Dear Jasper,
great, thanks for the quick reply.
On a slightly related topic:
Are you planning to extend Eddy with an option to resample or align the
diffusion data with another (e.g. T1w) data set? One would probably need
to specify a target and optionally a transformation matrix.
This would allow to resample/ realign the data set without an extra step
of interpolation.
Thanks for this great toolbox FSL!
Jan
On 03/17/2017 12:12 PM, Jesper Andersson wrote:
> Dear Jan,
>
> I use a 6 dof (rigid body) model for that.
>
> Jesper
>
>> On 17 Mar 2017, at 10:00, Jan Schreiber <[log in to unmask]> wrote:
>>
>> Dear Jasper,
>>
>> how many degrees of freedom are you using for aligning the average dwi:s
>> and b0:s?
>>
>> Thanks,
>> Jan
>>
>>
>>
>> On 01/31/2017 06:15 PM, Jesper Andersson wrote:
>>> Dear Vasilis,
>>>
>>>> Dear FSL experts and users,
>>>>
>>>> I recently updated my FSL version and gave a try to the new eddy version, which includes some really cool and useful features (like rms movement, bvec rotation). So I decided to estimate the motion of each participant using the restricted_rms_movement output as previously suggested in this list. However, I noticed that the ecc output is different than what I got with the previous eddy version (505 build) and it looks to me that it is worse in general (for some participants at least).
>>>>
>>>> The issue is that the algorithm detects a "jump" in motion when it goes from a b0 to a gradient volume.In most data I get restricted rms value of ~1 for 2nd volume compared to the 1st. The other values seem very small at ~0.1-0.3 range (relative to previous volume). Comparing the eddy_parameters files, there was also a big inconsistency on the values estimated by the two algorithms, with the new version assigning much higher values for the second volume.
>>>>
>>>> Visual inspecting the ecc files, I believe that the "old" version did a better job (at least in the data I checked so far). I would like to ask if someone else has experienced the same issue and if there is a way to "improve" or make the two outputs more similar?
>>>
>>> I _think_ this is what you see: In eddy all the dwi:s and the b0:s are aligned independently of each other. In the previous version of eddy there was no explicit alignment of the dwi:s with the b0:s. Instead it relied on the assumption that the first b0 and the first dwi (who are reference locations for the b0:s and dwi:s respectively) are acquired close in time, and that therefore there is no/little movement between the two. This would for example be the case if the first volume is a b0 and the second volume is a dwi.
>>>
>>> Since that release I came across several data sets where this was not the case, so I decided to include a “post eddy aligmnment of shells” (peas). This is based on averaging all dwi:s and all b0:s and do a mutual information alignment between the two. As with all mutual information alignment the accuracy is so-so (~0.3mm in the experiments I did, compared to ~0.1mm for the within shell precision), but at least it guarantees that there aren’t any really large misalignment between b0:s and dwi:s. As a “precaution” against large misalignments it performs the “peas” by default.
>>>
>>> If your data is of the kind that the first b0 and the first dwi really are very close you can probably do better by adding the --dont_peas flag. This turns off the “peas”, and you wont have the ~0.3mm uncertainty in relative location between the b0 and the dwi shells.
>>>
>>> Jesper
>>>
>>>
>>>
>>>
>>>>
>>>> I used the exact same call for both functions:
>>>> <eddy> --imain=subjXX --mask=eddy_mask --acqp=acqparams.txt --index=index.txt --bvecs=subjXX.bvec --bvals=subjXX.bval --topup=AnP_topup --fwhm=0 --flm=quadratic --out=subjXX_ecc
>>>>
>>>> Best regards,
>>>> Vasilis
>>>>
>>>> PS: I have prepared a folder with all the relevant files to upload if anyone is willing to take a look at this :)
>>>
>>
>
|