Automatic dental biofilm detection based on deep learning.

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Dental biofilm, commonly known as plaque, is a major factor in oral health issues.
• Accurate detection of dental biofilm is crucial for effective oral hygiene practices.
• Current methods for biofilm detection can be subjective and time-consuming.

Purpose of the Study:

• To evaluate the capability of the U-Net neural network for automated dental biofilm detection.
• To assess the performance of U-Net in segmenting biofilm from intra-oral images.
• To determine the feasibility of using AI for improved oral hygiene monitoring.

Main Methods:

• Utilized two datasets of intra-oral photographs (96 and 480 images) for validation and training.
• Employed a U-Net neural network for biofilm segmentation on images without disclosing agents.
• Validated expert annotations with an intra-class correlation coefficient of 0.93.
• Measured segmentation performance using accuracy, F1 score, sensitivity, and specificity.

Main Results:

• The U-Net model achieved 91.8% accuracy, 60.6% F1 score, 94.4% specificity, and 67.2% sensitivity.
• Accuracy was slightly higher (92.6%) in images with orthodontic appliances.
• The model demonstrated strong performance in automated biofilm detection.

Conclusions:

• Automated dental biofilm segmentation using U-Net is a feasible approach.
• This technology can support dental professionals and patients in identifying biofilm.
• Improved identification of dental biofilm can lead to enhanced oral hygiene and health outcomes.