False Discovery Rate Control in Magnetic Resonance Imaging Studies via Markov Random Fields
Magnetic resonance imaging (MRI) is widely used to study the population effects of covariates on brain morphometry. Inferences from these studies often require the simultaneous testing of millions of statistical hypotheses. Such scale of simultaneous testing is known to lead to large numbers of false positive results. False discovery rate (FDR) controlling procedures are commonly employed to mitigate against false positives. However, current methodologies in FDR control only account for the marginal significance of hypotheses and are not able to take into account spatial relationships, such as in MRI studies. In this article, we present a novel method for incorporating spatial dependencies in the control of FDR through the use of Markov random fields. Our method is able to automatically estimate the relationship between spatially dependent hypotheses by means of pseudo-likelihood techniques. We show that the our spatial FDR control method is able to outperform marginal methods in simulations of spatially dependent hypotheses. Our method is then applied to investigate the effect of aging on brain morphometry using data from the PATH study. The results of our investigation were found to be in correspondence with the brain aging literature.