Seminars/Colloquia/Invited Talks

Seminars

Thomas Nichols

Thresholding Methods for Small Group fMRI Studies

PLACE:	Clark 314
EVENT:	CIS Seminar Series
DATE:	February 08, 2005
TIME:	1:00 - 2:00

Abstract

Central to any fMRI experiment is the identification of significant voxels while controlling the chance of false positives. The standard modeling approach is a massively univariate one, where univariate models are fit independently at every voxel. Test statistics at each voxel then comprise a statistic image, and detection of a signal reduces to finding a threshold on the image. A valid familywise error threshold must control the chance of one or more false positives across the image. If the statistic image approximates a smooth random field under the null hypothesis, a rich collection of theory for finding thresholds is applicable. Keith Worsley and colleagues (1992) brought the Gaussian random field work of Adler and Hasofer to the functional neuroimaging community, and Worsley and colleagues has subsequently extended Adler's work by finding similar results for t, F, chi-squared and Hotelling's T^2 statistic images. In our work we examine the performance of random t field methods under small sample sizes. Using Monte Carlo simulations and permutation tests we assess the validity and power of the random field methods. While usually valid, we find that the random t field results can be quite conservative, especially for low smoothness or low degrees-of-freedom. A related set of random field methods, based on suprathreshold cluster size, demonstrate unstable performance (sometimes conservative, sometimes anteconservative) as smoothness drops. We explore the causes the random field poor performance and offer recommendations for when a nonparametric permutation test is to be preferred to the parametric random field methods.

Brief biography

Thomas Nichols has been an Assistant Professor of Biostatistics at the University of Michigan since 2000. He received his Ph.D. in statistics from Carnegie Mellon University where he also trained in cognitive neuroscience at the Center for the Neural Basis of Cognition. He has been active in the field of functional neuroimaging since 1992, when he joined the University of Pittsburgh's Positron Emission Tomography (PET) Center as a programmer and statistician. Dr. Nichols' research focuses on modeling and inference of functional neuroimaging data, including PET and Functional Magnetic Resonance Imaging (fMRI). He has developed methods and software for: Nonparametric analysis of PET fMRI data, inference methods which account for the multiplicity of searching the brain for changes in activity; exploration and diagnosis of massively univariate fMRI data and models; and high temporal resolution reconstruction methods for PET.