Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 168

account strength of activated voxels. The hemodynamic information offered by fMRI
provided the norm in the source space for solution of the inverse problem.
Studies concerning EEG informed fMRI aim to identify BOLD correlates of EEG
events. In case of epilepsy the early applications involved a spike-triggered ac-
quisition mode whereby each fMRI scan was acquired following visualization of
spike [Symms et al., 1999]. In a more advanced approach [Liston et al., 2006] the
semi-automatic system was proposed based on spatio-temporal clustering of inter-
ictal EEG events. The clusters of interictal epileptic discharges were correlated to
BOLD activations. In a second step signal space was used to project scalp EEG onto
dipoles corresponding to each cluster. This allowed for identification of previously
unrecognized epileptic events, the inclusion of which increased the experimental ef-
ficiency as reflected by significant BOLD activation.
The general idea behind ERP-informed fMRI analysis is to correlate active brain
regions identified by fMRI with amplitude modulation of individual ERP. In the ex-
periment concerning auditory oddball paradigm [Eichele et al., 2005] in the first step
the data were de-noised by means of application of wavelets and ICA. Then single-
trial N1, P2, and P3 amplitudes of ERP were found. These amplitude vectors were
convolved with the function of hemodynamic response and used as regressors to find
the BOLD time course.
In the simultaneous measurements of ERP and fMRI the signals are recorded
which are spatially and temporally mixed across the brain, since they are volume-
conducted and temporally extended by hemodynamic response. The data in both
modalities are generated by multiple, simultaneously active overlapping neural pop-
ulations. In solving problems, where specific hypotheses regarding spatial and tem-
poral relationships are lacking or are ill-specified, blind source separation methods
are useful. In order to address the problem of mutual ERP-fMRI relation the ICA
approach was proposed in [Eichele et al., 2008]. The ICA was used in parallel: to
recover spatial maps from fMRI in terms of spatial independent components (sIC)
and time-courses from ERP in terms of time related independent components (tIC).
Then the components were matched across modalities by correlating their trial-to-
trial modulation. Inspecting the results, only one tIC component was found which
predicted selectively the time-course of one sIC component, there were no other co-
variances between components corresponding to both modalities.
An alternative approach is to fuse the ERP and fMRI signals in a common data
space. In [Moosmann et al., 2008] joint independent component analysis was pro-
posed for analysis of simultaneous single trial ERP-BOLD measurements from mul-
tiple subjects. The authors presented the results based on simulated data, which in-
dicated a feasibility of the approach. It seems that there is a long way to go to obtain
a satisfactory integration of EEG and fMRI modalities. Comparative experiments
involving intracranial recordings would be helpful in this respect.