Abstract
In vivo tumor labeling with fluorescent agents may assist endoscopic and surgical guidance for cancer therapy as well as create opportunities to directly observe cancer biology in patients. However, malignant and nonmalignant tissues are usually distinguished on fluorescence images by applying empirically determined fluorescence intensity thresholds. Here, we report the development of fSTREAM, a set of analytic methods designed to streamline the analysis of surgically excised breast tissues by collecting and statistically processing hybrid multiscale fluorescence, color, and histology readouts toward precision fluorescence imaging. fSTREAM addresses core questions of how to relate fluorescence intensity to tumor tissue and how to quantitatively assign a normalized threshold that sufficiently differentiates tumor tissue from healthy tissue. Using fSTREAM we assessed human breast tumors stained in vivo with fluorescent bevacizumab at microdose levels. Showing that detection of such levels is achievable, we validated fSTREAM for high-resolution mapping of the spatial pattern of labeled antibody and its relation to the underlying cancer pathophysiology and tumor border on a per patient basis. We demonstrated a 98% sensitivity and 79% specificity when using labeled bevacizumab to outline the tumor mass. Overall, our results illustrate a quantitative approach to relate fluorescence signals to malignant tissues and improve the theranostic application of fluorescence molecular imaging. Cancer Res; 77(3); 623–31. ©2016 AACR.
Introduction
Clinical translation of fluorescence agents that target cancer has the potential to guide surgical and endoscopy procedures, improving upon the limitations of human vision (1, 2). Several studies have received regulatory approvals to administer non–FDA-approved targeted fluorescent agents to humans (see ClinicalTrials.gov NCT01508572, NCT02113202, NCT01972373, NCT02129933, NCT01987375, NCT02415881). Applied topically or systemically, targeted fluorescence agents are expected to change the landscape of interventional guidance by steering biopsy, improving disease detection and driving accurate theranostics (3). Several studies have recently demonstrated the potential of using targeted fluorescent reporters to guide human surgery and endoscopy (3–7).
In addition to intraoperative guidance, in vivo cancer staining in patients scheduled for surgery or endoscopy may offer new insights in tumor physiology and agent biodistribution. Contrary to tissue histopathology that uses ex vivo staining, histologic analysis of human specimen stained in vivo can reveal functional characteristics of the tumor and its microenvironment associated with agent delivery, biodistribution and targeting, on a per-patient basis and at resolutions not available to macroscopic optical imaging or radiologic imaging.
However, an important and so far unsolved problem in in vivo fluorescence cancer imaging is the uncertain relation of fluorescence intensity to the underlying tumor extent. Fluorescence images obtained in vivo are diffusive (low resolution) in nature and do not delineate tumors with precision. Consequently, it becomes challenging to set an unbiased fluorescence threshold that will allow explicit differentiation of malignant from healthy tissue. As fluorescence molecular imaging is increasingly considered for clinical application, it becomes critical to develop methods that lead to accurate tissue classification and reveal the relation between administered agent and tumor extent.
In this work, we developed an analytic method that sought to deliver, objective criteria for fluorescence-based disease detection and differentiation from nondiseased tissues, toward precision fluorescence imaging. The study was based on clinical molecular imaging of breast cancer using fluorescence-labeled bevacizumab and had three objectives. First, it sought to establish the spatial relationship between the distribution of the antibody and the underlying tumor spatial extent, as it relates to the intraoperative identification of human breast cancer and cancer margins and quantitatively identify a fluorescence threshold for distinguishing malignant from nonmalignant tissue. Of particular interest herein was the development of a fluorescence threshold that is normalized and is not affected by variations of the amount of agent administered or agent dilution variations in each patient. Second, it inquired, in high-resolution, the spatial-pattern of labeled antibody within human cancer, a parameter not previously resolved by radiologic methods and generally unknown due to the effects of interstitial pressure gradients and abnormal tumor vascularization. Finally, the study identified the sensitivity and specificity of cancer detection using bevacizumab-IRDye800CW and quantified the ability to detect fluorescent agents administered at micro-dosing amounts. The underlying threshold analysis and biodistribution study was part of a clinical molecular imaging study of breast cancer using fluorescence-labeled bevacizumab in 20 patients with breast cancer of whom 19 patients were eligible for analysis. Tracer application was concluded to be safe and details are described in Lamberts and colleagues (1).
