RAFNI: Robust analysis of functional neuroimages with non-normal alpha-stable errors

Functional Magnetic Resonance Imaging (fMRI) is a non-inasive neuro-imaging method that is widely used in cognitive neuroscience. It relies on the measurement of changes in the blood oxygenation level resulting from neural activity. The technique is widely used in cognitive neuroscience. fMRI is known to be contaminated by artifacts. Artifacts are known to have fat tails and are often skewed therefore modeling the error using a Gaussian distribution is a not enough. In this paper, we introduce RAFNI, an extention of AFNI, which is an fMRI open source software for the Analysis of Functional NeuroImages. We are modeling the error introduced by artifacts using α-stable distribution. We demonstrate the applicability and efficiency of stable distributions on real fMRI. We show that the α-stable estimator gives better results than the OLS-based estimators.