Department of Biostatistics Seminar/Workshop Series

Multiple comparisons, fMRI imaging and one brave (but dead) Atlantic salmon

Jeffrey Blume, PhD

Associate Professor, Department of Biostatistics, Vanderbilt University School of Medicine

Wednesday, January 9, 2013, 1:30-2:30pm, MRBIII Room 1220

The attribution of “activation” to brain regions in fMRI images involves the analysis of data from many thousands of voxels (3D pixels), raising concerns about multiple comparisons. In 2009, Bennett et. al., used a standard fMRI brain mapping approach on a (dead) Atlantic salmon. The subsequent analysis of images revealed an increase in brain activity, in just a few voxels, associated with the presentation of human faces. Bennett et. al. argued that these results demonstrated the need for rigorous multiple comparison adjustments in fMRI studies.

But Bennett’s argument is problematic because findings of a similar magnitude in, say, a (dead) Minnow, would not have been vanquished by the adjustments (there are fewer voxels in the Minnow’s image). The problem is not that there are a large number of comparisons, but rather that the strength of statistical evidence is not well described by p-values.

An alternative approach that better characterizes “activated” voxels uses likelihood ratios to represent the strength of statistical evidence, obviating the need for multiple comparison adjustments. This approach requires forethought about the magnitude of brain activation, but this is preferable to using p-values which conflate the magnitude of brain activation with the magnitude of evidence against the null hypothesis (which is why dead fish sometimes appear deep in thought).