X-ray dose considerations

To assess dose compatibility with respect to future in-vivo imaging applications, the air dose was determined. It should be noted, however, that the setups were not specifically optimized for dose-this relates in particular to the imaging detectors, which featured a low detective quantum efficiency (DQE). For this reason, two dose values were determined: the actually delivered dose and a feasible dose value. The latter is calculated from the measured dose and is based on a higher DQE, corresponding to existing efficiency-optimized detectors, and thinner grating support (within the limits of technical feasibility), reducing the loss of x-rays due to attenuation. Details about the measurement of x-ray dose and derivation of the feasible dose value can be found in a dedicated subsection of the Materials and Methods section. Table 1 lists the determined actual and feasible dose values. As expected, the feasible dose value of 40 Gy for the high-performance synchrotron measurement is substantial and by far not compatible with in-vivo imaging. This is different for the low-dose measurement with a feasible dose value of 0.4 Gy. Similarly, for the tube source measurement, where the feasible dose value amounts to 1.1 Gy. For comparison, dose values of up to several hundred mGy for in-vivo microCT imaging of mice are reported in the literature[19]. Regarding the CT measurements performed for our studies, there is room for further dose reduction even beyond the feasible dose scenario. For example Zanette et al. recently reported a data acquisition and processing method that optimizes dose efficiency by a factor of 4, while maintaining image quality [20]. This means that-assuming a dose-optimized setup-the low-dose synchrotron and tube measurement would reach a range that is compatible with in-vivo imaging of small animals. In the light of these considerations, both dose-reduced measurements hence probe the maximum achievable image quality for phase-contrast CT in-vivo imaging applications.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Determination of x-ray air dose.

https://doi.org/10.1371/journal.pone.0058439.t001

Overview of imaging data and analysis

The abdomen of the ex-vivo mouse specimen with induced PDAC was imaged with the mentioned x-ray phase-contrast setups and MRI. Subsequently, histology images of a limited section of the mouse specimen were acquired. All synchrotron measurements were performed using a two-grating Talbot interferometer at the European Synchrotron Radiation Facility (ESRF, Grenoble, France). For the high-performance measurement the focus was on optimizing image quality disregarding x-ray dose, as opposed to the dose-reduced measurement with the focus on minimizing dose in a synchrotron setup. The second dose-reduced measurement was performed using a three-grating Talbot-Lau interferometer and conventional x-ray tube source. The x-ray energy of 35 keV for the synchrotron setup was chosen to optimize image quality. For the three-grating tube setup, the acceleration voltage was chosen to optimize visibility and was set to 35 kV. This acceleration voltage results in an effective energy of 23 keV, which is the design energy of the interferometer. Further details about the preparation of the mouse specimen and the technical parameters of the imaging setup can be found in the Materials and Methods section.

Fig. 1 gives an overview over the imaging data (except for the synchrotron low-dose measurement) of the abdominal area of the mouse. It shows coronal x-ray and MR images and exemplary axial histology slices. The high-performance synchrotron measurement with attenuation (left) and phase contrast (right) is shown in panel (A). Panel (B) shows the corresponding tube source measurement, (C) the T2-weighted (T2w) MRI data and (D) histology slices. In order to distinguish between the different x-ray imaging setups and image contrasts, the following notation will be used: ACI-attenuation contrast image, PCI-phase contrast image, index S for synchrotron radiation source and index T for tube source. Due to the differences in imaging parameters, there is a difference in effective pixel size between the high-performance synchrotron (30 ), the tube-based (120 ) and the MRI data (130 ). Bone as a strongly phase-shifting tissue causes streaking artifacts in transverse phase slices, similar to metal artifacts in conventional CT. These streaking artifacts appear as horizontally oriented intensity fluctuations in the shown coronal phase images. The pancreatic tumor induction gave rise to the formation of solid tumor tissue and various lesions as previously described [21], [22]. These included pancreatic intraepithelial neoplasia (PanIN) lesions progressing to invasive PDAC as well as dilated, cyst-like ducts probably due to regional obstruction. In the MRI reference image (panel C), this solid tumor tissue (a) and one large cystic lesion (b) are highlighted. The further analysis of the image data is based on transverse slices at the indicated positions (red dashed lines) and is divided into two parts. To examine the potential of x-ray phase-contrast CT for in-vivo applications, the visibility of the solid tumor tissue (a), under dose-reduced conditions, is considered. The potential for virtual microscopy applications on the other hand is based on the high-performance data of the indicated cystic lesion (b). For both cases, x-ray attenuation and phase images from one experimental setup are compared with one another. This comparison is done both visually and quantitatively using contrast-to-noise-ratios of selected regions-of-interest. Please note that this comparison is only performed for attenuation and phase images from the same imaging setting and not across settings as the experimental parameters and especially the spatial resolution are different.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Coronal slices of the acquired multi-modal tomographic imaging data of the abdominal area in the mouse and histology.

