Skip to content

Latest commit

 

History

History
198 lines (146 loc) · 9.1 KB

heuristics.rst

File metadata and controls

198 lines (146 loc) · 9.1 KB
Raw

Heuristics File

The heuristic file controls how information about the DICOMs is used to convert to a file system layout (e.g., BIDS).

Provided Heuristics

heudiconv provides over 10 pre-created heuristics, which can be seen here .

These heuristic files are documented in their code comments. Some of them, like convertall or ReproIn could be immediately reused and represent two ends of the spectrum in heuristics:

  • convertall is very simple and does not automate anything -- it is for a user to modify filenames in the prepared conversion table, and then rerun with -c dcm2niix.
  • reproin can be used fully automated, if original sequences were named according to its ReproIn convention.

Discover more on their user in the :ref:`Tutorials` section.

However, there is a large variety of data out there, and not all DICOMs will be covered by the existing heuristics. This section will outline what makes up a heuristic file, and some useful functions available when making one.

Components

infotodict(seqinfos)

The only required function for a heuristic, infotodict is used to both define the conversion outputs and specify the criteria for scan to output association. Conversion outputs are defined as keys, a tuple consisting of three elements:

  • a template path used for the basis of outputs
  • tuple of output types. Valid types include nii, nii.gz, and dicom.
  • None - a historical artifact (corresponds to some notion of annotation_class no living human is aware about)

The following details of the sequences could also be used as a {detail} in the conversion keys:

  • item: an index of seqinfo (e.g., 1),
  • subject: a subject label (e.g., qa)
  • seqitem: sequence item, index with a sequence/protocol name (e.g., 3-anat-scout_ses-{date})
  • subindex: an index within the seqinfo (e.g., 1),
  • session: empty (no session) or a session entity (along with ses-, e.g., ses-20191216),
  • bids_subject_session_prefix: shortcut for BIDS file name prefix combining subject and optional session (e.g., sub-qa_ses-20191216),
  • bids_subject_session_dir: shortcut for BIDS file path combining subject and optional session (e.g., sub-qa/ses-20191216).

Note

An example conversion key

('sub-{subject}/func/sub-{subject}_task-test_run-{item}_bold', ('nii.gz', 'dicom'), None)

or equivalent in --bids mode which would work also if there is a specified session

('{bids_subject_session_dir}/func/{bids_subject_session_prefix}_task-test_run-{item}_bold', ('nii.gz', 'dicom'), None)

The seqinfos parameter is a list of namedtuples which serves as a grouped and stacked record of the DICOMs passed in. Each item in seqinfo contains DICOM metadata that can be used to isolate the series, and assign it to a conversion key.

A function create_key is commonly defined by heuristics (internally) to assist in creating the key, and to be used inside infotodict.

A dictionary of {conversion key: series_id} is returned, where series_id is the 3rd (indexes as [2] or accessed as .series_id from seqinfo).

create_key(template, outtype)

A common helper function used to create the conversion key in infotodict. But it is not used directly by HeuDiConv.

filter_files(fl)

A utility function used to filter any input files.

If this function is included, every file found will go through this filter. Any files where this function returns True will be filtered out.

filter_dicom(dcm_data)

A utility function used to filter any DICOMs.

If this function is included, every DICOM found will go through this filter. Any DICOMs where this function returns True will be filtered out.

infotoids(seqinfos, outdir)

Further processing on seqinfos to deduce/customize subject, session, and locator.

A dictionary of {"locator": locator, "session": session, "subject": subject} is returned.

grouping string or grouping(files, dcmfilter, seqinfo)

Whenever --grouping custom (-g custom) is used, this attribute or callable will be used to inform how to group the DICOMs into separate groups. From original PR#359:

grouping = 'AcquisitionDate'

or:

def grouping(files, dcmfilter, seqinfo):
    seqinfos = collections.OrderedDict()
    ...
    return seqinfos  # ordered dict containing seqinfo objects: list of DICOMs

custom_seqinfo(wrapper, series_files)

If present this function will be called on each group of dicoms with a sample nibabel dicom wrapper to extract additional information to be used in infotodict.

Importantly the return value of that function needs to be hashable. For instance the following non-hashable types can be converted to an alternative hashable type: - list > tuple - dict > frozendict - arrays > bytes (tobytes(), or pickle.dumps), str or tuple of tuples.

POPULATE_INTENDED_FOR_OPTS

Dictionary to specify options to populate the 'IntendedFor' field of the fmap jsons.

When a BIDS session has fmaps, they can automatically be assigned to be used for susceptibility distortion correction of other non-fmap images in the session (populating the 'IntendedFor' field in the fmap json file).

For this automated assignment, fmaps are taken as groups (_phase and _phasediff images and the corresponding _magnitude images; consecutive Spin-Echo images collected with opposite phase encoding polarity (pepolar case); etc.).

This is achieved by checking, for every non-fmap image in the session, which fmap groups are suitable candidates to correct for distortions in that image. Then, if there is more than one candidate (e.g., if there was a fmap collected at the beginning of the session and another one at the end), the user can specify which one to use.

The parameters that can be specified and the allowed options are defined in bids.py:
  • 'matching_parameter': The imaging parameter that needs to match between the fmap and an image for the fmap to be considered as a suitable to correct that image. Allowed options are:
    • 'Shims': heudiconv will check the ShimSetting in the .json files and will only assign fmaps to images if the ShimSettings are identical for both.
    • 'ImagingVolume': both fmaps and images will need to have the same the imaging volume (the header affine transformation: position, orientation and voxel size, as well as number of voxels along each dimensions).
    • 'ModalityAcquisitionLabel': it checks for what modality (anat, func, dwi) each fmap is intended by checking the _acq- label in the fmap filename and finding corresponding modalities (e.g. _acq-fmri, _acq-bold and _acq-func will be matched with the func modality)
    • 'CustomAcquisitionLabel': it checks for what modality images each fmap is intended by checking the _acq- custom label (e.g. _acq-XYZ42) in the fmap filename, and matching it with: - the corresponding modality image _acq- label for modalities other than func (e.g. _acq-XYZ42 for dwi images) - the corresponding image _task- label for the func modality (e.g. _task-XYZ42)
    • 'PlainAcquisitionLabel': similar to 'CustomAcquisitionLabel', but does not change behavior for func modality and always bases decision on the _acq- label. Helps in cases when there are multiple tasks and a shared fmap for some of them.
    • 'Force': forces heudiconv to consider any fmaps in the session to be suitable for any image, no matter what the imaging parameters are.
  • 'criterion': Criterion to decide which of the candidate fmaps will be assigned to a given file, if there are more than one. Allowed values are:
    • 'First': The first matching fmap.
    • 'Closest': The closest in time to the beginning of the image acquisition.

Note

Example:

POPULATE_INTENDED_FOR_OPTS = {
        'matching_parameters': ['ImagingVolume', 'Shims'],
        'criterion': 'Closest'
}

If POPULATE_INTENDED_FOR_OPTS is not present in the heuristic file, IntendedFor will not be populated automatically.