FMT-PCCT: Hybrid Fluorescence Molecular ... - IEEE Xplore

1434

IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 33, NO. 7, JULY 2014

FMT-PCCT: Hybrid Fluorescence Molecular Tomography—X-Ray Phase-Contrast CT Imaging of Mouse Models Pouyan Mohajerani, Alexander Hipp, Marian Willner, Mathias Marschner, Marija Trajkovic-Arsic, Xiaopeng Ma, Neal C. Burton, Uwe Klemm, Karin Radrich, Vladimir Ermolayev, Stratis Tzoumas, Jens T. Siveke, Martin Bech, Franz Pfeiffer, and Vasilis Ntziachristos* Abstract—The implementation of hybrid fluorescence molecular tomography (FMT) and X-ray computed tomography (CT) has been shown to be a necessary development, not only for combining anatomical with functional and molecular contrast, but also for generating optical images of high accuracy. FMT affords highly sensitive 3-D imaging of fluorescence bio-distribution, but in standalone form it offers images of low resolution. It was shown that FMT accuracy significantly improves by considering anatomical priors from CT. Conversely, CT generally suffers from low soft tissue contrast. Therefore utilization of CT data as prior information in FMT inversion is challenging when different internal organs are not clearly differentiated. Instead, we combined herein FMT with emerging X-ray phase-contrast CT (PCCT). PCCT relies on phase shift differences in tissue to achieve soft tissue contrast superior to conventional CT. We demonstrate for the first time FMT-PCCT imaging of different animal models, where FMT and PCCT scans were performed in vivo and ex vivo, respectively. The results show that FMT-PCCT expands the potential of FMT in imaging lesions with otherwise low or no CT contrast, while retaining the cost benefits of CT and simplicity of hybrid device realizations. The results point to the most accurate FMT performance to date. Manuscript received February 20, 2014; accepted March 19, 2014. Date of publication March 24, 2014; date of current version June 27, 2014. This work was supported in part by the DFG Cluster of Excellence Munich-Centre for Advanced Photonics (MAP), in part by the DFG Gottfried Wilhelm Leibniz program, and in part by the European Research Council (ERC, FP7, StG 240142). This work was also supported in part by the Karlsruhe Nano Micro Facility, a Helmholtz Research Infrastructure at Karlsruhe Institute of Technology (KIT) and German Research Foundation(SFB 824). Asterisk indicates corresponding author. This paper has supplementary downloadable material available at http://ieeexplore.ieee.org, provided by the authors. P. Mohajerani, X. Ma, N. C. Burton, U. Klemm, K. Radrich, V. Ermolayev, and S. Tzoumas are with the Chair for Biological Imaging, Technische Universität München, 80333 Munich, Germany, and also with the Institute for Biological and Medical Imaging, Helmholtz Zentrum München, 85764 Munich, Germany. A. Hipp, M. Willner, M. Marschner, and F. Pfeiffer are with Chair for Biomedical Physics, Department of Physics, Technische Universität München, 80333 Munich, Germany . M. Bech is with the Chair for Biomedical Physics, Department of Physics, Technische Universität München, 80333 Munich, Germany, and also with the Department of Medical Radiation Physics, Lund University, 221 00 Lund, Sweden. M. Trajkovic-Arsic and J. T. Siveke are with II. Medizinische Klinik, Klinikum rechts der Isar, Technische Universität München, 80333 Munich, Germany. *V. Ntziachristos is with the Chair for Biological Imaging, Technische Universität München, 80333 Munich, Germany, and also with the Institute for Biological and Medical Imaging, Helmholtz Zentrum München, 85764 Munich, Germany (e-mail: [email protected]). Digital Object Identifier 10.1109/TMI.2014.2313405

Index Terms—Computed tomography (CT), fluorescence imaging, molecular imaging, phase contrast imaging, tomography.

I. INTRODUCTION

T

HE COMBINATION of multiple imaging modalities in hybrid implementations has been an indispensable development that merges their strengths while alleviating their limitations [1]–[8]. Multi-modality imaging has likewise proved of fundamental importance for fluorescence tomography, as it not only allowed the combination of images representing different contrast mechanisms but also enabled the generation of perhaps the most accurate 3-D fluorescence images. Indeed current developments in hybrid X-ray computed tomography (XCT) and fluorescence molecular tomography (FMT) have resulted in the use of the X-ray CT information in the fluorescence inversion problem in the form of statistical priors, leading to significant improvements in optical image quality and accuracy over standalone systems [3], [9]–[17]. The performance of FMT as a functional and molecular modality, however, is limited by the strong scattering of near-infrared photons. The use of anatomical information can improve the reconstruction accuracy by restricting the ill-posed nature of the inverse problem through structured regularization. However, a major limitation of hybrid implementations based on X-ray CT is the low soft-tissue contrast offered. In particular, while bone, lung, and the tissue outline can be visualized with good contrast, other internal organs and tissues are not visible with high contrast in X-ray CT images as they do not exhibit considerable absorption differences of X-ray photons—even if they may have very different optical properties. The use of X-ray contrast agents presents an interesting solution [18], [19]; however, contrast agents generally demarcate the vascular system and possibly biological features such as cancer-enhanced permeability and retention and not entire organs or tissue structures as it relates to photon propagation in the animal body. In addition to micro-CT, other anatomical and function modalities have also been previously employed and combined with FMT to improve the quality and accuracy of molecular imaging. A survey of these approaches is presented in [20]. In particular, magnetic resonance tomography (MRI) along

0278-0062 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

MOHAJERANI et al.: FMT-PCCT: HYBRID FLUORESCENCE MOLECULAR TOMOGRAPHY

with magneto-fluorescent nanoparticles has been proposed in conjunction with FMT [20]–[24]. While FMT can benefit from the excellent soft-tissue contrast of MRI, realization of hybrid FMT-MRI systems is limited due to several technical challenges [23], [25], [26]. Instead, we consider herein the combination of FMT with phase-contrast X-ray computed tomography (PCCT). Grating-based PCCT, originally developed at synchrotron radiation facilities [27], [28], has been more recently put into practice with common X-ray tubes [29] and has spurred the growth of phase-contrast CT. Though important benchmark studies have shown improved image contrast with monochromatic synchrotron radiation [30], [31] as well as polychromatic X-ray sources, one very important step towards preclinical deployment was achieved recently by integrating a PCCT imaging system into a rotating gantry toward realization of in vivo scanning of small animals. Grating-based PCCT shares the advantages of high resolution and relatively low cost with conventional micro-CT, but provides access to both electron density information [28], [32] and scattering caused by features on the micrometer length scale [33]. We hypothesized herein that FMT-PCCT can lead to a highly beneficial imaging modality, allowing for better anatomical contrast while improving the accuracy of FMT. Herein, FMT-PCCT is demonstrated using two different animal models; one with a subcutaneous tumor developed after animal inoculation with 4T1 breast cancer cells and an endogenous model of pancreatic ductal adenocarcinoma (PDAC). Both cases have an important role in cancer studies and drug development [3], [34], [35], where accurate molecular imaging plays a critical role [36]–[38]. The FMT and PCCT images were acquired subsequently for the two animal models and measurements were coregistered. The anatomical information segmented from the PCCT scans was used in the inversion of the FMT data. The imaging results were demonstrated for FMT-PCCT and compared with the current state-of-the-art FMT-XCT. This study was performed to assess the potential of FMT-PCCT in imaging specific uptake of fluorescent probes in lesions where there is low or no contrast with surrounding tissue in the conventional micro-CT. II. MATERIALS AND METHODS A. Animal Preparation All work with experimental animals was performed in agreement with Helmholtz Zentrum and district government of Upper Bavaria rules and regulations. A 93-day-old nude mouse was injected with one million 4T1 breast tumor cells subcutaneously in the dorsal thoracic region (caudal to the lungs), 10 days before imaging. Furthermore, a 63-day-old (CKP) [39], [40] model of pancreatic ductal adenocarcinoma (PDAC) was imaged using FMT as well. The 4T1 model and the PDAC model were injected with 2 nmol of IntegriSense 750 and IntegriSense 680 (Perkin Elmer, Waltham, MA, USA), respectively, in the tail vein 24 h prior to imaging. IntegriSense targets integrin which is known to be expressed in cancers of both models [3], [34].

