XClose

Medical Physics and Biomedical Engineering

Home
Menu

Subsequent key advances

 

Enhanced material discrimination with dark field imaging combined with deep learning

Dark field imaging relies on the detection of x-rays refracted in multiple directions by features smaller than the spatial resolution of the imaging system. As such it is sensitive to the microstructure of a material, and spatial variations of this microstructure create a texture in the dark field image that can be unique to that material. This is a perfect match with deep learning algorithms, which can be trained to recognise specific textures. We applied this approach to the detection of explosive materials, creating realistic, challenging scenarios where these were hidden in small quantities in electrical items placed inside bags alongside other “cluttering” objects, mimicking the content of a carry-on bag, and obtained a 100% detection rate. While this result is surely affected by the small scale of our study and by the fact it was carried out in a relevant, but not an operational environment, it indicates a significant potential for this new “combined” approach. Importantly, there is nothing special about explosives as such, and the method can be applied to any microscopically inhomogeneous material (Partridge et al Nat. Commun. 13 (2022) 4651. This article received significant media coverage, including all major UK outlets. A partial list can be found at https://nature.altmetric.com/details/135701425/news, which however does not include articles appeared on the Daily Mail, Daily Telegraph and others).

Xray images of a hairdryer and mobile phone
Examples images from the study, specifically of a hair drier (top) and a mobile phone (bottom), with conventional attenuation-based x-ray images (left) directly compared to their dark-field correspondent (right).
Low-energy x-ray microscopy with user-defined resolution 

Our phase-sensitive methods use absorbing masks with apertures to make an x-ray imaging system sensitive to refraction and ultra-small angle scattering. An additional advantage of an apertured pre-sample masks is that it allows achieving a resolution determined purely by the size of the apertures. These are typically regularly spaced, and since the sample parts corresponding to areas between adjacent apertures are not illuminated during an x-ray shot, they do not contribute to the image; their contribution can be recovered in subsequent shots by displacing the sample or the mask. This means the system resolution can be determined by the user by selecting a given aperture size. We exploited this property in combination with micron-size apertures to realise a laboratory-based microscopy system, using low x-ray energy (the Cu K line at ~8 keV) for maximum sensitivity to soft biological tissues. Coupled with a detector with a pixel size smaller than the mask period, this allows the simultaneous retrieval of attenuation, phase and dark-field images from a single image. Notably, since the latter image is sensitive to sub-resolution sample features, pushing the resolution down to the micron level makes dark-field sensitive to structures on the nanometric scale (Esposito et al Appl. Phys. Lett. 120 (2022) 234101). 

images of a bar pattern test object
Attenuation (top), differential phase (middle) and dark-field (bottom) images of a bar pattern test object, extracted from the same dataset acquired with the laboratory-based low-energy x-ray microscope. The bars in the pattern on the far right are 1 micron in size.
Integrating the detector mask into the detector itself

Most of our phase-based imaging methods use two masks, placed before imaged sample and detector respectively, to make an x-ray imaging system sensitive to refraction and ultra-small angle scattering. By working with Prosperity Partnership’s external partner Scintacor/Photonic Science, we have demonstrated that the second (detector) mask can be integrated directly into the detector itself, which simplifies setup complexity and alignment procedures. We did this by replacing the standard scintillator of an indirect conversion detector with a “structured” one, where trenches filled with scintillating material replace the apertures in the detector mask. The remainder of the detector area remains insensitive to x-rays, equivalently to the absorbing septa of the same mask. We have shown that use of this “structured” scintillator provides similar results to our standard double-mask configuration, and it allows a reliable retrieval of phase and attenuation (Massimi et al J. Appl. Phys. 131 (2022) 204501).

small wires quantitatively retrieved using a detector
Attenuation (far left) and differential phase (centre) of small wires quantitatively retrieved using a detector equipped with a structured scintillator replacing the absorbing mask normally placed before the detector itself. The image on the far right (also obtained with the structured detector) shows how the latter channel allows the visualisation of thin cartilage layers, invisible to conventional x-rays.
Monitoring tissue engineering protocols

