Dental fluorosis classification using multi-prototypes from fuzzy C-means clustering

Abstract

Dental fluorosis occurs in many parts of the world because of highly exposure to high concentration of fluoride in the teeth development stage. To help the health policy makers developing the prevention and treatment plans, a manual or automatic image-based dental fluorosis classification system is needed. In this paper, we develop an automatic dental fluorosis classification system using multi-prototypes derived from the fuzzy C-means clustering algorithm. The values from red, green, blue, hue, saturation, and intensity channels are utilized as features in the algorithm. We also set the dental fluorosis classification criteria from the amount of pixels belonging to each class. We found that the pixel correct classification rate is around 92% on the training data set and around 90% on the blind test data set when comparing the results with two experts. Three out of seven images in the training data set and eight out of fifteen images in the blind test data set are correctly classified into dental fluorosis classes.