Patients and Methods
Clinical study
We administered bevacizumab conjugated to the near-infrared (NIR) fluorescent dye IRDy800CW (i.e., bevacizumab-IRDye800CW) to breast cancer patients. Drug labeling was achieved at the University Medical Center Groningen under therein established Good Manufacturing Practices (GMP; ref. 2). Bevacizumab-IRDye800CW is already considered in clinical trials for surgical and endoscopic guidance (ClinicalTrials.gov numbers: NCT01508572, NCT01972373, and NCT02113202). Nineteen patients diagnosed with breast cancer at mean age 64.6 years (10.26 years SD) have been included in this analysis. Tumor sizes, evaluated at pathology, averaged at 20.1 mm (7.9 mm SD). Consistent with micro-dosing regulations, as described in the FDA guidelines “Guidance for Industry, Investigators, and Reviewers Exploratory IND Studies,” 4.5 mg of labeled bevacizumab (<30 nmol of labeled antibody) were administered to patients 3 days before surgery (see Supplementary Section for details).
Development of fSTREAM
To achieve the study objectives, we developed comprehensive analysis of fluorescent human tissue specimen at multiple scales (Fig. 1). Termed fSTREAM, the analysis coregisters (i) color images, (ii) fluorescence images, (iii) hematoxylin and eosin (H&E) mosaic–stained microscopy slices, and (iv) the pathologists demarcation border obtained from an excised cancer specimen, onto a common geometrical frame. Subsequently, statistical processing of the 4-modal hybrid image was carried-out to guide threshold selection by studying bevacizumab-IRDye800CW distribution in human breast cancer. The study obtained multiscale measurements from tissue specimen at the macroscopic, mesoscopic, and microscopic scales, that is:
Macroscopic tumor imaging, based on a real-time color and fluorescence composite EagleRay-V3 camera, custom-developed at the Helmholtz Zentrum München and Technical University of Munich (see Supplementary Material; refs. 3, 4) and approved for clinical use by the Institutional Review Board of the University Medical Center Groningen. EagleRay-V3 video and images were simultaneously acquired during surgery (Fig. 1B) at a field of view (FOV) of approximately 15 cm × 15 cm and resolution of 150 μm. Immediately after the excision, the surgical specimen was placed on a table and imaged (Fig. 1C) with the same EagleRay-V3 parameters.
Sensitivity . | Specificity . | Threshold of S (presence of maximum normalized fluorescence value) . |
---|---|---|
0.9000 | 0.9274 | 13.2% |
0.9500 | 0.8775 | 9.8% |
0.9800 | 0.7882 | 6.5% |
Sensitivity . | Specificity . | Threshold of S (presence of maximum normalized fluorescence value) . |
---|---|---|
0.9000 | 0.9274 | 13.2% |
0.9500 | 0.8775 | 9.8% |
0.9800 | 0.7882 | 6.5% |
Mesoscopic EagleRay-V3 color and fluorescence imaging was then performed on freshly excised 3-mm thick lamellae (Fig. 1E) using mesoscopic parameters, that is, 2 cm × 2 cm FOV and 20 micrometer resolution. Shortly after, the lamellae were fixed in paraffin. Paraffin blocks were also imaged with mesoscopic imaging settings (Fig. 1F) using the EagleRay-V3 camera or flat-bed scanning (Odyssey, LI-COR Biosciences). Details are outlined in the Supplementary Material “Fluorescence validation, H&E & VEGF-A Staining.”
Microscopic mosaicking imaging was then performed on consecutive 4-μm thick histologic slices obtained from the paraffin blocks, covering the same field of view as in mesoscopic imaging (Fig. 1G). The histology slices were stained with H&E and tumor demarcation was performed by experienced pathologists. All pathologists were blinded to the fluorescence signals during the classification process. Interleaved histology slices not processed by H&E were stained for VEGF-A expression using the polyclonal IgG VEGF A-20 Antibody (sc-152, Santa Cruz Biotechnology) and imaged with conventional microscopy; (For details see Supplementary Material section “Fluorescence validation, H&E & VEGF-A Staining”).
Color and fluorescence images, obtained from paraffin blocks, were registered onto the corresponding H&E mosaic image obtained from the same field of view. Then the pathologists' tumor segmentation on the H&E image was recorded and registered on the color, fluorescence and H&E images. Image registration was based on affine transformations utilizing six or more anatomic landmarks. The 4-modality coregistration was an essential fSTREAM step to streamline the spatial and intensity correlations between malignant tissue area and fluorescence signals.
