Knee osteoarthritis traditional Chinese medicine treatment scheme optimization method based on big data analysis

Abstract

The invention provides a knee osteoarthritis traditional Chinese medicine treatment scheme optimization method based on big data analysis. The knee osteoarthritis traditional Chinese medicine treatment scheme optimization method comprises the steps of firstly extracting entities and relationships from an unstructured knee osteoarthritis medical case text to form a structured triple; constructing a heterograph which comprises symptoms, syndromes, traditional Chinese medicines, prescriptions, curative effect nodes and side relations for connecting the nodes, such as treatment, inclusion and operation relations; distributing weights for the edges, wherein the weights cover frequency weights based on edge relation occurrence frequencies, efficacy weights based on curative effect quantitative scores and expert weights based on expert types; then meta-path analysis is carried out, and meta-paths are defined and traversed to carry out statistics on traditional Chinese medicine clinical rule modes; and finally, according to an analysis result, generating a quantized traditional Chinese medicine clinical rule. According to the method, the weighted heterogeneous graph is constructed and analyzed by extracting the structured triad, quantitative mining of traditional Chinese medicine clinical rules is realized, and the defect that hidden knowledge cannot be analyzed and deep rules cannot be quantized in the prior art is overcome.

Description

Technical Field

[0001] This application belongs to the field of TCM treatment plan optimization, and in particular relates to a method for optimizing TCM treatment plans for knee osteoarthritis based on big data analysis. Background Technology

[0002] Traditional Chinese medicine (TCM) treatment of knee osteoarthritis is based on syndrome differentiation and treatment principles, drawing knowledge from clinical case records. These records guide the development of diagnostic and treatment plans. The approach primarily relies on electronic medical record systems and traditional data mining methods, aiming to store and perform basic analysis of the case data.

[0003] Specifically, electronic medical record systems achieve persistent data preservation by digitally storing unstructured medical record texts (such as symptom and syndrome descriptions); traditional data mining methods analyze drug co-occurrence relationships in medical records through frequent itemset mining to obtain shallow association rules, thereby helping to discover common medication patterns.

[0004] Because medical records exist in the form of unstructured text, electronic medical record systems can only store data but cannot understand the semantics of the text, resulting in the implicit logic between syndrome differentiation and medication. Traditional data mining methods can only handle explicit co-occurrence relationships and lack the ability to quantify contextual semantics and deep clinical rules. Because they cannot extract complex relationships and modifying logic between entities, they form an "experience black box," which seriously restricts the optimization and promotion of TCM diagnosis and treatment experience. Summary of the Invention

[0005] The purpose of this application is to overcome the deficiencies in the prior art and provide a method for optimizing the TCM treatment plan for knee osteoarthritis based on big data analysis.

[0006] This application provides a method for optimizing traditional Chinese medicine treatment plans for knee osteoarthritis based on big data analysis, including:

[0007] Structured triples are obtained from medical case data, wherein the structured triples include entities and relations extracted from unstructured knee osteoarthritis medical case texts;

[0008] Based on the structured triples, a heterogeneous graph is constructed, which includes symptom nodes, syndrome nodes, traditional Chinese medicine nodes, prescription nodes, and efficacy nodes, as well as edge relationships connecting the nodes, including treatment relationships, inclusion relationships, and operation relationships.

[0009] Assign weights to the edges of the heterogeneous graph, the weight assignment including calculating frequency weights based on the frequency of edge relationships, calculating efficacy weights based on efficacy quantification scores, and calculating expert weights based on expert types;

[0010] Based on the heterogeneous graph and the weights, meta-path analysis is performed, which includes defining meta-paths and traversing the meta-paths to statistically analyze TCM clinical rule patterns.

[0011] Based on the results of the meta-path analysis, quantified TCM clinical rules are generated.

[0012] Optionally, structured triples are obtained from medical record data, including:

[0013] The unstructured medical record text in the medical record data is cleaned to generate cleaned text;

[0014] The cleaned text was segmented using a word segmentation tool combined with a custom TCM dictionary to generate word segmentation results.

