Diagnosis Of Diabetes By Tongue Analysis Using Image Processing

Abstract

The tongue is a major part of the human body to taste, speak and swallow food. The tongue portion is directly connected with our internal organ. If any problem happens in the internal organ, it reflects the effect through the tongue. The tongue center portion is connected with the stomach, pancreas. Side portions are connected with the liver. The tongue tip is connected with the heart, etc. In this research work, an efficient Decision Support System is used to detect diabetes based on the Characterization of tongue images using three soft computing methods. The proposed methods are namely Multiclass Support Vector Machine (MSVM), Sequential Learning Neural Network (SLNN) and Convolutional Neural Network with Long short time memory (CNN-LSTM) architecture. The binary classifier uses the hyper-plane, which is also called the decision boundary between two classes called a Multi-Class Support Vector Machine (MSVM). The SVM method determines the hyper plane in dividing two classes. The boundary is maximized between the hyper plane and the two classes. The samples nearest to the margin will be selected to determine the hyper plane called support vectors. Multiclass classification is also possible either by using one-to-one or one-to-many. The highest output function will be determined as the winning class. The characteristics of MSVM helped to speed up the training and testing of tongue images. The precision, sensitivity, specificity, accuracy and F1-score of MSVM based diabetic detection are 89.61%, 92.0%, 89.33%, 90.67% and 0.9179. Determining the various parameters associated with neural networks is not straight forward and finding the optimal configuration is a time and memory-consuming process. The Sequential Learning Neural Network (SLNN) algorithm is used to reduce time and memory. Since SLNN has a single hidden layer, the memory utilization will be less. Sequential learning is employed to reduce the memory space and also reduce thecomputation complexity.