Information Technology Reference
In-Depth Information
within cell assemblies. These currents circulate and return through secondary vol-
ume currents (see Section 8.4.4), which depend on the geometry and conductivity
properties of head tissues, as discussed in the upcoming sections.
8.4.2 Anatomical Modeling of Head Tissues and Neural Sources
To conduct subject-specific anatomical modeling, the geometry of head compart-
ments is obtained from the analysis of the T1-weighted MRI data. Head tissues,
such as gray and white matter, skull bone, scalp, cerebrospinal fluid (CSF), and fat
are classified from MRI data using segmentation techniques [17]. The geometry of
these components is represented using surface and volume tessellation techniques
for subsequent modeling of their electromagnetic properties as discussed in Sec-
tion 8.4.4.
The anatomical domain of sources can then be constrained to the gray matter
volume, or rather to the cortical surfaces since NSI cannot discriminate generators
from different cortical layers. With this surface approach, dipole orientations can
easily be constrained to point along the normal direction of the cortical surfaces
(i.e., in the direction of the apical dendrites of cortical pyramidal neurons). Non-
cortical structures can be modeled as volumetric source subspaces without dipole
orientations constraints.
If a subject's MRI is not available, a standardized MRI may be warped to opti-
mally fit the subject's anatomy based on the individual's digitized head shape points.
The warped brain anatomy can then be used as a standardized volumetric source
space and for standardized forward modeling [16].
8.4.3 Multimodal Geometric Registration
As a prerequisite to the modeling of EM signals originating from anywhere in the
source space, the sensor and source positions and orientations must be expressed in
the same coordinate system. This registration process is usually done by transform-
ing (i.e., using rigid-body translation and rotation based on anatomical landmarks)
the sensor positions and orientations to the coordinate system of the MRI, where the
NSI generators are modeled. Errors in the definition of the fiducial anatomical land-
marks in either the NSI or MRI modality can result in poor geometrical alignment
and therefore, critical errors in the modeling of generators. Improved registration
can be achieved by matching a larger number of fiducial points - beyond the three
typical nasion and auricular locations - such as a digitized head-shape or the loca-
tions of EEG electrodes to the skin surface extracted from MRI data. Careful align-
ment can help minimize geometrical registration errors within the range of under
5 mm.
