6 Biomechanical Modelling of the Brain for Neurosurgical Simulation. . .
149
We recommend that new real-time solution algorithms be verified against
well-established solution procedures implemented in commercial software. The
assumptions of the biomechanical model need to be evaluated against available
experimental data. The biomechanical model should then be validated by comparing
the solutions computed using established numerical procedures and experimental
results. If these hurdles are cleared and it can be also demonstrated that replacing
established numerical procedures with the specialised ones developed for real-time
applications does not affect the computed results, then one may treat the ‘software
system’ with some degree of confidence.
How accurate should the results of the computational biomechanics model of
the brain be? Accuracy of manual neurosurgery is not better than 1 mm. The voxel
size of the best available experimental tool for model validation – currently the
intra-operative MRI – is of the same order. Therefore, the computed intra-operative
displacements do not need to be more accurate than about 1 mm. We may note here
that, paradoxically, this accuracy requirement is much less stringent than those used
in traditional engineering disciplines. In image-guided surgery applications, we are
not interested in stress distributions, only in the displacement field. This is one of the
reasons why simple constitutive models of the brain tissue can be used. However,
for surgical simulation applications, we need to compute reaction forces on surgical
tools that will be fed back to the user through a haptic interface. At present there is
no consensus regarding how accurate this haptic feedback should be to facilitate
a realistic experience. Given the present state of biomechanical knowledge, the
best that can be achieved is qualitative agreement between real and computed
interaction forces; see, e.g. [71]. This is despite examples of excellent agreement
between computations and phantom experiments [90] as well as controlled in vitro
experiments [91].
6.3 Application Example: Computer Simulation of Brain
Shift
A particularly exciting application of nonrigid image registration is in intraoperative image-guided procedures, where pre-operative scans are warped onto
sparse intra-operative images [7, 58]. We are especially interested in registering
high-resolution pre-operative MRIs with lower-quality intra-operative imaging
modalities, such as multiplanar MRIs and intra-operative ultrasound. To achieve
accurate matching of these modalities, precise and fast algorithms to compute tissue
deformations are fundamental.
Here we present selected results of the analysis of 33 cases of craniotomyinduced brain shift representing different situations that may occur during neurosurgery [92, 93].
149
We recommend that new real-time solution algorithms be verified against
well-established solution procedures implemented in commercial software. The
assumptions of the biomechanical model need to be evaluated against available
experimental data. The biomechanical model should then be validated by comparing
the solutions computed using established numerical procedures and experimental
results. If these hurdles are cleared and it can be also demonstrated that replacing
established numerical procedures with the specialised ones developed for real-time
applications does not affect the computed results, then one may treat the ‘software
system’ with some degree of confidence.
How accurate should the results of the computational biomechanics model of
the brain be? Accuracy of manual neurosurgery is not better than 1 mm. The voxel
size of the best available experimental tool for model validation – currently the
intra-operative MRI – is of the same order. Therefore, the computed intra-operative
displacements do not need to be more accurate than about 1 mm. We may note here
that, paradoxically, this accuracy requirement is much less stringent than those used
in traditional engineering disciplines. In image-guided surgery applications, we are
not interested in stress distributions, only in the displacement field. This is one of the
reasons why simple constitutive models of the brain tissue can be used. However,
for surgical simulation applications, we need to compute reaction forces on surgical
tools that will be fed back to the user through a haptic interface. At present there is
no consensus regarding how accurate this haptic feedback should be to facilitate
a realistic experience. Given the present state of biomechanical knowledge, the
best that can be achieved is qualitative agreement between real and computed
interaction forces; see, e.g. [71]. This is despite examples of excellent agreement
between computations and phantom experiments [90] as well as controlled in vitro
experiments [91].
6.3 Application Example: Computer Simulation of Brain
Shift
A particularly exciting application of nonrigid image registration is in intraoperative image-guided procedures, where pre-operative scans are warped onto
sparse intra-operative images [7, 58]. We are especially interested in registering
high-resolution pre-operative MRIs with lower-quality intra-operative imaging
modalities, such as multiplanar MRIs and intra-operative ultrasound. To achieve
accurate matching of these modalities, precise and fast algorithms to compute tissue
deformations are fundamental.
Here we present selected results of the analysis of 33 cases of craniotomyinduced brain shift representing different situations that may occur during neurosurgery [92, 93].