1.

INTRODUCTION

X-ray Computed Tomography (CT) is a well-established technique, which is – in its conventional form – limited to attenuation contrast. There it allows high resolution imaging, fast acquisition times, and decent contrast for many purposes such as material research, inspection, and medical diagnostics. Throughout the last decade Talbot-Lau interferometry, which started as a synchrotron-only technology, gained attention as it unlocks additional contrast channels.³, ⁴ A Talbot-Lau interferometer, as illustrated in Fig. 1, allows simultaneous imaging of attenuation, refraction, and small-angle scattering of the transmitted object.⁵, ⁶

Figure 1.

Talbot-Lau interferometer in a CT gantry. It consists of three cylindrically bent gratings illustrated as line patterns, referred to as G₀, G₁, and G₂, which structure the X-ray beam as they absorb or phase-shift the radiation. By the Talbot effect a self-image of G₁ appears at the location of G₂. The object induces attenuation, refraction, and small-angle scattering, which subsequently change this self-image characteristically. Using the Moiré effect and the G₂ grating which has a period matching to the G₁ self-image these distortions can be retrieved with a common X-ray detector. Since this approach relies on coherent radiation, a G₀ source-grating is required to convert the incoherent source spot into a set of sufficiently coherent slit sources.

Particularly the access to sub-resolution tissue properties by analyzing the small-angle scattering power – also referred to as dark-field signal – of the material is a promising application in biomedical imaging. Previous work at laboratory setups demonstrated that the dark-field modality is useful for lung diagnostics as it enables the detection of micro-structural changes in lung parenchyma.^7–10

Recently the first clinical study on dark-field chest radiography of COPD and COVID-19 patients validated these pre-clinical results and clearly showed the potential of this additional image contrast.¹, ¹¹ This first in-vivo human dark-field imaging system can now retrieve otherwise undetectable changes in the porous lung tissue which are correlated with diseases progressions. However, this chest radiography provides no 3D information and anatomical structures overlap in image domain. To overcome these restrictions our approach is to install the interferometer into a clinical CT gantry. However, this step involves a variety of problems since the space on a clinical CT gantry is limited, an extremely large field of view must be covered by the gratings (compared to current state of the art dark-field implementations), and vibrations of the system can quickly degrade the measured signal. In this work we discuss our design of a Talbot-Lau interferometer which has been integrated into a Brilliance iCT (Royal Philips, The Netherlands) and produces the first dark-field CT reconstructions of human sized phantoms.²

2.

TALBOT-LAU INTERFEROMETER LAYOUT

In our presented design the most important design rules have been (1) to maintain the bore diameter of the conventional CT, which is 70 cm (blue dashed line in Fig. 1), (2) to not modify the conventional gantry and its primary imaging components, i.e., source and detector, and (3) to have at least 45 cm coverage which is 90% of a conventional CT which has an diameter of 50 cm (highlighted area and green dashed line in Fig. 1, respectively).

The presented Talbot-Lau interferometer consists of three gratings and must be optimized for the 80 kVp spectrum, which is the lowest available energy on the iCT platform. As this method only works for coherent radiation a source grating G₀ is required.¹² It splits the radiation from the incoherent X-ray source into several slit sources which fulfill the coherence requirement for the length of the interferometer setup. Because of the weak interaction between hard X-rays and grating the aspect ratio of the structures must be high – which is difficult to fabricate – and gold as a good absorber is used as the grating material. The grating G₁ introduces a fine intensity modulation on the incident radiation, which is basically a fine line pattern with a periodicity of a few micrometers. Attenuation of the radiation by the object causes a decrease of intensity, whereas refraction and small-angle scattering induce small distortions of the pattern. To resolve these small changes an analyzer grating G₂ is positioned in front of the detector and makes use of the Moiré effect to translate the changes to an intensity pattern which can be measured by the detector.

