Registration rotates and shifts functional data into a standard coordinate frame,
typically a Talairach-based stereotactic frame. This is done by aligning the functional
data with a reference data set. The reference data set may be a high-resolution T1-weighted data set
of the individual subject, or it can be an average data set obtained from averaging
across a population of subjects. The functional data consist of a time sequence of
2D slices. For registration, the slices of one timepoint may be selected from this sequence
and aligned with the reference data set. Alternatively, any other 2D slices that are
in geometric alignment with the functional data slices may be used for registration.
For instance, T1-high resolution slices are often obtained along with the functional
scans. If these T1-slices are in geometric alignment (i.e. have the same orientation in space)
then they may be used for registration instead of the functional slices.
The rotational and translational parameters obtained
from this operation are copied into a transformation matrix. In a subsequent step,
this transformation matrix is then used to transform the remaining time steps or
the zmap.
The registration is done in three steps:
1. In the first step, we select a time step from the fMRT timecourse
using 'vtimestep'.
Note: This step is not needed,
if T1-weighted 2D anatomical slices are used that are geometrically aligned with the
functional data.
2. In the second step, we compute a transformation matrix that registers the
2D slices with the 3D data set using 'vreg3d'. This matrix describes a rigid, affine linear
transformation using 3 translational and 3 rotational parameters.
3. In a third step, this transformation
matrix is applied to the zmap, the anatomical slices, or the
raw functional data. This is done by the program 'vdotrans', Optionally, you can also
check the registration accuracy by applying
the transformation to the 2D slices only, and comparing the result to
the reference image in the data base.
Here are these steps in detail:
You must first identify the high resolution reference image in the image data base to
which your present image should be registered. The images in the data base
have identification numbers. In order to obtain the data base id-number (the number) that belongs
to your test subject, check the webpage
https://newbdb.
- With standard search the number and ID can be retrieved,
- with extended search more details about the subjects can be obtained.
Note that some test subjects have more than one reference image in the data base.
In this case, you can choose either one.
The registration matrix describes a rigid, affine linear
transformation between the 2D slices and the 3D reference image.
The 2D slices must be in parallel and must have a fixed inter-slice distance.
The matrix contains 3 translational and 3 rotational parameters.
This matrix is computed using the program 'vreg3d'.
It uses the 2D slices and optionally also 2D EPI-T1 slices to
compute the transformation. Here is an example:
vtimestep -in functional_data.v -out tmp.v -step 0
vreg3d -in tmp.v -out transmatrix.v
-ref reference_image.v -type MI
vdotrans -in tmp.v -out image.v -trans transmatrix.v
-object all
vlv -in image.v reference_image.v
The first call selects a time step from the fMRT time course. In this example,
the first step was selected. The subsequent call to 'vreg3d' uses this data for alignment
with the reference image and places the transformation parameters into
the matrix called 'transmatrix.v'. The parameter '-type' specifies the goal function
to be used for alignment. If functional data are used (as in the above example) then
mutual information (MI) is the best choice. If 2D anatomical slices are used
then linear correlation is sufficient.
The third call applies this matrix to the data and produces a rotated image of the
functional data. The last call visualizes the results so that the registration accuracy
can be checked.
The parameter '-pitch' can be used to specify an initial pitch angle.
This angle is used as a starting point for the matching procedure.
This can sometimes help to get better registration results, especially if
the data set is not aligned with the CA/CP-axis.
For example, if the data set is aligned with the Sylvian fissure,
then use '-pitch -20', as the Sylvian fissure is usually tilted
by an angle of approximately -20 degrees.
If there is an EPI-T1 scan available in addition to the
MDEFT 2D slices, then this scan can be additionally
used for registration in a refinement step.
During refinement, new correlation values are computed, this time between the
EPI-T1 slices and the reference data set. These correlation values are
generally lower than the correlation values computed in the first
step, however, this does not mean that the registration is less accurate.
In general:
The correlation values do not necessarily provide an indication for whether the
registration was successful or not. The only way to be sure is to check
the registration manually by applying the registration matrix to the 2D
anatomical slices.
The EPI-T1 refinement can be enabled by setting '-refine true'.
By default, EPI-T1 refinement is disabled.