!Atlas construction and image analysis using statistical cardiac models

Center for Computational Imaging & Simulation Technologies in BiomedicineUniversitat Pompeu Fabra, Barcelona, Spain

Networking Center on Biomedical Research – Bioengineering, Biomaterials and Nanomedicinemathieu.decraene@upf.edu

www.cilab.upf.edu

M. De Craene, F.M. Sukno, C. Tobon-Gomez, C. Butakoff, R.M. Figueras i Ventura, C. Hoogendoorn, G. Piella, N. Duchateau, E. Muñoz-Moreno, R. Sebastián, O. Camara,

and A.F. Frangi

WHY DO WE NEED ATLASES OF THE HEART?

2

1

Why do we need atlases of the heart?

Looking at multiple levels

Global shape

Local shape

Motion / Deformation

3

1. Integrated image-based biomarkers

Why do we need atlases of the heart?

Probabilistic biomarkersEncode normalityP-value of abnormality

4

1. Integrated image-based biomarkers

d1 <?> d2 d1 d2

patient

atlas

Duchateau et al, STACOM (2010)

Why do we need atlases the heart?

5

2. Integrated multimodal information for patient-specific modeling

6

EVOLUTIONS AND

CHALLENGES2 IN HEART ATLAS CONSTRUCTION

…

Affine + nonrigid diffeomorphic registration

…

Apply inverse transforms

Average up to affine transform:

The atlas image

Reference

Segment Triangulate

Average non rigid transformation

Apply average non rigid transform

From single subject to population atlases

Ordas et al . Proc. SPIE Medical Imaging (2007)

From monomodal to multimodal atlases

8

MODEL TO IMAGE ADAPTATION/MATCHING

STATISTICAL INTENSITY MODEL

POINT DISTRIBUTION MODEL

Create automatically by image simulation

Tobon- Gomez et al. IEEE Trans on Medical Imaging, 27(11):1655-1667 (2008)

! Two parameterizations! a and b, subject and

cardiac phase! Each in their own space

with orthogonal basis

9

From spatial to spatiotemporal atlases

Hoogendoorn et al. Int J Comput Vis 85(3):237-252 (2009).

! Two parameterizations! a and b, subject and

cardiac phase! Each in their own space

with orthogonal basis

9

From spatial to spatiotemporal atlases

Hoogendoorn et al. Int J Comput Vis 85(3):237-252 (2009).

From single object to multi-objects atlases

! Multiple anatomical levels and topologies ! 4 chambers! Tissue properties! Muscle & Purkinje fibers! Coronaries

10

11

Anderson et al. Clinical Anatomy 22:64-76(2009)

From scalar objects to vectors & tensors! DTI-based fiber orientation

Muñoz-Moreno,& Frangi ICIP (2010, in press)

In vivo Human 3D Cardiac DTI Reconstruction 7

Fig. 5. Top: Joint histograms of the elevation (or helix) angle and the normalized

transmural distance from endo to epi. (1a) in-vivo interpolated results using Cartesian

coordinates, (1b) in-vivo interpolated results using PSS coordinates, and (1c) as a

reference, the fully sampled LV statistical atlas. The correlation is visible using PSS

coordinates. Bottom: Interpolated DTI slice color coded by eigenvector direction,

using Cartesian Coordinates (2a) and PSS (2b). (2c) is a streamline fibre tractography

result from the PSS interpolated tensor field.

5 Conclusions

In this paper, we demonstrated that shape adapted curvilinear coordinates –Prolate Spheroidal – are appropriate for the tensor reconstruction over the leftventricle wall volume. We set up a mathematical framework for the kernel re-gression of tensor data in those coordinates, using anisotropic kernel regressionwith an optimised bandwidth matrix. We have shown that the resulting inter-polated tensors better fit the physionomy of the heart. As the left ventricle hasa very characteristic ellipsoidal shape, its fibre architecture (and thus the un-derlying tensor field) has an important spatial coherence in PSS coordinates,whereas it is less sensible in Cartesian coordinates. We have shown that usingthe PSS anisotropic spatial coherence of a statistical cardiac DTI atlas as aprior information for in vivo tensor interpolation and regularization helps us toreconstruct full tensor information. We applied our method to reconstruct thefibre architecture of the left ventricle of a healthy volunteer, and, to the best ofour knowledge, it is the first time that the in vivo human 3D structure of theheart has been reconstructed. The in vivo results show a good correlation withliterature values of ex vivo human studies. We were able to reproduce the typicalpattern of transmural variation of the helix angle. The presented approach opensup possibilities in terms of analysis of the fibre architecture in patients.

Toussaint et al. Miccai (2010, in press)

CONCLUSIONS &

PERSPECTIVES12

3

Shape is not enough

13

14

Source: http://jcmr-online.com/imedia/1712877943433156/supp1.mpg

Motion is not enough

Erikson et al. JCMR, 12(9), (2010)

Perspectives

! On biomarkers! Towards complex indexes

! Integrate shape (local & global), electrical activation, motion/deformation, and flow

! Towards new probabilistic biomarkers! Distance to populations/manifolds

! On data integration! Multiple-layers visualization of heart

function! Multi-level patient specific models

15

Simulation

FEM Model

Functional Model

Patient-specific simulation and virtual populationsGeometrical

Model

MembraneIntracellular

Extramyocardial

Extracellular

FEM

Mod

elEl

ectr

ophy

siol

ogy

Electrical Multiscale Modeling

Romero et al. ABME, (2010)

Hoogendoorn et al. STACOM, (2010)

Thanks

17