High resolution propagation-based imaging system for in vivo dynamic computed tomography of lungs in small animals

We have developed an x-ray imaging system for in vivo four-dimensional computed tomography (4DCT) of small animals for pre-clinical lung investigations. Our customized laboratory facility is capable of high resolution in vivo imaging at high frame rates. Characterization using phantoms demonstrate a spatial resolution of slightly below 50 μm at imaging rates of 30 Hz, and the ability to quantify material density differences of at least 3%. We benchmark our system against existing small animal pre-clinical CT scanners using a quality factor that combines spatial resolution, image noise, dose and scan time. In vivo 4DCT images obtained on our system demonstrate resolution of important features such as blood vessels and small airways, of which the smallest discernible were measured as 55–60 μm in cross section. Quantitative analysis of the images demonstrate regional differences in ventilation between injured and healthy lungs.


M Preissner et al
Accepted quality standards and customized phantoms enable quality control and robust performance testing for clinical cone-beam computed tomography (DIN 2013, Steiding et al 2014. Currently, no such standards for pre-clinical (micro-CT) scanners exist. In order to address the issue of quality assurance in micro-CT, Kalender et al propose a method that tests a combination of dose, resolution, noise and scan time (Kalender et al 2005). To benchmark our system against current scanners, we adopt the quality factor, which amalgamates these important parameters into a single index for comparison between systems. Finally, we present in vivo 4DCT images of mouse lungs to demonstrate the quality of imaging that our system achieves for pre-clinical investigations. Figure 1 shows the laboratory set-up. The high-brightness x-ray source (Excillum D2 + , Excillum AB, Kista, Sweden) has a liquid-metal-jet anode (gallium alloy) which enables higher electron beam power (70 kVp, 250 W) with a micro-focus spot of between 15 and 20 μm. It has a polychromatic x-ray beam with characteristic x-ray peaks at 8 keV and 24 keV (Larsson et al 2011). A high-speed flat panel detector (PaxScan 2020, Varian Medical Systems, Palo Alto, CA, USA) is mounted at a distance of 3363 mm from the x-ray source. The geometric magnification (M) is adjusted by translating the stage for the distances R 1 and R 2 (Zaber Technologies, Vancouver, Canada). The detector has a pixel size of 194 × 194 μm and is capable of achieving a frame rate of up to 30 Hz with an 18 ms exposure time.

X-ray imaging set-up
The range of geometric magnification achievable with the current set-up of this system is 7.2-12.0, where M = (R 1 + R 2 )/(R 1 ). The corresponding range of field of view is between 16.1 and 27 mm. The effective pixel size of the projection images for this magnification range is between 16 and 27 μm. Phase enhancement is achieved via a propagation distance and can be seen in the fringes of the projection image in figure 1(B). The phase enhancement can also be quantified with an effective object-image distance, z ef f = R 1 R 2 /(R 1 + R 2 ) (Wilkins et al 1996, Mayo et al 2002. Our set-up has z eff values of between 257 mm and 402 mm, which is a practical compromise between exploiting the phase enhancement our system provides without compromising the flux required for dynamic imaging with short exposure times. Table 1 shows the two main combinations of x-ray spot size, power and projections per CT used for dynamic lung imaging. The standard setting is based on a larger spot size, fewer projections and lower power (with a lower radiation dose), whereas the maximum setting has a smaller spot size and a higher power, resulting in better image quality (as defined by the quality factor described below), with a higher radiation dose as the trade-off.

Phantom imaging and dose measurement
Resolution testing was carried out with phantoms. The CT resolution was measured using a line pair phantom (Micro-CT bar pattern phantom, QRM GmbH, Möhrendorf, Germany). A customized 3D printed rod (FullCure720, Objet Eden260V, Stratasys Ltd) was used to test noise. The contrast resolution was tested with a low contrast phantom with three inserts (Micro-CT Low Contrast Phantom V2, QRM GmbH, Germany). The radiation dose rate was measured using a pencil beam dosimeter (TNT 12000WD wireless detector and 500-100 CT ion chamber, Fluke Biomedical, Washington, USA).

