Eur Radiol (2003) 13:195–208

DOI 10.1007/s00330-002-1524-x E X P E R I M E N TA L

Vasilis Ntziachristos Fluorescence imaging with near-infrared light:

Christoph Bremer
Ralph Weissleder new technological advances that enable
in vivo molecular imaging

Received: 23 March 2002 Abstract A recent development in invasive imaging of the distribution
Accepted: 15 April 2002 biomedical imaging is the non-inva- of such probes. We illuminate the
Published online: 19 July 2002 sive mapping of molecular events in advantages and limitations of simple
© Springer-Verlag 2002 intact tissues using fluorescence. photographic methods and turn our
Underpinning to this development is attention to fluorescence-mediated
the discovery of bio-compatible, spe- molecular tomography (FMT), a
cific fluorescent probes and proteins technique that can three-dimension-
and the development of highly sensi- ally image gene expression by re-
tive imaging technologies for in vivo solving fluorescence activation in
fluorescent detection. Of particular deep tissues. We describe theoretical
V. Ntziachristos (✉) · C. Bremer interest are fluorochromes that emit specifics, and we provide insight in-
R. Weissleder
Center for Molecular Imaging Research, in the near infrared (NIR), a spectral to its in vivo capacity and the sensi-
Massachusetts General Hospital window, whereas hemoglobin and tivity achieved. Finally, we discuss
and Harvard Medical School, water absorb minimally so as to al- its clinical feasibility.
Building 149, 13th Street 5406, low photons to penetrate for several
Charlestown MA 02129–2060, USA
e-mail: vasilis@helix.mgh.harvard.edu centimetres in tissue. In this review Keywords Near infrared · In vivo ·
Tel.: +1-617-7266091 article we concentrate on optical Molecular imaging · Fluorescence
Fax: +1-617-7265708 imaging technologies used for non-

Introduction imaging by intravital microscopy [11, 12] or total inter-

nal reflection fluorescence microscopy [13]. Recently
Tissue observation with light is probably the most com- however, light has been used for in vivo interrogations
mon imaging practice in medicine and biomedical re- deeper into tissue using photographic systems with con-
search ranging from the simple visual inspection of a pa- tinuous light [14, 15, 16, 17] or with intensity-modulated
tient to advanced in vivo and in vitro spectroscopic and light [18] and tomographic systems [19, 20]. Potentially,
microscopy techniques [1]. While intrinsic tissue absorp- phased-array detection [21] can also be applied. This re-
tion and scattering yields significant information on cent focus in macroscopic observations of fluorescence
functional and anatomical tissue characteristics, signifi- in tissues has evolved due to the potential of transferring
cant attention has also been given to fluorescence inves- this technology to imaging through animals and humans
tigations of tissue since many biochemical markers can [2]. This technology has become feasible mainly due to
be retrieved due to fluorescence contrast, and many more the development of fluorescence probes emitting in the
can be targeted using appropriate fluorescent markers near-infrared spectrum where tissue offers low absorp-
[2, 3]. tion, highly sensitive detectors and monochomatic light
Numerous different optical imaging approaches can sources (lasers) with higher but nevertheless safely de-
be used for imaging fluorescence in vivo. Traditionally, livered power per wavelength compared with white-light
optical methods have been used to look at surface and illuminators. While macroscopic fluorescence imaging in
subsurface fluorescent events using confocal imaging the visible has also been attempted using fluorescent
[4, 5, 6], multiphoton imaging [7, 8, 9, 10] microscopic proteins, the penetration depth is limited to only 1–2 mm
196

[16], whereas it has been predicted that NIR fluorescent

light can penetrate for several centimeters [22]. Such an
approach could enable investigations available currently
only for in vitro studies to propagate in in vivo human
disease diagnosis and imaging of treatment response and
significantly enhance the field of molecular imaging [2].
In this article we discuss imaging techniques that use
the diffuse component of light for probing molecular
events deep in tissue. Specifically, we focus on reflec-
tance fluorescence imaging and fluorescence-mediated
molecular tomography (FMT), which are the two most
common approaches currently used for imaging fluores-
cent probes in deep tissues. We further discuss recent
findings that predict the capacity of near-infrared fluo-
rescent signals to propagate through human tissue for
non-invasive medical imaging and address feasibility is-
sues for clinical studies.

