Related Experiment Video
Updated: Nov 16, 2025
09:34
Semi-Automated Planimetric Quantification of Dental Plaque Using an Intraoral Fluorescence Camera
Published on: January 27, 2023
2.2K
Detecting white spot lesions on dental photography using deep learning: A pilot study
Haitham Askar1, Joachim Krois1, Csaba Rohrer1
1Department of Oral Diagnostics, Digital Health and Health Services Research, Charité - Universitätsmedizin Berlin, Germany.
Journal of Dentistry
|February 22, 2021
Summary
Deep learning accurately detects white spot lesions in dental images, especially fluorosis. While promising, model stability may be limited by the small dataset size, necessitating larger studies for broader applicability.
Area of Science:
- Dentistry
- Artificial Intelligence
- Medical Imaging
Background:
- White spot lesions (WSLs) are early indicators of enamel demineralization.
- Accurate detection of WSLs is crucial for timely intervention and preventing caries progression.
- Current diagnostic methods can be subjective and time-consuming.
Purpose of the Study:
- To evaluate the efficacy of deep learning models in detecting white spot lesions from dental photographs.
- To differentiate between fluorotic and other types of hypomineralized lesions.
- To assess the performance of deep learning models on cropped versus region-of-interest (ROI) images.
Main Methods:
- A dataset of 2781 tooth segments from 434 dental images was created with expert annotations.
- SqueezeNet architecture was utilized for deep learning model development.
- Models were trained to detect any WSLs, fluorotic lesions, and other-than-fluorotic lesions using cross-validation.
- Feature visualization was employed to identify salient detection areas.
Main Results:
- Deep learning models achieved mean accuracies between 0.81-0.84 for WSL detection.
- Specificities ranged from 0.85-0.86, with sensitivities between 0.58-0.66.
- Models demonstrated comparable performance on cropped and ROI images.
- Light reflections were identified as a common source of false positive detections.
Conclusions:
- Deep learning models show significant potential for automated WSL detection in dental imagery, particularly for fluorosis.
- The study highlights the need for larger, more generalizable datasets to improve model stability and clinical utility.
- Automated classification can assist dental practitioners in WSL diagnosis and management.
