The Study Of TCM Tongue Diagnosis Of Breast Cancer Recommendation Chinese Medicinal Technology Based On Knowledge Graph And Neural Network

ObjectiveBased on traditional Chinese medicine theories and knowledge,through cutting-edge computer artificial intelligence technology,explore the application of computer technology in the field of Chinese medicine tumor tongue diagnosis.This study uses knowledge graph technology,neural network technology,and implicit feedback collaborative filtering recommendation ideas,combined with the clinical logic of Chinese tongue-diagnosis and its role in tumors,reproduces the logic of tongue-diagnosis and learns the potential relationship between the tumor tongue picture and the drugs of Chinese medicine prescription.This paper gradually builds a model of Chinese medicine recommendation for breast cancer tongue diagnosis and realizes the recommendation of Chinese medicinal by simulating the prescription and medicine thinking of the doctors studied.Methods1.Research on TCM tongue diagnosis knowledge graph based on Knowledge Graph technology.Behind the thoughts of TCM syndrome differentiation and treatment is a systematic logic system,which includes everything from checking symptoms to prescribing prescriptions.Based on the role of the tongue in the process of TCM tumor differentiation and treatment,the knowledge graph technology is used to build a TCM tongue diagnosis knowledge graph,which based on tongue information through a top-down construction method.The research centers on the theory of TCM syndrome differentiation and treatment and designs the logic layer of the TCM tongue diagnosis knowledge graph from the four levels of TCM--theory,method,prescription and medicine.Relying on the entity type of the model layer,this research collects knowledge ontology from multi-source unstructured data such as TCM theoretical textbooks,national standards,and clinical outpatient records of Chinese medicine.After the steps of knowledge extraction,processing and knowledge fusion,the data layer of the TCM tongue diagnosis knowledge graph is formed.Finally,the entities and triples are imported into the Neo4j graph database through the Python programming language to realize the storage function of the TCM tongue diagnosis knowledge graph and use the visual display,graph query and path retrieval functions of the knowledge graph itself to verify the feasibility of it.On this basis,the MATCH query language of the knowledge graph is used to collect all the path information between the tongue image and the medicine.The TransH knowledge representation technology is used to process each entity and relationship in the graph to obtain its representation vector.It provides a path-vector data of the "tongue-medication" path for the construction of a breast cancer tongue diagnosis Chinese medicine recommendation model.2.Research on tongue image recognition model based on Convolutional Neural Network technology.In order to explore the potential relationship between the tumor tongue and traditional Chinese medicine,it is necessary to collect their respective characteristic representations.In terms of tongue image,according to the current research progress of convolutional neural network technology,the study chooses to use the MXNet deep learning network framework to build a tongue image recognition model with ResNet network structure as the main body of the network.The research refers to the three network structures of ResNet-18,ResNet-34 and ResNet-50 in the ResNet network to construct the tongue image recognition model.The tongue image data set used to train the model parameters was collected from the outpatient department of the Department of Traditional Chinese Medicine of Guangdong Provincial People’s Hospital.Select relatively complete picture of the front view of the tongue from the data set(not including the tongue of patients diagnosed with breast cancer).After cropping and artificial attribute marking,the original data set is scattered and randomly rearranged.Through the verification set method,the arranged data set is divided into a training set(samples occupy 80%of the data set)and a validation set(samples occupy 20%of the data set),and then the training set and validation set data are further cut and transformed and data enhancement processing.Based on the above data operation and processing,repeated experiments were carried out 10 times,with overall accuracy and overall Loss value as evaluation indicators,and the average of the highest overall accuracy of each of the 10 experiments and the highest overall accuracy of each model in a one time.These values are used as the standard to judge the comprehensive identification ability of the model and are used to evaluate the feasibility of the tongue image identification model based on the three model networks of ResNet-18,ResNet-34,and ResNet-50 and the tongue image attribute recognition result of each model.Through horizontal comparison to find the best model which has the best recognition result,as the basic structure