Reflectance imaging
Technology
Fig. 1 Typical fluorescence reflectance imaging (FRI) system for
in-vivo animal investigations. The illumination source and CCD
Simple “photographic methods”, in which the light camera are on the same side of the animal imaged. The compo-
source and the detector reside on the same side of the an- nents shown are usually enclosed into a light-tight box
imal imaged, are generally referred to as “reflectance im-
aging”. Reflectance imaging is currently the typical
method of choice for accessing the distribution of fluo- CCD chip resolution and dynamic range are not crucial
rescent probes in vivo, but the method can be applied factors in these types of systems. Also the noise require-
more generally to imaging fluorescent proteins or even ments of a CCD camera for fluorescence imaging are not
bioluminescence even if in the latter case no excitation as stringent since in the general case the detection sensi-
light is used [23]. tivity limit is set by the background tissue fluorescence
Near-infrared fluorescence reflectance imaging in (background probe distribution and auto-fluorescence),
particular operates on light with a defined bandwidth as and fluorescence strength can be adjusted above the
a source of photons that encounters a fluorescent mole- camera noise floor by optimizing illumination strength
cule (“optical contrast agent or molecular probe”), which and acquisition times. These features are in contrast to
emits a signal with different spectral characteristics, that bioluminescence measurements, where the detection sen-
can be resolved with an emission filter and captured by a sitivity is usually set by the CCD camera noise floor. De-
high-sensitivity CCD camera. tection systems that scan a single spot over a field of
A typical reflectance imaging system is shown in view and detect light using photo-multiplier tube detec-
Fig. 1. The light source can be either a laser at an appro- tors have also been reported [24]. Such systems may of-
priate wavelength for the fluorochrome targeted or white fer greater dynamic range than CCD-based systems but
light sources using appropriate low-pass filters. General- compromise resolution, acquisition frame rate, and over-
ly, laser sources are preferable because they offer higher all noise characteristics, and are not widely used in in vi-
power delivery at narrower and better-defined spectral vo fluorescence reflectance measurements.
windows (typically ±3 nm for laser diodes vs ±10 nm or Typically, the fluorescence image acquired (see exam-
more for filtered white-light sources). The laser beam is ple in Fig. 2a) is accompanied by a second image, shown
expanded on the animal surface with an optical system in Fig. 2b, which is measured without the fluorescence
of lenses (not shown). Narrow wavelength selection is filter to obtain a “photograph” of the animal for registra-
important, especially in the NIR, whereas the excitation tion purposes at the excitation wavelength. This intrinsic
and emission spectra overlap and it is likely that excita- light image also serves as a calibration measurement
tion photons can propagate into the fluorescent images. since it records the exact spatial distribution of the exci-
The CCD camera is usually a high-sensitivity camera tation light strength for later correction of excitation
since fluorescent signals are of low strength. On the oth- field inhomogeneities. Figure 2 depicts a measurement
er hand, since the targeted measurement is a diffuse-light from a nude mouse with a cathepsin B rich HT1080 fi-
measurement, emanating from a virtually flat surface, brosarcoma, which had been implanted (105 cells) into
 197

Fig. 2a–c Example of typical images involved in FRI. a Fluores-

cence image obtained from the animal. In this example the image
is obtained using a cathepsin-B sensitive fluorescent probe inject-
ed in an animal implanted with a HT1080 carcinoma in the upper
posterior thorax. b Intrinsic light image obtained without the fluo-
rescent band-pass filter. Such images are useful to outline with
greater detail the animal investigated. c Superposition of a onto b
improves the visual result

the posterior upper thorax of the mouse 7–10 days prior

to the experiment, at ~3 mm depth using a cathepsin B
sensitive imaging probe [25]. It is noteworthy that the in-
trinsic light image records the structure on the animal
surface and can yield high-resolution images after appro-
priate focusing, whereas the fluorescence (or lumines-
cence) image records fluorescence signals emanating
from under the surface and is by nature a low-resolution
image that uses mainly the diffuse component of light.
The side-by-side comparison or the superposition of the
two images (c.f. Fig. 2c) into a single image yields a
good visual effect since high resolution is retained.

Applications

Reflectance imaging has been used to image cathespin B

[25], cathepsin D [26] and matrix metalloproteinase 2
(MMP-2) [27] using activatable probes, i.e., appropriate-
ly engineered fluorochromes that are dark in their native
(quenched) state and fluoresce only upon interaction
with specific proteases. A representative mechanism of
activation is shown in Fig. 3. In the non-activated state Fig. 3 An example of an activatable probe. These probes are dark
in their native state, but after enzymatic cleavage of the backbone
fluorochromes are loaded in closed proximity onto a carrier they fluoresce when appropriately excited
graft copolymer consisting of a poly-lysine (PL) back-
bone through appropriate peptides and undergo mutual
energy transfer; thus, they exist in a quenched state. A characteristic example of this technology applied to
At the presence of the targeted enzyme, cleavage of the elucidate different expression levels of tumor MMP-2
peptide link releases the fluorochromes which results activity is shown in Fig. 4 using an MMP-2-positive hu-
in a bright fluorescence signal, which can be detected man HT1080 fibrosarcoma and an MMP-2-negative
in vivo. BT20 mammary adenocarcinoma [28]. Both types of
198

breast cancers activated the near-infrared probe, but the

activation at the HT1080 tumor was markedly higher
than the BT-20 tumor which was barely differentiable
from the background activation. The different levels of
enzyme secretion in the two tumor cell lines used were
further verified by zymographic studies [28] as shown in
Fig. 4c.
Reflectance imaging has also been used for targeting
cell-surface receptors in vivo using peptide near-infrared
dye conjugates [15, 29]. Figure 5 depicts fluorescence
reflectance images using a conjugate of tricarbocyanine
dye and an octapeptide (c.f. Fig. 5a) for targeting the so-
matostatin receptor in CA20948 tumor-bearing rats [29,
30]. The tumor cells were implanted subcutaneously in
the left flank of the rat demonstrating a marked fluores-
cence increase compared with surrounding tissues
(Fig. 5b, c). Similar observations were performed in a
separate study [15] that showed significant fluorescence
increase from tumor sites compared with normal tissues
using the somatostatin analog octreotate coupled to in-
docarbocyanine dyes. Specifically, the fluorescence de-
tected before the injection of the dye-peptide conjugate
administered in a tumor-bearing mouse (RIN38 pancre-
atic tumor expressing the somatostatin receptor subtype
2) and a control mouse are shown in Fig. 6a and b, re-
spectively. Figure 6c and d show the fluorescence imag-
es obtained from the same animals 6 h after probe injec-
tion. The administered probe dose was 0.02 µmol/kg
body weight. The receptor-binding indotricarbocyanine-
octreotate conjugate leads to a significantly elevated tu-
mor signal (Fig. 6c), whereas a control conjugate with no
receptor affinity does not generate contrast (Fig. 6d). The
relative fluorescence intensities of tumor tissue and nor-
mal tissue in the tumors studied with two different dye
conjugates are shown in Fig. 6e and f (indodicarbocya-
nine-octreotate, Fig. 6e; and indotricarbocyanine-octreo-
tate, Fig. 6f) at different times after application. The time
course of tumor-to-normal contrast is depicted for both
compounds in Fig. 6g. Figure 6h shows the results of a
biodistribution analysis of 125I-ITCC-octreotate in
RIN38 tumor-bearing nude mice 24 h post injection. Tis-
sue distribution of this double-labeled peptide derivative
is given as percentage injected dose per gram.
More generally, reflectance imaging can be used to
macroscopically assess endogenous gene products by us-
Fig. 4a–c Near-infrared imaging of a nude mouse implanted with
both an MMP-2-positive human HT1080 fibrosarcoma and an ing fluorescent proteins [16, 31]. Imaging of florescent
MMP-2-negative BT20 mammary adenocarcinoma. a Both tumors proteins (FP) can be done in analogy to near-infrared im-
measured approximately 2–3 mm. b Raw near-infrared image aging with the following exceptions:
shows that the fibrosarcoma generated strong fluorescent signal
intensity 2 h after intravenous injection of an MMP-2-sensitive
activatable probe, but the signal intensity of the BT20 tumor was
1. Absorption/excitation are shifted in, usually to the
only slightly higher compared with the background fluorescence visible light range
of the skin. c Zymographic measurements revealed a significant 2. The higher endogenous tissue absorption in the visi-
difference between MMP-2 secretions by the two tumor lines. ble wavelength yields significantly smaller penetra-
(Adapted from [28]) tion depths
3. No exogenous fluorochromes have to be administered
to visualize FP expression
 199

