In vivo validation of free-space fluorescencetomography using nuclear imaging

Anikitos Garofalakis,1,2 Albertine Dubois,1,2 Bertrand Kuhnast,1 Daniel M. Dupont,1,2 Isabelle Janssens,1,2

Nicolas Mackiewicz,1,2 Frédéric Dollé,1 Bertrand Tavitian,1,2 and Frédéric Ducongé1,2,*1Commissariat à l’énergie atomique et aux énergies alternatives (CEA), Institut d’Imagerie Biomédicale (I2BM),

Service Hospitalier Frédéric Joliot (SHFJ), 4 place du Général Leclerc, 91401, Orsay Cedex, France2Institut National de la Santé et de la Recherche Médicale (INSERM) U1023, Laboratoire d’Imagerie

Moléculaire Expérimentale, 4 place du Général Leclerc, 91401, Orsay Cedex, France*Corresponding author: frederic.duconge@cea.fr

Received June 18, 2010; revised August 3, 2010; accepted August 13, 2010;posted August 17, 2010 (Doc. ID 130364); published September 2, 2010

The performance of small animal photonic imaging has been considerably improved since the development offluorescence diffuse optical tomography (fDOT), which can reconstruct fluorescent probe distribution inside tissue.However, the quantification capabilities of this new technology are still a topic of debate, especially in comparisonto classical nuclear imaging techniques. Here, we present a method to in vivo calibrate the quantity and localizationof a probe provided by free-space fDOT (where no plate is compressing the mouse) with positron emissiontomography (PET) and x-ray computed tomography, respectively. This methodology allowed us to demonstrate astrong linear correlation (R2 ¼ 0:95) between fDOT and PET for probe concentrations ranging from 3 nM to 1 μM in adeep-seated organ. © 2010 Optical Society of AmericaOCIS codes: 170.0110, 170.3880, 170.6960, 110.6955, 110.4190.

In vivo molecular imaging needs to be quantitativelyvalidated in order to be widely used and accepted bythe scientific community. Currently, positron emissiontomography (PET) and single-photon-emission com-puted tomography (SPECT) represent well-validatedimaging technologies that allow the quantification ofradio-labeled probes independently of their localizationinside small animals and humans. PET and SPECT havebeen proven to achieve detection sensitivities in the pi-comolar and nanomolar ranges, respectively. However,access to nuclear imaging facilities remains limitedand their detection limits are time dependent, becausethey are connected to the radio-tracer decay. As an alter-native, planar imaging systems based on optics have beendeveloped and have encountered huge success in precli-nical research [1]. Such instruments use high-sensitivityCCD cameras that allow semiquantitative assessment ofsuperficial luminescence signals originating from a fewmillimeters inside small animal models. However, for sig-nals in deep-seated organs, these instruments are inher-ently not quantitative, because their measurements arestrongly dependent on the signal depth. In contrast to nu-clear imaging, the major problem in photonic imaging isthat light photons are subject to high amounts of scatter-ing when traveling through tissue. Fluorescence diffuseoptical tomography (fDOT) has addressed these issues,by the development of instruments that operate in a tran-sillumination excitation mode [2], and are supported bysophisticated reconstruction algorithms [3]. Hence, fDOTinstruments allow for reconstructing and quantifying afluorophore in three dimensions, regardless of its posi-tion inside the subject.Until now, the most common way to calibrate an fDOT

signal has been to employ phantoms of known geometrythat contain fluorophores of known concentrations [2].An alternative calibration method has been to placecapillaries filled with fluorophores of known concentra-tions inside an animal [4]. Such methods have demon-strated the quantitative capability of fDOT machines

from 250 nM to 1 μM [4,5]. However, these experimentswere often performed with very few points and are farfrom realistic cases, because a phantom model doesnot reflect the complexity of living tissue. Recently,Nahrendorf et al. [6] described the fusion of nuclear ima-ging (PET) and fDOT. They demonstrated that a goodcorrelation between these two imaging modalities can beobtained to measure the concentration of a probe in asuperficial xenografted model. However, when the probeis in deep tissue, the sensitivity limit of fDOT and its re-construction accuracy still remain questionable. To ad-dress this question, we measured the concentration of aprobe in a deep-seated organ (the kidney) using multimo-dal PET/fDOT imaging. Therefore, we calibrated a novelfDOT system that operates in a free-contact mode, wherethe mouse is not compressed into slab geometry [5]. Thesmall animal PET measurements acted as a referencefor quantification and micro-computed-tomography mea-surements as a reference for anatomical localization.

The probe used is a scramble, 20-O-methyl-modified,26-mer oligonucleotide (ON) derivatized at its 30 end witha thiohexyl motive and tagged with the positron-emitterfluorine-18 ([18F]ON) or the near-IR fluorescent dyeAlexa Fluor680 ([AF680]ON). The full tagging procedureshave been recently described in detail [7]. The choice ofthis probe was based on its lack of degradation in vivoand fast renal clearance with strong signal accumulationwithin few minutes after injection, as we reportedpreviously.

