 |
 |
John, Thanks for the reply. A little bit of further clarification if you could. If I'm using the images for subtractions and the rotations are not 90 degrees and the translations are not whole voxels what should I do? I'm not asking for a prescription necessarily, although that would be nice. What I really want is as a place to read about the mathematical framework. Here's what I gather from reading through spm code. I see that: 1. spm_read_vols.m uses nearest neighbor interpolation in its call to spm_slice_vol.m 2. spm_imcalc.m uses nearest neighbor also 3. spm_realign_ui.m when using sinc uses 9x9x9 4 spm_coreg_uil.m and spm_mireg_ui.m both use trilinear interpolation to reslice. I'm having a hard time reading the c code for interpolation. This makes sense to me in general. If you want to read data in as it is on disc then you should use nearest neighbor interpolation because that's what is actually there. If you want to read an arbitrary point in space, or an arbitrary slice, or write data out at onto an arbitrary grid then you need to interpolate and there is no "truth" there. It's all approximation. The built in matlab family of interp functions uses orders of linear interpolation (nearest, linear or cubic), or alternatively spline interpolation (which is different). As far as I can tell there is no built in Matlab sinc interpolation, but I might be wrong about this. Best, Souheil John Ashburner wrote: > > > I have a T1 weighted MR imaged acquired with aproximately 1x1x1 > > resolution. It's got a .mat file with a rotation in it, and I want to > > reslice it to axial (analyze compatible) image at 1x1x1 resolution also. > > > > My one remaining question is what order of sinc interpolation to use. > > The default in the realign routine for functional data is a 9x9x9 > > kernel. How should I compute what order sinc to use for interpolatin > > high resolution images? > > It depends what you want to do with the data. If it is just for > visualisation then you dont need to use many neighbouring voxels. If you > want to create subtraction images then you would use more. > > If you are merely rotating by multiples of 90 degrees and translating by > whole voxels then nearest neighbour interpolation will produce the same > results as sinc interpolation. > > > > > The simplicstic answer is higher is better, except I've only got so many > > voxels and there's some noise in the image so I can't go too high. > > > PS - how did John Ashburner et al arrive at 9x9x9 as the default for > > functional data motion correction? > > It was a fine balance that I figured would minimise the number of complaints > about the realignment running too slowly while simultaneously minimising > objections about interpolation effects. > > As our knowledge about the sources of the artifacts changes, then the methods > we use also evolve. For example, the view from the MR experts used to be > that most of the residual variance after realignment was from interpolation > artifact. Now, we know that much of it is due to motion by distortion > interraction (see Jesper's recent paper on the subject) and various other > interractions between image artifacts and movement, as well as the fact that > all slices are not acquired simultaneously. > > The amount of residual variance in the images after realignment is probably > not an ideal indicator of the effects of using different interpolation > methods. We found that the residual variance after using trilinear > interpolation was actually less than that obtained using a good windowed > sinc. This was due to the smoothing introduced by using trilinear. > > Best regards, > -John > > -- > Dr John Ashburner. > Wellcome Department of Cognitive Neurology. > 12 Queen Square, London WC1N 3BG, UK. > tel: +44 (0)20 78337491 or +44 (0)20 78373611 x4381 > fax: +44 (0)20 78131420 > http://www.fil.ion.ucl.ac.uk/~john > mail: [log in to unmask] -- Souheil Inati, PhD Research Assistant Professor Center for Cognitive Neuroscience Dartmouth College 6162 Moore Hall Hanover, NH 03755-3570 web: http://www.dartmouth.edu/~inati email: [log in to unmask] phone: (603) 646-0173 FAX: (603) 646-1181