Reproducibility of computer-aided image-analysis-derived estimates of the depth and area of radiolucencies in approximal enamel.

Abstract

In vivo studies of caries progression from serial bite-wing radiographs require the reproducible detection and measurement of approximal radiolucencies in enamel; the aim of this study was to evaluate the reproducibility of a new computer-aided image analysis method. Three series of radiographs depicting natural caries lesions were employed: films of 11-12-year-old (Hounslow) schoolchildren, dental students from Hong Kong (HKDS), and extracted Hong Kong teeth (EXT). Each surface was "searched" three times, resulting in data from 450, 600, and 180 searches, respectively. Lesion detection was consistent for all but two of the 1230 searches. Average values for the standard error of the mean (SEM) of triple determinations of depth % were between 2.0 and 3.5%, while the mean SEM for area was from 0.03 to 0.04 mm2. Average test/re-test correlations ranged from 0.963 to 0.871 for depth and from 0.963 to 0.884 for area, while average reliability coefficients ranged from 0.937 to 0.821 for depth and 0.963 to 0.886 for area. There was a significant difference (P less than 0.001) between the reproducibility of the two clinical series, with measurements of the Hounslow lesions being more consistent. When compared with the reproducibility of attempts to grade radiolucency size visually, the image analysis method appears to offer considerable advantages, in that consistent estimates of radiolucency size are now possible. This method may facilitate more sensitive assessments of the behavior of lesions over time than has hitherto been possible.