Does -fineblur in 3dAllineate correspond to -1blur_sigma, -1blur_rms or -1blur_fwhm in 3dmerge?

AFNI Message Board
1

Dear AFNI gurus,

Does -fineblur in 3dAllineate correspond to -1blur_sigma, -1blur_rms or -1blur_fwhm in 3dmerge?

I'm trying to visualize data as 3dAllineate sees it, and the blur seems like the most important thing.

Thanks for the help!
Sebastien

2

Hi, Sebastien-

I don't think that -fineblur applies to the output data, but to the intermediate processing during estimation of alignment.

Note that there will always be some blurring/interpolation incurred by regridding, as well as by alignment (which includes regridding). You can choose the interpolant used for creating the final output dataset to have different properties, depending on what you want. I think you would want to use -final wsinc5 to minimize the blurring in the final image. Mathematically, this should be the best function to use to preserve edges within the dataset. The slight tradeoff of using it is that it produces a very slight ringing, which can be seen typically in background outside the brain afterwards. That ringing typically does not affect much, and could be purged by playing around with masking and dilation+erosion if necessary.

From the 3dAllineate help description about this option:

 -final iii  = Defines the interpolation mode used to create the
               output dataset.  [Default == 'cubic']
            ** N.B.: If you are applying a transformation to an
                       integer-valued dataset (such as an atlas),
                       then you should use '-final NN' to avoid
                       interpolation of the integer labels.
            ** N.B.: For '-final' ONLY, you can use 'wsinc5' to specify
                       that the final interpolation be done using a
                       weighted sinc interpolation method.  This method
                       is so SLOW that you aren't allowed to use it for
                       the registration itself.
                  ++ wsinc5 interpolation is highly accurate and should
                       reduce the smoothing artifacts from lower
                       order interpolation methods (which are most
                       visible if you interpolate an EPI time series
                       to high resolution and then make an image of
                       the voxel-wise variance).
                  ++ On my Intel-based Mac, it takes about 2.5 s to do
                       wsinc5 interpolation, per 1 million voxels output.
                       For comparison, quintic interpolation takes about
                       0.3 s per 1 million voxels: 8 times faster than wsinc5.
                  ++ The '5' refers to the width of the sinc interpolation
                       weights: plus/minus 5 grid points in each direction;
                       this is a tensor product interpolation, for speed.

--pt

3

Thanks, though I am not concerned with blurring of 3dAllineate's output, it's the blurring of its input I want to visualize. As I understand, 3dAllineate takes some images as input, blurs them according to -fineblur, finds the best transformation, then applies the transformation to the original non-blurred images to produce the output images. I only want to visualize the blurred input.

thanks again for the help and
Have a great day!
Sebastien

4

-1blur_fwhm is what is used (see function GA_smooth in mri_genalign_util.c -- trust in the source, Luke).

5

Awesome, thanks!