[Insight-users] Nifti vs Dicom orientation

Mon Apr 6 16:34:14 EDT 2009

Mathieu,

I believe that the Coursolle definition is the opposite of the ITK
definition. The Coursolle definition is the same as the
OrientImageFilter direction.

In other words, ITK says that
X increasing from Right to Left
Y increasing from Anterior to Posterior
Z increasing from Inferior to Superior
is LPS.

Dicom/ITK name the axes based on the positive name of the axis.
OrientImageFilter (and Meta and nifti and Analyze) name the axes based
on the negative name of the axis.

So, if you tell OrientImageFilter that you want RAI, you will get an
ITK direction of LPS.

Confused??? See my recent e-mail.

Bill

On Mon, Apr 6, 2009 at 3:05 PM, Mathieu Coursolle
<mcoursolle at rogue-research.com> wrote:
> Hi,
> Just to clarify...what I meant by RAI is:
> X increasing from Right to Left
> Y increasing from Anterior to Posterior
> Z increasing from Inferior to Superior
> So is ITK the opposite (LPS) ? Or do we mean the same ?
> Thanks,
> Mathieu
> On 6-Apr-09, at 2:47 PM, Bill Lorensen wrote:
>
> I believe, that the OrientImageFilter defines its orientations the
> reverse of dicom/itk. In other words, to achieve RAI in ITK's world,
> you should specify LPS to OrientImageFilter. But, LPS is ITK's
> coordinate system. So you should not need to use OrientImageFilter.
>
> Confused... I am,. See my recent e-mail regarding this.
>
> Bill
>
> On Mon, Apr 6, 2009 at 2:37 PM, Mathieu Coursolle
> <mcoursolle at rogue-research.com> wrote:
>
> Hi,
>
> Thanks again.
>
> Maybe I did not understand the way ITK stores its direction and origin
>
> then...
>
> My NIfTI file doesn't have a qform, but it has a sform that defines the
>
> transform to MNI-152 space.
>
> Limited to the data size, this space is bound to :
>
> Left is -90 mm
>
> Right is +90 mm
>
> Posterior is -126 mm
>
> Anterior is 90 mm
>
> Inferior is -72 mm
>
> Superior is +108 mm
>
> What I am looking for is a 4 x 4 matrix I could use to multiply an index in
>
> RAI
>
> to get the location in MNI-152 space.
>
> I though that the origin was the location of the first voxel in space (in
>
> this case MNI-152).
>
> In this case, 126 is out of my space.
>
> I would expect an origin of (90, 90, -72), since this is the most Right,
>
> Anterior, Inferior point of
>
> the data.
>
> I typically read the file using itkNIfTIImageIO, then convert it to RAI
>
> using itkOrientImageFilter.
>
> I expect an origin of (90, 90, 72) mm at the ouput of itkOrientImageFilter,
>
> but I instead get (-90, -90, -72).
>
> What could I do to get the expected result ? I guess I am missing a
>
> transformation somehow...
>
> Thanks,
>
> MAthieu
>
>
>
>
>
>
> On 6-Apr-09, at 1:33 PM, Luke Bloy wrote:
>
> Hi Mathieu,
>
> So nothing happens to the data buffer, meaning that its ordering isn't
>
> rearranged. the only thing that is changed is the way we relate the i,j,k
>
> coordinates to 'world coordinates'.
>
> the 'origin' in both nifti and itk refers to the world coordinates of the
>
> first voxel in the data buffer (0,0,0).
>
> so with a qform =
>
> [1, 0, 0, -90]
>
> [0, 1, 0, -126]
>
> [0, 0, 1, -72]
>
> [0, 0, 0, 1]
>
> you get an origin of (-90 mm, -126 mm, -72 mm), since nifti qforms use RAS
>
> as the world frame this means that the (0,0,0) ijk voxel is located 90mm to
>
> the left, 126 posterior and 72 mm inferior, to whatever is at 0mm,0mm,0mm
>
> (in MRI this usually the scanner isocenter).  So in itk's LPS world frame
