[FieldTrip] reverse source interpolate?
jan-mathijs schoffelen
jan.schoffelen at donders.ru.nl
Thu Mar 13 13:00:51 CET 2014
Hi Vitória, Erick, Cornelius et al,
Allow me to chime in to this Brazilio-German interaction ;-).
It seems to me that the discussion consists of a few separable elements.
-The first one pertains to the specific question how to map source locations between two different representations (source with pos versus volumetric MRI-type image with a transform).
-The other one relates to some atlas related business, i.e. mapping locations that have a specific anatomical label to specific MEG source space locations.
-And then there's a third one, which is important for 2, and pertains to the usual complicated stuff of mapping between different coordinate systems.
There are several ways to achieve what Vitória is looking for, and I think that Erick already solved it, but recent changes in FieldTrip make it quite straightforward.
So here goes my recipe:
1 I would go along with Erick to do my beamformer source reconstruction on a subject-specific grid, which is inverse-warped from a regular grid defined on the MNI-template.
2 If we take this template grid, which can be found in ~/fieldtrip/template/sourcemodel, you can interpolate the atlas of your choice onto this sourcemodel, using ft_sourceinterpolate. Reading in of the atlas nowadays indeed should be done with ft_read_atlas (ft_prepare_atlas will be deprecated soon), and my personal taste would be NOT to use the afni atlas, because this one is defined in Talairach space, which is slightly different from MNI-space, and although FT accounts for the differences I am not sure how accurate this is. Anyway, one could do something like atlas = ft_read_atlas('~/fieldtrip/template/atlas/....'), load the template sourcemodel with the resolution you need (i.e. the resolution you used in your beamformer grid), and call ft_sourceinterpolate: cfg = []; cfg.interpmethod = 'nearest'; cfg.parameter = 'tissue'; sourcemodel2 = ft_sourceinterpolate(cfg,atlas,sourcemodel); % NOTE: ensure that the units are consistent in atlas and sourcemodel, if not use ft_convert_units
3 Step 2 should have given you sourcemodel2 with a field 'tissue', that represents the anatomical labels, according to the corresponding atlas. Atlas.tissuelabel gives the labels that correspond to the numbers in the tissue field.
4 Then, without the need to call ft_sourceinterpolate, nor ft_volumenormalise on your source-reconstructed data, you can do indx = find(sourcemodel2.tissue==x), where x is the number of your choice, and you directly get the indices of the source positions in your 'source' variable, that have this particular anatomical label.
5 Then, you can do whatever you like with it, e.g. take a max or mean or whatever...
I hope I was clear enough, and that someone feels inclined to make this into a FAQ on the FieldTrip wiki ;-).
Best wishes,
Jan-Mathijs
On Mar 12, 2014, at 5:50 PM, Vitoria Piai wrote:
> Dear Erick,
>
> "simply label and/or define my ROI in the MNI grid. When I warp to the individuals, everything is in place without a hitch" -> that sounds like a good thing to do.
> Is this a very complicated thing to do? Would you mind sharing a piece of code or some tips for how I could achieve it?
>
> Thanks once more, Vitória
>
> On 3/12/2014 11:37 AM, Erick Ortiz wrote:
>> Dear Vitoria,
>>
>> What I do usually, is to simply label and/or define my ROI in the MNI grid. When I warp to the individuals, everything is in place without a hitch.
>>
>> About the interpolation, I cannot help much more. I have been using my own tools for years, and I was not following the support of atlases in Fieldtrip.I tried it today, but had issues with coordinate system labeling when defining a ROI (I updated to the latest version by SVN):
>>
>> afni = ft_read_atlas('template/atlas/afni/TTatlas+tlrc.HEAD');
>> cfg = [];
>> cfg.method = 'ortho';
>> cfg.interactive = 'yes';
>> cfg.funparameter = 'coh';
>> cfg.atlas = afni;
>> cfg.coordsys = 'mni';
>> % cfg.roi = 'Middle Temporal Gyrus';
>> % cfg.inputcoord = 'tal';
>> ft_sourceplot(cfg, interp);
>>
>> Also note that I had to use ft_read_atlas, because ft_prepare_atlas does not generate the brick0label/brick1label fields.
