| Objective: To design and implement the TCM knowledge graph auxiliary decision-making system,including database,data management,data processing,data visualization and other functional modules,to realize the process from questionnaire to auxiliary diagnosis to auxiliary diagnosis result map display,and to provide some dialectical reference for HD patients to choose appropriate TCM treatment methods.At the same time,it provides physicians with certain remote symptom information collection and clinical decision-making auxiliary functions.Method:1.Using the seven-step method of domain ontology of Stanford University School of Medicine to construct the knowledge graph pattern layer(ontology),through the BILSTM-CRF model [1],the unstructured case information in the previous literature of CNKI,Wanfang,Weipu and the previous cases of the case system of Shandong Provincial Hospital of Traditional Chinese Medicine was identified,and the recognition results were embedded in the ontology layer to construct the symptom-name-disease name knowledge map of hemodialysis patients,and the "TCM Diagnosis","TCM Clinical Diagnosis and Treatment Terminology Syndrome Part" and "Identification Dialectic" were used by D2 R technology The existing structured information in the data is cleansed and transformed,and the TCM syndrome-proof-name knowledge map is constructed,and the symptom-factor-name knowledge map of hemodialysis patients is obtained by the ontology fusion of the above two knowledge maps.2.Through map knowledge reasoning,obtain case factor information,and conduct frequency and correlation analysis of evidence syndrome.3.For the standard syndrome(symptoms and signs)and factors obtained in the previous step,refer to Professor Zhu Wenfeng’s double-layer frequency weight reduction algorithm,calculate the weight of each syndrome,and simplify the weight,use the syndrome as the line name,and the factor as the list name,and construct a preliminary HD patient syndrome-factor weight matrix.Combined with the general weights of syndrome-evidence in "Dialectics of Evidence",the syndrome-factor weight matrix of HD patients is expanded.Then,the two weight matrices of general and HD patients are added to obtain the final HD patient syndrome-factor weight matrix.4.Design the syndrome questionnaire,input the matrix row names related to the syndromes selected by the patient in the questionnaire,assign the subblocks of non-input rows in the matrix as an empty matrix,sum all the rows of the matrix by columns,sort the column values from largest to smallest,and output the column names greater than the diagnostic threshold,that is,the elements,which are the identification results,and input the obtained element group into the element-certificate diagnosis knowledge map,and the top three testimonials in the number of connecting nodes are output as the testimony results.5.Select a suitable application framework to combine the knowledge graph with the questionnaire analysis,so that the questionnaire analysis results are dynamically displayed on the front end after the knowledge reasoning of the knowledge graph,and the system is run on the server to achieve service provision.6.Take 92 random cases in the case database and observe the accuracy of name recommendation.Outcome:1.The literature and the unstructured medical case text of hemodialysis patients in the hospital case system,after analyzing the medical case data,summarized 7 entity types.Automatic named entity recognition is realized by BILSTM-CRF model,and the f1 value of the model is 0.92314 through entity labeling training.The identification results of manual inspection were embedded in the ontology layer to construct a knowledge map of symptom-name-disease name of hemodialysis patients,including 5009 symptom entities of 632 species,398 of 268 types of symptom entities,398 of 12 types of disease names,and 9922 triples.Through D2 R technology,the existing structured information in "TCM Diagnosis","TCM Clinical Diagnosis and Treatment Terminology Syndrome Part" and "Dialectic of Evidence Dialectics" was cleaned and transformed,and the knowledge graph of TCM syndrome-proof-name was constructed,with 632 symptom entities,53 evidential entities,523 symptom entities,96 treatment entities and 4920 triples.By the ontology fusion of the above two knowledge maps,the syndrome-factor-name knowledge map of hemodialysis patients was obtained.Among them,there are5641 kinds of 632 kinds of syndrome,921 kinds of 532 types of witness entities,1531 types of 53 types of witness entities,96 legal entities,and 16230 ternary groups.2.A total of 398 patients from the source of literature,medical records and disease system were counted on the frequency of evidence and syndrome,and Table 7 and Table 8 were obtained.The frequency of kidney and spleen were319 and 288,respectively,which were important disease certificates for hemodialysis patients,and the frequency of wet disease,qi deficiency,yang deficiency and blood stasis were 301,294,152,143,respectively,which were important disease certificates for hemodialysis patients.The frequency of symptoms of qi deficiency and spleen deficiency such as fatigue and long-term lack of food and hunger was the highest,301 and 216 times,respectively.After the correlation analysis of the syndrome-testimina-testimony,it was found that the symptoms related to the "kidney" of the testimony were relatively the highest in terms of lichen whiteness,backache,pale tongue,anuria,pulse sinking,fine pulse,low back pain,fatigue,edema,long-term craving,greasy tongue,and sleep insecurity.3.For the 198 standard syndromes(symptoms and signs)and 39 factors obtained,refer to Professor Zhu Wenfeng’s double-layer frequency weight reduction algorithm,calculate the weight of each syndrome,and simplify the weight,use the syndrome as the line name,and the witness as the column name,and construct a preliminary HD patient syndrome-element weight matrix.Based on the universal weights of the syndrome-proofs in "Dialectics of Evidence",a universal weight matrix of 632 syndromes and 53 elements is established.The standard of symptoms of HD patients refers to the "Dialectics of Evidence",and the 198 syndromes after the specification belong to the subset of 632 syndromes,and 39 syndromes also belong to the subset of 53 symptoms.The remaining 434 syndromes and the rows and columns of 13 factors were assigned as null matrices,and the syndrome-factor weight matrix of HD patients was extended to 632 rows and 53 columns.Then,the two weight matrices of general and HD patients are added to obtain the final HD patient syndrome-factor weight matrix.4.Combined with the classification of dialectical syndrome to comprehensively collect patient information,design a questionnaire,divided into 20 questions,each question after a number of options,for multiple choices,if there is no abnormality can not be selected,the symptoms selected by the patient in the questionnaire are summarized,in the background HD patient syndrome-factor weight matrix in the relevant matrix row name assignment,the subblock of the non-input row in the matrix is assigned as an empty matrix,all the rows of the matrix are summed by columns,the column values are summed from largest to smallest,and the column names are output greater than the diagnostic threshold,that is,the elements,According to Professor Zhu Wenfeng’s three-level classification of elements,the weight of 14 is used as the demarcation,the weight of the single element is more than 14,the name of the output element is lower than 14,and it is not counted,and the output element group is the diagnosis result of the testimony5.Using the open source web application framework Django+vue+d3v6 to realize the separation of front-end and back-end,the back-end operation of questionnaire data,so that the questionnaire analysis results are dynamically displayed on the front-end after the knowledge reasoning of the knowledge graph,and the system is run on Tencent Cloud Server to achieve service provision.6.Accuracy of auxiliary decision-making results: A total of 92 cases of hemodialysis patients were included in the test.Using the syndrome-challenge questionnaire combined with the testimony-testimonial knowledge graph name recommendation system,the auxiliary identification name recommendation was carried out,and the results were as follows: 36 cases were accurate,32 cases were basically accurate,24 cases were wrong,and the model accuracy rate was73.9%.Conclusion: This thesis constructs an auxiliary decision-making system of TCM knowledge graph,including database,data management,data processing,data visualization and other functional modules,so as to realize the process from questionnaire to auxiliary diagnosis to auxiliary diagnosis result map display,and the results have reference value. |
