An intellectual caries segmentation and classification using modified optimization-assisted transformer denseUnet++ and ViT-based multiscale residual denseNet with GRU

Dental caries is also referred as tooth decay, which is a common teeth disorder for humans. Based on the statistics, most of the adults faced this dental caries issue. It could result in disfigurement, discomfort, and pain. The issue of dental caries can be avoided effortlessly by early detection. The implementation of a reliable approach for the detection and classification of this issue; thus it is a timely and efficient treatment. However, the manual treatments are complex to identify which influences the clinical experiment. The inadequate history details of the patient also affect the proper treatment. On the other hand, the deep learning model has become the optimal choice when considering the evaluation of medical images. However, the traditional computerized approaches still have several limitations such as more resource consumption, and computational burden. Hence in this work, a deep learning-assisted caries segmentation and classification model is developed to treat the caries at the early stage to avoid the tooth loss. The dental caries images are collected from the standard “Panoramic Dental Dataset, and MLUA Caries Image Dataset”. Those images are given to the segmentation phase. The image segmentation process increases the accuracy, and robustness and minimizes the processing time and cost. The caries segmentation is performed using the Adaptive Trans-Dense Unet++ (ATDUnet++). This network effectively segments the given images and supports for producing high-quality outcomes. Here, the images are resized and normalized, and the noise in the images is removed. The parameters from ATDUnet++ are optimized by employing the Modified Running City Game Optimizer (MRCGO). The segmented images are given as the input to the classification phase. The classification of dental caries is done by using the proposed Vision Transformers-based Multi-scale Residual DenseNet with Gated Recurrent Unit (ViT-MRDGRU). This network categorizes dental caries precisely to prevent tooth loss and assists in making timely decisions for treatment. Throughout the result analysis, the accuracy and precision rate of the developed model are 96.59% and 96.55%. The final outcomes of the suggested model are justified by comparing it with other classifiers and algorithms.