Interaction between noise and file compression and its effect on the recognition of caries in digital imaging.

Abstract

To determine the interaction between image noise and file compression, with special emphasis on the accuracy of caries diagnosis. Fifty-nine bitewing radiographs of patients were taken simultaneously with Ektaspeed Plus (Eastman-Kodak, Rochester, NY, USA) film without lead foil and the Digora storage phosphor system (Soredex, Helsinki, Finland). Three different levels of Gaussian noise were added to the original digital images which were then compressed with JPEG 53. Seven observers evaluated the presence and depth of caries lesions on selected approximal surfaces on a 5-point scale. The results of JPEG 27 compression from a previous study were also included. ROC analysis was used together with multivariate analysis of variance (MANOVA). JPEG 27 and 53 reduced the file size down to 7% and 4.6% of the original respectively. ROC curve analysis showed no significant difference between image conditions (original, JPEG 27, and JPEG 53) at the same noise level. JPEG 27 and 53 had larger Az scores than their original counterparts at the same noise level. However, MANOVA showed that for depth estimation of enamel lesions JPEG 53 resulted in a higher observer error. Both JPEG 53 and 27 could reduce some of the adverse effect of noise from the image by removing high spatial frequencies. JPEG 53, resulting in a compression ratio of 1:21, does not compromise the diagnostic performance in general. JPEG 53 compression may however affect the ability to detect enamel lesions.