effects of background fluorescence in fluorescence molecular tomography

Effects of background fluorescence in fluorescencemolecular tomography

Melisa Gao, George Lewis, Gordon M. Turner, Antoine Soubret, and Vasilis Ntziachristos

Recent advances in optical imaging systems and systemically administered fluorescent probes havesignificantly improved the ways by which we can visualize proteomics in vivo. A key component in thedesign of fluorescent probes is a favorable biodistribution, i.e., localization only in the targeted diseasedtissue, in order to achieve high contrast and good detection characteristics. In practice, however, there isalways some level of background fluorescence present that could result in distorted or obscured visual-ization and quantification of measured signals. In this study we observe the effects of backgroundfluorescence in tomographic imaging. We demonstrate that increasing levels of background fluorescenceresult in artifacts when using a linear perturbation algorithm, along with a significant loss of imagefidelity and quantification accuracy. To correct for effects of background fluorescence, we have appliedcubic polynomial fits to bulk raw measurements obtained from spatially homogeneous and heterogeneousphantoms. We show that subtraction of the average fluorescence response from the raw data beforereconstruction can improve image quality and quantification accuracy as shown in relatively homoge-neous or heterogeneous phantoms. Subtraction methods thus appear to be a promising route for adap-tively correcting nonspecific background fluorochrome distribution. © 2005 Optical Society of America

OCIS codes: 170.6960, 170.0110, 260.2510.

1. Introduction

Fluorescence molecular tomography (FMT) is a recentoptical imaging technique that can visualize and quan-tify the function of genes and the expression ofenzymes and proteins deep in tissues by using fluoro-chromes with molecular specificity.1 The method isbased on a generic theoretical mainframe developedfor optical investigations of diffuse media by severalinvestigators2–11 and in this particular implementa-tion uses measurements at both excitation and emis-

sion wavelengths to provide robust imaging per-formance even in highly heterogeneous media, such assmall animals.12 FMT can offer a significant imagingtool not only for small animal imaging research andsubsequent drug discovery but also in clinical applica-tions. Tomographic imaging with diffuse fluorescencemeasurements has been demonstrated in the pastwith simulated data and experimental measurementsfrom phantoms.13,14 In vivo capability was also re-cently shown by in vivo protease imaging of brain tu-mors.1 Furthermore, FMT has been found to be moreaccurate than planar imaging methods and has beenused to resolve tumor treatment.15

A key issue that allows for a bright outlook for thistechnology is the increasing availability of advancedfluorescence probes that can be systemically adminis-tered and subsequently localized in certain diseasedtissues to report on specific molecular activity.16 Twogeneric classes have been developed so far, i.e., tar-geted and activatable fluorescent probes.17 The firstclass encompasses probes that are engineered to bindwith high affinity to cellular receptors or other pro-teins. The nonbound probe is cleared from the bloodstream through the hepatobilliary or urinary tract.The second class uses probes that are engineered to bedark in their base state and can fluoresce only afterinteraction with specific enzymes. Both technologiesaim to minimize the effects of fluorescence from back-

M. Gao is with the Harvard–MIT Division of Health Sciencesand Technology Biomedical Optics Summer Institute, 77 Massa-chusetts Avenue, E25-519, Cambridge, Massachusetts 02139, andPrinceton University, Princeton, New Jersey. G. Lewis is with theHarvard–MIT Division of Health Sciences and Technology Bio-medical Optics Summer Institute, and the University of Miami,Coral Gables, Florida. G. M. Turner, A. Soubret, and V. Ntzia-christos ([email protected]) are with the Harvard–MIT Division of Health Sciences and Technology BiomedicalOptics Summer Institute, and the Laboratory for Bio-Optics andMolecular Imaging, Center for Molecular Imaging Research, Mas-sachusetts General Hospital and Harvard University, CNY149,13th Street, 5209, Charlestown, Massachusetts 02129.

Received 11 November 2004; revised manuscript received 28April 2005; accepted 5 May 2005.

