Overview

1
Overview
The European project SimBio developing a
bio-numerics environment
G. Lonsdale CC Research Laboratories NEC Europe
Ltd.

• Overview
• Introduction to the project SimBio environment
• Environment components applications recent
  results and developments
• Cooperation with external groups
• Concluding remarks

2
Impact Innovation
IST99 - Action Line IV.4.1 Project Duration 36
Months Commencement date 1st April, 2000
Promotion of HPC Numerical Simulation within
the medical sciences.
Expected Impact
Actual scan data from Patients used for
modelling, analysis predictive, physically
accurate simulation.
Innovation
3
Global overview
Discrete Representation
Analysis, Prediction Design
SimBio-external Data production
Using HPC Computational Technology
MRI, EEG, MEG,..
4
The Consortium
The SimBio Consortium
Collaboration Partner
caesar , Bonn D
5
An overview of the SimBio environment
SimBio Environment Components
6
Component interaction
Interoperable Components on distributed systems
Current prototype set-up
Req-DB
XML
Computation Server
User Site
P1
P2
P3
P4
PC-Cluster
Result Client
Result
Result
7
Visualisation Tool for SimBio
VM The Visualisation tool for the SimBio
components
Interaction for model definition (here vectors
for external forces)
Tensor representation tensors in detail ?
8
Section discrete representation
Discrete Representation
Image Segmentation registration
Mesh generation
Material Modelling (database)
9
segmentation overview
2 segmentation approaches followed
10
image processing toolbox
Image processing Toolbox

• Vista file format
• Implemented for all current and developing tools
• Dicom to Vista format implemented
• Registration tools
• Manual image alignment
• Multi-modality rigid registration
• Non-linear registration modules work in progress
• Segmentation tools
• Intensity based segmentation
• Model based segmentation work in progress

11
VGrid Mesh Generator
VGrid Mesh Generator Finite Element Mesh
Generator based on the input of segmented (images
as) voxel data.

• Focus of most recent activities
• interfacing to Numerical Solution System
• surface smoothing using node shifting, marching
  tetrahedra surface extraction

Ventricular system
12
overview template approach
Meshed Segmented Template Approach

• Create detailed meshed, segmented templates of
  target structures

Articular Cartilage

• Produce an image of that template use that
  image to register to the patients structure
• (thus, creating segmented, meshed model)

blue original segmented knee green new knee
shape after transformation
13
overview template approach
Template Approach

• Model template approach
• build model of structure to be segmented
• assign image intensity values to image components
  
• construct image from weighted components
• register image to actual subject image
• Post registration
• derived mapping function can be used to map model
  segments onto patient image.
• Thus mesh can be mapped directly

14
registering between MRI knee images
Registering MR Knee Images
Volunteer AW red Volunteer AM green
Well-registered yellow
Before registration
After registration
15
transformation mesh template
Transformed segments
Creating a hand-segmented high-quality mesh
template
Transforming template segments to a patient
specific knee AM blue original segmented
knee AW green new knee shape after
transformation
Articular Cartilage
Posterior view
16
material database intro
Material Database
Main techniques to be exploited within SimBio are
Magnetic Resonance Strain Imaging
(a combination of magnetic resonance
ultrasound)
Diffusion Tensor Imaging
( a pure magnetic resonance technique)
Literature Values
Example ongoing work on the determination of
anisotropic mechanical properties of brain white
matter
17
experimental set-up
Experimental setup
18
brain white matter sample results
Brain samples

• Cylindrical samples taken from nine pig brains,
  using a biopsy punch parallel and perpendicular
  to fibers

Typical sample shown from two different angles
(90)
Material Properties as input for subsequent
modelling

• Anisotropy
• Small deformation ? no anisotropy
• Large deformation ? pronounced anisotropy

