A multi-focus oral panoramic x-ray image dataset based on pixel-level annotations.

Area of Science:

• Dental imaging
• Artificial Intelligence
• Medical Informatics

Background:

• Dental artificial intelligence research requires large, annotated datasets.
• Existing datasets may lack comprehensive lesion annotations or scale.
• High-quality panoramic X-ray images are crucial for AI model training.

Purpose of the Study:

• To develop and present a large-scale, meticulously annotated dataset of panoramic dental X-ray images.
• To facilitate research in dental artificial intelligence, focusing on lesion detection and diagnosis.
• To provide a reliable resource for training and validating AI models in dentistry.

Main Methods:

• Collected 8,655 de-identified panoramic X-ray images from Changsha Stomatological Hospital.
• Utilized LabelMe software for manual, pixel-level annotation of over 30,186 lesion regions by 20 dental imaging specialists.
• Ensured annotation precision and clinical reliability through rigorous expert delineation of tooth contours and oral conditions.

Main Results:

• The dataset contains 8,655 panoramic X-ray images with 30,186+ pixel-level annotations of oral lesions.
• Annotations cover tooth segmentation and various common oral conditions, validated for precision.
• Deep learning models demonstrated strong generalization capabilities when trained on this dataset.

Conclusions:

• The presented dataset is a valuable, large-scale resource for advancing dental artificial intelligence.
• It supports diverse applications including tooth segmentation, lesion detection, and computer-aided diagnosis.
• The dataset's validated effectiveness indicates its potential to improve AI-driven dental diagnostics.