Caries-segnet: multi-scale cascaded hybrid spatial channel attention encoder-decoder for semantic segmentation of dental caries.

Abstract

Objectives: Dental caries is a prevalent oral health issue around the world that leads to tooth aches, root canal infections, and even tooth extractions. Existing dental caries diagnosis models may misdiagnose the disorder and take more time to segment the caries. This research work aims to provide an in-depth analysis of spatial and channel attention mechanism techniques used for semantic segmentation in an encoder-decoder network. For effective performance, the research implements novel techniques to segment the dental caries accurately.

Methods: Deep Fully Connected Residual Block (DFCR) is designed to provide relevant features without the loss of significant information. A novel Hybrid Spatial Channel Attention (HSCA) module is developed for combining significant features with the help of multi-scale spatial features and cross-dimensional channel features.

Results: The proposed methodology performs better than other cutting-edge algorithms by achieving 96.63 % accuracy, 95.77 % dice score, 96.28 % Intersection over Union (IOU) score for the caries dataset, and 96.93 % accuracy, 95.21 % dice value, and 96.1 % IOU for the Tufts dental dataset.

Conclusions: The developed model facilitates in detection of cavities precisely at an earlier stage with the help of dental images. The semantic segmentation of dental caries provides accurate diagnosis by assisting medical professionals.