Plugin Help - Cluster Thresholding
- Details
- Category: Tresholding: Multiple Comparisons Problem
- Last Updated: 11 April 2018
- Published: 11 April 2018
- Hits: 4383
Correction of multiple comparisons using cluster-size thresholding
Authors: Fabrizio Esposito, Rainer Goebel
The "ClusterThresh" plugin provides a method for the correction of multiple comparisons using cluster-size thresholding. It complements the existing methods in BrainVoyager QX for the correction of multiple comparions (Bonferroni, FDR) using voxel-based thresholding. This method exploits the fundamental assumption that areas of activity tend to stimulate signal changes over spatially contiguous groups of voxels rather than over sparsely isolated voxels. The implemented solution for cluster-level thresholding is based on the approach described by Forman and colleagues (1995) but has been extended and generalized from 2D to 3D statistical maps (Goebel et al., 2006).
In BrainVoyager QX, all 3D statistical maps are arranged in so-called VMP (Volume MaPs) files. A single VMP is created after running any voxel-level statistical test in the 3D space of VMRs and VTCs, e.g. when computing GLM or ANOVA contrasts and overlaying the results on an anatomical data set (VMRs).
The principal motivation for running the ClusterThresh plugin is the calculation of a specific cluster size threshold, which is then applied to the current VMP (the calculation is restricted to the first VMP of the list in case of multiple overlaid VMPs). By applying the cluster size threshold, the resulting map becomes “corrected for multiple comparisons” at a desired confidence level (e.g. alpha = 0.05). Since the calculation does not change the current single-voxel threshold, you might not notice any change in the overlayed VMP display - the only visible change is the removal of small clusters from the map due to the applied cluster-size threshold. The computation of the minimum cluster threshold is accomplished via MonteCarlo simulation of the random process of image generation, followed by the injection of spatial correlations between neighboring voxels, voxel intensity thresholding, masking (optional) and cluster identification. Starting from a manually adjusted voxel-level probability threshold, a minimum cluster size threshold is automatically set for the VMP which yields 5% (or less) protection against false positive detection at the cluster level.
References
Forman SD, Cohen JD, Fitzgerald M, Eddy WF, Mintun MA, Noll DC. Improved assessment of significant activation in functional magnetic resonance imaging (fMRI): use of a cluster-size threshold. Magn Reson Med. 33(5), 636-647.
Goebel, R., Esposito, F. & Formisano, E. (2006). Analysis of functional image analysis contest (FIAC) data with Brainvoyager QX: From single-subject to cortically aligned group general linear model analysis and self-organizing group independent component analysis. Human Brain Mapping, 27, 392-401.