Identification of a global threshold
The work herein sought to develop a method that could derive an objective, quantitative global threshold for optimal cancer delineation. The use of a normalized objective global threshold is critical for the clinical application of fluorescence imaging, that is, at conditions where real-time guidance is required. The following analysis was applied to the bevacizumab-IRDye800CW breast cancer study treated herein. Nevertheless, the proposed methodology is generalizable to any fluorescence imaging study.
The first step in deriving a normalized global threshold was image normalization, so that threshold application relates to comparable images independently of gains (intensity of illumination, camera amplification), background noise (ambient light, read noise etc.), and tissue variations or amount of agent administered and patient body weight. We therefore assumed the normalized fluorescence image Si, that is,
whereby Fij is the fluorescence intensity value of the j-th pixel of the |i$|-th image of the original (raw) fluorescence image, assuming a number of fluorescence images |i$| = 1…N obtained from different patients (e.g., Figs. 2E or 3B). The parameters |\alpha ,\;\beta \;$|modulate the normalization of image Fi by its mean value mean(Fi) and a threshold value thres(|{F_i}$|); the latter indicating the value that maximizes the inter-class variance for the |i$|-th image and was determined by the Otsu's method (5) for each Fi image. The parameter |\gamma $| adjusts the image offset, representative of a background constant value due to bias values typically present in CCD camera images. To derive a global threshold, we determined the values |[ {\alpha \;\beta \;\gamma } ]$| by minimizing the cost function C(|[ {\alpha ,\;\beta ,\;\gamma } ]$|) that is:
where G is a binary image, indicating the areas of malignant versus nonmalignant tissue on the image S, as obtained by the congruent H&E pathology segmentation, and AUC is the area under the receiver-operator-characteristics (ROC) curve (the normalized Mann–Whitney-Wilcoxon test). The optimal |[ {\alpha \;\beta \;\gamma } ]$| parameters used in image normalization (Eq. A) were estimated by an unconstrained nonlinear optimization.
ROC analysis
ROC analysis was performed on data from all specimen measured and processed by Eq. A, using the image normalization parameter set |[ {\alpha \;\beta \;\gamma } ]$| obtained by Eq. B. To inspect whether the use of a global threshold did not produce any outliers in the data set examined, we solved (minimized) Eq. B 21 times. Each time, the minimization used 21 samples to derive the global threshold and excluded a different sample from the training set.
Homogeneity of bevacizumab-IRDye800CW distribution
The study further examined the patterns of bevacizumab-IRDye800CW in human breast cancer. To quantify the spatial pattern observed, we calculated entropy values for areas of (0.5 mm)2 on the raw fluorescence tumor images obtained from the paraffin blocks. The bevacizumab-IRDye800CW distribution pattern at a higher resolution was also interrogated by analyzing images acquired from 4-μm slices obtained from each of the 3-mm thick lamellas used for the entropy analysis.
Results
Pattern of bevacizumab-IRDye800CW distribution in breast cancer
Fluorescence images obtained in situ or post-surgery from excised specimen (Fig. 2A and B) exhibited patterns of diffusive appearance. In breast cancer surgery a positive margin is defined as “ink-on-tumor” according to the 2014 SSO/ASTRO guidelines (6); however, a margin of several millimeters to centimeters is allowed around the tumor to ensure complete tumor resection. Therefore, diffusive appearance of fluorescence signals indicated a complete (R0) resection, that is, no ink on tumor. Signals from tumors as deep as at least 1 cm could be nevertheless detected (Fig. 2B).
Mesoscopic imaging of lamellae (Fig. 2C) offered first insights into the distribution parameters of the fluorescent bevacizumab. We observed an apparent, previously undisclosed homogenous distribution of bevacizumab-IRDye800CW throughout the tumor area. The images were obtained according to the step in Fig. 1F and demonstrate that sufficient fluorescence signal could be collected even after paraffin preservation. Occasionally, the processing of human tissue into lamellae would tear tissue in areas of low structural integrity, typically associated with necrotic areas in the tumor center, giving an appearance of openings in the middle of the tumor. Figure 2D and E show images from a paraffin embedded specimen from the rectangular area marked on Fig. 2C. These findings suggest that bevacizumab-IRDye800CW distributed in a diffusive manner throughout the tumor mass, apparently without strong influence by tumor interstitial fluid pressure or irregular perfusion (See also Figs. 3 and 4). The fluorescence images of the lamellae contain light scattering effects due to the sample thickness of 3 mm; therefore, the images in Fig. 2 are diffusive in nature. This is evidenced by inspecting fluorescence intensity profiles (Fig. 2J). Even though the borderline between tumor and healthy tissue on histologic slices is marked by pathologists as a sharp line (Fig. 2F), the fluorescence signals, as expected, do not exhibit sharp borders but a diffuse appearance with no distinct definition of the tumor margin (Fig. 2I and J).
