Bayesian-Optimized Caries Classification in Dental Radiographs Using Mask R-CNN

Abstract

In order to improve diagnostic speed and accuracy, dental image analysis has become a crucial topic of study in the field of dentistry, utilizing developments in computer vision and machine learning. Manual evaluation is frequently used in traditional dental diagnosis techniques, which can be laborious and prone to human mistake. Because dental diseases, especially caries (tooth decay), are becoming more complicated, automated solutions that offer trustworthy and impartial assessments are required. This study presents a special technique that effectively analyzes many dental images while placing great emphasis on the identification and classification of multiple types of caries. Using sophisticated deep learning techniques, the Mask Region-based Convolutional Neural Network (Mask R-CNN) is employed effectively, allowing for efficient isolation and accurate segmentation of individual teeth in multiple dental images. A variety of three different techniques after performing segmentation is also applied, to effectively extract many relevant features, including Histogram of Oriented Gradients (HOG), Gabor Filters, along with Extreme Features (X-Feat). Each feature extraction method catches special aspects of tooth shape, so the classification approach becomes more strong. The retrieved features are categorized with a Bayesian-Optimized Random Forest Classifier. This helps to increase the precision of identifying dental conditions. This all-all-embracing technique provides accurate and automatic evaluations for multiple dental issues and it simplifies customary dental diagnostics. The findings show that this method classifies dental diseases. It can classify caries and other dental diseases with a maximum accuracy of 98.08%.