[MINC-users] MINC support in Brainstorm
François Tadel
francois.tadel at mcgill.ca
Tue Aug 6 15:01:47 EDT 2013
Pierre, Andrew, Claude:
Thanks for you help. The support for the CIVET pipeline in Brainstorm is
almost finished:
http://neuroimage.usc.edu/brainstorm/Tutorials/SegCIVET
The MINC format is supported thanks to Pierre Bellec's code
(Matlab-based, hence cross-platform).
For now, the MRI prefix_t1_final.mnc still have to be manually converted
to MINC1 because the support for the MINC2 format is not finished.
Everything is already available for download as part of the Brainstorm
distribution: http://neuroimage.usc.edu/brainstorm3_register/download.php
I will post a final note on the CIVET mailing list when it's ready for
public use.
Cheers,
Francois
On 25/07/2013 10:40 AM, Pierre Bellec wrote:
> Dear François,
>
>
> For reading the nifti files, I'm using my own matlab-only function
> (attached here) that I wrote a few years ago. It's not supporting
> all the extensions of the nifti format, it reads only what we need
> in Brainstorm (volume, basic orientation of the cube, voxel size).
> So indeed, it's not that the voxel<=>world transformation is not
> present in the output of the mnc2nii file, it's just that I'm not
> reading it.
>
> Except for CIVET, all the other programs I know for MRI
> segmentation (FreeSurfer, BrainVISA, BrainSuite) are saving their
> surface files in MRI coordinates, there is no additional transform
> needed to get the correspondence MRI/surface. I guess this is why
> I never had to read those transformations properly.
>
> With these discussions, I realize it probably makes more sense for
> me to put efforts into reading correctly the transformation
> directly from the MINC format rather than going through the nifti
> format. I will try to use Pierre's code to get the information I
> need (voxel <=> world transformation).
>
>
> That would be the cleanest approach indeed.
>
> Apparently, it is also the easiest way for me to get a fully
> scripted function, with no compilation involved (easier to
> maintain and distribute). Matlab supports natively NetCDF and HDF5
> formats, no need to redistribute additional libraries.
>
> Agreed, that's why despite all the work involved and the overlap with
> some other projects (including NIAK) I still felt it was necessary to
> get the MOMINC library done eventually.
>
> Pierre: Can I distribute your code with the the main Brainstorm
> distribution in the "external" folder?
>
>
> Absolutely.
>
> Do you want to attach a copyright notice on top of what is already
> in the code?
>
> Nope, MIT license in the code is fine.
>
> I'll try to finish a beta version that works with both MINC1 and MINC2
> by next week. Best,
>
> Pierre
>
> Thank you for you help!
> Francois
>
>
>
>
> On 23/07/2013 11:38 PM, Pierre Bellec wrote:
>
> Dear François (and Andrew),
>
>
> 2) The mnc2nii solution is not going to be a valid
> long-term solution for
> one specific reason: the import of the CIVET output
> surfaces in Brainstorm.
> By converting to NIfTI, we lose the initial MINC
> coordinate system, on
> which all the CIVET output is based. If I don't have
> access to the origin
> of the referential, I don't think I can register correctly
> the cortical
> envelopes generated by CIVET with the corresponding T1 MRI.
> Therefore we will need a different solution for importing
> the MINC volumes.
>
>
> As Andrew said, the conversion from minc to nifti should
> retain the correct
> voxel-to-world transformation. Hopefully this will be fixed
> once and for
> all by Andrew with niftilib. Great to hear about that very
> useful future
> development.
>
>
> 3) Since HDF5 and netCDF are formats that are fully
> supported in many
> platform-independent languages (Java, Matlab, Python),
> will there be at
> some point a MINC reader available that does not rely on
> any external
> compiled library?
> I looked at Pierre Bellec's Matlab functions
> (http://code.google.com/p/**
> mominc/ <http://code.google.com/p/mominc/>), but for what
> I understood it
>
> does the reading of the HDF5/netCDF contents and does not
> provide any tool
> to re-orient the the volume in another standard coordinate
> system (such as
> NIfTI).
> There is probably not much to add to get a Matlab-based
> (or Java-based)
> version of mnc2nii (that reads the MINC volume into a
> NIfTI-oriented volume
> but with the full MINC header information).
> Are there other solutions readily available to help us
> with this MINC
> import?
> Is anybody working on or interested in those aspects at
> the present time?
>
>
> I have put some efforts in mominc from time to time, and I am
> confident a
> reliable reader can be implemented for minc1 and minc2 in
> matlab, and minc1
> for octave. I haven't prototyped the writer, but it should
> work. A lot of
> what's left to do for the reader is to run some systematic
> test (with some
> fancy slice-based intensity normalization) and finalize the
> fields of the
> header to retain compatibility with the reader/writer in NIAK
> (which is a
> different beast altogether based on system calls to the minc
> tools).
> Regarding the limitation you mentioned on the re-orientation,
> I have added
> a few features from NIAK in minc_read (for MINC1 files only
> for now, MINC2
> will crash). If you download the head revision of the
> code<https://code.google.com/p/mominc/source/checkout> (I'll
>
> also send it to you by email) and run something like:
>
> [hdr,vol] = minc_read('test.mnc');
>
> you'll get a field hdr.info.mat which is a 4 x 4
> transformation matrix from
> voxel coordinates to world coordinates. If you want to get a
> list of world
> coordinates for each voxel (and assuming vol is 3D):
>
> ind = find(true(size(vol)));
> [x,y,z] = ind2sub(size(vol),ind);
> coord_w = minc_voxel2world([x,y,z],hdr.info.mat);
>
> COORD_W is an array (nb_voxel x 3) where each row is the world
> coordinates
> of one voxel in the volume. You can use MINC_WORLD2VOXEL to go
> the other
> way around. If you want to have a volume that has been
> resampled such that
> the first dimension is "left to right", the second is "caudal
> to rostral"
> and the third is "ventral to dorsal", and the voxel to world
> transformation
> has a diagonal matrix for rotation, with diagonal elements
> coding only for
> voxel size, I can implement that as well.
>
> Let me know if you think this could solve your problem and, if
> so, I'll
> help wrap that up.
>
> Best,
>
> Pierre
>
> Those would be very helpful tools for us and probably for the
> developers of
>
> other software packages as well. We would be willing to
> help for any new
> developments in this direction, and more generally support
> the integration
> between the different software environments at the Neuro.
>
> Thanks
> Francois
>
>
> --
> François Tadel, MSc
> MEG / McConnell Brain Imaging Center / MNI / McGill University
> 3801 rue University, Montreal, QC H3A2B4, Canada
>
>
--
François Tadel, MSc
MEG / McConnell Brain Imaging Center / MNI / McGill University
3801 rue University, Montreal, QC H3A2B4, Canada
More information about the MINC-users
mailing list