Abstract

Oral pathologies, like dental cysts, present significant difficulties for timely diagnosis and treatment, affecting public health outcomes. This study proposes a novel deep-learning framework for automatically categorizing dental cysts in panoramic radiography pictures. The suggested model, DentViT-RCNN, incorporates Generative Adversarial Networks (GANs) for data augmentation, improving the variety and caliber of training data. A Mask R-CNN architecture is used for object identification and segmentation, with the segmentation mask providing pixel-level resolution and the bounding box defining the location of the cyst. To increase the precision and resilience of the diagnosis, the segmented cyst region is subsequently input into a Vision Transformer (ViT) for Classification. Grad-CAM is also applied to the ViT output to create attention heat maps, highlighting the model's emphasis areas and improving explainability for clinical application. The accuracy measure of the proposed model segmentation is 83.90%, 97.10%, 99%, and 99.99%. The accuracy measures of the classification model values are 92.40%, 95.20%, and 98%. By improving automated dental cyst diagnosis accuracy and interpretability, the suggested approach shows promise in offering dental practitioners practical decision help.