Detection of approximal radiolucencies in enamel: a preliminary comparison between experienced clinicians and an image analysis method.

Abstract

Abstract A new computer-aided image analysis (IA) method of estimating the size of radiolucencies in approximal enamel has recently been shown to be capable of making objective and reproducible measurements of lesion depth and area. If the IA method is to be used to monitor the behaviour of lesions over time it is important also to evaluate its performance in terms of detecting the presence or absence of lesions prior to measurement (that is ‘lesion detection’ rather than ‘lesion measurement’). The purpose of this study was to determine the diagnostic performance of the IA system and compare it with that of 10 dentally qualified observers who were experienced in radiographic caries diagnosis. Bitewing radiographs of extracted (Hong Kong) teeth were exposed, using a film-holder, to simulate clinical films. Radiographs depicting 20 macroscopically carious and 20 macroscopically sound surfaces were independently read on three occasions by the image analyser, on three occasions by an oral radiologist, and on two occasions each by nine teachers of conservative dentistry. Diagnoses were validated by reference to the ‘true’ histologically determined state of each surface. On sectioning, in addition to caries, numerous defects of maturation were disclosed. The image analyser's diagnoses concurred for 38 40 surfaces in contrast to the considerably greater variation seen between the repeated visual caries diagnosis. The human observer's diagnostic performance revealed visual diagnosis to be a somewhat insensitive, inconsistent but reasonably specific process. Overall, the diagnostic accuracy of the IA method was comparable with that achieved by the visual method, the IA system tended, however, to be more sensitive but less specific.