simulation of soft tissue deformation for medical...
TRANSCRIPT
Simulation of soft tissue
deformation for medical
applications
Hervé Delingette
[email protected] INRIA SOPHIA ANTIPOLIS March 20th , 2014
Context
in v
ivo Medical
Images
and
Bio-signals
The Digital Patient
CT ScanMRI
ECG
Medical Records
- 2
Context
Personalisation
in v
ivo Medical
Images
and
Bio-signals
Geometry
Physics
Physiology
Cognition
Computational
Models
&
Tools
in silico
Sta
tist
ics
The Digital Patient
- 3
Soft Tissue Deformation in Medicine
• Water content of human Body is 50-75%
- 4
Soft Tissue Deformation in Medicine
• Water content of human Body is 50-75%
• Cause of Deformation :
– Muscle :
- 5
MR Imaging ofKnee joint@3DAH
Soft Tissue Deformation in Medicine
• Water content of human Body is 50-75%
• Cause of Deformation :
– Muscle :
– Heart :
- 6
Cardiac MR Imaging
Soft Tissue Deformation in Medicine
• Water content of human Body is 50-75%
• Cause of Deformation :
– Muscle :
– Heart :
– Respiration :
- 7
Augmented Reality IHU Strasbourg
Soft Tissue Deformation in Medicine
• Water content of human Body is 50-75%
• Cause of Deformation :
– Muscle :
– Heart :
– Respiration :
– Pathologies
- 8
Simulation of Glioblastoma Growth
Soft Tissue Deformation in Medicine
• Water content of human Body is 50-75%
• Cause of Deformation :
– Muscle :
– Heart :
– Respiration :
– Pathologies
– Surgical tools
- 9Liver Surgery Simulation
Application of soft tissue deformation
• Image Registration :
- 10
Cardiac Motion Tracking based on Biomechanical model
Application of soft tissue deformation
• Image Registration :
• Image Segmentation
• Therapy Training
• Therapy Planning
- 11
Holy Grail of Soft Tissue Deformation
• The 4Ps:
– Precise
– Performant
– Personalized
– Predictive
- 12
Accurate Modeling
• Use Physically (=biomechanical) based models
– Model verification
– Simplest Suitable Model
- 13
Accurate Modeling
• Use Physically (=biomechanical) based models
• Image Based Validation :
– Huge amount of data acquired every day
– Only visible motion
- 14
Cine-MRI : visible motion tagged-MRI : “true” motion
Holy Grail of Soft Tissue Deformation
• The 4Ps:
– Precise
– Performant
– Personalized
– Predictive
- 15
Computational Speed
• Why is it important ?
– Models Compatible with clinical practice
• Training : Real Time !
• Diagnosis : Few minutes
• Planning : Few hours
– Important for
• Model Personalization
• Uncertainty Estimation
- 16
How to speed up computation
• Possible approaches (can be combined):
– Fast assembly of Force vectors / Stiffness matrices• Geometric View of Linear Finite Elements
- 17
TriangleTetrahedra
� � ���� � ���
�� � � �� ⋅ � ���
Shape Function Shape Vector
Displacement NodalDisplacement
How to speed up computation
• Possible approaches (can be combined):
– Fast assembly of Force vectors / Stiffness matrices• Geometric View of Linear Finite Elements• Use mesh topology to store matrices• Link between discrete & continuum mechanics
- 18
Established equivalence between :
• Linear Strain / Stress Elasticity • Spring mass systems on Triangles / Tetrahedra with tensile / angular
and volumetric springs
H. Delingette. Triangular Springs for Modeling Nonlinear Membranes .IEEE Transactions on Visualization and Computer Graphics, 14(2), March/April 2008
Compressible St Venant Kirchhoff
• Efficient stiffness matrix computation
������ � � ������ �� � �� ��� ���
- 19
AffineTransformation
Linear ElasticStiffness Matrix
Cope with inverted elements
Cope with Large Deformation
How to speed up computation
• Possible approaches (can be combined):
– Fast assembly of Force vectors / Stiffness matrices• Geometric View of Finite Elements• MJED
- 21
S. Marchesseau, T. Heimann, S. Chatelin, R. Willinger, and Hervé Delingette.Fast porous visco-hyperelastic soft tissue model for s urgery simulation: application to liver surgery .Progress in Biophysics and Molecular Biology, 103(2-3):185-196, 2010
Fast Assembly of Stiffness Matrices
• For Hyper-elastic materials– Existence of a strain energy W
• Multiplicative Jacobian Energy Decomposition– Decompose W according to :
• J=|F| Jacobian of deformation gradient � � ��• I1, I2, I3, invariants of Deformation tensor C = (Right Cauchy Green)
– Simplify term ��� and
�!�� !
