6 Biomechanical Modelling of the Brain for Neurosurgical Simulation. . .
139
Fig. 6.3 3D magnetic
resonance image presented as
a triplanar cross-section. 3D
rendering of ventricles is
shown in red. Public domain
software Slicer
(www.slicer.org), developed
by our collaborators from the
Surgical Planning Laboratory,
Harvard Medical School, was
used to generate the image
can be constructed with approximately 1 mm accuracy and that higher accuracy
is probably not required. One must then decide which brain structures should be
explicitly included in the model and which omitted. As described in Chap. 2,
anatomists recognise well over a thousand structures within the brain. However,
very little is known about the relative mechanical properties of these structures. The
‘maps’ of brain regional mechanical properties from MR elastography are not yet
reliable and are only valid for small strain (linear viscoelastic) [14, 15]. Therefore,
even the most sophisticated models in current use by the scientific community
typically only include brain parenchyma, ventricles, tumour (if present, for attempts
to estimate tumour mechanical properties, see, e.g. [16–18].) and skull. I refer the
reader to Chap. 4 of this book where mechanical properties of brain tissues are
discussed in great detail.
A necessary step in constructing patient-specific models of brain geometry is
medical image segmentation. Segmentation is a process that identifies different parts
of the brain on the image; see Fig. 6.4.
Unfortunately, despite years of effort by the medical image analysis community,
a generally accepted automatic segmentation method for brain MR images is not
yet available. In practice, very laborious semi-automatic or manual methods are
employed [10, 19]. It is clear that if one attempted to include many brain structures
in the patient-specific biomechanical model, then one would need to identify them
in the medical image and segment them. This would be a daunting task that at the
time of writing does not appear to be practical.
On the other hand, when a generic application that does not require patientspecific data is considered, the 3D geometry of essentially all structures that might
possibly be of interest can be imported from electronic brain atlases. For example, a
hippocampus is often of interest, and its geometry and location can be clearly seen
in Fig. 2.6 of Chap. 2.
139
Fig. 6.3 3D magnetic
resonance image presented as
a triplanar cross-section. 3D
rendering of ventricles is
shown in red. Public domain
software Slicer
(www.slicer.org), developed
by our collaborators from the
Surgical Planning Laboratory,
Harvard Medical School, was
used to generate the image
can be constructed with approximately 1 mm accuracy and that higher accuracy
is probably not required. One must then decide which brain structures should be
explicitly included in the model and which omitted. As described in Chap. 2,
anatomists recognise well over a thousand structures within the brain. However,
very little is known about the relative mechanical properties of these structures. The
‘maps’ of brain regional mechanical properties from MR elastography are not yet
reliable and are only valid for small strain (linear viscoelastic) [14, 15]. Therefore,
even the most sophisticated models in current use by the scientific community
typically only include brain parenchyma, ventricles, tumour (if present, for attempts
to estimate tumour mechanical properties, see, e.g. [16–18].) and skull. I refer the
reader to Chap. 4 of this book where mechanical properties of brain tissues are
discussed in great detail.
A necessary step in constructing patient-specific models of brain geometry is
medical image segmentation. Segmentation is a process that identifies different parts
of the brain on the image; see Fig. 6.4.
Unfortunately, despite years of effort by the medical image analysis community,
a generally accepted automatic segmentation method for brain MR images is not
yet available. In practice, very laborious semi-automatic or manual methods are
employed [10, 19]. It is clear that if one attempted to include many brain structures
in the patient-specific biomechanical model, then one would need to identify them
in the medical image and segment them. This would be a daunting task that at the
time of writing does not appear to be practical.
On the other hand, when a generic application that does not require patientspecific data is considered, the 3D geometry of essentially all structures that might
possibly be of interest can be imported from electronic brain atlases. For example, a
hippocampus is often of interest, and its geometry and location can be clearly seen
in Fig. 2.6 of Chap. 2.