(A) Synchrotron: Attenuation-contrast image () (left), phase-contrast image () (right). (B) Tube source: Attenuation-contrast image () (left), phase-contrast image () (right). (C) MRI with highlighted solid tumor (a) and cystic lesion (b). (D) Stack of histology slices. All images are displayed on a linear gray scale and are windowed for best visual appearance of the solid tumor and cystic lesion.

https://doi.org/10.1371/journal.pone.0058439.g001

Potential of phase-contrast CT for in-vivo imaging

As phase-contrast CT increases soft-tissue contrast, potential applications for preclinical imaging include non-invasive tumor detection and characterization. In the present mouse specimen, the solid tumor tissue is regarded for this purpose. Fig. 2 shows transverse slices (position indicated by profile in Fig. 1) for all three x-ray imaging settings [(A)–(C)] with attenuation contrast on the left and phase contrast on the right side. As a reference, the MRI image is shown in (D). Panel (A) contains the high-performance and (B) the low-dose SR-based data. Panel (C) depicts the tube-based images. The high-performance SR-based data is shown for completeness, but naturally the dose-reduced imaging data (B and C) is of most interest for evaluating the potential for in-vivo imaging. Tumor detection by means of imaging is mainly based on a detailed knowledge of anatomy and the morphology of abnormal tissue. For the present mouse specimen, solid tumor tissue was identified at the indicated position in the MRI reference image (D). In the high-performance SR-based data (A), this tissue can be identified clearly in both images, however recognizability is superior in the phase image. Visibility of this tissue is also given in the phase image of the low-dose SR-based data (B), but is very poor in the corresponding attenuation image. The same applies to the tube measurement (C)-also here, identification of the solid tumor mass is clearly possible on the basis of the phase image, but is strongly compromised in the attenuation image. This means that, for the imaging settings that assess preclinical in-vivo imaging applications, tumor visibility is strongly superior in the phase images. This observation demonstrates the potential of phase-contrast CT for tumor detection.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Assessment of solid tumor visibility.

(A) Synchrotron: (left), (right). (B) Synchrotron (low dose): (left), (right). (C) Tube source: (left), (right). (D) MRI, solid tumor part indicated by red dashed line. Regions-of-interest for the quantitative contrast-to-noise ratio analysis are indicated by colored circles. All images are displayed on a linear gray scale and are windowed for best visual appearance of the solid tumor.

https://doi.org/10.1371/journal.pone.0058439.g002

In order to objectify and quantify this evaluation of tumor visibility, a CNR analysis of selected ROIs was performed. Three ROIs were selected for each image: 1) covering the solid tumor, 2) covering surrounding tissue, 3) probing image noise. A separate ROI for determining noise was used because the standard deviation in a ROI reflects two opposing effects: image noise and tissue heterogeneity. Image noise should of course be considered in the CNR, unlike tissue heterogeneity that, if present within the ROI, will also increase the standard deviation and hence falsely appear as additional noise. The third ROI is hence placed within a homogenous region, and the standard deviation only reflects image noise. The CNR, based on the mean values of region 1 and 2 ( and ) and the standard deviation of region 3 (), is then calculated according to:(1)

