Has this issue reproduced across subjects? Have you tried using the HCP’s CCF protocol (http://protocols.
humanconnectome.org/CCF/) for the Prisma? It looks like that is what you are trying to do and we have gotten fine results with 2mm data on the Prisma. Some important differences: 1) We are not using any partial fourier (which does not reduce distortion but does increase dropout). 2) Are you achieving an echo spacing of 0.58ms?I do not recommend GRAPPA at 3T for fMRI, this will lose you a lot of SNR.Peace,Matt.From: FSL - FMRIB's Software Library <[log in to unmask]> on behalf of Johan <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Wednesday, August 23, 2017 at 10:24 PMTherefore some data where we could see SE fieldmaps, as they should look like - would be really valuable.Dear Matt,Of the fieldmaps:
The scanner is Magnetom Prisma, Software Version E11 (B)
Resolution is 106x106 2mm^2, slice thickness 2 mm, 72 slices. Partial Fourier 6/8, and no iPat and no MB.The functionals have MB 8.
This issue regarding the SE fieldmap performing worse than the GE versions - we have it not only for this scan (A) - but also for other scan versions, even when we use iPat of 2 (row D). Of course, distortions in iPat accelerated scans are less bad. The last column RelDist is just based on visual inspection.
Scan Type MB Grappa Matrix VoxSize (mm) Slices SliceTh (mm) TR (sec) FA (deg) NVOL Task RelDist (voxels) A 8 1 106 2 72 2 0.81 53 450 3.4 B 4 1 96 2.21 60 2.2 1.24 53 310 2.39 C 8 1 88 2.4 64 2 0.735 52 490 2.6 D 3 2 86 2.5 60 2.5 1.2 67 295 1 ThanksJohanOn Thu, Aug 24, 2017 at 11:07 AM, Matt Glasser <[log in to unmask]> wrote:What scanner and resolution are you using?Peace,Matt.From: FSL - FMRIB's Software Library <[log in to unmask]> on behalf of Johan <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Wednesday, August 23, 2017 at 7:59 PMDear Jesper,Thanks for looking at it!
Do you have a data sample (of preferably a Prisma scanner) of how things (EPI, SE Fieldmaps) should look like? Otherwise its not clear what quality we should be aiming for. It'd be tremendously helpful if I can run comparisons what we have with something like a 'gold standard'.
What I can try today is to use iPat of 2 for our current SE fieldmaps, and then take care of the discrepancy with the functional using the readoutlength as defined in the parameters file that topup uses like you suggested. If iPat of 2 exactly halves the readout time, then topup should solve for a less-distorted solution and applytopup could interpolate accordingly in the functionals. It'll still be helpful to have something to check our scans against, though.Thanks and best wishesJohanPS. It could be that in our case, the total readout time (with 106/106 matrix, 6/8th partial F, but no iPat) causes distortions to a higher degree (i.e., ~3 voxels... or maybe even higher) than topup can handle. Maybe that's the issue with your other dataset - or do you think it's something more subtle?On Wed, Aug 23, 2017 at 10:54 PM, Jesper Andersson <[log in to unmask]
k>wrote:Dear Johan,I have had a look at your data and I am afraid that neither fieldmap (the SE or GE) look very good. I am not 100% sure exactly why that is, but I think that the problem is the very large distortions. It (topup) has come up with a quite creative solution where instead of pushing back all of the frontal intensity that has been pulled out, it pushes back some and leaves some, while doing the opposite for the reverse PE direction. The end result is an image pair (the --iout output) that looks surprisingly sane, but where the field is not what one would expect from a human head.I have only seen this behaviour once before, and I haven’t got a solution for it. Essentially it is a situation where the non-linear optimisation (which is what topup does) has two minima, and topup falls into the wrong one. For future acquisitions I would suggest using some IPAT (a factor 2 should be sufficient) to reduce the distortions. For the current data I can only suggest experimenting with the config file to see if you can get it to take a different path and find the other minimum.Sorry I cannot be of more help.JesperOn 23 Aug 2017, at 04:09, Johan <[log in to unmask]> wrote:JohanDear Matt,I uploaded a tarball (a second one - the first one isn't good) to the FSL upload site for inspection. I also have a screenshot but it's too big for the FSL mailinglist to accept, so I also uploaded it to the FSL upload site.Thanks again and best wishesOn Wed, Aug 23, 2017 at 11:48 AM, Matt Glasser <[log in to unmask]> wrote:The images you sent were not sufficient to evaluate the issue. Perhaps if you have some mid sagittal slices after distortion correction in which you can see some clear anatomy and compare the T1w to the fMRI.Peace,Matt.JohanThanks for your time and best wishesI sent an example 'package' of our results (topup files for GE and SE + instructions), including T1 and T1 boundary to Jesper (well - will do in a few minutes). If you wish I can also send it to you for inspection.I did actually carefully check the scans how they compare to the Anatomy - the T1. I extracted the boundary between WM and GM of the T1 scan, and compared to that boundary (and also the T1). The thing is I see too much dropout in a region where normally you wouldn't expect any dropout normally - it's too much in the 'middle' of the EPI scan. As if somehow things get too compressed, or something outside of the brain where signal is more random, gets pushed inside (but this just a guess on my part). Also when I calculate the match between T1 and EPI, using fsl FLIRT's brr cost function, I get lower values for the GE-corrected scans, indicating a better fit.Dear Matt,On Wed, Aug 23, 2017 at 11:14 AM, Matt Glasser <[log in to unmask]> wrote:Like I said on the HCP list what is most important is comparing anatomy that is within the distorted region but not the dropped out region. This is how you evaluate the quality of the distortion correction, not on the size of the dropped out region.Peace,Matt.JohanThanks and best wishesThis scan is, from the 4 scans that we did, the most 'sensitive' to distortions (about 2-3 voxels) due to a res of 2mm ^3 and 106x160 matrix. Partial F is 6/8th for both fieldmap and functionals.I will upload a sample dataset that we have to the URL Jesper gave me!Dear Matt,We have (tried) to be really contentious w.r.t. matching echo dwell time, in-plane resolution, # k-lines (EPI factor or iPat), partial Fourier (it matches the EPI), voxel sizes, tot. slices, orientation and bandwidth. Anything basically that could taper with the amount of time spent reading out one slice of k-space (which is the main thing w.r.t. distortions). We also were careful in the analysis w.r.t. possible motion happening between acquisition of SE Fieldmap and following functionals too.