1435

B. FMT and PCCT Experimental Setups and Imaging The animals were imaged by 360 FMT scanning in vivo. Imaging was performed under isoflurane anesthesia using a previously developed FMT-XCT system located at the Chair for Biological and Medical Imaging (IBMI) at Helmholtz Zentrum München (HMGU) [10]. A schematic diagram of this system is shown in Fig. 1(a). Optical imaging was performed within a field of view centered on the region of interest at 24 equispaced gantry angles. The 680 and 750 nm lasers (B&W Tek, Newark, DE, USA) scanned the sample at, on average, 26 source locations per gantry angle, where at each location intrinsic and fluorescence images were obtained using a cooled CCD camera (Pixis 512B, Princeton Instruments, Trenton, NJ, USA). The in vivo FMT imaging lasted a total of approximately 45 min for each mouse. Conventional micro-CT scans were also acquired after the FMT acquisition for volumetric coregistration and comparison purposes and lasted approximately 20 min for each animal. The animals were sacrificed after the FMT imaging through intraperitoneal overdose injection of Ketamin und Xylazin while still anesthetized. The 4T1 and PDAC mice were then preserved in a 4% Paraformaldehyde (PFA) fixation solution for two weeks and four weeks, respectively. X-ray phase-contrast computed tomography (PCCT) was carried out for both mice using a Talbot–Lau interferometer [28] PCCT system located at the Chair for Biomedical Physics of the Technische Universität München. The block diagram of this system is demonstrated in Fig. 1(b). The setup employed was equipped with a rotating anode X-ray tube source (Enraf Nonius FR-591) and a photon counting detector (Dectris Pilatus II). The interferometer itself consists of three gratings: the source grating, which is necessary to create enough spatial coherence using a conventional X-ray tube source [29], the phase grating to create an interference pattern and the analyzer grating to determine the position of the interference pattern through transversal stepping. The sample was placed directly in front of the phase grating to reach the best possible sensitivity [41]. The phase stepping approach, requiring the movement of one grating perpendicular to the X-ray beam, was used to extract the phase information. The gratings were manufactured by the Institute of Microstructure Technology (Karlsruhe Institute of Technology) and Microworks GmbH (Karlsruhe, Germany) [42]. Sourceand analyzer grating were made of gold with a bar height of approximately 60 m. The phase grating was made of nickel with a bar height of 8 m, inducing a phase shift of to incident X-rays with an energy of 22.8 keV (design energy). All gratings have a period of 5.4 m and were arranged in a symmetric configuration with inter-grating distances of 80 cm. The entire interferometer was placed 60 cm away of the X-ray source, operating with a Molybdenum target at an acceleration voltage of 40 kV and an anode current of 70 mA. Spectral filtering was achieved using three 500- m Silicon filters and immersing the sample in a 4-cm-wide water bath to prevent phase wrapping [43], [44]. Slightly different settings were used for the two measurements here. The interferometer reached a visibility of % (for the 4T1 model) and % (for the PDAC model). For the tomographic scan of each specimen, 1200 projections were recorded

1436


Fig. 1. FMT-PCCT system components: (a) FMT system and the complementary conventional CT subsystem. (b) X-ray PCCT system based on a Talbot–Lau interferometer consisting of three gratings. Sample was imaged in vivo under isoflurane anesthesia in the FMT and ex vivo in a falcon tube immersed in a transparent water container (not shown) in the PCCT system.

over 360 . Eleven steps comprised each projection, each step with 5 s (for the 4T1 model) and 3 s (for the PDAC model) exposure time. The total acquisition time was approximately 30 h (for the 4T1 model) and 20 h (for the PDAC model). The pixel intensities in the PCCT grayscale slice images represent the phase shift induced to the X-rays by the tissue in each voxel and are proportional to the refractive index decrement. Validation was achieved ex vivo to confirm the in vivo studies using a cryostat (CM1950, Leica Microsystems GmbH, Wetzlar, Germany) [45] retrofitted with a fluorescence imaging system. A maximum 30 s exposure time was applied for acquiring fluorescence images of cryo-sections, as the fluorescence is strongly weakened after several weeks of immersion in PFA 4%. C. FMT-PCCT Processing Here we describe processing and coregistration of optical and phase-contrast measurements. Since the FMT processing makes use of the anatomical priors derived from the PCCT, the processing of PCCT scans is explained first. To reconstruct phase-contrast CT images, filtered backprojections using a Hilbert filter were applied to the recorded differential phase-contrast projections. Due to the magnification of the cone-beam setup, the effective pixel size was 100 m 100 m (physical detector pixel size was 172 m 172 m). Anatomical information was extracted from PCCT scan via a trace-based semi-automatic segmentation approach using Amira (Visage Imaging, Richmond, Australia). In this approach, the anatomical boundaries of a specific tissue region or organ are extracted in several transverse slices and a 3-D surface designating the organ is then created through spatial interpolation of the 2-D traces. A mouse atlas [46] as well as cryo-section photographs obtained ex vivo were consulted during the segmentation to help with recognition and localization of different organs.

The animal volume extracted from the PCCT was coregistered with the micro-CT volume, obtained after the FMT acquisition in vivo using the conventional micro-CT subsystem of Fig. 1(a) and employed as the volumetric reference for coregistration. It should be noted that the animal body essentially undergoes nonrigid deformations between the in vivo optical imaging and ex vivo phase-contrast imaging. Such a nonrigid transformation, however, does not lend itself to reliable characterization. Hence, a rigid transformation was employed in this work to geometrically map the PCCT scan to the animal volume as imaged in vivo by the FMT. Specifically, the coregistration was designed to achieve a maximum overlap between the skeletons in both scans using a rigid transformation calculated through an approach described previously in [47]. Different optical absorption and scattering coefficients noted in [48] were assigned to general tissue, bone, lungs, and heart for solving the forward problem. Other soft-tissue organs such as the tumor tissue, kidneys, spleen, or segmented parts of the gastrointestinal tract were considered as general tissue in the forward modeling. We have previously shown that the use of published optical property values does not introduce errors that are larger than errors accrued during calculations on a per animal basis [48]. Toward modeling the propagation of light in biological tissue, the tissue volume was discretized for each mouse using a tetrahedral mesh with an average edge length of 1.3 mm. The mesh was generated using the methods described in [49], [50]. The FMT problem can be then expressed as (1) where and are the fluorophore distribution vector, the weight matrix, the measurement vector, the regularization parameter, and the regularization matrix [10], [51], [52].