0003-6935/05/265468-07$15.00/0© 2005 Optical Society of America

5468 APPLIED OPTICS � Vol. 44, No. 26 � 10 September 2005

ground tissues by increasing the tumor-to-backgroundratio, hence allowing detection of diseased tissue withhigh sensitivity and specificity.

Regardless of the exact technology employed foreliminating nonspecific background fluorescencesignals, there is generally some residual signal invivo due to either imperfect clearance or the pres-ence of low concentrations of background targetsthat bind or activate the administered probe. In thisstudy, we seek to characterize the effect of back-ground fluorescence on the linear tomographic im-aging performance and quantification. To drawconclusions from well-characterized experiments,we base our observations on phantom measure-ments that simulate small animal imaging. We firstinvestigate the generic characteristics of back-ground fluorescence measured from relatively ho-mogeneous phantoms. Then, we generalize theseobservations with phantoms with higher back-ground spatial heterogeneity, which better reflectsthe in vivo case. We further examine methodologiesthat can potentially correct for the effects of back-ground fluorescence and improve imaging per-formed in the presence of background nonspecificfluorescence. The ultimate goal of this study is tosuggest an automated method for correcting for thepresence of background fluorescence in tomo-graphic reconstructions by using data preprocess-ing.

Section 2 describes the materials and methodsused. Section 3 summarizes the most important find-ings. Finally, in Section 4 we discuss our major find-ings and alternative methods for improving imagingperformance and future directions.

2. Materials and Methods

A. Imaging Setup

The experiments in this study used a previously re-ported imaging system,14 capable of performing (i)planar imaging with front illumination, also knownas fluorescence reflectance imaging (FRI), and (ii) to-mographic imaging, i.e., FMT. The system is shownin Fig. 1. Light from a 670 nm diode laser (B&WTEK, Inc., Newark, Delaware) was routed eitherto a reflectance branch for FRI or to a multichanneloptical switch (DiCon Fiberoptics, Inc., Richmond,California) for FMT in which the imaging chamber istransilluminated by 46 time-shared fibers. Thesesource fibers are equidistantly arranged over an areaof 1.8 cm � 1.5 cm on the movable back plate of theimaging chamber, each fiber is placed 3 mm fromits adjacent fibers. The front window is made from anantireflection-coated glass for the near infrared(Edmund Scientific, Tonawanda, New York). Reflec-tance or transillumination images were capturedthrough the glass window with a CCD camera(VersArray 512B, Roper Scientific, Trenton, New Jer-sey). A three-cavity bandpass filter with a center fre-quency at 670 � 10 nm combined with neutral-density filters with a total attenuation of 1.5 opticaldensity were used to collect the intrinsic measure-

ments. Fluorescence measurements were collectedwith a three-cavity bandpass filter centered at 710� 10 nm and a long-pass filter with a cutoff wave-length of 695 nm.

B. Phantoms

1. Homogeneous PhantomA hollow sphere of 5 mm diameter was developed tosimulate a tumor. The sphere was created by heatinga 100 �m plastic vinyl sheet and using a vacuum tocause it to collapse around a ball bearing, generatinga hemisphere. Two halves were glued together, andthe sphere was sandblasted to make it more diffusive.To simulate optical tissue properties, the hollowsphere phantom was filled with a solution of 1% In-tralipid, 50 parts in 106 (ppm) black India ink, andCy5.5 fluorescent dye (Amersham Biosciences, Pisca-taway, New Jersey; excitation peak of 670 nm, emis-sion peak of 710 nm) at a concentration of 200 nM.The sphere was suspended from an adjustable armand immersed in a parallel-plate imaging chamberset to a thickness of 13 mm. The chamber was filledwith 1% Intralipid and 50 ppm black India ink, re-sulting in optical properties similar to the aver-age optical properties of mice (approximately �a

� 0.3 cm�1 and ��s � 10 cm�1). Cy5.5 fluorochromewas further added to the background fluid at concen-trations of 0, 30, 60, and 90 nM to yield backgroundfluorescence. For each background fluorescence level,images were taken with the sphere at the front and atthe middle of the chamber to represent tumors at twodifferent tissue depths as shown in Fig. 2. The corre-sponding distances of the sphere center to the glasswindow were 2.5 and 6.5 mm, respectively. Eight ex-periments were performed in total with the homoge-neous phantoms.