Relation of bevacizumab-IRDye800CW distribution and breast cancer
A next step was to quantitatively relate bevacizumab-IRDye800CW distribution to malignant tissue. Results from patient #8 are shown on Fig. 3A–E for demonstration purposes. Color and fluorescence images (Fig. 3A and B) from the paraffin blocks were registered onto the corresponding H&E mosaic image (Fig. 3D) and the pathologists' tumor segmentation (Fig. 3E), based on affine transformations (see Materials and Methods). Raw fluorescence counts from the tumor and background tissue for patient #8 are shown on Fig. 3F, indicating a target (tumor)-to-background ratio of approximately 2.5.
The tumor-to-background ratio (TBR) for the entire sample analyzed was related to the pathologic Bloom–Richardson–Elston (BRE) tumor grade (Fig. 3G). The BRE is a breast cancer classification metric that indicates cancer aggressiveness and combines measurements of the severity of tubule formation, nuclear pleomorphism and mitotic count per area in the tumor sample analyzed. We observed that tumors of pathologic grade score 1 exhibited less fluorescence uptake compared to tumors of grade 2 and 3. Overall, the TBR values ranged from 1.8 to 9. Fluorescence signals from the paraffin sections of all patients were spatially correlated to the underlying cancer area identified on the coregistered H&E slices. A ROC curve was drawn for each patient sample, assuming different thresholds (Fig. 3H).
Application of a global threshold
Figure 3H depicts the ROC curve for each patient and presents the sensitivity and specificity by which fluorescence intensity patterns demarcate the tumor area, as confirmed by H&E analysis. However, a critical parameter in interventional fluorescence imaging relates to setting an intensity threshold on the fluorescence image to differentiate cancer from healthy tissue, in the absence of an H&E analysis. Today, thresholds are empirically assigned on the basis of image appearance and may inaccurately estimate the tumor extent and surrounding tissue (see Fig. 4A–D).
The derivation of an objective normalized global threshold (see Materials and Methods) determined the optimal |[ {\alpha \;\beta \;\gamma } ]$| parameters (see Eq. A and B), which were estimated to be [0.9240, 1.3103, and 0.0040] for the set of 22 paraffin block samples from the 19 patients examined (double paraffin samples were available in pathology from 3 patients with large tumors). The ROC analysis considered all specimen images processed by Eq. A using image normalization based on the optimal parameter set |[ {\alpha \;\beta \;\gamma } ]$|. The resulting ROC curve (Fig. 4E) achieved an AUC value of 0.97 for all 22 samples examined. We note that skin regions exhibited elevated fluorescence but were not included in the analysis (see Fig. 1 in Supplementary Material for details). To confirm the generality of the global threshold, we computed the AUC variation between an excluded sample and the remaining set of samples (see Materials and Methods). The AUC variation was ± 0.00389%, indicating that the application of the global threshold did not produce any outliers in any of the data sets examined.
Homogeneity of bevacizumab-IRDye800CW distribution
Figures 2 and 3 showcase a rather homogeneous distribution of bevacizumab-IRDye800CW throughout tumors. Entropy calculations (see Materials and Methods) quantified the spatial pattern observed on the raw fluorescence tumor images from paraffin blocks (Fig. 4F). The mean entropy value for all patient samples was 3.06 with an SD of ±0.81 for a (0.5 mm)2 neighborhood; a finding that confirms a homogeneous pattern of labeled drug distribution in all patient samples examined.
Entropy analysis of 4-μm slices obtained from each of the 3-mm thick lamellas employed was also considered for higher-resolution bevacizumab-IRDye800CW observations, because images from the 4-μm slices are virtually scatter-free. The corresponding voxels represented on Fig. 4G are of lower total volume compared with the voxels observed in Fig. 4A–C. The entropy value for a (0.5 mm)2 area within tumors was 6.4374 ± 0.81006. As expected, the entropy increases when observing higher resolution images, but nevertheless showcases also a homogeneous distribution, which remains overtly constant from patient to patient. We further observed that even in high-resolution view, the fluorescence pattern generally matched the H&E outlined tumor border, although it marginally overestimated the tumor border (Fig. 4D). A comparison of magnified views from 4-μm thick versus paraffin block images is shown in Supplementary Fig. S2.
