Add dicom public and private data repositories

[matthew-brett]

Use nipy data package mechanism for now

Collect releasable data into public package, probably hosted on sourceforge, as in:

http://nipy.sourceforge.net/data-packages/

With each dataset, include successful conversion by SPM or dcm2nii or other. Write comparison utilities of form:

our_converted_img = nibabel.dicom.nifti_from_files(example_data.some_data.dicom_files)
desired_img = nibabel.load(example_data.some_data.converted_fname)
assert_true(np.allclose(as_canonical(desired_img), as_canonical(our_converted_img))

Collect unreleasable data in private package, probably on Dropbox or similar, with restricted access. Collect datasets known to cause problems.

[neurolabusc]

This seems related to issue 977. Each of these repositories provides exemplars of particular types of DICOM (in) image as well as what is deemed as correct NIfTI translations (ref). The included shell script uses dcm2niix to translate the images (out) and tests that json is valid and the NIDS/NIfTI files match the reference images. This could easily be adapted to test other conversion tools. The readme file for each repository provides notes.

• Agfa/dcm4che obfuscation
• Slice Timing Correction
• Enhanced DICOMs (Philips, Siemens XA)
• Computerized Axial Tomography Sequences
• Arterial-spin labeling sequences
• Multi-echo sequences
• Vendor: Canon
• Vendor: GE
• Vendor: GE/Siemens
• Vendor: Philips
• Vendor: Siemens
• Vendor: Toshiba
• Vendor: UIH