| Background:Qi deficiency and blood stasis(QDBS)syndrome is a common clinical syndrome characterized by Qi deficiency,blood circulation weakness and blood stasis.It can appear in the occurrence and development of a variety of diseases,and can also be a common pathogenesis of different diseases or different stages of diseases.In the research and development of new TCM drugs with syndrome for the treatment of QDBS syndrome,the standardization,accuracy and consistency of syndrome diagnosis are the technical keys to the application and efficacy evaluation of new TCM drugs with syndrome.TCM syndrome differentiation depends on doctors’ personal experience,and there are big differences and poor repeatability in the diagnosis of syndrome made by different TCM doctors.AI can reduce the uncertainty caused by doctors’ personal experience through the objective operation of certain rules.Deep learning technology,especially convolutional neural network,has high accuracy in image recognition and has made many complex image recognition tasks close to automation.The basis of previous research confirmed that tongue image has important diagnostic significance and research value in the clinical diagnosis and treatment of QDBS syndrome,and occupies a great weight in the syndrome diagnosis of QDBS syndrome.Therefore,it is of great significance to explore the construction method of the diagnostic model of QDBS syndrome based on the characteristics of tongue image,and to establish a simple,fast and accurate intelligent auxiliary tool for TCM syndrome differentiation.Syndrome is the external manifestation of the internal pathological state of the body,which reflects the pathological essence of the human body at a certain stage in the.state of disease.Protein is the executor of the function of life activities.Its content,structure and function determine the functional state of the body.Proteomics infers the functional changes in the pathological state of the human body by analyzing the type,quantity and change law of protein expression.Syndrome and proteomics both have the characteristics of integrity and dynamics.Reflecting the essence of TCM syndrome through the overall change of protein is the intersection of TCM and Western medicine.At present,the research on QDBS syndrome is mostly based on disease,and the syndrome research is carried out under the control of disease.There are few studies on QDBS syndrome from the perspective of ’different diseases with the same syndrome’.Therefore,it is of great significance to study the biological basis of QDBS syndrome under the mode of ’different diseases with the same syndrome’,to explore the occurrence mechanism of syndrome.At the same time,it’s also the common intervention mechanism of new TCM drugs with syndrome for different diseases.Methods:The first study was based on three disease populations:the recovery period of ischemic stroke,stable angina pectoris of coronary heart disease and diabetic peripheral neuropathy.183 subjects were included.Syndrome diagnosis and tongue image collection were carried out every 14 days.Based on the diagnostic criteria of QDBS syndrome scientifically formulated by the team’s early study,QDBS syndrome and non QDBS syndrome were defined on the day when the subjects were included,14 days,28 days,42 days and 56 days after included.All subjects were defined as QDBS syndrome on the included day.After more than 14 days of TCM treatment,some subjects’ clinical manifestations of QDBS syndrome disappeared and were defined as non QDBS syndrome.A total of 1506 images,753 tongue and face images each,were collected during the 56 days follow-up period,including 499 images of QDBS syndrome and 254 images of non QDBS syndrome.After preprocessing the image data,the data set is established,and the training set,validation set and test set are randomly divided according to 6:2:2.The training set n=451,validation set n=151 and test set n=151.The model was trained by 4-fold cross validation method and early stop method.The data was analyzed by Spyder(Python 3.6)integrated development environment,and the image data was read by PyTorch 1.1.0 deep learning framework.Logistic regression,MLP neural network,simple convolution neural network,enhanced convolution+BN neural network and ResNet18 neural network structure are used as the basic framework of the diagnostic model.The model is trained with the training set.After a certain period of training,the validation set is used to preliminarily evaluate the effect of the model,adjust the model parameters,and compare different models.The test set is used to evaluate the final performance and generalization ability of the optimal model.It is significant to explore the method of constructing the diagnostic model of QDBS syndrome based on the characteristics of tongue image.The second study compared the protein expression differences between patients with QDBS syndrome in the recovery period of ischemic stroke,stable angina pectoris of coronary heart