Briefly, the vector represents the fluorophore concentration in the nodes of the inversion grid. The measurement vector consists of the so-called Born ratio for all source detector pairs, defined as the ratio of the fluorescence signal to the intrinsic signal. For the th measurement (for source and detector ) and the th node, is given as (2) is the Green’s function value between where source and node (node and detector ). The Green’s function is then estimated using finite element method (FEM)-based discretization of the diffusion approximation for light propagation modeling in tissue, as previously described in [53]. The inverse problem was then solved over a rectilinear grid with a resolution of 1 mm using the least squares method (LSQR) with 50 iterations [54]. After the iterations, the negative values of were set of zero. The segmented organs containing the cancerous lesions were used as a priori information to shape the regularization matrix in (1) using the weighted-segmented approach in the inverse problem [10]. The segment weights for the structured regularization were calculated using a two-step inversion approach, where in the first step the average concentrations in different anatomical segments are estimated and, in the second step, these initial estimations are used to penalize reconstructions in different segments [55]. A succinct description of this method is given here. Specifically, let through represent distinct anatomical sections of the tissue. The first inversion step consists of estimation of the average signal intensity in all of these segments, represented by a vector . This estimation is achieved via (3) is the corresponding weight matrix. Non-negwhere ativity is enforced in the optimization of (3). Next, the regularization matrix is formed as a diagonal matrix where is obtained from the first step results as

(4) where is the segment containing the node and is set to 0.066 [55]. The second inversion step is then carried out using (1). This approach has been previously shown optimal in reducing user input and offering a data-driven regularization scenario to avoid biasing of the solution [3]. For the 4T1 model, segment concentration averages and consequently segment weights were approximated for two distinct segments consisting of the tumor and the rest of the tissue. For the PDAC model, the estimations of the average fluorophore concentrations were performed for four segments consisting of the pancreatic tumor, kidneys, spleen, and the rest of the tissue. The spleen and kidneys constitute the main organs surrounding the pancreas which are not in the gastrointestinal

1437

tract. To set the regularization parameter, the L-curve (defined as the curve defined by the solution and residual norms for a large range of regularization parameters) was calculated and the point with maximum curvature was found and used as the operational corner point of the L-curve [56] for the reconstructions presented graphically in Section III. It should be noted that the same optical measurements were used for FMT-PCCT and FMT-XCT reconstructions in each case. D. A Performance Evaluation Metric Here we present a numerical metric to quantitatively analyze the imaging performance of FMT-XCT and FMT-PCCT and thereby evaluate the improvement in the reconstruction accuracy gained through inclusion of the anatomical priors afforded by PCCT in inversion of FMT. Moreover, the reconstruction error can be defined only if a “true” distribution is known. In this work, we use the fluorescence distribution obtained on cryo-sections during ex vivo validation as an approximation of the true distribution [45]. denotes the tissue volume and denotes a subset of corresponding to the volume of the target lesion, tumor, or organ. For a given distribution of fluorophores within and denoted by the vector , a reconstruction error metric is designed by incorporating a relative contrast measure and a dispersion measure. Specifically, drawing upon the concepts of variation coefficient [57] and localized signal-to-noise ratio [58], the relative contrast in the target volume relative to the rest of the tissue and is defined herein as (5) and denote average concentrations in target where volume and the background tissue - , respectively, and is the standard deviation of . A second measure is defined to quantify the dispersion of the fluorophore concentration within the target volume as (6) where is the standard deviation of the reconstruction within the target . Both of the contrast and the dispersion measures defined in (5) and (6) are invariant under scaling and translation. This property is deemed essential since the fluorescence detection ex vivo on cryo-sections—employed here for reference as the “true” distribution—can be an attenuated version of the actual (yet inaccessible) fluorescence distribution at the exact imaging time point. This is particular important due to the long fixation times in PFA. The measures and defined above are independent of such changes by the virtue of being invariant under linear transformations. It should be added that the fluorescence detected on cryo-sections after fixation in PFA can potentially be a nonlinear function of the original in vivo distribution—however, the invariance of the proposed measured to linear distortions can mitigate the impact of discrepancies between the measured fluorescence ex vivo and the true in vivo distribution.

1438


Fig. 2. Intensity profiles of phase-contrast CT and micro-CT images: (a) Micro-CT slice of the 4T1 model. (b) PCCT slice and (c) the intensity profiles of the micro-CT image (red curve) and the PCCT image (blue curve) along the white dotted lines. Corresponding results for the PDAC model are shown in (d)–(f). Parts of the intensity curves corresponding to different tissues are labeled in (c) and (f), where black (red) labels correspond to PCCT (micro-CT) intensity profiles. The x-axes in (c) and (f) denote distance in millimeters along the line segments, where distance 0 corresponds to the upper white squares in all cases.

Two measures are defined in (5) and (6) to achieve a metric of the reconstruction error for a given distribution and target volume as (7) where denotes the “true” distribution. The metric has a low value if and only if both the contrast and the dispersion of the distribution in the target volume are close to the respective values of the true distribution . The reconstruction accuracy metric of (7) is used in Section III-D to analyze the performance of FMT-PCCT as compared to FMT-XCT in imaging the two models examined in this work. III. RESULTS A. Reconstruction Results for Subcutaneous Tumor Fig. 2 shows intensity profiles of the XCT and PCCT images along sample lines across selected mouse images, plotted in a comparative manner. Fig. 2(a) and (b) shows XCT and PCCT cross-sectional images for the 4T1 model. The images are shown with the same orientation to facilitate comparisons, although an exact co-registration is not possible due to some bodily deformations that occur during imaging by the different scanners. Fig. 2(c) shows pixel intensity profiles along the dotted white lines shown on Fig. 2(a) and (b), where the red and blue curves correspond to XCT and PCCT images, respectively. The dif-

ferent tissues along the plotted profiles are marked as intervals on the intensity profile curves of Fig. 2(c). As seen, there are significant gaps between tumor, fat and bone tissue in the PCCT intensity profile (blue curve), while only bone tissue has conspicuous contrast relative to background in the micro-CT intensity profile (red curve). The local 16% contrast in micro-CT intensity between fat and tumor tissue [blue and red marks in Fig. 2(e)] is barely distinguishable in the red curve in Fig. 2(c). Both intensity profiles are normalized to the respective maximums. The higher soft tissue contrast of the PCCT scans is used then to segment the anatomical images. Specifically, bones, heart, lungs, adipose tissue, and the subcutaneous tumor are segmented from the PCCT scans. Although XCT has a much lower contrast than PCCT, the contrast is still large enough to allow for segmentation of the same tissue regions from the XCT images. Fig. 3 demonstrates imaging results of the subcutaneous 4T1 tumor mouse model, prepared as described in Section II-A. Fig. 3(a) shows in grayscale a PCCT cross section of the animal along an area passing through the top of the heart and lung, which contains the subcutaneous tumor. Superimposed in color is the reconstructed fluorescence signal due to accumulation of IntegriSense 750 in the tumor. There is good congruence between the anatomical appearance of the tumor and fluorescence signal reconstructed using FMT-PCCT. A corresponding color and fluorescence image obtained by cryoslicing for validation purposes is shown in Fig. 3(b). The fluorescence image captured by the cryoslicer camera has been enhanced for contrast


1439

Fig. 3. Hybrid FMT-PCCT imaging of a subcutaneous 4T1 tumor model: (a) FMT-PCCT reconstruction as fluorescence signal (pseudo-orange) imposed over a PCCT transverse slice; (b) cryoslicing-based validation where the fluorescence cryoimage is superimposed as pseudogreen transparency over the cryo-section color photograph; (c) PCCT slice and (d) the corresponding CT slice. The relative soft-tissue contrast between the tumor and surrounding tissue (regions marked with red and blue markers in (c) and (d), respectively) were 85% and 16% for the PCCT slice (c) and the micro-CT slice (d), respectively. (e) 3-D FMT-PCCT reconstruction showing the high concentration of IntegriSense 750 in the tumor—heart, lungs, and adipose tissue are shown in red, blue, and dark green, respectively; (f) animal volume extracted from micro-CT showing the view angle of (e). Coordinate system is designated by D (dorsal), V (ventral), Cr (cranial), Cd (caudal), L (left), and R (right). Video demonstration of (e) is available in supplementary video FMT-PCCT_4T1.wmv.

through background subtraction and is overlaid as transparency in pseudogreen on the cryoslice color photograph confirming the specific uptake of IntegriSense 750 in the subcutaneous tumor. Corresponding PCCT and conventional micro-CT transverse slices are shown in Fig. 3(c) and (d), respectively. As seen, the tumor has a much higher soft-tissue contrast in the PCCT image than the conventional CT image. Specifically, soft-tissue contrasts of 16% and 85% were observed for a point inside the tumor (red marks) in Fig. 3(c) and (d), relative to the adjacent background (blue marks), for the CT and PCCT slices, respectively. Fig. 3(e) shows the 3-D rendering of FMT-PCCT showing the bone structure, lungs, heart and parts of adipose tissue segmented from the PCCT images and depicted in gray, blue, red, and green colors, respectively. Fig. 3(f) shows a rendering of the mouse outer boundary (skin). While Fig. 3 demonstrated the basic features of FMT-PCCT imaging, we further investigated a model of pancreatic ductal adenocarcinoma to examine our hypothesis that the better soft-tissue contrast offered by PCCT could improve hybrid fluorescence imaging. B. Reconstruction Results for Deep-Seated Pancreatic Tumor Here we present the imaging results for the more challenging and interesting case of the PDAC model. Fig. 2(d)–(f) shows intensity profiles of the XCT and PCCT images for the PDAC model. The line segments in Fig. 2(e) starts (from the upper white square) in skin and crosses consecutively fat, muscle, fat, spleen, and pancreas and ends at the lower white square in the colon.

