Diabetes Mellitus Diagnosis based on Tongue Images using Machine Learning

Abstract

Diabetes is a condition that impairs the body's ability to generate the hormone insulin, altering carbohydrate metabolism and raising blood glucose levels. Diabetes is defined by high blood sugar levels. Early diagnosis of such abnormalities extends the patient's life. High blood sugar is a defining characteristic of diabetes. All around the world, diabetes mellitus affects more than 45% of newborns, 65% of middle-aged persons, and 88% of elderly people. So, method to propose a non-invasive diagnosis of diabetes from a tongue picture using a convolutional neural network that predicts rapid findings with improved accuracy. The method is divided into three steps. In the initial step, images of the tongue are processed using standard image processing methods to extract two types of quantitative data, chromatic and textural metrics. By measuring the spatial variation in pixel intensities, such as roughness, smoothness, silkiness, or bumpiness, a texture analysis aims to measure the apparent consistency of a surface. Utilizing an edge detector and the region expanding approach, the second shape detection phase extracts the tongue's shape. Final step, to find acne and cracks, color intensity extraction is performed. In that color intensity extraction, it predicts white, yellow, grey, black, and red points. Modified five layer algorithm is used based on VGG16 algorithm is used in this technique. It contains three convolutional layers and two hidden layers. During the training module the convolutional neural network model then learns on its own using the input supplied by the incoming data. When the training model is finished, the validation is start and convolutional neural network model test by itself. After completion of validation module, the accuracy of a convolutional neural network model is generated. The accuracy level is 98% based on the prediction of healthy tongue and diabetic tongue.