Discussion
fSTREAM was proposed as methodology for standardized analysis of fluorescence images obtained from tissues after the administration of fluorescent agents in vivo. The method was applied to analyze breast cancer specimen labeled with bevacizumab-IRDye800CW, obtained from a phase I clinical trial and guide the selection of a normalized objective global threshold, that is, a single normalized value derived to optimally separate malignant tissue from surrounding healthy tissue in the entire study. It was found that derivation of an objective global threshold required appropriate image normalization, performed by Eq. A, so that the intensity seen on different images is calibrated to the same reference (standard). The derivation of a normalized threshold further ensures a metric that is independent of the exact amount of agent administered, patient body weight or absolute fluorescence intensity values, since it classifies tissues based on relative intensities and in relation to the overall image statistics (Eq. A). It is nevertheless expected that the fSTREAM analysis is applied on a per-study basis, because the threshold value will depend on the particular type of tracer used and possibly the tissue type targeted. Therefore, data collected during a phase I study conducted for a new tracer could be analyzed to derive a threshold for use in subsequent exploratory phases or during interventional procedures. Naturally, additional data obtained from phase II/III studies may be retrospectively used to further optimize the threshold.
The proposed methodology does not involve modification of the specimen analyzed, because it uses samples in paraffin blocks and conventional histology slices; therefore, it can be seamlessly incorporated into the routine histopathological analysis workflow.
Today, fluorescence images using targeted agents are evaluated on the basis of the assumption that stronger signal implies malignancy. Correspondingly, it is typical to render the stronger fluorescence intensities of a fluorescence image in pseudocolor, together with a color image as shown in Fig. 3C. This rendering operation implies the application of a threshold on the fluorescence image depicted. However, due to the diffusive nature of fluorescence photons collected from tissues, the exact relation of fluorescence intensity/threshold and malignancy is not known; thresholds are typically user-dependent, empirical in nature, and therefore prone to errors. Observations of Fig. 2G–J showcase that there exists no clear border for cancer differentiation when performing fluorescence imaging of tissues, due to photon diffusion effects that reduce the resolution of the fluorescence image. Therefore, the selection of an objective global threshold is critical for the clinical application of fluorescence imaging, especially in association of guiding cancer resection and observing for tumor borders.
A central fSTREAM target was the rigorous relation of fluorescence signals to cancer. Therefore, an important parameter in the analysis proposed was the registration of the tumor demarcation by pathologists on H&E slices and corresponding fluorescence images. Affine transforms allowed a per-pixel analysis of fluorescence intensities and the underlying presence of malignant versus nonmalignant cells. Then automatic statistical analysis on a pixel-to-pixel basis enabled accurate and statistically significant correlation between pathological classification and tumor-to-background fluorescence intensity ratios.
The analysis intrinsically also observed the unknown distribution pattern of bevacizumab-IRDye800CW in human breast cancer, demonstrating broad presence in the entire tumor mass. We confirmed that the labeled drug can be detected at micro-dosing amounts (30 nmol/patient) and can be assessed at macroscopic and mesoscopic scales. ROC analysis calculated a sensitivity of 98% and a specificity of 79% when using fluorescence signals to outline the tumor mass. Breast cancer surgery is reported to yield up to approximately 30% positive tumor margins, directing secondary procedures (7). Hence, the findings herein preliminary showcased that bevacizumab-IRDye800CW may be valuable for intraoperative breast cancer margin detection, a hypothesis that is currently investigated in a follow-up phase II clinical study (ClinicalTrials.gov Identifier: NCT02583568). It was suggested that fluorescent bevacizumab selectively labels cancer in xenograft animals (8) and can be used for imaging studies, even if it only targets soluble VEGF-A (9, 10). Staining for VEGF-A expression revealed elevated VEGF levels in all tumors studied herein. Nevertheless, it is likely that the homogeneous distribution pattern observed is influenced by enhanced permeability and retention effects.
We also observed bevacizumab-IRDye800CW presence in areas not rich in VEGF expression including the skin and breast ducts. The sensitivity and specificity data reported were calculated only for the area (volume) around the tumor, as defined by the surgical specimen, not for the entire breast tissue, and it therefore relates more closely to intraoperative observations and not breast cancer diagnosis in the radiological sense.
