MEDICAL IMAGE COMPRESSION AND VECTOR QUANTIZATION

In this paper, we describe a particular set of algorithms for clusteri ng and show how they lead to codes which can be used to compress image s. The approach is called tree-structured vector quantization (TSVQ) a nd amounts to a binary tree-structured two-means clustering, very much in the spirit of CART. This coding is thereafter put into the larger framework of information theory. Finally, we report the methodology fo r how image compression was applied in a clinical setting, where the m edical issue was the measurement of major blood vessels in the chest a nd the technology was magnetic resonance (MR) imaging. Measuring the s izes of blood vessels, of other organs and of tumors is fundamental to evaluating aneurysms, especially prior to surgery. We argue for digit al approaches to imaging in general, two benefits being improved archi ving and transmission, and another improved clinical usefulness throug h the application of digital image processing. These goals seem partic ularly appropriate for technologies like MR that are inherently digita l. However, even in this modern age, archiving the images of a busy ra diological service is not possible without substantially compressing t hem. This means that the codes by which images are stored digitally, w hether they arise from TSVQ or not, need to be ''lossy,'' that is, not invertible. Since lossy coding necessarily entails the loss of digita l information, it behooves those who recommend it to demonstrate that the quality of medicine practiced is not diminished thereby. There is a growing literature concerning the impact of lossy compression upon t asks that involve detection. However, we are not aware of similar stud ies of measurement. We feel that the study reported here of 30 scans c ompressed to 5 different levels, with measurements being made by 3 acc omplished radiologists, is consistent with 16:1 lossy compression as w e practice it being acceptable for the problem at hand.