Biomedical Engineering Reference
In-Depth Information
To model the spatial nature of the imaging data, SPM techniques make use of
random field theory. In simultaneously recorded EEG and fMRI, features of interest
from the EEG can serve as regressors in the GLM. Traditionally, the regressors of
the GLM model consist of behavioral and experimental conditions (appropriately
convolved with a standardized HRF). When investigating the hemodynamic corre-
lates of various spectral bands of the EEG, the GLM can be appropriately modified
to incorporate the band power as a regressor in generating the statistical fMRI
image. This approach has been used to study postmovement beta rebound [78], pos-
terior alpha rhythm [73], and so forth. Figure 12.14 schematically depicts the vari-
ous steps involved in generating the EEG regressors for the GLM model [73].
12.2.3.2 Epilepsy: Spike-Correlated fMRI Analysis
Simultaneous EEG-fMRI is an emerging tool for studying the focus and spread of
epileptic activity, as well as to gain a better understanding of its hemodynamic cor-
relates. The methods used for epileptic spike-correlated fMRI analysis are some-
what similar to those described in the previous section, but with some marked
differences. These arise from the fact that spike activity in the EEG is neither spa-
tially nor temporally regular, and the “regressors” for fMRI therefore are not as
simple as those for an EEG rhythm. Additionally, the HRF during epileptic activity
100
0
−
100
1
0
1
2
3
4
5
6
7
8
9
10
HF and gradient
artifact removal
100
0
2
−
100
0
3
5
6
7
8
9
10
1
2
4
5
Wavelet analysis
10
15
3
20
1
2
3
4
5
6
7
8
9
10
1
Average over
frequencies of
Alpha-Rhythm
0.5
0
4
0
1
2
3
4
5
6
7
8
9
10
Time(s)
Regressor for SPM
Convolution with
hemodynamic
response function
Figure 12.14
Analysis of EEG acquired simultaneously with fMRI. This figure gives a schematic representation
of the different steps (indicated by arrows) in the data analysis. Step 1, application of algorithms for MR artifact
correction; step 2, time-frequency decomposition by wavelet analysis; step 3, estimation of the alpha power by
averaging alpha-band frequencies; step 4, convolution with the hemodynamic response function to estimate a
predictor for the BOLD signal. (
From:
[73]. © 2003 Elsevier. Reprinted with permission.)
Search WWH ::
Custom Search