fMRI signal issues

fMRI signal issues

What are all the things/artifacts/issues that can show up in fMRI data, especially with an eye to spatial accuracy and mesoscopic measurement of BOLD response patterns?

From T2* to percent signal change (PSC)

Let's talk about standard gradient-echo T2*-weighted EPI imaging (i.e. fMRI):
A parameter, known as T2* (a time constant) is what determines the signal you measure for each voxel. As your experimental demands alter local oxygen needs/consumption, T2* varies and thus your measured signal fluctuates. The rise and fall of the signal is due to variation in T2*

B = measured signal = S0 * exp(-TE/T2s) + noise;
A = during activation = S0 * exp(-TE/T2snew) + noise;
PSC = percent signal change = (A-B)/B * 100
Notes/caveats:
  • Most fMRI is just looking at the magnitude component
  • Noise is Rician (this will be more of a problem for very high spatial resolution data when you approach the noise floor)
  • TE is just at the "center" line of k-space, so it's actually varying for other lines of k-space
  • When you sample this, you are collecting different regions of k-space (aka different spatial frequencies) at different points along this decay curve.
  • Veins have low T2* (for susceptibility reasons (and dependent on field strength)); the simulation shows that for a fixed TE and a fixed delta-T2*, this implies that veins will enjoy large percent signal change. (But note that veins may also have a large delta T2* which may be an additional reason that PSC goes way up for veins.)
  • Across the brain, different voxels will have different T2* due to: different vascular density, partial voluming (e.g. gray matter mixed with white matter); large susceptibility artifacts (e.g. ear canals); it could also vary across cortical depth due to myelination, vascular density; orientation of the vessels relative to the B0 direction
  • Multi-echo fMRI attempts to estimate the S0 and T2* at each voxel. This can be done in an anatomical sense (long acquisition) or in a dynamic fMRI sense.
  • The theory for multiecho denoising is that changes in S0 are "artifact" and that what we want to get at are the changes in T2*.

References for some of the discussed points:

  • T2* (R2*) across magnetic field streghts, See Figure 1 Panel A: Uludag, K., Müller-Bierl, B., & Ugurbil, K. (2009). An integrative model for neuronal activity-induced signal changes for gradient and spin echo functional imaging. NeuroImage, 48(1), 150–165.  https://doi.org/10.1016/j.neuroimage.2009.05.051 
  • Irati Markuerkiaga, José P. Marques, Lauren J. Bains, David G. Norris. An in-vivo study of BOLD laminar responses as a function of echo time and static magnetic field strength bioRxiv 2020.07.16.206383; doi:  https://doi.org/10.1101/2020.07.16.206383 
  • Koopmans, P. J., Barth, M., Orzada, S., & Norris, D. G. (2011). Multi-echo fMRI of the cortical laminae in humans at 7T. NeuroImage, 56(3), 1276–1285.  https://doi.org/10.1016/j.neuroimage.2011.02.042 

From PSC to beyond

What happened to your data, fMRI user?
  • raw EPI volumes
  • motion correction (achieved through interpolation) [we can lump into this correction for spatial geometric distortion]
  • slice time correction (achieved through interpolation)
  • then, some analysis of the fMRI time-series data

GLM

  • What's the deal with using derivative of GLM predictors?

Miscellaneous

  • Ernst optimal flip angle: flip-angle = acos(exp(-TR/T1))/pi*180. T1 at 3T is 1333.3333 ms (see below link). (At 7T, T1 is 2132 ms.) So, you just substitute your proposed TR into the equation, and the resulting flip angle should provide maximum longitudinal signal strength (after equilibriation).
  • Also see  http://www.mritoolbox.com/ParameterDatabase.html  for information on parameters...