Título de la tesis


REGISTRATION-BASED MOTION AND DEFORMATION ANALYSIS OF CARDIOVASCULAR IMAGE SEQUENCES

Doctorando


Estanislao Oubel

Director


Alejandro Frangi

Descripción


The estimation of morphological changes of biological tissues over time is an
ubiquitous problem in medical imaging. Image registration methods are suit-
able to solve this kind of problems since they allow to establish dense point cor-
respondences between images, which in turn can be used to quantify deforma-
tion. Given a discrete image sequence I (x, t) = I (x, 0), I (x, 1) · · · , I (x, N − 1), in
the context of this thesis, sequence registration means to find a transformation
T(x, t) : (x, 0) → (x ′ , t) that puts into correspondence points belonging to the
same sequence. This term must not be confused with its meaning in the context
of intersubject sequence registration [1], where the objective is to find a transforma-
tion T12 (x1 , t1 ) : (x1 , t1 ) → (x2 , t2 ) that establishes correspondences between points
of different sequences I1 (x1 , t1 ) and I2 (x2 , t2 ). In this thesis we have focused on
two challenging applications such as wall motion estimation in cerebral aneurysms
from Digital Subtraction Angiography (DSA), and deformation estimation of the
heart from Tagged Magnetic Resonance Imaging (T-MRI) [2, 3]. Figures 1 and 2
show examples of the images used in this thesis.

The quantification of pulsation in cerebral aneurysms is important for studying
the connection between haemodynamics and rupture. One of the hypothesis that
explain the rupture of aneurysms is the stress concentration on the vessel wall. This
can be quantified by computing wall shear stress values from computational fluid
dynamics (CFD) simulations. Wall motion information can be used in this context
to impose boundary conditions in CFD simulations performed on non-rigid models
as described in Chapter 1.

Wall motion estimation in cerebral aneurysms is also important since the po-
tential connection between pulsation and risk of rupture, as reported in [4–6]. The
underlying hypothesis is that rupture owes to a weakness of the vessel wall, which
should change the pulsation amplitude of the aneurysm. To study the interrela-
tionships between rupture and pulsation it is necessary to quantify this pulsation,
which can be performed by measuring displacements of the vessel wall over the
cardiac cycle.

Most of wall motion values reported in the literature correspond to experiments
with phantoms [5, 7], simulated images [8], or experimental models [9]. Only a
few attempts of in-vivo quantification have been reported in human beings [4, 10].
In this thesis we have developed an automatic method to quantify wall motion in
cerebral aneurysms from DSA sequences. This method was then applied to study
the relationship between rupture status and pulsation (Chapter 2).

Methods for cardiac deformation estimation are important for studying the be-
havior of the heart under normal, pathological, and simulated conditions. Among
other applications, these methods are useful for studying the mechanical effects of
cardiac diseases [11] and for the development of electro-mechanical models. Pair-
wise registration maximizing the Mutual Information (MI) between component vol-
umes of T-MRI sequences has been successfully applied for recovering displacement
fields of the heart [12, 13]. This motivated us to study extensions of methods based
on information theory both for inclusion of spatial and temporal information.

Normally, the similarity metrics used in image registration methods are based
on pixel intensity information. Such metrics ignore spatial information in the pixel
neighborhood that could provide valuable information for guiding the registration
process. We have explored the use of metrics based on wavelets transforms for non
rigid registration of 2D T-MRI sequences (Chapter 3).

Joint sequence registration refers to the simultaneous alignment of all the frames
in the sequence, opposite to the pairwise approach in which all frames are succes-
sively registered (one by one) to the first (or previous) one of the sequence. We have
developed a novel method for recovering cardiac displacement fields by joint align-
ment of T-MRI sequences (Chapter 4), and applied it for studying regional strain
differences between patients with myocardial infarction and a group of healthy
subjects (Chapter 5).

The results of this research have been presented in several conferences, and
submitted to scientific journals for their publication. A list of these publications can
be found after the Chapter 5. In the same list, we have added other publications
resulting from colaborations with external groups, to which we have contributed
with the experience obtained during these years in the field of image registration.


Estado


Defendida - 2009

Calificación


Sobresaliente Cum Laude



Validación de código - XHTML 1.1
Validación de CSS
Validación de Accesibilidad Web