Automated segmentation of dental restorations using deep learning: exploring data augmentation techniques.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Digital Dentistry

Background:

• Deep learning models are crucial for automated dental radiograph segmentation.
• Model performance depends on training data quality and diversity.
• Data augmentation artificially expands datasets to improve model generalization.

Purpose of the Study:

• To automatically segment implants, prostheses, and fillings in panoramic dental images.
• To evaluate the impact of various data augmentation techniques on segmentation performance.
• To compare the effectiveness of nine different deep learning segmentation models.

Main Methods:

• Utilized nine deep learning segmentation models for dental image analysis.
• Applied eight distinct data augmentation techniques to training datasets.
• Assessed model performance using Intersection over Union (IoU) and Dice coefficient metrics.

Main Results:

• Deep learning models achieved IoU scores between 0.62-0.82 and Dice scores between 0.75-0.9.
• Data augmentation improved segmentation performance by up to 3.37% (implants), 5.75% (prostheses), and 8.75% (fillings).

Conclusions:

• Data augmentation enhances the accuracy of automated dental image segmentation.
• The choice of augmentation strategy is dependent on the specific deep learning model and the type of dental structure being analyzed.