The size of the ROIs was chosen to cover an area of 1 , which is as large as reasonably possible to only cover the specific tissue of interest. In Fig. 2, the chosen ROIs are displayed in each image as colored circles: yellow covering solid tumor tissue (1), green covering surrounding tissue (2), and blue covering a homogeneous region (3). Table 2 lists the individual CNRs as well as their respective ratio that reflects the relative contrast improvement of phase over attenuation contrast, i.e. the relative contrast gain. The uncertainty of the CNR () was determined by applying standard error propagation to the equation of the CNR (equation 1). The required uncertainties are the standard error (SE) of each mean value () and the standard error of the noise estimate (). denotes the number of voxels in the corresponding ROI [23]. Subsequently, error propagation was also applied to the relative contrast gain values, based on the determined uncertainties of each CNR (). The uncertainties for all CNRs and relative contrast gain values are quoted as errors in Table 2. The relative contrast gain values range from 4 for the low-dose SR-based images, over 10 for the tube images, up to 28 for the high-performance SR-based images. The tube data are obviously of particular interest for in-vivo applications. The contrast improvement of 10 for the phase images from the tube setup compared to the attenuation images clearly underlines the strong potential of phase-contrast CT for early tumor detection for in-vivo applications. As a reference, the absolute CNR of the magnetic resonance images amounts to 8.0 (not listed in Table 2). Please note that a comparison of CNRs across different imaging setups is not directly meaningful due to the difference in spatial resolution.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Contrast-to-noise ratio analysis for solid tumor visibility and tissue composition discernibility.

https://doi.org/10.1371/journal.pone.0058439.t002

Potential of phase-contrast CT for virtual microscopy applications

Increased soft-tissue contrast in combination with the capability of high spatial resolution renders phase-contrast CT attractive for virtual microscopy applications. This potential area of application was examined on the basis of the high-performance SR-based CT data of the previously indicated cystic lesion. In tumor biology in general, and specifically for tumor characterization and therapy response monitoring, tissue heterogeneity is recognized as a characteristic feature. This is, for example, because differences in tissue composition result in regional differences in therapy response. Such a difference in tissue composition is for example present around the mentioned cystic lesion and is examined in detail in the following. Fig. 3 shows transverse slices of the SR-based high-performance attenuation (A) and phase-contrast data (B) and the corresponding histology slice (C). It should be noted that the orientation of the x-ray and histology slice are marginally tilted with respect to one another, with best agreement at the lower right part of the cystic lesion, which is investigated. Moreover, in the histologic slice the upper right part of the wall of the lesion is locally deformed due to the cutting procedure. Moreover, strain fields in histological slices can lead to further deformations of the original shape [24]. Most relevant for this study is the well-known tissue shrinkage caused by dehydration. While the misalignment and deformation were small enough to be tolerable for the analysis in the present case, suitable registration algorithms exist that can be used to align the tomography images with the 2D histology data and remove distortions in the general case [25]. The mentioned difference in tissue composition, either fibrotic or cell-rich stroma, can be observed in the zoomed area of the histologic slice (C). The arrows indicate these two areas and the border in between is displayed in the extra zoom panel. This difference in tissue composition is also accessible in the phase image (B), i.e. at the indicated positions the cell-rich stroma appears darker and the fibrotic part brighter. In the attenuation image this information is not accessible and the surrounding tissue of the cyst appears homogeneous.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Assessment of tissue composition discernibility.

(A) (left), Zoom (right). (B) (left), Zoom (right). The arrows highlight positions of differences in tissue composition. (C) Histology (left), Zoom (right). Regions-of-interest for the quantitative contrast-to-noise ratio analysis in the x-ray images are indicated by colored circles. All images are displayed on a linear gray scale and are windowed for best visual appearance of the cystic lesion.

https://doi.org/10.1371/journal.pone.0058439.g003

This visual impression was also analyzed quantitatively as before on the basis of the three indicated ROIs in Fig. 3 (area of 0.3 ). The corresponding CNRs and associated uncertainties are also listed in Table 2. The very low CNR of 0.06 in the attenuation image, and the significant CNR of 7.2 in the phase image reflect the described visual impression that the discrimination of tissue composition is practically impossible in the attenuation image, but is well represented in the phase images. The relative contrast gain was not determined for this set of images as, in the attenuation image, the error in the CNR of 0.08 is on the same order as the CNR itself. This demonstrated capability of phase-contrast CT of visualizing such subtle differences in tissue composition underlines the potential of phase-contrast CT for virtual microscopy applications.

