Thank you, Mark, for your helpful explanation! I think I'll take the
applyxfm4d output then for the anaylsis.
Lea
Am 08.10.2007 19:21 Uhr schrieb "Mark Jenkinson" unter
<[log in to unmask]>:
> Sorry - I should have said that they are the same if
> you apply sinc interpolation in flirt (with a sincwidth
> kernel size of 7 - which is the default size). If you just
> use the defaults in flirt then they are not the same, as
> flirt will use trilinear which will be smoother (as you've
> seen). If you smooth your data after this then it
> doesn't matter than much, although you will be
> effectively getting a bit less total smoothing from the
> applyxfm4D output than would typically occur in the
> FEAT pipeline (which doesn't use sinc).
>
> All the best,
> Mark
>
>
>
>
> On 8 Oct 2007, at 18:12, Lea Krugel wrote:
>
>> Hi Mark,
>>
>> thanks a lot for your answer!
>>
>> In the meantime I calculated the difference between the flirt_applyxfm
>> output and the applyxfm4D output, using fslmaths. The resulting
>> difference
>> has a mean intensity that corresponds to about 2 % of the mean
>> intensitity
>> of the original time series (I looked that up in fslview).
>> I also noticed 2 other things: first, the flirt_applyxfm output
>> looks a bit
>> smoother than the applyxfm4d output, and second, the applyxfm4d
>> procedure
>> takes much longer than the flirt_applyxfm operation.
>>
>> The 2 % seem a rather high difference to me, so I now have to
>> decide which
>> of the two possibilities I choose... I have the intuition that the
>> applyxfm4D might be the more exact one because it takes longer to
>> calculate:-)... What would you suggest?
>>
>> Cheers,
>> Lea
>>
>>
>> Am 08.10.2007 18:46 Uhr schrieb "Mark Jenkinson" unter
>> <[log in to unmask]>:
>>
>>> Hi,
>>>
>>> That's right.
>>> There shouldn't be any real difference (although sometimes there
>>> are very slight numerical differences, but nothing big) if you only
>>> want to apply a single transformation to all the volumes in a 4D
>>> series.
>>>
>>> All the best,
>>> Mark
>>>
>>>
>>> On 8 Oct 2007, at 10:45, Lea Krugel wrote:
>>>
>>>> Dear FSL people,
>>>>
>>>> I'd like to know whether "applyxfm4D" with only one transformation
>>>> matrix
>>>> for all the volumes of a functional data set leads to somehow
>>>> different
>>>> results than flirt in combination with "applyxfm" and "init".
>>>> I'm asking because I used "applyxfm" for my functional data and
>>>> FEAT has no
>>>> problems in recognising the correct number of time points, so I
>>>> guess the
>>>> "applyxfm" command works fine here, also for functional data. My
>>>> understanding of "applyxfm4D" would then be that it is the right
>>>> tool when
>>>> applying more than one transformation matrix to the data, but with
>>>> only one
>>>> there is no difference between "applyxfm" and "applyxfm4D". Is that
>>>> right or
>>>> might there be problems with my output in later analysis steps?
>>>>
>>>> Thanks a lot,
>>>> Lea
>>>>
>>>>
>>>>
>>>> -------------------------------------------
>>>> Lea Krugel
>>>>
>>>> Max Planck Institute for Human Development
>>>> Neurocognition of decision making
>>>> Lentzeallee 94
>>>> 14195 Berlin
>>>> Tel.: +49-30-82406618
>>>> Fax: +49-30-82406616
>>>> email: [log in to unmask]
>>
>> -------------------------------------------
>> Lea Krugel
>>
>> Max Planck Institute for Human Development
>> Neurocognition of decision making
>> Lentzeallee 94
>> 14195 Berlin
>> Tel.: +49-30-82406618
>> Fax: +49-30-82406616
>> email: [log in to unmask]
-------------------------------------------
Lea Krugel
Max Planck Institute for Human Development
Neurocognition of decision making
Lentzeallee 94
14195 Berlin
Tel.: +49-30-82406618
Fax: +49-30-82406616
email: [log in to unmask]
|