>>
>> It works well, but if I try to define a ROI (uncommenting the two lines), I get the error "The field cfg.inputcoord is required". There seems to be some confusion between inputcoord and coordsys.
>>
>> It can be because I was doing something wrong: maybe a Fieldtrip developer could comment on that?
>>
>> Best,
>> Erick
>>
>>
>>
>> On Wed, Mar 12, 2014 at 9:24 AM, Vitoria Piai <v.piai.research at gmail.com> wrote:
>> Dear Erick,
>>
>> Thanks for your detailed reply.
>> I've been using individual MNI-warped grids (point 4 below) and I'm always using ft_convert_unit so those points are alright in my data.
>>
>> The reason why I do need to interpolate (at least, I think I need), is because I'm looking for max 'pow' in anatomically defined regions given the afni atlas. I managed to get it to work by interpolating to the mri and looking for max 'pow' given a label (e.g., 'Middle Temporal Gyrus') using ft_volumelookup.
>> Do you have any experience with this approach (searching within anatomical labels) without having to interpolate?
>>
>> Thanks once more, Vitória
>>
>>
>>
>>
>> On 3/11/2014 6:35 PM, Erick Ortiz wrote:
>>> Dear Vitória and Cornelius,
>>>
>>> this is a common problem, and maybe it would be good to clarify the strategies that can be used here. I hope this can help others in a similar situation, which I have found many times.
>>>
>>> 1) You are trying to convert voxel indexes to head (CTF) coordinates. In this case, there are two mistakes in the code:
>>> 1a) pos = warp_apply( pinv( sourceNAIInt.transform ), posInt, 'homogeneous' );
>>> ... should be done with sourceNAIInt, not mri. The transforms are different, in general; e.g. note the downsample=2.
>>> 1b) use cfg.locationcoordinates = 'head', as commented by JM in a previous post. Or remove this line, since 'head' is the default.
>>>
>>> 2) But if you are working with the same subject/grid, there is no need for interpolation at all. Simply find the grid position with highest 'pow' and take its position.
>>> [dummy,ind] = max(source.avg.pow);
>>> pos = source.pos(ind,:);
>>>
>>> 3) However, this is not possible when using different grids, e.g. in a grand average in MNI coordinates. Find the peak ('pos') in the interpolated volume, as in Cornelius' code, then find the index of the nearest grid position for a common (again, MNI) grid. This is also in JM's message.
>>>
>>> dpos = source.pos - repmat( pos, size(source.pos,1), 1 );
>>> [dummy,ind] = min(sum(dpos.^2,2));
>>>
>>> 4) What I have been using, and strongly recommend, is an MNI grid warped to the subject's anatomy, according to this procedure:
>>> http://fieldtrip.fcdonders.nl/example/create_single-subject_grids_in_individual_head_space_that_are_all_aligned_in_mni_space
>>>
>>> This would give you both the benefits of working in MNI coordinates, and the ease of having a common grid across subjects (for instance, enabling solution #2). Still, some projects do call for one of the other options.
>>>
>>> Just a couple of details:
>>> - Pay utmost attention to coordinate systems and units. For instance, it is common to confuse unlabeled cm for mm, and either of these with voxels. Try to also label all structures as CTF or SPM (MNI).
>>> - Using pinv instead of inv is usually a good idea. Here, it is a not an issue, but matrix inversion in MATLAB can lead to a world of hurt to the unwary.
>>>
>>> Best,
>>> Erick
>>>
>>>
>>>
>>> On Tue, Mar 11, 2014 at 4:04 PM, Vitoria Piai <v.piai.research at gmail.com> wrote:
>>> Dear FT-ers,
>>>
>>> I found an old posting to the mailing list (see below) that never got answered so my question still holds: given a location I found in the interpolated source, how do I find back the same location in .pos of the non-interpolated data?
