The effect of wavelet and discrete cosine transform compression of digital radiographs on the detection of subtle proximal caries. ROC analysis

Abstract

The study compared diagnostic performances of 2 different image compression methods: JPEG (discrete cosine transform; Joint Photographic Experts Group compression standard) versus JPEG2000 (discrete wavelet transform), both at a compression ratio of 12:1, from the original uncompressed TIFF radiograph with respect to the detection of non-cavitated carious lesions. Therefore, 100 approximal surfaces of 50 tooth pairs were evaluated on the radiographs by 10 experienced observers using a 5-point confidence scale. Observations were carried out on a standardized viewing monitor under subdued light conditions. The proportion of diseased surfaces was balanced to approximately 50% to avoid bias. True caries status was assessed by serial ground sectioning and microscopic evaluation. A non-parametric receiver operating characteristic analysis revealed non-significant differences between the 3 image modalities, as computed from the critical ratios z not exceeding +/-2 (JPEG/JPEG2000, z = -0.0339; TIFF/JPEG2000, z = 0.251;TIFF/JPEG, z = 0.914). The mean area beneath the curve was highest for TIFF (0.604) followed by JPEG2000 (0.593) and JPEG (0.591). Both intra-rater and inter-rater agreement were significantly higher for TIFF (kappa(intra) = 0.52; kappa(inter) = 0.40) and JPEG2000 images (kappa(intra) = 0.49; kappa(inter) = 0.38) than for JPEG images (kappa(intra) = 0.33; kappa(inter) = 0.35). Our results indicate that image compression with typical compression algorithms at rates yielding storage sizes of around 50 kB is sufficient even for the challenging task of radiographic detection of non-cavitated carious approximal lesions.