Biomechanical Modeling of the Brain

1
Biomechanical Modeling of the Brain

• Ron Kikinis

MRI Divison, Department of Radiology, Image
Guided Therapy Program Neuroimaging Analysis
Center, a NCRR national resource center
2
Acknowledgements
Black, P. McL Grimson, E. Guttmann, C. Halle,
M.Hynynen, K. Jolesz, F.A. Nabavi, A.
Panych, L. Shenton, M. Tempany, C. Tannenbaum,
A. Warfield, S. Wells, W. Westin, C.F.
NCRR, NCI, NLM, NSF, DOD et al.
Http//www.spl.harvard.edu
3
Clinical Application

• Improve Surgery

4
Conventional Surgery See the surface
Provided by Nakajima, Atsumi et al.
5
Image Guided Surgery See under the surface
Provided by Leventon et al.
6
Imaging of Intraop. Changes
Craniotomy
Initial
7
Minor Technical Problems

• Gather ALL diagnostic information
• Integrate, process and enrich
• Update during procedure

8
Template moderated statistical Classification
Intensity and Anatomical Feature Space
Provided by Kaus
9
Hierarchical Control Strategy
Skin Brain
Ventricles Tumor
Provided by Kaus et al.
10
Improving fMRI Activation Detection Mutual
Information (MI)

• Test the hypothesis that the experimental
  protocol signal and the voxel intensities are (or
  are not) statistically independent using MI.

MI is the optimal test statistic (using
Neyman-Pearson criteria) !
GLM
MI
Provided by Wells
11
Tensor Imaging
Water diffusion in myelinated fibers
Provided by Sierra et al.
12
Tensor Display
Close-line
Close-plane
Close-sphere
3 classes
Provided by Sierra et al.
13
Tensor Visualization
Provided by Sierra et al.
14
Int. Capsule and Corpus
Provided by Westin, Meier
15
Dislocation of WM tracts
1.5T
0.5T
Oligodendroglioma
Provided by Meier, Mamata, Westin et al.
16
Distruction of WM tracts
red optic radiation green genuculocalcarine
tract light green auditory radiation
green corpus callosum fiber disruption by GBM
Provided by Westin, Mamata, et al.
17
Minor Technical Problems

• Gather ALL diagnostic information
• Integrate, process and enrich
• Update during procedure

18
Data Fusion
Provided by Nabavi, Wells et al.
19
fMRI

Provided by Wible,Nabavi, Gering, Odonnell
20
Non-Imaging Data TMS
Provided by Leventon et al.
21
Atlas

• Mapping atlas knowledge into patient data sets
• Guessing the location of the cortico-spinal tract

Provided by Kaus, Nabavi, Warfield
22
Colon Segmentation and Flattening
Provided by Haker, Tannenbaum
23
Polyps-Highlighted Rendering
Provided by Haker, Tannenbaum
24
White Matter Curvature Map
Provided by Haker, Tannenbaum
25
White Matter Flattening
Provided by Haker, Tannenbaum
26
Minor Technical Problems

• Gather ALL diagnostic information
• Integrate, process and enrich
• Update during procedure

27
Intraoperative MRI
Open configuration MR scanner
Operating Room
Provided by Pergolizzi
28
Total procedures 656 Craniotomy
447 Biopsy 143 Others
66 (Laser ablation 13,
transsphenoidal resection 22, etc)E.g. Resection
of Low Grade Glioma
Provided by Black, Nabavi, Talos
29
Surgery or Science?

• Clinical goal Map all the preoperative data into
  the patients brain as it changes during surgery
• Basic science goal In-vivo measurement of
  mechanical properties under mechanical and
  pharmacological stress
• Research is only doable if patient care is not
  affected in a negative way

30
Brain Shift During Surgery
Provided by Warfield, Ferrant et al.
Before surgery (1)
Dura opened (2)
Small resection (3)
Tumor resected (4)
Dura closed (5)
31
Segmentation of the Brain
Provided by Warfield, Ferrant et al.
Before surgery (1)
Dura opened (2)
Small resection (3)
Tumor resected (4)
Dura closed (5)
32
How to map pre-op. information into intra-op.
data sets Brain Shift happens
Courtesy A. Nabavi, S. Warfield
33
More Minor Technical Problems

• Clinical goals Real time requirements

34
Method
Provided by Warfield, Ferrant et al.

• Pathology is specific to the patient a patient
  specific model is required.

