Computational Parametric Mapping: A Bayesian and Brain-centric Framework for Model-based fMRI

Specialeforsvar ved Line Rosendal Meldgaard Pedersen

Titel:  Computational Parametric Mapping: A Bayesian and Brain-centric Framework for Model-based fMRI

Abstract: Functional magnetic resonance imaging (fMRI) is a non-invasive technique for studying neural activity through changes in the blood-flow in the brain, and in the  previous decades it has become subject for increasing interest within neuroscience especially in combination with psychology. We will in this thesis formulate a novel approach, Computational Parametric Mapping (CPM), within a Bayesian framework for investigating cognitive processes in terms of an underlying computational model. The novelty of this approach lies within the aim of computing a posterior distribution, rather than a point estimate, for the latent parameters within the computational model from the fMRI. Furthermore the posterior distribution is used for formalizing a criteria for selecting voxels of interest. Through a simulation study we will examine the robustness of CPM before applying the methodology to a data set from a study of reinforcement learning with the purpose of estimating the learning rate parameter of a computational model specified by the Rescorla-Wagner learning rule. The simulation study showed to be acceptable for detecting the true parameter of an underlying computational model, even under violations of the data being independent and identically distributed.The results obtained from the CPM analysis on the fMRI data differed widely among subjects, where our selection criteria proved to discard a large proportion of the considered voxels even within a pre-dened region of interest given by the anterior cingulate cortex. Furthermore the estimated learning rate parameters showed great divergence across subjects. In extension of the CPM analysis we investigated if the learning rates showed evidence of a gradient in either of the brains three dimensions - the anterior-posterior, left-right and inferior-superior dimension - using a Gaussian random effects model. Though some subjects showed the indication of a linear trend of learning rate as an effect of the position in the brain, the lack of validation of this modes means the results hereof remain inconclusive

Vejledere:  Bo Markussen
                  Kristoffer Madsen
Censor:      Søren Møller, SDU