of the tongue diagnosis identification model of this study.It provides tongue image attribute information and tongue image feature vector for the training of the breast cancer tongue diagnosis Chinese medicine recommendation model.3.Research on the recommendation model of Chinese medicine for breast cancer tongue diagnosis based on knowledge graph and neural network.Based on knowledge graph knowledge representation learning technology,neural network technology and implicit feedback collaborative filtering recommendation ideas,a tongue-like drug recommendation model is constructed,and the potential relationship between tumor tongue-like images and prescription drugs is explored and learned.In terms of acquiring and processing experimental data sets,this study uses tongue images of breast cancer patients in the Department of Traditional Chinese Medicine Cancer Outpatient Department of Guangdong Provincial People’s Hospital as the image data set.Together with the corresponding TCM prescription data,they serves as the original data set for training the drug recommendation model.For the processing of prescription data,based on the distinction of interactive data types in the collaborative filtering recommendation algorithm,combined with the characteristics of traditional Chinese medicine in formulating prescriptions and distributing medicines,the study processes the original prescription data attributable to Doctor A and Doctor B’s prescription data according to implicit feedback data.Tongue image data uses the tongue image recognition model for tongue image attribute recognition and image feature vector acquisition.In terms of model framework construction,Python language is used under the MXNet deep learning framework to build a Tongue-image Recommended Herb Model(TRHM)based on the idea of Neural Collaborative Filtering.The study uses the accuracy of the TOP-10 ranking results of the recommended results and the evaluation of the drugs recommended by the model by TCM experts(Doctor A,Doctor B)as evaluation indicators,which is used to verify the feasibility of the TRHM model based on neural network and knowledge graph technology for the task of recommending traditional Chinese medicine and to evaluate the recommendation performance of the model.Results1.Research on TCM tongue diagnosis knowledge graph based on Knowledge Graph technology.Based on the clinical logic of TCM tongue diagnosis,the research designed a total of 12 entity types for the logic layer,including Tongue,Tongue attribute,Tongue_part,Organ,Cause_disease,Symptom,Syndrome,Herb,Treatment,medicinal properties_four Qi(Property),medicinal properties_toxicity(Toxicity)and medicinal properties_wuwei(Taste).There are 12 kinds of semantic relationsm,it includes part relationship(Tongue_Tongue_part),reflection relationship(Tongue_part_Organ),disease position relationship(Organ_Syndrome),syndrome element relationship(Tongue_image_Cause_disease),disease relationship(Cause_disease_Syndrome),symptom relationship(Symptom_Syndrome),therapies relationship(Syndrome_Treatment),pesticide effect relationship(Treatment_Herb),medicine-property relation_Four Qi(Herb_Property),medicine-property relation_taste(Herb_Taste),medicine-property relation_Zangfu Guijing(Herb_Organ)and medicine-property relation_toxicity(Herb_Toxicity).After the steps of knowledge extraction,processing,and knowledge fusion,the data layer of the TCM tongue diagnosis knowledge graph contains 2,686 nodes under 12 types of labels,and 19,215 relations under 12 types of relations.The final TCM tongue diagnosis knowledge map relies on its own MATCH query language to realize functions such as map query and path retrieval,and uses MATCH query sentences to collect all the tongue attributes and the path between Chinese medicine([tongue attributes]->[Disease]->[Syndrome]->[Therapeutic Method]->[Chinese Medicinal]),a total of 9860 paths.TransH knowledge representation technology respectively carried out vectorized representation for 2686 nodes and 12 relation types.2.Research on tongue image recognition model based on Convolutional Neural Network technology.Combining the content of the tongue diagnosis of traditional Chinese medicine,the study design the tongue image classification label for the tongue image recognition model.According to the characteristics of the clinical tongue picture information collected,the tongue picture attributes are divided into 22 categories,which are divided into two major categories:tongue body and tongue fur.The tongue body is based on two types of attributes:tongue color and form of the tongue and the tongue fur is based on two types of attributes:color and texture.Specifically:pale tongue,pale red tongue,dark red tongue,red tongue,bluish-purple tongue,old tongue,teeth-marked tongue,spotted tongue,ecchymosis tongue,fissured tongue,white fur,yellow fur,gray and black fur,thin fur,thick fur,dry fur,moist fur,slippery fur,slimy fur,exfoliated fur,partial fur and whole fur.Through the experimental results,it is found that the training of the three model