Department of Biostatistics Seminar/Workshop Series

Controlling Global Error Rates in fMRI Data Analysis

Hakmook Kang, PhD

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

Wednesday, December 12, 1:30-2:30pm, MRBIII Room 1220

Standard fMRI analyses use a voxel based linear regression model after applying spatial smoothing and common approaches to adjust for the multiple testing problem in fMRI include random field theory (RFT), false discovery rate (FDR), and FDR based on empirical distribution (E-FDR). The primary limitation of these approaches is that they are still conservative in finding positive voxels, even though they are known to be less conservative than the Bonferroni correction. Moreover, the probability of a false positive is fixed at a certain level (e.g., 0.05) and does not change at all regardless of the amount of data. To overcome these two main disadvantages of current approaches, we propose an approach based on likelihood ratio at each voxel, which allows the probabilities of both false positive and negative results to converge to zero as the sample size grows. The characteristics of this approach are illustrated via simulation study and real data analysis.