Deep learning models for the detection of dental-findings and tooth-types using video data

Area of Science:

• Artificial Intelligence in Dentistry
• Deep Learning for Medical Imaging
• Oral Health Technology

Background:

• Oral diseases affect 35.5% globally, with limited access to dental care.
• Existing AI dental diagnostics often require high-quality images from clinical settings.
• This research addresses the need for AI diagnostics using accessible, everyday oral care devices.

Purpose of the Study:

• To develop and validate Deep Learning (DL) models for dental findings and tooth identification.
• To utilize video data from a toothbrush-mounted intraoral device for AI diagnostics.
• To assess the performance of AI models in detecting common dental issues and classifying tooth types.

Main Methods:

• 708 videos captured with an electronic toothbrush integrated with an intraoral camera.
• Extracted and filtered image frames for dental findings (7,963) and tooth types (3,799).
• Employed YOLOv8s architecture, with expert annotation of findings (decay, fillings, plaque, staining) and tooth types.

Main Results:

• Dental-findings model achieved 83.0% mAP and 77.9% F1-score, excelling in detecting amalgam fillings (94.5% mAP) and decay (84.2% mAP).
• Tooth-type model achieved 88.6% mAP and 86.2% F1-score, with high accuracy for molars (97.7% mAP) and premolars (96.7% mAP).
• Models demonstrated robust performance despite variations in video capture without standardization.

Conclusions:

• Establishes proof of concept for integrating AI diagnostics into daily oral care devices.
• Accurate detection of clinical findings from user-captured video enables remote monitoring and early intervention.
• Potential to optimize healthcare resource allocation, especially in underserved areas.