Selected amounts of [AF680]ON (ranging from 2 to10,000 pmol) were mixed with 200 μCi of the [18F]ON,ranging from 80 to 7000 pmol) and then injected in an an-esthetizedmouse just before imaging. Eighteenmicewereused to cover a wide dynamic range of fluorescent probeconcentrations. To minimize the autofluorescence signalfrom the intestinal area, the mice were nourished with achlorophyll-free diet (Diet 210, SAFE, France). Thebiodis-tribution of the probe was sequentially measured by PET,fDOT, PET, and CT (Fig. 1). The anesthetized animals

3024 OPTICS LETTERS / Vol. 35, No. 18 / September 15, 2010

0146-9592/10/183024-03$15.00/0 © 2010 Optical Society of America

were positioned on the PET scanner bed immediatelyafter injection, and a dynamic scan was acquired for30 min. The mouse base was then positioned in the near-by fDOT scanner for 30 min and, during this time, two tofour fDOT scans in the kidney area were performed. Thesupporting plate was then again placed in the PETmachine, and a second dynamic scan of 30 min was per-formed. Finally, a CT scan was performed after the intra-venous injection with 200 μL of Iopamidol (Iopamiron370, Schering AG) for the anatomical visualization ofthe kidneys.To allow coregistration of multimodal measurements,

we built a mouse holder with a transparent Plexiglasplate that can be fixed in the three different imaging mod-alities and exhibits minimal attenuation. Four sources ofmultimodality fiducial markers were fixed on the platearound the mouse, and they allowed for the coregistra-tion of sequentially acquired images of the same animal.For coregistration of images and analysis, a user-

friendly, in-house toolbox has been designed and devel-oped. This toolbox was integrated within the Brainvisamedical imaging processing software (http://brainvisa.info/index_f.html). Coregistered images clearly showthat the highest reconstructed signal correlated betweenPET and fDOT and was colocalized within kidneys asvisualized by using anatomical CT imaging (Fig. 1). More-over, metabolism of the ON in urine was analyzed usingelectrophoresis and no metabolites were observed, de-monstrating that both PET and fDOT signals correspondto intact ONs (data not shown). For each modality,

volumes of interest (VOIs) were drawn in order to esti-mate the kidney volumes as well as to measure the fluor-escence and radioactivity signals. VOIs were drawnmanually on the fDOT and CT datasets, while an unsuper-vised segmentation method based on voxel dynamicswas applied for PET [8].

Because PET and fDOT measurements were not per-formed at the same time, PET data were extrapolatedto cover the time window of the fDOT scanning (Fig. 1).Overall, 45 fDOT data points had signals greater than thefluorescent background signal, covering 3 orders of mag-nitude of probe quantity when correlated with PET(Fig. 2) (ranging between 0.7 and 300 pmol). With theuse of PET VOIs, we calculated the percentage of the[18F]ON injected dose (%IDPET) residing in the targetedorgan. Hence, given that we know the quantity (Qi) of[AF680]ON in the injected product, we calculated thequantity of fluorescent probe as %IDPET × Qi. The resultsallowed us to demonstrate a strong linear correlation(R2 ¼ 0:95) between the quantity of probe measuredby PET and the fluorescence signal measured by fDOT[Fig. 2(a)].

It is important to point out that, when the CT data wereused as the accurate anatomical reference, the PET seg-mentation was found to overestimate the kidney volumes

Fig. 1. (Color online) Timing of the different imaging acquisi-tions. In the top row, typical reconstructed images of each mod-ality are shown. The graph points below illustrate the variationof the probe quantity residing inside the two kidneys as calcu-lated by two separate dynamic PET scans on the same mouse.The gap under the two PET scans was filled by a nonlinear re-gression, and the fitted line served for extracting the probevalue that corresponds to the time point of the fDOT measure-ment. The vertical dotted lines indicate the repositioning of themouse bed from the one modality to the other (Media 1).

Fig. 2. (a) Probe quantities in picomoles as calculated by PETcorrelated with the optical reconstructed signal, in arbitraryunits (a.u.) and plotted in logarithmic axes. The errors in thepoints correspond to the regression errors of the extrapolationprocess of Fig. 1. The equation of the linear regression is½pmol� ¼ ð0:0045� 0:0002Þ × ½a:u:� − ð1:0074� 5:7776Þ with R2

¼ 0:95. (b) Correlation of concentrations between PET andfDOT. The equation of the linear regression is ½C� ¼ ð0:0099�0:0005Þ × ½a:u:� − ð0:0002� 0:0244Þ with R2 ¼ 0:95.