The grating periods and distances define the sensitivity of the system to small-angle scattering. Basically, smaller periods and longer distances lead to a more sensitive setup. Note that a too small sensitivity results in a poor signal to noise ratio and a too high sensitivity can lead to artefacts due visibility starvation. From previous work in¹³, ¹⁴ a sensitivity range that is meaningful for lung imaging can be derived. To achieve a sufficient sensitivity the position and the period of the grating structures must be selected carefully. Certainly higher sensitivity is preferred, however, there are limitations in the grating fabrication and the available space on the CT gantry which limit the maximum sensitivity of our prototype.

In Fig. 2 two sketches of different geometries illustrate possible implementations of the three gratings into a CT gantry. Increasing inter-grating distances as well as decreasing the grating periods improve sensitivity. Evidently, it is therefore a good design choice to maximize the distances within the constraints given by the CT platform.¹⁵ Consequently G₀ and G₂ are positioned as close to the X-ray source and the detector, respectively, and G1 as close to the bore as possible. The inverse geometry is advantageous for our CT implementation as it requires only one large coverage G₂ with relatively coarse period and the gratings G₀ and G₁ are small enough to be manufactured as single parts.

Figure 2.

Asymmetric and inverse interferometer geometry. a, in an asymmetric design the gratings G₁ and G₂ are both behind the patient and must cover a large area due to the large field of view. Furthermore, the periods of these two gratings are rather fine, compared to the G₀ grating. b, in the inverse geometry, G₀ and G₁ are in front of the patient, their size is relatively small due to the strong divergence of the setup, and the period of the largest grating G₂ is coarse, which makes fabrication easier. Another advantage of this design is its dose efficiency because G₁ is positioned in front of the patient.

3.

INTEGRATION INTO THE CT GANTRY

In our implementation we position a combined mount which holds G₀ and G₁ into the collimator box which is positioned downstream the X-ray source before the patient. This assembly is shown in Fig. 3 where the two gratings are cylindrically bent to focus towards the X-ray source spot of the CT. This is an important design characteristic since grating G₀ exceeds an aspect ratio of 100 (relation between height and width of the lamellae). In an un-bent state, a significant fraction of the radiation would traverse the grating not with a perpendicular incident angle because of the strong divergence. This would lead to a degradation of the slit-sources by shadowing, which should be generated by the G₀, and consequently the performance of the interferometer vanishes.

Figure 3.

G₀ and G₁ bending frame for CT implementation. The two gratings are rigidly mounted on a machined structure which ensures a precise inter-grating distance and bends the gratings cylindrically. This focuses the trenches between the grating lamellae towards the X-ray source spot to compensate for the strongly divergent radiation and reduce shadowing artefacts.

The large G₂ grating must cover about 80 cm of arc length which is realized by combining several smaller tiles to one large grating. We developed a specialized mount which allows us to adjust the tiles inside the gantry individually and which additionally bends each tile to focus towards the source spot. It is shown in Fig. 4 covering 90% of the CT detector columns and 32 pixel rows which leads to a total coverage of 20mm in the iso-center and a reconstructible volume diameter of 450 mm.

Figure 4.

Modular G₂ implementation covering 80 cm width. The specialized mount allows adjustment of each G₂ tiles individ-ually to minimize the gap between the tiles and to ensure parallel alignment of the G₂ lamellae to the line pattern produced by the G₁. Again, each grating tile is bent cylindrically and also the combined assembly follows this curvature to focus into the source spot, which reduces shadowing artefacts.

4.

OPTIMIZATION OF THE GRATING PARAMETERS

Based on the state of the art fabrication technology for X-ray absorption gratings, which are compatible with the 80 kVp spectrum used in our dark-field CT prototype and a ballpark sensitivity estimation that shows that we will not reach the upper sensitivity limit for human lung imaging where signal saturation occurs, the period of G₀ is fixed to 4.8 μm. From this parameter and the introduced inverse interferometer geometry we derive the optimal grating parameters using a wave propagation simulation. It is similar to previous work by¹⁶ and includes the geometrical effects of the diverging beam, the selected X-ray source spot, the clinical 80 kVp spectrum, its filtration, and the imperfections in the extended gratings, i.e. bridges in the layout.

