Deep Learning-Based Multi-Stage System for Automated Tooth Detection and Segmentation in Orthodontic Photography.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Orthodontic Technology

Background:

• Manual tooth charting is time-consuming, error-prone, and inconsistent.
• Inaccurate manual segmentation can compromise dental diagnosis and treatment planning.
• There is a need for automated solutions to improve efficiency and accuracy in dental imaging.

Purpose of the Study:

• To develop and evaluate an automated AI system for tooth detection and segmentation from orthodontic photographic plates.
• To enhance clinical outcomes and workflow efficiency in dentistry through AI.
• To provide a reliable tool for large-scale annotation and digital charting.

Main Methods:

• A multi-stage AI system combining YOLOv11 for localization and detection with Segment Anything Model (SAM) for segmentation.
• Training and validation on a dataset of 2000 composite dental images, with external validation.
• Zero-shot segmentation masks generated without additional training for each detected tooth.

Main Results:

• High accuracy in arch classification (99.5% mAP@0.5) and tooth detection (mAP@0.5 of 0.943 for maxilla, 0.918 for mandible).
• 94% of generated segmentation masks met clinical usability criteria without correction.
• The system processed each image in an average of 2.2 seconds, demonstrating practical applicability.

Conclusions:

• The integrated YOLO and SAM system offers a fast, accurate, and fully automated solution for tooth analysis in orthodontic photography.
• This AI-driven approach supports efficient large-scale annotation, digital charting, and AI-assisted diagnostics.
• The system has the potential to significantly benefit both orthodontic and general dental practices.