Fig. 1. Schematic of a FMT imaging setup. Light from a 670 nmlaser diode (iii) is routed with a two-channel optical switch (iv) toeither a single source for reflectance imaging (v) or a multichanneloptical switch for FMT (viii). A beam splitter (vi) redirects a smallportion of that light to the front of the chamber to serve as a laserpower reference. The source fibers transilluminate the imagingchamber (i), where the phantom is suspended from an adjustablearm. Light emitted from the photon is imaged through the frontwindow of the chamber (ii) by a CCD camera (ix) with appropriatebandpass filters (x).

10 September 2005 � Vol. 44, No. 26 � APPLIED OPTICS 5469

2. Heterogeneous PhantomTo more closely match in vivo imaging conditions, wefurther created a heterogeneous phantom by sur-rounding the hollow sphere described in the previousparagraph by four diffusive tubes of 3 mm diameter,containing 1% Intralipid, 50 ppm black India ink,and varying levels of Cy5.5 fluorochrome. The tubeswere sealed at both ends with hot glue. The exactconfiguration of the tubes is shown in Fig. 2(c), andthe fluorescent concentrations used are listed inTable 1. This construct of tubes was immersed in theimaging chamber and was surrounded by a fluid con-sisting of 1% Intralipid, 50 ppm black India ink, and30 nm of Cy5.5 dye concentration to establish back-ground scattering, absorption, and fluorescence prop-erties similar to those found in mice in vivo.

C. Data Collection and Processing

For spatial coregistration purposes, front-illuminationplanar (reflectance) images of the phantoms were ob-tained at both the fluorescent excitation and emissionwavelengths in the absence of surrounding Intralipidsolution. Transillumination images for each of the 46source fibers were then obtained at the emission andexcitation wavelengths. The data were combined toform the normalized Born ratio, which is the fluores-cent measurements divided by measurements at theexcitation wavelength after bleed-through signal sub-traction,12 i.e.,

U nB�rs, rd� �Ufl�rs, rd� � �f Uinc�rs, rd�

Uinc�rs, rd�, (1)

where Ufl�rs, rd� and Uinc�rs, rd� are the photon fieldsmeasured at the emission and excitation wave-lengths, respectively, at location rd for a source atlocation rs. �f is a coefficient that corrects for residualtransmittance of the excitation field through the flu-orescence filter. This normalized Born field serves asthe input to a linear forward scheme based on theformation of a weight matrix as described in Ref. 12and using the algebraic reconstruction technique fordata inversion. The reconstruction mesh selected forall the inversions in this study was 19 � 19 � 21voxel elements over a 22 mm � 22 mm � 13 mm�17 mm� volume for the homogeneous (heteroge-neous) backgrounds.

Two image reconstruction schemes were examined.The first used raw ratios as calculated from Eq. (1).The second approach employed a data preprocessingstep to examine whether average signal subtractioncould improve imaging performance. This was accom-plished by fitting a cubic polynomial to the measuredBorn ratio for each source after plotting all the datacontained in each CCD image acquired as a functionof radial distance on this image. A cubic polynomialwas selected because it allowed enough degrees offreedom to provide for an adequate approximation ofthe background information. The center of the radiusin each image was assumed to be the pixel in theimage that corresponded to a point on the front glasswindow with the same x–y coordinates as the x–ylocation of the light source on the back plate. Thiscenter can therefore be thought of as the projection ofthe source location onto the glass window and corre-sponds to the maximum signal collected when illumi-nating through homogeneous diffusive media.

