State-of-the-Art Research on Loess Microstructure Based on X-ray Computer Tomography

Abstract

Computer tomography (CT), combined with advanced image processing techniques, can be used to visualize the complex internal structures of living and non-living media in a non-destructive, intuitive, and precise manner in both two and three-dimensional spaces. Beyond its clinical uses, CT has been extensively employed within the field of geotechnical engineering to provide both qualitative and quantitative analyses of the microstructural properties of loess. This technology has been successfully applied in many fields. However, with the rapid development of CT technology and the expansion of its application scope, a reassessment is necessary. In recent years, only a few documents have attempted to organize and review the application cases of CT in the field of loess microstructure research. Therefore, the objectives of this work are as follows: (1) to briefly introduce the development process of CT equipment and the basic principles of CT and image processing; (2) to determine the current state and hotspots of CT technology research based on a bibliometric analysis of the literature from the past three decades in the Web of Science Core Collection and CNKI databases; and (3) to comprehensively review the application of CT to explore the microstructural characteristics (such as particle size, shape, arrangement, and the connectivity, orientation, and pore throats of pores, etc.) and the evolution of structural damage in loess within geotechnical science. In addition, the progress and deficiencies of CT applications in the field of loess microstructure are summarized, and future prospects are proposed.

1. Introduction

Loess is characterized by well-developed vertical joints, high strength, and a porous and loose nature in its natural state. In China, loess is primarily distributed in the areas around the Kunlun Mountains, the Qinling Mountains, Mount Tai, and the Lu Mountains, concentrated in the arid to semi-arid regions north of 34° to 35° in latitude, and is most extensively spread in the middle reaches of the Yellow River, forming the Loess Plateau with unique topography. The plateau covers an area of 276,000 square kilometers and is the largest loess plateau in the world [1]. The pores in loess are one of the main characteristics that distinguish it from other soil types and determine its engineering properties, such as permeability and collapsibility. They also determine its complex collapse mechanisms. Disasters caused by loess collapse have long affected the construction and safe operation of engineering projects in the loess areas of this country, presenting a significant risk to the integrity and security of human life and possessions and constraining construction and economic development [2]. The occurrence of these disasters is closely related to the microstructural characteristics of the loess.

The study of the microstructure of geomaterials has largely benefited from the rapid development of advanced precision instruments and testing technology levels. Currently, common methods for observing the microstructure of geomaterials include optical microscopy, scanning electron microscopy (SEM), mercury intrusion porosimetry (MIP), and computed tomography (CT), among others. CT has been widely applied in the study of the microstructure of geomaterials due to its non-destructive, quantitative, repeatable, high-resolution capabilities and its ability to obtain the true geometric characteristics of particles and pores. Additionally, three-dimensional reconstruction and visualization provide strong technical support for the refined characterization of the microstructure of geomaterials [3,4]. In recent years, a multitude of researchers have used CT scanning to construct the three-dimensional structure of loess, achieving a wealth of research findings.

Considering the significance and broad application prospects of CT, as early as 1988 [5], Avinash Kak in 2001 [6] and Duliu in 1999 [7] described in detail the principles and applications of computerized tomography. Building upon this foundation, Carlson [8] conducted an in-depth summary and elucidation of the specialized terminology that may emerge during the CT scanning process. It is worth noting that, possibly due to technological limitations in earlier years, their research was predominantly focused on rock and soil particles. Despite the inherent necessity of X-ray CT in visualizing granular geomaterials, coarse sand, or glass beads, its capacity to reveal the intricacies of soil structure and pore space distribution remains an area that requires further exploration. Consequently, conclusions derived from research based on idealized systems may not accurately reflect the characteristics of real soil [9]. Pires et al. [10] offered a synthesis of the advancements in the application of CT in soil physics, yet their review was confined to applications pertinent to Brazil. The swift progression of computed tomography (CT) has broadened its sphere of influence, simultaneously introducing novel challenges to the methodologies of data acquisition and image analysis within the domain. The situation is especially evident as advancements in both hardware and analytical methods are made, and research questions in the field of geotechnical bodies become more complex and in-depth. Moreover, while several review articles have been published in recent years, they have predominantly focused on technological advancements in the imaging of plant roots and biological processes [11,12], as well as the evolution of microstructures [13]. There is a relative scarcity of comprehensive case study reviews that delve into the application of computed tomography (CT) in the integrated fields of soil and plant research [14,15]. Specifically, there is a dearth of review articles that concentrate on the application of CT in the study of the microstructure of loess within geotechnical bodies.

