Mask R-CNN (ResNet-50 vs ResNet-101): A Deep Learning Framework for Instance-Level Enamel Segmentation

Dental cavities constitute a major global health issue and must be diagnosed reliably to enable timely and effective treatment. The identification of dental caries at an early stage is essential, as lesions typically begin at the enamel surface and, over time, progress into the deeper tooth structures, including dentin and pulp. Advancements in dental imaging, combined with artificial intelligence-based methodologies, offer promising solutions for improving diagnostic accuracy and efficiency. Therefore, the present study evaluates the performance of Faster Region-based Convolutional Neural Network (Faster R-CNN) and Mask Region-based Convolutional Neural Network (Mask R-CNN) with ResNet-50 and ResNet-101 backbones for automatic enamel detection and segmentation. All models exhibited excellent detection performance, obtaining perfect Average Precision (AP) scores at IoU thresholds of 0.50 (AP50) and 0.75 (AP75). Faster R-CNN has achieved an AP of 95.92%, while both Mask R-CNN variants, ResNet-50 and ResNet-100, achieved near-perfect bounding box detection with an AP of approximately 99%. For segmentation, Mask R-CNN with a ResNet-50 backbone achieved an AP of 86.30%, whereas the deeper ResNet-101 backbone significantly improved segmentation performance, achieving an AP of 98.44%. These results demonstrate that the Mask R-CNN architecture surpasses Faster R-CNN in detection accuracy and provides superior segmentation performance. Overall, Mask R-CNN with a ResNet-101 backbone can be considered the most effective model for enamel detection and segmentation. Nevertheless, the proposed model should be improved and externally validated. This work can be further carried out to detect carious lesions in the enamel portion for early detection and treatment.