ABSTRACT
Objective To propose a classification method for small sample tongue images based on transfer learning and fully connected neural network, so as to solve the problems of large amount of data, high requirement of training equipment and long training time of deep learning in the classification of tongue images. Methods Effective features such as tongue points and lines of tongue images were extracted by the convolution Inception_v3 network after training on the massive data set of ImageNet. The above features were classified by the fully connected neural network, and the image knowledge acquired by the deep learning network was transferred to the tongue image recognition task, and then the tongue data set were used to train and test the efficiency of the network. Results Compared with the typical tongue image classification method such as K-nearest neighbor (KNN) algorithm, support vector machine (SVM) algorithm and convolutional neural network (CNN) deep learning method, the two methods (Inception_v3＋2NN and Inception_v3＋3NN) in our experiment had higher classification rates for tongue images, with the accuracy rates being 90.30% and 93.98%, respectively, and had shorter training time for the sample. Conclusion Compared with KNN algorithm, SVM algorithm and CNN deep learning method, the tongue image classification method based on transfer learning and fully connected neural network can effectively improve the accuracy rate of tongue image classification and shorten the training time.
ABSTRACT
Objective To investigate the correlation between body mass index (BMI) and features of tongue manifestation.Methods A total of 1 655 physical examination individuals were selected,and height and body weight were measured to calculate BMI.Traditional Chinese Medicine tongue inspection was performed,and map of tongue manifestation was photographed by a digital camera.Results Among the 1 655 physical examination individuals,prevalences of normal weight,overweight/obesity and low weight were 50.94％,43.74％ and 5.32％,respectively.The prevalences of individuals had dark red tongue (37.98％ vs.24.56％;x2=40.883,P＜0.001),purplish tongue (5.94％ vs.3.44％;x2=8.133,P=0.017),dark purplish tongue (2.35％ vs.0.59％;x2=10.398,P=0.006),yellow fur (63.95％ vs.58.84％;x2=9.570,P=0.008),thick greasy fur (24.86％ vs.17.20％;x2=18.698,P＜0.001) in the overweight/obesity group were significantly higher than those in the normal weight group,and the prevalence of individuals had exfoliative fur/little fur/less fur in the overweight/obesity group was significantly higher than that in the normal weight group (1.93％ vs.3.91％;x2=6.454,P=0.040).The prevalences of individuals had red or deep red tongue (48.86％ vs.24.72％;x2=24.608,P＜0.001),red dot tongue (7.95％ vs.1.93％;x2=12.254,P=0.002),thin whitish fur (22.73％ vs.9.53％;x2=14.030,P=0.001) in the low weight group were significantly higher than those in the normal weight group.Binary logistic regression analysis showed positive correlation between dark red tongue (OR 1.830,95％ CI 1.515-2.211;P＜0.001),purplish tongue (OR 1.874,95％ CI 1.211-2.901;P=0.005),dark purplish tongue (OR 4.259,95％ CI 1.610-11.266;P=0.003),yellow fur (OR 1.294,95％ CI 1.093-1.531;P=0.003),gray-black fur (OR 6.147,95％ CI 1.816-20.811;P=0.004),thick greasy fur (OR 1.597,95％ CI 1.289-1.978;P＜0.001) and BMI,and negative correlation between pink tongue (OR 0.766,95％ CI 0.637-0.920;P=0.004),red or deep red tongue (OR 0.659,95％ CI 0.549-0.789;P＜0.001),red dot tongue (OR 0.548,95％ CI 0.326-0.921;P=0.023),thin whitish fur (OR 0.695,95％ CI 0.536-0.900;P=0.006),whitish fur (OR 0.821,95％ CI 0.683-0.986;P=0.035) and BMI.Conclusions There is some correlation between BMI and features of tongue manifestation.Abnormal tongue color is mainly dark red tongue in the overweight/obesity individuals,and red or deep red tongue in low weight individuals.Abnormal tongue fur is mainly thick yellow fur.The tongue shape is normal in different body weight groups.