To apply the background correction suggestedabove, we selected the normalized Born field at thepresence of the 30 nM background fluorochromeshown in Fig. 3 (� symbols) for a source known to bedistant from the spherical inclusion. This measure-ment was selected as indicative of the backgroundfield only and was fitted as described above to obtaina distance-dependent average background fluores-cence signal; this fit is shown in Fig. 3 (solid curve).This distance-dependent average value was then sub-tracted from the Born value at each source, an exam-

Table 1. Heterogeneous Background Setup: FluorescenceConcentration Combinations for the Four Tubes of the Heterogeneous

Phantom Shown in Fig. 2(c)

Experiment Tube 1 Tube 2 Tube 3 Tube 4

1 30 nM 30 nM 30 nM 30 nM2 90 nM 30 nM 30 nM 30 nM3 30 nM 60 nM 60 nM 30 nM4 60 nM 60 nM 60 nM 60 nM

Fig. 2. Experimental setup in an imaging chamber (top view). (a)The phantom, a 5 mm diameter hollow vinyl sphere containing200 nM fluorochrome, was suspended against the front plate of theimaging chamber, which was filled with 1% Intralipid fluid andhomogeneous background concentrations of Cy5.5 fluorochrome(see Section 2). (b) Same as (a) but with the sphere suspended inthe middle of the chamber. (c) To simulate heterogeneous back-ground fluorescence, four tubes were inserted containing fluores-cence concentrations ranging from 0 to 90 nM (see Table 1) whilethe background level of fluorochrome in the chamber was heldconstant at 30 nM. In all the experiments the concentration offluorochrome in the sphere was held constant at 200 nM.


ple of which is given for the case of source 1 in Fig. 3.The corrected data of Fig. 3 (diamonds) appear sim-ilar to the data obtained at the absence of backgroundfluorochrome (filled circles). The subtraction schemedemonstrated in Fig. 3 was applied to examine thebasic performance of data preprocessing in improvingimage reconstruction as shown in Section 3. We notethat selection of a distant source is not practical forthe high-background-heterogeneity case (or the invivo case) since no measurement is known a priori tobe representative of the background. To generalizeour observations for the heterogeneous phantom, cu-bic polynomials were fit to the Born value at each ofthe 46 sources, and the radial distance-dependentaverage of all the polynomials was used for back-ground subtraction.

3. Results

A. Homogeneous Background

Figure 4 depicts reconstructed slices passing nearthe center of the spheres at depths of 2.3 mm [Fig.4(a), front of the chamber] and 6.5 mm [Fig. 4(b),middle of the chamber]. As observed, the sphere iswell reconstructed at 0 nM background fluoro-chrome concentration in both positions (front andmiddle) of the chamber. At the front of the chamber[Fig. 4(a)] it can be seen that increasing the back-ground fluorochrome concentration results in in-creasing levels of reconstructed background signalbut that in general the sphere is still well recon-structed even at the highest levels of backgroundfluorochrome. The correction method, however, issuccessful at removing this background signal, andthe quantification results discussed below furtherillustrate this improvement. In the middle of the

chamber [Fig. 4(b)] it is seen that the presence ofbackground fluorochrome has a dramatic effect onthe reconstructed images. Even with as little as 30nM background fluorochrome there is significantdistortion of the reconstructed sphere, and thisworsens as the background concentration increases.The corrected reconstructions, however, showmarked improvement. The sphere is visible (al-though slightly diminished in intensity) and nearlyall of the background has been removed. There aresome artifacts to the right of the sphere that arepresent to some degree in the 60 nM backgroundlevel but worsen for the 90 nM background level.

The quantification performance as a function ofbackground fluorescence for the homogeneousphantom is summarized in Figs. 5(a) and 5(b). Bothgraphs plot the reconstructed concentration of thesphere within a region of interest defined by the0 nM background reconstruction. These results arenormalized to the 0 nM background case to plotrelative changes as a function of background fluo-rescence and corresponding correction. The recon-structions based on uncorrected data are shownwith black bars, and the reconstructions based onbackground-corrected measurements are shownwith gray bars. In the case of the sphere at the frontof the chamber [Fig. 5(a)] the quantification error isseen to increase with increasing background con-centration. In each case, underestimation of thesphere’s fluorochrome concentration is observed,reaching nearly 40% in the 90 nM background case.The corrected results improve upon the uncorrectedresults in each case and are within 10% of the true

Fig. 3. Illustration of background fluorescence correction for ho-mogeneous phantom source 1, showing the normalized Born fieldcollected in the absence of background fluorescence (filled circles)and with 30 nM background fluorochrome (� symbols). The solidblack curve is a cubic polynomial fit to the data with backgroundfluorescence. The corrected Born field (diamonds) is shown aftersubtraction of the cubic polynomial from the data. The cubic poly-nomial is subtracted from all of the sources, yielding Born valuesthat are nearly identical to that of the case without background.

