Development of a Unified Deep-Learning Model for Dental Abnormality Examination Using Digital Photographs
Ramya Mohan, A. Rama, Deepak Nallaswamy
Abstract
Maintaining good Oral Health (OH) is important for an individual's general health. Clinical-level examination of the OH using recommended protocol is time consuming, and hence computerized algorithm-supported methods are commonly adopted in recent years. Owing to its increased accuracy, Deep Learning (DL) based OH testing has become a popular procedure in recent times. This study suggests a revolutionary Unified DL (UDL) technique that uses digital photos taken from actual patients to assess the state of the teeth. With a clinically meaningful level of accuracy, the suggested UDL model applies the DL scheme, for instance, segmentation, tooth recognition, and classification. Oral image collection, a novel UDL modelbased assessment, DL-segmentation to extract the tooth and the caries section, and performance verification using the selected image database are the phases of this instrument. MIDNet18 serves as the foundation for three distinct UDL-schemes that are proposed in this study. Three distinct UDL-models can be achieved by combining MIDNet18 with the Region Proposal Network (RPN) and ResNet101. Metrics such as F1-score and error are used to validate the effectiveness of the suggested tool, and the results of the experiment show that the UDLII-model performs superior to other models taken into consideration in this investigation.