On Thu, Jun 9, 2016 at 5:49 AM, Thomas Nichols <t.e.nich...@warwick.ac.uk>
wrote:
> Hi Tim,
>
> Thanks for taking the time to delve into this.
>
> wb_command does have surface to volume mapping now.
>>
>
> Just to be clear, which command are talking about? Using the output of
> wb_command -cifti-export-dense-mapping in combination with a (say)
> *32k_fs_LR.surf.gii file?
>
-metric-to-volume-mapping and -label-to-volume-mapping. If you choose to
go this way (since your new stuff is surface based, I would choose to
compare in surface space instead, if possible), you will need to use
-cifti-separate first to get single-hemisphere gifti files, and you may
need to get a bit creative in combining the volume files for each
hemisphere (-volume-math with 'left + right * (left == 0)', concatenating
the label tables from -label-export-table, and then use
-volume-label-import).
> However, going that direction, especially with group average data, means
>> you aren't comparing apples to apples. Volume-based analysis generally
>> ignore places where the cortical ribbon isn't overlapping between subjects,
>>
>
> I'm not sure I know what you mean by 'ignore'. In volume land, where the
> alignment isn't perfect you get blurring of the inter-subject average.
> This isn't great, but we also add more blurring with some smoothing. So we
> always know in a typical volume analysis of fMRI data, if you are thinking
> at fine-grained level, that a given voxel is a mixture of different bits of
> cortex and WM.
>
Perhaps 'ignore' isn't the right word - basically, if volume registration
doesn't get the cortex to overlap, the volume methods generally say "eh,
close enough" and don't try to establish any better mapping between
subjects' cortical ribbons. Smoothing can increase signal overlap where
there is little overlap to begin with, but the resulting signal overlap is
still weaker, with a longer spatial tail, than places where cortex was well
aligned, and of course it mixes in csf and white matter signal everywhere
(tolerable for BOLD activation, but you wouldn't want to make group average
myelin maps this way), and if you smooth by more than, say, the cortical
thickness, you smooth straight across sulcal banks, which is bad.
> but in return, the location of the cortex doesn't move around during group
>> averaging.
>>
>
> I'd say any intersubject volumetric warping is very much 'moving things
> around', but I fear I may be misunderstanding you.
>
Right, the warping during volume registration moves things, but when you
average after that, that doesn't introduce additional movement. In
contrast, if you average the surface coordinates of subjects after
registration, the resulting surface coordinates CAN move away from the
coordinates where any subject's cortex actually was, even the coordinates
after volume registration. This makes group average surfaces more of a
challenge to work with.
> Surface-based analysis acknowledges the volume mismatch of the placement
>> of the cortical ribbon, and removes it from the problem of cross-subject
>> alignment, but in return presents a problem if you try to put it back into
>> a volume space, as there was not agreement in the first place of what
>> coordinates each feature should be at, and the obvious workarounds
>> generally result in the coordinates being moved in the direction of average
>> local curvature.
>>
>
> I think I understand this imperfection in moving a surface back to the
> volume in a common atlas space. But I also think of the various inevitable
> imperfections, in particular those subjects that have idiosyncratic
> foldings that no diffeomorphic registration (surface or volume) will be
> able to perfectly map to a common space. So we're always working with
> imperfect mappings.
>
If you align based on areal features (MSMAll), you don't have to match
folding patterns when they are different (or at all). Yes, you can still
have mismatches in topologies of area arrangements, but as a 2D
registration of only cortex (even using only folding for registration), the
problem is simpler and can't do same the kind of cortex/csf/white matter
overlap that volume registration is stuck dealing with.
> I don't see a good way around this problem, other than coming up with a
>> way to compare parcellations based on how they parcellate data in their own
>> native representation, rather than how they align with each other in a
>> coordinate space. I was thinking maybe you could use the signed distance
>> functions from the subjects to artificially add the folding definition back
>> into the surfaces, but this again suffers due to volume misalignment
>> problems, especially for the thin between-surface areas in pial and white
>> surfaces, which probably makes it completely unworkable. This is beside
>> the fact that it wouldn't be really matching folding patterns to their
>> "correct" locations with respect to the parcellation (not least because the
>> parcellation may not be defined in relation to folding patterns at all!).