The only thing different, as far as we could see, is that we used the Scanner's own SE - EPI instead of the HCP SE-EPI. I'm trying to get those installed so I can compare those with what we have so far.On Tue, Aug 22, 2017 at 9:55 PM, Matt Glasser <[log in to unmask]> wrote:Also, did you match the geometry between SE and GE images and the echo spacing? What kind of scanner do you have?Peace,Matt.Dear Johan,that does indeed look unusual. I think the best would be if I could take a look at your data. To do so, can you please send me the input files into topup for the SE and the GE cases and also a small sample (a couple of volumes) of the GE data that you want to correct? Please put all file in a tar-ball and upload it toCan you please also send me an email with the exact commands you used for running topup and applytopup?JesperOn 22 Aug 2017, at 09:44, Johan <[log in to unmask]> wrote:JohanDear Jesper,I could send you the Magnitudes that I have for you to inspect. Here the SE stuff does look a bit better/more tidy. However the EPI, when topup-SE-corrected have quite big 'holes' frontally that I really don't trust is normal dropout - or at least, it looks quite unsubtle.This figure is about the EPI - left = SE unwarped right = GE unwarped.Best wishes<image.png>On Tue, Aug 22, 2017 at 6:02 PM, Jesper Andersson <[log in to unmask]
k>wrote:Dear Johan,what does the --iout output look like for SE and GE-EPI respectively. I would expect the former to look better, i.e. less difference between AP and PA after topup. If not, I think something must be going wrong.When then applying these maps to the GE-EPI images those wont _look_ as good, because the GE-EPI images also has dropout in those same areas as there are distortions. It can be surprisingly difficult to distinguish dropout from distortion, no matter how trained your eye.JesperOn 22 Aug 2017, at 08:33, Johan <[log in to unmask]> wrote:Hi Stephen,Yes, we did carefully check all major dropout-susceptible regions of the brain and still found that the GE AP/PA 'way' somehow produced more 'complete/better looking' EPIs. The output we get from the GE AP/PA looks quite reasonable, whereas the SE AP/PA produced scans looking actually a bit worse with more dropout where we didn't suspect it. We got the same kind of result for four different kinds of EPI scans, making me thing that either our SE AP/PA protocol isn't optimal in some way (or I am just doing something wrong in the analysis).
But I still wish to figure out what's the deal with our SE AP/PA scans (se_ep2d based), exactly because in theory SE approach should work better due to absence of signal dropout in the AP and PA scans.In our analysis, we just used basic topup to get the unwarping data - then did applytopup to unwarp functionals - being careful about orientation issues. Just as written in the tutorials. We didn't mask or anything.Could I possibly as a test obtain one of your SpinEchoFieldMap AP/PA scans as well as a functional (and how it should be looking once it's unwarped)?Thanks and best wishesJohanOn Tue, Aug 22, 2017 at 4:00 PM, Stephen Smith <[log in to unmask]> wrote:Hi - GE AP/PA data is not recommended because this has the serious dropouts in the areas being most affected also by distortions - have you checked in those areas for performance of the unwarping?This email list is the best venue for sending these questions,Cheers, Steve.On 22 Aug 2017, at 06:54, Johan <[log in to unmask]> wrote:Dear experts,If I have a rather complicated question regarding Topup, SpinEcho Fieldmaps with AP / PA phase encoding directions, who would you suggest I contact?I have some data that i'd like to send. We obtained GE and SE AP and PA scans, and although SE AP/PA combination pair is recommended, for our case GE AP/PA seems to work better. We're not confusing this with the dual-echo B0 Fieldmaps.Thanks and best wishesJohan