Hi,Try this tool,2014-09-12 3:07 GMT+08:00 Reid, Robert I. (Rob) <[log in to unmask]>:There are other things you can do, but none of them are particularly easy or guaranteed to work. dipy can be made to fit tensors much like dtifit, with the difference that the weights are iteratively calculated, trading speed for more accuracy. It does not guarantee positive definiteness (at least it did not the last time I checked), but it has been shown that positive definite fitting does not have much if any statistical advantage over fitting with well determined weights. Since negative eigenvalues mean that whatever signal attenuation there was in the voxel due to diffusion was overwhelmed by noise and/or registration error, better registration (eddy instead of eddy_correct, for example), a larger b, and/or larger voxels would probably help. You could also denoise the images before fitting tensors but I do not know what that would do to TBSS.
Rob
--
Robert I. Reid, Ph.D. | Sr. Analyst/Programmer, Information Technology
Aging and Dementia Imaging Research | Opus Center for Advanced Imaging Research
Mayo Clinic | 200 First Street SW | Rochester, MN 55905 | mayoclinic.org
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Barbara Kreilkamp
Sent: Thursday, September 11, 2014 1:43 PM
To: [log in to unmask]
Subject: Re: [FSL] DTI: Getting rid of negative eigenvalues before or after TBSS
Thank you very much.
Yes, I am already using the --wls option in dtifit.
Unfortunately I find the negative eigenvalues in ROIs I am interested in.
Otherwise I used standard FSL DTI processing with appropriate updating of the b-vector table. Other than using eddy instead of eddy_correct I assume there's not much else I can do. Right?
Thanks,
Barbara
On 11/09/2014 19:35, Reid, Robert I. (Rob) wrote:
Have you tried fitting the tensors with weighted least squares, i.e. the -w option of dtifit? It does a much better job of getting reasonable results from noisy data (and makes almost no difference if noise is not the problem). As you noted, dtifit does not guarantee positive definite tensors, but that can be tolerated if the spurious eigenvalues are limited to unimportant regions.
Rob
--
Robert I. Reid, Ph.D. | Sr. Analyst/Programmer, Information Technology
Aging and Dementia Imaging Research | Opus Center for Advanced Imaging Research
Mayo Clinic | 200 First Street SW | Rochester, MN 55905 | mayoclinic.org
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Barbara Kreilkamp
Sent: Thursday, September 11, 2014 12:39 PM
To: [log in to unmask]
Subject: Re: [FSL] DTI: Getting rid of negative eigenvalues before or after TBSS
Dear all,
I would appreciate some advice on this. Nevertheless, I have thought about the problem again and I have decided to do this:
To correct the tensor outputted from FSL with a tool that will resample it to the closest valid tensor (thus to avoid loosing data by thresholding negative eigenvalues).
After this I will recompute the scalar measurements and give those as input to TBSS.
Is this a valid approach?
All the best and thanks,
Barbara
On 08/09/2014 15:40, Barbara Kreilkamp wrote:
edit:
Of course I mean L1-L3 instead of V1-V3.
Thanks
On Mon, Sep 8, 2014 at 2:21 PM, Barbara Kreilkamp <[log in to unmask]> wrote:
Dear FSL experts,
I have noticed (as others before me), that the model that FSL uses is not positive-definite when estimating the tensor for DWI data.
I would appreciate some help in understanding how I can deal with this:
1. Should I threshold negative eigenvalues from V1, V2, V3 whole brain maps (before TBSS) and recompute FA/MD and give these as input to TBSS
2. Or should I register all unmodified FA/MD maps through TBSS first, apply this registration to V1,V2,V3 with tbss_non_FA as well, find the negative eigenvalues and mask the TBSS outputted FA/MD maps with this?
Please let me know if the approaches would result in different registration results etc.
Thank you very much,
Barbara