Detection of subtle interstitial abnormalities of the lungs on digitized chest radiographs: acceptable data compression ratios

To determine acceptable compression ratios for digital radiography, we evaluated the effect of data compression on the detection of subtle interstitial lung abnormalities using digitized chest radiographs. Screen-film chest radiographs of 38 patients with subtle interstitial lung abnormalities and 40 patients with normal lung parenchyma were digitized (spatial resolution, 0.175 mm; 2000 x 2000 pixels; 10 bits per pixel) and compressed with the discrete cosine transform method at ratios of 10:1, 20:1, and 30:1. Five chest radiologists and five radiology residents examined the uncompressed and compressed digital images and rates the presence of interstitial lung abnormalities with a five-level scale of confidence. Results were analyzed by receiver operating characteristic methods. Overall, the interpretation of images with a compression ratio of 30:1 was significantly less accurate than that of uncompressed images (p < .05). For the five chest radiologists, interpretation of images with a compression ratio of 20:1 or 30:1 was significantly less accurate than that of uncompressed images (p < .05). However, for the five residents, no significant difference between interpretations of compressed and uncompressed images was noted (p > or = .05). These results suggest that a 10:1 data compression ratio does not influence the detection of subtle interstitial lung abnormalities. However, information that is lost with a 20:1 data compression ratio might be essential for interpretation by experienced chest radiologists.