And now to add to my own answer,
MJ just pointed out that when downsampling in flirt he will smooth the images slightly prior to the actual interpolation step. That means that each voxel in the smoothed image will now have some information also from all it's neighbouring voxels. Hence, when you do the interpolation even though you are not using some voxels, these will implicitly be considered by virtue of some of their information now being included in the voxels that are used.
So, in my original mail I said you would pay a price in SNR but the situation is a little different to that. You will pay some price in SNR, because different voxels will be used to a different extent, but not as much as you would without that smoothing. On the other hand you will loose some resolution.
Jesper
On 26 Oct 2011, at 23:17, Jesper Andersson wrote:
> Dear Antonios,
>
> just to add to Wolf's answer. When you downsample you might want to consider using applywarp with the -s option instead. When downsampling flirt will only calculate the interpolated intensities for the voxel centres of the new (lower) resolution, which means that many of the original voxels will effectively be ignored. If you use applywarp with the -s option you will calculate interpolated values for a grid twice as dense (eight times as many points) and then in a second step sum over neighbouring voxels. This way you will have considered all the voxels in your original volume and end up with better SNR in your downsampled image.
>
> Good luck Jesper
>
> On 26 Oct 2011, at 19:39, wolf zinke wrote:
>
>> Hi,
>>
>> You can use as a starter the ApplyXFM gui, choose output size based on voxel dimensions. Keep the identity matrix option selected. In the shell you see the required steps after executing it from the gui.
>>
>> First, you need to create a new header with the intended voxel dimensions. This header is used as a reference for flirt. The original data will then be resampled by a 'coregistration' to the new header with the identity matrix.
>>
>> I hope this helps,
>> wolf
>>
>> On 26/10/11 20:27, Antonios-Konstantinos Thanellas wrote:
>>> Hello fsl users,
>>>
>>> I do have MRI scans of 166x256x196 (pix dim: 1.2023x0.9436x0.9477) and I would like to downsample them in only one (i.e 166x128x196) or only two dimensions (166x128x98).
>>> Is there any fsl utility to do that and how it should be applied to my data?
>>>
>>> Thank you
>>> Antonios Konstantinos
>>
>
|