The effect of wavelet and discrete cosine transform compression of digital radiographs on the detection of subtle proximal caries. ROC analysis.
OBJECTIVES To study the effect of lossy image compression on caries detection. Null hypothesis states no difference in caries detection efficacy among observers using original digital images versus images compressed at various rates of compression. METHODS Digital images of 41 extracted posterior teeth were obtained with a storage phosphor DenOptix (Gendex DenOptix Imaging system) system. Images were exported in Tagged Image File Format (TIFF) and compressed with Joint Photographic Experts Group File Interchange Format (JFIF), as provided by the software of the imaging system. The compressions options JFIF100%, JFIF75% and JFIF50% resulted in reducing the image size to 1:2, 1:11 and 1:16, respectively. Eight observers evaluated the presence or absence of caries on a 5-point confidence scale. The actual caries status of each proximal surface was determined by ground section histology. Responses were evaluated by receiver operating characteristic (ROC) analysis. Areas under the curves (Az) were assessed using analysis of variance (ANOVA). RESULTS The mean Az scores were 0.85 for original/uncompressed images, and 0.89 for JFIF100%, 0.88 for JFIF75% and 0.88 for JFIF50% images. These differences were not statistically significant (P = 0.26). Differences between observers were also not statistically significant (P = 0.12). CONCLUSIONS JFIF compression at the level of 1:16 can be used without significant deterioration in diagnostic accuracy for proximal caries detection.