Caries detection using YOLOv3 on bitewing radiographs
Assessment of YOLOv3 for caries detection in bitewing radiographs based on the ICCMS™ radiographic scoring system
Authors: Wannakamon Panyarak, Wattanapong Suttapak, Kittichai Wantanajittikul, Arnon Charuakkra, Sangsom Prapayasatok
Summary:
This study evaluated the feasibility of YOLOv3 for automatic caries detection and ICCMS™-based radiographic classification on bitewing radiographs under two IoU thresholds (0.50 and 0.75). A total of 994 annotated images were used for training, 256 for validation, and 175 for testing. Performance was reported across binary (non-carious vs carious), 4-class (0/RA/RB/RC), and 7-class (0, RA1–RA3, RB4, RC5–RC6) settings. YOLOv3 achieved acceptable results at both IoU50 and IoU75; metrics decreased for the 7-class task. The model localized and classified advanced caries (RC6) well, but struggled with initial enamel lesions (RA1).
Check the original publication:
Key Words: YOLOv3, Bitewing radiograph, ICCMS, Caries detection, Object detection, AP, mAP, Precision, Recall
Extracted Data
Year: 2023 (published online Nov 28, 2022)
Modality: Bitewing radiographs (intraoral X‑rays)
Dataset: 994 - 17,530 teeth annotated
Dataset Split: Train 994 / Val 256 / Test 175 images
Network Architecture: YOLOv3 (Darknet‑53 backbone)
Metrics: Precision, Recall, F1‑score, AP, mAP at IoU50 and IoU75
AP – Professional Qty: 3 (consensus annotation)
AP – Supervisor Presence: No information
AP – Experience Level: 7–25 years
AP – Expertise Area: Oral and Maxillofacial Radiology (OMFR)
AP – Tool or System: LabelImg (annotation);
ML Task: Object detection + multi‑class caries classification by ICCMS™ (2‑, 4‑, and 7‑class)
Project Objective: Assess YOLOv3 performance for caries detection and ICCMS™ classification under IoU50 and IoU75
Clinical Relevance
YOLOv3 can assist dentists by detecting and staging caries on bitewing radiographs using ICCMS™ rules with acceptable accuracy, particularly for moderate/advanced lesions. Early enamel lesions remain challenging; class imbalance and small‑object limitations warrant further refinement and data balancing.
Watch full video on YouTube