In vivo 4DCT imaging
The use of eight-week old BALB/c female mice for in vivo imaging was approved by the local Animal Ethics Committee of Monash University (Monash University Research Platform, Melbourne, VIC, Australia) and conducted in accordance with the guidelines set out in the Australian code of practice for the care and use of animals for scientific purposes. Mice were anaesthetized with an intraperitoneal injection of ketamine (Parnell Australia Pty Ltd, Alexandria NSW, Australia) and xylazine (Xylazil-20, Troy Laboratories Pty Ltd, Smithfield NSW, Australia), surgically intubated and placed upright in a custom sample mount with ventilator attachments, which is 3D printed and can therefore be made to the size and shape of the sample, e.g. mouse or small rat. Mice were ventilated at a peak inspiratory pressure of 20 cm H 2 O and zero positive end-expiratory pressure with an inspiratory-expiratory ratio of 1:1 (300 ms : 300 ms). The animal was placed on a ventilator (AccuVent, Notting Hill Devices, Melbourne, Australia) and secured on the rotating stage. Image acquisition was synchronized with the ventilator. Projection images were obtained over multiple breath cycles for a 360° rotation of the sample. The respiratory rate of the animal (RR, breaths per minute) determines the number of phases (n) acquired in the 4DCT: n = 60f/RR, where f is the image acquisition rate (Hz). The rotational speed of the stage (ω, degrees per second) is determined by the total number of projections required for the 4DCT: ω = 360 • f /np, where p is the number of projections per phase, i.e. 400 projections for the standard setting or 800 projections for the maximum setting (table 1). The projections are allocated (binned) into discrete phases of the respiratory cycle. The binned projection images are then reconstructed using filtered back-projection based on the Feldkamp-Davis-Kress cone-beam CT reconstruction algorithm to obtain 3D cross-sectional images (Feldkamp et al 1984, Yang et al 2006.

CT resolution
The resolution for CT images was tested using a high precision bar pattern (line pair) phantom for micro-CT (QRM GmbH, Möhrendorf, Germany) at the standard and maximum settings (table 1). The images are shown in figure 2. Note that the corner artifacts are due to beam hardening. Based on visual inspection of the line pairs in the image, the spatial resolution lies between 10 line pairs per mm (lp mm −1 ), i.e. 50 μm line widths, where  (see table 1). A phase fringe is visible at the boundary between the air (left) and the rod material (right), and is shown by the vertically averaged profile plot in (C), whereby a spike in intensity at the boundary is observed.  (2001). The 10% MTF was calculated as 11 lp mm −1 for the standard source setting and 12 lp mm −1 for the maximum source setting.

Radiation dose
Reducing the radiation dose received by the animal is important to avoid unwanted interference in experimental results or ill-effects from radiation, particularly in the case of longitudinal studies (Boone et al 2004, Vande Velde et al 2015. The dose rate was measured as 5.01 mGy s −1 air kerma using a pencil-beam dosimeter (TNT 12000WD wireless detector and 500-100 CT ion chamber, Fluke Biomedical, Washington, USA). The total dose for a CT obtained with the standard setting is 30 mGy, whereas the total dose for a CT on the maximum setting is 60 mGy. Figure 3 shows CT images of a phantom with three inserts of low contrast (due to air bubbles) with known contrast levels of −3%, −6% and −9% (+/− 0.1%) as compared to the background material made of a proprietary epoxy resin (QRM GmbH, Möhrendorf, Germany). Each contrast level has three inserts of varying diameter sizes: small (0.5 mm), medium (1 mm) and large (2 mm). Intensity values were measured for each of the inserts (nine in total) using ImageJ software (Schindelin et al 2012). The contrast (c) was determined by comparing the intensity values (I) of the inserts to the background material, c = 1 − ((I insert − I air )/(I resin − I air )). These were plotted against the known contrast values and shows excellent agreement between the measured and known values. This demonstrates that the system is capable of detecting low contrast levels in samples such as the tissueair contrast found in the lungs.

Noise
We adopted an approach similar to Kalender et al for the calculation of noise for use in their quality factor equation, whereby noise was defined as the standard deviation (in HU) for a 10 mm 2 region of interest in a 32 mm diameter water phantom (Kalender et al 2005). We used a 3D printed 24 mm diameter rod (FullCure720, Objet Eden260V, Stratasys Ltd) made of a material that has properties similar to an acrylic, and as such, has properties similar to water (Ionita et al 2014). A region of interest was converted to Hounsfield Units (HU) with ImageJ (Schindelin et al 2012) using the following relationship between intensity (I) and HU: (1) For our system, the value for HU shift is 3072 and the value for HU range is 8191, based on calibrations of the attenuation coefficient (μ) for water and air at an effective photon energy of 20 keV (Seltzer 1993), and using A B the standard formula for calculating Hounsfield Units: HU = 1000(µ material − µ air )/(µ water − µ air ) (Kalender 2011). The standard deviation was determined to be 520 HU for the standard setting and 198 HU for the maximum setting, using equation (1).

