Jump to content

Neuroimaging Data Processing/Processing/Steps/Physiological Noise Regression

From Wikibooks, open books for an open world
Neuroimaging Data Processing/Processing/Steps
Field Map Correction Physiological Noise Regression Temporal Filtering

Physiological Noise

[edit | edit source]

Physiological noise can seriously confound the signal measured by fMRI. While very low frequency fluctuations due to vascular / metabolic oscillations (< 0.01 Hz) are usually removed by Temporal Filtering high frequency confounds from breathing (~0.3 Hz) or heartbeat (~1.0 Hz) are harder to deal with. In standard fMRI sequences these frequencies are often undersampled (according to the Nyquist theorem) and thus aliased into lower frequencies [1]. Furthermore, breathing and cardiac rate fluctuate at low frequencies, affecting the cerebral blood flow (through CO2 vasodilation or blood pressure respectively) and thus eventually the BOLD response. [2][3]. Physiological fluctuations can thus be expressed in the frequency range of interest in resting state fMRI (< 0.1 Hz), possibly introducing spurious connectivity between simultaneously measured time series. A problem to bear in mind is that the physiological fluctuations and neural activity of interest might be coupled temporally. In that case, removing the former would remove (at least parts of) the latter as well.

Physiological Noise Regression

[edit | edit source]

Time series representing physiological noise can be included as nuissance regressors into a GLM. The part of the signal that is explained by the nuissance regressors will be removed from the residuals (which is the data of interest for rsfMRI). By removing a structured, non-random part of the residuals, nuissance regressors also render it more normally distributed or “white”. This helps fulfilling one basic assumption of the GLM, namely identically and normally distributed error terms. Unfortunately, the more regressors in the GLM, the fewer its degrees of freedom (= observations(voxels) – regressors), leading to a the more conservative significance testing of the model and single parameter weights

Obtaining Regressors

[edit | edit source]

There are different ways to arrive at reasonable regressors for physiological noise. The most straightforward is to acquire physiological measurements during the scan, using chest straps for the respiratory and a pulse oximeters for the cardiac rate. These measures are then fed into the analysis software to model appropriate nuissance regressors. However, getting the timing of MRI and physiological data straight can be very tricky.

If these measures are not available, the time series of the physiological noise can be modelled in different ways. Since the signal fluctuation of interest should primarily be located in the grey matter, one approach is to extract physiological noise time series of voxels located in white matter or ventricles.[4] Another method, called CompCor[5], focuses on the voxels showing highest variabilty, subsequently reducing the resulting time series to dominating components using PCA. It is also possible to apply ICA to separate noise components from components of interest. However, this always relies on the assumption that one is able to correctly tell them apart. Automated ICA based physiological noise removal methods like CORSICA [6] or PESTICA [7] should therefore be handled with care.

Resting State fMRI

[edit | edit source]

As mentioned above, resting state analysis is especially vulnerable to physiological artifacts, since they often manifest in the frequency range of interest (0.01 - 0.1 Hz). [8] Therefore, physiological noise regression is far more common in resting state than in task-based fMRI, which often relies on simple temporal filtering.

Implementation

[edit | edit source]

3dretroicor[9] uses an image-based method called RETROICOR [10] which estimates the phase of the cardiac and respiratory cycles at which an imaging slice is acquired and models a low-order Fourier series of this phase data for regression. The default order is 2 but can be adjusted via the -order option, as can the threshold for the detection of R-wave peaks in the input via -threshold. The input is 1D -resp / -card files. A command which also outputs the calculated respiration and cardiac wave for control (-respphase/-cardphase) would look like this

3dretroicor -resp 1resp_file -card card_file -cardphase cardphase.1D -respphase respphase.1D INPUTFILE

The phase outputs can be inspected with 1dplot

NOTE that the algorithm uses slice timing information for calculation. Therefore any steps destroying slice timing information, e.g. 3dvolreg motion correction, should be avoided before this step. Also, when some first volumes have been discarded, those time periods have to be discarded from the physiological data accordingly.

In afni_proc.py, this is not a default step but but can be included using the do block -ricor option. The default solver is OLS, polynomial order 2*runlength. The respective options to remove n timepoints from the physiology [default = 0], and apply the PHYSFILES in the regression respectively are:

-ricor_regs_nfirst n 
-ricor_regs PHYSFILE

PNM [11] also provides cardiac and respiratory RETROICOR regressors of requested order and additionally allows for cardiac-respiratory interaction regressors to be specified. Furthermore, it is possible to receive alternative physiological regressors like RVT (respiration volume per time) [12], HeartRate [13] and CSF regressors. It is strongly recommended in the manual, to provide scanner triggers (1/volume) to ensure timing accuracy between scanning and physiological data. Input is required as a single text file with different columns representing cardiac, respiratory and trigger information, along with a file describing these columns. It is possible and recommended to manually check accuracy of peak detection.

