A DEEP LEARNING FRAMEWORK INTEGRATING ENHANCED SWIN TRANSFORMER EMPLOYING WINDOW-BASED AND SHIFTED WINDOW MULTI-HEAD SELF-ATTENTION WITH RESIDUAL MLPS FOR ACCURATE DENTAL CARIES DETECTION

Abstract

Dental caries is widely recognized as a major oral health concern worldwide for affecting individuals of all age groups. Dental decay problem continues rapidly and emphasizing early diagnosis system and importance of making reliable clinical decisions. Early detection of dental caries enables timely treatment and prevent further deterioration of tooth structure to improves patient outcomes. In this study, an approach of deep learning framework which is based on Enhanced Swin transformer with Hybrid Shifted Window Attention and a Residual Multi-Layer Perceptron (MLP) is proposed for dental caries detection. This research model effectively captures the contextual features of dental images which is both local and global images by integrating hybrid attention mechanism with hierarchical features extraction. Furthermore, the enhanced features representation was introduced by using residual MLP module which is having better stability.

This research proposes the novel hybrid model for automated dental caries detection in orthodontics treatment by using the conventional image processing with the combination of deep learning architecture. The approach is based on the local features and Swin transformer for capturing local and global features extraction and early diagnosis to detect the caries. The proposed system was evaluated better effectiveness assessed using common metrics such as accuracy, precision, recall and F1 score by implementing ten-fold cross validation technique. The model results were demonstrated effectively and the outcomes were evaluated against conventional deep learning architecture. And this study describes the state-of-the-art methods which is reported in the literature that indicates that the proposed approach achieves the better diagnostic performance, attaining a maximum accuracy of 95.36%, automated dental caries detection in clinical environment and also highlighting its potential for reliability. This study uses the dataset of 1272 dental radiographic images which includes panoramic images, intraoral images and bitewing images also. The dataset is preprocessed to improve the performance metrics. The findings also indicated that the performance of the local and global features extraction and also contribute towards the advancement of AI based dental diagnosis from intelligent diagnosis system.