[0015] The entities and relations are extracted based on the word segmentation results to generate the structured triples.

[0016] Optionally, structured triples are obtained from medical record data, including:

[0017] The medical record text in the medical record data is input into a pre-trained language model to generate text vectors;

[0018] The text vectors are processed using a sequence labeling model to extract the entities and relations;

[0019] Output the structured triplet.

[0020] Optionally, structured triples are obtained from medical record data, including:

[0021] A multi-head self-attention mechanism is introduced into the relation extraction model to calculate the attention weights between characters and generate attention information;

[0022] Based on the attention information, semantic relationships are parsed to generate relational data;

[0023] The structured triples are generated based on the relational data.

[0024] Optionally, the efficacy weights based on the efficacy quantification score are calculated, including:

[0025] The efficacy description is quantified into a score to generate quantitative efficacy data;

[0026] The average score corresponding to the quantitative efficacy data is calculated to generate the efficacy weight.

[0027] Optionally, meta-path analysis is performed, which includes defining meta-paths and traversing the meta-paths, including:

[0028] Define the syndrome-prescription-operation-traditional Chinese medicine pathway as a meta-path to generate path definitions;

[0029] The metapath is traversed to statistically analyze the TCM clinical rule pattern to generate rule statistics results.

[0030] Optionally, based on the results of the meta-path analysis, quantified TCM clinical rules are generated, including:

[0031] Calculate the semantic similarity of syndromes to generate similarity data;

[0032] The confidence score is generated by combining efficacy weights and frequency weights using a confidence formula.

[0033] Treatment plans are generated based on the confidence scores, serving as the quantified TCM clinical rules.

[0034] Optionally, structured triples are obtained from medical record data, including:

[0035] The identified entities are linked to a standard glossary to generate standardized entities;

[0036] The entities and relations are extracted from the standardized entities to generate the structured triples.

[0037] Optionally, calculate expert weights based on expert type, including:

[0038] Different weight values ​​are assigned to expert types to generate expert weight data;

[0039] The expert weights are calculated based on the expert weight data.

[0040] Optionally, based on the results of the meta-path analysis, quantified TCM clinical rules are generated, including:

[0041] The new patient's symptoms are mapped onto the heterogeneous graph to generate mapping results;

[0042] Based on the mapping results, retrieve associated treatment plans to generate a plan list;

[0043] The quantified TCM clinical rules are generated based on the list of schemes.

[0044] The beneficial effects of this application are:

[0045] Invention Point 1: Multi-dimensional weighted heterogeneous graph construction and weight allocation mechanism

[0046] Invention Point 2: A Quantitative Clinical Rule Generation Method Based on Meta-Path Analysis

[0047] This application provides a method for optimizing TCM treatment plans for knee osteoarthritis based on big data analysis, comprising: obtaining structured triples from medical case data, wherein the structured triples include entities and relationships extracted from unstructured knee osteoarthritis medical case texts; constructing a heterogeneous graph based on the structured triples, wherein the heterogeneous graph includes symptom nodes, syndrome nodes, TCM nodes, prescription nodes, and efficacy nodes, as well as edge relationships connecting the nodes, wherein the edge relationships include treatment relationships, inclusion relationships, and operation relationships; assigning weights to the edges of the heterogeneous graph, wherein the weight assignment includes calculating frequency weights based on the frequency of edge relationship occurrences, calculating efficacy weights based on efficacy quantification scores, and calculating expert weights based on expert types; performing meta-path analysis based on the heterogeneous graph and the weights, wherein the meta-path analysis includes defining meta-paths and traversing the meta-paths to statistically analyze TCM clinical rule patterns; and generating quantified TCM clinical rules based on the results of the meta-path analysis. This application achieves quantitative mining of TCM clinical rules by extracting structured triples from unstructured medical case texts, constructing weighted heterogeneous graphs, and performing meta-path analysis. This overcomes the shortcomings of existing technologies that cannot parse implicit knowledge and quantify deep rules. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the AI-based medical respiratory system diagnostic assistance process in this application. Detailed Implementation

