Re-Normalised Sine Interpolation For Rapid Reslicing Of MRI Data

Introduction It has long been established that the appropriate way of reslicing volume MR images is to use the method of sine interpolation [l, 31. We have have found, like other authors before us, that large convolution kernels are needed in order to produce accurate reslice data. However, we have found that the major component of the error introduced from interpolation with small kernels, is due to a first order normalisation problem introduced by truncation. We demonstrate the characteristics of this problem and show how it. can be eliminated, so that accurate reslice data can be obtained with small kernels. The required changes in computation are simple and significantly reduce the processing requirement for any given interpolation accuracy. Methods The decomposable sine interpolation algorithm from X, Y, 2 to Z, y, z co-ordinates with a cosine Hann (Hamming) window is defined as follows,