disease and diabetic peripheral neuropathy and healthy people.A total of 16 serum samples from patients and healthy people were collected.712 proteins were detected by proteomics technology,and 665 protein quantitative data with high reliability were obtained after screening.The protein with statistically significant changes(>1.2 times)in the expression level of the three disease groups was used as the differential expression protein.The common differential expression proteins of patients with QDBS syndrome of three different diseases were obtained through Venn analysis.The common differential expression proteins were screened and evaluated through GO database enrichment analysis,KEGG database enrichment analysis,protein interaction network analysis and CytoNCA centrality analysis.The common core node protein’s function and possible pathological mechanism of QDBS syndrome were discussed and explained,to explore the biological basis of QDBS syndrome under the guidance of ’different diseases with the same syndrome’ theory.Results:In the research on the construction of tongue diagnosis model of QDBS syndrome,the results show that the performance of convolutional neural network with deep learning is significantly better than that of logical regression and artificial neural network of traditional machine learning in tongue image data recognition and classification.Meanwhile,the performance of the diagnostic model can be significantly improved by BN and establishing the residual structure to improve the gradient explosion and gradient disappearance problems.The enhanced convolution+BN neural network model has the best model performance in the diagnosis of QDBS syndrome based on tongue image.It can construct the syndrome diagnosis model in the case of unbalanced sample categories better.The tongue diagnosis model of QDBS syndrome based on the enhanced convolution+BN neural network has a higher accuracy in the diagnosis of QDBS syndrome,a better ability to distinguish between QDBS syndrome and non QDBS syndrome,and a better generalization ability of the model.The average AUC value of the diagnostic model is 0.741,F1 Score is 0.825,and the total accuracy of diagnosis reaches 75.5%.It can be used as an intelligent auxiliary tool for syndrome differentiation of TCM.In the biological basic research of QDBS syndrome,PCA shows that the protein expression of patients with QDBS syndrome and healthy people represents a clear separation effect,and there is common protein expression in patients with QDBS syndrome of three different diseases.Cluster analysis of differential proteins shows that the consistency of each group is good,and the characteristics of protein expression are similar,while there are significant differences in protein expression between patients with QDBS syndrome and healthy people.Venn analysis obtained 31 common DEPs in patients with QDBS syndrome of three different diseases,including 18 up-regulated proteins and 13 down regulated proteins.After screening and evaluation,14 core node proteins associated with QDBS syndrome were obtained.Research analysis show that the pathological mechanism of QDBS syndrome may be closely related to glycolysis and energy metabolism disorders,apoptosis,immune and coagulation dysfunction,etc.Core node proteins include key enzymes in glycolysis and energy metabolism,important regulators of apoptosis,important regulators of immune and coagulation functions,etc.,suggesting that these proteins have the potential to be biomarkers of QDBS syndrome.They’re expected to become potential targets for the diagnosis and treatment of QDBS syndrome.Conclusions:The tongue image diagnosis model of QDBS syndrome based on deep learning technology is expected to become a practical diagnosis tool.The model is simple,fast and accurate.Meanwhile,the common differential expression protein of patients with QDBS syndrome of different diseases is the material basis of ’different diseases with the same syndrome’.The biological basis of QDBS syndrome is related to glycolysis and energy metabolism disorders,apoptosis,immune and coagulation dysfunction.The study developed and verified the diagnosis model of QDBS syndrome based on the tongue image for the first time’ confirming that the research method of applying deep learning to the diagnosis of TCM syndrome is practical,providing ideas and basis for the rapid diagnosis and screening of QDBS syndrome based on tongue image in clinical practice and scientific research.Meanwhile,the common pathological mechanism of QDBS syndrome of different diseases under the theory of ’different diseases with the same syndrome’ is interpreted from the molecular level,which proves that different diseases with similar clinical characteristics have common material basis,adds a molecular basis to the theory of ’different diseases with the same syndrome’,and verifies the feasibility of the ’syndrome dominating disease’ model in the research of new TCM drugs with syndromes. |
PDF Full Text Request