>>> I can try and go around it by finding the corresponding MNI coordinates in the interpolated data (which by now I have to do by hand, so not optimal anyways), and then look for those coordinates in .pos. But I was wondering, like Cornelius was in his post, whether there is a neater way to get this information.
>>>
>>> Thanks a lot for the help,
>>> Vitória
>>>
>>>
>>> >>>>>>>>>>>>>>
>>>
>>> Hi mailing list,
>>>
>>> I'm still struggling with the conversion of coordinates between source and
>>> interpolated source space.
>>> For example, if i have the position of a single grid point how do i get the
>>> corresponding voxel in the interpolated source structure. Or the other way
>>> around, how to get the grid point which corresponds to a voxel (e.g the max
>>> voxel) in the interpolated source structure.
>>>
>>> I think this should be a common problem when working with virtual
>>> electrodes, shouldn't it?
>>>
>>> To make things easier i put together a example with the data from the
>>> source tutorial where i tried to calculated the grid pos of the max voxel
>>> in the interpolated source.
>>>
>>> clear all;
>>> load sourcePost_nocon; % source structure from tutorial
>>> mri = ft_read_mri('Subject01.mri'); % mri of subject01 from tutorial
>>>
>>> sourceNAI = sourcePost_nocon;
>>> sourceNAI.avg.pow = sourcePost_nocon.avg.pow ./ sourcePost_nocon.avg.noise;
>>> sourceNAI=rmfield(sourceNAI,'freq'); % had to remove that to let
>>>
>>> cfg = [];
>>> cfg.downsample = 2;
>>> cfg.parameter = 'avg.pow';
>>> sourceNAIInt = ft_sourceinterpolate(cfg, sourceNAI , mri);
>>>
>>> % Find position of max activity
>>> [dum, maxindx] = max(sourceNAIInt.avg.pow(:));
>>> [xi, yi, zi] = ind2sub(sourceNAIInt.dim, maxindx);
>>> posInt=[xi, yi, zi];
>>>
>>> % Plot interpolated source with position of max activity
>>> cfg = [];
>>> cfg.method = 'ortho';
>>> cfg.funparameter = 'avg.pow';
>>> cfg.locationcoordinates = 'voxel';
>>> cfg.location = posInt; % location of max activity is marked
>>> correctly.
>>> figure;
>>> ft_sourceplot(cfg,sourceNAIInt);
>>>
>>> % Transform coordinate back to uninterpolated source???
>>> dpos = warp_apply(inv(mri.transform), posInt, 'homogeneous');
>>>
>>>
>>> % Plot uninterpolated source with position aquired from interpolated source
>>> cfg = [];
>>> cfg.method = 'ortho';
>>> cfg.funparameter = 'avg.pow';
>>> cfg.locationcoordinates = 'voxel';
>>> cfg.location = dpos; % unfortunately this coordinate is
>>> obviously wrong!!! WHY?
>>> figure;
>>> ft_sourceplot(cfg,sourceNAI);
>>>
>>>
>>> Unfortunately the solution of Jan Mathijs did not work, nor did mine :(
>>>
>>> Any ideas???
>>>
>>> Cornelius
>>>
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> fieldtrip mailing list
>>> fieldtrip at donders.ru.nl
>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>
>>>
>>>
>>> _______________________________________________
>>> fieldtrip mailing list
>>> fieldtrip at donders.ru.nl
>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>
>>
>> _______________________________________________
>> fieldtrip mailing list
>> fieldtrip at donders.ru.nl
>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>
>>
>>
>> _______________________________________________
>> fieldtrip mailing list
>> fieldtrip at donders.ru.nl
>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
Jan-Mathijs Schoffelen, MD PhD
Donders Institute for Brain, Cognition and Behaviour,
Centre for Cognitive Neuroimaging,
Radboud University Nijmegen, The Netherlands
Max Planck Institute for Psycholinguistics,
Nijmegen, The Netherlands
J.Schoffelen at donders.ru.nl
Telephone: +31-24-3614793
http://www.hettaligebrein.nl
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20140313/be7712ba/attachment-0002.html>
More information about the fieldtrip
mailing list