– Allow for some precomputation
– Extended for Visco-elasticity, anisotropy
- 22
MJED Computational Speed -Up
- 25
On average 2.7 times faster !
Models for hyperelasticity
How to speed up computation
• Possible approaches (can be combined):
– Fast assembly of Force vectors / Stiffness matrices• Geometric View of Finite Elements• MJED
– Reduced Models (POD)
- 26
How to speed up computation
• Possible approaches (can be combined):
– Fast assembly of Force vectors / Stiffness matrices• Geometric View of Finite Elements• MJED
– Reduced Models (POD)
– Parallelization (MT, GPU)
– Dedicated Software
- 27
Excalibur SOFA
SOFA : www.sofa -framework.org
• Developed by several INRIA teams since 2004
• API for medical simulation :
– Focused on but not limited to real-time applications
– Modular : components structured inside a graph
– Support for GPU ( Cuda / Opencl)
– Well developed for Mechanical deformation (solid, fluid,
FEM. CG methods), Collision Detection, Visualization, Haptics
28
SOFA in Action
- 29
Deformable Augmented Reality@Shacra – IHU Strasbourg
Haptic Feedback@Shacra
Pre-stressed Cutting
With Shacra Team, Inria Lille
00 MOIS 2011EMETTEUR - NOM DE LA PRESENTATION - 32
Hugo Talbot
EndoVascular Simulator of Cardiac RadioFrequency Ablation
Holy Grail of Soft Tissue Deformation
• The 4Ps:
– Precise
– Performant
– Personalized
– Predictive
- 33
Parameters
Electromechanical Model
Equations
SimulatedObservations
Measured Observations
Patient Data
Data
processing
,...),,( 0 Kµσ
GlobalParameters
Calibration
LocalParameters
Local
Personalization
Model Personalization• Amounts to solve an inverse problem
- 34
Parameter Observability
• Not all parameters can be estimated from observations
35
dx
Cannot estimate spring stiffness kfrom dx!!
dx
Fk =
?k
Parameter Observability
• Can estimate combination of parameters from observation
36
dx
Only estimate spring stiffness k1+k2from dx and F!!
k1
k2
F
Parameter Observability
• Can estimate combination of parameters from observation
37
dx2
Can estimate the ratio of spring stiffness k1/k2from displacements !!
k1 k2k1k2
dx1
Biophysical Model Personalization
• Not just “Parameter Fitting” :
– Sensitivity analysis to extract most important params
– Parameters constrained by physics and physiology
• Avoid overfitting by adapting model complexity to that of the
measurements
38
solid mechanics
Clinical applications
Diagnosis
Therapy planning
blood flow
Cardiac data
Personalizationelectro-physiology
perfusion & metabolism
Physiological Modeling of the Heart
Cardiac modeling
anatomy
- 39
A Multiphysics Problem
- 40
Electrophysiology Modeling
SimulateAction PotentialPropagation
Mechanical Modeling
Action PotentialControls
Active Stress
Orthotropic PassiveMaterial
Flow Modeling
Arterial Pressure
Valve Opening / Closure
Strong Anisotropy due to the cardiac
fibers
Simulating the Cardiac Cycle
- 41
IsovolumetricContractionEjectionIsovolumetricRelaxationFilling
Stéphanie Marchesseau
Complex Muscle Modeling
42
Contractile Sarcomere
Energy dissipation in SarcomereDue to friction
Elasticity of the Z-line (titine)
Elasticity of the Collagen
Energy dissipation in the Collagen
[Bestel 2009,Chapelle 2012]
Longitudinal Motion:Apico-basal Shortening
Radial Motion:Wall Thickening
Simulating the Healthy Heart
43S. Marchesseau, H. Delingette, M. Sermesant, M. Sorine, K. Rhode, S.G. Duckett, C.A. Rinaldi, R. Razavi, & N. Ayache. Preliminary Specificity Study of the Bestel-Clément-Sorine Electromechanical Model of the Heart using ParameterCalibration from Medical Images. Journal of the Mechanical Behavior of Biomedical Materials, 2012.
