3dPolyfit


Usage: 3dPolyfit [options] dataset

* Fits a polynomial in space to the input dataset and outputs that fitted dataset.

* You can also add your own basis datasets to the fitting mix, using the
  '-base' option.

* You can get the fit coefficients using the '-1Dcoef' option.

Options:

-nord n    = Maximum polynomial order (0..9) [default order=3]
              [n=0 is the constant 1]
              [n=-1 means only use volumes from '-base']

-blur f    = Gaussian blur input dataset (inside mask) with FWHM='f' (mm)

-mrad r    = Radius (voxels) of preliminary median filter of input
              [default is no blurring of either type; you can]
              [do both types (Gaussian and median), but why??]
              [N.B.: median blur is slower than Gaussian]

-prefix pp = Use 'pp' for prefix of output dataset (the fit).
              [default prefix is 'Polyfit'; use NULL to skip this output]

-resid  rr = Use 'rr' for the prefix of the residual dataset.
              [default is not to output residuals]

-1Dcoef cc = Save coefficients of fit into text file cc.1D.
              [default is not to save these coefficients]

-automask  = Create a mask (a la 3dAutomask)
-mask mset = Create a mask from nonzero voxels in 'mset'.
              [default is not to use a mask, which is probably a bad idea]

-mone      = Scale the mean value of the fit (inside the mask) to 1.
              [probably this option is not useful for anything]

-mclip     = Clip fit values outside the rectilinear box containing the
             mask to the edge of that box, to avoid weird artifacts.

-meth mm   = Set 'mm' to 2 for least squares fit;
             set it to 1 for L1 fit [default method=2]
              [Note that L1 fitting is slower than L2 fitting!]

-base bb   = In addition to the polynomial fit, also use
             the volumes in dataset 'bb' as extra basis functions.
              [If you use a base dataset, then you can set nord]
              [to -1, to skip using any spatial polynomial fit.]

-verb      = Print fun and useful progress reports :-)

Notes:

* Output dataset is always stored in float format.

* If the input dataset has more than 1 sub-brick, only sub-brick #0
  is processed. To fit more than one volume, you'll have to use a script
  to loop over the input sub-bricks, and then glue (3dTcat) the results
  together to get a final result. A simple example:
     #!/bin/tcsh
     set base = model.nii
     set dset = errts.nii
     set nval = `3dnvals $dset`
     @ vtop = $nval - 1
     foreach vv ( `count 0 $vtop` )
       3dPolyfit -base "$base" -nord 0 -mask "$base" -1Dcoef QQ.$vv -prefix QQ.$vv.nii $dset"[$vv]"
     end
     3dTcat -prefix QQall.nii QQ.0*.nii
     1dcat  QQ.0*.1D > QQall.1D

m QQ.0*
     exit 0

* If the '-base' dataset has multiple sub-bricks, all of them are used.

* You can use the '-base' option more than once, if desired or needed.

* The original motivation for this program was to fit a spatial model
  to a field map MRI, but that didn't turn out to be useful. Nevertheless,
  I make this program available to someone who might find it beguiling.

* If you really want, I could allow you to put sign constraints on the
  fit coefficients (e.g., say that the coefficient for a given base volume
  should be non-negative). But you'll have to beg for this.

-- Emitted by RWCox

++ Compile date = Jul 10 2020 {AFNI_20.2.01:linux_ubuntu_16_64}