dcmstack fails to convert certain sets of DICOM files

[chrisgorgo]

I'm trying to convert this dataset - https://dl.dropbox.com/u/412959/dicoms.tar.gz Unfortunately I get an error:

---------------------------------------------------------------------------
InvalidDicomError                         Traceback (most recent call last)
/Users/filo/<ipython-input-10-9c30552a7407> in <module>()
      2 for src_path in glob.glob("*.dcm"):
      3     print src_path
----> 4     src_dcm = dicom.read_file(src_path)
      5     stack.add_dcm(src_dcm)

/Library/Frameworks/EPD64.framework/Versions/7.3/lib/python2.7/site-packages/dicom/filereader.pyc in read_file(fp, defer_size, stop_before_pixels, force)
    623     try:
    624         dataset = read_partial(fp, stop_when, defer_size=defer_size,
--> 625                                             force=force)
    626     finally:
    627         if not caller_owns_file:

/Library/Frameworks/EPD64.framework/Versions/7.3/lib/python2.7/site-packages/dicom/filereader.pyc in read_partial(fileobj, stop_when, defer_size, force)
    529     """
    530     # Read preamble -- raise an exception if missing and force=False

--> 531     preamble = read_preamble(fileobj, force)
    532     file_meta_dataset = Dataset()
    533     # Assume a transfer syntax, correct it as necessary

/Library/Frameworks/EPD64.framework/Versions/7.3/lib/python2.7/site-packages/dicom/filereader.pyc in read_preamble(fp, force)
    502             fp.seek(0)
    503         else:
--> 504             raise InvalidDicomError("File is missing 'DICM' marker. "
    505                                     "Use force=True to force reading")
    506     else:

InvalidDicomError: File is missing 'DICM' marker. Use force=True to force reading

But when I try to force conversion I get the another error:

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
/Users/filo/<ipython-input-11-11b6ef0c3691> in <module>()
      3     print src_path
      4     src_dcm = dicom.read_file(src_path, force=True)
----> 5     stack.add_dcm(src_dcm)

/Library/Frameworks/EPD64.framework/Versions/7.3/lib/python2.7/site-packages/dcmstack-0.6.1-py2.7.egg/dcmstack/dcmstack.pyc in add_dcm(self, dcm, meta)
    525         nii_wrp = None
    526         if not is_dummy:
--> 527             nii_wrp = NiftiWrapper.from_dicom_wrapper(dw, meta)
    528             if self._ref_input is None:
    529                 #We don't have a reference input yet, use this one

/Library/Frameworks/EPD64.framework/Versions/7.3/lib/python2.7/site-packages/dcmstack-0.6.1-py2.7.egg/dcmstack/dcmmeta.pyc in from_dicom_wrapper(klass, dcm_wrp, meta_dict)
   1487 
   1488         #The Nifti patient space flips the x and y directions

-> 1489         affine = np.dot(np.diag([-1., -1., 1., 1.]), dcm_wrp.get_affine())
   1490 
   1491         #Make 2D data 3D

/Library/Frameworks/EPD64.framework/Versions/7.3/lib/python2.7/site-packages/nibabel-1.3.0-py2.7.egg/nibabel/nicom/dicomwrappers.pyc in get_affine(self)
    282         # direction cosine for changes in row index, column 1 is

    283         # direction cosine for changes in column index

--> 284         orient = self.rotation_matrix
    285         # therefore, these voxel sizes are in the right order (row,

    286         # column, slice)

/Library/Frameworks/EPD64.framework/Versions/7.3/lib/python2.7/site-packages/nibabel-1.3.0-py2.7.egg/nibabel/onetime.pyc in __get__(self, obj, type)
     41            return self.getter
     42 
---> 43        val = self.getter(obj)
     44        #print "** setattr_on_read - loading '%s'" % self.name  # dbg

     45        setattr(obj, self.name, val)

/Library/Frameworks/EPD64.framework/Versions/7.3/lib/python2.7/site-packages/nibabel-1.3.0-py2.7.egg/nibabel/nicom/dicomwrappers.pyc in rotation_matrix(self)
    165         assert np.allclose(np.eye(3),
    166                            np.dot(R, R.T),
--> 167                            atol=1e-6)
    168         return R
    169 

AssertionError: 

At the same time dcm2nii is able to convert this dataset with the following result https://dl.dropbox.com/u/412959/20101222_131954BTfMRIcontrolB4s004a1001.nii.gz I hope this is something minor since I love using dcmstack. Thanks in advance for any help!

[moloney]

The rotation matrix generated from the ImageOrientationPatient tag was failing the test for orthogonality in nibabel.nicom.dicomwrappers. Its close enough (~1e-5) that its probably not worth normalizing. The referenced pull request reduces the threshold from 1e-6 to 2e-5.

[chrisgorgo]

Thanks!