IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, PREPRINT 1
Multi-Scale Modeling for Image Analysis of
Brain Tumor Studies
Stefan Bauer, Student Member IEEE, Christian May, Dimitra Dionysiou, Georgios Stamatakos, Member IEEE,
Philippe Bu¨chler and Mauricio Reyes, Member IEEE
Abstract—Image-based modeling of tumor growth combines
methods from cancer simulation and medical imaging. In this
context, we present a novel approach to adapt a healthy brain
atlas to MR images of tumor patients. In order to establish
correspondence between a healthy atlas and a pathologic patient
image, tumor growth modeling in combination with registration
algorithms is employed. In a first step, the tumor is grown in the
atlas based on a new multi-scale, multi-physics model including
growth simulation from the cellular level up to the biomechanical
level, accounting for cell proliferation and tissue deformations.
Large-scale deformations are handled with an Eulerian approach
for finite element computations, which can operate directly on the
image voxel mesh. Subsequently, dense correspondence between
the modified atlas and patient image is established using non-
rigid registration. The method offers opportunities in atlas-
based segmentation of tumor-bearing brain images as well as
for improved patient-specific simulation and prognosis of tumor
progression.
Index Terms—Brain Tumor, Glioma, Image Analysis, Tumor
Growth Modeling, Tumor Biomechanics
I. INTRODUCTION
COMPUTATIONAL ONCOLOGY is recently gaining in-creased attention among the research community. This
field aims to investigate computational models for tumor pro-
gression, which can help to better understand the phenomenon
of cancer and finally provide better diagnosis and treatment
plans for patients. Current approaches range from the molec-
ular, to the cellular, up to the macroscopic level including
biomechanics. In Stamatakos et al. [1] an “Oncosimulator”
has been proposed, which aims to model cancer progression
on a biological level, taking into account cell proliferation.
Konukoglu et al. [2] used physical models to better understand
the progression of gliomas by adapting reaction-diffusion
dynamics. As gliomas also exhibit a significant mass-effect on
the surrounding tissues, biomechanics should be considered as
presented by Hogea et al. [3].
Brain tumor image analysis is a more established field
than computational oncology, however active research is being
Manuscript received April 9, 2011; revised May 31, 2011; accepted July
25, 2011.
S. Bauer, C. May, P. Bu¨chler and M. Reyes are with the Institute for Surgical
Technology and Biomechanics, University of Bern, Bern, Switzerland. e-mail:
fstefan.bauer, christian.may, philippe.buechler, mauricio.reyesg@istb.unibe.ch
Dimitra Dionysiou and Georgios Stamatakos are with the Institute of Com-
munication and Computer Systems, National Technical University of Athens,
Athens, Greece. e-mail: dimdio@esd.ece.ntua.gr, gestam@central.ntua.gr
Funding by the European Union within the framework of the ContraCan-
crum project (FP7 IST-223979) is gratefully acknowledged.
Copyright (c) 2010 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending an email to pubs-permissions@ieee.org.
conducted to handle the varying appearance of brain tumors,
which makes generic tumor-bearing brain segmentation and
registration a challenging task. While the majority of methods
are mostly concerned with tumor segmentation, e.g. Verma et
al. [4], fewer work has been done on aligning brain tumor
images with a standard template using registration. One of the
latest efforts to adapt a registration and segmentation method
for brain tumor images was done by Zacharaki et al. [5]. In
this work, a purely macroscopic biomechanical tumor growth
model is used to simulate tumor growth in a healthy atlas,
which is subsequently registered to the patient image using
deformable registration algorithms.
The aim of this paper is to present a novel multi-scale
method for patient-specific adaptation of a healthy brain atlas
to tumor patient images, which also offers implicit segmen-
tation of the brain tissues. A major problem is to establish
correspondence between a healthy atlas and a tumor-bearing
patient image in a generic way. To this end, we combine
patient-specific tumor growth simulation with medical image
analysis. The approach comprises a multi-scale, multi-physics
model of tumor growth and progression, from the cellular, up
to the biomechanical level, and state-of-the art methods for
non-rigid registration of brain images.
II. METHODS
In a first step, we simulate tumor growth in a healthy brain
atlas. We use the publicly available SRI24 atlas provided by
Rohlfing et al. [6], which is an average of 24 normal adult
subjects. This atlas provides different modalities, including
label maps. It exhibits increased sharpness, making it suitable
for atlas-based segmentation purposes.
After applying a preprocessing pipeline to the patient image,
including intensity normalization, customized skull-stripping1,
edge-preserving smoothing and bias-field correction, initial
correspondence between the atlas and the patient image is es-
tablished using an affine registration method. Next, the tumor
area in the patient can either be delineated manually or using
classification methods on multi-modal magnetic resonance
(MR) images as shown in [4]. A physically realistic seed for
tumor growth is automatically chosen in the vicinity of the
center of mass of the patient tumor, with the patient tumor
outline also delimiting the growth process later.
1software tool available at
www.istb.unibe.ch/content/surgical technologies/medical image analysis/software