Against this backdrop, the objective of this research is to offer a contemporary assessment of the utilization of computed tomography (CT) within the sphere of loess microstructure investigation. This paper begins with an exposition of the essential concepts and historical context of CT, complemented by an overview of the fundamental tenets of image processing techniques. Subsequently, it assesses the noteworthy progress achieved in the realm of CT scanning over the past years, and it synthesizes an array of specialized methodologies applied in the context of CT scanning and image processing for pertinent applications. Furthermore, bibliometric analysis is applied to delineate the research trends and hotspots for CT applications. This review also systematically revisits the application instances of CT technology in the study of loess microstructures in recent years.

2. The Development History of CT Technology

2.1. The Origin of X-ray CT

Since Wilhelm Röntgen first discovered X-rays, this radiation technology has been widely applied in numerous research fields. The principle of X-ray CT in detecting the structure of an object mainly utilizes the Compton scattering and photoelectric effect that occurs when the rays pass through the scanned object, during which the rays undergo energy attenuation [16]. When the ray beam passes through a substance with an attenuation coefficient greater than zero, its intensity attenuates exponentially, and the attenuation coefficient is determined by the variations in the material density and atomic composition of the scanned object [17]. Assuming that the incident beam is monochromatic and the irradiated object is homogeneous (i.e., it has a constant attenuation coefficient), then mathematically, it can be represented by the Beer–Lambert law of attenuation [18], which denotes the penetration intensity I of monochromatic X-rays as they pass through an object:

$$I=I_0{e}^{-\int \mu (s)ds}$$

(1)

Here, I₀ represents the initial intensity of the incident beam, and μ(s) is the local linear attenuation coefficient along path s of the X-ray through the object.

This property was quickly applied to both medical and non-medical uses. In the field of geotechnical science, primarily over the past 30 years, numerous scholars have utilized X-ray imaging technology to extract and analyze the internal microstructures of a multitude of soil samples [19,20,21,22,23,24,25,26].

2.2. The Detection Principle of CT Technology

Computerized Tomography technology, often referred to simply as CT, is a technique that utilizes an X-ray source to rotate around the object under test and collect information on the attenuation of the rays to reconstruct two-dimensional or three-dimensional images of the object. Compared to traditional radiographic imaging techniques, CT technology can image a specific slice of an object being tested, effectively avoiding interference from other tissue structures. This capability provides clearer images of the internal structure. Additionally, CT technology is highly sensitive when identifying material density and atomic number, making it widely applicable in the field of non-destructive testing.

The X-rays emitted by the signal source of a CT device follow a specific equation when they penetrate an object:

$$I=I_0\exp (-{\mu }_m\rho x)$$

(2)

In Formula (2), Ι₀ is the initial X-ray intensity before the material is penetrated (ev/m²$\cdot$s); Ι is the X-ray intensity after penetrating the material (ev/m²$\cdot$s); ${\mu }_m$ is the mass attenuation coefficient of the material (cm²/g); $\rho$ is the density of the material (g/cm³); and $x$ is the length that the incident X-ray travels through the material (cm).

Under normal circumstances, the absorption characteristics of a substance to X-rays can be described by the per unit mass attenuation coefficient, which is closely related to the wavelength of the incident X-rays. For X-rays of a specific wavelength, the wavelength remains constant. Therefore, the per unit mass attenuation coefficient can be combined with the density of the substance to derive the per unit volume attenuation coefficient (µ) as follows:

$$\mu ={\mu }_m\rho$$

(3)

Under computer control, the X-ray tube S and the detector D rotate around the center of rotation O. For each rotation position β, a certain detector receives the projection data $P\left(\beta ,\alpha \right)$ that penetrates the measured material as follows:

$$P(\beta ,\alpha )={{\displaystyle \int }}_{(\beta ,\alpha )}f(x,y)dl={{\displaystyle \int }}_{(\beta ,\alpha )}f(r,\phi )dl$$

(4)

In Formula (4), $\beta$ is the position parameter on the line, representing the distance from the origin along the line to a certain point, and $\alpha$ is the angular parameter, representing the rotation angle of the line relative to the x-axis. $P\left(\beta ,\alpha \right)$ is a function concerning position $\beta$ and angle $\alpha$.

The equation above represents the Radon transform, where CT images are generated from a multitude of projection data to determine the density distribution function $f\left(x,y\right)$ of the material under examination.

