Tongue Feature Classification for Computer Aided Tongue Analysis System

Tongue observation has been an integral part of traditional medicine across various cultures along with other non-invasive diagnosis techniques. Computerized algorithms can facilitate healthcare personnel to train and practice this traditional form of diagnosis with confidence. This paper presents DenseNet 121 architecture for classification of Prakrati of an individual i.e. Vatta, Pitta and Kapha. Three tongue features coating, cracks and red dots are considered for classification. Three-fold cross validation is implemented on the model; performance enhanced using bagging operation with Random Forest classifier along with soft majority voting. The proposed DenseNet121 model showed an accuracy of 72 %. Multiple tongue features can be associated with any diseased condition, as well as it could also be a combination of vatta, pitta and kapha imbalance responsible for any particular disease. In future work aim is to collect comprehensive data with smartphone cameras pertaining to all tongue features mentioned in traditional medicine necessary to be accounted for diagnosis and interpretation of prakrati to build an automated tongue analysis system.