 |
 |
Thanks John and Donald for taking the time to reply Just to explain this thread fully for the archive I should add that the differences in realignment parameters were NOT to do with 3D and 4D conversion but actually due to changing the image origin. On 28 November 2011 13:12, John Ashburner <[log in to unmask]> wrote: > The transformation parameters are expected to be different. Here is a > little snippet of code to illustrate the fact that the same > transformation matrix (M) will give different voxel-to-voxel mappings > when the data have different voxel-to-world mappings (R0 and R1).... > > R0=eye(4); > R1=R0; > R1(1,4)=20; > M=spm_matrix(randn(1,6)); > R0\M*R0 > R1\M*R1 > > Using motion parameters as regressors would give very slightly > different results that are dependent on the reference space, but I > don't think the effect is likely to be large. The reason for the > different results is that rigid-body motion is actually a nonlinear > transformation (it can not be defined by a linear combination of > spatial basis functions). The matrix logarithm of a rigid-body > transform can however be described by a combination of basis > functions. > > M1 = [zeros(4,3) [1 0 0 0]']; > M2 = [zeros(4,3) [0 1 0 0]']; > M3 = [zeros(4,3) [0 0 1 0]']; > M4 = [0 1 0 0;-1 0 0 0; 0 0 0 0; 0 0 0 0]; > M5 = [0 0 -1 0;0 0 0 0; 1 0 0 0; 0 0 0 0]; > M6 = [0 0 0 0;0 0 1 0; 0 -1 0 0; 0 0 0 0]; > > p = randn(1,6); > R = expm(p(1)*M1+p(2)*M2+p(3)*M3+p(4)*M4+p(5)*M5+p(6)*M6) > > The nonlinearities could be reduced by computing the transformations > relative to some mean rigid-body transform. If you are really > interested, this (Karcher mean) could be computed as described in > Roger Woods' 2003 paper "Characterizing volume and surface > deformations in an atlas framework: theory, applications, and > implementation". > > Best regards, > -John > > > On 25 November 2011 23:37, MCLAREN, Donald <[log in to unmask]> wrote: >> John, >> I was just looking into this and found the same thing. Different rp file >> values with reoriented images. >> hdr(1).private.mat >> -3.0000 -0.0000 0 97.5000 >> -0.0000 3.0000 0.0000 -96.9129 >> 0.0000 0 4.5000 -56.4129 >> 0 0 0 1.0000 >> hdr(1).private.mat0 >> -3.0000 0 0 97.5000 >> 0 3.0000 0 -96.9129 >> 0 0 4.5000 -56.4129 >> 0 0 0 1.0000 >> and reoriented file: >> hdr2(1).private.mat >> -3.0000 -0.0000 0.0000 97.5000 >> 0.0000 3.0000 0.0000 -56.9129 >> -0.0000 0 4.5000 -86.4129 >> 0 0 0 1.0000 >> hdr2(1).private.mat0 >> -3.0000 0 0 97.5000 >> 0 3.0000 0 -96.9129 >> 0 0 4.5000 -56.4129 >> 0 0 0 1.0000 >> It looks like the .mat fields are slightly different; however, the same for >> all volumes. The question is thus, why are the realignment parameters >> different if the same data is used, just different transformation matrices. >> Additionally, since the rp files are different, using the motion regressors >> as covariates would change the models as well. >> Are the differences due to interpolation to produce the mean image resulting >> in slightly different values? >> >> Best Regards, Donald McLaren >> ================= >> D.G. McLaren, Ph.D. >> Postdoctoral Research Fellow, GRECC, Bedford VA >> Research Fellow, Department of Neurology, Massachusetts General Hospital and >> Harvard Medical School >> Office: (773) 406-2464 >> ===================== >> This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED >> HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is >> intended only for the use of the individual or entity named above. If the >> reader of the e-mail is not the intended recipient or the employee or agent >> responsible for delivering it to the intended recipient, you are hereby >> notified that you are in possession of confidential and privileged >> information. Any unauthorized use, disclosure, copying or the taking of any >> action in reliance on the contents of this information is strictly >> prohibited and may be unlawful. If you have received this e-mail >> unintentionally, please immediately notify the sender via telephone at (773) >> 406-2464 or email. >> >> >> >> On Fri, Nov 25, 2011 at 3:43 PM, John Ashburner <[log in to unmask]> >> wrote: >>> >>> I don't have any ideas here. I think you really need to take a close >>> look at the data. You can read the fields in the headers by: >>> >>> P='blah.nii'; >>> Nii=nifti(P) >>> >>> If you want to see the actual values, then you can look at a hidden >>> field of the nifti object: >>> >>> Nii.hdr >>> >>> Computing difference images via ImCalc, and displaying them via >>> CheckReg can also help to give clues. >>> >>> Best regards, >>> -John >>> >>> On 24 November 2011 10:53, Joshua Balsters <[log in to unmask]> wrote: >>> > Dear SPMers >>> > >>> > I found one previous discussion on this point but not much else, so >>> > apologies if I'm repeating what everyone already knows. >>> > >>> > I've recently starting using 4D files straight from the scanner (Philips >>> > Acheiva) in SPM rather than splitting them into 3D files using fslsplit. I'm >>> > finding in most cases there are subtle (+/-0.5mm) to drastic (over 2mm) >>> > differences in my realignment parameters depending on whether I use the 4D >>> > or 3D versions of the same datasets. These differences are only in the >>> > translations and not the rotations. >>> > >>> > It would be great if anyone could explain why this happens. I had a >>> > quick look for differences between the 3D and 4D files and I think the issue >>> > is that fslplit is automatically changing my datatype from int16 to float32, >>> > and i read in a previous discussion that it also changes the scaling factor >>> > which is probably the bigger issue. >>> > >>> > any advice on this issue would be greatly appreciated! >>> > >>> > Best >>> > Josh >>> > >> >> -- Joshua Henk Balsters, PhD Visiting Research Fellow Trinity College Institute of Neuroscience Lloyd Building Trinity College Dublin Dublin 2 Ireland Tel: +353 1 896 8405 http://people.tcd.ie/balsterj http://tcd.academia.edu/JoshuaBalsters http://www.tcd.ie/Neuroscience/niel/people/research-associates/j-balsters.php