Hi Mark, Thanks for the link. According to the formula (36), M = T2*inv(T1) - I = xmat_1_to_2 - I when I put T1 = identity and T2 = xmat_1_to_2 M = inv(xmat_1_to_2) - I when I put T2 = identity and T1 = xmat_1_to_2; By this, M seems not invariant to the order of T1 and T2? Best, Leo On Tue, Jan 31, 2012 at 3:00 PM, Mark Jenkinson <[log in to unmask]> wrote: > Hi, > > No, that's not the implementation that we use. > See: http://www.fmrib.ox.ac.uk/analysis/techrep/tr99mj1/tr99mj1/node5.html > > All the best, > Mark > > > On 31 Jan 2012, at 22:56, Yiou Li wrote: > >> Hi Mark, >> >> Thanks for the prompt feedback! As I understand, the rmsdiff measure >> is invariant to the order of transform matrices just as |X1 - X2|^2 is >> symmetric to both variables? >> >> Thanks! >> Leo >> >> On Tue, Jan 31, 2012 at 2:44 PM, Mark Jenkinson <[log in to unmask]> wrote: >>> Dear Leo, >>> >>> There is no right answer. >>> The two different ways just average over a slightly different >>> spherical ROIs. However, the difference will be pretty minor >>> and I can't think of an example where these small differences >>> would matter. >>> >>> All the best, >>> Mark >>> >>> >>> >>> On 31 Jan 2012, at 22:08, Yiou Li wrote: >>> >>>> Hi Mark, >>>> >>>> I want to use rmsdiff to measure the difference between image X1 and >>>> X2 in terms of affine linear transformation, to do that, I ran flirt >>>> to register X1 to X2, taking X2 as reference image and get the >>>> transformation xmat_1_to_2, now I got two slightly different number by >>>> running: >>>> >>>> 1. rmsdiff xmat_identity xmat_1_to_2 X2 >>>> 2. rmsdiff xmat_1_to_2 xmat_identity X2 >>>> >>>> Can you please advise which one is the correct order in this case? >>>> >>>> Thanks in advance! >>>> Leo >>>> >>