networks of ResNet-18,ResNet-34,and ResNet-50 quickly reached the fitting.In a further horizontal comparison analysis,it is found that the average values of the highest overall accuracy of the three models reached 86.77%,86.97%,and 87.13%,respectively,and they were able to identify tongue attributes well.There is a statistical difference in the recognition accuracy of the three models(P<0.01).The comprehensive recognition ability of the model can be sorted according to the numerical value to get ResNet-50>ResNet-34>ResNet-18,of which the ResNet-50 model has the best performance,with the highest overall accuracy rate reaching 87.19%.3.Research on the recommendation model of Chinese medicine for breast cancer tongue diagnosis based on knowledge graph and neural network.After adjusting the various hyperparameters,the TRHM model obtained the highest accuracy rates of 54.11%and 52.11%on the data sets of Doctor A and Doctor B,respectively.Further analysis revealed that the medications of the samples with low scores were significantly different from the original prescriptions,which was caused by the deviation of the input tongue attributes.For other scoring samples,the drugs recommended by them that are not on the original prescription can also be applied to the input tongue-image.In the evaluation of TCM experts(Doctor A,Doctor B)on the drugs recommended by the model,92.15%and 73.6%of the results of the two data sets A and B are considered to be most or even completely suitable for the corresponding tongue.Conclusion1.Research shows that it is feasible and effective to design and construct a knowledge map of TCM tongue diagnosis from the four levels of "theory,method,prescription and medicine" based on tongue diagnosis information.From tongue attribute information to traditional Chinese medicine(tongue attribute-disease nature-syndrome-treatment-Chinese medicinal),there is a complete path of speculation in TCM syndrome differentiation.This realizes functions such as visual display,graph query and path retrieval,and provides a visual platform for querying,analyzing and displaying the thinking logic prescriptions based on tongue images.Through the vectorized transformation of knowledge representation technology,the TCM tongue diagnosis knowledge graph contains the path information of TCM syndrome differentiation and treatment logic to provide the"tongue-drug" vector data for the breast cancer tongue diagnosis TCM recommendation model.2.The tongue image recognition model based on the ResNet-50 network structure has good recognition of the tongue image attributes,which can provide the attribute information and feature vector data about the tongue image for the construction of the breast cancer tongue diagnosis Chinese medicine recommendation model,at the same time,it improves the intelligence and automation of model.3.The breast cancer tongue diagnosis TCM recommendation model based on knowledge graph knowledge representation learning technology,neural network technology and implicit feedback collaborative filtering recommendation ideas can be effectively applied to the task of recommending TCM for breast cancer patients.TCM experts have a high degree of recognition for the drugs recommended by the model.4.The breast cancer tongue diagnosis Chinese medicine recommendation model contains the path data of the knowledge graph,which makes the knowledge graph have the function of expressing knowledge and logic in a graphical form,and becomes an entry point for the information of Inquiry,Palpation,etc.This will explore the potential connection between the tongue image and the medicine and futher improve the potential connection between the patient’s body information characteristics and the medicine,so as to better recommend the medicine for the patient.5.The prerequisite for digging and learning the potential relationship between two objects must be a large number of high-quality training data.The key to training a recommendation model is to provide it with accurate feature data,whose accuracy is reflected in quantity.Diversified description information helps the model to obtain characteristic data from it,so that the network can learn more effectively the non-linear interaction relationship between users and items,and ultimately obtain better recommendation results.6.Research on the modernization of Chinese medicine cannot pay attention to the theory of numbers only.Although the numbers are the most intuitive and reliable in statistics.But when computer science is applied to traditional Chinese medicine,the numbers do not reflect the real situation.In order to obtain real results,it is necessary to combine multiple factors to make a comprehensive judgment,such as the subjective evaluation of TCM professionals.7.The technology and research ideas used in this study provide methodological and technical references for the formulation of individualized diagnosis and treatment plans for tumor patients in traditional Chinese medicine,and also open up new ideas for the study of tumor tongue information.