Construction Of TCM Constitution Identification Model Based On Tongue,Face And Inquiry Information Fusion

Object:To explore the change characteristics of tongue diagnosis and face diagnosis image feature information of different constitution groups;Analyze the correlation between tongue,face,inquiry characteristic information and different constitutions;The TCM constitution identification model integrating tongue,face and inquiry information was constructed,and the characteristic variables of tongue,face and inquiry information with high contribution to the model were screened.It aims to realize the intelligence and convenience of TCM constitution identification,and provide objective and quantitative evaluation methods for TCM constitution identification.Methods:1.Using the portable intelligent diagnostic apparatus of traditional Chinese medicine independently developed by Shanghai University of Traditional Chinese Medicine to collect tongue diagnosis,face diagnosis images and inquiry information of different physical groups,and then using the intelligent diagnostic system software of traditional Chinese medicine to analyze the characteristic parameters of tongue and face diagnosis images.2.Using one-way ANOVA to analyze the differences of tongue and face diagnosis image characteristic parameters among different physical groups,to explore the changes of tongue and face diagnosis characteristic information among different physical groups.3.The Apriori algorithm of association rules is used to construct the association rule model between different constitutions and tongue,face and inquiry feature information,and the correlation between different constitutions and tongue,face and inquiry feature information is analyzed.4.The extreme gradient boosting(XG Boost)machine learning algorithm is used to construct seven types of TCM constitution identification models,including qi deficiency,yin deficiency,yang deficiency,phlegm dampness,damp heat,qi stagnation and gentleness,which integrate tongue,face and inquiry characteristic information.The prediction accuracy of the model is given,and the tongue,face and inquiry characteristic variables with high contribution to the model are screened.The prediction accuracy of the model is compared with the Decision Tree and Linear Regression method.Results:1.There were differences in the characteristic parameters of tongue and facial images among people with different constitutions(P<0.05):(1)Comparison of facial color parameters:Compared with each group of L values of facial color,yang deficiency was the highest(P<0.05),and qi deficiency was the lowest(P<0.05),and the trend was yang deficiency>yin deficiency>qi deficiency>peace>phlegm-dampness>dampness-heat>qi stagnation;A value of each group,yin deficiency was the highest(P<0.05),the trend was:yin deficiency>phlegm dampness>yang deficiency>gentleness>qi stagnation>qi defici-ency>damp heat.Compared with each group of b value of facial color,the damp-heat quality was the highest(P<0.05),and the yin deficiency quality was the lowest(P<0.05).The trend was as follows:damp-heat quality>qi stagnation quality>calm quality>phlegm-dampness quality>qi deficiency quality>yang deficiency quality>yin deficien-cy quality.(2)Comparison of lip color parameters:The L value of lip color in each group was the lowest(P<0.05),and the trend was yang deficiency>qi deficiency>yin deficiency>gentleness>damp heat>phlegm dampness>qi stagnation.The lip color a value of each group was the highest in yin deficiency constitution(P<0.05),and the trend was yin deficiency constitution>yang deficiency constitution>phlegm dampness constitution>qi stagnation constitution>qi deficiency constitution>peace constitutio-n>damp heat constitution;The b value of lip color in each group was the highest in damp-heat constitution(P<0.05),and the lowest in yin deficiency constitution(P<0.05).The trend was damp-heat constitution>qi stagnation constitution>phlegm-dampness constitution>yang deficiency constitution>qi deficiency constitution>calm constitution>yin deficiency constitution.(3)Comparison of facial gloss index: Compared with the mild constitution,the gloss index of yin deficiency,yang deficiency and phlegm-dampness constitution decreased(P<0.05),and the trend was:mild constitution>damp-heat constitution>qi deficiency constitution>qi stagnation constitution>phlegm-dampness constitution>yang deficiency constitution>yin deficiency constitution;The glossless index of qi deficiency,yin deficiency,phlegm dampness,dampness heat and qi stagnation increased(P<0.05),and the trend was as follows:dampness heat> phlegm dampness>qi stagnation>qi deficiency>yin deficiency>yang deficiency>peace.(4)Comparison of tongue color parameters:Comparison of tongue color L value in each group,qi deficiency was the highest(P<0.05),qi stagnation was the lowest(P<0.05),the trend was:qi deficiency>yang deficiency>yin deficiency>damp heat>gentleness>phlegm dampness>qi stagnation.The a value of tongue color in each group was the highest in yin deficiency constitution(P<0.05),and the lowest in qi deficiency constitution(P<0.05).The trend was yin deficiency constitution>qi stagnation constitut-ion>phlegm dampness constitution>yang deficiency constitution>peace constitution>damp heat constitution>qi deficiency constitution;Compared with each group,the b value of tongue color in qi deficiency group was the highest(P<0.05),and the trend was:qi deficiency>gentleness>yang deficiency>phlegm dampness>damp heat>yin deficiency>qi stagnation.