SIAM News Blog

Image-Guided Neurosurgery: Computation of Brain Deformations

By Karthika Swamy Cohen

In a minisymposium on computational methodologies for mathematical models of human brain electrophysiology, Adam Wittek of University of Western Australia gave a talk titled “Computation of Brain Deformations for Image-Guided Neurosurgery: Progress and Uncertainty.” 

Wittek presents his work - from image to computational mechanics model. Staff photo.
In the talk, he demonstrated that sufficiently accurate predictions of brain deformations during surgery exposed brain surface during surgery. To all intents and purposes, the computed deformations within the brain are insensitive to uncertainty in patient-specific properties of the brain tissue data.

In traditional methods, going from image to computational biomechanical models is a fairly time-consuming process since it involves segmentation, generation of mesh and measurement.

Meshless methods, on the other hand, allow more speed. Here, patient-specific biomechanical models are generated semi-automatically directly from medical images. The concept utilizes Galerkin-type meshless discretisation of equations of solid mechanics in combination with statistical tissue classification.

3D patient-specific meshless computational model. Photo credit: Adam Wittek presentation AN16/LS16
Here, nodes are inserted in the domain and stress parameters are assigned at integration points of the computational grid. The process uses a lot less time when compared with segmentation and fine element mesh generation. 

Since the image is amorphous, it is divided or classified by tissue classification rather than segmentation, which divides the brain into parenchyma tissue, ventricles and tumor tissue. 

How well does this meshless model work? Wittek’s group evaluated accuracy in three cases. To do this, the group acquired MRI images from surgery and validated and qualitatively evaluated them by comparing sets of edges in the predicted image with the intraoperative images.

Distinct overlap was seen between the intraoperative and predicted images. For quantitative evaluation, the edge-based hausdorff distance was used, which gives a measure of how many edges are within a given percentile. 60% of edges were seen to register correctly. Thus the group’s biomechanical model was seen to give results very close to intraoperative image, which was at least as good as traditional methods.

Karthika Swamy Cohen is the managing editor of SIAM News.

blog comments powered by Disqus