The application of absorption-based X-ray computed tomography (CT) techniques in geoscience and material science has greatly enhanced our understanding of complex materials such as rocks (e.g., Cnudde et al., 2006; Gualda et al., 2010; Boone et al., 2011; Baker et al., 2012; Cnudde and Boone, 2013; Wildenschild and Sheppard, 2013). X-ray CT produces volumes that represent the spatial distribution of an object’s constituents (used here to refer to compositional phases within a multi-phase material). Within a reconstructed absorption-based X-ray CT volume, the gray intensity value for each volume pixel (voxel) represents the linear attenuation coefficient, µ, of the constituent that occupies that space. When a CT scan is acquired using a monochromatic X-ray source, µ is a constant that has a direct correspondence with the absorbing constituent (Cnudde et al., 2006). A heterogeneous object imaged with absorption-based X-ray CT will be well resolved (i.e., with a sufficient contrast) when its constituents’ linear attenuation coefficients are sufficiently different from both one another and the experimental background so that when the volume is reconstructed, each constituent has a unique gray intensity value (Wildenschild and Sheppard, 2013).

In general, when imaged with absorption-based X-ray CT, two classes of materials have contrast-related limitations: poorly-absorbing materials and hard materials. Poorly-absorbing materials (e.g., biological tissues, including wood, many synthetic or bio-polymers) composed of low-Z elements (where Z is the atomic number) have low µ values at the X-ray energies typically used for laboratory- or synchrotron radiation (SR)-based X-ray CT scanning (Hubbell and Seltzer, 1995). Thus, there is often minimal contrast between the experimental background and the scanned object. Hard materials, such as many geomaterials, are typically highly X-ray absorbing. However, many of them contain multiple constituents which can attenuate X-rays in a similar manner. This results in very similar µ values which complicates downstream data processing and ultimately, constituent identification and localization.

Phase-contrast X-ray imaging and phase-retrieval techniques have been developed over many years as a way to overcome contrast-related limitations (e.g., Cloetens et al., 1996; Wilkins et al., 1996; Cloetens et al., 1999). Such techniques have been employed to study soft materials including biological matter and polymers. They result in improved contrast leading to better downstream data analysis (Groso et al., 2006; Pfeiffer et al., 2008; McDonald et al., 2009; Zhu et al., 2010). In the geosciences, phase-contrast techniques have been used before to study volcanic pyroclasts with improved contrast (Zandomeneghi et al., 2010). More widespread use has been limited because most geomaterials are composed of numerous constituents and require complex phase-retrieval techniques (Wildenschild and Sheppard, 2013). Here we present grating-based synchrotron radiation computed tomography (GB-SRCT) on a geomaterial (i.e., a basaltic rock). To our knowledge grating-based phase-contrast X-ray imaging has not been documented for geomaterials. We show how GB-SRCT can be used to differentiate between constituents with similar absorption properties in a hard, highly absorbing material.

X-ray CT investigations of geomaterials often focus on elastic properties, porosity and permeability, as well as wettability and/or particle packing (e.g., Song et al., 2001; Totten et al., 2002; Naveed et al., 2013; Yang et al., 2017; Glatz et al., 2018; Ju et al., 2018; Marone et al., 2020; Chenlu et al., 2021). Since the physical and mechanical properties of an object are dictated by the nature and distribution of its constituents, it is necessary to be able to differentiate between them. With X-ray CT, image segmentation typically occurs after volume reconstruction to individualize the object’s constituents. Manual segmentation is rarely used; rather semi-automated or fully-automated classification schemes are instead employed to segment image on the basis of gray intensity values (Iassonov et al., 2009). However, automated thresholding, or other means of algorithmic classification, require sufficiently different µ-determined gray intensity values to distinguish constituents from one another. For CT scans collected at the same X-ray energy, constituents with similar chemical compositions and densities will therefore be hardly resolved. If these constituents have, for a same chemical composition, variable texture and morphologies, protocols based on texture and shape will not improve their identification (Iassonov et al., 2009). Linear attenuation coefficients are also influenced by the presence of pores. Accordingly, depending on its texture and porosity, the same constituent can have different µ values which complicates its identification and localization in heterogeneous samples.

