Accuracy of computer-automated caries detection in digital radiographs compared with human observers

The aim of this study was to compare diagnostic accuracy of a caries detection program with that of human observers. A total of 190 extracted teeth were radiographed with two Trophy RVG (RadioVisioGraphy) digital sensor systems. Four observers scored the approximal surfaces in all images on a disease severity scale. Each observer thereafter used the Logicon Caries Detector (LCD) program to analyse the surfaces in the digital images and recorded their outcome. To determine the true absence or presence of caries, histological validation was used. Sensitivities, specificities, positive and negative predictive values were calculated and differences between the diagnostic methods tested. Specificities for the outcome with the LCD were significantly lower for three observers than when they themselves assessed the RVG images and, correspondingly, the positive predictive values were lower for the LCD outcome for three of the observers. Sensitivity was also lower for two observers on the diagnostic threshold caries in dentine. It was concluded that the automated caries detection program is less accurate than human observers in detecting approximal caries lesions.