Tooth Detection and Segmentation on Occlusal Surfaces in Early Mixed Dentition Using Deep Learning

Bruna Cristine Dias — 2026-05-27

Artificial intelligence, particularly Convolutional Neural Networks (CNNs), has shown great potential for the detection of oral health conditions, including dental caries. Objectives: To evaluate the performance of CNNs in the detection and segmentation of posterior teeth during the early phase of mixed dentition. Materials and Methods: A total of 945 images of posterior teeth were collected in a school-based setting. The dataset was divided into training and testing subsets. Three CNN architectures — YOLOv11, U-Net, and DeepLabv3 — were compared based on precision, recall, and their ability to detect partially erupted teeth. Results: YOLOv11 achieved a precision of 0.967 and a recall of 0.938, successfully identifying 96.1% of partially erupted permanent teeth. U-Net demonstrated a precision of 0.953 and a recall of 0.951, detecting 78.4% of partially erupted teeth, while DeepLabv3 achieved a precision of 0.963 and a recall of 0.944, detecting 76.5% of these teeth. Conclusions: All three CNNs demonstrated high accuracy in detecting and segmenting posterior teeth in children. However, YOLOv11 outperformed both U-Net and DeepLabv3 in detecting partially erupted teeth, highlighting its potential for use in studies involving early mixed dentition. Clinical Relevance: Identifying the most suitable CNN architecture for detecting caries in mixed dentition may help standardize diagnosis and reduce clinical time.

Journal of Digital Health and Biomedical Informatics

Tooth Detection and Segmentation on Occlusal Surfaces in Early Mixed Dentition Using Deep Learning