Looking and listening to light: The evolution of whole-body photonic imaging

Abstract

Optical imaging of live animals has grown into an important tool in biomedical research as advances in photonic technology and reporter strategies have led to widespread exploration of biological processes in vivo. Although much attention has been paid to microscopy, macroscopic imaging has allowed small-animal imaging with larger fields of view (from several millimeters to several centimeters depending on implementation). Photographic methods have been the mainstay for fluorescence and bioluminescence macroscopy in whole animals, but emphasis is shifting to photonic methods that use tomographic principles to noninvasively image optical contrast at depths of several millimeters to centimeters with high sensitivity and sub-millimeter to millimeter resolution. Recent theoretical and instrumentation advances allow the use of large data sets and multiple projections and offer practical systems for quantitative, three-dimensional whole-body images. For photonic imaging to fully realize its potential, however, further progress will be needed in refining optical inversion methods and data acquisition techniques.

Full text

PERSPECTIVE
NATURE BIOTECHNOLOGY VOLUME 23
NUMBER 3
MARCH 2005 313
Looking and listening to light: the evolution of
whole-body photonic imaging
Vasilis Ntziachristos
1
, Jorge Ripoll
1,2
, Lihong V Wang
3
& Ralph Weissleder
1
Optical imaging of live animals has grown into an important
tool in biomedical research as advances in photonic
technology and reporter strategies have led to widespread
exploration of biological processes in vivo. Although much
attention has been paid to microscopy, macroscopic imaging
has allowed small-animal imaging with larger fields of view
(from several millimeters to several centimeters depending
on implementation). Photographic methods have been the
mainstay for fluorescence and bioluminescence macroscopy in
whole animals, but emphasis is shifting to photonic methods
that use tomographic principles to noninvasively image optical
contrast at depths of several millimeters to centimeters with
high sensitivity and sub-millimeter to millimeter resolution.
Recent theoretical and instrumentation advances allow the use
of large data sets and multiple projections and offer practical
systems for quantitative, three-dimensional whole-body images.
For photonic imaging to fully realize its potential, however,
further progress will be needed in refining optical inversion
methods and data acquisition techniques.
Small-animal imaging is rapidly becoming a cornerstone in biomedi-
cal investigation, serving as an important translation tool between in
vitro research and clinical application. Recently, key advances in the in
vivo reporting of genomics and proteomics have intensified the devel-
opment of dedicated small-animal imaging systems and strategies
1–6
.
Linked to these developments is an emerging shift from traditional in
vitro assay–based research to in vivo imaging–based research. In vivo
imaging can improve our ability to probe complex biologic interactions
dynamically and to study disease and treatment responses over time in
the same animal, thus offering the potential to accelerate basic research
and drug discovery using fewer animals.
Two major approaches have been adopted in the development of
in vivo small-animal imaging. The first is an elegant adaptation of
proven clinical imaging technologies to the smaller animal dimen-
sions. It includes all of the major radiological modalities; that is,
positron emission tomography (PET), single photon emission com-
puted tomography (SPECT), magnetic resonance imaging (MRI),
ultrasound, X-ray computed tomography (CT) and multi-modality
approaches
1,5,6
. The resulting imaging systems attain higher resolu-
tion and detection sensitivity compared with their clinical counter-
parts because of the smaller field of view used and the corresponding
modification of the operating characteristics (for example, higher
field strengths for MRI or higher frequencies for ultrasound).
The second approach focuses on new imaging technologies, primarily
macroscopic imaging based on photonics. These novel methods extend
beyond three-dimensional in vivo microscopy (such as multi-photon or
confocal microscopy), which is not well suited for whole-body imaging
because of its limited penetration depth (<1 mm) and the restricted
field of view typically achieved. Here, progress in instrumentation and
methodology is combined with ingenious advances in fluorescence or
bioluminescence reporter gene/reporter probe
1,4,7
strategies, activatable
fluorescent probes and targeted fluorescent probes
8,9
for in vivo molecu-
lar sensing. Whole-body fluorescence and bioluminescence imaging have
transformed the ways gene expression and protein functions are visual-
ized in vivo
10,11
. In most cases, optical detection has been accomplished
using photographic methods, using low-light cameras and appropriate
filters. Photon attenuation, however, is strongly nonlinear as a function
of depth and of the optical heterogeneity of tissue, which obscures signal
quantification. Planar imaging is further complicated by the inability to
resolve depth and by tissue scattering, which limits spatial resolution. For
these reasons, although planar methods are useful, they do not harness
the true potential of the optical imaging technologies.
The emergence of mathematical models that describe photon propa-
gation in tissues, combined with advanced illumination and detection
schemes, and appropriate tomographic principles, can significantly
improve visualization capacity in tissue. Tomography enables quan-
titative three-dimensional volumetric imaging of opaque media and
can overcome planar imaging limitations. The combination of these
advances with adept contrast mechanisms using highly specific fluo-
rescent probes or the photoacoustic phenomenon has facilitated a new
generation of photonic imaging systems that is rapidly maturing and can
greatly facilitate small-animal research by complementing the laboratory
microscope, spectrophotometer and flow cytometer with whole-body
in vivo molecular imaging capacity.
This review focuses on novel macroscopic photonic imaging tech-
nologies that, in combination with emerging reporter strategies, promise
to provide researchers with unprecedented power to visualize biological
processes. We compare photographic and tomographic optical imaging
and describe the three major optical domains of optical tomography
1
Center for Molecular Imaging Research (CMIR), Massachusetts General
Hospital & Harvard Medical School CNY149, 13
th
street 5406, Charlestown,
Massachusetts 02129, USA.
2
Institute for Electronic Structure and Laser,
Foundation of Research and Technology-Hellas, P.O. Box 1527, 71110
Heraklion, Greece.