19
Section numerical solution system overview
Numerical Solution System
Building on Existing Codes/Software
20
sw-hw overview
SimBio sw-hw overview
SimBio Material DB
SimBio GUI
SimBio Environment Software (CORBA) - for
distributed environment
Numerical Solution System
Inverse Tools
PAM-SAFE?
Head-FEM
Neuro- FEM
Future Additions
DRAMA (mesh partitioning)
Parallel Linear Solver Libs
Communication Lib. (MPI)
OS System Software
Parallel HW
Standard/Graphics HW e.g. workstation/PC
21
sw details
Numerical Solution System
22
Section Apps Tools Overview
Evaluation Validation Applications Related
Simulation/Analysis Tools

• Electromagnetic Source Localisation
• Inverse Problem Toolbox
• NeuroFEM
• Brain Mechanics Analysis from Time-series data
• Nonlinear Registration Material Modelling
• Head Mechanics for Maxillo-facial surgery
• HeadFEM
• Knee Analysis Prosthesis Design
• PAM-SAFE

23
Reminder source localisation
Reminder Electromagnetic Source Localisation
A widespread technique to investigate the
functional state of the brain is to record its
electromagnetic activity by means of
electroencephalography (EEG) and magnetoencephalog
raphy (MEG)
Major Goal within SimBio FE modelling Only the
Finite Element Method enables the modelling of
internal brain structures and their physical
properties, eg anisotropic conductivity!

• Purpose
• identify brain regions of disfunction (e.g.
  caused by tumor growth)
• diagnose epilepsy.

dipolar current sources in the brain
24
Inverse Toolbox for Source Localisation
Inverse Problem Toolbox for Source Localisation
The Inverse Toolbox offers Novel software
architecture for flexible combination of
methodsUse of inhomogeneous and anisotropic head
models (NeuroFEM)

• Progress
• General structure completed
• First set of methods implemented
• Selection of methods and I/O via three-shell user
  interface
• Strong coupling to NeuroFEM

25
Inverse toolbox sensitivity analysis
Inverse Problem Toolbox for Source Localisation
The Sensitivity Analysis allows The fast and
flexible estimation of sensitivity of inverse
results due to inaccuracies, errors or
simplifications of the model used for simulations.
Error of reconstructed source positions due to
underestimation of skull conductivity

• Progress
• A flexible, general framework for sensitivity
  analysis is defined and implemented
• First application examples

26
source localisation validation
Source Localisation Experimental Validation
Source localisation carried out using EEG MEG
measurements for 2 known cortical sources
Distance between cortical sources anatomical
2-3 mm source localisation 2.5 /- 0.5 mm
Computed Electric Field Source 1, Peak P1
Measured Electrical Signal Source 1, Peaks P1
P2
27
source localisation validation
Source Localisation Experimental Validation
Electrical Signal Source 1, Peaks P1 P2
Electric Field Source 1, Peak P1
Magnetic Field Source 1, Peak P1
Magnetic Signal Source 1
Distance between cortical sources anatomical
2-3 mm source localisation 2.5 /- 0.5 mm
Source 2
28
Inverse Methodology for Time-series analysis
Inverse Biomechanical models Brain Mechanics
Analysis from Time-series data
Objective develop inverse methods for the
identification of structural changes underlying
bio-mechanical forces Method register anatomical
time-series data via (non-) linear vector field
transformations. Incorporate both realistic
material parameters consistency with behaviour
of deformable elastic solids
Results of ongoing study

• Car-accident patient
• red blue colouring indicate inward-outward
  deformation
• Critical areas highlighted in green locations
  which can be related to local brain matter
  destruction in adjacent areas.

