On the Use of Linear-Modelling-based Algorithms for Physiological Noise Correction in fMRI Studies of the Central Breathing Control.

Abstract

A full characterization of the physiological behavior of human central chemoreceptors through fMRI is crucial to understand the pathophysiology of central abnormal breathing patterns. In this scenario, physiological noise and activity of interest may be naturally correlated. Here, we examined the adequacy of linear-modelling-based retrospective physiological noise correction for studies of the central breathing control. We focused on the relationship between a nonlinear model of BOLD response, hypothesized to describe neuronal specific activity, and noise modelled by correction algorithms. Analyses were performed on fMRI acquisitions from healthy subjects during a breath hold task. A general linear model including static nonlinearities in the response to end-tidal CO2 was applied to data preprocessed both with and without physiological noise correction. Relations between physiological noise and PETCO2 were explored both with linear and nonlinear measures. Lastly, parametric maps of noise spatial distribution were extracted. Our results evidenced that correction algorithms based on linear modelling remove components that are both linearly and nonlinearly related to end-tidal CO2, whereas uncorrected data showed spurious activations in regions outside gray matter. Thus, despite a correction step is fundamental, these algorithms are shown to be over-conservative approaches to noise correction and need to be adapted to the specific purpose.