Multimedia information, in particular medical imaging, consists of enormous quantities of data. No wonder, health care providers are submitted to the pressing constraint of keeping the medical image interchange and storage practicably manageable, without losing sight of legal requirements. Data compression offers a useful solution but always involves a certain amount of quality reduction due to irreversible loss of information or the possible introduction of artefacts. The required image quality depends on the context in which the data will be used and hence determines which kind or degree of compression has to be applied. In addition, hospitals deal with a serious archival problem and need to consider how to draw the line between on-line and off-line storage systems. Wavelet compression might offer an answer but legal implications have to be taken into account. Euromed project manager Andrew Marsh outlines the current situation.
Obviously, health care information transmission for diagnostic use requires a completely different type of compression than the simple browsing of a medical images file, just to check out its contents. In fact, there exist two major classes of compression algorithms: the lossless or reversible kind, which allows exact reconstruction of the original information, and the lossy or non-reversible one, where information fragments selectively are rejected during the process. The lossless algorithms have to suit exactly the data that need compressing in order to obtain a reduction in the data volume instead of an expansion. Reconstruction of lossy compressed information on the other hand, only offers an approximation to the original images and moreover, degradation largely increases with higher compression ratios. Yet, lossy compressed high quality information may still be superior to an initial lower quality image.
The choice between lossless and lossy compression and their various degrees finally depends on three important factors. First, the modality which has been used to acquire and digitise an image; second, the purpose for which the information has to serve; and third, the characteristics of the depicted object. Only for the second one, the user has an input. The level of image quality can be adapted to the final application and is divided into six categories:
- original acquisition quality
- a "gold standard", which can be equalled by lossless and approximated by lossy compression but never surpassed; compression ratio= 2:1
- perceptually lossless
- a quality reduction designed to match the information content to the limits of human perception; compression ratio=12:1
- functionally lossless
- a further quality reduction designed to match the data to the purpose for which the image will be used; compression ratio=25:1
- matched to the reproducing device
- a variable criterion since quality standards are established according to the used device
- thumbnail quality
- no fine detail needed for just recognising the object if browsed at speed; compression ratio=200:1
- icon quality
- very low quality, no information content required, simply used to indicate whether or not the object is present; compression ratio= >200:1
Hospital archival needs constitute yet another challenge, given the fact that over a one year period, a disc storage of 6.000.000 MB is required for one single general hospital. Legal requirements include that images should be stored for as long as they might be needed, which makes it impossible to run a cost-effective on-line storage system. Time-depth and capacity limits force hospital staff to organise an off-line archive. Besides, statistics are revealing that less than 10% of the images are accessed ever again after the first year. Therefore, it should be reasonable to provide on-line archive storage for at least two years of image production. Older images can be accessed with the help of a human operator to insert the storage media into the global system. In this case, automatic instructions have to be supplied on how to identify the location of the concerned media and the action to be performed.
It is foreseeable that large archiving sites will be set up, allowing hospitals to download the majority of their data to an external site. This again raises the question of data compression with relation to its legal implications. A great deal is expected from wavelet compression techniques, which currently are investigated at the University of Bucharest in Romania within the frame of the Euromed partnership. More information about compression and archival solutions can be found at the Euromed web site as well as in the VMW-analysis contribution "Euromed work in progress".