Compression research continues as image volume grows

February 23, 2007

Storage and transmission of medical images still pose significant challenges as the number and size of imaging studies both continue to grow, stimulating further compression research.

Storage and transmission of medical images still pose significant challenges as the number and size of imaging studies both continue to grow, stimulating further compression research.

One new paper from Taiwan (Comput Med Imaging Graph 2007;31(1):1-8) proposes an adaptive prediction-based compression algorithm for lossless compression that aims to overcome some of the shortcomings of traditional wavelet-based coders.

"Compression of medical images must be lossless because a minor loss may result in a serious diagnostic consequence," said Din-Chang Tseng, Ph.D., a computer scientist at Taiwan's National Central University. "A high-ratio lossless compression method is our pursuit."

Having a predictor function that adapts to each image's individual characteristics would result in higher compression (in most cases) versus using a fixed prediction scheme for all images.

"Instead of relying on a fixed number of predictors on fixed locations and using only one prediction equation, we propose an adaptive prediction approach that employs three prediction equations for different wavelet subbands to achieve a more accurate prediction," Chen said.

This means prediction-based coders produce higher compression ratios when the mathematical prediction function can accurately model the original pixel values, said computer engineer April Khademi, Ph.D., of Ryerson University in Toronto.

The predicted image is subtracted from the original image, and the difference between these images is then coded, Khademi said.

"The benefits of this are that the overall dynamic range of the prediction residual is much less than that of the original image," she said.

For example, a 16-bpp (bits per pixel) image is reduced to a 2-bpp prediction residual.

Other scientists, however, believe lossy compression is the better path. There is only so much redundancy in images that can be exploited, and hence compression levels achievable with lossless methods are limited.

Lossless methods typically are unable to compress with ratios beyond 1:2 to 1:4 (depending on image type), while lossy methods can achieve compression ratios of 1:60 and more while still preserving image diagnostic quality, according to computer scientist Gerald Schaefer, Ph.D., of Aston University in Birmingham, U.K.

"Such visually lossless techniques clearly have a much higher potential, as they can, for example, be employed in remote diagnosis even over ordinary low-bandwidth telephone connections," Schaefer said.