Using Workbench Command

   The CIFTI format is a new data file format intended to make it easier to
   work with data from multiple disjoint structures at the same time - often
   this means both hemispheres of cortex as surface data, and other structures
   as voxel data (amygdala, thalamus, putamen, etc).  Additionally, it can
   exclude locations that are uninteresting for the task at hand (medial wall,
   white matter, csf), preventing them from taking up room in the data of the
   file.  The set of structures and the locations in them that are used in a
   cifti file are referred to as 'brainordinates', or for the specific case of
   'all gray matter locations', 'grayordinates'.

   However, to explain the cifti format, it is easiest to work from the
   opposite direction, as it is conceptually simpler.  A single cifti file is a
   single rectangular data matrix (usually 2 dimensions, but supports 3, and
   may support more in the future), where each dimension is labeled with what
   we call a 'mapping', each of which uses one of (currently) five possible
   'mapping types'.  It is the combinations of these mapping types that give
   rise to the diverse types of cifti files.  A single mapping of type 'brain
   models' (also known as 'dense') can represent both hemispheres and all
   subcortical structures simultaneously, meaning that only a single matrix
   dimension is used to represent over a dozen structures, both surface-based
   and voxel-based.  Each mapping contains all information needed to figure out
   what every index along the matrix dimension means.  By putting a dense
   mapping along both dimensions in a 2D cifti file, you get a brainordinates
   by brainordinates matrix, which is used for connectivity measures.  Notably,
   even if two dimensions use the same mapping *type*, they can have different
   information in them, for example a connectivity matrix between two different
   parcellations.

   The other mapping types that currently may be used in a cifti file are:
      Parcels: each index refers to a named subset of the brainordinates (i.e.
         'V1', and the surface vertices in V1)
      Scalars: each index is simply given a name (i.e. 'Myelin')
      Series: each index is assigned a quantity in a linear series (i.e., a
         timeseries of 0 sec, 0.7 sec, 1.4 sec, ...)
      Labels: each index is assigned a name (i.e., 'Visual Areas'), but also a
         list of labels that maps integer data values to names and colors (i.e.
         {(5, 'V1', #ff0000), (7, 'V2', #00ff00), ...})

   The commands that operate on cifti files often require you to specify which
   dimension they should operate on.  Because cifti files can contain 3
   dimensions, we specify them as which dimension to operate along, that is, 
   the ROW dimension refers to the mapping along the length of a row.
   Additionally, the ROW dimension is the *first* dimension in a cifti file,
   unlike 2D matrices in linear algebra.  This means that increasing the value
   of the first index moves rightwards in the matrix, not downwards.

   The common types of cifti files and the mapping types they use are:
      dconn: ROW is dense, COLUMN is dense
      dscalar: ROW is scalars, COLUMN is dense
      dtseries: ROW is series, COLUMN is dense
      dlabel: ROW is labels, COLUMN is dense
      pconn: ROW is parcels, COLUMN is parcels
      pdconn: ROW is dense, COLUMN is parcels
      dpconn: ROW is parcels, COLUMN is dense
      pscalar: ROW is scalars, COLUMN is parcels
      ptseries: ROW is series, COLUMN is parcels

   For the full details of the CIFTI format, see
      http://www.nitrc.org/projects/cifti/