Hi Barney,
it seems only fair that I try to disentangle the mess I made ;) although
I am not sure I have the answers:
> In reading the recent thread on this list titled 'More Realign, Coreg
> questions' I noted Marko Wilke's advice to carry out realignment then
> coregistration using estimates only in each case and then do a single
> interpolation as part of normalisation.
I think it is generally advisable to avoid or at least combine
interpolation steps whenever possible. This is, by the way, one of the
drawbacks when using a fieldmap: the non-linear deformations cannot be
stored in the header so the 'u' files have to be written out.
> 1. If I want to do slice-timing correction (which according to the
> explanation in the Huettel, Song and McCarthy text is necessary since
> my images were acquired using an interlaced acquisition order) but
> want to take Marko's advice to reduce the number of interpolations,
> should I do the slice timing correction before the realignment step?
Excellent question :) What I am going to offer is a not-well-informed
opinion: I think it depends on your assessment of whether the errors
introduced by motion are bigger than those introduced by slice
acquisition differences. If you have motion of 3mm, slice timing is not
likely to do a good job (because the "next" slice [in time] is already
another slice [in space]). If you have motion of 0.3mm, it just may be
helpful. People used to suggest to do slice-timing before doing anything
else to the data, but I believe the argument above is valid in this
regard. Ideally, one could also incorporate the effect of slice
acquisition differences during spatial normalization, but this, I have
been explained some time back, is not as easy as it sounds.
> 2. Why is it that 'The authors of SPM do not generally suggest that
> [slice timing] correction be used'? (This question has been asked a
> couple of times recently but doesn't seem to have been answered).
Several people tried, to be fair... I think it is theoretically a great
idea but, as it invariably involves reslicing, it invariably degrades
your data. The question then is, does your gain outweigh your data
degredation? In easy settings (long block designs, short TR), the answer
seems easy: don't do it. In difficult settings (ER, long TR, interleaved
acquisition), the answer is, well, difficult :) You could actually check
the effect in your data if you want to know for sure... Daniel did a
while back and liked what he saw (http://i8t.de/rwtxxdp6) so perhaps it
works for your setting, too.
Hope this helps (but am not sure at all :)
Marko
--
=====================================================================
Marko Wilke (Dr.med./M.D.)
[log in to unmask]
Universitäts-Kinderklinik University Children's Hospital
Abt. III (Neuropädiatrie) Dept. III (Pediatric neurology)
Hoppe-Seyler-Str. 1, D - 72076 Tübingen
Tel.: (+49) 07071 29-83416 Fax: (+49) 07071 29-5473
=====================================================================
|