3
Optical Imaging Laboratory, Department of Biomedical
Engineering, Texas A&M University, 3120 TAMU, College Station, Texas
77843-3120, USA. Correspondence should be addressed to V.N.
(vasilis@helix.mgh.harvard.edu).
Published online 4 March 2005; doi:10.1038/nbt1074
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

PERSPECTIVE
314 VOLUME 23
NUMBER 3
MARCH 2005 NATURE BIOTECHNOLOGY
(that is, time, frequency and constant intensity). Particular weight is
given to recent noncontact imaging approaches and tomographic spe-
cifics that significantly improve imaging performance and enable the
development of practical and accurate systems. State-of-the-art in vivo
imaging examples are illustrated. Bioluminescence and photoacoustic
tomography are also reviewed, together with the expected impact of
these technologies in biomedical research.
Planar imaging
Planar imaging, the simplest technique for detecting optical reporter
molecules in vivo, uses photographic principles to capture light
emitted from the animals
12–14
and has broadly affected biomedical
research
7,11,15
. For fluorescence imaging, the animal is typically illu-
minated with a broad light beam tuned to the excitation wavelength
of the fluorochrome of interest and is photographed at the emission
wavelengths by a highly sensitive and low noise charged-coupled device
(CCD) camera using appropriate filters and large aperture lenses for
high photon collection efficiency (Fig. 1a). These low-light images
are superimposed on mouse photographs obtained at the excitation
wavelength or with white light illumination. This is the most com-
mon whole-body fluorescence imaging approach today and is termed
fluorescence reflectance imaging (FRI). Advanced forms of FRI use
spectral information to differentiate between different fluorochromes.
Yang et al.
14
, for example, demonstrated that the use of a highly sensi-
tive color CCD camera can detect green fluorescent proteins expressed
by tumors implanted superficially in living animals. More recently, to
improve detection contrast (see Fig. 2), Gao et al.
16
have used spectral
un-mixing, a technique that can differentiate multiple fluorochromes
from nonspecific background fluorescence on the basis of their spectral
characteristics (for details, see ref. 17). Bioluminescence signals can be
similarly detected, but in the absence of external illumination light. In
contrast, fluorescence transillumination imaging illuminates the animal
with a broad light beam or a raster scan of focused beam and images are
collected from the opposite side of the illumination source (Fig. 1b).
Planar imaging offers an attractive tool for high-throughput imaging
and it is technically easy to implement. However, it also has important
limitations, such as the single projection viewing, the restricted penetra-
tion depth of a few millimeters and the nonlinear relationship between the
signal strength and the depth and the tissue optical properties. These fea-
tures limit the applicability of the method primarily to superficial obser-
vations and may lead to erroneous interpretation of the data collected if
the nonlinear effects are not explicitly corrected or accounted for. Box 1
outlines some examples of planar imaging limitations in relation to depth
and tissue optical properties and contrasts them with tomography.
Fluorescence tomography
Tomographic reconstruction of fluorescence biodistribution has
its roots in the early 1990s when the first theoretical frameworks
for tomography of diffuse media were proposed to spatially resolve
intrinsic tissue contrast (primarily absorption and scattering) in the
Reflectance Transillumination
Tomography Diffuse pattern
Input
Input
Input
Output
Output
Output
1 cm
1 cm
Diffusive photons
ab
cd
Figure 1 Modes of data collection. (a,b) Planar
imaging. Fluorescence reflectance imaging.
Excitation light (input) is expanded on the object
surface and fluorescence light (output) collected
from the same side of the object. Scattered photon
trajectories are simplistically demonstrated with
a few lines to demonstrate the typical volume
sampling. (b) Transillumination illuminates the
object from the opposite side from the data
collection so that light propagates through the
object. Photon scattering is also indicated with
a few lines and indicates the volume sampling
differences in relation to a. Transillumination can
be obtained with a single point source (green), a
summation of multiple point sources or with an
expanded planar beam (light gray). (c) Tomography.
Illustration of data collection where multiple
point-source transillumination data are time-shared
around a cylindrical geometry. Different geometries
and the use of reflected data can also be used for
tomographic purposes. The direction indicated
by the arrows shows the general photon trajectory
established. The pattern of data collected is, in
fact, diffusive as is evident from the experimental
measurements shown in (d) obtained from a
transilluminated homogeneous diffusive cylinder.
Figure 2 Spectral imaging applied to in vivo
fluorescence detection. (a) Standard color image
obtained from a green fluorescence protein
(GFP)-expressing mouse implanted with a red
fluorescent protein (RFP)-expressing tumor.
(b) Spectral imaging and processing improves
visualization of the RFP signal, which can
be separated from the mouse intrinsic auto-
fluorescence and GFP fluorescence. Images and
analysis were provided by Cambridge Research &
Instrumentation (CRI); samples were provided by
Anticancer.
ab
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

PERSPECTIVE
NATURE BIOTECHNOLOGY VOLUME 23
NUMBER 3
MARCH 2005 315
context of studying hemodynamics or organelle concentration
18–21
.
This work followed original observations that near-infrared light
(NIR: 650 nm–900 nm) can penetrate several centimeters into tissues
because of the low photon absorption in this spectral window
22,23
.
In contrast to high-energy rays, however, NIR photons are highly
scattered in tissue and they become diffuse within approximately a
millimeter of propagation
24
. For tomography, multiple points on the
tissue boundary are illuminated in a time-sharing fashion (Fig. 1c)
and diffuse light patterns are collected around the boundary (Fig. 1d)
using photodetector sets or a CCD camera. Each source-detector
pair effectively implements a different projection through the tissue,
albeit following diffusive propagation patterns . Fluorescence mea-
surements can be obtained using appropriate filters in front of the
detectors, although the same generic tomographic principles are used
for reconstruction of intrinsic tissue contrast, that is, absorption or
scattering. These measurements are then combined in a tomographic
scheme, which can be written as a system of equations that are solved
for the unknown, spatially dependent fluorochrome concentration.
