The first possibility is to sample artificial data sets, given as 3D analytic functions, like a sphere or the Marschner-Lobb function [11], on a bcc grid. Generally, data sets obtained via voxelization [17] can straightforwardly be generated on a bcc grid. This is especially useful for evaluating the applicability of bcc grids as the frequency content of such data can directly be controlled.
Second, there is of course the possibility to resample data sets on Cartesian grids to a bcc grid. Since this resampling step has to be performed only once when generating the new data set, an arbitrarily good reconstruction kernel (e.g., a rather wide windowed sinc) can be used.
The third and most interesting possibility is to take raw data from modalities like CT or MRI and directly generate bcc grid data sets from them. Raw data from tomography data sets (CT, PET, SPECT) is typically given by many 2D or 1D projections. Hence adapting the reconstruction algorithm for bcc grids has the potential of speeding up these typically very costly operations by almost 30%. Likewise image data acquired in the frequency domain (e.g. MRI) could be (re)sampled onto an fcc grid. We could easily acquire the samples in the frequency domain on a face centered cubic grid and use a modified inverse FFT to generate a bcc grid data set. That would allow either faster acquisition times or more accurate images when samples are acquired on a bcc grid.