Hi,
Sorry to take so long to reply.
There are not obvious things to tune in robustfov, but in the past it has done well on a wide range of images, however, it does make a few assumptions and it sounds like your image domain PI might be breaking them. It is definitely possible that lost of exactly zero voxels could cause a problem, but it does depend on how they are distributed. The main assumption that robustfov makes is that there is only noise above the head and not any strong artefacts or highly non-Gaussian distributions (though it should ignore the zeros). It then determines the first slice containing head by the fact that the intensities associated with the head are a long way into the tail of the distribution.
So it is certainly possible that this is a problem but seeing some pictures would be helpful - both some axial slices and some sagittal and coronal slices. You can’t attach things to emails on the list, so it is better to just provide a link to a place where you can host the pictures.
One workaround that might work (or not, as I can’t really tell without seeing the images) is just to add a very small amount of noise to the image, as that can often make robustfov happier, although it obviously slightly degrades the quality. So you could just add noise to the problem ones, run robustfov on the noisy version, apply the same crop to the original (without noise added), and then feed this result into fsl_anat with the —nocrop option. This would probably work for all of your images (assuming it works for these unusual cases).
So do send us some pictures and I’ll take a look.
All the best,
Mark
> On 19 Jul 2023, at 04:11, Patrick Sadil <[log in to unmask]> wrote:
>
> I've recently processed around 1000 T1w scans with fsl_anat (default settings), most of which go through just fine. However, fsl_anat is not producing a complete set of outputs for a subset of the scans. These failing scans do not exhibit any obvious quality issues. The processing seems to go haywire early, with the FOV cropping by robustfov; the "T1" output on these failing images all have inferior portions of the brain cropped (whether fsl_anat reports an error varies, but one error that sometimes occurs is an inability to segment the brainstem during the FIRST step).
>
> Most of the images that are processed successfully were collected with k-space parallel imaging, but the ones that fail were all collected with image domain PI. I suspect that the image domain PI might be throwing off the heuristics used by robustfov by causing many of the background voxels to have an intensity that is far from the intensity of the brain voxels (most voxels in the background are exactly 0 whereas brain voxels have an intensity of around 10k). When calling robustfov alone, the slice along the z-axis that is reported as the top of the brain is far from the skull.
>
> So, a couple of questions
> - Does that (lots of 0-intensity background voxels) sound like a plausible source of the issue?
> - If so, is there a recommended way to configure fsl_anat to handle these kinds of images?
>
> It seems like there are several ways in which we can get things to work (e.g., skip cropping, do the cropping ourselves, replace the 0 intensity voxels with some helpfully random value, etc), but also it would be nice to process these images without giving them special treatment.
>
> Let me know if it would be useful to see pictures or samples.
>
> Thanks!
>
> ########################################################################
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
>
> This message was issued to members of www.jiscmail.ac.uk/FSL, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
This message was issued to members of www.jiscmail.ac.uk/FSL, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
|