This generic mainframe, combined with appropriate fluorescent
molecules with specificity to cellular and sub-cellular processes,
has led to the development of fluorescence molecular tomography
(FMT), a technology directed towards noninvasive quantitative
Box 1 Performance characteristics of planar and tomographic imaging
A side-by-side comparison of the performance of planar and
tomographic imaging in visualizing tissue-mimicking phantoms
reveals some of the likely drawbacks of the former approach in
imaging small animals. Figure 6a shows planar and tomographic
imaging of two ∼1.5-mm diameter fluorescent tubes containing
cyanine 5.5, which are immersed in a diffusive fluid with the
average optical properties of small animals and imaged through a
glass window. Both planar and FMT resolve the tubes when they
are placed in contact with the glass window, although FMT offers
better resolution. However, planar imaging detection becomes
highly challenging as the tubes move deeper in the diffuse
medium, away from the glass window, even at the 3 mm depth.
Depth sensitivity depends strongly on the light strength used and
the size and concentration of the fluorochrome used, but the
images demonstrate the superior ability of tomography to look
deeper into diffuse media with higher resolution, using in this
case identical hardware to that used with the planar imaging. The
color images are reconstructed slices obtained at different depths
and superimposed on gray scale photographs of the tubes for
visualization purposes
52
.
Similarly, imaging fluorochromes with varying background optical
properties is shown in Figure 6b. Both planar and FMT accurately
resolve the 2:1 relation in fluorochrome concentration between
the left and right tubes when the background absorption is the
same in both tubes (top row). However, when India ink is added
in the left tube to simulate a threefold increase in vascularization
(absorption), the planar image erroneously reports a ∼1:1 cyanine
5.5 concentration in the two tubes (bottom row). This is because
the added ink absorbs more fluorescent photons. In contrast, FMT
can correct for the added absorption and demonstrates more robust
performance, reporting 1.8:1 relation in this case
52
.
Finally, Figure 6c demonstrates the capacity of FMT to image
fluorochromes in a highly diffusive medium that simulates tissue
optical heterogeneity. The image labeled nBorn shows that the
use of normalization methods (in this case, as described in
ref. 41) can accurately resolve the fluorescence distribution,
despite the highly heterogeneous background. The absence of
data normalization, however, yields images that are affected by
the background heterogeneity, as shown in the image labeled
hBorn. The planar imaging could not detect the presence of the
fluorochrome in this case.
Overall, these findings indicate that planar imaging should be
used with caution. Signal intensity relates linearly to fluorochrome
concentration but nonlinearly to depth, size and optical properties,
and its measurement is further complicated by the highly scattering
nature of tissue. FMT has the potential to circumvent some of these
limitations and offers more robust and accurate imaging.
a
b
c
Figure 6 Performance of planar and tomographic imaging. (a) Two ∼1.5mm
diameter fluorescent tubes (500 nM cyanine 5.5, 3 mm apart) immersed
at different depths in an intralipid and India ink solution, simulating the
optical properties of small animals. (b) Planar and reconstructed images
of two fluorescent tubes, the left containing twice as much cyanine 5.5 as
the right tube (400 nM versus 200 nM). The tubes are immersed in the
same diffuse fluid as in a. (c) A single fluorescing tube is asymmetrically
surrounded by five absorbers at twofold and threefold the background
absorption. The tube is reconstructed using a normalized (nBorn) and not
normalized method (hBorn).
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

PERSPECTIVE
316 VOLUME 23
NUMBER 3
MARCH 2005 NATURE BIOTECHNOLOGY
molecular imaging of whole animals and tissues. Tomography can
overcome several of the limitations of planar imaging, as summa-
rized in Box 1.
Optical domains. There are three distinct illumination-detection tech-
nology domains for optical tomography; that is, the time-domain (TD),
the frequency domain (FD) and the continuous wave (CW) domain. Each
has distinct advantages and disadvantages, and the selection of the appro-
priate technology largely depends on the specific application (Table 1).
For molecular investigations where the goal is to localize and quantify
fluorescent probes, CW imaging offers excellent detection characteristics.
CW domain methods use light of constant intensity and require simple
and low-cost optical components
25
. They further offer optimum signal-
to-noise performance because CW light sources and detectors are typi-
cally more stable and have low noise characteristics compared with those
sources and detectors used in TD and FD methods. The major disad-
vantages of CW domain methods include the difficulty of resolving the
tissue absorption from scattering and the inability to image fluorescence
lifetime. Another caveat is that, unlike TD or FD methods, resolution is
entirely dependent on a tissue’s optical properties and geometry; it can
not otherwise be optimized.
When independent measurements of tissue
absorption, scattering or fluorochrome life-
time are required, the use of TD or FD tech-
nology becomes essential
22,26,27
. TD methods
illuminate tissue with ultrafast (femtosecond
to picosecond) photon pulses and resolve the
arrival of the photons as a function of time at
different locations around the tissue boundary.
In contrast to CW methods, they can use early
arriving photons to improve resolution because
highly diffusive photons are rejected
28–30
. On
the downside, TD methods are less sensitive
than CW methods because of the lower duty
cycles achieved (that is, the length of time the
laser beam and detector is on), resulting in
dimmer average light intensity available for imaging. In addition, TD
instrumentation is noisier than CW systems due to time and intensity
fluctuations that are associated with ultrafast switching electronics and
pulsing lasers.
