Object-based coding of volumetric medical data

Medical data are increasingly represented as volumes. The huge amount of data to be handled every day in clinical practice makes compression an unavoidable step in picture archiving and communication systems (PACS). We propose an object-based coding scheme for 3D datasets, providing full range quality scalability for each semantically meaningful region. An integer three-dimensional separable wavelet transform is performed on the whole volume. An analytic model based on hybrid ellipsoids is used to describe the different objects, which are then coded independently allowing random access. The adopted coding strategy is the 3D extension of the well known EZW approach. First results show that 3D EZW outperforms both JPEG and the bi-dimensional version of the same coding algorithm in both lossless and lossy modes.