Fig. 4. Representative reconstructed tomographic slices of homo-geneous background experiments with a sphere at (a) the front ofthe chamber (slice depth of 2.3 mm) and (b) the middle of thechamber (slice depth of 8.5 mm). The different background concen-trations used are arranged in labeled columns, with the uncor-rected (Raw) and corrected reconstructions grouped in labeledrows.


value for both the 30 and 60 nM background cases.Although there is improvement in the 90 nM case,the corrected results demonstrate nearly a 20% un-derestimation in this case. The correspondingchanges for the case of the sphere in the middle ofthe chamber are shown in Fig. 5(b). Here, the un-corrected reconstruction yields an overestimationthat is higher than 80% for the 60 nM backgroundlevel and 150% for the 90 nM background level. Thecorrected results reveal a 25% underestimation inthe 30 nM background case and less than 10% un-derestimation in the 60 and 90 nM backgroundcases.

Overall the data demonstrate that for objects onthe surface even the �2:1 target:background contrastratio can be well resolved in all cases. However in themiddle of the chamber an �6:1 ratio is marginallydetected without correction. Data correction is neces-sary to resolve objects at an �3:1 ratio, whereas at an�2:1 ratio image fidelity deteriorates.

B. Heterogeneous Background

Figure 6(a) depicts a planar fluorescence image of thephantom used for the heterogeneous background flu-orescence study experiments before it was sur-rounded by the diffuse fluorescent fluid (here shownfor experiment 1 as indicated on Table 1). Figure 6(b)shows representative slices through the center of thesphere (depth of 8.5 mm) for experiments 1, 3, and 4on Table 1. Experiment 2 was omitted for brevity, butit exhibited behavior similar to the other three exam-ples. Figure 6(b) (top row) shows the effects of back-ground fluorescence increase on the reconstructedimages. Similar to the homogeneous backgroundcase, a marked background increase is noted, prefer-entially localized in the periphery of the image. Thereconstructed phantom is more intense than thecontrol image with no background fluorescence. Inall the cases examined, the sphere and the more-concentrated surrounding tubes were well distin-guishable when the entire reconstructed volume wasanalyzed.

Reconstructions performed on corrected Born val-ues are seen in Fig. 6(b) (bottom row) and demon-strate better image fidelity than reconstructionsperformed on raw Born values in all four cases ex-amined. These results demonstrate an improvementin image reconstructions of the sphere since the im-ages are virtually artifact free. Conversely an under-estimation in intensity and size is observed comparedto the control reconstruction.

The calculated concentrations (Fig. 7) were affectedby the presence of background fluorescence to a sim-ilar degree as in the homogeneous background casewith the sphere in the middle of the chamber. With-out correction, the percent error was typically in therange of 50% overestimation, which was less than the

Fig. 5. Quantification of homogeneous experiments with (a) asphere at the front of the chamber and (b) a sphere in the middleof the chamber. Black bars indicate relative reconstructed concen-trations for uncorrected reconstructions, and gray bars indicateconcentrations from corrected reconstructions. In both (a) and (b)the reconstructed object concentration is normalized to the recon-structed object concentration obtained from an experiment with nobackground fluorescence.

Fig. 6. Representative slices of heterogeneous background recon-struction. (a) Fluorescence reflectance image of the phantom withoutIntralipid showing blocking tubes obscuring a direct view of (b)reconstructed slices taken at a chamber depth of 8.5 mm with la-beled columns to indicate experiment number (see Table 1) and withthe top row (Raw) containing uncorrected slices and the bottom rowcontaining slices reconstructed with background correction.


error for the homogeneous experiments, but withworse image quality as exhibited in Fig. 6(b) (toprow). Polynomial correction resulted in a notable un-derestimation of the sphere concentration (nearly45%) for the first three experiments, as is also evidentin Fig. 6(b) (bottom row). Somewhat better resultswere obtained in experiment 4, with only 15% over-estimation in the corrected image.

