ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

Customized convolutional neural network to detect dental caries from radiovisiography(RVG) images

Journal: International Journal of Advanced Technology and Engineering Exploration (IJATEE) (Vol.9, No. 91)

Publication Date:

Authors : ; ;

Page : 827-838

Keywords : RVG images; CNN; Image processing; Deep learning; Image augmentation; Dental caries; Machine learning.;

Source : Downloadexternal Find it from : Google Scholarexternal

Abstract

Dental caries is amongst the most common tooth disease that people of all age groups face around the world. Detection of dental caries from dental x-ray images or radiovisiography (RVG) images in its initial stages is a challenging task. Deep learning is being used in almost all medical fields to predict or detect different diseases. This growing trend of deep learning algorithms like artificial neural network, back propagation neural network, convolution neural network (CNN) and recurrent convolution neural network (RCNN) was also applied into dentistry but the size of dataset is small. However, no standard large dataset is available for dental RVG images. Therefore, a dataset comprising 1336 samples has been prepared for the proposed work, and the dataset size has been augmented by 13582 using the image augmentation technique. In this paper, a customized convolution neural network (CCNN) was proposed that learns features automatically to classify the tooth is having dental caries or not from dental x-ray images. We've committed to old approaches like input data, data preparation, segmentation, feature extraction, model building, model training, model testing, model consequences, and model interpretation. Out of total 1336 images, 1104 images were used for training, 111 images were used for validation testing and 121 images were used for testing purpose. With a learning rate of 0.0001 and 100 epochs, a CCNN with six convolution layers achieves an average precision of 94.59 %, recall of 95.89 %, specificity of 91.66 %, f1-score of 94 %, and average testing accuracy of 94.2 %.

Last modified: 2022-08-08 17:46:24