 |
 |
Dear Jesper, after further analysis I have some follow up questions on this issue. I haven't been able to run eddy using the new 5.0.11 patch yet. However, I wanted to assess the effect on MD and FA of the spikes I was having due to the interleaved B0s and the post eddy alignment error. So I manually removed any interleaved B0s, leaving only the first B0 and the 30 diffusion gradient volumes and redid topup, eddy and DTIfit. First of all, could this be a valid approach to elude the issue of having to realign between shells? Or is there anything methodologically wrong in doing this that I might be missing? I'm attaching a couple of figures showing the resulting movement parameters for 1 subject either leaving ( https://drive.google.com/open?id=1sDHOJEeLiB_Zlq0eRqNNLN-mR_CDx6pe) or removing the interleaved B0s ( https://drive.google.com/open?id=1_ikUSbqatF507pWmv2c3WPkPIH6-6KHz). I then compared the results with those I originally got when I analyzed the data including the B0s, and running topup, eddy (with "peas") and DTIfit. This comparison between the data analyzed with and without the B0s is part of a reproducibility analysis I am running, so I took 2 control scans per subject and took the absolute difference between both scans for MD and FA. I found that mean reproducibility (i.e. error between scans) for MD and FA across grey and white matter is much larger if I don't include the B0s in the preprocessing steps than if I do. I'm attaching some figures illustrating this in case it helps. https://drive.google.com/open?id=1Dcm_a3Zz1R_C7cWJG1kFR_3LeH_5YcdK https://drive.google.com/open?id=1yrhgtcsZFOtPWpigJEe77XXiA-IE6oAH How could this be? I would have expected better reproducibility (less error between the scans) if I didn't include the B0s in the preprocessing (since movement parameters don't have the "peas" systematic error and I would have expected a more faithful motion correction). Does this make sense? Any comments or insights on this issue will be greatly appreciated. Thanks again. Regards, Florencia. 2017-12-12 14:15 GMT-03:00 Jesper Andersson <[log in to unmask]> : > Dear Florencia, > > use --dont_peas and/or update to 5.0.11. That will solve the problem. > > I know there is presently no Ubuntu version of 5.0.11, but I hope that the > official 5.0.11 release is not too far away. > > Jesper > > > On 12 Dec 2017, at 16:12, Florencia Jacobacci < > [log in to unmask]> wrote: > > Thanks for the answer. In our study, we are particularly interested in > looking at longitudinal changes in MD, so that makes me wonder how much MD > may be affected. > Looking back at some preliminary results we got, there was a huge and very > significant cluster in the parietal area when we ran the statistics with > randomise (https://drive.google.com/open?id=1mAFlv-v93f3WYCo-c_ > nuzWkc6wmt2M3t). This cluster was somewhat unexpected and we suspected it > may be some sort of artefact. > Could these results be driven by this misalignment? > Do you think it would be a good idea to get rid of the interleaved B0s > prior to running topup and eddy or would it be sufficient to use the > *--dont-peas* option to avoid having this issue? > Thanks! > > 2017-12-12 12:55 GMT-03:00 Jesper Andersson <[log in to unmask] > uk>: > >> Dear Florencia, >> >> Dear Jesper, >> Is it possible that this misalignment between the b0 and the dwi images >> will potentially affect the output of the eddy corrected images and, >> subsequently, the scalar values calculated in DTIfit? >> >> >> it is still a quite small misalignment (~0.3mm if I remember your figure >> correctly) so the impact would be small. In general FA is (almost) immune >> to misalignment between b0 and DWI, whereas for example MD is more affected. >> >> Jesper >> >> >> Thanks again for the help. >> Regards, >> Florencia. >> >> Ing. Florencia Jacobacci. >> PhD student >> Instituto de Fisiología y Biofísica (IFIBIO) - Bernardo Houssay >> Laboratorio de Fisiología de la Acción, Facultad de Medicina >> Universidad de Buenos Aires >> Argentina >> >> 2017-12-11 16:20 GMT-03:00 Florencia Jacobacci < >> [log in to unmask]>: >> >>> Dear Jesper, thank you very much for your answer. >>> Up to this point, I've been using 5.0.9 since I'm running my pipeline on >>> Ubuntu and I believe there's no patch for this OS yet. Is this correct? >>> I've been using the package that I downloaded from NeuroDebian, as >>> suggested in the installation instructions. Is there anywhere else I can >>> get the last version of FSL and eddy? >>> I am planning on re-running the pipeline so I will definitely take into >>> account using the --dont-peas flag. >>> However, I would like to be able to draw some preliminary conclusions >>> from the pipeline as it is. Do you think I could still work with these >>> movement parameters, in spite of the B0 uncertainty introduced