The third mode, FD technology, uses light of a modulated intensity
at a frequency f, which establishes a photon wave of the same fre-
quency in the diffuse medium
31
. Measurements of the light intensity
and the phase shift of the photon wavefront away from the source or
the excited fluorochromes reveal information about the tissue opti-
cal properties and fluorochrome bio-distribution
32
. FD methods are
less affected by ambient light than CW and TD methods. However,
they require frequencies of several hundred MHz or higher to achieve
improvements in resolution over CW. They are also less robust than
CW methods because of the reduced signal-to-noise detection
involved in sensing high frequencies. Data obtained at multiple fre-
quencies improve FD imaging performance and can become equiva-
lent to TD data via the inverse Fourier Transform.
Improving spatial sampling. Previous fluorescence tomography
investigations focused primarily on feasibility studies with simple tis-
sue-mimicking phantoms and on algorithmic validation. Typical sys-
tems used a limited number of sources and
detectors arranged around the tissue boundary
using light-guiding fibers. In several instances,
a matching fluid was used to surround the
tissue and achieve optimal photon coupling
and a simplified experimental arrangement;
otherwise, fibers were brought in contact with
the tissue. This technology typically yielded
coarse spatial photon-sampling on the bound-
ary, resulting in 10
2
–10
3
total measurements,
which is generally insufficient for high-fidel-
ity volumetric imaging. In addition, the fiber-
based technology complicated experimental
procedures because either the tissue had to be
surrounded by fluids or meticulous engineering
had to be exercised to ensure the optimal con-
tact of each individual fiber with the tissue
33
.
This set of approaches compromised imaging
performance and reduced overall enthusiasm
for optical tomography.
More recently, it has been shown that sub-
millimeter spaced arrays of sources and detec-
tors (that is, data sets on the order of 10
4
–10
6
measurements or more) are necessary for
high-fidelity, small-animal imaging
25
. To
achieve such large data sets, researchers have
Table 1 Optical domains
Domain Time Frequency Continuous wave
Resolution
a
0.5–1 mm 0.5–1 mm 1 mm
Sensitivity
a
Picomoles Picomoles Picomoles-
femtomoles
Depth
b
<30 mm <30 mm <50 mm
Contrast T, A/S, F/L T, A/S, F/L T, F, B
Fluorochrome quantification Yes Yes Yes
Signal-to-Noise Ratio Medium Medium High
Cost High Medium/high Low/medium
a
Reported for fluorochromes at the center of 15 mm thick tissues.
b
Assuming ∼10 picomoles of an organic fluorochrome.
A, absorption; S, scattering; F, fluorescence concentration; L, fluorescence lifetime; T, attenuation, B, bioluminescence.
Figure 3 Fluorescence reconstruction of a fluorescent tube inserted in a euthanized animal obtained in
the absence of contact detection. Image reconstruction is based on mathematical models that describe
the composite photon propagation in tissue and in air (for details see ref. 36). The animal surface was
captured using photo-grammetry, that is, the mathematical combination of photographs obtained under
different angles to deduce the physical dimension of the animal.
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

PERSPECTIVE
NATURE BIOTECHNOLOGY VOLUME 23
NUMBER 3
MARCH 2005 317
introduced free-space and noncontact imaging approaches. Free-space
tomography is based on appropriate forward models that predict the
composite propagation of photons in diffuse media and in air
34
and is
enabled by noncontact collection methods, such as direct lens coupling
of CCD cameras onto nonbounded tissue
35
. Free-space imaging is fur-
ther combined with surface capture optical methods to yield an accurate
description of the arbitrary tissue boundaries. These collection schemes
offer high-quality data sets, high-spatial photon sampling and experi-
mental simplicity because they both avoid the use of matching fluids/
complex fiber interfaces and eliminate the associated fiber-tissue cou-
pling issues. The feasibility of this approach has been recently showcased
with phantoms
35
and animals
36
. Figure 3 depicts reconstructed images
of a 1.5-mm diameter fluorescent tube (500 nM of cyanine 5.5) inserted
through the esophagus of a euthanized animal. With these advances, it is
now possible to obtain practical, complete projection (360°), noncontact
systems that, similar to other tomographic modalities (e.g., X-ray CT),
can offer optimum imaging performance.
Forward problem and inversion. Two factors have a significant influ-
ence on tomographic performance: first, the selection of appropriate
mathematical models that describe photon propagation in tissues (that
is, the forward problem); and second, the selection of image reconstruc-
tion algorithms (that is, the inverse problem). Typical forward problems
used for fluorescence tomography of tissues are based on numerical or
analytical solutions of the diffusion equation
37–41
solved for the exci-
tation and fluorescence fields. Forward models based on approximate
solutions to the radiative transport equation
42
or on diffusion equation
solutions merged with radiosity principles
43
have been proposed for
regimes where solutions of the diffusion equation become less accurate
(such as, situations using early photons, millimeter-sized, source-detec-
tor separations or in void, nondiffusive regions).
A particular scheme that recently enabled in vivo application
44
has
been the inversion of normalized data (that is, by solving for the ratio of
fluorescence measurements over excitation measurements to minimize
the sensitivity to tissue heterogeneity and to theoretical inaccuracies
41
(see Box 1). Such methods are computationally fast, robust and simple to
implement and can be used with analytical and numerical solvers. More
integrated approaches are based on iterative numerical solutions, and
they handle heterogeneity explicitly by solving first for the background
absorption and scattering and then implementing this information for
fluorescence solutions
40
. Overall, the need for fast forward and inversion
algorithms is becoming ever more important as data sets increase in size
as a result of the application of newer generation noncontact instru-
ments. The use of fast analytical solvers
45
or the acceleration of numeri-
cal solutions using, for example, multi-grid methods
46,47
are important
contributions to achieving practical inversion schemes.
