Neural Network-Based Detection of Dental Caries Using Roboflow and Jetson Nano

Abstract

This study developed a convolutional neural network for the automated detection of dental conditions—specifically caries, restorations, root canals, and prostheses—on radiographic images using the Roboflow platform and deployed on the NVIDIA Jetson Nano. A dataset of 1,933 dental X-ray images, provided by the Bright Smile clinic and expanded to 4,639 through data augmentation techniques, was used. The model was trained and tested in five iterative versions, gradually incorporating each dental condition class. Preprocessing techniques, including a 20% increase in image saturation and targeted augmentations, significantly improved detection performance. The final model achieved a mean Average Precision (mAP) of 97.7%, with notable improvement in the identification of dental caries—previously the most challenging class. These results demonstrate that optimized preprocessing combined with YOLO-based training in Roboflow and deployment on Jetson Nano constitutes an effective pipeline for real-time dental diagnostics.