Fig. 5 a Molecular structure of cytate, a conjugate of tricarbocya- Advantages and limitations

nine dye for near-infrared imaging and an octapeptide for targeting
somatostatin receptor which is up-regulated in neuroendocrine tu-
mors. b Fluorescence intensity imaging of a CA20948 tumor-bear- Reflectance imaging is an ideal tool for high-throughput
ing rat 27 h post administration of cytate via the tail vein. Tumor imaging and screening of surface fluorescent events in
cells were implanted subcutaneously in the left flank of the rat. vivo or in excised tissues. It offers simplicity of opera-
c Fluorescence-intensity image of selected organ parts excised tion and high sensitivity for molecular events that are
24 h post injection of cytate. The somatostatin receptor-positive
tumor, CA20948, preferentially retains cytate. (Adapted from [29]
close to the surface. Typical acquisition times range from
and [30]) a few seconds to a few minutes. For laboratory settings,
multiple animals can be simultaneously imaged. Hard-
ware development and implementation is also straight-
forward and relatively inexpensive. Reflectance systems
In a similar paradigm, bioluminescence imaging [32] is do not use ionizing radiation, employ safe laser powers,
based on emission of visible photons at specific wave- can be made portable and attain small space require-
lengths based on energy-dependent reactions catalyzed ments to be ideal for the laboratory bench.
by luciferases. Luciferase genes have been cloned from a On the other hand, reflectance imaging has funda-
large number of organisms, including bacteria, firefly mental limitations both as a research or clinical tool.
(Photinus pyralis), coral (Renilla), jellyfish (Aequorea), Firstly, the technique attains only small penetration
and dinoflagellates (Gonyaulax). In the firefly, luciferase depths (a few millimeters). This is because the appear-
utilizes energy from ATP to convert its substrate lucife- ance of deeper lesions is significantly blurred and the
rin to oxyluciferin, with the emission of a detectable signal detected from them is significantly attenuated as a
photon. Sensitive imaging systems have been built to de- function of lesion depth while the background noise re-
tect quantitatively a small number of cells or organisms mains constant in reflectance mode since it consists of
expressing luciferase as a transgene [33]. fluorescence signals from superficial layers and intrinsic
light contaminating the fluorescence measurement due to
imperfect filtering. This is in contrast to tomographic
Multi-spectral imaging methods (discussed in the next section) where both sig-
nal and noise reduces so that signal to noise remains sig-
An exciting approach in optical imaging is capitalizing nificant for lesions that are much deeper than a few mil-
on the spectral differentiation of different fluorochromes limeters. Secondly, reflectance imaging is not capable of
in order to perform concurrent imaging of multiple tar- quantification as illustrated in Fig. 8. A small structure
gets [34]. This can be achieved by using appropriate of high fluorochrome concentration that is deeper into
filters in front of the CCD detector that can selectively tissue could yield the same appearance on the surface as
acquire images at different wavelengths. Figure 7 exem- a larger structure of low fluorochrome concentration that
plifies this approach by showing images of GFP and is closer to the surface. This is due to the nature of prop-
cathepsin B expression obtained from a nude mouse agation of diffuse photon density waves into tissue. The
implanted subcutaneously with GFP positive 9L glioma photon count reading at a surface of an animal due to a
tumors on the right side and GFP negative 9L glioma tu- lesion on or under the surface depends on the lesion
mors on the left side [34]. Both tumors were expressing depth, the lesion volume and the optical properties of
cathepsin B, which was imaged with a cathepsin-B sen- both the lesion and the surrounding tissue; therefore, im-
sitive probe [25] loaded with the Cy5.5 fluorochrome ages from different animals, or from the same animal at
(Amersham Pharmacia Biotech, Piscataway, N.J.). The different time points, are generally insufficient for yield-
images were acquired consecutively using appropriate ing quantitative insights.
filters in front of the detector for imaging at the GFP
spectral window (500–550 nm) and at the Cy5.5 dye
channel (690–740 nm) and demonstrate the ability to im-
age separate gene expressions in vivo
200