[0049] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is to be understood that various forms of implementation of the present disclosure are intended and should not be limited to the embodiments set forth herein. Rather, the embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0050] This application uses rule mining based on multi-dimensional weighted heterogeneous graphs and meta-path analysis to quantify deep clinical rules (such as "addition and subtraction according to symptoms") from medical case data, thereby supporting the generation of personalized treatment plans.

[0051] The first step is heterogeneous graph construction, which involves using a graph database to construct a heterogeneous graph structure that includes node types such as symptoms, syndromes, Chinese medicine, prescriptions, and efficacy, as well as edge relationships (such as treatment and inclusion). This structure provides a relational representation framework for TCM knowledge.

[0052] Based on the construction of heterogeneous graphs, desired effects are achieved by assigning additional edge weights and performing meta-path analysis.

[0053] Specifically, edge weight assignment includes assigning weights to edges based on the frequency of relationship occurrence, efficacy quantization scores, and expert types, which introduce dimensions of clinical efficacy and authority; meta-path analysis defines and traverses specific paths (such as syndrome - formula - traditional Chinese medicine path) for statistical analysis of traditional Chinese medicine rule patterns.

[0054] Edge weight assignment gives the edges of the heterogeneous graph multi-dimensional clinical significance. Meta-path analysis uses weighted edges for path traversal, enabling the heterogeneous graph to quantify specific clinical rules of traditional Chinese medicine, such as identifying the "modification according to syndrome" rule with high probability.

[0055] Please refer to Figure 1 As shown, this application provides an optimization method for traditional Chinese medicine treatment plans for knee osteoarthritis based on big data analysis, which is applied to the field of traditional Chinese medicine. It is used to solve the problems of unstructured text understanding, insufficient depth mining of clinical rules, and scientific nature of plan recommendation in the optimization of traditional Chinese medicine treatment plans for knee osteoarthritis. The method includes:

[0056] S101. Obtain structured triples from medical record data, where the structured triples include entities and relationships extracted from unstructured knee osteoarthritis medical record texts.

[0057] Collect unstructured medical record texts of knee osteoarthritis patients from hospital information systems and electronic medical records, such as "The patient has knee pain that worsens when exposed to cold, with a syndrome differentiation of liver and kidney deficiency, and is given modified Duhuo Jisheng Decoction", as well as structured information such as age and K-L grading.

[0058] Next, perform text cleaning on the unstructured medical record texts, using regular expressions to remove irrelevant characters and generate cleaned texts.

[0059] Then, use the word segmentation tool jieba combined with a custom traditional Chinese medicine dictionary to segment the cleaned texts and generate word segmentation results. After word segmentation, link the identified entities with the Thesaurus of Traditional Chinese Medicine Subject Headings to achieve entity standardization. For example, standardize "Radix Achyranthis Bidentatae" to "Achyranthes bidentata", and generate standardized entities.

[0060] After that, input the medical record texts into a pre-trained language model, specifically using the bert-base-chinese model, convert each character into a 768-dimensional vector, and generate text vectors. Use a sequence labeling model to process the text vectors, which includes an input layer, an encoding layer, an attention layer, and a label decoding layer.

[0061] In the encoding layer, input the output vectors of BERT into a bidirectional long short-term memory network to capture context information.

[0062] In the attention layer, the multi-head self-attention mechanism is introduced to calculate the attention weights of each character relative to other characters, thereby dynamically perceiving the context. For example, it can identify that "when getting cold" is the adverbial of "aggravation", and "aggravation" modifies "knee pain", generating attention information.

[0063] Based on the attention information, semantic relationships are parsed to generate relationship data.

[0064] In the label decoding layer, the conditional random field is used for sequence labeling to precisely define entity boundaries, such as B-Syndrome, I-Syndrome, and 0, and to judge relationships.

