A simplified dental caries segmentation using Half U-Net for a teledentistry system

Abstract

High-reliability diagnostic equipment efficiently supported by a computer-based diagnostics system. For instance, a computational approach establishes a model that can diagnose diseases. Artificial intelligence has been applied to aid in the field of medical imaging. Classification, prediction, and localisation of lesions or dental caries greatly minimise the load and difficulties for clinical practitioners. In this study, U-Net architectures are simplified to propose the feature reduction of the decoder layers. This simplification of U-Net architectures is utilised for segmented dental caries images. This paper simplified the U-Net decoder layers into the level of blocks Half-UNet () and Half-UNet (). The Half-UNet structural model surpasses the U-shaped structural model in terms of efficiency and segmentation capabilities. The simplification of the UNet architecture outperformed using Half-UNet 0.83% of the dice coefficient. The Half-UNet design is able to preserve model performance in segmenting actual images and ground truth against expected ground truth.