Bayesian log-Gaussian Cox process regression with applications to fMRI meta-analysis
Year: 2017
Type: preprint
Abstract: A typical neuroimaging study will produce a 3D brain statistic image that summarises the evidence for activation during the experiment. However, for practical reasons those images are rarely published; instead, authors only report the (x,y,z) locations of local maxima in the statistic image. Neuroimaging meta-analyses use these foci from multiple studies to find areas of consistent activation across the human brain. However, current methods in the field do not account for study-specific characteristics. Therefore, we propose a fully Bayesian model based on log-Gaussian Cox processes that allows for the inclusion of study-specific variables. We present an efficient MCMC scheme based on the Hamiltonian Monte Carlo algorithm to simulate draws from the posterior. Computational time is significantly reduced through a parallel implementation using a graphical processing unit card. We evaluate the method on both simulated and real data.
Source: arXiv (Cornell University)
Language: English
Subfield: Applied Mathematics
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2578267374