ROC analysis of acid demineralized artificial caries.
- Author:
Byung Cheol KANG
1
Author Information
1. Department of Oral and Maxillofacial Radiology, College of Dentistry, Chonnam National University, Korea.
- Publication Type:Original Article
- Keywords:
ROC analysis
- MeSH:
Bicuspid;
Consensus;
Dataset;
Dentists;
Humans;
Molar;
ROC Curve*;
Sensitivity and Specificity;
Tooth;
Weights and Measures
- From:Journal of Korean Academy of Oral and Maxillofacial Radiology
1997;27(2):7-14
- CountryRepublic of Korea
- Language:English
-
Abstract:
This study is designed to determine the artificial incipient proximal caries lesion detectability by dentists on Ektaspeed Plus film using ROC analysis. Sixteen premolars and 30 molars, which have 52 proximal caries-like demineralized lesions using acid-gel technique were added to 20 sound premolars and 30 sound molars to make 24 plaster bolcks. Each block with 4 teeth and 6 contacting proximal surfaces was placed in an optical bench to take 12 bitewing radiographs with Ektaspeed Plus film. Thirty-six dentists acted as observers to evaluated the proximal lesions using five rating scales for ROC analysis. They were also asked to determine the presence or absence of the proximal caries. The true status of the proximal caries was established by the consensus of three oral and maxillofacila radiologists. For evaluation of intra-observer agreement, 9 dentist reread the radiographs at an interval of 1 month. The Pearson correlation coefficient for the intra-observer agreement was 0.746(good agreement). Ten observer's data set were degenerated. The mean area under ROC curve from 26 obsevers was 0.806 and standard deviation was 0.061. The sensitivity and the specificity of the binary response were 0.17(SD=0.11) and 0.78(SD=0.17) respectively. The binary response only reveal a single values of sensitivity and the specificity. The ROC analysis to assess the diagnostic accuracy in caries detection, which producing estimates of sensitivities for all specifities, yield more comprehensive measures of diagnostic performance than single values for sensitivity and specificity.
