Computer aided Diabetes Diagnosis using Textural Features of Saliva Crystallogram Images

Abstract

Diabetes mellitus (DM) is a major cause of morbidity and fatality across the world. DM is a chronic disease where the patient suffers from higher concentrations of glucose in blood over a persistent period. Diabetes is identified by analysis of features extracted from saliva crystallogram images. Student’s t-test yielded p-values of 0.02, 0.00 and 0.00 for Fractal Dimension, Shannon Entropy and Lacunarity respectively. Furthermore, performance of machine learning classification algorithms are compared. Classification algorithms such as Support Vector Machine, Linear Discriminant Analysis, Quadratic Discriminant Analysis, KNearest Neighbor, Decision Tree and Bagged Tree were analyzed based on their accuracy, precision, sensitivity, specificity and F1-score. Bagged Tree classifier outperformed other classifiers under study. It achieved an accuracy of 0.993, sensitivity of 0.983, F1-score of 0.993 and execution time of 2.70 sec. Whereas, KNN classifier has the lowest execution time of 0.824 sec.