Different intervals corresponding to these tissues can be readily distinguished in the PCCT intensity profile (blue curve) in Fig. 2(f). However, the micro-CT has noticeable contrast only in colon [coinciding with the lower white square in Fig. 2(d)]. The dark spot in the lower right part of Fig. 2(d) is due to an air bubble arising in the gastrointestinal tract. The large soft tissue contrast of PCCT allows for segmentation of several organs from the PCCT images. Specifically, the bone, heart, lung, pancreas, stomach, spleen, kidneys, liver, colon, cecum, as well as part of panniculus adiposus were segmented from the PCCT images, as described in Section II-C. The lower contrast of XCT allows for reliable segmentations of bones, heart, and lungs as well parts of the colon. Fig. 4 shows the results of imaging the PDAC model using FMT-PCCT. Specifically, Fig. 4(a) and (b) demonstrate two cross-sectional slices from the abdomen of a PDAC animal which render PCCT images in grayscale and superimpose reconstructed fluorescence images of IntegriSense 680 in color. Only the fluorescence images are superimposed with a transparency scheme that plots the higher values in orange and renders the lower values transparent. The image exhibits rich soft tissue contrast by PCCT and a fluorescence signal that is distributed along the pancreatic tumor. Fig. 4(c) and (d) depicts the corresponding slices from the FMT-XCT reconstructions. Here, the characteristic absence of soft-tissue contrast significantly limits the ability to add accurate priors to the FMT inversion problem. As a result the fluorescence images, shown superimposed in color, only exploit few anatomical priors, such

1440


Fig. 4. Hybrid FMT-PCCT imaging of a PDAC model: (a) and (b) FMT-PCCT reconstructions of the PDAC model imposed over two transverse PCCT slices specific concentration of IntegriSense 680 in the pancreatic tumor. (c) and (d) Corresponding FMT-XCT reconstruction. (e) and (f) Fluorescence cryoslice images (enhanced for contrast) imposed over cryo-section photographs. (g) and (h) Enlarged views of PCCT slices, where the white, orange, light green, pink, and black markers designate pancreatic tumor, spleen, left kidney, stomach, and colon, respectively. (i) Left sagittal photograph of the mouse. (j) and (k) 3-D FMT-PCCT of the pancreatic tumor (extremities designated by white arrows) where segments from PCCT are color coded as indicated in the upper right color legend (micro-CT skeleton is used in (j) and (k), due to its slightly larger field of view). (l) Mouse outer boundary (skin) with green and gray arrows showing view angles of (j) and (k), respectively. (m) 3-D representation of cryoslicing-based validation. Coordinate system is designated by D (dorsal), V (ventral), Cr (cranial), Cd (caudal), L (left), and R (right). Video demonstration of (e) available is in supplementary video FMT-PCCT_PDAC.wmv.

as the animal surface and bone structure and appear different in shape and placement than the ones observed in Fig. 4(a) and (b). Regardless of the surface bias seen on Fig. 4(c) and (d), the fluorescence activity shown appears close to the same area of the animal containing the pancreas. Fig. 4(e) and (f) show confirming cryoslicing images, revealing marked similarity

with the findings of the FMT-PCCT images. This result offers a clear assessment on the benefits of using elaborate soft-tissue contrast as obtained by PCCT, as is common in abdominal imaging applications. Enlarged portions of the PCCT slices are shown in Fig. 4(g) and (h), where the spleen, pancreas, stomach, left kidney, and colon are marked with orange, white,


1441

Fig. 5. Profiles of the fluorescence signals reconstructed with FMT-XCT and FMT-PCCT versus cryoslicing-based validation results: (a) FMT-PCCT reconstruction of the 4T1 model with a priori anatomical information extracted from micro-CT showing fluorescence signal as transparent pseudo-orange overlay on the PCCT slice. (b) FMT-XCT reconstruction of the 4T1 model with a priori anatomical information extracted from micro-CT. (c) Cryoslicing-based validation showing contrast-enhanced fluorescence image of the cryo-section as pseudogreen overlay on the color photograph. (d) Fluorescence signal profiles of FMT-PCCT (blue curve), FMT-XCT (red curve), and cryoslicing-based validation (green curve) normalized to their respective maxima, along the dotted line segments on (a)–(c), respectively. (e)–(h) Results for the PDAC model, where no a prior information was available to the FMT-XCT due to lack of soft-tissue contrast in CT.

green, pink, and black, respectively. Fig. 4(i) shows a Born normalized transillumination image (defined as the measured fluorescence image divided by the measured intrinsic image [59]), for a typical source location. The specific accumulation of the fluorescence probe in the pancreatic tumor has contributed to the increased signal intensity. Fig. 4(j) illustrates a 3-D demonstration of the FMT-PCCT reconstruction results of the PDAC model. The results show a large concentration of IntegriSense 680 in the pancreatic tumor. The view angles of Fig. 4(j) and (k) are marked on the animal outer boundary extracted from micro-CT (due to its larger field of view), as shown in Fig. 4(l). Finally, Fig. 4(m) shows a 3-D demonstration of the validation results, whereby the signal shown in color in the upper part of the image shows fluorescence concentration in the PDAC and the lower signal is due to probe accumulation in the bladder and peritoneal cavity.

shown in Fig. 5(d), where blue, red, and green curves denote FMT-PCCT, FMT-XCT and validation results, respectively, versus distance (in mm) from rightmost white square markers. Anatomical priors were used in inversion of both FMT-PCCT and FMT-XCT. As observed both hybrid modalities achieve satisfactory conformity with the cryoslicing-based validation results. Corresponding results for the PDAC model are shown in Fig. 5(e)–(h), where anatomical priors extracted from PCCT were used in FMT-PCCT inversion [Fig. 5(e)], and only boundary information as available from XCT was employed for the FMT-XCT results of Fig. 5(f), due to lack of soft-tissue contrast. As a result, while the FMT-PCCT reconstructions closely match validation results as seen in Fig. 5(h), the FMT-XCT reconstructions correctly resolve the approximate area of high fluorescence activity but offer a surface bias which leads to an inaccurate spatial appearance of the signals.

