NISANOVA(1)                      NIS 3.4                September 8, 2000

NAME
     nisanova - voxelwise multi-factor analysis of variance of neuroimaging data

SYNOPSIS
     nisanova [-vh] [-s stem] [-t thresh] [-f dfile] [-d outdir] [factor names]

DESCRIPTION
     nisanova does voxelwise multi-factor analysis of variance
     on neuroimaging datasets.  nisanova is based upon
     the UNIX|STAT program anova, and many of its core procedures
     come directly from its source code.  Thus, nisanova
     follows the UNIX|STAT conventions both in terms of allowable designs
     and input formats, except where described below.  Some familiarity
     with UNIX|STAT anova is assumed in this documentation.  Please refer
     to the UNIX|STAT manual for further information on anova.


     Input Format. The input format follows as closely as possible to that of
     UNIX|STAT anova. However, an important change was made to this format
     to accommodate the need to perform the ANOVA for each voxel in an image.
     Thus, the input to nisanova consists of the filename of each scan on
     every line, rather than the individual datum.  As in UNIX|STAT anova,
     the file name is listed in the last column of the line, with index labels
     for each factor preceding it.

     For example:
                               block1 naming 2 scan001

     indicates that "scan001" is the file containing the image data for
     level "2" of the factor indexed by column three and at level "naming" 
     of the factor indexed by column two, in "block1" which is a level of 
     the random factor indexed in column 1.  A description of an experiment 
     will typically consist of a series of lines like the one above.  As in 
     UNIX|STAT, the order of the lines does not matter and replications are 
     allowed (and are averaged before analysis). 


     Also like UNIX|STAT, names of independent and dependent variables can be 
     supplied to nisanova on the command line, in order to provide mnemonic 
     labels for the output.  For example, in a three factor design, the call 
     to nisanova:
                   nisanova  block  task load activity

     would give the name "block" to the random factor, "task" and
     "load" to the next two, and "activity" to the dependent variable.
     If names are not specified, the default names for the independent
     variables are A, B, C, D, etc.

     Data can be read from the standard input (or as output from a pipe),
     or can redirected from a text file.

     Output Format. The output from nisanova obviously diverges from that 
     of UNIX|STAT in that a summary table cannot printed for every voxel in 
     an image.  Instead, output consists of image files composed of the 
     F-ratio associated with each voxel in the image. Separate files are 
     produced for each main effect and interaction term.  Each output file 
     contains the prefix Fmap (or <stem> if the -s option is used), and a 
     suffix which denotes the main effect or interaction being tested. Thus, 
     for the above call, three output files would be generated:

                      Fmap.task
                      Fmap.load
                      Fmap.task_x_load

OPTIONS

     -h   show usage.

     -v   verbose output to stdout; pipes cannot be used with this option

     -s stem
          uses stem as output file prefix, rather than Fmap.

     -t thresh
          only performs ANOVA on voxels whose mean signal intensity is above 
          thresh.  Provides a way of skipping background voxels to speed up 
          analysis.

     -f dfile
          design file (optional; will read from standard input if not
          included).

     -d outdir
          output directory (optional; will use current working directory
          if not included).



LIMITS AND DIAGNOSTICS
     At most 40000 image files are allowed.  Other limits are the same as in
     UNIX|STAT anova.     
     

     nisanova follows the same error messages as UNIX|STAT anova.  Warnings 
     are given if sums of squares or mean squares are invalid or equal to 
     zero.  This occasionally occurs for background voxels whose signal 
     intensity is near zero, and show little variance.