SegmentAnyTooth: An open-source deep learning framework for tooth enumeration and segmentation in intraoral photos.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Deep Learning for Oral Health

Background:

• Preventive dentistry is crucial for public oral health.
• Inequalities in dental care access persist, particularly in underserved regions.
• Image-based dental analysis using intraoral photographs offers a scalable solution.

Purpose of the Study:

• To develop SegmentAnyTooth, an open-source deep learning framework for automated tooth enumeration and segmentation.
• To address the challenge of accurate tooth identification across five standard intraoral views.
• To lay the groundwork for advanced dental applications and enhance access to preventive care.

Main Methods:

• A dataset of 5000 intraoral photos from 1000 sets was annotated.
• You Only Look Once 11 (YOLO11) nano models were used for tooth localization and enumeration.
• Light Segment Anything in High Quality (Light HQ-SAM) with active learning was employed for segmentation.

Main Results:

• SegmentAnyTooth achieved high segmentation accuracy across multiple views (e.g., DSC of 0.983 ± 0.036 for upper occlusal).
• Models demonstrated strong performance on lateral views (DSCs of 0.939 ± 0.070 right, 0.945 ± 0.056 left).
• Statistically significant improvements were observed compared to baseline models (U-Net, nnU-Net, Mask R-CNN; P < 0.01).

Conclusions:

• SegmentAnyTooth offers accurate, multi-view tooth segmentation to advance dental care and diagnosis.
• The open-source framework facilitates integration into clinical and public health workflows.
• Future work focuses on enhancing the generalizability of the SegmentAnyTooth model.