[0065] Finally, structured triples are output, such as extracting (knee pain, has the property of, getting cold and aggravating), (this medical record, is diagnosed as, liver and kidney deficiency), and (Duhuo Jisheng Decoction, performs the operation of, addition and subtraction) from the medical record.

[0066] The whole process accurately extracts entities and relationships from unstructured text, solving the problem of text understanding.

[0067] The regular expression used in text cleaning is:

[0068] [^\u4e00-\u9fa5a-zA-Z0-9,。;:()《》]

[0069] Among them, ^ means to match any character not within the square brackets; \u4e00-\u9fa5 means to match all Chinese characters (Unicode encoding range); a-zA-Z means to match all English letters (both uppercase and lowercase); 0-9 means to match all digits; "。", ":", ";", "()", "《》" mean to match specific Chinese punctuation marks.

[0070] The above formula is used to remove punctuation marks and special characters.

[0071] It is determined that the traditional Chinese medicine dictionary includes the names of prescriptions in Treatise on Febrile Diseases. Entity standardization is linked to Traditional Chinese Medicine Thesaurus (TCMMeSH). In the sequence labeling model, the multi-head self-attention mechanism adopts an 8-head self-attention mechanism. In relationship extraction, entity types include symptoms, properties, syndromes, prescriptions, operations, and traditional Chinese medicines, and edge relationship types include has_property, treats, contains, has_operation, added_herb, and has_effect, etc. For example, the triple (Duhuo Jisheng Decoction, has_operation, addition and subtraction) corresponds to the operation relationship.

[0072] S102. Based on the structured triples, construct a heterogeneous graph, which includes symptom nodes, syndrome nodes, traditional Chinese medicine nodes, prescription nodes, and efficacy nodes, as well as edge relationships connecting the nodes, including treatment relationships, inclusion relationships, and operation relationships.

[0073] Heterogeneous graphs are constructed using the graph database Neo4j, where node types include symptoms, syndromes, traditional Chinese medicines, prescriptions, and therapeutic effects, and edge relationships include has_property, treats, contains, and has_operation.

[0074] For example, from structured triples, entities are mapped to nodes and relations are mapped to edges. For example, the triple (knee pain, has property, aggravated by cold) corresponds to the has_property edge between the symptom node "knee pain" and the property node "aggravated by cold".

[0075] After the heterogeneous graph is constructed, a network representation of traditional Chinese medicine knowledge is formed, which facilitates subsequent analysis.

[0076] The complete types of edge relationships in heterogeneous graphs include has_property for symptom-property relationships, treats for syndrome-prescription relationships, contains for prescription-traditional Chinese medicine relationships, has_operation for prescription-operation relationships, added_herb for operation-traditional Chinese medicine relationships, and has_effect for prescription-efficacy relationships.

[0077] The complete classification of entity types includes symptoms such as knee pain and swelling, nature such as aggravation by cold and morning stiffness, syndromes such as liver and kidney deficiency and cold-dampness obstruction, prescriptions such as Du Huo Ji Sheng Tang, manipulations such as additions and subtractions, and Chinese medicines such as Coix seed and Achyranthes bidentata.

[0078] S103. Assign weights to the edges of the heterogeneous graph. The weight assignment includes calculating frequency weights based on the frequency of edge relationships, efficacy weights based on efficacy quantification scores, and expert weights based on expert types.

[0079] When calculating frequency weights, the number of times each edge relationship appears in the data is counted, and then normalization is performed using the maximum and minimum values ​​to generate the frequency weights.

[0080]

[0081] When calculating efficacy weights, efficacy descriptions such as "significantly effective," "effective," and "ineffective" are quantified to 5, 3, and 1 points, respectively, generating quantified efficacy data. Then, the average score of the efficacy corresponding to the edge relationship is calculated to generate efficacy weights.

