NISCONTRAST(1) NIS 3.4 September 8, 2000 NAME niscontrast - post-hoc or focused contrast tests performed on neuroimaging data SYNOPSIS niscontrast [-vh] [-c contrast] [-s stem] [-t thresh] [-o outdir] [-f dfile] [roifile] DESCRIPTION niscontrast performs voxelwise contrast tests on neuroimaging datasets. These contrasts can be performed post-hoc after omnibus statistical tests have already been performed, or directly as a focused contrast, when the contrast is hypothesis-driven. As a post-hoc analysis, niscontrast works in complement with nisanova, and shares many of its formatting conventions (which are in turn based on the UNIX|STAT program anova). Some familiarity with UNIX|STAT anova and nisanova is assumed in this documentation. Please refer to the UNIX|STAT manual for further information on anova, and to the accompanying documentation on nisanova. Input Format. The input format follows closely to that of nisanova. However, an important limitation of niscontrast is that it can only perform a contrast on one factor at a time, thus for multi-factorial designs the input specification must be filtered so that it collapses across the other factors which are not being tested in the contrast. In other words, in multi-factorial designs, input specifications that were used for nisanova must be edited before use with niscontrast. Furthermore, niscontrast assumes that the input specification will be formatted so that the levels of the factor are listed in the order in which they are to be tested. niscontrast follows the convention that in all tests, the levels of the factor are in increasing order. Both t-test values and F-ratios reflect this in the output. For example, in the following specification: block1 low scan001 block1 high scan002 block2 low scan003 block2 high scan004 niscontrast tests whether the level "high" is greater than the level "low" of the factor listed in the second column. The converse test can also be performed, but the levels should remain in the same order (and the contrast specification in the input should change). Likewise, in linear and quadratic contrast tests, the levels must be presented to niscontrast sorted in the order in which they are to be tested. For example, in the following specification: block1 1 scan001 block1 2 scan002 block2 3 scan003 block2 4 scan004 niscontrast will test for a monotonic contrast that the levels 1,2,3,4 are monotonically arranged, and for a quadratic contrast that they form a U-shape (and again the contrast specification determines the type of test): monotonic contrast quadratic contrast 4 1 3 2 2 3 1 4 2 3 1 4 1 4 2 3 increasing decreasing upside down right side up Thus, to repeat, the factor levels must be pre-sorted such that they will be tested in the proper order. Input specifications can be read from the standard input (or as output from a pipe), from the -f flag, or can redirected from a text file. If the test is to be performed as a post-hoc analysis, an image map can be provided as additional input on the command line. This image will likely be a map of the data showing pixels which have been previously tested for statistical significance in an omnibus analysis (and/or clustered as contiguous particles), via nisanova and/or nisroi. For post-hoc analyses, only pixels above threshold in the input image will be tested in the contrast. In this case, computational time is significantly reduced, and there is greater protection against Type I errors. Output Format. The output from niscontrast obviously diverges from that of nisanova in that tests are only performed on a single factor. Thus, output consists of a single image file composed of pixels which show statistical significance in the contrast test performed. If the test is performed as a simple contrast (C or c), the value of each significant voxel is set at a constant level (5), and those not significant are set to background levels. Otherwise, significant voxels are set to the t-value of the test, and nonsignificant voxels are set to zero. The output file contains the prefix Fpost (or if the -s option is used), and a .post suffix. OPTIONS -h show usage. -c contrast option. At present, niscontrast offers three types of contrasts: Monotonic/Linear trend analysis (M), Quadratic trend analysis (Q), and Pairwise/t-test contrasts (T). M,m: This contrast will test whether the N means of the factor monotonically arranged from level 1 to N. "M" tests for increasing order, and "m" tests for decreasing order. Q,q: This contrast will test whether the N means of the factor are arranged in a quadratic (U-shaped) manner. The contrast examines whether of levels 1 to (N/2)-1 are monotonic in one direction, and (N/2)+1 are monotonic in the other direction. "Q" tests for an upside-down U-shape, and "q" tests for a right-side-up U-shape. T,t: This contrast will test whether 2 means of a factor are significantly different (i.e. if there are more than 2 levels of a factor, then only the 2 levels to be tested should be included as input). "T" will test if level 2 is greater than level 1, and "t" will test if level 2 is less than level 1. If performed as a focused test, it is equivalent to a paired t-test, and the voxels displayed will have as their value the resulting *t-value*. C,c: This contrast will test the directional significance of the difference between the 2 means of a factor when their omnibus significance has already been determined with nisanova (but when the factor has only 2 levels). This version performs the equivalent of the T option but is computationally simpler (and faster) when the relevant preconditions have already been satisfied (omnibus significance). "C" will test if level 2 is greater than level 1, and "c" will test if level 2 is less than level 1. -s stem uses stem as output file prefix, rather than Fpost. -t thresh only performs contrast on voxels above thresh. In post-hoc analysis, thresh relates to the input image and is used to set a significance threshold. In focused contrasts, thresh relates to the raw functional data and provides a way of skipping background voxels to speed up analysis. -v verbose output to stdout; pipes cannot be used with this option. -d outdir output directory (optional; will use current working directory if not included). -f dfile design file (optional; will read from standard input if not included). LIMITS AND DIAGNOSTICS File and factor limits are the same as in nisanova and UNIX|STAT anova. niscontrast follows the same error messages as nisanova.