In actual scanning, it is necessary to calibrate and normalize $P\left(\beta ,\alpha \right)$ [27] to eliminate the non-uniformity of various detectors and obtain absolute measurement values. This is achieved by performing a pre-scan of the air and a standard water film:

$$P={\displaystyle \frac{P-P_A-P_W}{P_W}}$$

(5)

In Formula (5), P represents the preprocessed projection data; P_A is the projection value from the air scan; and P_B is the projection value from scanning a standard water film.

There are numerous methods for reconstructing images, with the commonly used Convolution Back Projection (CBP) method, the calculation formula for which is the following:

$$P(\beta ,\alpha )={{\displaystyle \int }}_0^{2\pi }{\displaystyle \frac{1}{L^2}}P(\beta ,\alpha )g(\alpha )d\beta$$

(6)

where function g ($\alpha$) is the convolution kernel function. The design of this function aims to minimize the artifacts caused by detector spacing while sacrificing some density resolution, thereby reducing the computational differences between central and peripheral back projections [28].

In Equation (4), $f\left(x,y\right)$ corresponds to the X-ray absorption coefficient of the point being examined, commonly referred to as the CT number (H value). By comparing the CT numbers after image reconstruction with the inherent CT numbers of the object’s cross-section, one can determine the damage (defects) within various parts of the object’s cross-section. Professor Housfield, the inventor of CT, defined the CT values for air, pure water, and ice as −1000, 0, and −100, respectively. Therefore, the relationship between the X-ray absorption coefficient of the object under examination and the CT number is converted as follows [29]:

$$CT(H)={\displaystyle \frac{\mu -{\mu }_m}{{\mu }_m}}\times K$$

(7)

In Equation (7), K is the scaling factor used to convert the differences in the absorption coefficient to Hounsfield units. In practical applications, this value is commonly set to 1000 to facilitate the representation of CT values within an easily understandable and comparable numerical range.

During CT scanning, the X-ray source and the detector rotate synchronously around the central axis of the object. The X-rays emitted by the source penetrate the cross-section of the object and are captured by the detector, which then performs photoelectric conversion. Subsequently, an Analog-to-Digital Converter (ADC) is used to convert the analog signal into a digital signal. These digital projection data are processed using a predefined image reconstruction algorithm to obtain quantitative information that reflects the X-ray absorption coefficients at various points within the object’s cross-section, thereby constructing a digital image of the object’s cross-section with µ values.

2.3. Development of CT Equipment

X-ray CT was initially developed by Hounsfield in the late 1960s to the early 1970s and was applied to the field of medical imaging, known as medical CT. By the end of the 1970s, researchers discovered that CT technology could effectively be used for industrial material structure detection, and the value of CT in industries and geology gradually emerged, leading to the birth of industrial CT. In the 1990s, due to significant improvements in image quality and detail resolution capabilities, industrial CT began to play an important role in the detection of cracks, porosity, material density, 3D petrophysical studies, and the 3D visualization of pore structures in natural building materials [30]. Currently, industrial CT has been widely applied in various fields, such as industrial engineering, pedology, geology, materials science, food science, archaeology, aerospace, coal, petroleum, natural gas, chemical engineering, nuclear industry, and geotechnical engineering [31], gradually becoming a common technology for inspecting geological and engineering materials.

Medical CT and industrial CT are based on the same fundamental physical principles [32], but due to different application purposes, their system architectures and designs vary; this paper compiles and organizes comparative information on the differences among various CT devices, as shown in

Table 1. Differences among various CT devices.

CT Category	Resolution	Common Current	Common Voltage	Equipment Schematic	Scanning Time
Medical CT	600 um	10~600 mA	120~140 KV		1~10 s
Industrial CT	50~500 um	10~1000 mA	225~600 KV		5~60 min or longer
Micro-CT	0.5~50 um	5~100 mA	30~160 KV		15~60 min or longer
Nano-CT	50~0.5 um	30 uA~50 mA	8~150 KV		1~60 min or longer

Note: The current, voltage, and scanning time used during CT scanning are contingent upon the scanning conditions.