Fig. 6 In vivo fluorescence

images of tumor-bearing mice
(RIN38 pancreatic tumor ex-
pressing the somatostatin re-
ceptor subtype 2) a, b before
and c, d 6 h after injection of
dye-peptide conjugate at a dose
of 0.02 µmol/kg body weight.
c The receptor-binding
indotricarbocyanine–octreotate
conjugate leads to a significant-
ly elevated tumor signal,
whereas d a control conjugate
with no receptor affinity does
not generate contrast. Relative
fluorescence intensities of tu-
mor tissue and normal tissue
in RIN38/SSTR2 tumors from
nude mice after intravenous
injection e of ITCC-octreotate
or f ITCC-(M2M7) octreotate
at different times after applica-
tion. g The time course of
tumor-to-normal contrast is
depicted for both compounds.
h The results of a biodistribu-
tion analysis of 125I-ITCC-
octreotate in RIN38/SSTR2
tumor-bearing nude mice 24 h
post injection. Tissue distribu-
tion of this double-labeled pep-
tide derivative is given as per-
centage injected dose per gram.
(Image courtesy of K. Licha is
taken from [15])
 201

Fig. 7 Dual-channel imaging

of genetically identically 9L
glioma tumors except for GFP
expression (positive on ani-
mal’s right flank, negative on
animal’s left flank). Both tu-
mors are expressing cathepsin
B. From left to right: white-
light image, GFP image, and
cathepsin B obtained virtually
simultaneously using spectrally
separated fluorescence reflec-
tance images. (Image courtesy
of U. Mahmood adapted from
[34])

significantly attenuated and the resolution impaired. Dif-

fuse optical tomography has been used for quantitative
imaging of absorption and scattering [41, 42, 43, 44, 45,
46] as well as concentration and fluorochrome lifetime
measurements with simulated or phantom measurements
[47, 48, 49, 50, 51, 52]. Recently, DOT has been applied
clinically for imaging tissue oxy- and deoxy-hemoglobin
concentration and blood saturation [53, 54, 55] and con-
trast agent uptake in absorption mode [56].
Fig. 8 Different strength and size fluorescent lesions embedded in A particular development that facilitates in vivo mo-
a diffuse medium at different lengths can have the same appear- lecular interrogations of tissue is the use of fluorescence
ance on the surface impeding quantification using reflectance im- molecular tomography (FMT) [20]. The method produc-
aging es quantified three-dimensional reconstructions of fluo-
rescence activation or concentration using measurements
of fluorescent molecular probes at both the emission and
Fluorescence-mediated molecular tomography excitation wavelengths. The theoretical mainframe has
been described elsewhere [52]. The main advantage of
In order to resolve and quantify fluorochromes deep in using both intrinsic and fluorescence contrast is that no
tissue tomographic approaches are necessary. The gener- absolute photon-field measurements are required, yet ab-
al framework of reconstruction techniques using diffuse solute fluorochrome concentrations can be reconstructed.
light has been developed during the last decade where Furthermore, the method does not require any measure-
rigorous mathematical modeling of light propagation in ments obtained before contrast agent administration,
tissue, combined with technological advancements in which is a crucial parameter for in vivo imaging of
photon sources and detection techniques has made possi- probes that require long circulation times to achieve suf-
ble the application of tomographic principles [35, 36, 37, ficient accumulation in their intended targets.
38, 39, 40] for optical imaging. The technique, generally
termed diffuse optical tomography (DOT), uses multiple
projections and measures light around the boundary of FMT imaging systems
the illuminated body. It then effectively combines all
measurements into an inversion scheme that takes into In order to perform fluorescence tomography, tissue has
account the highly scattered photon propagation to de- to be illuminated at different projections and multiple
convolve the effect of tissue on the propagating wave, measurements should be collected from the boundary of
even though high frequency components are generally the tissue of investigation. FMT, in its simplest imple-
202

Fig. 9a–c Example of an FMT

scanner. a The scanner uses
multiple sources and detectors
to effectively collect light mea-
surement at multiple projec-
tions around the animal’s body.
For details see text. b Photo-
graph of the optical bore used
for animal investigations.
The source fibers (blue) and
detection fiber bundles (black)
are shown installed. c Photo-
graph of the fiber bundle array
arranged for imaging by the
CCD camera

mentation, needs only measurements of fluorescence and ing fluid” (also shown in Fig. 9b contained in the bore),
intrinsic light using constant-wave sources in order to re- such as a water solution of a scatterer (e.g., TiO2 parti-
solve the single quantity targeted, namely fluorochrome cles) and an absorber (India Ink or similar) that matches
concentration. This is in contrast to DOT which, in its the optical properties of the tissue investigated. The
general form, requires decomposition of absorption from matching fluid serves virtually the same function as gel
scattering and therefore requires more elaborate photon to ultrasound. Fiber bundles (v) can be used to collect
technologies such as intensity-modulated sources or photons through the turbid medium and direct them onto
short photon pulses and appropriate detection systems. the CCD camera (viii) either by direct fiber coupling or
The ability of FMT to operate with constant wave sys- by an appropriate positioning arrangement (vi) so that
tems allows for a practical and relatively inexpensive the fiber output can be imaged via a lens system as de-
implementation of multiple detection channels using picted in Fig. 9c. Appropriate filters (vii) are necessary
CCD cameras. In CW mode, CCD cameras offer very to reject background ambient light and intrinsic or fluo-
high sensitivity and low noise levels. Nevertheless, the rescence signals according to the measurement per-
use of advanced photon sources and detection systems formed. A reference measurement could also be intro-
can enhance tomographic performance by potentially im- duced to account for temporal variations in laser intensi-
proving resolution and resolving lifetime as well [48, ty. The use of an optical bore and matching fluid is a
49]. convenient approach for matching photons onto tissue
A CCD-based FMT system example is shown in but by no means limiting. Different schemes, including
Fig. 9a. The light source (i) is generally a laser diode at direct fiber positioning on the tissue surface, at arbitrary
the appropriate wavelength to excite the targeted fluoro- geometries, including geometries, that could be used by
chrome. Multiple laser diodes at different wavelengths endoscopic probes could be devised.
could be combined in the same system (either by time
sharing or spectral separation) to excite multiple fluoro-
chromes simultaneously. The light from the laser diode In vivo imaging
can be directed to an optical switch (ii) for time sharing
one input to many outputs and directed with optical fi- The FMT has been used recently for imaging of cathep-
bers (iii) at different points around the body of investiga- sin B activity within deep structures. Figure 10 shows a
tion or a specially designed “optical bore” (iv). Our im- representative experiment of the results obtained from
plementation of an optical bore [57] is shown in Fig. 9b. 9L gliosarcomas stereotactically implanted into unilater-
The optical bore contains the body of examination simi- al brain hemispheres of nude mice, as cathespin B activi-
lar to a CT or MR scanner. Typically, such an implemen- ty had been implicated in glioma invasion [58, 59, 60].
tation would require the animal immersed into a “match- Correlative MR imaging was performed to determine the
 203