The progress of regenerative medicine towards clinical use is currently hindered by the lack of a suitable non-destructive, 3D imaging method. This is necessary to perform a series of assessments essential to clinical translation, e.g. of the integrity and suitability of the scaffold (e.g. that it is completely cell-free if it has been obtained through a decellularization method), of the correct cell-scaffold interactions during the organ generation process and, last but not least, to ensure that the final regenerated organ fulfils the necessary requirements before implantation in a patient. Histology is used in current studies, and this is both destructive and 2D; conventional micro-CT cannot be used because it lacks the necessary soft-tissue contrast. We have shown that the high soft-tissue sensitivity of our phase-contrast CT methods provides outcomes comparable to histology, while remaining fully three-dimensional and non-destructive. We therefore expect its adoption to lead to significant progress in the clinical translation of regenerative medicine and tissue engineering approaches. Original paper: https://www.sciencedirect.com/science/article/abs/pii/S1742706122000319


Figure showing CT of decellularized oesophageal scaffold and corresponding histology slice.
Comparison between a phase contrast CT of a decellularized oesophageal scaffold (a) and the corresponding histology slide (b). Half the slide has been turned to greyscale, to highlight the similarity between the visibility of the soft tissue detail in the two cases. (a), however, is non-destructive and fully 3D.
Digital histology of breast tumors

By trading-off on acquisition speed, the scanner we developed for our intra-operative study of wide local excisions in breast conserving surgery (https://www.ucl.ac.uk/news/2021/feb/advances-x-ray-imaging-can-help-pati...) can also be used to perform digital histology with much higher resolution and detail definition. This is because the pre-sample mask placed before the sample to make the system sensitive to phase effects redefines the resolution properties of the system: with a proper acquisition scheme where the sample is laterally displaced at every projection angle as well as rotated, a final resolution equal to the size of the apertures in the pre-sample mask can be achieved, practically allowing a “user-defined” resolution. Application of this approach to tumour-bearing breast specimens revealed a wealth of previously undetectable details – notably necrosis and tissue response to chemotherapy, alongside much finer tumour strands. Notably this can be performed with exactly the same machine, simply by swapping masks and setting up a slightly different acquisition sequence. Original paper: https://ieeexplore.ieee.org/document/9661394

Breast tumor phase and histology images
Detail of a breast tumour images with the high resolution, high sensitivity phase scanner (a) and direct comparison to the histology gold standard (b). The yellow arrows point at (reduced-density) regions where the tissue responded to chemotherapy, and the red one at a residual infiltrating ductal carcinoma nodule.
Ptychography with laboratory sources

Ptychography has been one of the hottest topics in x-ray research over recent years. It is a lensless x-ray microscopy method, in which the lens is replaced by algorithms. Far-field coherent diffraction patterns are collected as a probe is scanned over a sample, and processed through an iterative algorithm to produce images in which resolutions of a few nanometers can be reached. Due to its extreme coherent requirements, so far ptychography was considered restricted to highly specialised facilities like 3rd generation synchrotrons. We have demonstrated that, by developing an appropriate set up based around a high-brilliance laboratory source, it can also be made to work in a conventional laboratory environment (Batey et al Phys. Rev. Lett. 126 (2021) 193902).

Figure showing coherent diffraction patterns and their result after processing through ptycographic algorithms
Left: two examples of the coherent diffraction patterns collected by Cipiccia’s team with a laboratory source. Right: their processing through a ptychographic algorithm yields an image in which the sub-micron spokes of a star-shaped resolution pattern are clearly resolved.
Dynamic multi-modal x-ray imaging

As well as in CT, the need to laterally displace the sample in steps equal to the aperture size to achieve maximum resolution is a significant limitation also in dynamic imaging. We have solved this problem by continuously translating the pre-sample mask as images are acquired. The use of a high-resolution detector allowed retrieving attenuation, phase and scatter from each individual frame, and therefore in co-registered attenuation, phase and scatter videos of evolving phenomena. Application to real-time laser melting of aluminium powder (additive manufacturing) showed how phase and scatter can detect molten regions much earlier than attenuation; the tracking of powder particles allowed identifying accumulation regions where droplet subsequently formed, before the melting process actually began (Massimi et al Phys. Rev. Lett., in press)

