Computer-automated caries detection in digital bitewings: consistency of a program and its influence on observer agreement.

The aim of this study was to evaluate a decision-support, caries detection program and its influence on observer agreement in caries diagnosis. 130 patients were examined by digital bitewing radiography (RVG XL sensor, Trophy Radiologie Inc.). Fifty-four approximal surfaces (27 in premolars and 27 in molars) were selected by the author: 24 surfaces (9 in molars and 15 in premolars) scored as sound, 16 surfaces (9 in molars and 7 in premolars) scored as carious in enamel, and 14 surfaces (9 in molars and 5 in premolars) scored as carious in dentine. The Logicon Caries Detector (LCD) program (Logicon Inc., USA) was assessed by repeating the automated analysis ten times for each surface. The two most varying outcomes for lesion probability (Lp(min) and Lp(max)) were saved. Five observers scored the 54 surfaces independently as sound, caries in enamel or caries in dentine before and after the use of LCD. In more than one third of all surfaces the program indicated different lesion probability, from sound at Lp(min) to the presence of a carious lesion at Lp(max). The 5 observers changed their caries score after the use of LCD in a total of 31 surfaces (only 2 of these were in the same surface). Mean kappa value for inter-observer agreement for caries scores before the use of LCD was 0.47 (range 0. 39-0.61) and after LCD 0.48 (range 0.37-0.69). It was concluded that the automated caries detection program was not very consistent and provided different opinions on the caries status in a surface. Inter-observer agreement in caries diagnosis did not improve using the program.