Compression of medical images using prediction and classification

We have developed and subjectively evaluated a lossy classified vector quantization (CVQ) using a subsampling and prediction decollation scheme. Both interframe and intraframe codings were evaluated using a sequence of x-ray CT images. Both 10:1 and 15:1 compression ratios were evaluated using nine head images from three patients. Thirteen radiologists evaluated the quality of images by viewing them on film, and comparing them to the original images on film. Although there are large variations in individual evaluations of image quality, there was overall agreement among all readers to a statistically significant level. With the proposed algorithm, the interframe coding approach produces better quality than the intraframe at the same level of data compression. Even though some data compressions were not statistically significant from the originals, the average responses were slightly worse than those for the original image. The effect of data compression on diagnostic accuracy was not evaluated.