Biomedical Engineering Reference
In-Depth Information
can then be performed just as is done for fMRI data, but with far
richer access to neuronal dynamics
(12, 13)
. The popular ECD
method of analysis, using point-like models for the generators,
will not work with real time MEG signals because rarely, if ever,
the activity of the normal brain is dominated by a single focal
source. Early MEG experiments with one or only a few sensors
could only obtain signal topographies by repeating a task or
delivering the same stimulus many times while recording the
resulting MEG signal. Although this is no longer necessary with
multi-channel systems, averaging remains a simple and powerful
way of drastically reducing the complexity of the data. Averaging
emphasizes activity that is precisely time-locked to an external
stimulus which is more likely to be dominated by contributions
that are fairly focal. As a result, ECD analysis often provides a very
good fit for the average evoked response. Some of the apparently
good fits obtained by ECD modeling of average data do reflect
the true nature of the generators. However, the average makes
up only a tiny part of the single trial MEG signal. Even the actual
average MEG signal is often a collection of distinct responses that
do not belong together
(5, 24)
and the apparent success of the
ECD model could be a mirage with some of the real generators
a fair distance away from the ECD loci
(25, 26)
. The likelihood
always remains that a very large part of the activity related to
the processing of the stimulus remains unexplained, lost in the
process of averaging before any analysis is made, as recent studies
have shown
(27, 28)
, but see reference
(29)
for a different view.
Despite the huge potential of MEG, its usefulness has been
questioned. MFT has been for many years the only technique
capable of real-time tomographic reconstructions, initially
because its implementation demanded what, at the time, was
super-computing resources
(30)
. What single trial MFT solutions
described (tomographic estimates of real-time brain activity) was
often confused with descriptive measures of power of the MEG
signal and ECD fits to average data. Tapping into single trial data
tomographically was giving a view of brain activity that was far
more dynamic than the smooth evolution produced by fits to
average data using ECDs that the MEG and EEG communities
were familiar with for decades. Although this dynamic view was
much more in step with invasive measures of activity, it was
not adopted initially because the experience from ECD analysis
and the non-uniqueness of the inverse problem had convinced
many practitioners that tomographic analysis was impossible. In
recent years, other techniques have emerged making better use
of the information in the raw MEG signal than the averaging.
Results obtained with these methods have vindicated many of the
early MFT results. Beamformer techniques, in particular, have
been increasingly used recently with good effect
(8)
. One of the
most exciting results of MFT analysis was the identification of