Fig. 10a–i In-vivo FMT of cathepsin B expression levels in 9L FMT orientation and animal positioning. The in vivo im-
gliosarcomas stereotactically implanted into unilateral brain hemi- aging data correlated well with surface-weighted reflec-
spheres of nude mice. a, b Axial and sagittal MR slices of an ani-
mal implanted with a tumor, which is shown in green after gado- tance imaging of the excised brain. Figure 10g depicts
linium enhancement. c, e, f, Consecutive FMT slices obtained the axial brain section through the 9L tumor examined
from top to bottom from the volume of interest shown in b by thin with white light using a CCD camera mounted onto a
white horizontal lines. d Superposition of the MR axial slice pass- dissecting microscope, Fig. 10h shows the same brain
ing through the tumor a onto the corresponding FMT slice c after
appropriately translating the MR image to the actual dimensions
section imaged using a previously developed reflectance
of the FMT image. g, h Axial brain section through the 9L tumor imaging system [14] at the excitation wavelength
imaged with white light and with monochromatic light at the exci- (675 nm), and Fig. 10i is the fluorescence image ob-
tation wavelength (675 nm), respectively, and i fluorescence im- tained at the emission light wavelength using appropriate
age of the same axial brain section demonstrating a marked fluo- three-cavity cut-off filters that demonstrates a marked
rescent probe activation, congruent with the tumor position identi-
fied by gadolinium-enhanced MRI and FMT. (Reproduced from fluorescent probe activation, congruent with the tumor
[20]) position identified by gadolinium-enhanced MRI and
FMT. The above results confirm that cathepsin B can be
used as an imaging marker [25] since the protease is pro-
presence and location of tumors prior to the FMT imag- duced in considerable amounts by tumor cells and by re-
ing studies. Figure 10a and b depicts the gadolinium- cruited host cells [61]. Cathepsin B expression in the tu-
enhanced tumor (enhancement is shown in a green color mors was further confirmed by immunohistochemistry,
map superimposed onto a T1-weighted image) on axial Western blotting and RT-PCR.
(Fig. 10a) and sagittal (Fig. 10b) slices. Figure 10c, e,
and f depicts the three consecutive FMT slices obtained
from top to bottom of the volume of interest. The loca- Quantification: resolution and coregistration
tion and volume covered by the three slices is indicated
in Fig. 10b by thin white horizontal lines. Figure 10c The feasibility to three-dimensionally reconstruct and
shows marked local probe activation relative to adjacent quantify fluorochromes embedded in diffuse media has
slices, congruent with the location of the tumor identi- been demonstrated in the past [47, 48, 49, 52]. It has
fied on the MR images. Figure 10d shows a superposi- been further shown that FMT can yield linear responses
tion of the MR axial slice passing through the tumor over a wide area of physiologically relevant fluoro-
(Fig. 10a) onto the corresponding FMT slice (Fig. 10c) chrome concentrations [57]. Figure 11 depicts recon-
after appropriately translating the MR image to the actu- structions obtained for the range 1–800 nM concentra-
al dimensions of the FMT image. For coregistration pur- tion of the Cy5.5 fluorochrome obtained with a system
poses we used special water-containing fiducials and similar to the one shown in Fig. 9. The geometry em-
body marks that facilitated matching of the MR and ployed is shown in Fig. 11a. A 2-mm radius tube was
204

immersed into the optical bore (shown in Fig. 9) and

contained various concentrations of fluorochromes. A
typical reconstructed image is seen in Fig. 11b. Fig-
ure 11c plots the reconstructed fluorochrome concentra-
tion as a function of the expected concentration. The
reconstructed value demonstrates a remarkably linear re-
sponse for the whole range of examined concentrations
[57].
The resolution limits of diffuse light imaging tech-
niques have been studied theoretically, but there is limit-
ed experimental demonstration, mainly due to the lack of
high source-detection efficiency systems. Increasing the
number of sources and detectors improves resolution not
only due to the higher spatial sampling but also due to
the improvements in signal-to-noise ratio achieved in in-
creased data sets [62]. Naturally, improvements of
source and detector density do not scale linearly with
resolution improvements. It is expected that the resolu-
tion for small animal imaging would be of the order of
1–2 mm, whereas for larger tissues it would approach
4–5 mm [62].
The FMT can be combined with high-resolution “re-
flectance” imaging in superimposing surface architec-
tural features and markers on the underlying three-
dimensional reconstruction to correlate the tomographic
results with anatomical features, especially in animal
imaging. Such an example is shown in Fig. 12.
Figure 12a depicts a merged axial FMT/MRI image ob-
tained from the nude mouse shown in Fig. 2. Figure 12b
depicts the sagittal MRI image of the same animal and
the three FMT slices imaged. Figure 12c depicts a
three-dimensional rendering of the superposition of the
three-dimensional FMT images with the two dimen-
sional intrinsic light image (shown in Fig. 2b) after ap-
plication of appropriate thresholds. Furthermore, the
combination of FMT with high-resolution imaging
methods that primarily target structure, such as X-ray
CT or MRI, could yield a superior hybrid imaging mo-
dality in which highly sensitive molecular information
and high resolution is achieved. Such an approach is
feasible due to the optical component compatibility
with most other medical imaging modalities and has
been recently applied to performing clinical investiga-
tions of the breast using a combined MRI and DOT
system [19].
Fig. 11a–c Reconstruction linearity for varying fluorochrome
concentrations. a Experimental phantom used for the demonstra-
tion. The small circle represents a tube of 100 µl volume contain-
ing different concentrations of Cy5.5 dye. b Typical image recon- Clinical feasibility of fluorescence imaging
structed. c Reconstructed vs real fluorochrome concentration. The
linearity achieved spans more than two orders of magnitude of bi- Clinical fluorescent imaging would require detection of
ologically relevant fluorochrome concentrations. (Adapted from
[57]) fluorescent signals that have propagated for several cen-
timeters into tissue. Figure 13a demonstrates the photon
attenuation rate of different human tissues. In particular,
it is shown that breast tissue and the adult lung attenu-
ates NIR light one order of magnitude every 2.5 cm, the
attenuation rate in denser breast approaches one order
 205

