Advances and Applications in Statistics
Volume 51, Issue 6, Pages 397 - 426
(December 2017) http://dx.doi.org/10.17654/AS051060397 |
|
BAYESIAN 3-DIMENSIONAL SPATIAL VARIABLE SELECTION MODELING OF VOXEL-SPECIFIC HRFS FOR LOCALIZATION IN FMRI TIME SERIES DATA
Nasrin Borumandnia, Hamid Alavi Majd, Farid Zayeri, Ahmad Reza Baghestani, Mahmood Reza Gohari, Seyyed Mohammad Tabatabaei and Fariborz Faeghi
|
Abstract: This research describes a new fully Bayesian spatiotemporal model to analyze fMRI studies. We have considered recent developments in Bayesian spatiotemporal models for detecting neuronal activation in fMRI experiments. The complete relationship between the neuronal activation and the blood oxygenation level dependent signal has not been fully modeled yet. Our goal is to provide an analytical framework that considers the complex temporal and spatial correlation structures of fMRI data as well as the complex relationship between neuronal activity and its hemodynamic response function, HRF. In the temporal dimension, we assumed an autoregressive structure on error terms and also parameterize the HRF’s shape parameter. So we modeled the data using the HRFs with the voxel-dependent shape parameters. We account for the complex three-dimensional spatial correlation structure of the voxels using an Ising prior on parameters that are for selecting the activated voxels. For inference, we combine the component wise MCMC technique with the auxiliary variable method. We investigate the properties of the model through its performance on three-dimensional simulated data sets with the block design. Also, we implemented the method on two real data. The first one is the n-back from MyConnectome project and the second one is the auditory data. |
Keywords and phrases: Bayesian spatiotemporal model, Ising prior, hemodynamic response function, activation inference, fMRI data. |
|
Number of Downloads: 462 | Number of Views: 1358 |
|