Biomedical Engineering Reference
In-Depth Information
about the model order is not available a priori, especially for complex brain map-
ping conditions, and the resulting localization of higher order cortical functions can
sometimes be unreliable.
An alternative approach is whole-brain source imaging methods. These methods
apply voxel discretization over a whole brain volume, and assume a fixed source at
each voxel. These methods estimate the amplitudes (and directions) of the sources
by minimizing a cost function. One classic method of this kind is the minimum-norm
Since the number of voxels is usually much larger than the number of sensors,
the cost function should contain a constraint term that is derived from various prior
information about the nature of the sources. We have developed a powerful imaging
algorithm that incorporates a sparsity constraint that facilitates the sparse source
configurations. This algorithm, called the Champagne algorithm, is described in
The other class of imaging algorithm is the spatial filter. The spatial filter esti-
mates source amplitude at each location independently. It is often called a virtual
sensor method, because it forms a virtual sensor, which scans an entire source space
to produce a source image. The most popular spatial filter algorithms are adaptive
spatial filtering techniques, more commonly referred to as adaptive beamformers.
Adaptive beamformers are the topic of Chap.
3
in which a concise review on adap-
tive beamformers is presented. A comprehensive review of these algorithms is found
in [
2
]. The implementation of adaptive beamformers is quite simple, and they have
proven to be powerful algorithms for characterizing cortical oscillations. Therefore,
they are popular for the source localization from spontaneous brain activity, par-
ticularly resting-state MEG. We have recently proposed novel scanning algorithms
[
3
,
4
]. These methods have shown performance improvements over the adaptive
beamformers. One of these methods, called the Saketini algorithm, is described in
various source imaging algorithms in a unified framework and reveals a close rela-
tionship between algorithms that are considered quite different [
5
].
An enduring problem in MEG imaging is that the brain evoked responses to
sensory or cognitive events are small compared to the interfering magnetic field.
Typical sources of such interference include the background room interference from
power lines and electronic equipment, the interference with biological origins such as
heartbeat, eye blink, or other muscle artifacts. Ongoing brain activity itself, including
the drowsy-state alpha rhythm, is also a major source of interference. All existing
methods for brain source imaging are hampered by such interferences present in
MEG data. Several signal-processing methods have been developed to reduce inter-
ferences by preprocessing the sensor data before submitting it to source localization
algorithms. One such algorithm, called partitioned factor analysis, is described in