Many minerals have similar densities or chemistries thereby producing similar or overlapping linear attenuation coefficients (Wildenschild and Sheppard, 2013; Figure 1 and Supplementary Figure S1). For example, basalt, an igneous silicate rock, is primarily composed of plagioclase feldspars and pyroxenes, with olivine and iron oxides as accessory minerals (see Supplementary Table S1 for respective formula). When weathered, these minerals alter into clays (mainly smectite), carbonates, zeolites, and other secondary minerals. Clays (ρ = 2–3 g cm^-3) and plagioclase feldspar (ρ = 2.6–2.8 g cm^-3) attenuate X-rays to a similar extent (Deer et al., 2013; Figure 1 and Supplementary Figure S1). Therefore, they can be nearly indistinguishable in absorption-based X-ray CT scans collected on synchrotron and laboratory-based systems (Figures 2A,B, respectively). This is a fundamental problem because plagioclase feldspar is a primary mineral, while clays are isochemically produced during low-temperature weathering.

GB-SRCT can be used to elucidate constituents with poor absorption contrast. This technique has been available for many years at the TOMCAT beamline of the Swiss Light Source (SLS) at the Paul Scherrer Institute (PSI) in Switzerland (Weitkamp et al., 2005; David et al., 2007; Marone et al., 2020). To improve beamline capability for GB-SRCT, the setup was upgraded to a permanently mounted flexible interferometer named DPC2.0 (Patera et al., 2017; Supplementary Figure S2). The results presented herein are some of the first results measured using DPC2.0. The technique enables the simultaneous acquisition of three signals, thus producing three data volumes: the transmission signal (absorption) which depends on µ, the differential phase-contrast, DPC, signal (small angle refraction), and the dark-field signal (result of local scattering). The DPC signal is sensitive to the refractive properties of a constituent. In basalt, the refractive properties of the two similarly X-ray attenuating constituents (i.e., the plagioclase feldspar and the clays; Figure 1) differ due to their crystallographic structure and packing properties. Clays are made of layered, platy particles that are packed together whereas plagioclase feldspar is a plain crystalline solid (Deer et al., 2013). This theoretically allows them to be differentiated by using the DPC signal. The dark-field signal is more sensitive to abrupt changes in density within an object (Pfeiffer et al., 2008; Strobl, 2014), and can provide structural information at a sub-voxel level. Therefore, it can be used to identify small fractures and cracks, as well as interfaces between basalt constituents and pores.

Figure 3 compares high-resolution SRCT in absorption mode (Figure 3A) with GB-SRCT performed using the DPC2.0 endstation (Figures 3B–D). The voxel sizes for the two techniques are 0.325 and 6.5 µm, respectively. Three different gray intensity values associated to mineral structures can be observed in slice data from both absorption-based SRCT (Figure 3A) and the transmission signal collected using GB-SRCT (Figure 3B). The layered structure of the clay is observable in the higher resolution SRCT scan, but clays are not associated with a different gray intensity value compared to plagioclase feldspars (Figure 3A). A fourth gray intensity value appears in the DPC signal (Figure 3C) which can be associated to clay minerals. Compared to plain plagioclase minerals, the nano-scale layered structure of clays causes a higher amount of small angle refraction leading to a different distribution of gray intensity values in the DPC signal image with respect to the absorption-based signals (Supplementary Figure S4). Commonly used segmentation methods (e.g., watershed transformation, histogram-based thresholding, edge detection) failed to segment 4 constituents from both absorption-based data volumes (Supplementary Figures S5A,B). However, the DPC signal provides the ability to segment the plagioclase phase (Figure 3C, #2) from the clay phase (Figure 3C, #4) using global thresholding methods (Supplementary Figure S5C). Figure 3D presents the dark-field signal which shows, thanks to local scattering, the propagation of microfractures and cracks as well as mineral boundaries and interfaces between basalt constituents and pores. If the dark-field image is processed using only automated segmentation techniques, the detected fractures or pores will occupy a larger volume than found in higher resolution scans due to imaging constraints (i.e., signal scattering; Supplementary Figure S4D). However quantitative dark-field imaging can be performed when the autocorrelation length of the instrument setup is known (Strobl, 2014).

Although samples were hydrated in the present GB-SRCT experiment, a single test scan on a sample in air showed minimal phase wrapping artefacts and produced visually comparable data quality (Supplementary Figure S6). This suggests that rock samples do not necessarily need to be hydrated and immersed in water. This is further substantiated by additional work performed using earlier versions of the DPC system (Patera et al., 2017).

