Multi-Source Dental X-Ray Dataset Using Image-to-Image Transformation
Description
The Teeth View X-ray Image Dataset is a collection of dental X-ray images gathered from different dental clinics. It is designed for machine learning tasks such as object detection. The dataset is organized into one main folder: the object detection dataset. The object detection folder contains 1,674 augmented images with corresponding labels in JSON format. By applying a diffusion model image to image, we have generated additional X-ray images of teeth scans to enhance the Teeth View X-ray Image Dataset to be deeper, more diverse, and beneficial for model tuning. The synthetic image contains various anatomy, density, and acquiring conditions of the tooth and thereby enhances the model's robustness for unusual dental conditions. The increased size of the dataset works to counterbalance class imbalance by having more examples of the under-represented classes and reducing model bias. Also, diffusion models demonstrate high effectiveness in reproducing noise patterns, making training more robust; improving the clarity of images through denoising processes; and enabling highly precise segmentation mask predictions to facilitate efficient boundary detection. Variables: BDC-BDR Teeth -208 Caries Teeth - 342 Fractured Teeth -52 Healthy Teeth - 893 Impacted Teeth - 348 Inflection Teeth - 92 Folder structure: Teeth view X-ray Image dataset | |---Objectect Detection | |----image |----Labels.
Files
Steps to reproduce
The Teeth View X-ray Image Dataset is a curated collection of dental X-ray images obtained from multiple dental clinics. The dataset was sourced from three different medical institutions, ensuring diversity in patient demographics, imaging techniques, and clinical conditions. Standardized X-ray imaging protocols were followed across all sources to maintain consistency in resolution, exposure settings, and anatomical coverage. To enhance the dataset and improve model robustness, we applied an image-to-image diffusion model for data augmentation. Annotation was performed using Roboflow, an AI-powered annotation tool that streamlines dataset preparation for object detection and segmentation tasks. Images were manually labeled by dental experts to ensure high annotation accuracy. Each image was annotated with bounding boxes and class labels, stored in JSON format for easy integration with machine learning models.