Dental caries in bitewing radiographs
The dataset contains 100 images of bitewing radiographs, each image showing several human teeth. Carious lesions were annotated by rectangular bounding boxes drawn around the lesions by 8 annotators - 5 experienced and 3 less experienced dentists. Images are of size 1068x847 pixels, sometimes with black top and bottom margins, in the PNG format. Annotations are in the COCO JSON format and contain the bounding boxes created by all annotators. The data can be used for evaluating the variability between human annotators as well as to evaluate machine learning algorithms for carious lesion detection. An article "Tichý, Kunt, Nagyová, Kybic: Automatic caries detection in bitewing radiographs. Part II: Experimental comparison" using this dataset was submitted to the Clincial Oral Investigations journal.
Steps to reproduce
Gather bitewing radiographs from a hospital information system. Rescale and pad them to the same size. Use the CVAT annotation tool to draw bounding boxes, each annotator independently.
Všeobecná Fakultní Nemocnice v Praze
Ministerstvo Školství, Mládeže a Tělovýchovy