General Discussion

For the assessment of the potential of phase-contrast CT for preclinical imaging, we investigated different x-ray phase-contrast imaging settings. As very well-established reference technique with high intrinsic soft-tissue contrast, MRI was used. In this regard, we would like to indicate a technical limitation of our study. The employed instrumentation of a human MR scanner with microscopy coil is not ideally suited for microscopic imaging and dedicated small animal MR scanners are available in principle. These scanners are optimized for smaller samples and perform superior with respect to spatial resolution and contrast. For our purposes however-to include a well-established soft-tissue imaging technique as a reference-the performance of the MR scanner was sufficient. For an elaborate comparison of phase-contrast CT and MRI, the reader is referred to a recent study by Schulz et al. [26], comparing phase-contrast CT, magnetic resonance microscopy and histology of the human cerebellum.

Concerning the signal-to-noise ratio of attenuation and phase contrast in general, it has been reported in the literature that phase contrast benefits from comparably small detector pixel sizes [27]–[29]. In preclinical imaging systems, pixel sizes naturally are smaller than in human CT scanners, rendering phase contrast especially attractive for preclinical imaging applications.

When comparing attenuation and phase-contrast images from a grating-based setup in general, one aspect relating to the method should be noted. In a conventional CT setup, there are no gratings present and the dose-relevant final grating (G2) absorbs approximately half of the x-rays that would otherwise also contribute to the image signal. This means that attenuation-contrast images from a grating-based setup require approximately twice the dose needed for comparable images from a conventional setup. On the other hand, the grating-based setup yields-from a single CT scan with a certain radiation dose-two perfectly registered images with complementary contrast mechanisms.

Ex-vivo mouse specimen

A mouse with pancreas specific activation of oncogenic Kras was obtained by breeding knock-in mice with animals [21]. To model PDAC, a 19 months old (CK) animal was used. A T2-weighted anatomy MRI scan of CK mice of tumor-bearing age was performed to confirm the presence of a solid tumor mass. Subsequently the animals were kept in isoflurane narcosis and a median laparotomy was performed, followed by a perfusion fixation protocol. In brief, the left ventricle was canulated with a 22 G needle, followed by clipping of the right atrium. Then, 10 ml of phosphate buffered saline was manually infused to flush out all blood from the vasculature, followed by manual infusion of 20 ml 4% paraformaldehyde (PFA). Thereafter the animal was emerged for 72 hours in 200 ml 4% PFA, briefly washed in 70% ethanol and transferred into a 50 ml Falcon tube in 70% ethanol. Animal care and experimental protocols were conducted in accordance with German animal protection laws and approved by the Institutional Animal Care and Use Committee at the Technische Universität München. For histology, the mouse was removed from the plastic container. The abdomen was cut off, dehydrated in an automated tissue processor and embedded in paraffin. After embedding, the spine was removed and the whole abdomen was cut into 4 thick serial sections. Afterwards, the sections were stained with hematoxylin and eosin and digitalized with a slide scanner (DotSlide, Olympus).

X-ray grating-based phase-contrast CT

Conventional x-ray attenuation-contrast images and phase-contrast images were obtained using a grating interferometer [2], [4], [5]. The synchrotron radiation source measurements were carried out at beamline ID 19 of the European Synchrotron Radiation Facility (ESRF) in Grenoble, France. Images were recorded with a two-grating Talbot interferometer and scintillator/lens-coupled CCD detector (FReLoN) [30]. The interferometer was operated with a monochromatic x-ray beam of 35 keV (Si(111) double-crystal monochromator, energy bandwidth ). The exact interferometer and tomography parameters are listed in Table 3 and Table 4 respectively. For the low-dose dataset, the detector pixels were binned prior to readout, resulting in an effective pixel size of . The tube source measurement was performed at the Physics Department of the Technische Universität München, Munich, Germany. A rotating anode (Enraf Nonius FR 591, molybdenum target) with an effective focal spot size of was operated at 35 kVp and 70 mA. With the three gratings in the beam (each on a silicon wafer of 500 thickness acting as a filter), this results in an x-ray spectrum with its center at the design energy of 23 keV of the interferometer. Phase- and attenuation-contrast images were obtained with a three-grating Talbot-Lau interferometer and a Pilatus II silicon-based photon-counting imaging detector. For detailed parameters, see Table 3 and Table 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Interferometer characteristics.

https://doi.org/10.1371/journal.pone.0058439.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. X-ray CT data acquisition parameters.