4. Discussion

Improving the ability to image fluorescent objects inthe presence of broad and inhomogeneous back-ground fluorescence is important for in vivo tomo-graphic applications. In this study we sought tocharacterize the effect of different levels of homo-geneous and heterogeneous background fluores-cence on the reconstruction of a fluorescent targetby using phantoms of known fluorescence distribu-tion. An important aspect of the study was to ex-plore whether average background subtractionschemes could improve imaging and relative quan-tification performance.

Our observations show that objects are tomo-graphically detectable at an �3:1 target: backgroundcontrast ratio deep in diffuse media of small animaldimensions, but detection ability worsens at smallercontrast ratios. Generally, increased background flu-orescence resulted in significant loss of relative quan-tification accuracy, erroneous size estimation, and anincreased number of artifacts. However, the use ofnormalized methods minimize significantly the sen-sitivity to background optical properties and hetero-geneity. These observations generally apply both inthe homogeneous and heterogeneous case. Mean rel-ative quantification error for all the studies examinedwas of the order of 50%. At a low target:backgroundcontrast ratio, relative quantification errors reachedas high as 150% for reconstructions on uncorrectedraw data.

The study of heterogeneous phantoms is importantin order to generalize findings to situations more

closely resembling the in vivo case. The phantomsselected cover a reasonable range of optical variationand spatial heterogeneity. We selected a worst-casescenario of heterogeneity arrangement where thespatial variation is highly asymmetric with respect tothe target sphere (experiments 3 and 4) and wherethe inhomogeneity selected obstructs direct viewingof the target sphere as seen in Fig. 6(a). These com-binations are certainly not representative of all invivo situations but nevertheless allow for insight intoheterogeneous background situations.

The subtraction method examined herein wasfound to significantly improve the detection and rel-ative quantification of the target sphere for bothhomogeneous and heterogeneous phantoms. The re-sults from the homogeneous cases showed virtualelimination of the effects of the background fluores-cence, independently of the background concentra-tion examined. Improvements of the heterogeneouscases included images with fewer artifacts and im-proved quantification values. The premise of the cor-rection is that background signals will appear in allthe measurements and therefore can be approxi-mated by an average fitted function. Conversely,stronger perturbations will still be observable abovethis average fit and more accurately reconstructedfollowing data subtraction. Relative quantificationerrors after correction attained a mean value of 25%.Maximum error reached 50% underestimation ob-served in the heterogeneous studies, using the poly-nomial correction method.

In conclusion, background fluorescence can signif-icantly affect tomographic image quality in FMT ap-plications when the target:background ratio is below�6:1. Typical effects include image artifacts, quanti-fication errors, and detection loss. Correction meth-ods have demonstrated capacity to improve onreconstructions, and they can be used to correct forbackground signals due to extrinsically administeredfluorochromes or intrinsic autofluorescence due to tis-sue fluorochromes. The particular method selectedherein, a straightforward subtraction of average poly-nomial fits through the data, was found to improve onimage fidelity and yielded greater quantitative accu-racy and better detection characteristics. In thisstudy we found that objects with an �3:1 target:background contrast could still be accurately de-tected after correction. In addition, this resultindicates that more advanced subtraction schemes,including model-based methods, could potentiallyfurther improve reconstruction performance, andsuch methods are currently being pursued.

The authors thank Joshua Dunham and AshrafIssa-Samarou for invaluable help with the experi-ments performed; the staff of the Harvard–MIT Di-vision of Health Sciences and Technology BiomedicalOptics Summer Institute; and its director, ThomasDeutsch. This research was sponsored by NationalInstitutes of Health grant RO1 EB 000750-1, NASA–National Cancer Institute contract BAA-NO1-CO-

Fig. 7. Quantification of the heterogeneous background phantomfor uncorrected (black bars) and corrected (gray bars) reconstruc-tions.