>>
>
> I am facing a really rather basic question: I am using clustering methods
> on the dense connectome (and volume data) that know *nothing* about space.
> I now want to know how well they produce clusters that are phsiologically
> reasonable. I'm not looking for "truth". I'm looking for a (possibly
> imperfect) measure of biologically feasibility of multiple partitions to be
> able to compare the different methods.
>
>
>> Yes, I'm aware that it seems like we are being difficult, but if you want
>> to compare things fairly and correctly, mapping group data from surface to
>> volume is probably not the answer. Likewise, mapping from volume to
>> surface using a group average surface is not recommended, but might not be
>> as bad as going from surface to volume? On the other hand, if it is only
>> the parcellation being mapped, and you aren't worried about the borders
>> between areas shifting somewhat, these problems might not be as bad.
>>
>
> Exactly! Modest shifting of boarders doesn't trouble me so much, as that
> is a error that affects my 'truth' and should simply lower the floor on how
> good each of my dense connectome clusterings will work, but shouldn't
> invalidate the comparison between different clustering methods.
>
To compare cifti-based clusterings to each other relative to some standard,
I would suggest using another cifti-based parcellation, even if you have to
make it imperfectly from dilating a volume parcellation and mapping it to
group average surfaces. You don't want to have to degrade the cifti
clusterings by putting them through surface to volume mapping.
Alternatively, you could try using a single subject's surface to map the
volume parcellation onto, avoiding some of the problems of a group average
surface.
> -Tom
>
>
>>
>> Tim
>>
>>
>> On Wed, Jun 8, 2016 at 2:19 PM, Dierker, Donna <do...@wustl.edu> wrote:
>>
>>> Tom has a volumetric parcellation in MNI space that he has used for
>>> other analyses, so he needs to use the same parcellation for comparison
>>> with the dconn and dscalar fancy graph stuff.
>>>
>>> It might be possible to map the parcellation volume to the S900
>>> midthickness and use the mapped parcellation for analysis in surface-land.
>>> You can gauge the viability of this method by viewing the parcellation
>>> volume over its corresponding mean T1/T2 with the S900 midthickness contour
>>> overlaid.
>>>
>>> Freesurfer has mri_surf2vol, and caret5 has an option for projecting
>>> from surface to volume, but I am not aware of a feature like this in
>>> wb_command.
>>>
>>>
>>> On Jun 8, 2016, at 12:59 AM, "Glasser, Matthew" <glass...@wustl.edu>
>>> wrote:
>>>
>>> > Hi Tom,
>>> >
>>> > We don’t explicitly do anything to correct surface coordinates for
>>> shrinkage. Tim might have an idea if this were possible, but I think it
>>> may be an uncorrectable nonlinear geometrical effect of averaging
>>> incompatible folding patterns (though it nicely shows where areas and folds
>>> are well aligned and folds are consistent across subjects by being much
>>> more similar to the average volume in early sensory areas, insula). It
>>> definitely affects pial surfaces more than white matter surfaces. I’ve
>>> attached an example (green whitematter surface, red midthickness surface,
>>> blue pial surface). This is what happens to neuroanatomically aligned
>>> corresponding points in the brain (i.e. MSMAll matched surface vertices in
>>> individuals) when you average in the volume.
>>> >
>>> > Matt.
>>> >
>>> > From: <ten.pho...@gmail.com> on behalf of Thomas Nichols <
>>> t.e.nich...@warwick.ac.uk>
>>> > Date: Wednesday, June 8, 2016 at 12:21 AM
>>> > To: Matt Glasser <glass...@wustl.edu>
>>> > Cc: HCP Users <hcp-users@humanconnectome.org>
>>> > Subject: Re: [HCP-Users] Location of CIFTI vertices in standard space
>>> >
>>> > Hi Matt,
>>> >
>>> > Thanks for the clarification. I'm aware of the merits of the surface
>>> approach (no one would put up with all it's headaches otherwise!) but I'd
>>> guess I'm not alone in needing "an answer" of the question how to move from
>>> surface back to volume space. I know there are limitations, but "don't do
>>> it" can't be the answer.
