178
K. Miller et al.
be used to solve for the internal deformation field of a patient-specific epilepsy case
and warp the pre-operative MRI to the intra-operative position. Comparing the poor
resolution and detail of the CT in Fig. 7.14a, to the overlaid image in Fig. 7.14c, it
is clear to see the vast improvement in accuracy that is possible with patient-specific
modelling.
7.4 Conclusions
The surgical treatment of epilepsy has the potential to permanently cure seizures,
but the process is hindered by the inability to confidently locate the seizure-onset
zone (SOZ) in the planning stage. Through the analysis of a real, patient-specific
case from the Boston Children’s Hospital, we have shown that it is possible to apply
biomechanical modelling and finite element methods to compute the deformation
field within the brain arising from invasive electrode placement and warp a preoperative MRI into the intra-operative configuration of the brain. This provides a
highly detailed map of the electrodes relative to neurological landmarks, making
it easier to confidently identify what tissue to resect and how. The generation of
the finite element mesh and model input took an experienced analyst approximately
2 days. This is acceptable in the research environment but too long for compatibility
with existing clinical workflows. Further work into more efficient methods of
patient-specific model generation is clearly needed [11, 28] (see Chaps. 10 and 11
of this book).
The model was analysed in Abaqus for 100 simulation seconds, which took
approximately 2 hours to complete and a further hour to process the results and
register the deformation field. As close to real-time processing speeds are not
demanded by this application, these simulation and analysis times are compatible
with existing clinical workflows, and further improvements, while helpful, are not
strictly necessary. These timeframes would fit within the 5–7-day period of data
collection, while the electrodes are on the brain.
The accuracy of segmentation is a limitation that affects the accuracy of the
model geometry. Segmentation remains a challenging and subjective process that
does not guarantee repeatability. Improvements in this area can only really come
from research into improved imaging and more robust segmentation algorithms;
however it is likely that manual input will always be required in this process.
For example, intensity and label fusion algorithms have demonstrated very high
reproducibility and accuracy indistinguishable from that of human experts [2, 3,
25].
Because manual surgery cannot achieve accuracy better than 1 mm, even slight
deviations by one or two voxels can still produce a better method of localisation
than is currently used in practice.
Finally, another area of interest in the surgical treatment of epilepsy is in the
modelling of depth electrodes. These are very long and slender needles that pierce
deep into the parenchyma and are extremely difficult to control and locate during
K. Miller et al.
be used to solve for the internal deformation field of a patient-specific epilepsy case
and warp the pre-operative MRI to the intra-operative position. Comparing the poor
resolution and detail of the CT in Fig. 7.14a, to the overlaid image in Fig. 7.14c, it
is clear to see the vast improvement in accuracy that is possible with patient-specific
modelling.
7.4 Conclusions
The surgical treatment of epilepsy has the potential to permanently cure seizures,
but the process is hindered by the inability to confidently locate the seizure-onset
zone (SOZ) in the planning stage. Through the analysis of a real, patient-specific
case from the Boston Children’s Hospital, we have shown that it is possible to apply
biomechanical modelling and finite element methods to compute the deformation
field within the brain arising from invasive electrode placement and warp a preoperative MRI into the intra-operative configuration of the brain. This provides a
highly detailed map of the electrodes relative to neurological landmarks, making
it easier to confidently identify what tissue to resect and how. The generation of
the finite element mesh and model input took an experienced analyst approximately
2 days. This is acceptable in the research environment but too long for compatibility
with existing clinical workflows. Further work into more efficient methods of
patient-specific model generation is clearly needed [11, 28] (see Chaps. 10 and 11
of this book).
The model was analysed in Abaqus for 100 simulation seconds, which took
approximately 2 hours to complete and a further hour to process the results and
register the deformation field. As close to real-time processing speeds are not
demanded by this application, these simulation and analysis times are compatible
with existing clinical workflows, and further improvements, while helpful, are not
strictly necessary. These timeframes would fit within the 5–7-day period of data
collection, while the electrodes are on the brain.
The accuracy of segmentation is a limitation that affects the accuracy of the
model geometry. Segmentation remains a challenging and subjective process that
does not guarantee repeatability. Improvements in this area can only really come
from research into improved imaging and more robust segmentation algorithms;
however it is likely that manual input will always be required in this process.
For example, intensity and label fusion algorithms have demonstrated very high
reproducibility and accuracy indistinguishable from that of human experts [2, 3,
25].
Because manual surgery cannot achieve accuracy better than 1 mm, even slight
deviations by one or two voxels can still produce a better method of localisation
than is currently used in practice.
Finally, another area of interest in the surgical treatment of epilepsy is in the
modelling of depth electrodes. These are very long and slender needles that pierce
deep into the parenchyma and are extremely difficult to control and locate during