Image data compression in magnification hand radiographs.

Abstract

A study was conducted on the use of an irreversible compression technique, Fourier quantization, to reduce the amount of digital data needed for an image. The effect of image compression was studied in radiographs obtained to diagnose subperiosteal bone resorption. Magnification radiographs of 44 hands were digitized to an array size of 4,096 X 4,096 X 12 bits. Selected subregions containing a single middle phalanx were compressed and reconstructed. A subset of the resultant 12-bit uncompressed and 8-, 7-, and 6-bit compressed images were read by four radiologists whose responses were analyzed with receiver operating characteristic (ROC) techniques. There were 81 images used in the ROC analysis, of which 48 were normal and 33 showed subperiosteal resorption. No statistically significant loss of diagnostic quality was detected for 8- or 7-bit compressed images, with average compression ratios of 16:1 and 28:1, respectively. Diagnostic quality was lost with 6-bit images, with an average compression ratio of 107:1.