Comparison Between FWE and FDR Corrections for Threshold Free Cluster Enhancement Maps

Abstract

Threshold Free Cluster Enhancement (TFCE) (Smith 2009) is a successful attempt to avoid the problem of selecting a cluster forming threshold in topological inference procedures. In this transformation, excursions sets are integrated over all the possible thresholds resulting in a new set of values reflecting both cluster height and support. In the original paper, Smith et al. proposed permutation based Family Wise Error (FWE) correction as a thresholding method. This, however, controls for the probability of making at least one false positive error, an assumption which is known to be conservative. A different approach is to control for the proportion of all the false positive errors among all the discoveries - False Discovery Rate (FDR). In this paper we investigate the power and false positive error rates achieved by both methods. Additionally we focus on a randomisation method suitable for single subject application, namely time points instead of group labels shuffling.