In vivo applications. Figure 4 summarizes FMT studies from our
group using noncontact approaches for fluorescence tomography of
small animals. The results are contrasted with planar imaging methods
to illuminate differences. Figure 4a–f depicts findings from an in vivo
imaging study of inflammatory lung disease. In this study, pulmonary
inflammation was induced by intratracheal lipopolysaccharide (LPS)
instillation in a BALB/c mouse using a previously described procedure
48
(LPS administration has previously been shown to upregulate cathepsins
in macrophages as well as other pro-inflammatory pathways
49,50
). The
a
ghi jkl
bcdef
Figure 4 In vivo fluorescence imaging. (a–e) Imaging of proteolytic activity in LPS-induced pulmonary inflammation. Fluorescence reflectance images of the
LPS-challenged and nonchallenged mouse (control) injected with a cathepsin-sensitive fluorescent probe (a,b). Fluorescence transillumination images of
control and LPS-challenged mouse, respectively (c,d). Fluorescence Molecular Tomography slice (e) and corresponding T1-weighted MR image (f).
(g–i) Imaging of lung tumors. In vivo fluorescence reflectance image (g). Post mortem fluorescence reflectance image after skin and rib-cage removal
(h). (i) FMT slice at 2 mm depth under the surface. (a–i) Images are courtesy of investigators at the Laboratory for Bio-optics and Molecular Imaging/CMIR.
(j–l) Differential imaging of treatment effects using an annexin V-Cy5.5 probe. In vivo planar reflectance image (j), two consecutive FMT slices obtained at
1 mm and 1.8 mm under the animal surface (k,l) (see text for details).
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

PERSPECTIVE
318 VOLUME 23
NUMBER 3
MARCH 2005 NATURE BIOTECHNOLOGY
animal was imaged 24 h after challenge follow-
ing administration of a fluorescence-activatable
probe sensitive to major cathepsins (B > S, K,
L) associated with inflammation
51
. Although
the fluorescence reflectance images, shown
in Figure 4a,b, are unable to resolve protease
activity from the lung, transillumination images
(Fig. 4c,d) do depict a marked difference in
fluorescence distribution between the control
and the LPS-treated animal. Correspondingly,
a tomographic slice (Fig. 4e), obtained 3 mm
under the surface, better demarcates the lung
inflammation and demonstrates good congru-
ence with the anatomical magnetic resonance
image shown in Fig. 4f, which was obtained
under identical placement conditions. Fig. 4e
does not depict signal from the liver because the
reconstruction algorithm automatically rejects
fluorescence signals corresponding to excitation
patterns that have been highly absorbed.
In a related example from our laboratory, a
Lewis Lung carcinoma tumor (LLC), was grown
in the lung of a nu/nu female after direct injec-
tion of 10
6
LLC cells mixed with MatrigelHC
and targeted with the same fluorescent activatable probe used in Fig.
4a–e. The tumor was not visible on fluorescence reflectance images (Fig.
4g) in vivo but could be seen by FRI when the skin and the front rib cage
were removed after the mouse was killed (Fig. 4h). Conversely, FMT
(Fig. 4i) could detect contrast congruent with the appearance of the
tumor. Tomographic feasibility to resolve molecular activity in animal
brain tumors has also been shown based on fiber-based systems
44
.
Finally, in Figure 4j–l, we illustrate the ability of tomography to image
treatment effects in vivo. In this study
52
, a mouse bearing LLC tumors
sensitive and resistant to cyclophosphamide was treated with the drug
and injected with a phosphatidylserine (PS)-sensing annexin V–based
probe conjugated to the cyanine 5.5 fluorochrome to probe apoptosis. In
Figure 4j–l, the drug-sensitive and drug-resistant tumors are in the left
and right mammary area, respectively. Because of the superficial nature
of the tumors, both the planar (Fig. 4j) and the two tomographic slices
obtained from two adjacent depths (Fig. 4k,l) resolve the tumors. The
tomography, however, allows more accurate quantification, tumor-to-
tissue contrast, depth and size estimation. Planar imaging did not cor-
rectly indicate levels of apoptosis in some instances because of unequal
absorption between the two tumors
52
.
Bioluminescence tomography
Bioluminescence imaging uses enzymes, which convert unique sub-
strates into light in the presence of oxygen and other factors (e.g.,
ATP, Mg)
4
. The propagation of emitted photons can be modeled,
similarly to fluorescence photon propagation, as a diffusion process.
Bioluminescence tomography can therefore be based on the same
framework used for fluorescence tomography but light is collected from
the subject in the absence of external illumination sources. Because
internal bioluminescent light is continuously on during the measure-
ment, bioluminescence tomography operates only in CW mode. One
of the major advantages of the technique is that there is no inherent
background bioluminescence in most tissues, which yields high imag-
ing contrast. Methods for bioluminescence tomography have recently
been reported
53,54
, and there is a great impetus for in vivo tomographic
applications for improving localization and quantification beyond what
has been achieved by planar methods.
However, the unavailability of external illumination sources com-
plicates the tomographic problem, compared to fluorescence tomog-
raphy, because it becomes mathematically more difficult and possibly
less accurate, to resolve problems of internal sources (bioluminescence)
compared with problems using external sources (fluorescence) because
of the fewer source-detector pairs (projections) available. This problem
is also common in other tomography methods (e.g., SPECT versus CT),
although the experimental and image formation characteristics are dif-
ferent for high-energy and near-infrared photons. A combination of
multi-view angle (360°) imaging with a priori information on tissue
heterogeneity could improve the performance of the bioluminescence
inverse problem
53
. In vivo applications of bioluminescence tomography
have not yet been reported; however this technology is being actively
researched and may become available in the future.