Apart from the sensitivity (which is defined by distances and periods), another key parameter of the interferometer is the so-called fringe visibility, which directly translates into signal to noise ratio of the dark-field image.¹⁷ This visibility is strongly influenced by the height, shape, and duty cycle of the grating G₁. In Fig. 5 simulation results of these free parameters reveal significant performance differences, particularly for two different G₁ grating profiles. We find that the best system performance can be achieved with a triangular shaped G₁ grating profile.¹⁸ The elegance of a triangular G₁ profile is that it requires only about 18 μm high structures and thus, X-ray flux through this grating remains high whereas the alternative – an absorption G₁ – would reduce the flux at least by 50%. Furthermore, an absorption G₁ would require a high aspect ratio which makes its fabrication difficult.

Figure 5.

Wave optical simulation results of the interferometer for different G₁ parameters (using gold as a grating material). a, shows the performance for a conventional Talbot-Lau interferometer with rectangular or binary grating profile. Here we find no significant performance in the height range typical for phase-shifting G₁ (below 25 μm) merely a design with an absorption G₁ (height exceeding 200 μm) would give good performance. b, only for a triangular G₁ grating profile with a height of 18 μm a good performance exceeding 26% visibility is found.

The dark-field CT prototype is equipped with a triangular G₁ on a flexible polyimide substrate. The flexibility of the substrate enables us to bend the structure as discussed earlier and ensures that the performance is not degraded, e.g. by scattering of radiation in the substrate material. G₀ and G₂ use graphite substrates which are also flexible for bending to the required radii.

5.

PERFORMANCE OF THE TALBOT-LAU INTERFEROMETER

The design utilizes the standard iCT detector and can be rotated with rotation times between 0.27 and 1.5 s which are the standard settings for this clinical CT model. From the centrifugal acceleration and the vibrations induced by the X-ray tube as well as other sources of instabilities the interferometer is periodically leading to a corresponding distortion of the fringe pattern. Fortunately, the frame rate of the detector is high enough to resolve the distortions with high accuracy, thus, with a sophisticated processing framework the attenuation and dark-field signal can be extracted from the measurements.

The visibility of the presented interferometer, which is the most important measure for the performance of a Talbot-Lau interferometer, is around 22% for the central 50% of the field of view. Fig. 6 shows the flat-field performance of the interferometer where a characteristic reduction towards the left and right is induced by remaining partial shadowing in G₀ due to the large source spot. For larger fan angles, the width of the X-ray source spot appears to be bigger, because the actual source area is an inclined plane on the anode target surface. Only under the projection through the iso-centre the source spot is smallest.

Figure 6.

Performance evaluation of the Talbot-Lau interferometer during rotation. a, the intensity has a characteristic maximum in the center and decreases to the left and right due to partial shadowing in the G₀. b, the system visibility is highest in the center exceeding 26%, but due to partial shadowing there is a drop to the left and right. c, the interferometer phase is optimized for about 10 pixels per fringe for ideal phase sampling during processing and reconstruction.

6.

CONCLUSION

We conclude that the presented design is most suitable for clinical dark-field CT implementation within the limitations of current grating fabrication and the geometrical boundary conditions imposed by the desire to use a conventional clinical CT as a platform for the prototype. We analyzed the impact of the most important design parameters to identify potential problems and could verify that a triangular grating profile for G₁ has the best performance, besides an absorption G₁.

In our implementation shadowing is avoided by cylindrically bending all gratings to focus into the source spot. Nevertheless, partial shadowing in the G₀ due to the extended source causes a visibility decrease towards larger fan-angles.

With this proposed design, we demonstrate that implementation of a Talbot-Lau interferometer into a clinical CT is feasible with state of the art gratings. This development now brings dark-field CT an important step closer to the clinics as it demonstrates that up-scaling to human sized field of views works and sub-second scan times are now easily possible.