Modeling and Activation Detection in fMRI Data Analysis

We employ our previously proposed framework [18] for the analysis of event-related functional magnetic resonance imaging (fMRI) data. In [18], we use a Gaussian blurring kernel to explicitly model the spatial correlation introduced by the scanner. In the present paper, we propose an improved strategy for estimating the extent of this spatial blurring. We also propose a new algorithm for performing activation detection. We illustrate the promise of our algorithm by comparing it with the widely used general linear model (GLM) method. In synthetic data experiments, under the same probability of false alarm, the probability of correct detection of our method is up to 40% higher than GLM. In real data experiments, through anatomical analysis and benchmark testing using block paradigm results, we demonstrate that our algorithm tends to produce fewer false alarms than GLM.