An edge-based prediction approach for medical image compression

The types of edges found in volumetric medical image data are typically smooth due to a phenomenon known as the Partial Volume Effect (PVE). These smooth edges are very different to the sharp edges typically found in real world images. Consequently, it is not appropriate to use conventional edge-based prediction schemes with medical image data. This paper proposes a novel edge-based prediction scheme that is optimised for use with the types of edges associated with the PVE. This technique exploits prior anatomical knowledge and the characteristics of the PVE regions to accurately predict unknown voxel data. Our experimental results show that the performance of this technique in edge regions is significantly better than previously documented edge-based prediction techniques. We also show that the inclusion of our proposed technique in a standard compression scheme improves the overall level of compression that can be achieved.