Fig. 12a–c An example of

merging two-dimensional with
three-dimensional information
for improving visualization.
a The FMT image of the mouse
shown in Fig. 2. b An MRI
sagittal image. The three-plane
FMT images are contained
between the white horizontal
lines. The arrow indicates
the position of the tumor.
c Rendering of the FMT volu-
metric reconstruction superim-
posed with the two-dimension-
al intrinsic light image after
application of appropriate
thresholds

of magnitude every 1.5 cm and the attenuation rate in

brain and muscle is approximately one order of magni-
tude every 1.0 cm. In signal strength terms Fig. 13 plots
fluorescence counts per second of exposure and square
millimeter of of CCD detector area, as a function of or-
gan diameter. This calculation is for a tumor-like lesion
of 100 µl volume and 100 nM of Cy5.5 concentration
(10 pico-moles) located near the center of the various
organs simulated. The simulation is based on experi-
mental measurements obtained from a diffuse volume
(similar to the one shown on Fig. 11a but with varying
outer diameter) using 30 mW of illuminating power
[22].
There are also three horizontal lines plotted, marking
the 10-, 20-, and 30-dB signal-to-noise ratio (SNR)
achieved assuming shot-noise limited detection. In reali-
ty, background fluorescence (primarily due to non-spe-
cific dye distribution) limits the SNR achieved, since
background signals can be seen as “biologically induced

Fig. 13 Average fluorescent photon counts expected at the periph-

ery of different organs due to a fluorochrome located 0.5 cm off
center as a function of organ diameter. Three signal-to-noise levels
for shot-noise limited detection are also plotted. (From [22])
206

noise” after appropriate subtraction schemes [63]. Data proton density, relaxation times, absorption, scattering)
set optimization [62] and appropriate algorithms [64] can and/or occasionally physiological parameters as the main
be used to image at low SNR and such considerations source of image signal. We describe herein a new set of
should be addressed on a case-to-case basis. This study, technologies that could allow imaging of specific molec-
however, demonstrates that fluorescence imaging of hu- ular markers in vivo. This has been enabled by several
man tissues becomes an issue of uptake-to-background developments:
contrast rather than of propagation feasibility. Important
1. The development of “smart” and targeted fluorescent
to the latter argument is that fluorescence offers mecha-
molecular probes developed for an increasing number
nisms for multifold background suppression by using
of targets identified by gene-array profiles
quenching and activation (dequenching) of fluorescent
2. Miniaturization and development of highly sensitive
probes [25, 65] or fluorescence resonance energy trans-
imaging equipment that allows recording of fluores-
fer [66]. These technologies significantly minimize
cent signals at pico- to femto-mole accumulations
background signals in vivo and, combined with advances
3. The ability to produce tomographic images of near-
in drug delivery and hardware improvements, could sur-
infrared probe distribution in whole bodies
pass the predictions of this study.
The technology has been currently applied to imaging
small animals, but it could be expanded to clinical use as
Conclusion well, due to the penetration of near-infrared fluorescent
signals for several centimeters in tissues. Furthermore,
Despite advances in medical imaging technologies over the optical technology is relatively inexpensive, can be
the past two decades, we are currently still limited in our made portable, and uses non-ionizing radiation and sta-
ability to detect tumors or other diseases in their earliest ble molecular markers. These features can allow easy
stage of formation, phenotype tumors during complex laboratory and bench-top use and enable monitoring of
cycles of growth, invasion and metastases, to use imag- molecular events repeatedly and over time. The combi-
ing techniques to speed up drug testing, and to use imag- nation of molecular beacons and optical imaging tech-
ing as an objective endpoint for tailoring therapies in a nologies is expected to play a fundamental role in bio-
given individual. Similar limitations also exist in neuro- medicine in the next decade and continued developments
degenerative, cardiovascular, and immunologic diseases. in probe design and imaging technologies will undoubt-
Traditional cross-sectional imaging techniques, such as edly further expand current capabilities.
MR, CT, and ultrasound, primarily rely on physical (e.g.,