C. Comparisons with Ex Vivo Validations

D. Contrast Analysis

Fig. 5 shows the intensity profiles of FMT-PCCT and FMT-XCT reconstructions for both animal models as compared to cryoslicing-based validation results Fig. 5(a) shows an FMT-PCCT slice, with the fluorescence signal as transparency overlay on the PCCT slice, and Fig. 5(b) and (c) depicts, respectively, FMT-XCT reconstructions and validation results for the 4T1 model. The fluorescence intensity profiles along the dotted white line segments in all three images are

Here we further investigate the imaging performance difference between FMT-XCT and FMT-PCCT. Fig. 6 presents the results of the application of the reconstruction error metric (presented in Section II-D) to hybrid FMT reconstructions of both animal models. The contrast and dispersion measures and were calculated for the cryoslicing-based validation results within an axial field of view of 6 mm along the pancreatic tumor for the PDAC model and a field of view of

1442

Fig. 6. Reconstruction error (hybrid FMT versus cryoslicing-based validation) for imaging the PDAC model using FMT-PCCT (black solid curve) and FMT-XCT (red solid curve) and the normalized curvature of the L-curve (blue solid curve) expressed in percentage and plotted versus the regularization parameter ; the point of maximum curvature occurring at the 135th , employed in the reconstruction demonstrations of Figs. 3 and 4 is denoted on the curves with diamond markers. Corresponding results for the 4T1 model are shown with dashed curves, where the point of maximum curvature occurring at the 148th is denotes on the three dashed curves with square markers. The red and green dashed curve denotes the reconstruction error for imaging the 4T1 model using FMT-XCT without and with inclusion of anatomical information extracted from micro-CT. Displayed range of is limited for better demon’s logarithmically spaced stration to the 70th and the 200th of a total of and —the reconstruction errors for beyond the between displayed range of remain constant and hence not shown.

4 mm covering the subcutaneous tumor for the 4T1 model. The contrast and dispersion measures were calculated as well in the same axial fields of view for the FMT-XCT and FMT-PCCT reconstruction. The reconstruction error was found according to (7) for the entire range of the regularization parameter [refer to (1)] for both models. As seen in Fig. 2, the target region is not recognizable in the XCT scan of the PDAC model. Therefore, to calculate the reconstruction error metric for the FMT-XCT imaging of the PDAC model, the target region was obtained from the PCCT volume. That is, the reconstruction error for FMT-XCT imaging of the PDAC model was obtained using the geometrical boundary of the PCCT as a volumetric reference. The solid black and red curves in Fig. 6 denote reconstruction error for the PDAC model for using FMT-PCCT and FMTXCT, respectively, and the solid blue curve shows normalized curvature of the L-curve, with the diamond markers showing the 135th corresponding to the point of maximum curvature. Results corresponding to the 4T1 model are shown with dashed curves and square markers with the point of maximum curvature occurring at the 148th . Since micro-CT has detectable contrast in the region of the subcutaneous tumor—although with much lower soft-tissue contrast than PCCT as observed in Fig. 4—it is possible to achieve equally accurate reconstructions using FMT-XCT in this case. The green dashed curve of Fig. 6 depicts the reconstruction error when the micro-CT anatomical information is employed in inversion of FMT-XCT, as described


in Section II-C. The points of maximum curvature do not necessarily correspond to points of minimum reconstruction error, as observed in Fig. 6. For the FMT-PCCT reconstruction, the points of maximum curvature marked on, correspond to reconstruction errors of 49% and 54% for the 4T1 and PDAC models, respectively, while the respective minima across the entire range of are 31% and 46%. The corresponding reconstruction errors for FMT-XCT results are 132% and 110% for 4T1 and PDAC models, respectively, with global minima of 111% and 110%. In all cases, the reconstruction error for large values of is dominated by the error in the contrast measure (with respect to the contrast of the true distribution) and by dispersion error for small values of created by large artifacts. The regularization parameter corresponding to the points of maximum curvature were used for the reconstruction presented in Figs. 2 and 3. FMT-XCT shows a larger error than FMT-PCCT for small ’s because of different number and strength of artifacts arising in the two cases. IV. DISCUSSION We demonstrated for the first time hybrid fluorescence molecular tomography—X-ray phase-contrast computed tomography (FMT-PCCT) for noninvasive, 3-D molecular imaging of biodistribution of fluorophores small animals. The results were demonstrated here for two animal experiments at two different excitation wavelengths. The PCCT images showed a much higher soft-tissue contrast than the conventional CT images. This capability of PCCT afforded accurate reconstruction of cancerous lesions in both models through incorporation of anatomical information segmented from the PCCT in the FMT reconstruction process. FMT-PCCT demonstrated imaging performance, not otherwise achievable or demonstrated with the state of the art in fluorescence tomography. As noted previously, other anatomical imaging modalities have also been employed and combined with FMT in sequential mode or as hybrid systems to improve image quality. In particular, MRI, due to its excellent soft-tissue contrast, can contribute to the quality of FMT imaging. While it is not the focus of this work to compare different aspects of CT and PCCT versus MRI, we believe FMT-PCCT can offer several improvements over FMT-MRI. Cost efficiency is a first significant advantage of PCCT over MRI. Our prototype PCCT system uses similar imaging components as a general micro-CT device. In fact, it is possible to construct a PCCT system by slightly modifying a conventional commercially available small-animal micro-CT scanner and equipping it with gratings [60]. Since the only extra added components are the gratings and motors for the stepping procedure, the eventual cost of a commercial PCCT system is expected to be comparable to a commercial CT system. Hence, FMT-PCCT is expected to be considerably less expensive than FMT-MRI. Most importantly, hybrid integration of FMT and MRI is hindered by several technical challenges [20], [25]. To our knowledge, all previous hybrid FMT-MRI systems have consisted of placing an FMT imaging system within the MRI bore. The technical integration is complicated, as the optical components


and detectors must be MR-compatible and the FMT subsystem should spatially fit inside the micro-MRI bore [25], [26]. While MR-compatible cameras have been developed and employed in the MR bore for purposes such as monitoring, guided intervention and motion compensation [61], [62], integration of a cooled CCD camera with the high magnetic field of MRI can be technically hindering [25]. As a result, the combinations of optical tomography systems with MRI have been mostly limited to application of optical fibers for light detection [21], [22], [63], [64]. Such a detection approach can compromise image quality due to the limited number of measurements acquired [20]. An MR-compatible single-photon avalanche diode (SPAD)-based system has been proposed in [25], operating in front-illumination single-view geometry. A modification of this system using a MR-compatible CMOS camera has been recently presented in [65]. Nevertheless, it is known that the imaging quality of FMT is considerably improved when acquisition is performed in 360 mode [23], [66]. Image acquisition using cooled CCD cameras also results in a much larger data quantity and information than fiber-based systems. However, due to spatial constrains of the MRI as well as the complications of a using a cooled-CCD camera in the high magnetic field of the MRI bore [25], realization of a multi-view, CCD-based FMT-MRI device is technically hindered and has not been demonstrated so far. While MRI and multi-view, CCD-based FMT can be performed sequentially, simultaneous hybrid imaging is preferred due to several advantages [23]. Although in this work we demonstrate FMT-PCCT in sequential mode, a single-view hybrid FMT-PCCT imaging system is readily foreseeable as the three extra gratings can be practically embedded in a FMT-XCT device. This potential constitutes a considerable advantage of FMT-PCCT over FMT-MRI systems. As discussed, a PCCT system is realizable by extending a conventional micro-CT system with gratings whose periods are much smaller than the pixel-size of the micro-CT systems. Therefore, the resolution of a PCCT system remains the same as for a conventional CT system, in all three spatial dimensions. This advantage can make PCCT a more valuable option as the resolution of micro-MRI is fundamentally limited [67], [68]. Moreover, PCCT delivers information about the refractive index decrement, which is directly dependent on the electron density. With a proper calibration, PCCT is capable of quantitative imaging which can easily be scaled into phase-contrast Hounsfield units for clinical purposes [43], [60], [69]. In addition, effective atomic numbers can be evaluated by exploiting the complementary information content of phase and attenuation contrast [70], [71]. Another benefit of PCCT is the darkfield signal which is obtained simultaneously as third imaging modality and can reveal sub-pixel microstructures within an object [33]. As seen, while sustaining the high resolution and low cost of conventional CT, PCCT improves the soft- tissue contrast. This feature can play a critical role in future evolution of FMT-PCCT as a preclinical whole body in vivo imaging modality. An important current issue toward wider deployment of FMT-XCT is