The partial volume effect (PVE) dictates that the gray intensity value of a voxel is an average of the linear attenuation coefficients of all constituents and air-filled or water-filled voids present in the area to which that voxel is spatially associated (Kerckhofs et al., 2008). Due to this effect, small pores, thin fractures, as well as boundaries between a spatially small constituent and its surroundings, which are the targets of many studies, are often poorly resolved using X-ray CT, unless very high resolutions are used. This can complicate the study of heterogeneous hard materials. Absorption-based SR and laboratory X-ray CT can be used to acquire scans with voxel sizes ranging from 20 to 30 nm to tens of millimeters. The DPC2.0 setup offers a voxel size limited to 3.25 µm due to the analyzer-grating period (G2; Supplementary Figure S2) and the current combination of the microscope and detector available at the TOMCAT beamline. To some extent, the lower resolution produced by the GB-SRCT technique can be made up by the dark-field signal, which shows significant promise as a technique to locate small fractures and cracks at a sub-voxel level (Supplementary Figure S5D). While grain boundaries are also highlighted together with microfractures and cracks by the dark-field signal, these two types of features can be unambiguously discriminated by combining the dark-field signal with the DPC signal that selectively images grain boundaries (Supplementary Figure S5B). Alternatively, as presented in Figure 2A, Figure 3A, small pores, thin fractures, and similarly X-ray attenuating constituents can sometimes be observed in high-resolution SRCT scans. However, they usually do not differ enough in their gray intensity values to be readily segmented (Supplementary Figures S5A,B). Such a limitation of absorption-based X-ray CT can nevertheless be overcome in some cases by the use of dual energy X-ray CT (Coenen and Maas, 1994; Van Geet et al., 2000; De Man et al., 2001; Van Geet et al., 2001), now widely used in the geosciences (e.g., Iovea et al., 2009; Purcell et al., 2009; Ghorbani et al., 2011; Tsuchiyama et al., 2013; Al-Owihan et al., 2014; Bauer et al., 2021). Dual-energy scans that make use of the energy dependence of the attenuation coefficient (Figure 1) are collected using X-ray beams of two different energies. This allow the density and effective Z of an object’s constituents to be assessed and has shown significant improvement in differentiating minerals (Coenen and Maas, 1994). Appropriate sets of standard materials for absolute interpretation of the linear attenuation coefficients and data volumes’ registration are nevertheless required with this approach (Van Geet et al., 2000) whereas the standard-free GB-SRCT technique provide the three different signals that allow identification of the similarly X-ray attenuating constituents within the same data volume.

Although they have lower spatial resolution, the three data volumes generated by GB-SRCT may ultimately be more informative than conventional SRCT when studying heterogeneous hard materials with mineralogically and/or chemically different yet similarly X-ray attenuating constituents. In the case of geomaterials, clay minerals and plagioclase feldspars are generally indistinguishable on the basis of gray intensity values in absorption-based (SR)CT scans (Figure 2, Figure 3A). Typically, phase-retrieval methods are required to differentiate between them. Since grayscale contrast between different constituents is necessary for automated segmentation techniques, the DPC signal greatly enhances the ability to study mineral distribution in crystalline basalt (Figure 3C), and other heterogeneous materials comprised of constituents with similar or overlapping µ.