>>> >
>>> > So! What what I'll try now is using the MNI coordinates for the S900
>>> Group Average space from Q1-Q6_R440.{L,R}.midthickness.32k_fs_LR.surf.gii
>>> (in the "900 Subjects Group Average Data" package).
>>> >
>>> > The only remaining question is whether there is any volume-based
>>> correction for the 'shrinkage' you mention... is there some affine
>>> transform that should be applied to the coordinates from the
>>> *.midthickness.32k_fs_LR.surf.gii?
>>> >
>>> > -Tom
>>> >
>>> > On Wed, Jun 8, 2016 at 5:53 AM, Glasser, Matthew <glass...@wustl.edu>
>>> wrote:
>>> >> Hi Tom,
>>> >>
>>> >> I think Tim’s earlier e-mail is a good explanation. It is not
>>> possible to maintain the spatial localization standards and quality of the
>>> analysis approach when making a comparison with average/atlas data in
>>> volume MNI space. In contrast one can go from an individual’s physical
>>> volume space to standard CIFTI space via MSMAll and back again without a
>>> loss of quality aside from interpolation effects. This difference is due
>>> to the information loss that comes with group averaging misaligned volume
>>> data. Surface mesh XYZ coordinates suffer the same issues (e.g.
>>> topologically incompatible folding patterns, differences between folds and
>>> cortical areas, etc), and as a result group average surfaces shrink to some
>>> degree (a shrinkage that we can correct for on the surface for some
>>> computations using average vertex areas) and the folding patterns are
>>> blurred in many higher cortical regions. As I mentioned below, there are
>>> group average midthickness surfaces in the "900 Subjects Group Average
>>> Data” package that have average XYZ coordinates in MNI space for display
>>> purposes. However, I wouldn't recommend mapping directly between them and
>>> MNI space, as this causes well documented problems to the underlaying
>>> neuroanatomy (e.g Glasser and Van Essen 2011, Van Essen et al 2012). These
>>> are the same problems caused by volume-based cross-subject averaging of
>>> cortical data in any study.
>>> >>
>>> >> Matt.
>>> >>
>>> >> From: <ten.pho...@gmail.com> on behalf of Thomas Nichols <
>>> t.e.nich...@warwick.ac.uk>
>>> >> Date: Tuesday, June 7, 2016 at 11:11 PM
>>> >> To: Matt Glasser <glass...@wustl.edu>
>>> >> Cc: HCP Users <hcp-users@humanconnectome.org>
>>> >>
>>> >> Subject: Re: [HCP-Users] Location of CIFTI vertices in standard space
>>> >>
>>> >> Hi Matthew,
>>> >>
>>> >> We've already done a bunch of comparisons with data in volume MNI
>>> space, and so we need to continue to use the same volume-based labels with
>>> the dense brainordinate connectome.
>>> >>
>>> >> -Tom
>>> >>
>>> >> On Tue, Jun 7, 2016 at 11:30 PM, Glasser, Matthew <glass...@wustl.edu>
>>> wrote:
>>> >>> Hi Tom,
>>> >>>
>>> >>> Couldn’t you just use something already in surface space like the
>>> FreeSurfer labels? Which “known anatomical labels” are you wanting to use?
>>> >>>
>>> >>> Peace,
>>> >>>
>>> >>> Matt.
>>> >>>
>>> >>> From: Thomas Nichols <ten.pho...@gmail.com>
>>> >>> Date: Tuesday, June 7, 2016 at 5:28 PM
>>> >>> To: Matt Glasser <glass...@wustl.edu>
>>> >>> Cc: Thomas Nichols <t.e.nich...@warwick.ac.uk>, HCP Users <
>>> hcp-users@humanconnectome.org>
>>> >>> Subject: Re: [HCP-Users] Location of CIFTI vertices in standard space
>>> >>>
>>> >>> Thanks Tim, Matt, Donna for the speedy replies!
>>> >>>
>>> >>> Here's what we are doing: We are using fancy graph methods to
>>> cluster the a dense connectome by various methods (i.e. each surface vertex
>>> is assigned to a latent group). Now we want to compare our clustering to
>>> known anatomical labels in volume MNI space.
>>> >>>
>>> >>> So is a file like
>>> 100307_tfMRI_EMOTION_level2_hp200_s2_MSMAll.dscalar.nii in a standard
>>> coordinate space? (I know this isn't a dense connectome, but I'm also
>>> helping someone do something similar with task data.) The "900 Subjects
>>> Group Average" space? And I'll find a surf.gii file there?
