Dental Images' Segmentation Using Threshold Connected Component Analysis.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Dentistry

Background:

• Deep learning advances medical imaging analysis, particularly in identifying patterns in radiographs.
• Deep learning excels in dental image segmentation, a critical step for diagnosing dental caries.
• Challenges in dental image segmentation include complex structures, poor image quality, and irregular lesion borders.

Purpose of the Study:

• To develop and evaluate a robust method for segmenting dental radiographs to aid in caries diagnosis.
• To overcome limitations of existing deep learning models in segmenting challenging dental carious lesions.

Main Methods:

• A dental segmentation method combining thresholding and connected component analysis was employed.
• Images underwent preprocessing with Gaussian blur filters and enhancement via erosion and dilation morphology operations.
• Segmentation was finalized using thresholding, with connected components identified to extract the Region of Interest (ROI) of teeth.

Main Results:

• The proposed method achieved 93% precision and 93% recall on an augmented dataset.
• The evaluation involved training on 10,090 images and testing on 1,024 images from a total of 11,114 dental images.

Conclusions:

• The developed method demonstrates high performance in segmenting dental radiographs, offering a promising tool for caries diagnosis.
• The approach effectively addresses challenges associated with dental image quality and lesion complexity.
• This technique contributes to the advancement of artificial intelligence applications in dental diagnostics.