Brain Imaging & Analysis Center

usage: -regress cond flag nbins arg out

Regression mapping calculates an index of how much of the signal variability in each voxel time course can be attributed to a particular reference condition. A reference condition is any time series that has a value for every MR image acquisition timepoint. Typical reference conditions are cardiac or respiratory oscillations, head motion, behavioral performance parameters, or it can be any other signal recorded during the scan. Reference condition time series can be generated automatically from the MR images or read from paradigm file (see PDIGMFILE in “fscan help”), or they can be entered using the -reference command line options. These are referenced by name (see COND and REF_CONDS.)

Regression mapping (-simulate) and regression filtering (-set) use the same algorithms to display and filter the data, respectively. If a linear regression is selected (see below), it plots the correlation coefficient between the voxel time course and the reference time course. For a non-linear regression, the voxel data are binned according to the reference time course, first into 100 bins, which are then combined into ARG nodes evenly weighted across the reference parameter range. A smooth curve is then fitted through the nodes using a cubic spline interpolation. This curve is voxel signal change as a function of the reference condition. Each value in the reference time course is then replaced by its associated value in the splined curve. The deviation of this new curve from a straight line is plotted as specified by the FLAG argument.

• COND - one of the reference condition variables (see REF_CONDS).
• FLAG - specifies how to generate the regression index (add to combine):
  • 1 - Linear regression – plot the correlation coefficient (x 1000). Default is non-linear regression – plot the standard deviation of splined line, normalized by dividing by the standard deviation of the voxel.
  • 2 - For non-linear regression, plot the average magnitude of the splined line instead of the normalized standard deviation. The magnitude is calculated as the square root of the sum of the squared difference from the mean.
  • 4 - For non-linear regression, don't normalize map values by dividing by overall voxel standard deviation.
• NBINS - specifies how many bins to use for sorting the regressed data; the more bins the better the ability to resolve signal differences, but the greater the likelihood for random noise. NBINS defaults to 8 (max 100).
• ARG - used to affect the filtering. For a linear regression (FLG=4) ARG is a scaling factor to increase the reference values (default=1). For a non-linear regression ARG is the number of nodes to use for splining the interpolated smooth curve.
• OUT - specifies that the regression maps are to be saved in a file. OUT can be a filename or if using an analysis file it an be a record number. The word size for the output is short integers (16 bits). All regression maps will be stored in the same file, with all the maps for each slice grouped together; within each group the images are in the same order that the -y options were specified on the command line. The OUT option need only be specified once, it will automatically be applied to all the regression maps.

Note: Up to 10 different -set and -simulate options can be combined in a single fScan command.

See Also:
fScan Home, fScan Manual, CmdLineOptions, Manual Help