FDTooth: Intraoral Photographs and CBCT Images for Fenestration and Dehiscence Detection

Area of Science:

• Dental diagnostics
• Medical imaging
• Computer vision

Background:

• Fenestration and dehiscence (FD) present significant challenges in dental care, impacting oral health.
• Cone-beam computed tomography (CBCT) offers precise diagnostics but is limited by time and radiation exposure.
• A lack of public datasets combining intraoral photographs and CBCT images hinders deep learning for automated disease detection.

Purpose of the Study:

• To introduce FDTooth, a novel dataset for developing deep learning models for automated fenestration and dehiscence detection.
• To provide a valuable resource for interdisciplinary dental diagnostics research.
• To facilitate non-invasive, efficient early screening of dental conditions.

Main Methods:

• Creation of the FDTooth dataset, comprising intraoral photographs and CBCT images from 241 patients (ages 9-55).
• Annotation of 1,800 bounding boxes on intraoral photographs with ground truth from CBCT.
• Development of a baseline deep learning model for automated FD detection.

Main Results:

• The FDTooth dataset includes paired intraoral photographs and CBCT images with detailed annotations.
• A baseline model for automated FD detection from intraoral photographs was successfully developed.
• The dataset and model serve as foundational resources for advancing automated dental diagnostics.

Conclusions:

• FDTooth addresses the critical need for a public dataset combining intraoral images and CBCT data.
• The developed dataset and baseline model support research in automated detection of fenestration and dehiscence.
• This work promotes non-invasive and efficient methods for early dental screening and diagnosis.