626 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 25, NO. 5, MAY 2006
Deformation-Based Mapping of Volume Change
From Serial Brain MRI in the Presence of Local
Tissue Contrast Change
Colin Studholme* , Member, IEEE, Corina Drapaca, Bistra Iordanova, and Valerie Cardenas, Member, IEEE
Abstract—This paper is motivated by the analysis of serial
structural magnetic resonance imaging (MRI) data of the brain to
map patterns of local tissue volume loss or gain over time, using
registration-based deformation tensor morphometry. Specifically,
we address the important confound of local tissue contrast changes
which can be induced by neurodegenerative or neurodevelopmental
processes. These not only modify apparent tissue volume, but also
modify tissue integrity and its resulting MRI contrast parameters.
In order to address this confound we derive an approach to the
voxel-wise optimization of regional mutual information (RMI) and
use this to drive a viscous fluid deformation model between images
in a symmetric registration process. A quantitative evaluation
of the method when compared to earlier approaches is included
using both synthetic data and clinical imaging data. Results show a
significant reduction in errors when tissue contrast changes locally
between acquisitions. Finally, examples of applying the technique
to map different patterns of atrophy rate in different neurodegen-
erative conditions is included.
Index Terms—Deformation morphometry, magnetic reso-
nance imaging, mutual information, tissue contrast, viscous fluid
registration.
I. INTRODUCTION
S
ERIAL structural magnetic resonance imaging (MRI) of the
brain [1], [2] is an increasingly useful tool in the study of
neurodegenerative conditions [3]–[10]. Multiple high resolution
structural images of a subject’s brain are acquired over time, al-
lowing the progression of tissue volume changes to be tracked
in fine detail at all anatomical locations. The imaging technique
has been used to detect and characterize disease and has promise
in tracking the effects of treatments over time, and thus pro-
viding a surrogate marker for cognitive decline, that may be
more sensitive than conventional repeated neuropsychological
testing. As a result, the analysis of serial structural MRI data has
Manuscript received October 12, 2005; revised January 25, 2006. This work
was supported by the Whitaker Foundation under Grant RG-01-0115 and in part
by the National Institute of Mental Health under Grant MH65392-01. Asterisk
indicates corresponding author.
*C. Studholme is with the Department of Radiology, University of California
San Francisco, Northern California Institute for Research and Education, Vet-
erans Affairs Medical Center, 4150 Clement Street, San Francisco, CA 94121
USA (e-mail: colin.studholme@ieee.org).
C. Drapaca is with the Magnetic Resonance Research Lab, Mayo Clinic,
Rochester, MN 55905-0001 USA (e-mail: Drapaca.Corina@mayo.edu).
B. Iordanova is with the Department of Biological Sciences, Carnegie Mellon
University, Pittsburgh, PA 15213 USA (e-mail: biordano@andrew.cmu.edu).
V. Cardenas is with the Department of Radiology, University of California
San Francisco, Northern California Institute for Research and Education, Vet-
erans Affairs Medical Center, San Francisco, CA 94121 USA (e-mail: valerie.
cardenas-nicolson@ucsf.edu).
Digital Object Identifier 10.1109/TMI.2006.872745
been an active area of image analysis research for many years,
with methods ranging from global tissue volume estimation, to
finer scale mapping of local change.
Traditional computer vision approaches of extracting
anatomical boundaries or other simplified object representa-
tions from the two scans can be applied. However, because of
the complexity of the brain anatomy and the problems posed
by limited spatial resolution and contrast, an approach relying
on a specific geometric feature, such as a single boundary, is
inherently limited in its ability to explore the atrophic process.
A more direct approach is to use the first scan as a reference
template, containing all subtle structures in the individual, and
look for residual geometric differences between that image
and later time points. Initial methods taking this approach
were based on the analysis of intensity differences between
the images, after global rigid alignment of the scans to remove
variations in patient positioning [1], [3], [11], [12]. These
approaches then estimated global volume changes from the
residual intensity differences after global rigid alignment, either
through detection of significant intensity difference [12] or
through integration over regions of interest [11].
One of the main issues with such approaches is the confound
caused by local displacements of tissue over time, which may
arise when loss of tissue integrity leads to a collapse of gyral
structures. This can induce volume preserving deformations in
the neighboring tissue structure, that create significant residual
differences between globally rigidly aligned images, leading to
artifactual estimates of tissue loss. The most direct form of this
in aging is the loss of underlying white matter (WM) leading
to the local collapse and displacement of cortical gray matter
(GM). The approach to addressing this problem has been to em-
ploy nonrigid registration to simultaneously resolve both local
tissue size and displacement differences between the two scans
[13]–[15]. Nonrigid registration essentially aims to capture all
geometric differences between scans in terms of a spatial trans-
formation. The components of tissue displacement can then be
decomposed from true volume loss, creating a so called defor-
mation-based morphometric analysis [16]. A dense field regis-
tration-based technique also offers a way of using any subtle
contrasts and textures that may be common to both images.
Methods for serial MRI analysis published so far have made
use of an image similarity term derived directly from intensity
differences, to drive the image alignment process.
A. Motivation for This Work
An important practical confound in serial MRI imaging of
both neurodegeneration and development, is the presence of lo-
0278-0062/$20.00 © 2006 IEEE