Multi-Class Malocclusion Detection on Standardized Intraoral Photographs Using YOLOv11.

Area of Science:

• Artificial Intelligence in Dentistry
• Computer Vision for Medical Imaging
• Orthodontic Diagnostics

Background:

• Dental malocclusion identification from clinical photos is time-consuming and prone to variability.
• Automating this process can enhance efficiency and diagnostic consistency.
• Deep learning offers a potential solution for accurate, automated malocclusion detection.

Purpose of the Study:

• To evaluate a YOLOv11-based deep learning model for automatic malocclusion detection in intraoral photographs.
• To test if training on a structured protocol enables reliable detection of multiple malocclusions.
• To assess the model's performance across various clinically relevant malocclusion classes.

Main Methods:

• A dataset of 5854 intraoral photographs was annotated with bounding boxes for 17 malocclusion classes based on the Index of Orthodontic Treatment Need (IOTN).
• A YOLOv11 model was trained using augmented data and evaluated on a held-out test set.
• Performance was measured using mean average precision (mAP50), macro precision (macro-P), and macro recall (macro-R).

Main Results:

• The YOLOv11 model achieved 87.8% mAP50, 76.9% macro-P, and 86.1% macro-R across 15 analyzed malocclusion classes.
• High per-class performance was noted for Deep bite (98.8%), Diastema (97.9%), and Angle Class II canine (97.5%).
• Performance varied across classes, with lower detection rates for Posterior crossbite and Big overjet, potentially due to limited examples and visualization constraints.

Conclusions:

• A YOLOv11-based deep learning system can accurately detect multiple clinically significant malocclusions from routine intraoral photographs.
• The system supports efficient screening and standardized documentation of dental malocclusions.
• Further improvements in robustness can be achieved through larger datasets, protocol standardization, and multimodal inputs.