References
1. Chance B (1991) Optical method. 6. Gonzalez S, Rajadhyaksha M, 12. Monsky WL, Fukumura D, Gohongi T
Annu Rev Biophys Biophys Chem Rubinstein G et al. (1999) Character- et al. (1999) Augmentation of transvas-
20:1–28 ization of psoriasis in vivo by cular transport of macromolecules and
2. Weissleder R (2001) A clearer vision reflectance confocal microscopy. nanoparticles in tumors using vascular
for in vivo imaging. Nat Biotechnol J Med 30:337–536 endothelial growth factor. Cancer Res
19:316–317 7. Konig K (2000) Multiphoton micros- 59:4129–4135
3. Ntziachristos V, Chance B (2001) copy in life sciences. J Microsc 13. Toomre D, Manstein DJ (2001) Light-
Probing physiology and molecular 200:83–104 ing up the cell surface with evanescent
function using optical imaging: 8. Buehler C, Kim KH, Dong CY et al. wave microscopy. Trends Cell Biol
applications to breast cancer. (1999) Innovations in two-photon deep 11:298–303
Breast Cancer Res 3:41–46 tissue microscopy. IEEE Eng Med Biol 14. Mahmood U, Tung CH, Bogdanov A et
4. Korlach J, Schwille P, Webb WW et al. Mag 18:23–30 al. (1999) Near infrared optical imag-
(1999) Characterization of lipid bilayer 9. So PT, Dong CY, Masters BR et al. ing system to detect tumor protease ac-
phases by confocal microscopy and (2000) Two-photon excitation fluores- tivity. Radiology 213:866–870
fluorescence correlation spectroscopy. cence microscopy. Annu Rev Biomed 15. Becker A, Hessenius C, Licha K et al.
Proc Natl Acad Sci 96:8461–8466 Eng 2:399–429 (2001) Receptor-targeted optical imag-
5. Rajadhyaksha M, Grossman M, 10. Masters BR, So PT, Gratton E (1997) ing of tumors with near-infrared
Esterowitz D et al. (1995) In vivo Multiphoton excitation fluorescence fluorescent ligands. Nat Biotechnol
confocal scanning laser microscopy of microscopy and spectroscopy of in 19:327–331
human skin: melanin provides strong vivo human skin. Biophys J 16. Yang M, Baranov E, Jiang P et al.
contrast. J Invest Dermatol 72:2405–2412 (2000) Whole-body optical imaging of
104:946–952 11. Dellian M, Yuan F, Trubetskoy VS et green fluorescent protein-expressing
al. (2000) Vascular permeability in a tumors and metastases. Proc Natl Acad
human tumour xenograft: molecular Sci USA 97:1206–1211
charge dependence. Br J Cancer
82:1513–1518
 207

