Modeling and Suppression of Respiration-Related Physiological Noise in Echo-Planar Functional Magnetic Resonance Imaging Using Global and One-Dimensional Navigator Echo Correction Robert L. Barry 1,2 and Ravi S. Menon 1–3 * A major source of noise in functional magnetic resonance im- aging (fMRI) arises from modulations in the local magnetic field in the head due to motion of the subject’s chest through the respiratory cycle, and this physiologic noise can nullify the gains in statistical power expected by the use of higher mag- netic fields for fMRI. In particular, fMRI data acquired using echo-planar imaging (EPI) are very sensitive to these spatially and temporally varying respiration-induced frequency offsets. In this study, accurate 3D magnetic field maps in the head were measured and used to determine the frequency offsets at the two extremes of the respiratory cycle. From these maps, spa- tially dependent frequency variations from about 1.0 Hz to 1.5 Hz were measured in the brain through the respiratory cycle. Simulations of a typical axial EPI fMRI experiment ac- quired in the presence of this measured field variation were performed, demonstrating regional image intensity variations between 1 and 5% in single pixel time series. The inadequacy of either global or 1D navigator echo corrections to measure and suppress respiratory-induced noise in fMRI time series is dem- onstrated. The nature of the spatial variations observed sug- gests that 2D approaches should be considered. Magn Reson Med 54:411– 418, 2005. © 2005 Wiley-Liss, Inc. Key words: physiologic noise; respiration; navigator echo; func- tional magnetic resonance imaging; time series analysis First proposed by Ogawa et al. (1) in 1990 and later dem- onstrated in the human brain (2– 4), blood oxygenation level-dependent (BOLD) functional magnetic resonance imaging (fMRI) is currently the dominant technique for imaging human brain function. Since fMRI measures the very small modulations in BOLD signal intensity that oc- cur during changes in brain activity, it is also very sensi- tive to small signal intensity variations caused by subject movement during the scan (5). Gross subject motion can be reduced by immobilizing the subject’s head using foam padding and a bite-bar, but sporadic events such as swal- lowing or yawning can still change functional activation maps in undesirable and unpredictable ways (6). However, even if a subject’s head could be maintained completely stationary, other sources of physiologic noise will always be present in fMRI experiments (7). Research has shown that components of physiologic noise in the fMRI time series arise from: (1) fluctuations in cerebral metabolism, (2) cerebral blood flow, (3) cerebral blood volume, (4) respiration, and (5) cardiac pulsatility. These components are signal-dependent and may or may not be influenced by the echo time TE used in the scan (8). Physiologic noise levels have also been shown to vary between gray matter and white matter (8) and different cortical layers (9). There is considerable evidence that the first three components mentioned may well be reflective of underlying neural or vascular fluctuations. While they do contribute to signal fluctuation, the nature of those fluctuations likely plays an important role in approaches such as resting brain connec- tivity (10), and as such their removal should be treated with care. However, respiratory and cardiac movements can safely be considered extraneous sources of fluctuation in the fMRI time series. The measurement and suppression of physiologic mo- tion caused by the respiratory and cardiac cycles are of great importance, more so in fMRI at 3 T and beyond, where the mechanisms at play can reduced the statistical gains one might expect on the basis of the demonstrated increases in signal-to-noise ratio (SNR) and BOLD contrast (8,11–13). Cardiac pulsation and respiration cause inter- cranial movement of the brain through the cerebrospinal fluid (CSF) (14 –16) and also introduce complex spatial and temporal changes in the cerebral hemodynamics (17,18). A Fourier transform of a typical fMRI time course would reveal noise peaks at the frequencies of the respi- ratory and cardiac cycles (and their respective harmonics) (11,10), and attempts have been made at suppressing these components using digital filters (19). Filter approaches require that the cardiac and respiratory signals not be aliased, something typically not achieved with the repeti- tion times used for whole-head fMRI. In addition to direct mechanical modulation of the brain fMRI intensity by respiratory and cardiac activity, the gross movement of the chest and the organs within the chest cavity induces spa- tial and temporal fluctuations in the magnetic field throughout the entire head (20,21). The fluctuations in the local magnetic fields manifest themselves as dynamic off- resonance effects that distort the acquired MR image, and since it is observed that physiologic noise increases with 1 Laboratory for Functional Magnetic Resonance Research, Robarts Research Institute, London, Ontario, Canada. 2 Department of Biomedical Engineering, University of Western Ontario, Lon- don, Ontario, Canada. 3 Department of Diagnostic Radiology and Nuclear Medicine, University of Western Ontario, London, Ontario, Canada. Grant sponsor: NIH; Grant number: 1RO1EB002739; Grant sponsor: CIHR; Grant number: MOP-64399; Grant sponsor: Canada Research Chairs Pro- gram. *Correspondence to: Ravi S. Menon, Laboratory for Functional Magnetic Resonance Research, Robarts Research Institute, P.O. Box 5015, 100 Perth Drive, London, Ontario N6A 5K8, Canada. E-mail: rmenon@imaging.robarts.ca Received 14 December 2004; revised 16 March 2005; accepted 28 March 2005. DOI 10.1002/mrm.20591 Published online in Wiley InterScience (www.interscience.wiley.com). Magnetic Resonance in Medicine 54:411– 418 (2005) © 2005 Wiley-Liss, Inc. 411