Simulating the Healthy Heart
44
Circumferential Motion:Twist / Torsion, Inverse Rotation between Base and Apex
Personalization from In vivo Clinical
Measurements
King’s College, division of Imaging SciencesThe Guy's, King's and St Thomas' School of Medicine
St Jude Ensite
K. Rhode A. RinaldiR. Rezavi
- 45
Parameter Observability
46
Cine-MRI
Estimate ratio of stiffnesses
and contractilities
Cine-MRI
+
LV Pressure
Estimate stiffnesses
or contractilities
Personalization of Local Contractility
Observations
=
LV AHA
Regional Volumes
LV
barycenter
Vreg
To optimize17 local contractility parameters after calibration of up to 7 global parameters
- 47
Measured vs. SimulatedRegional Volumes
Measurements
Personalized
Simulation
- 48
Mechanical Personalization
- 49
Marchesseau, S., Delingette, H., Sermesant, M., Cabrera-Lozoya, R., Tobon-Gomez, C., Moireau, P., Figueras, R., Lekadir, K., Hernandez, A., Garreau, M., Donal, E., Leclercq, C., Duckett, S., Rhode, K., Rinaldi, C., Frangi, A., Razavi, R., Chapelle, D., and Ayache, N. Personalization of a Cardiac Electromechanical Model usingReduced Order Unscented Kalman Filtering from Regional Volumes. Medical Image Analysis 2013
EuheartProject
Holy Grail of Soft Tissue Deformation
• The 4Ps:
– Precise
– Performant
– Personalized
– Predictive
- 50
51
Predictive Value?• Predict the effect of a Cardiac Resynchronization
Therapy (CRT)
Currently, up to 30% of implantations are not successful
53
Virtual Pacemaker
before afterLV endocardia
Coronary sinus
RV endocardia
dP/dt
measuredsimulated
measuredsimulated
dP/dt
Simulated CRT
resynchronization
Importance of Estimating Uncertainty
• Predicting the future is difficult !!
• Estimate source of uncertainty
– Image / Data Noise or distorsion
– Image Processing
– Model Errors (False hypothesis)
– Errors in parameters / BC / IC
– Discretization errors
- 54
Conclusion
• Need for soft tissue models to match clinical constraints in
terms of speed and accuracy.
• Must adapt model complexity to each given problem but
keeping a predictive value.
• Personalization leads to difficult inverse problems :– Parameters observability
– Data assimilation techniques
• Access to rich experimental data is key
- 55
AcknowledgmentsPost-doc / engineer : Erik Pernod, Federico SpadoniPhd Students : Hugo Talbot, Stéphanie Marchesseau,
Tommaso Mansi, Jatin Relan, Jean-Marc Peyrat, Florence Billet, Loic Le Folgoc, Adityo Prakosa
Asclepios INRIA : Maxime Sermesant, Nicholas Ayache, Reo INRIA : Miguel Fernandez, Jean-Frédéric Gerbeau, Macs INRIA : Dominique Chapelle, Philippe MoireauSisyphe INRIA : Michel SorineShacra INRIA : Stéphane Cotin, Christian DuriezKCL : N. Smith, K. Rhode, R. Razavi, Toronto HSC : M. Pop, G. Wright,Creatis : P. Croisille, P. Clarysse
Funding : EuHeart, MedYMA, Health-e-Child, INRIA
- 56
"In theory there is no difference between theory and practice.
In practice there is.“
Yogi Berra
- 57