Abstract
Despite an obvious demand for a statistical test adapted to classification images, none has been proposed yet. The Stat4Ci Matlab toolbox (http://mapageweb.umontreal.ca/gosselif/Stat4Ci.html) performs all the computations necessary for the application, to classification images, of the Pixel and Cluster tests, both based on Random Field Theory (Adler, 1981; Worsley, 1994, 1995, 1996). These tests are easy to apply, requiring a mere three pieces of information. Furthermore, they are sensitive, producing p-values and thresholds usually much lower than those produced by the standard Bonferroni correction. The Stat4Ci toolbox comprises ReadCid.m that reads Classification Image Data at the individual trial level (in CID format); BuildCi.m that performs least-square multiple linear regression on this data; hrCi2cIm.m that transforms a 24 bits classification image into a color image, for storage; cIm2hrCi.m that does the opposite; SmoothCi.m that convolves a classification image with a Gaussian filter; ZTransCi.m that Z-transforms a smoothed classification image; CiVol that calculates a vector of spherical intrinsic volumes for the search region in the classification image; HalfMax.m that computes the FWHM of the Gaussian filter used to smooth the classification image; stat_threhold.m that applies the Pixel and Cluster tests on a Z-transformed and smoothed classification image; and DisplayCi.m that displays the statistically thresholded classification image and ouputs a summary table. We illustrate the workings of the Stat4Ci toolbox on a set of representative classification images from Gosselin and Schyns (2001), Sekuler, Gaspar, Gold and Bennett (2004), and Adolphs, Gosselin, Buchanon, Tranel, Schyns and Damasio (2004).