Figure showing multiple contrast channels of metal being melted in sequential imaging frames
Dynamic sequence showing how the formation of liquid droplets becomes apparent much earlier in the phase (middle row) and dark-field (bottom row) images compared to conventional x-ray attenuation (top row). Time (in milliseconds) from when the laser hits the powder bed is reported in all images.
Intra-operative specimen imaging

In breast conserving surgery, only the tumour is resected rather than the entire breast as was the case in mastectomies. However, there is a need to ensure that the entire tumour has been resected: where this is not the case, recurrencies can occur. At the moment this is done through histopathological analysis, which takes days and can lead to re-operations when tumour involvement at the margins is found. By building our development of fast lab CT (see above) and our demonstration that edge illumination allows the realisation of compact XPCI CT systems (Havariyoun et al Phys. Med. Biol. 64 (2019) 235005), we showed that tumour involvement can be detected in real time by using XPCI CT. Real time detection would allow the surgeon to resect more tissue, consequently reducing the number of re-operations (Massimi et al Sci. Rep. 11 (2021) 3663)

Figure showing the comparability of non-destructive EI-CT slice with destructive histology slicing
Figure showing the comparability of non-destructive EI-CT slice with destructive histology slicing
Cycloidal computed tomography

In edge-illumination XPCI, the ultimate resolution is determined by the size of the apertures in the pre-sample mask. However, to access it, it is necessary to laterally displace the sample in steps equal to the aperture size, until the full gap between two adjacent apertures is covered. We demonstrated that comparable results can be obtained by roto-translating the sample instead, which is compatible with fast flyscans in CT. The ability to increase the resolution at no dose expense offers prospects also to reduce the dose at the “original” resolution (Hagen et al Phys. Rev. Appl. 14 (2020) 014069, see also feature stories in Physics Today and Physics World).

Figure showing the variation in image quality when using different sampling approaches.
Image quality obtained with a full lateral scan of the sample (a), rotation only (b), roto-translation (cycloidal) CT (c ). The dose and exposure time are the same for (b) and (c), while they are eight times higher for (a).
Detection of individual sub-pixel features

Scatter imaging is known to provide a signal proportional to the amount of inhomogeneity present in a sample on the sub-resolution level. While up to this point it was assumed that a distribution of “micro-scatterers” was needed to generate a detectable signal, here we show how also a single sub-pixel feature can create a detectable scattering signal (Matsunaga et al J. Phys. D: Appl. Phys. 53 (2020) 095401)

Figure showing that a sub-resolution feature (a crack) becomes visible in dark field contrast channel
A single line crack is detected in the scatter image (broken yellow ellipse in (c)) and invisible in the corresponding attenuation (a) and differential phase (b) images.
Enhanced damage detection in composite materials

Composites are a mainstay of the aerospace and the transport industry in general but their very nature makes them difficult to test. We have shown that a much better characterisation of impact damage can be obtained through XPCI, that the attenuation/phase/scatter signal show complementary aspects of the damage, and that the latter in particular is capable of revealing damage well beyond the abilities of state-of-the-art conventional CT (Shoukroun et al Compos. Part B-Eng. 181 (2020) 107579)

Figure showing complimentarity of the different X-ray contrast channels (attenuation, refraction and dark field)in an additive manufacturing sample.
Complementarity among attenuation (a), differential phase (b) and scatter (c) images, best appreciated in panel (d) where the three channels ate represented in blue, green and red, respectively.
Pre-clinical imaging

Most drug development is based on small animal models, with the animals (typically mice and rats) being regularly imaged to study the effect of the administered drugs. In current pre-clinical CT, this is hindered by the limited soft tissue sensitivity of conventional x-rays. We provided a first demonstration of the advantages that can be brought by the introduction of edge-illumination XPCI in this field (Hagen et al Mol. Imaging Biol. 22 (2019) 539-48)

CT scan of a mouse thorax.
Increased soft tissue sensitivity and detailed visibility of phase contrast (b) vs conventional (a) CT slice across a mouse thorax.
 
