Based on a study we did, including stage 4 gives you better results so if you have enough computation time I would recommend doing it. However, if you are trying to save time and the motion is not excessive using the default option is fine.
Best,
Ivana
On 23 Dec 2010, at 16:00, Michael Waskom wrote:
> Thanks Ivana, that's very helpful.
>
> What are the considerations that would bear on using a final stage of
> searching with sinc interpolation? Is it always going to give you
> better results, and is it just a question of whether you want to give
> up the increased computation time? Or are there diminishing returns
> if you're not working with a population that has excessive motion?
>
> Cheers,
> Mike
>
> On Wed, Dec 22, 2010 at 7:00 PM, FSL automatic digest system
> <[log in to unmask]> wrote:
>> Date: Wed, 22 Dec 2010 22:39:51 +0000
>> From: Ivana Drobnjak <[log in to unmask]>
>> Subject: Re: MCFLIRT sinc options
>>
>> Hello,
>>
>> Option -stages refers to the process of finding the parameters of the transformation (translations, rotations). When calculating the dissimilarity between the reference image (IREF) and the image we want to register to the reference image (IM), interpolation needs to be applied to IM. The interpolation choices in MCFLIRT are: trilinear, sinc and nearest neighbour. Different choices of the interpolation function are used for different stages of the algorithm, and sinc is used for the 4th stage (which is optional).
>>
>> Once the minimisation process is finished and all of the transformation parameters are estimated, a transformation is applied to the image IM. This transformation can be applied using either trilinear or sinc or the nearest neighbour interpolation method (the default in MCFLIRT is trilinear interpolation). If you specify sinc_final it will be sinc.
>>
>> SO in short: sinc in stages helps to find the right transformation, sinc_final applies sinc interpolation to the image when applying this transformation.
>>
>> Hope this helps,
>> Ivana
>>
>>
>> On 22 Dec 2010, at 19:01, Michael Waskom wrote:
>>
>>> Hi Mark (or whoever else might know),
>>>
>>> Could you help me understand the various options pertaining to sinc
>>> interpolation in MCFLIRT? Looking at the usage in version 4.1.6, it
>>> seems that there are two different options, "-sinc_final" and setting
>>> the "-stages" argument to 4. Running MCFLIRT with a permutation of
>>> these flags (in other words, calling it three times with the same
>>> input and reference volume but using both, or either one in isolation)
>>> seems to produce three different images (according to a diff between
>>> them). But, what is actually going on differently in the algorithm?
>>> Unfortunately, the online MCFLIRT user guide seems not to have any
>>> information on the -sinc_final option. Thanks!
>>>
>>> Cheers,
>>> Michael
>>>
>>
>
|