17016-32, and U.S. Army Medical Research andMateriel Command grant W81XWH-04-1-0239.

References1. V. Ntziachristos, C. Tung, C. Bremer, and R. Weissleder,

“Fluorescence-mediated tomography resolves protease activityin vivo,” Nat. Med. 8, 757–760 (2002).

2. M. A. Oleary, D. A. Boas, X. D. Li, B. Chance, and A. G. Yodh,“Fluorescence lifetime imaging in turbid media,” Opt. Lett. 21,158–160 (1996).

3. B. B. Das, F. Liu, and R. R. Alfano, “Time-resolved fluorescenceand photon migration studies in biomedical and model randommedia,” Rep. Prog. Phys. 60, 227–292 (1997).

4. M. J. Eppstein, D. E. Dougherty, T. L. Troy, and E. M. Sevick-Muraca, “Biomedical optical tomography using dynamicparameterization and Bayesian conditioning on photon migra-tion measurements,” Appl. Opt. 38, 2138–2150 (1999).

5. H. B. Jiang, “Frequency-domain fluorescent diffusion tomog-raphy: a finite-element-based algorithm and simulations,”Appl. Opt. 37, 5337–5343 (1998).

6. J. Chang, H. L. Graber, and R. L. Barbour, “Imaging of fluo-rescence in highly scattering media,” IEEE Trans. Biomed.Eng. 44, 810–822 (1997).

7. S. R. Arridge, “Optical tomography in medical imaging,” In-verse Probl. 15, R41–R93 (1999).

8. R. Barbour, S. Blattman, and T. Panetta, in Dynamic OpticalTomography: a New Approach for Investigating Tissue-Vascular Coupling in Large Tissue Structures, OSA TechnicalDigest (Optical Society of America, 2000), pp. 336–338.

9. D. A. Boas, L. E. Campbell, and A. G. Yodh, “Scattering and

imaging with diffusing temporal field correlations,” Phys. Rev.Lett. 75, 1855–1858 (1995).

10. A. D. Klose and A. H. Hielscher, “Iterative reconstructionscheme for optical tomography based on the equation of radi-ative transfer,” Med. Phys. 26, 1698–1707 (1999).

11. M. O’Leary, D. Boas, B. Chance, and A. Yodh, “Experimentalimages of heterogeneous turbid media by frequency-domaindiffusing-photon tomography,” Opt. Lett. 20, 426–428 (1995).

12. V. Ntziachristos and R. Weissleder, “Experimental three-dimensional fluorescence reconstruction of diffuse media usinga normalized Born approximation,” Opt. Lett. 26, 893–895(2001).

13. A. Godavarty, A. B. Thompson, R. Roy, M. Gurfinkel, M. J.Eppstein, C. Zhang, and E. M. Sevick-Muraca, “Diagnosticimaging of breast cancer using fluorescence-enhanced opticaltomography: phantom studies,” J. Biomed. Opt. 9, 488–496(2004).

14. E. Graves, J. Ripoll, R. Weissleder, and V. Ntziachristos, “Asub-millimeter resolution fluorescence molecular imagingsystem for small animal imaging,” Med. Phys. 30, 901–911(2003).

15. V. Ntziachristos, E. Schellenberger, J. Ripoll, D. Yessayan, E.Graves, A. Bogdanov, Jr., A. L. Josephson, and R. Weissleder,“Visualization of anti-tumor treatment by means of fluores-cence molecular tomography using an annexin V–Cy5.5 con-jugate,” Proc. Natl. Acad. Sci. U.S.A. 101, 12294–12299 (2004).

16. H. R. Herschman, “Molecular imaging: looking at problems,seeing solutions,” Science 302, 605–608 (2003).

17. R. Weissleder and V. Ntziachristos, “Shedding light onto livemolecular targets,” Nat. Med. 9, 123–128 (2003).


effects of background fluorescence in fluorescence molecular tomography

Documents