Quantitative scatter imaging

Interest in scatter or “dark field” imaging has grown over the years and first attempts at quantitation yielded scatterer concentrations over certain areas, typically through some calibration procedure. Albeit within some validity limits, we have developed a theoretical model that links the scatter signal to the unit decrement of the real part of the refractive index of the sample, and showed its inversion can yield the average size of the scatterers as well as their concentration (Modregger et al Phys. Rev. Lett. 118 (2017) 265501).

Scatter Diagram
Extraction of a quantitative “size parameter” from scatter images and its correlation with the average particle size of the scatterer.
Reliable phase detection from highly scattering samples

Phase is notoriously difficult to detect effectively from highly scattering samples, which tend to excessively broaden and/or flatten the sensitivity curves used in phase retrieval procedures. We showed this could be overcome by using more than one detector aperture to analyse the beam created by one sample aperture. Concurrently, we show the linearity of the scatter signal vs scatterer thickness which makes it suitable to be cast as a line integral and therefore reconstructed in a CT procedure (Endrizzi et al Phys. Rev. Lett. 118 (2017) 243902)

Scatter
Reliable retrieval of the phase shift (left) generated by a plastic cylinder buried underneath increasingly thick layers of highly scattering material (paper, right).
Large field of view, high-energy scanner

Most early XPCI systems only allowed small fields of view, especially at low energy. By building on our previous development of the “asymmetric mask” concept (Endrizzi et al Sci. Rep. 6 (2016) 25466), which allowed a full retrieval of attenuation/phase/scatter in a single acquisition so long as the sample is scanned through the x-ray beam, we developed the first scanner capable of acquiring images of objects up to 20 x 50 cm2 in size at spectral energies of up to 120 kVp (Astolfo et al Sci. Rep. 7 (2017) 2187).

Xray of a laptop
Image of a full laptop acquired at 120 kVp - attenuation (left) and scatter (right) are shown.
Fast lab-based phase contrast CT

Laboratory-based phase contrast CT typically takes several hours. CT requires quantitative voxel contents obtained through phase retrieval; this typically entails displacing an optical element at multiple positions for each CT projection, preventing flyscans, adding dead times and significantly lengthening the scans. The adaptation of our “single image retrieval” method (see above) to a laboratory setting removed all these limitations, allowing lab-based XPCI CT scans in minutes for the first time (Diemoz et al Phys. Rev. Appl. 7 (2017) 044029; see also focus story “3D Images 10 Times Faster” on “APS Physicshttps://physics.aps.org/articles/v10/48).

CT scan of a rat heart
XPCI CT scan of a rat heart sample acquired in just over 3 minutes.
Minimising dose in mammography

There has been wide speculation on the dose-saving potential of XPCI and here we provide an experimental demonstration that reductions of more than one order of magnitude are within reach. Higher energy x-rays are more penetrating, meaning their probability of being stopped (thus depositing dose) in tissue is reduced. A “sweet spot” exists where the reduction in the number of x-rays that are stopped is not overcompensated by the higher amount of energy they release in tissue when they are: here the dose reduction is maximal. In mammography, conventional attenuation contrast vanishes at such a high energy, but phase contrast is still detectable. (Diemoz et al Phys. Med. Biol. 61 (2016) 8750-61)

image of a tumour in breast tissue
(a) conventional image of a tumour in breast tissue. If an image of the same specimen is acquired at the much higher (monochromatic) energy of 60 keV, practically all features disappear (b). However, in a differential phase contrast image acquired with edge illumination, interfaces between tissue types are still visible (c). Its integration (d) leads to an image at least as good as (a), but at approximately 20 times less dose.
Image quality comparable to the synchrotron gold standard

