New Filtering Approaches to Improve the Classification Capability of Resting-state fMRI Transfer Functions

Abstract

Resting-state functional magnetic resonance imaging (fMRI) uses spontaneous regional brain activity to identify functional networks. Transfer functions (TF) can evaluate the amplification of resting-state fMRI signal frequency components from one brain region to another, but are highly susceptible to noise spikes. Resting-state fMRI’s low-temporal resolution implies that the high frequency noise characteristics necessary to implement Weiner filtering are not available. We investigated new approaches that replace the standard Weiner filter noise parameter with an alternative outlier suppression parameter (OSP) to identify and remove inaccurate TF estimates. When compared to standard TF approaches, our new filtering approaches shows an improved ability to distinguish optic neuritis (ON) patients from healthy volunteers, as well as patients experiencing ON as a clinically isolated syndrome (CIS) from ON patients with relapsing-remitting multiple sclerosis (RRMS).