge graph, with the initial weights of the edges set to default values ​​(e.g., 0.5). A pre-trained translation embedding model (TransE model) is introduced to perform semantic embedding mapping on medical entities and their entity relations, generating high-dimensional vector representations to obtain semantic embedding vectors. The pre-trained medical named entity recognition (NER) model, relation extraction model, and pre-trained translation embedding model (TransE model) are all techniques well-known to those skilled in the art and will not be elaborated upon here.

[0024] S2. A dynamic multidimensional semantic-driven relation weighting algorithm is introduced to dynamically assign weights to entity relations in the knowledge graph, resulting in dynamically adjusted entity relation weights. Based on newly added heterogeneous data of periodontal patients, the entity relation weights are iteratively updated to obtain the updated knowledge graph.

[0025] In the knowledge graph of periodontal patients, medical entities (such as "periodontitis" and "gingival bleeding") are connected by edges, and edges represent entity relationships (such as symptom-diagnosis). However, the strength and importance of entity relationships vary depending on the data source, semantic association, patient feature similarity, and data freshness. Unweighted entity relationships cannot accurately reflect the priority and credibility of medical knowledge, which may lead to bias in diagnosis or treatment recommendations. Therefore, a dynamic multidimensional semantic-driven relationship weighting algorithm is introduced to assign weights to entity relationships in the knowledge graph in order to quantify the association strength between medical entities (such as symptoms, diagnoses, and treatments). The dynamic multidimensional semantic-driven relationship weighting algorithm, based on the periodontal patient knowledge graph, generates dynamically adjusted entity relationship weights by comprehensively considering the semantic similarity of medical entities, information frequency weights, and time decay factors. The semantic similarity is calculated by the inner product between the semantic embedding vectors of two medical entities (such as "gingival bleeding" and "periodontitis"), which is used to measure the closeness of medical entities in the semantic knowledge base. To ensure that the relation weight values ​​are within a reasonable range, the inner product is input into the Sigmoid function and mapped to the interval between 0 and 1. The Sigmoid function smooths the inner product through exponential transformation to ensure that the calculation results are interpretable. The information frequency weights are calculated by comprehensively considering the co-occurrence frequency of entity relationships and the credibility of the data source. The co-occurrence frequency refers to the number of times an entity relationship (e.g., "gingival bleeding" points to "periodontitis") appears in the medical knowledge base or comprehensive feature vector, reflecting the commonness of the entity relationship. To avoid common medical entities dominating the relationship weight, the co-occurrence frequency of the entity relationship is divided by the geometric mean of the occurrence frequency of each of the two medical entities for normalization. The data source credibility is used to reflect the reliability of the source of semantic information on which entity relationships in the knowledge graph depend. Specifically, it is used to evaluate the reliability of entity relationships from different data channels such as authoritative medical knowledge bases, clinical guidelines, professional literature, or highly subjective medical records. It is set according to expert experience and is not limited here. When entity relationships exist in multiple data sources, a fusion strategy is used to calculate the credibility of the overall data source. Specifically, a weighted average method can be used to fuse the co-occurrence frequency and data source credibility value of different sources to reflect the comprehensive reliability of multiple sources or the representativeness of authoritative sources. This is a technique well known to those skilled in the art and will not be elaborated here. Furthermore, the co-occurrence frequency of entity relationships is multiplied by the data source credibility to obtain the information frequency weight, which reflects the reliability and authority of entity relationships in clinical statistical significance. At the same time, an adjustment parameter is introduced to control the contribution of the information frequency weight. The time decay factor, calculated by multiplying the time difference by a time decay parameter (typically 0.1), is used to dynamically adjust the edge weights of entity relationships to reflect their timeliness and freshness. The denominator uses 1 plus the time decay factor to ensure that the entity relationship weights (i.e., edge weights) do not approach zero due to excessive time differences, while maintaining computational stability. The time difference is the difference in time (in days) between the extraction of two medical entities from heterogeneous periodontal patient data. For example, in a medical record, if the symptom of "gingival bleeding" appears on August 1, 2025, and the diagnosis of "periodontitis" appears on August 3, 2025, the associated time difference is 2 days. A large time difference indicates that the entity relationship is historical information, and its value for judging the current condition may be reduced. Therefore, it should be appropriately decayed in the entity relationship weight calculation to enhance the knowledge graph's responsiveness to recent data. The formula for calculating entity relationship weight is:

[0026] in, Represents medical entities and The relationship weight between entities is dynamically adjusted and has a value range of (0,1]. Represents the Sigmoid function, defined as , It is the independent variable of the Sigmoid function; Representing two medical entities and semantic embedding vector and The inner product; Represents medical entities and Semantic similarity between them; Indicates transpose; The adjustment parameter, used to control the contribution of information frequency weight, was obtained through experiments and its value range is [0,1]. Represents medical entities and Information frequency weights of entity relationships between entities , It is a medical entity and The number of times the entity relationship between entities co-occurs is a positive integer. , Medical entities and The number of times each appears, This represents the reliability of the data source, with a value range of [0,1]. The time decay parameter is set based on expert experience, with a value range of [0,1]. To maintain dimensional consistency in the formula, the time decay parameter is set as follows: The unit is "1 / day" to ensure Dimensionless, thus ensuring the mathematical rationality of the denominator in the calculation of entity relation weights and the consistency with engineering implementation; Represents medical entities and The time difference is in days; Indicates the time decay factor; In the above formula, semantic similarity provides the minimum threshold for establishing entity relationships; if the semantics are completely unrelated, the entity relationship weight is extremely low. Information frequency weight and data source credibility provide evidence of group experience, avoiding bias due to scarce samples. The introduction of a time decay factor ensures that the knowledge graph can be dynamically updated with medical developments and changes in patient conditions. The final output relationship weight value is between 0 and 1, with higher values ​​representing stronger relationships and higher credibility. The above dynamic multidimensional semantic-driven relationship weighting algorithm generates corresponding weight values ​​for each edge in the knowledge graph and can provide multidimensional contribution explanations. Because periodontal patient heterogeneous data is dynamic and changes over time, and medical knowledge is constantly evolving, relationship weights may lose their representativeness due to data aging. Knowledge graphs without updated relationship weights may lead to diagnostic biases or inaccurate treatment recommendations. Therefore, it is necessary to introduce new heterogeneous data on periodontal patients and dynamically update entity relationship weights to ensure that the knowledge graph reflects the latest patient status and medical advancements. This enhances the timeliness and clinical applicability of the knowledge graph and improves the reliability of clinical decision support. Therefore, when the medical information system receives new heterogeneous data on periodontal patients (such as new clinical examination results, diagnostic records, treatment plans...), it should... When processing medical records (such as medical reports and imaging reports), new entity relation weights are obtained through preliminary processing. To achieve iterative updates of entity relation weights in the knowledge graph, preserving both long-term medical experience and real-time data value, a dynamic evolution algorithm for the knowledge graph is introduced to iteratively update historical entity relation weights. This algorithm uses a linear combination formula to proportionally fuse historical and new entity relation weights to obtain updated entity relation weights, thereby generating an updated knowledge graph. This ensures the knowledge graph can evolve over time and dynamically reflect the latest clinical information. The specific update formula is as follows:

[0027] For medical entities and The updated entity relationship weights between them; Medical Entities and The weights of historical entity relationships between them are derived from the existing periodontal patient knowledge graph; The new entity relation weights are calculated based on newly added heterogeneous data of periodontal patients. It is the balance coefficient, with a value range of [0,1]. Based on expert experience, it can be set to a value of 0.7. By weighting and balancing the stability and dynamism of the knowledge graph, it is ensured that the knowledge graph can reflect the latest heterogeneous data of periodontal patients (such as periodontal pocket depth, medical record text, and imaging features) and the progress of medical knowledge, while preserving the reliability of historical data. The updated entity relation weights optimize the importance representation of medical entity relations in the knowledge graph, supporting more accurate clinical queries, diagnostic inferences, and treatment recommendations. Using the updated knowledge graph, the correlation strength between a certain symptom and periodontitis can be identified more accurately, assisting doctors in decision-making.