[33,34,35]. For medical CT, the scanning subject is a patient, with the goal being human tissue. During scanning, patients lie still and flat, and the X-ray scanner rotates rapidly around the scanning area, with the scanning table moving forward and backward to achieve volumetric scanning. However, the limitation of this scanning method is that it is difficult to ensure high resolution. With technological advancements, the increase in the “number of rows” of medical CT detectors and the enhancement of the data acquisition system’s ability to synchronize the acquisition of image “slices” has led to an increased resolution (up to 600 um). This allows for more layers of images to be scanned in a shorter time during patient scanning, ensuring minimal radiation dosage and enabling more accurate diagnosis. Currently, more advanced CT technologies, such as 256-slice CT, dual-source CT, and PET-CT, have shown broad application prospects, mainly in non-invasive angiography, tumor diagnosis, and macro and micro disease diagnosis [36]. Additionally, some scholars have applied them to the fields of wood [37] and geotechnical engineering [38,39,40,41]. In contrast, the scanning subjects of industrial CT are more diverse, including metals, ceramics, rocks, soil, composite materials, and synthetic materials. Samples can rotate at a certain angle on the stage or adjust along the Z-axis of the stage while the X-ray source remains fixed, typically using fan-beam flat-scanning or cone-beam scanning methods. Compared to medical CT, industrial CT usually does not focus on reducing the X-ray radiation dosage but instead uses high-intensity X-rays to pursue higher scanning resolution and accuracy [18]. Therefore, the hardware requirements for industrial CT systems are higher, and the cost is also higher. It should be noted that the higher the resolution, the more the sample size is limited.

In the early stages, due to the widespread availability of medical CT and the high cost and limited acquisition of industrial CT, researchers conducted a series of geotechnical experiments using medical CT. These included studies on the damage and deformation patterns of frozen soil, the collapsible nature of loess, the evolution of rock fractures, and the monitoring and simulation of seepage in geotechnical bodies. Some important results were achieved, laying a solid foundation for research in this field. However, due to the resolution limitations of medical CT, the aforementioned studies could only be conducted at the microscale, observing only millimeter- or sub-millimeter-scale changes in pores and fractures. Moreover, they were mostly based on two-dimensional CT slices and numbers to represent various hydraulic parameters. For geotechnical bodies, their microstructural characteristics and changes are fundamental to determining macroscopic properties, and the anisotropy at the three-dimensional scale is a key and challenging point in revealing the true hydraulic properties.

With the evolution and iteration of CT technology, from the first generation to the fifth generation CT, and from planar scanning to cone-beam scanning, the development of high-energy, high-precision, and high-safety CT systems, such as micro-focus CT, nano-CT, and synchrotron radiation micro-CT, along with the advancement of three-dimensional reconstruction and visualization techniques, has greatly promoted the extensive application and popularization of industrial CT in various fields [36]. As for the definitions of micro-focus CT and nano-focus CT or the differences between them, they are currently somewhat ambiguous, lacking clear industry standards. Typically, the industry classifies devices with a resolution lower than 0.5 um as micro-focus CT, while those with a resolution higher than 0.5 um are referred to as nano-focus CT [34,35], as shown in Table 1. In the field of geotechnical materials, a multitude of significant research and application achievements have emerged, such as three-dimensional reconstruction and the quantitative characterization of soil structure [3,18,19,39,42,43,44,45,46]; the identification of soil macropores and the study of macropore flow and migration mechanisms [47,48]; the characterization, extension patterns, and seepage simulation of fractures and pores in granite, sandstone, oil shale, and coal rock bodies [49,50,51,52,53,54]; the mechanism of gas percolation in coal seams [55]; and the identification of paleosol pores and fractures in loess and the study of their hydrological significance for slope stability [43,44,55,56,57].

Loess is pervasively characterized by the development of macropores, and while there is no universal size standard for these macropores, summaries, and classifications by Fan et al. [58] suggest that the pore diameter of macropores ranges from 32 um to 500 um. The resolution of medical CT (600 um) is clearly inadequate for such a detailed investigation. Nowadays, the most advanced forms of micro-CT and nano-CT can achieve a resolution of 0.5 um or higher values. Therefore, these technologies can be used to describe the microstructure of loess more accurately. However, researchers must keep in mind that the limitation of the micro-CT and nano-CT, i.e., to achieve higher resolution of a smaller sample size, is required according to the principles of CT technology [34,35]. For instance, to attain a resolution of 0.5 um, the sample diameter must be less than 1 mm.

Consequently, the study on the microstructure of loess has primarily benefited from the rapid development of advanced precision instruments and testing technology levels. In addition, three-dimensional reconstruction and visualization have provided robust technical support for the detailed characterization of the microstructure of loess.

