Analysis of measures to determine diagnostic accuracy in detecting proximal enamel radiolucency

Abstract

Measures available for the diagnostic value of radiographic images were examined in the detection of the caries-like proximal enamel radiolucency. Systematic variations used in this study were the recording systems, mean density of the proximal enamel and horizontal projection angle. Performance was expressed in terms of a true positive (TP) answer coupled with the true negative (TN) answer and the receiver operating characteristic (ROC) analysis, obtainable by ten observers who were asked to detect the induced lesions in the radiographic phantom. The results showed that the TP value was inversely correlated to the TN value in the systematic changes of the recording systems and density, although both values seemed to express the nature of each observer in detecting the caries-like lesions. The ROC analysis data were correlated to the variations used. An analysis of the variance revealed that the observer variance was significant but had a relatively small effect compared to the other factors within our variations. These findings indicated that the ROC analysis complemented by the TP and TN values was more reliable for the evaluation of this specific diagnostic performance.