Biomedical Engineering Reference
In-Depth Information
neous fluctuations in EEG features that can be considered as vigilance/alertness
monitors would be used to inform the fMRI model. In one such study [93] the poste-
rior alpha rhythm (which is known to desynchronize with attention and increase in
amplitude with mental fatigue and restfulness) has been used as a regressor in the
GLM for fMRI image reconstruction. The main premise of these efforts is to deal
with the fact that there is no idealized resting state, an assumption that is fundamen-
tal to most functional MR imaging techniques.
12.2.5 The Inverse EEG Problem: fMRI Constrained EEG Source Localization
The apparent appeal of combining EEG and fMRI is to get the best of both worlds,
that is, higher temporal and spatial resolution attributed to each modality respec-
tively. The inverse EEG source localization problem is highly ill posed, is heavily
dependent on the forward model assumptions, and has infinite possible solutions to
account for any given boundary condition. The use of BOLD response data to
somehow constrain the solution of the inverse problem seems appealing and has
generally yielded better results. But one needs to be mindful of the fact that EEG and
BOLD changes are on different temporal scales, and more importantly, the presence
of changes in one of them does not necessarily imply detectable changes in the other.
For instance, neural activities in deeper structures of the brain are easily detect-
able by fMRI but make minimal or no contribution to the EEG. Similarly, dipoles
oriented tangentially to the scalp surfaces or opposing dipoles on either bank of a
deep sulcus, for instance, result in negligible contributions to the EEG [94]. Con-
versely, one could record a large EEG contribution due to synchronous activity of
only a few neurons but with minimal metabolic load and hence negligible contribu-
tion to the BOLD signal. Recent evidence from single-unit neuronal recordings
using microelectrodes suggests that the substrates for neuronal activity and that of
BOLD changes do not exactly match spatially. It is important to keep in mind the
mismatches between the electrical and hemodynamic signals of the same neuronal
event. However, incorporating fMRI data into the model used for solving the
inverse EEG problem has the potential to improve spatial localization without
compromising temporal resolution.
12.2.6 Ongoing and Future Directions
EEG and fMRI provide complementary information regarding neuronal dynamics,
albeit at different spatial and temporal scales. Existing methodologies largely rely
on using one of these two methodologies to study the other. However, recent meth-
ods that make joint inferences about neuronal activity from both electrical and
hemodynamic data seem to offer additional benefits that traditional techniques can-
not. These include using fMRI to inform the EEG forward model (for generating the
inverse source localization problem), as well as using EEG to inform the fMRI for-
ward model (by using EEG features as regressors in the GLM). A recent method that
jointly analyzes both classes of data simultaneously is the N-PLS (multiway partial
least squares) algorithm [95], which decomposes the multidimensional EEG and
fMRI data into a sum of time-frequency “atoms.” This is based on singular value
decomposition of the covariance matrix between fMRI (spatial and temporal activ-
Search WWH ::
Custom Search