by the >>> *peas* step, to be able to determine if all subjects move comparably? >>> Thanks again! >>> Regards, >>> Florencia. >>> >>> Ing. Florencia Jacobacci. >>> PhD student >>> Instituto de Fisiología y Biofísica (IFIBIO) - Bernardo Houssay >>> Laboratorio de Fisiología de la Acción, Facultad de Medicina >>> Universidad de Buenos Aires >>> Argentina >>> >>> >>> 2017-12-11 16:03 GMT-03:00 Jesper Andersson < >>> [log in to unmask]>: >>> >>>> Dear Florencia, >>>> >>>> I am analyzing the movement parameters that eddy outputs in the >>>> <Bassname>.eddy_parameters file. >>>> >>>> My diffusion data has 30 directions + 2B0s at the beginning, and 1 B0 >>>> interleaved every 5 b1000 volumes, for a total of 38 volumes. I've noticed >>>> that all movement parameters spike exactly at each B0 (I'm attaching a >>>> sample: https://drive.google.com/open?id=1sDHOJEeLiB_Zlq0eRqNNLN-mR_ >>>> CDx6pe). My subjects are young controls, so I wouldn't expect a >>>> massive amount of movement. >>>> >>>> I have searched through other threads in the mailing list for the cause >>>> and I found some useful info on the following two: >>>> >>>> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1210&L=fsl >>>> &D=0&P=256966 >>>> >>>> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1711&L=fsl >>>> &D=0&1=fsl&9=A&I=-3&J=on&X=955C1E2EC705B729A1&Y=fjacobacci%4 >>>> 0fmed.uba.ar&d=No+Match%3BMatch%3BMatches&z=4&P=320336 >>>> <https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1711&L=fsl&D=0&1=fsl&9=A&I=-3&J=on&X=955C1E2EC705B729A1&[log in to unmask]&d=No+Match;Match;Matches&z=4&P=320336> >>>> >>>> However, I'm not quite sure about the cause of these spikes. Is it the >>>> difference in contrast as the first thread suggests or is it that movement >>>> estimation hasn’t converged for the b=0 data, as Jesper specified in his >>>> message on the second thread? Are these parameters still useful to >>>> characterize subjects' motion in spite of these spikes or do I need to >>>> eliminate them prior to making any comparisons or running stats over them? >>>> >>>> >>>> the cause of the spikes is a misalignment between the b0 and the dwi >>>> images. The magnitude looks about right for the precision of those >>>> estimates that eddy had prior to the latest ( >>>> http://fsl.fmrib.ox.ac.uk/fsldownloads/patches/eddy-patch-fsl-5.0.11/) >>>> release. >>>> >>>> Eddy has an internal model for how the different diffusion weighted >>>> images “should” look, so is able to do a very good job of aligning those. >>>> It doesn’t have (nor is it possible to have) a model relating the b0 images >>>> to the DWIs. So for that alignment there are some options >>>> >>>> 1. Assume that the first b0 and the first DWI images are aligned. This >>>> is often a good option, especially if you are scanning cooperative subjects >>>> and your scan starts with one or two b0 followed by the first DWI. You get >>>> this option by specifying the --dont_peas option. >>>> >>>> 2. A Mutual Information (MI) based registration can be performed >>>> between the average (aligned) b0 and the average (aligned) DWI after all >>>> other alignments has been performed. This is what this is what eddy does by >>>> default since 5.0.10. >>>> >>>> Both these options are improved in 5.0.11. The MI based registration >>>> has been tweaked so that it has ~twice the precision of 5.0.10. If one does >>>> not do the MI registration (i.e. if one specifies --dont_peas), 5.0.11 >>>> still performs some MI based translation along the PE-direction, and that >>>> is a good thing. >>>> >>>> So, for you, I recommend >>>> >>>> 1. Download the 5.0.11 version >>>> 2. Run it with the --dont_peas flag, since your data seems to be suited >>>> for that. >>>> >>>> Jesper >>>> >>>> >>>> Thanks in advance for any info you can give me on this issue! >>>> Cheers, >>>> Florencia. >>>> >>>> >>>> -- >>>> Ing. Florencia Jacobacci. >>>> PhD studen >>>> Instituto de Fisiología y Biofísica (IFIBIO) - Bernardo Houssay >>>> Laboratorio de Fisiología de la Acción, Facultad de Medicina >>>> Universidad de Buenos Aires >>>> Argentina >>>> >>>> >>>> >>> >> > > > -- > Ing. Florencia Jacobacci. > > Instituto de Fisiología y Biofísica (IFIBIO) - Bernardo Houssay > Laboratorio de Fisiología de la Acción, Facultad de Medicina > Universidad de Buenos Aires > Paraguay 2155, C.A.B.A.(C1121ABG), Argentina > ☎ 5950 9500 int 2132 > > http://www.physiologyofactionlab.info/en/about-the-lab/ > https://ar.linkedin.com/in/florenciajacobacci/en > > > -- Ing. Florencia Jacobacci. Instituto de Fisiología y Biofísica (IFIBIO) - Bernardo Houssay Laboratorio de Fisiología de la Acción, Facultad de Medicina Universidad de Buenos Aires Paraguay 2155, C.A.B.A.(C1121ABG), Argentina ☎ 5950 9500 int 2132 http://www.physiologyofactionlab.info/en/about-the-lab/ https://ar.linkedin.com/in/florenciajacobacci/en