1443

the difficulty associated with segmentation of organs from the CT images due to the low contrast of target lesions relative to the background or the complete lack thereof (as observed in our demonstration of the PDAC model in Fig. 4). This issue especially poses problems for the end users of current FMT systems, as the segmentation of conventional CT segments are mostly based on semi-automatic methods which require user interference. This issue will likely play a smaller role in FMT-PCCT, as the higher soft-tissue contrast will make possible or improve segmentation automation. As observed in Fig. 5, while FMT-XCT and FMT-PCCT have comparable performance in imaging the 4T1 model, FMT-PCCT makes a much higher level of imaging accuracy possible in the case of the PDAC model. The reconstruction errors for FMT-PCCT imaging results, shown by black curves in Fig. 6, are significantly lower for both cases than the respective FMT-XCT results (depicted by red curves) in the vicinity of the maximum curvature points of the L-curves. It should be noted that for the 4T1 model, there is sufficient soft-tissue contrast between the subcutaneous tumor and surrounding tissue in the micro-CT images—a fact which makes FMT-XCT potentially capable of achieving the same performance as FMT-PCCT for this case. In fact, as seen in both Figs. 5 and 6, imaging of the 4T1 model using FMT-XCT when inverted with the help of anatomical information extracted from micro-CT does achieve almost exactly the same reconstruction error as FMT-PCCT. As noted before, a potential advantage of FMT-PCCT over FMT-XCT for the case of the 4T1 model is the increased possibility of automatic segmentation of PCCT compared to XCT. On the other hand, the imaging performance of FMT-PCCT in the case of the PDAC model cannot be otherwise achieved without the anatomical information made possible by PCCT—a case that clearly substantiates the merits of the proposed hybrid approach. Due its depth and surrounding soft tissue organs, pancreatic tissue does not generally lend itself to accurate imaging planar or tomographic optical techniques. As a result, this particularly representative demonstration of ability of FMT-PCCT for molecular imaging in soft tissue is of significant importance for a variety of pancreatic cancer studies. FMT-PCCT’s capability in imaging deep-seated pancreatic tissue along with the use of molecular probes targeting pancreatic beta islet cells, make FMT-PCCT a promising molecular imaging tool for diabetic studies as well [72]–[74]. A current drawback of PCCT is the long acquisition time (20–30 h). The lengthy scanning time is due to yet suboptimal components and acquisition methodology, as the PCCT technology is still in the early development phase. Interestingly, it is noteworthy that one of the first micro-CT scan demonstrated in 1984 also lasted 30 h [75]. Several technological developments can advance the current suboptimal components and lengthy procedures and, accordingly, lead to reduction in PCCT imaging time. Ongoing research on acquisition procedures other than the stepping approach—which constitutes a time consuming part in our PCCT system—aims to extract the phase-information with a notable reduction in scanning time [76]. New iterative reconstruction techniques are being investigated, which aim at sig-

1444


nificantly reducing the required number of angular projections per tomography scan [77]. Additionally, the detector used in this work has a relatively low quantum efficiency compared to detectors used in medical CT. The use of more efficient detectors along with more reliable grating structures can, therefore, enable shorter exposure times. Such optimization of system parameters and components and imaging methodology toward in vivo phase-contrast imaging is a topic of ongoing research [60], [78]. In general, X-ray phase-contrast imaging and tomography are rapidly evolving technologies, which are still undergoing intense development. The first in vivo phase-contrast (nontomographic) imaging of mice has been recently demonstrated [79]. Different aspects of X-ray phase-contrast imaging and tomography are under investigation for different preclinical and clinical applications [80], and particularly in comparison with MRI in terms of resolution and soft-tissue contrast [81], [82]. The current development is likely to lead to in vivo PCCT in future. It should be noted that a post-mortem fixation time of several weeks was required in this study for two reasons; first, to be able to preserve the sample in the room temperature during the lengthy acquisition of the current PCCT system—a goal which can be achieved after few days of fixation—and secondly, to prevent artifacts arising due to air bubbles occurring in the gastrointestinal tract. It is envisaged that with future PCCT system capable of in vivo imaging along with artifact removal algorithms [83], the need to fixate the animal can be removed as well. The optical imaging and phase-contrast imaging were carried out in our work in vivo and ex vivo, respectively. The consequent volumetric bodily deformations, which are not compensable using spatial rigid transformations, contribute to increased forward modeling error in FMT-PCCT. Nevertheless, the results and the accompanying processing framework presented in this work substantiate the potential of this hybrid modality in small animal molecular imaging. Fully in vivo imaging is expected to reduce such modeling imperfections and, hence, further improve imaging quality of FMT-PCCT. Overall, FMT-PCCT was presented and demonstrated for the first time for imaging animal models. The FMT and PCCT imaging were performed in vivo and ex vivo, respectively, and the imaging results were cross-validated with ex vivo cryoslicing. FMT-PCCT demonstrates the ability to effectively reconstruct the target fluorescence signal through incorporation of anatomical information afforded by PCCT in lesions, where low or no contrast relative to surrounding tissue is observed in conventional CT images. Due to low cost, high sensitivity and ability to tomographically image fluorescence in soft tissue lesions, we believe FMT-PCCT equipped with better segmentation methods afforded by high soft-tissue contrast of PCCT will be an important tool for preclinical molecular imaging in future. ACKNOWLEDGMENT P. Mohajerani acknowledges and thanks S. Glasl, A. Oancea, M. Koch, and A. Ale at HMGU for their helps and discussions.

REFERENCES [1] B. Brooksby et al., “Imaging breast adipose and fibroglandular tissue molecular signatures by using hybrid MRI-guided near-infrared spectral tomography,” in Proc. Nat. Acad. Sci., 2006, vol. 103, pp. 8828–8833. [2] B. A. Brooksby, H. Dehghani, B. W. Pogue, and K. D. Paulsen, “Nearinfrared (NIR) tomography breast image reconstruction with a priori structural information from MRI: Algorithm development for reconstructing heterogeneities,” IEEE J. Sel. Topics Quantum Electron., vol. 9, no. 2, pp. 199–209, Mar./Apr. 2003. [3] A. Ale, V. Ermolayev, E. Herzog, C. Cohrs, M. H. de Angelis, and V. Ntziachristos, “FMT-XCT: In vivo animal studies with hybrid fluorescence molecular tomography-X-ray computed tomography,” Nat. Methods, vol. 9, pp. 615–620, Jun. 2012. [4] V. Ntziachristos, C. Bremer, and R. Weissleder, “Fluorescence imaging with near-infrared light: New technological advances that enable in vivo molecular imaging,” Eur. Radiol., vol. 13, pp. 195–208, Jan. 2003. [5] T. Beyer et al., “A combined PET/CT scanner for clinical oncology,” J. Nucl. Med., vol. 41, pp. 1369–1379, 2000. [6] M. S. Judenhofer et al., “Simultaneous PET-MRI: A new approach for functional and morphological imaging,” Nature Med., vol. 14, pp. 459–465, 2008. [7] V. Ntziachristos, “Going deeper than microscopy: The optical imaging frontier in biology,” Nature Methods, vol. 7, pp. 603–614, Aug. 2010. [8] M. Nahrendorf et al., “Hybrid PET-optical imaging using targeted probes,” Proc. Nat. Acad. Sci., vol. 107, pp. 7910–7915, 2010. [9] V. Ntziachristos, J. Ripoll, L. V. Wang, and R. Weissleder, “Looking and listening to light: The evolution of whole-body photonic imaging,” Nature Biotechnol., vol. 23, pp. 313–320, 2005. [10] R. B. Schulz et al., “Hybrid system for simultaneous fluorescence and x-ray computed tomography,” IEEE Trans. Med. Imag., vol. 29, no. 2, pp. 465–473, Feb. 2010. [11] S. C. Davis, H. Dehghani, J. Wang, S. Jiang, B. W. Pogue, and K. D. Paulsen, “Image-guided diffuse optical fluorescence tomography implemented with Laplacian-type regularization,” Opt. Exp., vol. 15, pp. 4066–4082, 2007. [12] D. Hyde et al., “Hybrid FMT-CT imaging of amyloid-beta plaques in a murine Alzheimer’s disease model,” Neuroimage, vol. 44, pp. 1304–1311, Feb. 15, 2009. [13] P. K. Yalavarthy, B. W. Pogue, H. Dehghani, C. M. Carpenter, S. Jiang, and K. D. Paulsen, “Structural information within regularization matrices improves near infrared diffuse optical tomography,” Opt. Exp., vol. 15, pp. 8043–8058, 2007. [14] Y. Lin, H. Gao, O. Nalcioglu, and G. Gulsen, “Fluorescence diffuse optical tomography with functional and anatomical a priori information: Feasibility study,” Phys. Med. Biol., vol. 52, p. 5569, 2007. [15] F. Leblond, S. C. Davis, P. A. Valdés, and B. W. Pogue, “Pre-clinical whole-body fluorescence imaging: Review of instruments, methods and applications,” J. Photochem. Photobiol. B: Biol., vol. 98, pp. 77–94, 2010. [16] D. Kepshire et al., “A microcomputed tomography guided fluorescence tomography system for small animal molecular imaging,” Rev. Sci. Instrum., vol. 80, pp. 043701–043701-10, 2009. [17] P. Mohajerani and V. Ntziachristos, “Compression of Born ratio for fluorescence molecular tomography/X-ray computed tomography hybrid imaging: Methodology and in vivo validation,” Opt. Lett., vol. 38, pp. 2324–2326, Jul. 1, 2013. [18] X. Zhang, C. T. Badea, G. Hood, A. W. Wetzel, J. R. Stiles, and G. A. Johnson, “Free-space fluorescence tomography with adaptive sampling based on anatomical information from microCT,” in Proc. Soc. Photo Opt. Instrum. Eng., 2010, vol. 7757. [19] X. Liu, B. Zhang, J. Luo, and J. Bai, “4-D reconstruction for dynamic fluorescence diffuse optical tomography,” IEEE Trans. Med. Imag., vol. 31, no. 11, pp. 2120–2132, 2012. [20] F. Stuker, J. Ripoll, and M. Rudin, “Fluorescence molecular tomography: Principles and potential for pharmaceutical research,” Pharmaceutics, vol. 3, pp. 229–274, 2011. [21] S. C. Davis et al., “MRI-coupled fluorescence tomography quantifies EGFR activity in brain tumors,” Acad. Radiol., vol. 17, pp. 271–276, 2010. [22] S. C. Davis et al., “Magnetic resonance—Coupled fluorescence tomography scanner for molecular imaging of tissue,” Rev. Sci. Instrum., vol. 79, pp. 064302–064302-10, 2008.


