 |
 |
Hi Jesper - Thanks for your thoughtful analysis. Do you think it might be possible to reduce this effect by producing my own "older brain" FA template? If so, can you let me know if this method seems kosher: I recall on the list one method of doing is is to first warp each individual FA image to the template, produce the mean FA image, and then warp each individual FA image to this mean FA, produce a new mean FA, and repeat until there is little difference between the successively produced mean images. If this does not reduce the problem, the ultimate question is how usable this data is. I have deprojected the skeleton into each subjects native space to assess where the voxels are originating from, and it looks reasonable. If my voxel-wise analysis does not produce significance in the area of these bands, I feel I can trust the rest of the results. On Wed, Oct 13, 2010 at 12:49 PM, Jesper Andersson <[log in to unmask]>wrote: > Dear K, > > I've had a look at your data now. I agree that the immediate impression > when looking at the registered images is that something has gone seriously > wrong. > > However, looking more carefully I think it is actually just an effect of > your voxel size in combination with a quite severely atrophied brain. If I > for example look at the splenium in your original images it forms a > structure that is effectively one voxel thick that runs diagonally in a > transversal plane. Depending on exactly how it cuts a given voxel it may be > one voxel thick (with quite high FA in that voxel) or it may be two voxels > thick (with intermediate in FA in those voxels). When this diagonal line is > then warped (made thicker to match the template) and up-sampled to 1x1x1mm > one observes the strange looking pattern you have in your images. > > It is a little reminiscent of the aliasing you get when you try and > represent diagonal lines on a computer monitor. > > I think the best solution in your case would be to increase your image > resolution when imaging subjects with atrophy as severe as this. Though of > course I realize that is easier said than done. > > Good luck Jesper > > On 2 Oct 2010, at 17:30, K Rogerson wrote: > > Thanks again Jesper. I have done so and sent you an email. > > On Sat, Oct 2, 2010 at 2:54 AM, Jesper Andersson <[log in to unmask]>wrote: > >> Dear K, >> >> Thanks very much for your reply. I have looked at the original FA volumes >> visually, and can't see any indication of slice to slice signal variations >> in most of these participants. I did have one subject with that issue which >> completely threw off warping, and I have previously dropped them from the >> analysis.The severity of this banding/ringing does vary from volume to >> volume in all_FA, but is present in most of the volumes to some extent. >> >> At this point I'm wondering what other options there are to salvage this >> data - perhaps it is simply not suitable for warping to the template? >> Smoothing the eroded images (after tbss_1_preproc) makes the banding less >> obvious, but I am hesitant to smooth given how this may affect analyses. >> >> >> it is conceivable also that it is an effect of relatively large voxels in >> the original volume being resampled to much finer voxels. I think I would >> need to see the data to be able to say anything with certainty. Could you >> please upload (http://www.fmrib.ox.ac.uk/cgi-bin/upload.cgi) the original >> FA image together with the warp-file so that I can look at them? When you do >> you will be given a reference number that you can email me "off" the >> mailbase. >> >> Jesper >> >> >> >> On Tue, Sep 28, 2010 at 7:53 AM, Jesper Andersson <[log in to unmask]>wrote: >> >>> Dear K, >>> >>> >>> I am looking at some DTI data in TBSS. I followed tbss_2_reg and >>>> tbss_3_postreg scripts to transform the data to MNI space. After looking at >>>> all_FA.nii I noticed some significant banding distortion in a number of the >>>> images, though the alignment it self looks to be ok. >>>> Any idea what could be causing these problems? They are present in a lot >>>> of the participants volumes and I'm worried they are significantly impacting >>>> results (though perhaps the skeleton minimizes their contribution somewhat). >>>> Note I was aligning to the standard template. I've attached an example >>>> image to this message >>>> >>> >>> I have noticed these in the past as well, and in those cases they have >>> been caused by slice-wise signal variations that fnirt then rotated away >>> from the plane and into these diagonal bands. Can I ask you to first of all >>> go back to your original image and see if you can identify them there? They >>> are much less conspicuous when running parallel with the planes. >>> >>> Jesper >>> >> >> >> > >