Temporal correlation compensation (whitening) of the fMRI data using Akaike Information Criterion

Abstract

This work concerns the noise present in BOLD-fMRI (Blood-Oxygen-Level-Dependent: functional Magnetic Resonance Imaging, a technique that evaluates the levels of oxygen in the blood vessels of the brain. It is known that there is a temporal correlation present in the BOLD-fMRI's noise signal, complicating the estimation of the active regions of the brain due to an external stimulus. Using SPM-GLM methods (Statistical Parametric Mapping — General Linear Methods), denoised signal and response coefficients from all voxels (Volume Element) are estimated. The comparison between both signals gives an approximation of its noise signal. Using Akaike Information Criterion, this technique estimates the best model's order to decorrelate the noise and pre-whiten. Moreover, this algorithm recalculates new unknown parameters until a minimum threshold is achieved. Final results obtained were analyzed and concluded to have less false-positives, allowing a better definition of the real active regions.