Influence of the validation method on diagnostic accuracy for caries. A comparison of six digital and two conventional radiographic systems

Objective: To evaluate the influence of the validation method on the diagnostic accuracy and the relative comparison of eight radiographic systems for caries detection. Methods: Three hundred and thirty-eight approximal and 145 occlusal surfaces were radiographed under standardised conditions using six CCD-based sensor systems: MPDx (Dental/Medical Diagnostic Systems Inc., Woodland Hills, CA, USA), Dixi (Planmeca, Helsinki, Finland), Sidexis (Sirona, Bensheim, Germany), RVG(old) (Trophy, Paris, France, 1994 model), RVG(new) (Trophy, Paris, France, 2000 model) and Visualix (Gendex, Milan, Italy) and two film systems: Ektaspeed Plus and Insight (Eastman Kodak, Rochester, NY, USA). Four observers examined the radiographs for approximal and occlusal caries using a five-point confidence scale. The presence of caries was validated histologically and radiographically. Diagnostic accuracy was evaluated using ROC curve areas (A(z)). Results: For both approximal and occlusal caries the mean A(z) of the eight radiographic systems was significantly higher using radiographic than histological validation (P<0.001). Using histological validation for approximal caries, Dixi (A(z)=0.71) and Ektaspeed Plus (A(z)=0.7) were not significantly different, but Dixi was significantly more accurate than the other digital systems and the Insight film. Using radiographic validation for approximal caries, Ektaspeed Plus (A(z)=0.87) was significantly more accurate than Dixi (A(z)=0.82). Dixi was significantly more accurate than MPDx (A(z) = 0.74), RVG(old) (A(z) = 0.77), RVG(new) (A(z) = 0.77) and Visualix (A(z)= 0.76). Corresponding variations were found for occlusal caries depending on the validation method. Using histological validation, MPDx (A(z)=0.76) was significantly less accurate than Dixi (A(z)=0.81), Sidexis (A(z)=0.8), Ektaspeed Plus (A(z)=0.82) and Insight (A(z)=0.81). Using radiographic validation, MPDx (A(z)=0.83) was also significantly less accurate than RVG(old) (A(z)= 0.89) and RVG(new) (A(z) = 0.9). Conclusion: A(z) obtained from radiographic validation was significantly higher than A(z) obtained from histological validation. Comparison of the diagnostic efficacy for caries of the eight radiographic systems was strongly influenced by the validation method.