Labeled drugs have been considered for imaging studies due to their general availability and expectation that the biodistribution and targeting profile of the labeled molecule will not significantly vary over the unlabeled counterpart, especially when labeling antibodies (11–13). Radionuclide-labeled drugs are often assessed with positron emission tomography to quantitatively determine the dose of an administered therapeutic molecule delivered into different organs, whereas other, nontherapeutic molecules are considered for diagnostic and staging purposes (14, 15). Examples include imaging of anti-folate agents (16) or monoclonal antibodies (8, 10). However, while nuclear imaging studies enable full-body scans, the low spatial resolution achieved only allows assessment of the macro-distribution at volume sampling of approximately 0.1–1 cm3 (17). fSTREAM comes with the potential to study the labeled molecule at higher-resolution, within the tumor microenvironment and better understand long-term biodistribution within the cancer lesion, albeit it can only be applied as an invasive technique within surgery and endoscopy.
Overall, fSTREAM can be employed for the systematic analysis of new classes of fluorescent agents considered for human use and broader disease targets (8, 11). Moreover, it can be considered as an alternative method to elucidate human cancer biology by observing personalized readings of cancer pathophysiology on excised specimen following fluorescence-guided procedures using targeted agents. The high sensitivity of the method allows micro-dosing observations, possibly relaxing regulatory parameters. Such application could capitalize on the emerging clinical practice of using fluorescent agents for diagnostic purposes (18–22).
Disclosure of Potential Conflicts of Interest
J.S. de Jong is a consultant/advisory board member for SurgVision. W.B. Nagengast reports receiving a commercial research grant from SurgVision unrestricted research grant. G.M. van Dam reports receiving a commercial research grant and is a consultant/advisory board member for SurgVision BV. V. Ntziachristos has ownership interest (including patents) in Equity holder (SurgVision BV) and is a consultant/advisory board member for scientific advisory board. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: L.E. Lamberts, A.L.L. Adams, C.P. Schröder, S.G. Elias, A.J. Witkamp, W.P.T.M. Mali, E. Van der Wall, P.J. van Diest, E.G.E. de Vries, G.M. van Dam, V. Ntziachristos
Development of methodology: M. Koch, J.S. de Jong, J. Glatz, P. Symvoulidis, A.L.L. Adams, A.G.T. Terwisscha van Scheltinga, M.N. Lub-de Hooge, C.P. Schröder, S. Oliveira, P.J. van Diest, E.G.E. de Vries, G.M. van Dam, V. Ntziachristos
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Koch, J.S. de Jong, J. Glatz, P. Symvoulidis, L.E. Lamberts, A.L.L. Adams, M.E.G. Kranendonk, M. Aichler, L. Jansen, J. de Vries, M.N. Lub-de Hooge, C.P. Schröder, E. de Boer, B. van der Vegt, W.B. Nagengast, S. Oliveira, A.J. Witkamp, W.P.T.M. Mali, P.B. Garcia-Allende, P.J. van Diest, E.G.E. de Vries, A. Walch, G.M. van Dam
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Koch, J.S. de Jong, J. Glatz, P. Symvoulidis, L.E. Lamberts, M.E.G. Kranendonk, M. Aichler, S. Oliveira, E. Van der Wall, P.J. van Diest, A. Walch, G.M. van Dam, V. Ntziachristos
Writing, review, and/or revision of the manuscript: M. Koch, J.S. de Jong, J. Glatz, P. Symvoulidis, L.E. Lamberts, A.L.L. Adams, A.G.T. Terwisscha van Scheltinga, M. Aichler, J. de Vries, M.N. Lub-de Hooge, C.P. Schröder, A. Jorritsma-Smit, B. van der Vegt, W.B. Nagengast, S.G. Elias, A.J. Witkamp, E. Van der Wall, P.J. van Diest, E.G.E. de Vries, G.M. van Dam, V. Ntziachristos
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Koch, P. Symvoulidis, M.E.G. Kranendonk, A.G.T. Terwisscha van Scheltinga, M.N. Lub-de Hooge, A. Jorritsma-Smit, M.D. Linssen, E. de Boer, B. van der Vegt, P.B. Garcia-Allende, G.M. van Dam
Study supervision: G.M. van Dam, V. Ntziachristos
Grant Support
V. Ntziachristos received funding from the Leibniz Prize 2013 (NT 3/10-1). E.G.E. de Vries, P.J. van Diest, and W.P.T.M. Mali received funding from the Center for Translational Molecular Medicine, project MAMMOTH (grant 03O-201), E.G.E. de Vries has received funding from the ERC Advanced grant OnQView and S.G. Elias from the Dutch Cancer Society KWF by a research fellowship.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.