Photoacoustic tomography
When a short laser pulse, typically in the nanosecond range, is
spatially broadened and then used to irradiate biological tissue, it
produces a temperature rise on the order of milli-Kelvin in a short
time frame. Consequently, thermoelastic expansion causes emission
of acoustic waves, referred to as photoacoustic waves, that can be
measured by wideband ultrasonic transducers around the sample.
This phenomenon, discovered by Alexander Graham Bell, has been
recently exploited for small-animal imaging, because the acquired
photoacoustic waves can be combined mathematically to reconstruct
the distribution of optical energy absorption
55,56
. The technique,
termed photoacoustic tomography (PAT), also referred to as opto-
acoustic or thermoacoustic tomography, attains the advantage of
combining ultrasonic-scale spatial resolution with high sensitivity to
tissue light absorption
57,58
and can yield information on physiology
or on exogenously administered light absorbers.
This technique has been used recently for visualization of the brain
structure and lesions, of cerebral hemodynamic responses to hyper-
oxia and hypoxia and of cerebral cortical responses to neuroactivi-
ties induced by whisker stimulations in rats
57
. Figure 5 depicts two
images of the superficial cerebral cortex of the rat after stimulation
of the left-side and the right-side rat whiskers, respectively. These
ab
Figure 5 Visualization of brain structure and function using photoacoustic tomography (modified from
ref. 57. (a,b) Functional maps of brain activities corresponding to the left-side (a) and right-side (b)
whisker stimulations, respectively, acquired with the skin and skull intact.
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

PERSPECTIVE
NATURE BIOTECHNOLOGY VOLUME 23
NUMBER 3
MARCH 2005 319
images were obtained after subtraction of the image without whisker
stimulations from the two PAT images with whisker stimulations.
This differential contrast is attributed to the altered blood volume
and oxygenation associated with the whisker stimulations. Similarly,
noninvasive in vivo imaging of exogenous contrast agents in the rat
brain using indocyanine green stabilized with polyethylene glycol has
been also demonstrated
56
.
Outlook
The advancement of new whole-animal imaging technologies pres-
ents new opportunities for biomedical research. Tomographic meth-
ods can transform macroscopic optical observations of tissues from
a crude qualitative tool to an accurate three-dimensional imaging
technique. This is important for capitalizing on the advantages of
fluorescence and the bioluminescence methods: first, high molecular
specificity; second, the use of nonionizing radiation that simplifies
the operation and the chemical synthesis of reporter probes; third,
the ability to use optical switches (molecular beacons) for achiev-
ing high sensitivity and specificity; fourth, optical probe stability
because there is no intensity decay over time; and fifth, the potential
for simultaneous investigations of multiple targets using spectral
differentiation of probes.
These imaging principles can be applied to different biomedical
research areas, including cancer, cardiovascular, immunologic/inflam-
matory and neurodegenerative diseases
5,10,11
. Important to these new
developments is the accessibility of a larger number of sites and organs
compared to the number that can be assessed using planar imaging.
While small animal photonic tomography does not reach the resolution
of optical microscopy, molecular activity is detected based on the high
specificity of the optical probes employed (similarly to PET and SPECT),
even if single cells and molecules are not explicitly resolved.
One other advantage of fluorescence tomography is that it can be
combined with microscopy in a straightforward manner. Because it uses
the same optical probe as fluorescence microscopy, a sample can first be
visualized and quantified volumetrically in vivo as a function of time
and then observed at high resolution with microscopy using excised
samples or surgical intervention in the case of tissues that are not typi-
cally accessible by microscopy (that is, when samples are deeper or larger
than ∼400–700 µm).
The combination of modalities with complementing features offers
an attractive future direction for study. Although optical tomography
yields high molecular contrast versatility and specificity, photoacous-
tic imaging offers improved resolution to imaging light absorp-
tion. The combination of the two techniques could not only yield a
straightforward superposition of optical and photoacoustic images,
but also use PAT images as a priori information for constructing more
accurate models for photon propagation in tissues, which will fur-
ther mitigate the optical inversion problem. Other imaging modali-
ties, such as CT, MRI or ultrasound, may also be used to register
PAT images or fluorescence images onto high-resolution anatomical
images, thereby improving the information content of the combined
imaging approach.
Overall, these new technologies will continue to emerge and diversify,
and new clinical applications will be identified; for example, in imaging
human joints or the breast, in eye healthcare and in endoscopic appli-
cations, where appropriate photon models and limited angle projec-
tions would improve imaging performance. Currently, only a few sets
of approaches and ideas about photonic imaging methodology have
been explored. Light offers a plethora of contrast mechanisms and can
be manipulated in several ways to further improve the performance and
capability of these methods.
ACKNOWLEDGMENTS
V.N. is supported in part by National Institutes of Health (NIH) grants RO1 EB
000750-1, 1-NO1-CO027105 and R33 CA 91807. J. Ripoll acknowledges support
from EU Integrated Project “Molecular Imaging” LSHG-CT-2003-503259. R.W. is
supported in part by NIH grants P50 CA86355, R24 CA92782, R33 CA091807, PO1
AI054904, PO1 CA69246 and grants from the Donald W. Reynolds Foundation and
Siemens Medical Systems.
COMPETING INTERESTS STATEMENT
The authors declare competing financial interests (see the Nature Biotechnology
website for details).
Published online at http://www.nature.com/naturebiotechnology/
1. Massoud, T.F. & Gambhir, S.S. Molecular imaging in living subjects: seeing funda-
mental biological processes in a new light. Genes Dev. 17, 545–580 (2003).