(5)Comparison of moss color parameters:Compared with each group of moss color L value,qi deficiency was the highest(P<0.05),and qi stagnation was the lowest(P<0.05).The trend was:qi deficiency>yin deficiency>yang deficiency>gentlene-ss>damp heat>phlegm dampness>qi stagnation;Compared with each group with a value of moss color,yin deficiency was the highest(P<0.05),and qi deficiency was the lowest(P<0.05).The high and low trends were as follows:yin deficiency>qi stagna-tion>phlegm-dampness>yang deficien-cy>peace>dampness-heat>qi deficiency.The b value of moss color in each group was the highest in damp-heat constitution(P<0.05).The trend was:damp-heat constitution>qi stagnation constitution>yang defici-ency constitution>calm constitution>qi deficiency constitution>phlegm-dampness constitution>yin deficiency constitution.2.Association rules analysis results,according to the size of the correlation index to enhance the order :(1)The correlation between different constitutions and tongue,face,inquiry single characteristics:Qi deficiency constitution is mainly associated with white face(3.401),pale tongue(2.524),abdominal distension(1.523),fatigue(1.422);Yin deficiency mainly has strong correlation with red tongue(7.359),tongue crack(5.785),lip red(4.301),hand-foot heart heat(3.853).Yang deficiency mainly has strong correlation with urine length(3.487),fear of cold or cold hands and feet(1.603),pale tongue color(1.603),pale lip color(1.548).Phlegm-dampness was mainly correlated with fat tongue(3.08),thick fur(2.412),sticky mouth(1.633),sticky stool(1.398),etc.The damp-heat constitution is strongly correlated with lipstick red(4.552),face yellow(3.023),moss yellow(2.824),fat tongue(2.573)and tongue red(2.519).Qi depression is mainly associated with chest fullness(3.481),pale tongue(2.759),annoyance or depres-sion(2.379);Peaceful quality is mainly associated with pale red tongue(3.915),glossy face(2.708),fatigue(2.069),not afraid of cold and heat(1.854).(2)The correlation between different constitutions and the combination of tongue,face and inquiry:Qi deficiency constitution is mainly associated with white face+pale tongue+pale lips(4.981),fatigue and fatigue+weakness of waist,knees,legs and feet+pale mouth(2.748).Yin deficiency is mainly associated with facial red+tongue red+yellow fur(8.422),hand-foot heart heat+ night sweating+thirst(6.211);Yang deficiency is mainly associated with frequent nocturnal urination+active sweating+fear of cold or cold hands and feet(5.145),pale lips+pale tongue(2.227);Phlegm-dampness was mainly strongly correlated with glossy darkness+fat tongue+white fur(4.464),mouth stick +stool sticky+alopecia(1.965);There was a strong correlation between damp-heat constitution and lipstick+fat tongue+thick moss(9.278),sticky stool+yellow urine+bitter mouth(5.086).Qi stagnation mainly has strong correlation with chest wall fullness+insomnia+worry or depression(4.542),and face green+tongue darkness+thin fur(4.292);Temperament is mainly strongly correlated with glossy red tongue+glossy face+lip glossy red(7.278),glossy red tongue+glossy face+lip glossy red(3.527).3.Physical model construction: XG Boost algorithm、Decision Tree、Linear Regression method were used to construct a physical identification model integrating tongue,face and question information.The overall prediction accuracy of the model constructed by Xg Boost algorithm was the highest,up to 95.16 %.The classification and prediction accuracy of seven kinds of constitutions were:peace(98.99%),qi deficiency(92.03%),yin deficiency(91.74%),yang deficiency(94.28%),phlegm dampness(96.25%),damp heat(95.16%)and qi stagnation(97.67%).Screening out the characteristic variables which have important contribution to the classification and prediction of seven kinds of constitution,according to the order of importance: Qi deficiency:fatigue,pale tongue,active sweating,abdominal distension after eating,tongue color L value,white moss,moss color L value,white color,etc.Yin deficiency:hand-foot heart heat,night sweating,red tongue,A value of tongue color,dry mouth thirst,etc.Yang deficiency constitution:afraid of cold or cold hands and feet,L value of face color,pale tongue,L value of tongue color,loose stools,white face color,frequent nocturnal urine,clear urine.Phlegm wet constitution:thick moss,sticky stool,sticky mouth,L value of moss color,white moss.Damp heat:sticky stool,thick moss,yellow moss,bitter mouth,sticky mouth,tongue color A value,red tongue,etc.Depression:annoyance/depression,chest fullness,tongue L value,breast swelling,pale tongue,etc.Peaceful quality:thin moss,no fatigue and fatigue,normal sweating,pale red tongue,normal stool,not afraid of cold and heat,normal face,facial luster,etc.Conclusion:1.The parameters and characteristic distribution of tongue and facial images of people with different constitutions have certain changes,which provides a characteristic quantitative parameter basis for the identification of TCM constitutions.2.Through the data mining method of association rules,the tongue,facial features and inquiry symptoms with high correlation with constitution are screened out,which are in line with the clinical pathogenesis characteristics of different constitutions.It is found that the correlation strength of tongue,face and inquiry multi-feature information combination with constitution is significantly higher than that of single feature,which lays a foundation for the establishment of constitution identification model based on tongue,face and inquiry information fusion.3.The prediction effect of the physical identification model based on the combination of tongue + face + inquiry information is better than that of the combination of tongue +face consultation,tongue + inquiry and face + inquiry.On the basis of inquiry information,the fusion of tongue and face consultation information can significantly improve the prediction accuracy of the model.The traditional Chinese medicine constitution identification model based on the information fusion of tongue,face and inquiry constructed by XG Boost algorithm has high prediction accuracy,and realizes the intelligent and digital identification of seven constitutions.