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.本数据集包含100张咬翼片X光片（bitewing radiographs），每张图像均展示多颗人类牙齿。龋损（carious lesions）的标注由8名牙科医师完成，他们通过在病变区域绘制矩形边界框来标记病变，其中5名为经验丰富的牙医，3名为经验相对不足的牙医。图像分辨率为1068×847像素，部分图像带有上下黑色边距，格式为PNG。标注采用COCO JSON格式，包含所有标注者生成的边界框信息。该数据集既可用于评估人类标注者之间的标注差异，也可用于评估龋损检测相关的机器学习算法性能。使用本数据集的论文《Tichý, Kunt, Nagyová, Kybic: Automatic caries detection in bitewing radiographs. Part II: Experimental comparison》已提交至《Clinical Oral Investigations》期刊。

搜集汇总

背景与挑战

背景概述

该数据集包含100张咬翼X光片图像，用于龋齿检测研究，每张图像由8位牙医标注矩形边界框以标记病变区域，旨在评估标注者差异和机器学习算法性能；图像尺寸统一为1068x847像素，采用PNG格式和COCO JSON标注格式。