82 CHINESE OPTICS LETTERS / Vol. 8, No. 1 / January 10, 2010
In-vivo fluorescence molecular tomography based on
optimal small animal surface reconstruction
Daifa Wang (uuu), Xin Liu (444 !!!), Yanping Chen (òòò²²²), and Jing Bai (xxx ÀÀÀ)
∗
Department of Biomedical Engineering, School of Medicine, Tsinghua University,
Beijing 100084, China
∗
E-mail: deabj@tsinghua.edu.cn
Received February 26, 2009
Accurate small animal surface reconstruction is important for full angle non-contact fluorescence molecular
tomography (FMT) systems. In this letter, an optimal surface reconstruction method for FMT is proposed.
The prop osed method uses a line search method to minimize the mismatch between the reconstructed three-
dimensional (3D) surface and the projected object silhouette at different angles. The results show that the
mean mismatches of the 3D surfaces generated on three live anesthetized mice are all less than two charge
coupled device (CCD) pixels (0.154 mm). With the accurately reconstructed 3D surface, in-vivo FMT is
also p erformed.
OCIS co des: 170.6280, 170.6960, 170.3010.
doi: 10.3788/COL20100801.0082.
Fluorescence molecular tomography (FMT) is an emerg-
ing diagnosis tool for small animal research and drug dis-
covery. By tagging regions of interest with target specific
fluorescent probes, FMT may resolve three-dimensional
(3D) locations and geometries of target regions, such
as tumors. With the development of fluorescent probe
technologies, this new technology has been widely used
for gene function, proteins, enzymes, metastasis, drug
discovery, and cancer detection in-vivo
[1]
.
In the past years, FMT systems have evolved from
the early fiber-based systems
[2]
to the non-contact slab-
shaped system using charge coupled device (CCD)
[3]
.
However, the fixed geometries of the above systems re-
quire matching fluids, which is inconvenient for experi-
ments. At the same time, only limited projection angles
were used in slab-shaped based systems
[3]
. Non-contact
FMT system with full-angle projections overcomes the
limited projection angles of the slab geometry. Therefore,
it results in more accurate localization and quantification
information of the fluorescence target
[4]
. Additionally,
it avoids using the matching fluid and simplifies the
experimental setup. For imaging systems with fixed
geometry, a lot of reconstruction methods have been
proposed, including the adaptive mesh base method
[5,6]
,
Newton type method
[7]
, maximum likelihood method
[8]
,
diffuse optical tomography guided method
[9]
, and meth-
ods for time-domain FMT
[10−12]
. However, when apply-
ing these reconstruction methods in in-vivo experiments
on non-contact full-angle FMT systems, accurate mouse
surface should firstly be obtained. With this surface
information, the light transportation in the mouse can
be predicted using the diffusion equation (DE)
[2−10]
. In
Ref. [13], the mouse surface was obtained by using a
photogrammetric 3D camera. Structured light was also
used for obtaining the mouse surface in Xenogen IVIS
Imaging System 200 Series
[14]
and the hyperspectral
imaging system
[15]
. In full-angle non-contact imaging
systems
[4,16]
, the mouse surface can be reconstructed
from white light images
[17,18]
or shadow images
[4]
cap-
tured at different angles. This technique takes full ad-
vantage of the full-angle non-contact imaging system.
Thus, no extra equipment such as a photogrammetric
3D camera
[13]
is needed.
Radon transform with Sheep Logan filter
[18]
or con-
stant filter
[4]
has been used for reconstructing the mouse
surface. Reconstructing a two-dimensional (2D) contour
using Radon transform from projection lines is to find
one circumscribing polygon of the real 2D geometry.
This essence determines the theoretical optimal thresh-
old parameter for extracting the 2D contour from the
back projected 2D image. However, in practical in-vivo
experiments, the optimal parameter will be different
from the theoretical one as the result of mouse breath
movements and mechanical errors. A line search method
is introduced in this letter to obtain an optimal param-
eter, which minimizes the mismatch between the recon-
structed 2D contour and the projection line. Therefore,
the obtained surface will be more accurate compared
with that obtained with other Radon transform based
surface reconstruction methods
[4,18]
. The optimal sur-
face obtained will lead to better mathematical modeling
of light propagation in live mice.
The full-angle non-contact imaging system
[16]
used in
this letter is shown in Fig. 1. The imaged object was
positioned on a stage which could rotate and rise. Light
from a 250-W halogen lamp (7-star, Beijing) traveled
through a 715-nm long-pass filter (Andover Corpora-
tion, Salem, NH) and a 775± 23-nm band-pass filter
(Semrock, Rochester, NY) to provide the excitation
light. The excitation light was then coupled into a 4-
mm inner diameter optical fiber. The excitation light
from the optical fib er was focused on the mouse back
surface with a diameter less than 2 mm and a power of
13 mW. The focus point was maintained over a 9-mm
depth of field, which was appropriate for irregular mouse
shape. Photons which propagate out of the mouse front
surface were acquired by a 512 × 512 element EMCCD
array (Andor, Belfast, Northern Ireland, UK) which was
1671-7694/2010/010082-04
c
° 2010 Chinese Optics Letters