2. Blasberg, R.G. In vivo molecular-genetic imaging: multi-modality nuclear and optical
combinations. Nucl. Med. Biol. 30, 879–888 (2003).
3. Budinger, T.F., Benaron, D.A. & Koretsky, A.P. Imaging transgenic animals. Annu. Rev.
Biomed. Eng. 1, 611–648 (1999).
4. Contag, C.H. & Bachmann, M.H. Advances in in vivo bioluminescence imaging of gene
expression. Annu. Rev. Biomed. Eng. 4, 235–260 (2002).
5. Piwnica-Worms, D., Schuster, D.P. & Garbow, J.R. Molecular imaging of host-pathogen
interactions in intact small animals. Cell. Microbiol. 6, 319–331 (2004).
6. Weissleder, R. Scaling down imaging: molecular mapping of cancer in mice. Nat. Rev.
Cancer 2, 11–18 (2002).
7. Hoffman, R. Green fluorescent protein imaging of tumour growth, metastasis, and
angiogenesis in mouse models. Lancet Oncol. 3, 546–556 (2002).
8. Bornhop, D.J., Contag, C.H., Licha, K. & Murphy, C.J. Advance in contrast agents,
reporters, and detection. J. Biomed. Opt. 6, 106–110 (2001).
9. Tung, C., Mahmood, U., Bredow, S. & Weissleder, R. In vivo imaging of proteolytic
enzyme activity using a novel molecular reporter. Cancer Res. 60, 4953–4958
(2000).
10. Herschman, H.R. Molecular imaging: looking at problems, seeing solutions. Science
302, 605–608 (2003).
11. Weissleder, R. & Ntziachristos, V. Shedding light onto live molecular targets. Nat.
Med. 9, 123–128 (2003).
12. Reynolds, J.S. et al. Imaging of spontaneous canine mammary tumors using fluores-
cent contrast agents. Photochem. Photobiol. 70, 87–94 (1999).
13. Mahmood, U., Tung, C., Bogdanov, A. & Weissleder, R. Near infrared optical imaging
system to detect tumor protease activity. Radiology 213, 866–870 (1999).
14. Yang, M. et al. Whole-body optical imaging of green fluorescent protein-expressing
tumors and metastases. Proc. Natl. Acad. Sci. USA 97, 1206–1211 (2000).
15. Contag, C.H. & Ross, B.D. It’s not just about anatomy: in vivo bioluminescence imaging
as an eyepiece into biology. J. Magn. Reson. Imaging 16, 378–387 (2002).
16. Farkas, D.L.
et al. Non-invasive image acquisition and advanced processing in optical
bioimaging. Comput. Med. Imaging Graph. 22, 89–102 (1998).
17. Gao, X., Cui, Y., Levenson, R.M., Chung, L.W.K. & Nie, S. In vivo cancer targeting and
imaging with semiconductor quantum dots. Nat. Biotechnol. 22, 969–976 (2004).
18. Arridge, S.R., Schweiger, M., Hiraoka, M. & Delpy, D.T.A. Finite-Element Approach
For Modeling Photon Transport In Tissue. Med. Phys. 20, 299–309 (1993).
19. Graber, H.L. & Barbour, R.L. High-resolution near-infrared (nir) imaging of dense
scattering media by diffusion tomography. FASEB J. 7, A720–A720 (1993).
20. Schotland, J.C. & Leigh, J.S. Photon diffusion imaging. FASEB J. 6, A446–A446
(1992).
21. Yodh, A.G. & Chance, B. Spectroscopy and imaging with diffusing light. Phys. Today
48, 34–40 (1995).
22. Chance, B. Optical Method. Annu. Rev. Biophys. Biophys. Chem. 20, 1–28 (1991).
23. Jobsis, F.F. Noninvasive, infrared monitoring of cerebral and myocardial oxygen suf-
ficiency and circulatory parameters. Science 198, 1264–1267 (1977).
24. Patterson, M.S., Chance, B. & Wilson, B.C. Time resolved reflectance and transmit-
tance for the noninvasive measurement of tissue optical-properties. Appl. Opt. 28,
2331–2336 (1989).
25. Graves, E., Ripoll, J., Weissleder, R. & Ntziachristos, V.A. Sub-millimeter resolution
fluorescence molecular imaging system for small animal imaging. Med. Phys. 30,
901–911 (2003).
26. Alfano, R.R. et al. Time-resolved and nonlinear optical imaging for medical applica-
tions. Ann. NY Acad. Sci. 838, 14–28 (1998).
27. Sevick, E.M., Chance, B., Leigh, J., Nioka, S. & Maris, M. Quantitation of time-
resolved and frequency-resolved optical-spectra for the determination of tissue oxy-
genation. Anal. Biochem. 195, 330–351 (1991).
28. Cai, W. et al. Optical tomographic image reconstruction from ultrafast time-sliced
transmission measurements. Appl. Opt. 38, 4237–4246 (1999).
29. Chen, K., Perelman, L.T., Zhang, Q.G., Dasari, R.R. & Feld, M.S. Optical computed
tomography in a turbid medium using early arriving photons. J. Biomed. Opt. 5,
144–154 (2000).
30. Turner, G., Zacharakis, I., Soubret, A. & Ntziachristos, V. Complete angle projection
diffuse optical tomography using early photons. Optics Letters 30, 409–411 (2005).
31. Boas, D.A., Oleary, M.A., Chance, B. & Yodh, A.G. Scattering of diffuse photon density
waves by spherical inhomogeneities within turbid media - analytic solution and applica-
tions. Proc. Natl. Acad. Sci. USA 91, 4887–4891 (1994).
32. Godavarty, A. et al. Fluorescence-enhanced optical imaging in large tissue volumes using
a gain-modulated ICCD camera. Phys. Med. Biol. 48, 1701–1720 (2003).
