Detection of non-cavitated approximal caries lesions in digital images from seven solid-state receptors with particular focus on task-specific enhancement filters. An ex vivo study in human teeth.

The aim of this study was to compare detection of non-cavitated approximal caries lesions in images from seven solid-state intraoral digital receptors, with particular focus on two task-specific enhancement filters. One hundred and sixty approximal non-cavitated surfaces were radiographed under standardized conditions using the following seven intraoral solid-state digital receptors: two CMOS systems, Schick CDR-APS, and Kodak RVGui; and five CCD systems: Visualix, VistaRay, Dixi2, Sidexis, and Dr. Suni Plus. The Kodak RVGui digital images were captured with two task-specific, predefined enhancement filters ("dento-enamel" and "periodontal"). Eight observers examined the digital images for the presence or absence of approximal carious lesions. The teeth were subsequently sectioned for histological analysis, which served as the "gold standard" for the radiographic examination and allowed for a calculation of sensitivity, specificity, and overall accuracy. Both RVG modalities obtained the highest sensitivity values, which were significantly higher than with Visualix, Sidexis, and VistaRay images (all P<or=0.03). The RVG "periodontal" images also had a higher accuracy than Dixi2, Dr. Suni Plus, and CDR-APS (P<or=0.05). The RVG modalities also showed the largest observer variation, and their high sensitivities were mainly due to one observer. The sensitivity for VistaRay images was the lowest of all modalities and significantly lower than almost all the other digital systems (P<or=0.02; except for Visualix). Besides, Visualix images had lower sensitivity than CDR-APS, Dixi2, and Dr. Suni Plus images (P<or=0.003). On the other hand, the two RVGui image types showed the lowest mean specificity values. For overall accuracy, the differences among the modalities were non-significant (P>0.05).