Alongside demonstrating the usefulness of XPCI as a characterisation and monitoring tool in regenerative medicine, in this paper we provide the first comparison between images of the same sample type obtained with a laboratory edge-illumination system and with the synchrotron gold standard, showing comparable image quality (Hagen et al Sci. Rep. 5 (2015) 18156).

images of a decellularized rabbit oesophagus
Synchrotron (left) and laboratory (right) images of a decellularized rabbit oesophagus demonstrating the same degree of contrast between different soft tissue types.
Achromatic phase contrast imaging

Many XPCI methods depend heavily on the x-ray energy, for example methods based on crystals only work with monochromatic beams, and grating interferometer have a “design” energy at which their performance is optimal. We showed that, in the limit of fully absorbing masks, edge-illumination is a fully achromatic method (Endrizzi et al Opt. Exp. 23 (2015) 16473-80).

Histograms from the scatter signal
Histograms from the scatter signal originating from a collection of microsphere obtained from monochromatic images weighted based on the used spectral entries (green) and by summing all collected data regardless of energy (red). The correspondence between the histograms demonstrates the achromaticity of the system – no energy is preferentially weighted when all collected bins are simply summed together. Yellow and blue histograms show the same for the background values, the corresponding images are shown as insets.

Previous Advances

Phase-retrieval with polychromatic beams

Conventional x-ray sources emit multiple wavelengths at the same time: this means the retrieved phase is necessarily a weighted average over the polychromatic spectrum. We developed a framework to calculate this average, depending on the XPCI method and the sample (Munro and Olivo Phys. Rev. A 87 (2013) 053838)

Dark-field or “ultra-small angle scatter” imaging

Rather counter-intuitively, phase-based x-ray methods can enable access to sample features well below the spatial resolution of the imaging system. Marco Endrizzi has developed an acquisition procedure and analysis method that enables this using the UCL XPCI setup (Endrizzi et al. Appl. Phys. Lett. 104 (2014) 024106)

X-ray contrast images of a rose
Attenuation (left), phase (centre) and dark-field (right) images of a rose. As can be seen the images are complementary and highlight different characteristics of the sample.
Low-dose phase-enhanced mammography

XPCI allows improved detection of breast cancer, as demonstrated by the in vivo study underway at the Trieste synchrotron, partly based on Olivo’s early research. We have demonstrated that the UCL method allows similar results to be obtained with conventional sources, opening the way to clinical translation (Olivo et al Med. Phys. 40 (2013) 090701, – see also feature on Physics Today)

X-ray images of a mammogram
(a) conventional vs (b) lab-based XPCI (b) imaging of tissue samples, with emphasis on the detection of thin structures (in this case stromal trabeculae, along which the tumour tends to spread), invisible to conventional methods. Arrows in (b) point at obvious examples.
Phase Tomography

The special features of the UCL XPCI method mean that tomography can be performed quickly and at low dose. Charlotte Hagen demonstrated this first at synchrotrons (Hagen et al Opt. Exp. 22 (2014) 7989), then with our laboratory sources achieving record-low dose delivery (Hagen et al Med. Phys. 41 (2014) 070701, editor’s pick and cover feature).

3D X-ray image rendering of a wasp
3D rendering of the x-ray phase shift created by a domestic wasp
Phase-enhanced tomosynthesis

tomosynthesis is an emerging modality in various fields – notably breast imaging where the use of conventional CT has limitations. Magdalena Szafraniec demonstrated it can be easily combined with the UCL XPCI method while maintaining dose and exposure time compatible with clinical requirements (Szafraniec et al Phys. Med. Biol. 59 (2014) N1, see also feature on Medical Physics Web

Nano-radian sensitivity at synchrotrons

Having developed a method that generates strong phase signals with incoherent sources, we asked ourselves the question “what happens if we bring it back to coherent sources”. Paul Diemoz demonstrated it leads to unprecedented phase sensitivity of 2 nrad: the angle subtended by 2 micron at a distance of 1 km (Diemoz et al Phys. Rev. Lett. 110 (2013) 138105)