[0082]

[0083] When calculating expert weights, weight values ​​are assigned according to expert type. The relationship weight in the medical records of renowned traditional Chinese medicine practitioners is 1.5, while the relationship weight in the medical records of ordinary physicians is 1.0. Expert weight data is generated, and expert weights are calculated based on this data.

[0084] The weighting process takes into account frequency, efficacy, and expert knowledge, giving the edges of the heterogeneous graph a multidimensional measure of importance.

[0085] The efficacy quantification rules are: 5 points for significant effect, 3 points for effective effect, and 1 point for ineffective effect. The expert weighting rules are: 1.5 points for cases by renowned veteran TCM doctors and 1.0 points for cases by ordinary physicians. The frequency weighting formula is: [formula missing], where [value missing] represents the number of times the edge relationship occurs.

[0086] S104. Based on the heterogeneous graph and the weights, perform meta-path analysis, which includes defining meta-paths and traversing the meta-paths to statistically analyze TCM clinical rule patterns.

[0087] Define metapaths, such as syndrome-prescription-operation-traditional Chinese medicine path, as metapaths to generate path definitions.

[0088] Then, the metapath is traversed to statistically analyze the TCM clinical rule patterns. For example, by using the path Z--treats-->F--has_operation-->"addition and subtraction"--added_herb-->H, the "addition and subtraction according to symptoms" rule is statistically analyzed, and the rule statistics results are generated.

[0089] Metapath analysis transforms the conventional functions of heterogeneous graphs into quantitative clinical rule mining, enabling deep knowledge discovery.

[0090] The unique meta-path identified in the meta-path analysis is the syndrome-prescription-operation-traditional Chinese medicine path.

[0091] When traversing the metapath, the system starts from each syndrome node in the heterogeneous graph, searches for prescription nodes connected by treatment edges, then for each prescription node, searches for operation nodes connected by operation edges, and finally for each operation node, searches for Chinese medicine nodes connected by adding Chinese medicine edges, and counts the frequency of such paths to quantify clinical rules.

[0092] For example, during the traversal, the system will identify the treatment relationship from syndrome nodes to prescription nodes, the operation relationship from prescription nodes to operation nodes, and the addition relationship from operation nodes to Chinese medicine nodes, and discover rule patterns such as "addition and subtraction according to syndrome" by counting path instances.

[0093] S105. Based on the results of the meta-path analysis, generate quantified TCM clinical rules.

[0094] For new patients, their symptoms are mapped onto a heterogeneous graph, generating a mapping result. Based on the mapping result, all associated treatment options are retrieved, and a list of options is generated.

[0095] Then, the semantic similarity of the syndromes is calculated. The Wu & Palmer algorithm is used to calculate the semantic similarity of the two syndromes in the ontology, generating similarity data. Combining efficacy weights and frequency weights, confidence values ​​are generated using the confidence formula:

[0096]

[0097] Where Z represents the current symptom and T represents the treatment plan. To identify similar symptoms to symptom Z that can be reached within two steps on the graph, The semantic similarity between the two symptoms is calculated based on the Wu & Palmer algorithm. For effectiveness weight, is the frequency weight, and 'a' is a hyperparameter, which is set to 0.45 after grid search optimization.

[0098] Based on the confidence score, the schemes are sorted in descending order, and the top 3 schemes and corresponding suggestions for modification are output as quantified TCM clinical rules.

[0099] For example, it was found that "when using Du Huo Ji Sheng Tang for liver and kidney deficiency syndrome, if there is also dampness, there is a high probability of adding coix seed." This process integrates semantic similarity, efficacy evidence, and frequency to generate high-confidence solutions, thereby improving the personalization and scientific rigor of treatment plans.

[0100] The formula aims to calculate the reliability of scheme T for syndrome Z by comprehensively considering the similarity of symptoms, evidence of efficacy, and frequency of occurrence, through weighted summation and logarithmic amplification. This process not only integrates the objectivity of big data statistics but also captures the flexibility of TCM syndrome differentiation through semantic similarity, thereby enhancing the scientific rigor and relevance of the scheme.

