Efficient Experimental Designs Under a Nonlinear Model for Event-Related fMRI

• Activity Number: 36
• Type: Contributed
• Date/Time: Sunday, August 2, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Ming-Hung Kao*+ and Dibyen Majumdar and Abhyuday Mandal and John Stufken
• Companies: The University of Georgia and University of Illinois at Chicago and The University of Georgia and The University of Georgia
• Keywords: Bayesian designs ; design efficiency ; genetic algorithms ; hemodynamic response function ; local optimal designs ; nonlinear models

Abstract

Efficient designs for event-related functional magnetic resonance imaging (ER-fMRI) are crucial, but obtaining them is arduous; it requires considerations of statistical and practical issues, and major computational efforts. Kao, Mandal, Lazar and Stufken, 2009, NeuroImage, 44: 849-856, propose an efficient approach for finding good ER-fMRI designs. As other methods known hitherto, their approach fails to take into account the link between two common statistical goals, namely estimation of the hemodynamic response function and detection of active brain voxels. In contrast, here we consider a nonlinear model to accommodate both goals in a unified setting. Under this model, we define design criteria to evaluate designs and search for local and Bayesian optimal designs via a knowledge-based genetic algorithm. We also compare designs that we obtain to those currently in use by researchers.