On Tue, 16 Oct 2007, Matt Glasser wrote:
> For an estimate of noise wouldn't that be as simple as measuring the SNR of
> the diffusion weighted images? Susceptibility distortions often cause
> issues when trying to register a DTI dataset to a structural one. Perhaps
> you could evaluate the quality of your registration of the FA to the T1
> structural image, should you have acquired one?
I thought of using the transformation matrix obtained from
nonlin reg in TBSS (although it will confound the anatomical
variability with distortion!). I'm not very fsl savvy
though; and hence the message! :)
I will try Tim's suggestion and will report if something
interesting came up.
Thanks all for participating in discussion.
> -----Original Message-----
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
> Of Tim Behrens
> Sent: Tuesday, October 16, 2007 9:03 AM
> To: [log in to unmask]
> Subject: Re: [FSL] DTI: noise estimation
> The only thing I can think of it to use the variance map from bedpostx
> flsmaths merged_phsamples -Tstd std_ph
> flsmaths merged_thsamples -Tstd std_th
> fslmaths std_ph -add std_th std_both
> fslstats std_both -M
> you could even use a mask of the susceptibility region,
> but I don't know who accurate a measure this will be, particularly
> given that you will only have 50 samples at each voxel.
> Might be worth a try tho'
> On 16 Oct 2007, at 13:40, Najmeh Khalili M. wrote:
> > Hi,
> >>> Hi,
> >>> Is there a way to get the residuals of eddy_correction or is
> >>> there any way to estimate a metric of suseptibility distortions
> >>> present in DTI?
> >> Eddy_correction is not a model-fitting (in the way that dtifit is),
> >> so you don't get "residuals" - however, you can look at the temporal
> >> variance of the uncorrected and corrected datasets, though remember
> >> that this includes variation due to the real valid diffusion effects.
> >> To do it on either 4D dataset:
> >> fslmaths <4Ddata> -Tstd dataSTD
> >> However the majority of the variance you see here will not be related
> >> to susceptibility distortions. The geometric distortions would be
> >> modelled, for example, by using the FUGUE tool and a fieldmap (see
> >> the manual for more details).
> > Thanks for this suggestion.
> > Unfortunately I don't have the fieldmaps and this is why I was
> > hoping to get a measure of noise in each set.
> >>> More generally, I need to model the imaging noise in the
> >>> between-subject DTI analysis and I wonder if you have any
> >>> tools < or philosophy > that address this problem.
> >> Again - I'm not sure this I follow - this is another separate issue -
> >> the majority of between-subject variation will probably be neither
> >> due to susceptibility distortions nor within-subject acquisition
> >> noise, but due to real subject-subject variability in tract geometry
> >> and FA values.
> > Sure. But susceptibility distortions affect the tractography,
> > so I was looking for an objective way to control for that. (I
> > have of course excluded those subjects with gross distortions.)
> > Cheers
> > Naj