>>> >>>
>>> >>> -Tom
>>> >>>
>>> >>> On Jun 7, 2016, at 9:28 PM, Glasser, Matthew <glass...@wustl.edu>
>>> wrote:
>>> >>>
>>> >>>> Hi Tom,
>>> >>>>
>>> >>>> We should probably list the valid structure names in wb_command
>>> -cifti-export-dense-mapping, but you want CORTEX_LEFT and CORTEX_RIGHT.
>>> >>>>
>>> >>>> As for what surface to use, you would need to be using some kind of
>>> group average midthickness surface (go in the DB and find something called
>>> "900 Subjects Group Average Data").
>>> >>>>
>>> >>>> I would recommend giving me a few more details (off list if
>>> necessary), on what you are doing though just to make sure you are all of
>>> the advice that I can provide. ;)
>>> >>>>
>>> >>>> Peace,
>>> >>>>
>>> >>>> Matt.
>>> >>>>
>>> >>>> From: <hcp-users-boun...@humanconnectome.org> on behalf of Thomas
>>> Nichols <t.e.nich...@warwick.ac.uk>
>>> >>>> Date: Tuesday, June 7, 2016 at 3:19 PM
>>> >>>> To: HCP Users <hcp-users@humanconnectome.org>
>>> >>>> Subject: [HCP-Users] Location of CIFTI vertices in standard space
>>> >>>>
>>> >>>> Hi folks,
>>> >>>>
>>> >>>> I'm trying to extract the exact location of vertices in the
>>> standard brainordinates space, and I've been told the process works like:
>>> >>>> • Use wb_command's cifti-export-dense-mapping to get the indices.
>>> >>>> • Look up the XYZ location of the indices in a surf.gii file.
>>> >>>> Is this right?
>>> >>>>
>>> >>>> If so, I've got two problems.
>>> >>>>
>>> >>>>
>>> >>>> First wb_command's cifti-export-dense-mapping won't work for me.
>>> E.g. From
>>> >>>>> wb_command -file-information
>>> 100307_tfMRI_EMOTION_level2_hp200_s2_MSMAll.dscalar.nii
>>> >>>> I can see that there is a CortexLeft CortexRight structure, as well
>>> as subcortical voxel structures. But when I do
>>> >>>>> wb_command -cifti-export-dense-mapping
>>> 100307_tfMRI_EMOTION_level2_hp200_s2_MSMAll.dscalar.nii COLUMN -surface
>>> CortexLeft out.txt
>>> >>>> I get
>>> >>>>> ERROR: invalid structure name: 'CortexLeft'
>>> >>>> And likewise no love for CortexRight and Cortex and '*', but I
>>> *can* export the voxel coordinates successfully with -volume-all.
>>> >>>>
>>> >>>>
>>> >>>> Second, how do I know which surf.gii I should be using. I was
>>> directed to this standard_mesh_atlases GitHub repo, but I'm at a loss for
>>> which surf.gii I should use (one of
>>> >>>>> {L,R}.sphere.{32,59}k_fs_LR.surf.gii or
>>> >>>>> fsaverage.{L,R}_LR.spherical_std.164k_fs_LR.surf.gii
>>> >>>> ?). I was hoping that the wb_command -file-information output
>>> would tell me, but I can't see it.
>>> >>>>
>>> >>>> I looked in the FAQ and searched the list but didn't find enough
>>> breadcrumbs.
>>> >>>>
>>> >>>> Thanks in advance!
