Tests of Significance on Brain Imaging Data (A bootstrap approach)

Abstract

In brain imaging studies it is of interest to test for changes in imaging data among subjects in different groups. Testing hypotheses voxel by voxel results in a multiple comparisons problem for which solutions should take into account the spatial correlation structure inherent in the imaging. Statistical Parametric Mapping (SPM) and the permutation test have become popular in this setting but they rely on parametric and exchangeability assumptions, respectively, which are not always satisfied in practice. We propose two bootstrap approaches (L1 and L2) that are free of the parametric assumptions made by SPM and also are more flexible than the permutation test. Not only can our proposed methods be applied to the imaging data, they can also be applied to general functional data. For the L2 method, we present sufficient conditions that ensure asymptotic control of the family-wise error rate.