A Study On Facial Color Diagnosis Of Hypertension And Liver-Fire Hyperactivity Syndrome Based On Deep Learning

ObjectiveClarify and organize the concept and connotation of "color" in TCM;search and comb the modern scholars’ objective research on "color" in TCM to discover the current status and shortcomings of modern scholars’ research on "color" in TCM;use"MTCNN" the Face Detection Algorithm to defines the position of the Color Department of hypertension and liver-fire hyperactivity syndrome,and uses the "Deep Learning-Lightweight Network Xception" algorithm in the field of computer vision to construct a classification model for the facial color diagnosis of hypertension and liver-fire hyperactivity syndrome.Method1.Research on the connotation of "color" in traditional Chinese medicineResearch on "color" in graphetics and philology.First,the literary meaning of color is interpreted;Second,the content of "color" in "The Yellow Emperor’s Canon"is retrieved,selected,and extracted,and its connotation is analyzed in accordance with the context.Third,Contrastive analysis of the interpretation of "color" by later Chinese medical doctors was performed;Third,clarify and sort out the meaning of"color" in traditional Chinese medicine;Finally,using the keywords of "objectification of color diagnosis","objection of observation diagnosis" and "color department" as keywords,a fuzzy search was performed on HowNet,Wanfang,and Weipu databases.Relevant documents are organized and sorted.2.Establish a databaseThe facial images of three groups of subjects,including hypertension and liver-fire hyperactivity syndrome,hypertension and non-liver-fire hyperactivity and health people,were collected through the "tongue,facial pulse information collection constitution recognition system",and the images were cleaned and preprocessed to establish a facial image database.3.Define the Color Department of hypertension and liver-fire hyperactivity syndromeAfter establishing a database of the collected information,based on observing the facial characteristics of three groups of people,combining human characteristics and computer automation characteristics,after exploring and verifying,MTCNN(Multi-task Cascaded Convolutional Networks)face detection algorithm was used.Based on the combination of disease and TCM Syndromes,a Color Department definition method for the syndrome of hypertension and liver-fire hyperactivity syndrome is proposed.4.Construction of classification model based on "lightweight network Xception"for facial color diagnosis of hypertension and liver-fire hyperactivity syndromeThe "Deep Learning Method-Lightweight Network Xception" algorithm in the field of computer vision is used to construct a construction of classification model for facial color diagnosis of hypertension and liver-fire hyperactivity syndrome.First,the Color Departments located by the MTCNN face detection algorithm are extracted and a sub-database is established;second,data expansion is performed to divide the training set and test set;then,the image data is normalized;and then model training;Finally,make classification tests.Result1.The original meaning of "color" in traditional Chinese medicine is the qi of face,and its connotation includes color,luster,color department,color trend,and facial expression.2.A total of 374 patients were studied,including 126 patients with liver-fire hyperactivity syndrome in the hypertensive disease group,130 patients without liver-fire hyperactivity syndrome in the hypertensive group and 118 patients in the health group.Based on the observation of three groups of facial features,combining the characteristics of human body and computer identification features,this article first used the MTCNN algorithm face detection technology,and proposed the Color Department of liver-fire hyperactivity syndrome in hypertension based on the combination of disease and TCM syndrome.The positioning of the Color Department adopts the natural marker positioning method,and the skin color patches proportional to the face are extracted with the left pupil directly below the left pupil and the horizontal junction of the left nasal wing as the origin.The size of the skin patches varies from person to person,which reflects the proportional unit of middle finger of the Chinese medicine and individualized thought of natural marker placement.The position is specifically:the left eye is translated 3/4 W to the left,the 1/2 h is moved down to be the starting position coordinate,and the left eye is moved 5/4 h to be the end position coordinate.3.Establish an sub-database for the three sets of images extracted by the MTCNN algorithm,expand the data,randomly cut and expand each skin block with different sizes,and then screen the skin block to remove skin blocks with insignificant features and randomly generate a training set and a test set according to a 7:3 ratio.At this time,the total number of trained skin block pictures is 4,835,and the total number of test skin block pictures is 2454;then normalize the experimental image data;then use"Deep Learning-Lightweight Network Xception" for model training and test data.In model training,a total of 4,835 skin block images were selected for training in the three sets of data A,B,and C,and 0,1,2 were used as labels to indicate the category to which the skin block image belongs.In the experiment,the basic parameters were set to:A total of 70 epochs(training parameter:1 epoch equals to training once with all samples in the training set),the first 10 times,set the batchsize to 64(training parameter:sampling the number of training samples in the training set for each training),set the learning rate For 1e-3(training parameters:supervise and guide the model to adjust the network weight),the next 60 times,set the batchsize to 32 and the learning rate to 1e-4;and fine-tune and train based on the ImageNet weight coefficient pre-trained model Classifier parameters to get the final training model.Finally test the data,perform forward propagation in the optimized neural network to obtain the network output,and compare the actual output with the label data value to determine the category to which the image belongs,and count the correctly classified images Data,calculate the correct rate.The final verification accuracy of the model is 85.24%.The accuracy rate of the first category of hypertension and liver-fire hyperactivity syndrome recognition rate is 80.79%,the second type of hypertension and the non-liver-fire hyperactivity syndrome recognition accuracy rate is 85.91%,and the third category of health people is 88.70%.It is worth mentioning that deep learning is different from the models established by ordinary neural networks.The model parameters are huge,hidden in the computer background,invisible.Therefore,it is also called a "black box" by deep learning systems.Conclusion1.The original meaning of "color" in Chinese medicine is facial complexion,and its connotation includes color,luster,color department,color trend,and facial expression.TCM’s understanding of "color" is a comprehensive dynamic understanding of human body parts,human body whole,and human psychology,which reflects the unique perspective of traditional Chinese medicine in understanding the laws of life movement and disease change based on the holistic,perpetual,and dialectical concepts.2.Based on the”MTCNN face detection algorithm",the color department location method for hypertension and liver-fire hyperactivity syndrome has the characteristics of fast and simple,automatic identification,automatic positioning,high accuracy,and strong individualization.It explores the combination of disease and TCM syndrome of color department location,which can be extended to other diseases and TCM syndromes.3.The "deep learning-lightweight network Xception" algorithm is used to construct a classification model for facial color diagnosis of liver-fire hyperactivity syndrome in hypertension,which has operability and high accuracy,and explores the path to objective diagnosis of liver-fire hypersensitivity syndrome in hypertension.4.The model constructed by "Deep Learning-Lightweight Network Xception"can reflect the idea of individualized diagnosis and treatment of traditional Chinese medicine,explore the application of artificial intelligence in the field of color diagnosis of traditional Chinese medicine,and can be promoted to other color diagnosis research of other diseases and syndromes and other aspects of traditional Chinese medicine research field.