[0101] The above description of the embodiments is provided to enable those skilled in the art to understand and apply this application. Those skilled in the art will readily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without inventive effort. Therefore, this application is not limited to the above embodiments, and any improvements and modifications made to this application based on the disclosure thereof should be within the scope of protection of this application.

Claims

1. A method for optimizing traditional Chinese medicine treatment plans for knee osteoarthritis based on big data analysis, characterized in that, include: Structured triples are obtained from medical case data, wherein the structured triples include entities and relations extracted from unstructured knee osteoarthritis medical case texts; Based on the structured triples, a heterogeneous graph is constructed, which includes symptom nodes, syndrome nodes, traditional Chinese medicine nodes, prescription nodes, and efficacy nodes, as well as edge relationships connecting the nodes, including treatment relationships, inclusion relationships, and operation relationships. Assign weights to the edges of the heterogeneous graph, the weight assignment including calculating frequency weights based on the frequency of edge relationships, calculating efficacy weights based on efficacy quantification scores, and calculating expert weights based on expert types; Based on the heterogeneous graph and the weights, meta-path analysis is performed, which includes defining meta-paths and traversing the meta-paths to statistically analyze TCM clinical rule patterns. Based on the results of the meta-path analysis, quantified TCM clinical rules are generated.

2. The method according to claim 1, characterized in that, Structured triples are obtained from medical record data, including: The unstructured medical record text in the medical record data is cleaned to generate cleaned text; The cleaned text was segmented using a word segmentation tool combined with a custom TCM dictionary to generate word segmentation results. The entities and relations are extracted based on the word segmentation results to generate the structured triples.

3. The method according to claim 1, characterized in that, Structured triples are obtained from medical record data, including: The medical record text in the medical record data is input into a pre-trained language model to generate text vectors; The text vectors are processed using a sequence labeling model to extract the entities and relations; Output the structured triplet.

4. The method according to claim 1, characterized in that, Structured triples are obtained from medical record data, including: A multi-head self-attention mechanism is introduced into the relation extraction model to calculate the attention weights between characters and generate attention information; Based on the attention information, semantic relationships are parsed to generate relational data; The structured triples are generated based on the relational data.

5. The method according to claim 1, characterized in that, Calculating efficacy weights based on efficacy quantification scores includes: The efficacy description is quantified into a score to generate quantitative efficacy data; The average score corresponding to the quantitative efficacy data is calculated to generate the efficacy weight.

6. The method according to claim 1, characterized in that, Performing meta-path analysis, which includes defining meta-paths and traversing the meta-paths, includes: Define the syndrome-prescription-operation-traditional Chinese medicine pathway as a meta-path to generate path definitions; The metapath is traversed to statistically analyze the TCM clinical rule patterns to generate rule statistics results.

7. The method according to claim 1, characterized in that, Based on the results of the meta-path analysis, quantified TCM clinical rules are generated, including: Calculate the semantic similarity of syndromes to generate similarity data; A confidence score is generated by combining efficacy weights and frequency weights using a confidence formula. Treatment plans are generated based on the confidence scores, serving as the quantified TCM clinical rules.

8. The method according to claim 1, characterized in that, Structured triples are obtained from medical record data, including: The identified entities are linked to a standard glossary to generate standardized entities; The entities and relations are extracted from the standardized entities to generate the structured triples.

9. The method according to claim 1, characterized in that, Calculating expert weights based on expert type includes: Different weight values ​​are assigned to expert types to generate expert weight data; The expert weights are calculated based on the expert weight data.

10. The method according to claim 1, characterized in that, Based on the results of the meta-path analysis, quantified TCM clinical rules are generated, including: The new patient's symptoms are mapped onto the heterogeneous graph to generate mapping results; Based on the mapping results, retrieve associated treatment plans to generate a plan list; The quantified TCM clinical rules are generated based on the list of schemes.

Images