Quality factor versus scan time
Kalender et al utilize a combination of dose, resolution and noise to characterize various systems in high speed mode and high quality mode, and represent this as the quality factor, Q, given by equation (2): where; ρ 10% = 10 % MTF (lp mm −1 ), σ = noise (standard deviation as HU), D = dose (mGy) (Kalender et al 2005).
The ratio of quality factor to scan time (Q/T) in high speed mode for one CT is shown in figure 4 for our system for both the standard and maximum settings. These values are compared to four commercially available micro-CT scanners (Scanners (A)-(D) in figure 4) (Kalender et al 2005), as well as for a current state-of-theart commercially available micro-CT scanner (Scanner (E) in figure 4), for which the values of dose (19 mGy), resolution (4.85 lp mm −1 ) and scan time (26 s, based on 514 projections per CT) were obtained from the current literature and technical notes available from the manufacturer (Behrooz et al 2016, Ghani et al 2016. The noise for this system was estimated as the same as for the Excillum at the maximum setting (i.e. 198 HU). As scan time increases, the quality factor does not increase. This is due to the squared weighting of the resolution (10% MTF) and the increased dose; longer scan times do not significantly improve resolution, but they do significantly increase the radiation dose received by the sample, thus resulting in a lower quality factor. For terminal in vivo imaging, where the dose is less significant than for longitudinal imaging, we compared the devices without the dose component in the quality factor ( figure 4(B)), i.e. Q = 1000(ρ 2 10% )/σ.

In vivo 4DCT images of mice
The CT slices in figure 5 are from a typical in vivo 4DCT image obtained on our system (supplementary material 5 (stacks.iop.org/PMB/63/08NT03/mmedia)). The images are of healthy BALB/c mice. The images were taken at the maximum setting (table 1), i.e. 800 projections per CT and a scan time of 32 s at an image acquisition rate of 30 Hz and an 18 ms exposure time (with inspiratory and expiratory times of 300 : 300 ms). These in vivo CT images demonstrate the high quality images that are achievable with this laboratory set-up. Pertinent features in the lung parenchyma, such as the small airways (or alveolar clusters) can be discerned in our 4DCT images. A selection of these were manually measured (figure 5) using ImageJ and found to be approximately 55-60 μm in cross section, which is consistent with other studies (Irvin and Bates 2003). There are some artifacts (blurring) around the heart and at edges of the lungs and rib-cage bones, due to motion blur ( figure 5(A)). The images in figures 5(C)-(H) demonstrate the application of Hounsfield Units for calculating the volume of air in the lungs. In these images, the relative volumes are mapped onto the CT slices. The ability to accurately quantify absolute lung volumes is important in order to determine lung function, for example, in determining the functional residual capacity on a regional level. There are many potential applications for high quality dynamic CT reconstructions, for example to be used for 3D x-ray velocimetry analysis, such as that based on the technique by Dubsky et al (2012). High quality images also enable the accurate segmentation of airways (Kim et al 2016, Dubsky et al 2017 and pulmonary vasculature , which is essential for obtaining regional (e.g. lobar or sublobar) information about the lungs in order to determine heterogeneity in disease models (Stahr et al 2016).

Conclusion
The research presented in this paper demonstrates the optimization of a customized laboratory facility for the purposes of lung x-ray imaging in small animals. The key feature of our system is imaging at high speed with high image quality (10% MTF of 12 lp mm −1 for a CT taken at 30 frames per second), which is a prerequisite for functional imaging. The ratio of quality factor to scan time (Q/T) demonstrates that the facility has an excellent balance between imaging speed, resolution, contrast and radiation dose, which is essential for quality control in small animal imaging. Such information is necessary in deciding on the design of future in vivo studies with the capabilities of the system in mind and for translation to the clinic. The advantages of a customized in-house facility enables complex and longitudinal animal studies that would otherwise not be possible. Furthermore, the system can be configured and adapted to the specific imaging requirements. For example, the effective propagation distance can be adjusted from pure absorption to phase enhancement, and the magnification and field of view can be configured for specific studies (e.g. mouse, rat), or for a specific region of interest.