Hi,
Using "-nosearch" is not the same as doing no optimisation.
If you really don't want it to do optimisation, then just use
the -applyxfm mode as this will simply apply the initial
transform calculated from the s/qforms.
The -displayinit option also allows you to see what
flirt is using as the initial matrix.
If you are still having trouble and can't get a good
registration then please let us know.
All the best,
Mark
On 15 May 2009, at 12:46, Erlend Hodneland wrote:
> Hi,
>
> Thank you very much for you reply!
> My "sform_code = 1" in both the refence image and the input image.
>
> My registration results are not good, even if I use "-usesqform" or
> not. The
> results are similar, but dont look good.
>
> My wish is to make flirt use ONLY the sform information, and not
> performing
> any optimization. I tried this by setting "-nosearch" in addition to
> "-usesqform" but the results are still not good, as I obtain in
> Matlab with
> my own code.
>
> Do you have any suggestions?
>
> Erlend
>
> On Fri, 15 May 2009 09:30:30 +0100, Mark Jenkinson <[log in to unmask]
> > wrote:
>
>> Hi,
>>
>> Flirt does not use the sform or qform information for registration by
>> default.
>> It does use (and relies upon) the voxel sizes in mm.
>> If you turn on the -usesqform option then it will initialise the
>> registration
>> with the sform or qform information, but *only* if one of the qform
>> or
>> sform
>> codes is non-zero (otherwise they are both considered "unknown" and
>> they are ignored).
>>
>> So if you see no difference when using -usesqform then I would
>> imagine
>> that
>> either the codes are both zero (in either image) or that the initial
>> alignment
>> imposed by this is sufficiently similar to the default, or one of the
>> many initial
>> starting positions that flirt tries at the start of the optimisation,
>> that it leads to
>> the same solution. This would make sense if the registration was
>> succeeding.
>> Is the registration working well?
>>
>> All the best,
>> Mark
>>
>>
>> On 15 May 2009, at 09:25, Erlend Hodneland wrote:
>>
>>> I am trying to to multimodal registration using flirt. When I open
>>> the input
>>> nifti files I see they have header information for the real space
>>> coordinates inside the "sform" field. My questions are as following:
>>>
>>> 1. Is flirt using any header information for registration? To me, it
>>> seems
>>> so, since I get different results if I run flirt the same .nii image
>>> with
>>> and without the header.
>>>
>>> 2. If header information is used, why are the results not improved
>>> when I
>>> use the option "-usesqform" in flirt, compared to the situation
>>> without this
>>> option?
>>>
>>> Best regards
>>>
>>> Erlend Hodneland
>>>
>
|