X-ray image of cells demonstrating nanoradian phase sensitivity
Single cells lining up in the veins of a flower petal individually resolved through ultra-sensitive phase-based x-ray imaging
Single-image retrieval

A key limitation of XPCI methods is that they require more than one image to disentangle phase from attenuation – the latter does not disappear because a system is made sensitive to phase. Inspired by pioneering work by Paganin et al (J. Microsc. 206 (2002) 33), Paul Diemoz developed a single-shot phase retrieval method (Diemoz et al J. Synchrotron Radiat. 22 (2015) 1072)

X-ray images of a rabbit oesophagus retrieved usinga single and multiple exposure
Single shot (left) vs. “conventional” (double-shot) retrieval of the phase shift induced by soft tissue structures inside a rabbit oesophagus. As can be seen, image quality is very close in the two cases.
Beam tracking

The UCL XPCI method uses x-ray masks to achieve sensitivity to phase. Fabio Vittoria demonstrated that, if a detector with sufficient resolution is available, not only can one of these masks be eliminated, but attenuation, phase and dark-field can be acquired in a single shot (Vittoria et al Appl. Phys. Lett. 106 (2015) 224102 and Sci. Rep 5 (2015) 16318)

Colour-contrast X-ray image of a bamboo shoot
Microstructure of bamboo wood visualized through the beam tracking method, which simultaneously provides attenuation, phase and dark-field signals. These have been “fused” in a single image through use of different colours (red, blue and green respectively), which enables to appreciate their complementary nature – e.g. dark-field emerges in correspondence to the finest features.
New imaging agents

Many x-ray procedures are based on agents such as iodine, which increase x-ray contrast but can cause severe allergic reactions. Other modalities like ultrasound use harmless agents such as microbubbles, which are invisible to x-rays. Tom Millard proved XPCI methods enables their visualization (Millard et al Appl. Phys. Lett. 103 (2013) 114105 and Sci. Rep 5 (2015) 12509)

X-ray images of microbubbles in cuvettes
Cuvettes with and without microbubbles in suspension, imaged in conventional absorption (left) and dark-field (right). The two cuvettes look identical in absorption, but dark-field clearly enhances the presence of microbubbles.
Phase imaging with misaligned optical elements

One of the key challenges in the translation of XPCI into real-world environments is to keep optical elements (e.g. x-ray masks) aligned during imaging. Marco Endrizzi developed a methodology that enables achieving high quality, quantitatively correct results also with misaligned masks (Endrizzi et al Appl. Phys. Lett. 107 (2015) 124103)

Raw and retrieved X-ray images from Edge Illumination system
Panels (a) and (e) show field of view without a sample for aligned and misaligned x-ray masks respectively, with misalignment causing Moire fringes that distort the illumination field significantly. This notwithstanding, indistinguishable attenuation (b and f), phase (c and g) and dark-field (d and h) images are extracted in both cases.
Phase microscopy at high x-ray energy with laboratory sources

X-ray microscopy is a challenging area which often requires specialized sources and is usually implemented at low x-ray energy. Marco Endrizzi demonstrated that micrometric resolution, plus quantitative phase retrieval, can be achieved at high x-ray energies with standard laboratory equipment (Endrizzi et al Opt. Lett. 39 (2014) 3332).

X-ray microscopy image of an insect leg
Microscopic high-energy phase image of an insect leg

 

Quantitative phase retrieval with incoherent sources

This was considered a virtually impossible task, but was made possible by Peter Munro’s combined analysis of pairs of phase images acquired with the UCL method (Munro et al PNAS 109(2012) 13922), inspired by Dean Chapman’s pioneering work with crystals (Chapman et al Phys. Med. Biol. 42 (1997) 2015)

images of a ground beetle
Attenuation (left) and phase (right) images of a ground beetle. The arrow in the bottom right corner of the latter image shows how thin hairs on the insect’s leg are visualized thanks to phase effects.