>
> the origin should be set to (90,126,-72).
>
> This is unrelated to the itkOrientImageFilter, which does do actually
>
> flipping and transposing of the data buffer.
>
> -Luke
>
>
> Mathieu Coursolle wrote:
>
> Hi,
>
> Thanks for your clarifications.
>
> However, I must admit that I am still a bit confused.
>
> Since NIfTI and ITK have different Left/Right and Anterior/Posterior
>
> directions,
>
> I admit that the direction matrix needs to be modified.
>
> Since my image dimension is 181 x 217 x 181, a flip in X means that to
>
> change
>
> the direction, the new origin is -90 + (181 - 1) = 90.
>
> A flip in Y however means that the new origin should be -126 + (217 - 1) =
>
> 90.
>
> Should I then expect instead a GetDirection() and GetOrigin() result  of:
>
> [-1, 0, 0, 90]
>
> [0, -1, 0, 90]
>
> [0, 0, 1, -72]
>
> [0, 0, 0, 1]
>
> Just inverting the sign of the origin would then mean that the world
>
> coordinates
>
> are centered on (0, 0, 0)mm, which is not the always the case.
>
> Am I right ?
>
> Thanks!
>
> Mathieu
>
> On 6-Apr-09, at 12:13 PM, Luke Bloy wrote:
>
> Hi all,
>
> So this is my understanding of how orientations and ITK work.
>
> the GetDirection and GetOrigin methods in itk::Image return the directional
>
> cosines in an LPS world frame. Thus an identity matrix returned by
>
> GetDirection() would mean the image is stored in LPS.
>
> The QForm and SForm matrices in the nifti file format  are in the RAS world
>
> coordinate frame. See
>
> http://nifti.nimh.nih.gov/nifti-1/documentation/nifti1fields/nifti1fields_pages/qsform.html
>
> So Mathieu, what you are reporting is the expected behavior.
>
> qform (shown by fslhd etc) of
>
> [1, 0, 0, -90]
>
> [0, 1, 0, -126]
>
> [0, 0, 1, -72]
>
> [0, 0, 0, 1]
>
> should yield a GetDirection() and GetOrigin of
>
> -1  0  0   90
>
> 0  -1  0   126
>
> 0   0  1  -72
>
> 0   0  0   1
>
>
> Antoine,
>
> Your situation is more confusing, I'm not familiar with dcm2nii, but it
>
> isn't unheard of to have dicom converters change the orientation of the
>
> output image, to bring it into some consistent frame (LAS is a favorite of
>
> labs that have lots of old code expecting analyze files). From what you
>
> posted it seems that dcm2nii converted your image from an LPS arranged dicom
>
> to an LAS arranged Nifti file. I would investigate the nifti file's qform
>
> and sform matrics using either nifti_tool, fslhd or pynifti to confirm this.
>
> Hope this helps,
>
> Luke
>
> Mathieu Coursolle wrote:
>
> Hi,
>
> I am having a similar issue, which may be related.
>
> I have a NIfTI file, which I load using itkNIfTIImageIO.
>
> By stepping into ITK, I can see that the s-form (sto_xyz) is:
>
> [1, 0, 0, -90]
>
> [0, 1, 0, -126]
>
> [0, 0, 1, -72]
>
> [0, 0, 0, 1]
>
> However, when it read the origin and orientation of the resulting itkImage,
>
> I get:
>
> [-1, 0, 0, 90]
>
> [0, -1, 0, 126]
>
> [0, 0, 1, -72]
>
> [0, 0, 0, 1]
>
> This data is in the MNI-152 space. I know that the index (0, 0, 0) should be
>
> (-90, -126, -72) mm in LPI  coordinate system (NIfTI's default I believe).
>
> Since I want my data in the RAI coordinate system, I then use the
>
> itkOrientImageFilter.
>
> Assuming it converts the image from LPI to RAI, I would expect the ouput to
>
> be:
>
> [-1, 0, 0, 90]
>
> [0, -1, 0, 90]
>
> [0, 0, 1, -72]
>
> [0, 0, 0, 1]
>
> But I get:
>
> [1, 0, 0, -90]
>
> [0, 1, 0, -90]
>
> [0, 0, 1, -72]
