Tomographic image compression using multidimensional transforms

摘要

Since image compression in general and transform coding in particular have been studied most intensively by researchers outside the medical field, many of the resulting techniques are non-optimal when the constraints and characteristics associated with medical applications are applied. Tomographie images obtained using Positron Emission Tomography (PET) and Magnetic Resonance (MR) imaging are particularly suitable for development of application-specific compression algorithms because their high dimensionality can be exploited in the algorithm design. While a PET or MR study can be considered as a collection of 2D images and compressed accordingly, a better approach is to apply transform compression using all available dimensions. This takes maximum advantage of redundancy of the data, allowing significant increases in the compression efficiency and performance. The standard practice of modeling images using separable covariance functions is shown to be increasingly unsuitable as the number of dimensions increases. A numerical solution based on an isotropic covariance matrix gives much better results, particularly at low bit rates.