[0028] In summary, a knowledge graph-based method for managing periodontal patient data has been developed.

[0029] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0030] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0031] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for managing periodontal patient data based on knowledge graphs, characterized in that, Includes the following steps: S1. Obtain heterogeneous data of periodontal patients and preprocess it to generate a comprehensive feature vector; based on the heterogeneous data of periodontal patients and the comprehensive feature vector, extract medical entities and identify entity relationships to construct a knowledge graph of periodontal patients; S2. A dynamic multidimensional semantic-driven relation weighting algorithm is introduced to dynamically assign weights to entity relations in the knowledge graph, resulting in dynamically adjusted entity relation weights. Based on newly added heterogeneous data of periodontal patients, the entity relation weights are iteratively updated to obtain the updated knowledge graph.

2. The method for managing periodontal patient data based on knowledge graphs according to claim 1, characterized in that, S1 specifically includes: The heterogeneous data of periodontal patients includes structured numerical indicators, semi-structured text descriptions, and unstructured medical images. The heterogeneous data of periodontal patients is preprocessed to obtain numerical feature vectors, text feature vectors, and image feature vectors, and the numerical feature vectors, text feature vectors, and image feature vectors are concatenated to generate a comprehensive feature vector.

3. The method for managing periodontal patient data based on knowledge graphs according to claim 2, characterized in that, S1 specifically includes: Based on the semi-structured text descriptions in heterogeneous periodontal patient data, medical entities are extracted as nodes in the knowledge graph. Then, combined with comprehensive feature vectors, semantic relationships between medical entities are identified as edges in the knowledge graph. Semantic embedding mapping is performed on medical entities and their relationships to obtain semantic embedding vectors.

4. The method for managing periodontal patient data based on knowledge graphs according to claim 3, characterized in that, S2 specifically includes: The dynamic multidimensional semantic-driven relationship weighting algorithm, based on the periodontal patient knowledge graph, combines the semantic similarity of medical entities, information frequency weights, and time decay factors to generate dynamically adjusted entity relationship weights.

5. The method for managing periodontal patient data based on knowledge graphs according to claim 4, characterized in that, S2 specifically includes: The semantic similarity is obtained by calculating the inner product between the semantic embedding vectors of medical entities.

6. The method for managing periodontal patient data based on knowledge graphs according to claim 5, characterized in that, S2 specifically includes: The information frequency weight is calculated by introducing adjustment parameters and combining the co-occurrence frequency of entity relationships and the credibility of the data source.

7. The method for managing periodontal patient data based on knowledge graphs according to claim 6, characterized in that, S2 specifically includes: The co-occurrence count is the number of times the entity relationship appears after normalization; the data source credibility is preset according to different data sources. When the entity relationship exists in different data sources, a fusion strategy is used to calculate the credibility of the overall data source.

8. The method for managing periodontal patient data based on knowledge graphs according to claim 7, characterized in that, S2 specifically includes: The time decay factor dynamically adjusts the entity relationship weights by introducing a time decay parameter and combining it with the time difference between the extraction of two medical entities from heterogeneous periodontal patient data.

9. A method for managing periodontal patient data based on knowledge graphs according to claim 8, characterized in that, S2 specifically includes: The newly added heterogeneous data of periodontal patients are processed according to the above steps to obtain new entity relation weights; based on the new entity relation weights, a knowledge graph dynamic evolution algorithm is introduced to iteratively update the historical entity relation weights in the knowledge graph.

10. A method for managing periodontal patient data based on knowledge graphs according to claim 9, characterized in that, S2 specifically includes: In the implementation of the knowledge graph dynamic evolution algorithm, a balance coefficient is introduced to fuse the weights of historical entity relationships and the weights of new entity relationships, thereby obtaining the updated entity relationship weights and generating the updated knowledge graph.

Images