JPEG2000 primer

Advanced Imaging Magazine recently ran a nice article by Stewart Taylor on the JPEG2000 codec. If you are interested in compression technology you will find it very informative.

One of the key decisions which must be made for medical imaging systems is what type of compression to use. Medical images are large, so some type of compression is desirable, but the resolution and detail of medical images is critical, so image corruption and artifacts must be avoided.

These challenges are present for most types of medical images, but they are especially important for Pathology images obtained with virtual microscopy. A typical microscope slide tissue sample may be 1" x 1/2" in size, or about 30mm x 15mm. A high-resolution scanner like the Aperio ScanScope with a 40X objective generates images with a resolution of about 1/4 micron/pixel, yielding an image of 120,000 x 60,000 pixels. Since there are 3 bytes per pixel, this is 22GB of image data!

JPEG2000 is a great compression technology based on wavelet technology which strikes a great balance between high compression and preserving image quality; compression ratios of 25:1 are often achievable. This reduces the size of a 22GB image to 850MB, still very large but much more manageable. (By comparison, JPEG compression usually yields about 15:1, with more image artifacts.)

JPEG2000 is computationally-intense compression technology, and implementations often rely on dedicated hardware to speed up compression. (Aperio uses the Matrox Morphis board for this purpose.) A nice attribute of JPEG2000 is that the computation requires is asymmetric; far more computation is required for compression than for decompression. As a result, hardware acceleration is really only needed for compression; viewing images can be done with ordinary computer hardware.