Research On Correlation Between Tongue Image And Disease Based On Pattern Recognition

In the diagnosis of Chinese medicine,tongue diagnosis is an important part of it.Tongue diagnosis is an effective and non-invasive diagnosis method to determine the disease by observing the quality of the tongue,the shape and color of the tongue coating,etc.However,there is a certain degree of subjectivity in the diagnosis process.For example,Chinese medicine doctors have different ways of diagnosing patients,and different doctors have different diagnosis of the same patient.Therefore,objective indicators are needed to improve the accuracy of TCM diagnosis.With the development of computer science and technology,modern tongue diagnosis can achieve objective,quantitative and automated auxiliary analysis to help doctors make better diagnosis.The modernization of TCM tongue diagnosis is mainly divided into three research parts:tongue region segmentation,feature extraction,and disease analysis.Tongue body region segmentation is to obtain an accurate tongue image from the original tongue image using a segmentation algorithm.Then in the feature extraction link,the required TCM tongue features are extracted according to the TCM diagnosis method.Finally,the extracted features are used to classify and diagnose diseases.The research work of this paper is as follows(1)An improved tongue region segmentation network is proposed based on the U-net network:firstly,the tongue image is preprocessed and image enhanced,and then the data set is expanded through operations such as rotation,movement,and expansion to obtain more training use tongue image data;then,improve the traditional U-net deep learning neural network,design a training strategy to obtain a U-net segmentation network suitable for the tongue area,and realize the segmentation of the tongue area(2)Propose a multi-feature extraction method based on the separation of fur:for the extraction of tongue features,after the separation of the tongue,a number of different color space models are used to analyze the color of the tongue and the color of the tongue after the separation of the fur.The extraction is performed separately,and the texture feature of the tongue coating is extracted using the gray matrix at the same time.Finally,according to the diagnosis method of Chinese medicine on the tongue,the characteristics of cracks,fat and thin were extracted.(3)Propose a method for diagnosis and classification based on random forest classifier and SVM classifier:For the diagnosis and classification of diseases,the extracted features are first fused,so that the feature information of each field is complementary,and the identification ability of the method is improved;and then through the feature.The algorithm is selected for screening,and the feature with higher information content is found,and the classifier is trained based on this feature;the experiment verifies the feasibility and effectiveness of its diagnosis of diseasesThe experimental results of the above methods show that the classification accuracy(ACC)between diabetic patients and healthy people reaches 94.2%,and the area under curve(receiver operating characteristic curve,ROC curve),AUC)value reached 97.9%.The ACC and AUC between the two TCM gastric symptoms(weakness of the spleen and stomach and discordance of the liver and stomach)reached 64.4%and 69.4%,respectively.It is proved that the modernization of the diagnosis of tongue in Chinese medicine can provide a valuable reference for doctors’ diagnosis.