17. Zaheer A, Lenkinski RE, Mahmood A 30. Bugaj JE, Achilefu S, Dorshow RB et 45. Pogue BW, Testorf M, McBride T et al.
et al. (2001) In vivo near-infrared fluo- al. (2001) Novel fluorescent contrast (1997) Instrumentation and design of a
rescence imaging of osteoblastic activi- agents for optical imaging of in vivo frequency domain diffuse optical to-
ty. Nat Biotechnol 19:1148–1154 tumors based on a receptor-targeted mography imager for breast cancer de-
18. Reynolds JS, Troy TL, Mayer RH et al. dye-peptide conjugate platform. tection. Opt Expr 1:391–403
(1999) Imaging of spontaneous canine J Biomed Opt 6:122–133 46. Schmitz CH, Graber HL, Barbour RL
mammary tumors using fluorescent 31. Wouters FS, Verveer PJ, Bastiaens PI (2000) A fast versatile instrument
contrast agents. Photochem Photobiol (2001) Imaging biochemistry inside for dynamic optical tomography.
70:87–94 cells. Trends Cell Biol 11:203–211 In: Advances in optical imaging and
19. Ntziachristos V, Yodh AG, Schnall M 32. Contag CH, Spilman SD, Contag PR et photon migration, OSA. Technical
et al. (2000) Concurrent MRI and dif- al. (1997) Visualizing gene expression Digest. Optical Society of America,
fuse optical tomography of breast after in living mammals using a biolumines- Washington DC, pp 94–96
indocyanine green enhancement. Proc cent reporter. Photochem Photobiol 47. Chang JH, Graber HL, Barbour RL
Natl Acad Sci USA 97:2767–2772 66:523–531 (1997) Imaging of fluorescence in
20. Ntziachristos V, Tung C, Bremer C et 33. Sweeney TJ, Mailander V, Tucker AA highly scattering media. IEEE Trans
al. (2002) Fluorescence-mediated to- et al. (1999) Visualizing the kinetics of Biomed Eng 44:810–822
mography resolves protease activity in tumor-cell clearance in living animals. 48. Oleary MA, Boas DA, Li XD et al.
vivo. Nat Med 8:757–760 Proc Natl Acad Sci 96:12044–12049 (1996) Fluorescence lifetime imaging
21. Chance B, Kang K, He L et al. (1993) 34. Mahmood U, Tung CH, Tang Y et al. in turbid media. Opt Lett 21:158–160
Highly sensitive object location in tis- (in press) Feasibility of in-vivo 49. Paithankar DY, Chen AU, Pogue BW
sue models with linear in-phase and multi-channel optical imaging of gene et al. (1997) Imaging of fluorescent
anti-phase multi-element optical arrays expression: an experimental study yield and lifetime from multiply scat-
in one and two dimensions. Proc Natl in mice. Radiology tered light reemitted from random me-
Acad Sci USA 90:3423–3427 35. Schotland J-C (1997) Continuous wave dia. Appl Opt 36:2260–2272
22. Ntziachristos V, Ripoll J, Weissleder R diffusion imaging. J Opt Soc Am 50. Chernomordik V, Hattery D, Gannot I
(2002) Would near-infrared fluores- 14:275–279 et al. (1999) Inverse method 3-D re-
cence signals propagate through large 36. Yodh AG, Chance B (1995) Spectros- construction of localized in vivo fluo-
human organs for clinical studies. copy and imaging with diffusing light. rescence: application to Sjogren syn-
Opt Lett 27:527–529 Phys Today 48:34–40 drome. IEEE J Selected Top Quantum
23. Contag CH, Jenkins D, Contag PR et 37. Hielscher AH, Klose AD, Hanson KM Electron 5:930–935
al. (2000) Use of reporter genes for op- (1999) Gradient-based iterative image 51. Sevick-Muraca EM, Lopez G,
tical measurements of neoplastic dis- reconstruction scheme for time-re- Reynolds JS et al. (1997) Fluorescence
ease in vivo. Neoplasia 2:41–52 solved optical tomography. IEEE Trans and absorption contrast mechanisms
24. Ramanujam N, Chen J, Gossage K et Med Imaging 18:262–271 for biomedical optical imaging using
al. (2001) Fast and noninvasive fluo- 38. Barbour RL, Graber HL, Chang JW et frequency-domain techniques. Photo-
rescence imaging of biological tissues al. (1995) MRI-guided optical tomog- chem Photobiol 66:55–64
in vivo using a flying-spot scanner. raphy: prospects and computation for a 52. Ntziachristos V, Weissleder R (2001)
IEEE Trans Biomed Eng new imaging method. IEEE Comput Experimental three-dimensional fluo-
48:1034–1041 Sci Eng 2:63–77 rescence reconstruction of diffuse me-
25. Weissleder R, Tung CH, Mahmood U 39. Colak SB, van der Mark MB, Hooft dia using a normalized Born approxi-
et al. (1999) In vivo imaging of tumors GW et al. (1999) Clinical optical to- mation. Opt Lett 26:893–895
with protease-activated near-infrared mography and NIR spectroscopy for 53. Pogue BW, Poplack SP, McBride TO et
fluorescent probes. Nat Biotech breast cancer detectio. IEEE J Selected al. (2000) Hemoglobin imaging of
17:375–378 Top Quantum Electron 5:1143–1158 breast tumors with near-infrared to-
26. Tung CH, Mahmood U, Bredow S et 40. Arridge SR (1999) Optical tomography mography. Radiology 214:G05H
al. (2000) In vivo imaging of proteolyt- in medical imaging. Inverse Prob 54. Hillman EMC, Hebden JC, Schweiger
ic enzyme activity using a novel 15:R41–R93 M et al. (2001) Time resolved optical
molecular reporter. Cancer Res 41. Oleary MA, Boas DA, Chance B et al. tomography of the human forearm.
60:4953–4958 (1995) Experimental images of hetero- Phys Med Biol 46:1117–1130
27. Bremer C, Tung C, Weissleder R geneous turbid media by frequency- 55. Benaron DA, Hintz SR, Villringer A et
(2001) Imaging of metalloproteinase2 domain diffusing-photon tomography. al. (2000) Noninvasive functional im-
inhibition in vivo. Nat Med 7:743–748 Opt Lett 20:426–428 aging of human brain using light.
28. Bremer C, Bredow S, Mahmood U et 42. Jiang HB, Paulsen KD, Osterberg UL J Cereb Blood Flow Metab 20:469–477
al. (2001) Optical imaging of matrix et al. (1998) Improved continuous light 56. Ntziachristos V (2000) Concurrent dif-
metalloproteinase-2 activity in tumors: diffusion imaging in single- and multi- fuse optical tomography, spectroscopy
feasibility study in a mouse model. target tissue-like phantoms. Phys Med and magnetic resonance of breast can-
Radiology 221:523–529 Biol 43:675–693 cer. University of Pennsylvania
29. Achilefu S, Dorshow RB, Bugaj JE et 43. Chang JW, Graber HL, Koo PC et al. 57. Ntziachristos V, Weissleder R (2002)
al. (2000) Novel receptor-targeted fluo- (1997) Optical imaging of anatomical CCD-based scanner for three-dimen-
rescent contrast agents for in vivo tu- maps derived from magnetic resonance sional fluorescence-mediated diffuse
mor imaging [in process citation]. images using time-independent optical optical tomography of small animals.
Invest Radiol 35:479–485 sources. IEEE Trans Med Imaging Med Phys 29:803–809
16:68–77
44. Ntziachristos V, Ma XH, Chance B
(1998) Time-correlated single photon
counting imager for simultaneous
magnetic resonance and near-infrared
mammography. Rev Sci Instrum
69:4221–4233
208

58. Demchik LL, Sameni M, Nelson K et 61. Koblinski JE, Ahram M, Sloane BF 64. Eppstein MJ, Dougherty DE, Troy TL
al. (1999) Cathepsin B and glioma in- (2000) Unraveling the role of proteases et al. (1999) Biomedical optical tomog-
vasion. Int J Dev Neurosci 17:483–494 in cancer. Clin Chim Acta raphy using dynamic parameterization
59. Yan S, Sameni M, Sloane BF (1998) 291:113–135 and Bayesian conditioning on photon
Cathepsin B and human tumor progres- 62. Culver J, Ntziachristos V, Holboke M migration measurements. Appl Opt
sion. Biol Chem 379:113–123 et al. (2001) Optimization of optode 38:2138–2150
60. Rempel SA, Rosenblum ML, arrangements for diffuse optical 65. Tyagi S, Kramer FR (1996) Molecular
Mikkelsen T et al. (1994) Cathepsin B tomography: a singular value analysis. beacons: probes that fluoresce upon
expression and localization in glioma Opt Lett 26:701–703 hybridization. Nat Biotechnol
progression and invasion. Cancer Res 63. Wu J, Perelman L, Dasari RR et al. 14:303–308
54:6027–6031 (1997) Fluorescence tomographic im- 66. Tyagi S, Marras SAE, Kramer FR
aging in turbid media using early-arriv- (2000) Wavelength-shifting molecular
ing photons and Laplace transforms. beacons. Nat Biotechnol 18:1191–1196
Proc Natl Acad Sci USA
94:8783–8788