Dynamic image data compression in spatial and temporal domains: theory and algorithm

D Ho,D Feng,K Chen

DOI: https://doi.org/10.1109/4233.681164

Abstract:Advanced medical imaging requires storage of large quantities of digitized clinical data. These data must be stored in such a way that their retrieval does not impair the clinician's ability to make a diagnosis. In this paper, we propose the theory and algorithm for near (or diagnostically) lossless dynamic image data compression. Taking advantage of domain-specific knowledge related to medical imaging, the medical practice and the dynamic imaging modality, a compression ratio greater than 80:1 is achieved. The high compression ratios are achieved by the proposed compression algorithm through three stages: 1) addressing temporal redundancies in the data through application of image optimal sampling, 2) addressing spatial redundancies in the data through cluster analysis, and 3) efficient coding of image data using standard still-image compression techniques. To illustrate the practicality of the proposed compression algorithm, a simulated positron emission tomography (PET) study using the fluoro-deoxy-glucose (FDG) tracer is presented. Realistic dynamic image data are generated by "virtual scanning" of a simulated brain phantom as a real PET scanner. These data are processed using the conventional [8] and proposed algorithms as well as the techniques for storage and analysis. The resulting parametric images obtained from the conventional and proposed approaches are subsequently compared to evaluate the proposed compression algorithm. As a result of this study, storage space for dynamic image data is able to be reduced by more than 95%, without loss in diagnostic quality. Therefore, the proposed theory and algorithm are expected to be very useful in medical image database management and telecommunication.