35
Connectivity of OR and Compute Servers
Cluster
IMRI Scanner
PR 5200 GE switch
Sunswitch GE
Visualization Workstation
Gigabit Ethernet
Provided by Warfield, Ferrant et al.
36
Compute Platforms
Wildfire Cluster Scalable shared memory
architecture built on Sun E5k and E6k chassis, 36
250MHz UltraSPARC-II CPUs and 9GB RAM.
4 Sun Ultra 80s each with 4 450MHz UltraSPARC-II
CPUs, 2-4GB of RAM and Fast Ethernet connectivity.
Provided by Warfield, Ferrant et al.
37
Data Acquisition
Provided by Warfield, Ferrant et al.
Slice 0
Slice 30
Time T2 Intra-operative image (just after
opening of the dura)
Time T3 Intra-op. image (brain has deformed)
Time T1 Pre-operative image
How to match 3D image at time T1 onto 3D image at
time T3, during a neurosurgery operation ?
38
Biomechanical Simulation
Provided by Warfield, Ferrant et al.

• Active Surface provides known displacements at
  some mesh vertices.
• FEM based modelig of volumetric deformations
  Sparse linear system of equations.
• Parallel implementation using PETSc, MPI.

39
Concept Sphere to Ellipsoid
Interpolate deformation field back onto the image
grid
Deform volume using surface deformation as a
constraint for a volumetric FE elastic model
Provided by Ferrant et al.
40
Overview
Provided by Warfield, Ferrant et al.
41
Timeline of Analysis

• Typical times with current data and
  implementations on current hardware.

42
Step 1 Segmentation
Provided by Warfield, Ferrant et al.

• Goal Identify critical structures
• Pre-operative segmentation as good as possible
• Intra-operative segmentation as fast as possible

43
Brain Segmentation Before and After OP
1
1
2
2
1
4
i
f
i
f
3
3
4
4
2
5
i
f
i
f
5
5
6
6
3
6
i
f
i
f
Provided by Warfield et al.
44
Step 2 FEM Construction

• Designed for generating multi-resolution FE
  meshes from medical image data
• First generate a uniform tetrahedralization
• Tetrahedral clipping and remeshing
• Adaptive refinement

Provided by Ferrant et al.
45
Step 3 Update of Topology
Provided by Warfield, Ferrant et al.
46
Step 3 in 3D
Provided by Warfield, Ferrant et al.

• Elements in the resected areas are removed from
  the mesh

47
Step 4 Volume Deformation (1)
Provided by Warfield, Ferrant et al.
48
Volume Deformation (2)
Provided by Warfield, Ferrant et al.
49
Visualization of volumetric deformation (3)
3 TO 4
4 TO 5
Provided by Warfield, Ferrant et al.
50
Visualization of Stress-Tensors (1)

• Stress-tensors color-coded based on the largest
  eigenvalue of the tensor.

2-3
1-2
Provided by Warfield, Ferrant et al.
51
Visualization of Stress-Tensors (2)
3-4
4-5
Provided by Warfield, Ferrant et al.
52
Parallelization
Twenty 250MHz CPU SMP 77511 equations
53
Volumetric FE Deformation
Volumetric Deformation
Provided by Ferrant et al.
54
Updating preoperative imaging

• The volumetric deformation field can be
  interpolated back onto the gray scale images

55
Step 5 Landmark Validation

• Landmarks were manually placed on slices and
  deformed with the algorithm
• Mean error was about 2.5 mm

56
The Role of HPC in IGT

• Computationally expensive algorithms become
  practical (from hours to minutes)
• Achieving speed for intraprocedural processing
  (from minutes to seconds)
• Accommodate huge memory requirements during
  exploratory phase of algorithm development

57
Conclusions

• Clinically motivated work Provide better mapping
  of preoperative and intraoperative information
• Could provide in-vivo mechanical parameters of
  the brain

58
URL
Http//www.spl.harvard.edu Http//www.slicer.org