Classification of dental diseases using CNN and transfer learning

Abstract

Automated medical assistance system is in high demand with the advances in research in the machine learning area. In many such applications, availability of labeled medical dataset is a primary challenge and dataset of dental diseases is not an exception. An attempt towards accurate classification of dental diseases is addressed in this paper. Labeled dataset consisting of 251 Radio Visiography (RVG) x-ray images of 3 different classes is used for classification. Convolutional neural network (CNN) has become a most effective tool in machine learning which enables solving the problems like image recognition, segmentation, classification, etc., with high order of accuracy. It is found from literature that CNN performs well in natural image classification problems where large dataset is available. In this paper we experimented on the performance of CNN for diagnosis of small labeled dental dataset. In addition, transfer learning is used to improve the accuracy. Experimental results are presented for three different architectures of CNN. Overall accuracy achieved is very encouraging.