Predicting Post-Treatment Brain Activity Using a Bayesian Hierarchical Model

• Activity Number: 455
• Type: Topic Contributed
• Date/Time: Wednesday, August 9, 2006 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): F. DuBois Bowman*+ and Ying Guo
• Companies: Emory University and Emory University
• Keywords: neuroimaging ; fMRI ; PET ; hierarchical model ; prediction

Abstract

There is growing interest in the use of functional neuroimaging data to help inform medical decisionmaking. For example, knowing the impact of treatment on distributed patterns of brain activity, measured using fMRI or PET, may shed light on whether a treatment is appropriate for a particular patient. The complication is that post-treatment scans are not at a physician's disposal when a treatment decision is made. We develop a Bayesian hierarchical model that enables the use of pre-treatment brain scans and subject-specific health characteristics to predict post-treatment brain function. The first level of the hierarchy models within-subject activation effects, and the second level models subject-specific effects in terms of population parameters. Estimation is performed using the EM algorithm. We evaluate the accuracy of our proposed prediction method using K-fold cross-validation.