Detection of subtle abnormalities on chest radiographs after irreversible compression.

Abstract

To assess the effect of wavelet-based compression of posteroanterior chest radiographs on detection of small uncalcified pulmonary nodules and fibrosis. Computed tomography (CT) of the chest was used to identify 20 patients with normal posteroanterior chest radiographs, 20 with a solitary uncalcified pulmonary nodule 1-2 cm in diameter, and 20 with fibrotic disease. A double-blind protocol for readings of original images and images compressed at 40:1 and 80:1 was analyzed by using the nonparametric receiver operating characteristic to measure differences in diagnostic accuracy and their statistical significance. There was no substantial difference in the overall diagnostic accuracy (measured by the area under the curve index) for both nodules and fibrosis between images compressed at 40:1 and 80:1 and uncompressed images. Readers tended to perform better on images compressed at 40:1 compared with uncompressed images. The "high-sensitivity" portion of the 80:1 compression curve for nodules was below that for the uncompressed curve, although this was not statistically significant. Lossy compression of chest radiographs at 40:1 can be used without decreased diagnostic accuracy for detection of pulmonary nodules and fibrosis. There is no statistically significant difference in diagnostic accuracy at 80:1 compression, but detection ability is decreased.