[23] M. Niedre and V. Ntziachristos, “Elucidating structure and function in vivo with hybrid fluorescence and magnetic resonance imaging,” Proc. IEEE, vol. 96, no. 3, pp. 382– 396, Mar. 2008. [24] B. J. Tromberg, B. W. Pogue, K. D. Paulsen, A. G. Yodh, D. A. Boas, and A. E. Cerussi, “Assessing the future of diffuse optical imaging technologies for breast cancer management,” Med. Phys., vol. 35, p. 2443, 2008. [25] F. Stuker et al., “Hybrid small animal imaging system combining magnetic resonance imaging with fluorescence tomography using single photon avalanche diode detectors,” IEEE Trans. Med. Imag., vol. 30, pp. 1265–1273, Jun. 2011. [26] F. Stuker, “Hybrid imaging: Combining fluorescence molecular tomography with magnetic resonance imaging,” Diss., ETH Zürich, Zürich, Switzerland, 2011. [27] A. Momose, S. Kawamoto, I. Koyama, Y. Hamaishi, K. Takai, and Y. Suzuki, “Demonstration of X-ray Talbot interferometry,” Jpn. J. Appl. Phys., vol. 42, pt. 2, pp. L866–L868, Jul. 15, 2003. [28] T. Weitkamp et al., “X-ray phase imaging with a grating interferometer,” Opt. Exp., vol. 13, pp. 6296–6304, Aug. 8, 2005. [29] F. Pfeiffer, T. Weitkamp, O. Bunk, and C. David, “Phase retrieval and differential phase-contrast imaging with low-brilliance X-ray sources,” Nature Phys., vol. 2, pp. 258–261, Apr. 2006. [30] F. Pfeiffer et al., “High-resolution brain tumor visualization using three-dimensional X-ray phase contrast tomography,” Phys. Med. Biol., vol. 52, pp. 6923–6930, Dec. 7, 2007. [31] G. Schulz et al., “High-resolution tomographic imaging of a human cerebellum: Comparison of absorption and grating-based phase contrast,” J. R. Soc. Interface, vol. 7, pp. 1665–1676, Dec. 6, 2010. [32] J. Herzen et al., “Quantitative phase-contrast tomography of a liquid phantom using a conventional X-ray tube source,” Opt. Exp., vol. 17, pp. 10010–10018, Jun. 2009. [33] F. Pfeiffer et al., “Hard-X-ray dark-field imaging using a grating interferometer,” Nature Mater., vol. 7, pp. 134–137, Feb. 2008. [34] M. Trajkovic-Arsic et al., “Molecular imaging of integrin avß3 for in vivo detection of precursor lesions and pancreatic cancer,” J. Nucl. Med., vol. 55, 2014. [35] A. J. Beer and M. Schwaiger, “Imaging of integrin v 3 expression,” Cancer Metastasis Rev., vol. 27, pp. 631–644, 2008. [36] R. Weissleder, “Molecular imaging in cancer,” Science, vol. 312, pp. 1168–1171, 2006. [37] R. Weissleder and M. J. Pittet, “Imaging in the era of molecular oncology,” Nature, vol. 452, pp. 580–589, 2008. [38] J. Condeelis and R. Weissleder, “In vivo imaging in cancer,” Cold Spring Harbor Perspect. Biol., vol. 2, 2010. [39] C. M. Ardito et al., “EGF receptor is required for KRAS-induced pancreatic tumorigenesis,” Cancer Cell, vol. 22, pp. 304–317, Sept. 11, 2012. [40] N. Bardeesy et al., “Both p16(Ink4a) and the p19(Arf)-p53 pathway constrain progression of pancreatic adenocarcinoma in the mouse,” in Proc. Nat. Acad. Sci. USA, Apr. 11, 2006, vol. 103, pp. 5947–5952. [41] T. Donath et al., “Inverse geometry for grating-based x-ray phase-contrast imaging,” J. Appl. Phys., vol. 106, Sep. 1, 2009. [42] W. Bacher, W. Menz, and J. Mohr, “The liga technique and its potential for microsystems—A survey,” IEEE Trans. Ind. Electron., vol. 42, pp. 431–441, Oct. 1995. [43] T. Donath et al., “Toward clinical X-ray phase-contrast CT: Demonstration of enhanced soft-tissue contrast in human specimen,” Invest. Radiol., vol. 45, pp. 445–452, Jul. 2010. [44] I. Zanette et al., “Quantitative phase and absorption tomography with an X-ray grating interferometer and synchrotron radiation,” Physica Status Solidi a-Appl. Mater. Sci., vol. 208, pp. 2526–2532, Nov. 2011. [45] A. Sarantopoulos, G. Themelis, and V. Ntziachristos, “Imaging the biodistribution of fluorescent probes using multispectral epi-illumination cryoslicing imaging,” Molec. Imag. Biol., vol. 13, pp. 874–885, Oct. 2011. [46] I. Takamasa, Y. Hiroshi, and H. Toshiyuki, A Color Atlas of Sectional Anatomy of the Mouse. Tokyo, Japan: Adthree, 2001. [47] D. J. Kroon and C. H. Slump, “MRI modalitiy transformation in demon registration,” in IEEE Int. Symp. Biomed. Imag.: From Nano to Macro, 2009, pp. 963–966. [48] D. Hyde, R. Schulz, D. Brooks, E. Miller, and V. Ntziachristos, “Performance dependence of hybrid X-ray computed tomography/fluorescence molecular tomography on the optical forward problem,” J. Opt. Soc. Am. a—Opt. Image Sci. Vis., vol. 26, pp. 919–923, Apr. 2009.