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology

PERSPECTIVE
320 VOLUME 23
NUMBER 3
MARCH 2005 NATURE BIOTECHNOLOGY
33. Boas, D.A. et al. Imaging the body with diffuse optical tomography. IEEE Signal
Process. Mag. 18, 57–75 (2001).
34. Ripoll, J., Schultz, R. & Ntziachristos, V. Free-space propagation of diffuse light:
Theory and Experiments. Phys. Rev. Lett. 91, 103901–103904 (2003).
35. Schultz, R., Ripoll, J. & Ntziachristos, V. Experimental fluorescence tomography of
arbitrarily shaped diffuse objects using non-contact measurements. Opt. Lett. 28,
1701–1703 (2003).
36. Schultz, R., Ripoll, J. & Ntziachristos, V. Fluorescence tomography of tissues with
non-contact measurements. IEEE Med. Imag. 23, 492–500 (2004).
37. Chang, J., Graber, H.L. & Barbour, R.L. Imaging of fluorescence in highly scattering
media. IEEE Trans. Biomed. Eng. 44, 810–822 (1997).
38. Eppstein, M.J., Hawrysz, D.J., Godavarty, A. & Sevick-Muraca, E.M. Three-dimen-
sional, Bayesian image reconstruction from sparse and noisy data sets: near-infrared
fluorescence tomography. Proc. Natl. Acad. Sci. USA 99, 9619–9624 (2002).
39. Jiang, H.B. Frequency-domain fluorescent diffusion tomography: a finite- element-
based algorithm and simulations. Appl. Opt. 37, 5337–5343 (1998).
40. Milstein, A.B. et al. Fluorescence optical diffusion tomography. Appl. Opt. 42, 3081–
3094 (2003).
41. Ntziachristos, V. & Weissleder, R. Experimental three-dimensional fluorescence recon-
struction of diffuse media using a normalized Born approximation. Opt. Lett. 26,
893–895 (2001).
42. Klose, A.D. & Hielscher, A.H. Fluorescence tomography with simulated data based on
the equation of radiative transfer. Opt. Lett. 28, 1019–1021 (2003).
43. Dehghani, H., Arridge, S.R., Schweiger, M. & Delpy, D.T. Optical tomography in the
presence of void regions. J. Opt. Soc. Am. A Opt. Image Sci. Vis. 17, 1659–1670
(2000).
44. Ntziachristos, V., Tung, C., Bremer, C. & Weissleder, R. Fluorescence-mediated tomog-
raphy resolves protease activity in vivo. Nat. Med. 8, 757–760 (2002).
45. Ripoll, J., Nieto-Vesperinas, M., Weissleder, R. & Ntziachristos, V. Fast analytical
approximation for arbitrary geometries in diffuse optical tomography. Opt. Lett. 27,
527–529 (2002).
46. Gu, X., Xu, Y. & Jiang, H. Mesh-based enhancement schemes in diffuse optical
tomography. Med. Phys. 30, 861–869 (2003).
47. Ye, J.C., Bouman, C.A., Webb, K.J. & Millane, R.P. Nonlinear multigrid algorithms
for Bayesian optical diffusion tomography. IEEE Trans. Image Process. 10
, 909–922
(2001).
48. Vernooy, J., Dentener, M., van Suylen, R., Buurman, W. & Wouters, E. Long-term
intratracheal lipopolysaccharide exposure in mice results in chronic lung inflammation
and persistent pathology. Am. J. Respir. Cell Mol. Biol. 26, 152–159 (2002).
49. Lautwein, A. et al. Inflammatory stimuli recruit cathepsin activity to late endosomal
compartments in human dendritic cells. Eur. J. Immunol. 32, 3348–3357 (2002).
50. Prin-Mathieu, C. et al. Enzymatic activities of bovine peripheral blood leukocytes and
milk polymorphonuclear neutrophils during intramammary inflammation caused by
lipopolysaccharide. Clin. Diagn. Lab. Immunol. 9, 812–817 (2002).
51. Weissleder, R., Tung, C.H., Mahmood, U. & Bogdanov, A. In vivo imaging of tumors
with protease-activated near-infrared fluorescent probes. Nat. Biotechnol. 17, 375–
378 (1999).
52. Ntziachristos, V. et al. Visualization of anti-tumor treatment by means of fluorescence
molecular tomography using an annexin V - Cy5.5 conjugate. Proc. Natl. Acad. Sci.
USA 101, 12294–12299 (2004).
53. Wang, G., Li, Y. & Jiang, M. Uniqueness theorems in bioluminescence tomography.
Med. Phys. 31, 2289–2299 (2004).
54. Gu, X., Zhang, Q., Larcom, L. & Jiang, H.B. Three-dimensional bioluminescence tomog-
raphy with model-based reconstruction. Opt. Express 12, 3996–4000 (2004).
55. Hoelen, C.G.A. & de Mul, F.F.M. Image reconstruction for photoacoustic scanning of
tissue structures. Appl. Opt. 39, 5872–5883 (2000).
56. Wang, X. et al. Noninvasive photoacoustic angiography of animal brains in vivo with
near-infrared light and an optical contrast agent. Opt. Lett. 29, 730–732 (2004).
57. Wang, X. et al. Noninvasive laser-induced photoacoustic tomography for structural
and functional in vivo imaging of the brain. Nat. Biotechnol. 21, 803–806 (2003).
58. Karabutov, A.A., Savateeva, E.V. & Oraevsky, A.A. Optoacoustic tomography: new
modality of laser diagnostic systems. Laser Phys. 13, 711–723 (2003).
© 2005 Nature Publishing Group http://www.nature.com/naturebiotechnology