September 15, 2010 / Vol. 35, No. 18 / OPTICS LETTERS 3025

(VOLPET=VOLCT ¼ 1:52� 0:42), whereas the fDOT seg-mentation underestimated them (VOLfDOT=VOLCT ¼0:76� 0:26) (Fig. 3). Therefore, these results disclosethat both PET and fDOT are able to accurately reportthe total probe quantity in a given organ despite a largediscrepancy in the estimation of geometrical volumes. In-deed, the average kidney volume estimated by fDOT wasapproximately half the one estimated by PET (VOLfDOT=VOLPET ¼ 0:51� 0:16). Consequently, we used the kid-ney volumes measured from the CT datasets for the cal-culation of probe concentration and we observed astrong linear correlation (R2 ¼ 0:95) between the PETand fDOT data for probe concentrations ranging from3 nM to 1 μM [Fig. 2(b)].The first small-animal fDOT apparatus was developed

by Ntziachristos et al. [2] in 2002 and evolved, in less thanten years, from systems where the animal was immersedinside an index-matching fluid [2] to fluid-free systemswhere the animal was still compressed between twotransparent plates [9] and, finally, to free-space systems[5]. Currently, the main problem of fDOT is the low re-solution, which does not allow perfect organ delineation;however, SPECT and PET display the same problem. Tosolve this drawback, PET and SPECT systems have beenfused with anatomical imaging modalities, such as CT[10]. A similar solution has also been proposed for fDOTand could also offer the a priori structural informationthat would facilitate the inverse problem of optical recon-struction [11,12]. In addition, several methods are underinvestigation to further improve the fDOT technology,

such as the incorporation of 360° angular collectiongeometries [13].

In conclusion, whereas development to further im-prove fDOT is still ongoing, our work provides clear evi-dence that free-space fDOT can already reach nanomolarsensitivity, even in a deep-seated organ. Although it is farfrom the sensitivity limit of PET, it is quite similar to thelimit of small animal SPECT.

The authors would like to thank Vincent Brulon for hisvaluable assistance during the PET measurements. Thiswork was supported by grants from the Hybrid Fluores-cence Molecular Tomography and X-ray Computed To-mography System and Method (FMT-XCT) Europeanprogram (grant agreement 201792) the EuropeanMolecu-lar Imaging Laboratory (EMIL) network [European Union(EU) contract LSH-2004-503569], and by the “AgenceNationale pour la Recherche” (projects ANR-RNTS:TomoFluo3D). D. M. D. acknowledges financial supportfrom the Danish Cancer Society (DP 07043).

References

1. R. Weissleder and M. J. Pittet, Nature 452, 580 (2008).2. V. Ntziachristos, C. Tung, C. Bremer, and R. Weissleder,

Nat. Med. 8, 757 (2002).3. M. A. O’Leary, D. A. Boas, X. D. Li, B. Chance, and A. G.

Yodh, Opt. Lett. 21, 158 (1996).4. E. E. Graves, D. Yessayan, G. Turner, R. Weissleder, and V.

Ntziachristos, J. Biomed. Opt. 10, 044019 (2005).5. A. Koenig, L. Hervé, G. Gonon, V. Josserand, M. Berger, J.

Dinten, J. Boutet, P. Peltié, J. Coll, and P. Rizo, J. Biomed.Opt. 15, 016016 (2010).

6. M. Nahrendorf, E. Keliher, B. Marinelli, P. Waterman, P. F.Feruglio, L. Fexon, M. Pivovarov, F. K. Swirski, M. J. Pittet,C. Vinegoni, and R. Weissleder, Proc. Natl. Acad. Sci. USA107, 7910 (2010).

7. B. Tavitian, F. Ducongé, R. Boisgard, and F. Dollé, MethodsMol. Biol. 535, 241 (2009).

8. R. Maroy, R. Boisgard, C. Comtat, V. Frouin, P. Cathier, E.Duchesnay, F. Dollé, P. E. Nielsen, R. Trébossen, and B.Tavitian, IEEE Trans. Med. Imaging 27, 342 (2008).

9. R. B. Schulz, J. Ripoll, and V. Ntziachristos, IEEE Trans.Med. Imaging 23, 492 (2004).

10. Y. Seo, B. Teo, M. Hadi, C. Schreck, S. L. Bacharach, and B.H. Hasegawa, Med. Phys. 35, 3086 (2008).

11. D. Hyde, R. Schulz, D. Brooks, E. Miller, and V.Ntziachristos, J. Opt. Soc. Am. A 26, 919 (2009).

12. Y. Lin, W. C. Barber, J. S. Iwanczyk, W. Roeck, O. Nalcioglu,and G. Gulsen, Opt. Express 18, 7835 (2010).

13. A. Garofalakis, G. Zacharakis, H. Meyer, E. N. Economou,C. Mamalaki, J. Papamatheakis, D. Kioussis, V.Ntziachristos, and J. Ripoll, Mol. Imaging 6, 96 (2007).

Fig. 3. (Color online) (a) and (b) present the axial and the cor-onal views of fused mesh volumes occupied by the recon-structed signal in the kidneys. The optical reconstructedmesh is depicted with green (b), the PET with red (a), andthe CT with granular gray [(a) and (b)]. The mesh volumesof the organs are rendered to the mesh volume of the mousebody region corresponding to the optical scanning frame.

3026 OPTICS LETTERS / Vol. 35, No. 18 / September 15, 2010