Low-frequency Functional Connectivity in fMRI - Biomedical Signal Analysis: Contemporary Methods and Applications - page 263

9

Low-frequency Functional Connectivity in fMRI

Low-frequency fluctuations ( < 0 . 08 Hz) temporally correlated between
functionally related areas have been reported for the motor, auditory,
and visual cortices and other structures [35]. The detection and quan-
tification of these patterns without user bias poses a current challenge in
fMRI research. Many recent studies have shown decreased low-frequency
correlations for subjects in pathological states or in the case of cocaine
use [199], which can potentially indicate normal neuronal activity within
the brain.
The standard technique for detecting low-frequency fluctuations has
been the crosscorrelation method. However, it has several drawbacks,
such as sensitivity to data drifts and choosing the reference waveform
when no external paradigm is present. The use of prespecified regions
of interest (ROI) or “seed clusters” has been the method of choice in
functional connectivity studies [35], [199]. The main limitation of this
method is that it is user-biased.
Model-free methods that have recently been applied to fMRI
data analysis include projection-based and clustering-based. The first
method, PCA [14, 242] and ICA [10, 77, 168, 170] extracts several
high-dimensional components from original data to separate functional
response and various noise sources from each other. The second method,
fuzzy clustering analysis [24, 53, 226, 285] or the self-organizing map
[84, 185, 285], attempts to classify time signals of the brain into pat-
terns according to temporal similarity among these signals.
Recently, self-organizing maps (SOM) have been applied to the
detection of resting-state functional connectivity [199]. It has been shown
that the SOM represents an adequate model-free analysis method for
detecting functional connectivity.
The present chapter elaborates this interesting idea and introduces
several unsupervised clustering methods implementing arbitrary dis-
tance metrics for the detection of low-frequency connectivity of the rest-
ing human brain. These techniques allow the detection of time courses
of low-frequency fluctuations in the resting brain that exhibit functional
connectivity with time courses in several other regions which are re-
lated to motor function. The results achieved by these approaches are
compared to standard model-based techniques.