>
> [0, 0, 0, 1]
>
> By the way, my image dimensions are 181 x 217 x 181.
>
> I know that index (0, 0, 0) is (90,  90, -72) mm in RAI.
>
> I am right that the matrix I get from ITK is not what is expected ?
>
> It looks to me that the itkNIfTIImageIO is doing some transform to
>
> the orientation matrix.
>
> What would be that transformation's purpose ?
>
> Thanks a lot!
>
> Mathieu
>
> ____________________ Mathieu Coursolle, M.Ing.
>
> Rogue Research Inc.        www.rogue-research.com
>
> <http://www.rogue-research.com/>
>
>
> On 6-Apr-09, at 10:46 AM, Antoine DUCHAMPS wrote:
>
> Hi all,
>
> I have a DWI sequence (brain) acquired with a Siemens scanner in Dicom
>
> format. After converting it to NIfTI with dcm2nii
>
> (http://www.sph.sc.edu/comd/rorden/mricron/dcm2nii.html) I tried to read
>
> it an recover the orientation by using ITK (I've copied the code below).
>
> The matrix I obtain is the following:
>
> 0.9997  0.0000  0.0252
>
> 0.0002 -1.0000 -0.0072
>
> -0.0252 -0.0072  0.9997
>
> However, the transforation matrix in the Dicom header is
>
> 0.9997  0.0000  0.0252
>
> 0.0002  1.0000 -0.0072
>
> -0.0252  0.0072  0.9997
>
> If the Dicom image orientation is LPS, this means that the image
>
> orientation in NIfTI is LAS (And not RAS as I believed). Could anybody
>
> clarify this please? Are there several possible orientations in NIfTI?
>
> If so, how can I know the specific image orientation?
>
> Antoine.
>
> #include "itkImageFileReader.h"
>
> #include "itkOrientedImage.h"
>
> const unsigned int Dimension = 4;
>
> typedef short PixelType;
>
> typedef itk::OrientedImage<PixelType, Dimension> ImageType;
>
> typedef ImageType::Pointer ImagePointerType;
>
> typedef itk::ImageFileReader< ImageType > ReaderType;
>
>
> int main(int argc, char* argv[])
>
> {
>
> ImageType::Pointer image = ImageType::New();
>
> ReaderType::Pointer reader = ReaderType::New();
>
> reader->SetFileName(argv[1]);
>
> reader->Update();
>
> image = reader -> GetOutput();
>
> std::cout << image->GetDirection() << std::endl; std::cout.flush();
>
> }
>
> _____________________________________
>
> Powered by www.kitware.com <http://www.kitware.com>
>
> Visit other Kitware open-source projects at
>
> http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ITK FAQ at:
>
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
>
> http://www.itk.org/mailman/listinfo/insight-users
>
>
>
>
>
>
>
>
>
> ------------------------------------------------------------------------
>
> _____________________________________
>
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
>
> http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ITK FAQ at:
>
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
>
> http://www.itk.org/mailman/listinfo/insight-users
>
>
>
> ____________________ Mathieu Coursolle, M.Ing.
>
> Rogue Research Inc.        www.rogue-research.com
>
> <http://www.rogue-research.com/>
>
>
>
>
>
>
>
>
>
> ____________________
>
> Mathieu Coursolle, M.Ing.
>
> Rogue Research Inc.
>
> www.rogue-research.com
>
>
>
>
>
>
>
>
> _____________________________________
>
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
>
> http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ITK FAQ at:
>
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
>
> http://www.itk.org/mailman/listinfo/insight-users
>
>
>
>
> ____________________
> Mathieu Coursolle, M.Ing.
> Rogue Research Inc.
> www.rogue-research.com
>
>
>
>
>
>
>
>