Detecting changes in nonisotropic images

if the noise component of image data is nonisotropic, i.e., if it has nonco nstant smoothness or effective point spread function, then theoretical resu lts for the P value of local maxima and the size of suprathreshold clusters of a statistical parametric map (SPM) based on random field theory are not valid. This assumption is reasonable for PET or smoothed fMRI data, but no t if these data are projected onto an unfolded, inflated, or Battened 2D co rtical surface. Anatomical data such as structure masks, surface displaceme nts, and deformation vectors ate also highly nonisotropic. The solution off ered here is to suppose that the image can be warped or flattened (in a sta tistical sense) into a space where the data are isotropic. The subsequent c orrected P values do not depend an finding this warping; it is sufficient o nly to know that such a warping exists. (C) 1999 Wiley-Liss, Inc.