SPM8 does not have a built-in method for dealing with physiological noise. However, there is a couple of extensions, that can be used for this purpose. Visit: http://en.wikibooks.org/wiki/SPM/Physio. The DRIFTER toolbox for SPM can also be applied without external physiological data if the temporal resolution allows. [14]

References

[edit | edit source]
  1. Pallab K. Bhattacharyya, Mark J. Lowe, Cardiac-induced physiologic noise in tissue is a direct observation of cardiac-induced fluctuations, Magnetic Resonance Imaging, Volume 22, Issue 1, January 2004, Pages 9-13, ISSN 0730-725X, http://dx.doi.org/10.1016/j.mri.2003.08.003.
  2. Rasmus M. Birn, Jason B. Diamond, Monica A. Smith, Peter A. Bandettini, Separating respiratory-variation-related fluctuations from neuronal-activity-related fluctuations in fMRI, NeuroImage, Volume 31, Issue 4, 15 July 2006, Pages 1536-1548, ISSN 1053-8119, http://dx.doi.org/10.1016/j.neuroimage.2006.02.048.
  3. Karin Shmueli, Peter van Gelderen, Jacco A. de Zwart, Silvina G. Horovitz, Masaki Fukunaga, J. Martijn Jansma, Jeff H. Duyn, Low-frequency fluctuations in the cardiac rate as a source of variance in the resting-state fMRI BOLD signal, NeuroImage, Volume 38, Issue 2, 1 November 2007, Pages 306-320, ISSN 1053-8119, http://dx.doi.org/10.1016/j.neuroimage.2007.07.037.
  4. Andreas Weissenbacher, Christian Kasess, Florian Gerstl, Rupert Lanzenberger, Ewald Moser, Christian Windischberger, Correlations and anticorrelations in resting-state functional connectivity MRI: A quantitative comparison of preprocessing strategies, NeuroImage, Volume 47, Issue 4, 1 October 2009, Pages 1408-1416, ISSN 1053-8119, http://dx.doi.org/10.1016/j.neuroimage.2009.05.005.
  5. Yashar Behzadi, Khaled Restom, Joy Liau, Thomas T. Liu, A component based noise correction method (CompCor) for BOLD and perfusion based fMRI, NeuroImage, Volume 37, Issue 1, 1 August 2007, Pages 90-101, ISSN 1053-8119, http://dx.doi.org/10.1016/j.neuroimage.2007.04.042.
  6. Vincent Perlbarg, Pierre Bellec, Jean-Luc Anton, Mélanie Pélégrini-Issac, Julien Doyon, Habib Benali, CORSICA: correction of structured noise in fMRI by automatic identification of ICA components, Magnetic Resonance Imaging, Volume 25, Issue 1, January 2007, Pages 35-46, ISSN 0730-725X, http://dx.doi.org/10.1016/j.mri.2006.09.042.
  7. Erik B. Beall, Mark J. Lowe, Isolating physiologic noise sources with independently determined spatial measures, NeuroImage, Volume 37, Issue 4, 1 October 2007, Pages 1286-1300, ISSN 1053-8119, http://dx.doi.org/10.1016/j.neuroimage.2007.07.004.
  8. Kevin Murphy, Rasmus M. Birn, Peter A. Bandettini, Resting-state fMRI confounds and cleanup, NeuroImage, Volume 80, 15 October 2013, Pages 349-359, ISSN 1053-8119, http://dx.doi.org/10.1016/j.neuroimage.2013.04.001.
  9. http://afni.nimh.nih.gov/pub/dist/doc/program_help/3dretroicor.html
  10. Gary H. Glover, Tie-Qiang Li, David Ress, Image-based method for retrospective correction of physiological motion effects in fMRI: RETROICOR, Magnetic Resonance in Medicine, Volume 44, Issue 1, Pages 1522-2594, doi: 10.1002/1522-2594(200007)44:1<162::AID-MRM23>3.0.CO;2-E
  11. http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/PNM
  12. Rasmus M. Birn, Jason B. Diamond, Monica A. Smith, Peter A. Bandettini, Separating respiratory-variation-related fluctuations from neuronal-activity-related fluctuations in fMRI, NeuroImage, Volume 31, Issue 4, 15 July 2006, Pages 1536-1548, ISSN 1053-8119, http://dx.doi.org/10.1016/j.neuroimage.2006.02.048.
  13. Catie Chang, John P. Cunningham, Gary H. Glover, Influence of heart rate on the BOLD signal: The cardiac response function, NeuroImage, Volume 44, Issue 3, 1 February 2009, Pages 857-869, ISSN 1053-8119, http://dx.doi.org/10.1016/j.neuroimage.2008.09.029. (http://www.sciencedirect.com/science/article/pii/S1053811908010355)
  14. http://becs.aalto.fi/en/research/bayes/drifter/