https://doi.org/10.1371/journal.pone.0058439.t004

Measurement of x-ray dose

For determining the x-ray dose in each measurement, a clinically approved dosimeter (Patient Skin Dosimeter, Unfors, Sweden) was used. This dosimeter is designed for monitoring the x-ray dose during fluoroscopic CT procedures and is calibrated for entrance skin dose at 90 kVp. When taking a conversion factor of 1.4 into account, the dose in free air can be determined from the displayed entrance skin dose (Unfors PSD, Patient Skin Dosimeter, User's Manual). As the difference in effective energy for the various experimental setups was not considered, the determined dose value is subject to some uncertainty. However, a precise dosimetry of the performed CT measurements was not the objective of the study and the dose value rather serves the purpose of evaluating the compatibility with preclinical in-vivo imaging. In the technical specifications, the manufacturer quotes an energy dependence of 15% (40 kVp–150 kVp) and an uncertainty of 6%. The error in the determined dose values, for good measure, is hence assumed to be within 25%. As both the synchrotron and tube source imaging setup are not optimized for dose, two dose values were determined: the actually delivered dose and a feasible dose value. The latter value is calculated from the experimentally measured one and is based on two premises: 70% detective quantum efficiency (DQE) and a thinner silicon wafer support of the gratings. A DQE of 70% is used since, for example state-of-the art CMOS imaging detectors reach such high DQE at small pixel sizes of 75 (Dexela 2923) [31]. The calculation of feasible dose is based on the actual DQE of the used detector and a subsequent scaling to 70%. Thinner silicon wafers are based on a thickness of 100 , instead of the actual thickness of 500 . The scaling effect in dose due to reduced attenuation in the thinner wafer is calculated mono-energetically and is based on the present synchrotron photon energy of 35 keV and 23 keV interferometer design energy for the tube setup. For the synchrotron setup, the CCD detector (FReLoN) sensitivity was determined from the DQE at zero spatial frequency , from Coan et al. [32] (Table 3, first column, 33 keV). The scintillator screen thickness of the FReLoN detector in our study was 30 , as opposed to a thickness of 100 in Coan et al. [32]. In order to account for the thinner scintillator screen-and related lower DQE-in our study, the reduced attenuation of x-rays in the thinner scintillator screen was considered and the DQE value of 0.30 was scaled accordingly. This scaling results in a DQE of 0.11 for our measurement. The thinner silicon wafers were taken into account by considering the reduced attenuation in both gratings of the synchrotron setup. For the tube source setup, the DQE was determined from the attenuation of 23 keV photons (interferometer design energy) in the sensitive silicon layer of 450 in the Pilatus II detector and amounts to 0.25. Thinner silicon wafers were again considered by the reduced attenuation in all three gratings. Both the actual and feasible dose value for the different imaging setups are quoted in Table 1.

Magnetic resonance imaging

High resolution MR imaging was performed on a 1.5 T clinical scanner (Philips Achieva) using the 47 mm diameter microscopy coil. The sequence was a 3D turbo spin echo (TSE) sequence with an isotropic resolution of 130 and the following parameters: FOV = , TR = 1000 ms, TE = 95 ms, turbo factor 15, echo train length 333 ms, 6 NSA and a total acquisition time of approximately 14 hrs. A DRIVE pulse was used to compensate for the short repetition time.

CT Reconstruction, image processing and data analysis

CT reconstruction was performed using a standard filtered backprojection algorithm. For attenuation contrast, a Ram-Lak filter and for phase contrast, a Hilbert filter was used [33]. Strongly phase shifting materials, such as bone, produce streaking artifacts. To reduce the magnitude of the streaks, a straightforward method was applied during reconstruction of the phase-contrast data, i.e. the differential phase sinogram was weighted with the squared value of the interferometer visibility. This approach suppresses the bone signal as bone scatters strongly and hence causes a low weighting of the corresponding area in the differential phase sinogram. This method however only reduces the magnitude and streaking artifacts are still present. All attenuation- and phase-contrast CT slices were post-processed with a sharpening filter. Minor deformations in the shape of the mouse and changes in the position of few air bubbles are apparent between the different measurements and image registration was performed manually. Please note that the attenuation and phase image of each tomography however are intrinsically perfectly registered with one another as both images originate from one common dataset. Each image is displayed on a linear gray scale and windowed for best visual appearance of the discussed features. Comparison of image contrast is then done visually and quantitatively using CNRs.