Absorption-based X-ray CT allows for data to be acquired at a higher magnification than GB-SRCT. However, the additional signal information provided by GB-SRCT allows for improved segmentation, fracture identification, and enhanced grain-edge resolution. When spatial resolution increases, the impact of the PVE should decrease (Figures 3A,B) hence leading to higher resolution and reliability with regard to the real nature of the imaged material. However, for heterogeneous materials, the sample size necessary to be able to acquire absorption-based SRCT scans with a voxel size of 0.325 µm (Figure 3A) is likely not a representative elementary volume (REV). Sample size, voxel size, and what is a REV are strongly dependent on the purpose of the X-ray CT study, and should be considered on a per-study basis (Al-Raoush and Papadopoulos, 2010). Accordingly, the trade-off between high resolution and sample size must always be considered. We acknowledge that a strict comparison between GB-SRCT, absorption-based SR and laboratory X-ray CT approach would have required the use of centimetre-sized samples for all. However, the highly absorbing nature of the basalt samples used in this study does not yield sufficient resolution using the SRCT setup at the TOMCAT beamline that could not be optimised for the collection of high-quality data for samples of that size. For the purpose of this work, the same samples were used for both absorption-based and GB-SRCT. However, the field of view available on the DPC2.0 setup can accommodate samples up to 12.6 × 4.5 mm². There are opportunities to migrate grating-based X-ray CT techniques into the laboratory where the different incident X-ray energies used may allow for increased sample size without imaging artefacts. Grating-based X-ray CT is actually being developed for laboratory-based systems (Birnbacher et al., 2016). Quantitative dark-field tomography on larger geological samples could greatly enhance porous and fractured media studies due to the identification of sub-voxel size fractures and pores in more representatively-sized samples. Fine fractures and small pores (in reference to feature versus voxel size) are often difficult to segment due to the PVE. However, with the dark-field signal they can be more readily identified and binarized (Supplementary Figure S5D), reducing the need for manual segmentation. Given that contrast-based limitations of absorption-based (SR)CT occur even with a voxel size of 0.325 µm, we suggest that GB-SRCT can represent an alternative means for phase-retrieval X-ray CT without the acquisition of many computationally heavy data sets. Further, GB-SRCT acquires the three signals–transmission, DPC, and dark-field–simultaneously. The resulting data volumes require no registration because they are already aligned on the same coordinate system.

Finally, combining chemical and/or mineralogical data with GB-SRCT data could allow for semi-quantitative to quantitative characterization in 3D. X-ray fluorescence (XRF) tomography is combined with absorption-based X-ray CT already (de Jonge and Vogt, 2010; Boone et al., 2011; Egan et al., 2015; Laforce et al., 2017), but measuring capabilities are limited below the surface of dense, X-ray absorbing materials that affect the depth at which emitted XRF can be detected. Surface XRF, SEM-EDS/BSE, or electron microprobe signals can be co-registered to the surface of the collected X-ray CT scans. Then chemical signals are correlated to gray intensity values from the CT scans. That correlation can be used to extrapolate the global 3D distribution of elements within an object. Such approaches can even be improved if, instead of chemical mapping, X-ray CT is combined with automated mineralogy (Schulz et al., 2020) that involves two dimensional mineral mapping using for example QEMSCAN, MLA, Mineralogic, TIMA, AMICS, INCA Mineral, or Q-WRMA systems (e.g., Corti et al., 2019; Wang and Miller, 2020; Warlo et al., 2021) and allows quantification of, among others, mineral composition, mineral association, and grain size distribution. However, similarly X-ray attenuating constituents can make such extrapolation difficult to impossible. Quantitative integration of X-ray CT with automated mineralogy remains challenged by the quality of X-ray CT data segmentation, especially for rocks with low porosity and complex mineralogy (Warlo et al., 2021). GB-SRCT data, with its three signals, can be more accurately compared with elemental analysis techniques such as SEM-EDS or XRF than just absorption data (Figure 4). Such comparison would allow for chemical maps to be co-registered to the DPC and transmission signals thereby improving segmentation and reducing the errors associated with extrapolating global element distribution only on the basis of gray intensity values. Material degradation, especially in the context of rock alteration, can often result in isochemical mineralogical transformations (i.e., implying primary and secondary phases with similar µ values), thus coupling elemental analysis techniques with multimodal GB-SRCT imaging can be highly informative.

Partially altered Icelandic basalt was the geomaterial selected for this study, in part because of its high content in plagioclase and clay. Yet many geo- and engineered materials (e.g., cement) are composed of similarly X-ray attenuating constituents, thus limiting the usefulness of absorption-based X-ray CT in the geosciences. Many of the benefits observed when GB-SRCT is used to image hard and soft materials such as laser-welded joints and biological samples are also present when it is used to image a heterogeneous hard material such as basalt (Herzen, 2010). The DPC signal can enhance contrast between similarly X-ray attenuating constituents, while the dark-field signal can be used to examine spatially sub-voxel features such as pores, voids as well as small fractures and cracks. The possibility of applying this technique to other geomaterials or heterogeneous hard materials opens up new possibilities in particular for rock alteration, as well as porous and fractured media studies.