29
Inverse Methodology for Time-series analysis
Inverse Biomechanical models Brain Mechanics
Analysis from Time-series data
Analysis of pre- to post-operative deformation
in maxillo-facial surgery
30
maxillo-facial summary
Skull Mechanics Simulation with HeadFEM
The Halo Frame
CT scan slice with halo fixed with screws
Dr. Dr. Hierl Klinikum Leipzig
Static analysis results in agreement with
clinical findings peripheral outward
deformation, inward at screws
31
maxillo-facial 1
Results Skull Mechanics Simulation with HeadFEM
Basic medical background
Cutting Lines
Dr. Dr. Hierl Klinikum Leipzig
The Halo Frame
CT scan slice with halo fixed with screws
32
maxillo-facial 2
Results Skull Mechanics Simulation with HeadFEM
Simulation results (Phase 1) Skull
deformation Inward at screw positions. Outward
at peripheral concentric rings (arrows). Full
agreement with clinical findings!
33
Knee modelling Prosthesis design
Knee Analysis Prosthesis Design
Validation and evaluation of the SimBio
Environment with respect to the knee joint
Goal
Latest Full Knee Model
Study of Finite Element modelling possibilities
for the meniscus
Constrained end of meniscal sample
34
MR Rig
Clinical Validation
MR Sequences
Purpose of rig is to permit patients to perform
light exercise against known loads while knee is
MR imaged
35
MR Sequences
Additional T1 sequence to image cruciate ligaments
Static T2 Sequence using GP Flex coil
Dynamic T2 Sequence using GP Flex coil
36
Section External Collaborations
Collaborations with groups external to the
SimBio Consortium
37
caesar collaboration
Collaboration() with caesar - Center of Advanced
European Studies and Research, Bonn, Germany

• The caesar research group for Surgical Simulation
  and Navigation will investigate 3-dimensional
  visualisation techniques for the SimBio
  applications.
• Visualisation techniques will use virtual reality
  environments and tools.
• Collaboration will produce demonstrations of the
  visualisation possibilities multi-media or
  video publicity material.

() within the project, covered by an agreement
with NEC
38
MPI-U.Linz collaboration
MPI collaboration with U. Linz, Austria
Use of the Parallel Algebraic Multigrid solver
parPEBBLES within NeuroFEM
Almost optimal linear speedup. Solution time of a
few seconds!
Segmented scan data slice multimodal MR-imag.
Axial isopotential lines of a NeuroFEM Solution
39
MPI-U.Linz - study
Source Localisation with NeuroFEM
MPI Leipzig Uni. Linz, 2001 AMG
parPEBBLES
axial surface Isopotential
lines Solution
Segmented scan data slice multimodal MR-imag.
40
mpi-linz - results
Parallel Performance NeuroFEMAMG
MPI Leipzig Uni. Linz, 2001 AMG
parPEBBLES
Almost optimal linear speedup. Solution time of a
few seconds!
41
BloodSim- I
Use of the SimBio Image Processing Tools within
the ESPRIT project BloodSim
Use of SimBio non-linear registration tool to
permit a fully-3D approach to coronary vessel
segmentation
Segmentation to F.E. meshes
Post non-linear registration of idealised cylinder
Fluid and solid meshes of the reconstructed pre-in
tervention geometry
42
BloodSim- I
Use of the SimBio Image Processing Tools within
the ESPRIT project BloodSim

• Use of SimBio non-linear registration tool to
  permit a fully-3D approach to coronary vessel
  segmentation
• BloodSim Background
• Important to reconstruct coronary blood vessels
  to identify patients requiring coronary stents to
  treat occlusions
• Use intravascular ultrasound (IVUS)
• Prior studies have used 2-pass 2-D approach to
  reconstruction

43
BloodSim- I
Use of the SimBio Image Processing Tools within
the ESPRIT project BloodSim
Post non-linear registration of idealised cylinder
Segmentation to F.E. meshes
Fluid and solid meshes of the reconstructed pre-in
tervention geometry
Results of 3d non-linear registration
44
VGrid for COPHIT data
VGrid other medical data
Upper airways not including oral cavity
Lower airways down to 8th generation
Data provided by COPHIT
1.34 million volume elements (TETs) Time 4mins
on a single processor Origin 2000 or 800MHz P3
requiring 1Gb Ram
45
Concluding Remarks
Concluding Remarks

• The SimBio environment will combine medical
  imaging FEM techniques with up-to-date HPC
  technology
• The component-based distributed interaction will
  enable interaction (networking) of
  technological expertise (clinical ó engineering ó
  computing )
• The evaluator applications will improve options
  in/for
• diagnosis and analysis of neurological disorders
• design of prostheses
• neurosurgery

For further information