>>> >>>>
>>> >>>> -Tom
>>> >>>>
>>> >>>> __________________________________________________________
>>> >>>> Thomas Nichols, PhD
>>> >>>> Professor, Head of Neuroimaging Statistics
>>> >>>> Department of Statistics & Warwick Manufacturing Group
>>> >>>> University of Warwick, Coventry CV4 7AL, United Kingdom
>>> >>>>
>>> >>>> Web: http://warwick.ac.uk/tenichols
>>> >>>> Email: t.e.nich...@warwick.ac.uk
>>> >>>> Tel, Stats: +44 24761 51086, WMG: +44 24761 50752
>>> >>>> Fx, +44 24 7652 4532
>>> >>>>
>>> >>>> _______________________________________________
>>> >>>> HCP-Users mailing list
>>> >>>> HCP-Users@humanconnectome.org
>>> >>>> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>>> >>>>
>>> >>>>
>>> >>>> The materials in this message are private and may contain Protected
>>> Healthcare Information or other information of a sensitive nature. If you
>>> are not the intended recipient, be advised that any unauthorized use,
>>> disclosure, copying or the taking of any action in reliance on the contents
>>> of this information is strictly prohibited. If you have received this email
>>> in error, please immediately notify the sender via telephone or return mail.
>>> >>>
>>> >>>
>>> >>> The materials in this message are private and may contain Protected
>>> Healthcare Information or other information of a sensitive nature. If you
>>> are not the intended recipient, be advised that any unauthorized use,
>>> disclosure, copying or the taking of any action in reliance on the contents
>>> of this information is strictly prohibited. If you have received this email
>>> in error, please immediately notify the sender via telephone or return mail.
>>> >>
>>> >>
>>> >>
>>> >> --
>>> >> __________________________________________________________
>>> >> Thomas Nichols, PhD
>>> >> Professor, Head of Neuroimaging Statistics
>>> >> Department of Statistics & Warwick Manufacturing Group
>>> >> University of Warwick, Coventry CV4 7AL, United Kingdom
>>> >>
>>> >> Web: http://warwick.ac.uk/tenichols
>>> >> Email: t.e.nich...@warwick.ac.uk
>>> >> Tel, Stats: +44 24761 51086, WMG: +44 24761 50752
>>> >> Fx, +44 24 7652 4532
>>> >>
>>> >>
>>> >>
>>> >> The materials in this message are private and may contain Protected
>>> Healthcare Information or other information of a sensitive nature. If you
>>> are not the intended recipient, be advised that any unauthorized use,
>>> disclosure, copying or the taking of any action in reliance on the contents
>>> of this information is strictly prohibited. If you have received this email
>>> in error, please immediately notify the sender via telephone or return mail.
>>> >
>>> >
>>> >
>>> > --
>>> > __________________________________________________________
>>> > Thomas Nichols, PhD
>>> > Professor, Head of Neuroimaging Statistics
>>> > Department of Statistics & Warwick Manufacturing Group
>>> > University of Warwick, Coventry CV4 7AL, United Kingdom
>>> >
>>> > Web: http://warwick.ac.uk/tenichols
>>> > Email: t.e.nich...@warwick.ac.uk
>>> > Tel, Stats: +44 24761 51086, WMG: +44 24761 50752
>>> > Fx, +44 24 7652 4532
>>> >
>>> >
>>> >
>>> > The materials in this message are private and may contain Protected
>>> Healthcare Information or other information of a sensitive nature. If you
>>> are not the intended recipient, be advised that any unauthorized use,
>>> disclosure, copying or the taking of any action in reliance on the contents
>>> of this information is strictly prohibited. If you have received this email
>>> in error, please immediately notify the sender via telephone or return mail.
>>> > _______________________________________________
>>> > HCP-Users mailing list
>>> > HCP-Users@humanconnectome.org
>>> > http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>>> > <GroupAverageSurfaceShrinkage.png>
>>>
>>>
>>> _______________________________________________
>>> HCP-Users mailing list
>>> HCP-Users@humanconnectome.org
>>> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>>>
>>
>> _______________________________________________
>> HCP-Users mailing list
>> HCP-Users@humanconnectome.org
>> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>>
>
>
>
> --
> __________________________________________________________
> Thomas Nichols, PhD
> Professor, Head of Neuroimaging Statistics
> Department of Statistics & Warwick Manufacturing Group
> University of Warwick, Coventry CV4 7AL, United Kingdom
>
> Web: http://warwick.ac.uk/tenichols
> Email: t.e.nich...@warwick.ac.uk
> Tel, Stats: +44 24761 51086, WMG: +44 24761 50752
> Fx, +44 24 7652 4532
>
>
_______________________________________________
HCP-Users mailing list
HCP-Users@humanconnectome.org
http://lists.humanconnectome.org/mailman/listinfo/hcp-users