1445

[49] Q. Q. Fang and D. A. Boas, “Tetrahedral Mesh Generation from Volumetric Binary and Gray-Scale Images,” in Proc. IEEE Int. Symp. Biomed. Imag.: From Nano to Macro, 2009, pp. 1142–1145. [50] P. Alliez, L. Rineau, S. Tayeb, J. Tournois, and M. Yvinec, 3D mesh generation, 2012, CGAL User and Reference Manual. [51] P. Mohajerani, A. A. Eftekhar, J. Huang, and A. Adibi, “Optimal sparse solution for fluorescent diffuse optical tomography: Theory and phantom experimental results,” Appl. Opt., vol. 46, no. 10, pp. 1679–1685, 2007. [52] H. Jiang, “Frequency-domain fluorescent diffusion tomography: A finite-element-based algorithm and simulations,” Appl. Opt., vol. 37, p. 5337, 1998. [53] S. R. Arridge, M. Schweiger, M. Hiraoka, and D. T. Delpy, “A finite element approach for modeling photon transport in tissue,” Med. Phys., vol. 20, p. 299, 1993. [54] C. C. Paige and M. A. Saunders, “LSQR: An algorithm for sparse linear equations and sparse least squares,” ACM Trans. Math. Software, vol. 8, pp. 43–71, 1982. [55] D. Hyde, E. L. Miller, D. H. Brooks, and V. Ntziachristos, “Data specific spatially varying regularization for multimodal fluorescence molecular tomography,” IEEE Trans. Med. Imag., vol. 29, no. 2, pp. 365–374, Feb. 2010. [56] P. C. Hansen, Rank-Deficient and Discrete Ill-Posed Problems: Numerical Aspects of Linear Inversion. Philadelphia, PA: SIAM, 1998, vol. 4. [57] H. Abdi, “Coefficient of variation,” in Encyclopedia of Research Design. Thousand Oaks, CA: SAGE, 2010, pp. 169–171. [58] L. M. Kegelmeyer, P. W. Fong, S. M. Glenn, and J. A. Liebman, “Local area signal-to-noise ratio (LASNR) algorithm for image segmentation,” Opt. Eng. Appl., pp. 66962H–66962H-9, 2007. [59] V. Ntziachristos and R. Weissleder, “Experimental three-dimensional fluorescence reconstruction of diffuse media by use of a normalized Born approximation,” Opt. Lett., vol. 26, pp. 893–895, 2001. [60] A. Tapfer et al., “Experimental results from a preclinical X-ray phasecontrast CT scanner,” in Proc. Nat. Acad. Sci. USA, Sep. 25, 2012, vol. 109, pp. 15691–15696. [61] P. Piferi, C. Keidl, D. M. Peterson, and D. Sayler, “Imaging device for MRI-guided medical interventional systems,” Ed: Google Patents 2012. [62] M. Aksoy et al., “Real-time optical motion correction for diffusion tensor imaging,” Magn. Reson. Med., vol. 66, pp. 366–378, 2011. [63] H. Xu, R. Springett, H. Dehghani, B. W. Pogue, K. D. Paulsen, and J. F. Dunn, “Magnetic-resonance-imaging-coupled broadband near-infrared tomography system for small animal brain studies,” Appl. Opt., vol. 44, pp. 2177–2188, 2005. [64] Y. Lin, M. Ghijsen, H. Gao, N. Liu, O. Nalcioglu, and G. Gulsen, “A photo-multiplier tube-based hybrid MRI and frequency domain fluorescence tomography system for small animal imaging,” Phys. Med. Biol., vol. 56, p. 4731, 2011. [65] A. Dikaiou, “Biomedical applications of a hybrid fluorescence molecular tomography/magnetic resonance imaging (FMT/MRI) system for small animal imaging,” Diss., ETH Zürich, Zürich, Switzerland , 2013. [66] E. E. Graves, J. P. Culver, J. Ripoll, R. Weissleder, and V. Ntziachristos, “Singular-value analysis and optimization of experimental parameters in fluorescence molecular tomography,” J. Opt. Soc. Am. A: Opt. Image Sci., Vis., vol. 21, pp. 231–241, 2004. [67] A. Tapfer et al., “Three-dimensional imaging of whole mouse models: Comparing nondestructive X-ray phase-contrast micro-CT with cryotome-based planar epi-illumination imaging,” J. Microsc., vol. 253, pp. 24–30, 2013. [68] J. M. Tyszka, S. E. Fraser, and R. E. Jacobs, “Magnetic resonance microscopy: Recent advances and applications,” Curr. Opin. Biotechnol., vol. 16, pp. 93–99, 2005. [69] M. Willner et al., “Quantitative breast tissue characterization using grating-based X-ray phase-contrast imaging,” Phys. Med. Biol., 2014, accepted for publication. [70] Z. Qi, J. Zambelli, N. Bevins, and G. H. Chen, “Quantitative imaging of electron density and effective atomic number using phase contrast CT,” Phys. Med. Biol., vol. 55, pp. 2669–2677, May 7, 2010. [71] M. Willner et al., “Quantitative X-ray phase-contrast computed tomography at 82 keV,” Opt. Exp., vol. 21, pp. 4155–4166, 2013.

1446

[72] B. W. Paty, S. Bonner-Weir, M. R. Laughlin, A. J. Mcewan, and A. J. Shapiro, “Toward development of imaging modalities for islets after transplantation: Insights from the national institutes of health workshop on beta cell imaging,” Transplantation, vol. 77, pp. 1133–1137, 2004. [73] J. H. Hwang et al., “In vivo imaging of islet transplantation using PLGA nanoparticles containing iron oxide and indocyanine green,” Magn. Reson. Med., 2013. [74] D. Lee et al., “Dual microfluidic perifusion networks for concurrent islet perifusion and optical imaging,” Biomed. Microdevices, vol. 14, pp. 7–16, 2012. [75] J. Elliott and S. Dover, “X-ray microtomography of bone,” Metabolic Bone Disease Related Res., vol. 5, p. 50, 1984. [76] I. Zanette et al., “Trimodal low-dose X-ray tomography,” Proc. Nat. Acad. Sci., vol. 109, pp. 10199–10204, 2012. [77] M. Nilchian, C. Vonesch, P. Modregger, M. Stampanoni, and M. Unser, “Fast iterative reconstruction of differential phase contrast X-ray tomograms,” Opt. Exp., vol. 21, pp. 5511–5528, Mar. 11, 2013.


[78] A. Yaroshenko et al., “Pulmonary emphysema diagnosis with a preclinical small-animal X-ray dark-field scatter-contrast scanner,” Radiology, May 21, 2013. [79] M. Bech et al., “In-vivo dark-field and phase-contrast x-ray imaging,” Sci. Rep., vol. 3, 2013. [80] A. Bravin, P. Coan, and P. Suortti, “X-ray phase-contrast imaging: From pre-clinical applications towards clinics,” Phys. Med. Biol., vol. 58, pp. R1–R35, Jan. 7, 2013. [81] T. Takeda et al., “Interferometer-based phase-contrast X-ray computed tomography of colon cancer specimens: Comparative study with 4.74-T magnetic resonance imaging and optical microscopy,” J. Comput. Assist. Tomogr., vol. 31, pp. 214–217, 2007. [82] G. Schulz et al., “Multimodal imaging of human cerebellum-merging X-ray phase microtomography, magnetic resonance microscopy and histology,” Sci. Rep., vol. 2, 2012. [83] F. E. Boas and D. Fleischmann, “CT artifacts: Causes and reduction techniques,” Imag. Med., vol. 4, pp. 229–240, 2012.