Unified Deep Learning Scheme for Dental Caries Detection and Tooth Segmentation from Pre-Processed Digital Images

Abstract

Ensuring proper dental hygiene is crucial for an individual’s overall health and wellness. Conventional clinical assessments for the oral-health (OH) need a prescribed clinical protocol including the examination of the tooth by an experienced dentist. Recently, a considerable number of automatic methods are developed to examine the tooth and its condition using the computer algorithms. This research aims to propose a Unified Deep-Learning (UDL) tool for the tooth and caries segmentation from the digital tooth photographs collected from the real patients. The stages in this tool consist; (i) collection, pre-processing and resizing of images, (ii) UDL-scheme for tooth and caries segmentation, and (iii) performance evaluation and validation using a comparison with similar existing methods. The UDL-scheme consist the MIDNet18 as the backbone along with the Region Proposal Network (RPN) and ResNet50 for effective detection and segmentation. The performance of the developed scheme is then compared against the other similar segmentation procedures found in the literature and its clinical significance is confirmed with the clinically collected images with and without the tooth abnormalities. The experimental outcome of this study confirms that the proposed UDL-scheme is efficient in detecting the caries and the tooth section with better accuracy.