868 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 23, NO. 7, JULY 2004
Estimating Topology Preserving and Smooth
Displacement Fields
Bilge Karaçalı, Member, IEEE and Christos Davatzikos, Member, IEEE
Abstract—We propose a method for enforcing topology preser-
vation and smoothness onto a given displacement field. We first
analyze the conditions for topology preservation on two- and
three-dimensional displacement fields over a discrete rectangular
grid. We then pose the problem of finding the closest topology
preserving displacement field in terms of its complete set of gra-
dients, which we later solve using a cyclic projections framework.
Adaptive smoothing of a displacement field is then formulated as
an extension of topology preservation, via constraints imposed on
the Jacobian of the displacement field. The simulation results in-
dicate that this technique is a fast and reliable method to estimate
a topology preserving displacement field from a noisy observation
that does not necessarily preserve topology. They also show
that the proposed smoothing method can render morphometric
analysis methods that are based on displacement field of shape
transformations more robust to noise without removing important
morphologic characteristics.
I. INTRODUCTION
I
N BIOMEDICAL image matching, a template is warped to
a given patient scan in order to maximize point correspon-
dences. The matching is expressed in terms of a displacement
field , where and are the two-dimensional
(2-D) or three-dimensional (3-D) domains of the template and
the patient scan , respectively. Given a template and a patient
scan, a displacement field is determined usually by solving an
energy minimization problem, where a cost function of form
(1)
is minimized through some iterative scheme. The term
enforces data closeness, where
is a collection of descriptive features of image at
location . Frequently, is chosen to be the brightness
value of image at , [1], [5], [17]. More extensive
feature representations have also been used with improved
matching accuracy [19]. Regularity on the solution is enforced
through which penalizes deviations from some measure
of smoothness.
Another common approach to determine a displacement field
is to establish point correspondences between the template and
the patient scan at a subset of points in the domain for which
“reliable” point representations can be obtained [4], [9]. These
Manuscript received October 29, 2003; revised March 10, 2003. The Asso-
ciate Editor responsible for coordinating the review of this paper and recom-
mending its publication was J. Pluim. Asterisk indicates corresponding author.
*B. Karaçaı is with the Section of Biomedical Image Analysis, Department of
Radiology, University of Pennsylvania, Philadelphia, PA 19104 USA (e-mail:
bilge@rad.upenn.edu).
C. Davatzikos is with the Section of Biomedical Image Analysis, Department
of Radiology, University of Pennsylvania, Philadelphia, PA 19104 USA.
Digital Object Identifier 10.1109/TMI.2004.827963
values are then extrapolated over the rest of the domain to con-
struct a complete displacement field.
The main intuition behind topology preservation in a dis-
placement field is the desire to maintain connectivity between
neighboring morphological structures. The template and the
patient scan are presumably comprised of essentially the same
components, dilated, shrunk, or translated at regions. More-
over, topology preservation is necessary for the invertibility
of a transformation. The transformation that maps a template
to an individual, for instance, should be invertible in order
to allow mapping of the individual’s images to a stereotaxic
space, a process often called spatial normalization.
Whether a given deformation field preserves topology is usu-
ally monitored by the determinant of the Jacobian over the do-
main. In a 2-D setting, a deformation field can be expressed
in terms of its Cartesian components as , where
and denote its value over the and
axes. The determinant of the Jacobian of at ,
denoted by and referred briefly to as its Jacobian, is
computed using
(2)
Topology preservation in image matching can then be enforced
on the estimated displacement field minimizing (1) by ensuring
positivity on the Jacobian at every iteration through selec-
tions of hard [5], [17] or soft constraints [1] using the properties
of the selected model for . A practical alternative to such con-
tinuous monitoring of is to enforce displacement field
characteristics associated with a positive Jacobian such as in-
vertibility [13], [20], which, however, does not mathematically
guarantee preservation of topology.
Christensen et al. [5] propose a fluid model representation for
deformations to bring two morphologies into correspondence.
The advantage of the model is that the deformations expressed
as such inherently preserve topology by construction. The reg-
istration problem is then solved by numerical approximations
to fluid diffusion. On the other hand, there is no guarantee that
the morphological variations across subject anatomies is well
approximated by a fluid model deformation.
The approach of Ashburner et al. considers a triangular
partition of a 2-D domain and characterize the Jacobian deter-
minant on the triangles, but is biased toward the direction of
diagonalization. In this approach, a square patch is partitioned
into two triangles, but the deformation of diagonal vertices
are treated differently, whereas our results show that this char-
acterization is incomplete. Completing this approach would
require a second triangulation along the opposite diagonal.
Furthermore, the approach is limited to 2-D cases.
0278-0062/04$20.00 © 2004 IEEE