2.4. Development of CT Technology

Since its inception, CT has been subject to a series of advancements in technology, proving indispensable in the examination of porous substances without causing damage. Despite these developments, achieving precision in measurements remains a complex task, necessitating ongoing refinement and calibration [59]. Discrepancies in the dimensional assessments conducted by CT can be attributed to a multitude of elements, such as the setup of the testing apparatus, the constraints on image resolution, the emissions of radiation, subsequent image processing, and the intrinsic characteristics of the tested materials [60]. Lin et al. [61] crafted an automated calibration technique that leveraged the pattern of sine waves to precisely rectify any misalignment of the rotational axis. Villarraga-Gomez et al. [62] noted that the scope of CT assessments is predominantly confined by the extent to which X-rays can penetrate various materials. They advocate for enhancements in the precision of CT by fine-tuning its resolution and the scope of its measurements. In [63], they further explore how to determine the optimal measurement angle for samples under a rotation axis tilt angle of 10–35° by simulating the penetration pathway of X-rays and spatial analysis to minimize discrepancies within dimensional measurement. With the introduction of spiral CT scanning technology [64], the traditional CT scanning method has been replaced by this new technology.

Furthermore, the improvement in CT imaging resolution is pivotal for the ongoing evolution of CT technology. Houston et al. [65] underscored the importance of image clarity in the identification and delineation of the pore space structure, a process that necessitates a higher resolution, albeit with the potential introduction of increased noise. The potency of the X-ray beams significantly influences the signal-to-noise ratio (SNR), and the fine-tuning of this ratio is imperative for mitigating the adverse effects of X-ray scatter on the integrity of the image [66,67]. Conventional reconstruction methods, exemplified by filtered back projection (FBP), are adept at generating high-fidelity images; however, their effectiveness is predicated on certain idealized conditions, such as uniform X-ray photon energy and unvaried photon trajectories, which are not always attainable in practical scenarios and may lead to reduced accuracy in the stages of image segmentation and object recognition [68]. In recent years, deep learning reconstruction (DLR) algorithms [69] have been introduced to correct noise issues in CT images. Especially in dual-energy CT (DECT) [70], image reconstruction technology based on the total variable (SART-TV) [71] has shown excellent performance in suppressing artifacts and resolving the issue of sinogram loss. Zhang et al. [72] made technical improvements to reduce the impact of the non-ideal focal spot size of the X-ray source on image quality. Reducing the number of scans step by step (e.g., high-speed CT scans [73]) can effectively avoid motion artifacts [74]. The post-scanning image summation technique, known as fused CT, diminishes image noise [75], and the incorporation of contrast agents significantly boosts the contrast and overall image quality of the examined region [76,77].

After reviewing the application and development of CT technology in the non-destructive testing of porous materials (such as geotechnical materials), the importance of obtaining high-quality CT images becomes evident. However, these images alone are inadequate for conducting a quantitative assessment; techniques for image analysis are also required to extract and interpret useful information. Techniques for imaging analysis still pose challenges in determining pore boundaries or analyzing errors and distortions caused by CT reconstruction [78]. The following will discuss in detail the methods of CT image processing and analysis, including key steps such as noise reduction, edge enhancement, and image segmentation, as well as how to use advanced segmentation techniques and machine learning algorithms to improve the accuracy and efficiency of the analysis. The application of these methods will provide us with deeper insights, helping us to better understand and quantify the structural characteristics of porous materials.

2.5. CT Image Processing and Analysis

The core components of the image processing sequence are noise suppression, contour intensification, image partitioning, subsequent treatment, and structural evaluation [79]. Within X-ray CT imaging, the most challenging and extensively studied aspect is typically rooted in thresholding and segmentation procedures. The precision of soil porosity calculations from X-ray CT analysis is intrinsically linked to the thresholding process, which is primarily due to the correlation between the computed soil porosity and the grayscale values. Gackiewicz et al. [80] identified that errors associated with the thresholding step can range from 15% to 40% when calculating the visible total soil porosity and the saturated hydraulic conductivity. Prior to analysis, images must undergo thresholding based on diverse histogram shapes, spatial clustering, entropy measures, object attributes, spatial correlations, and local gray-level surface information [81]. These six categories facilitate the differentiation between soil particles and pores, converting the grayscale image into a binary format that smooths the boundaries between particles and pores [82]. Throughout this process, a variety of statistical and mathematical techniques can be implemented to enhance the accuracy and quality of the segmentation.

Currently, threshold-based segmentation techniques, encompassing both local and global approaches [79,83], are extensively utilized in image analysis. These methods discretize images into distinct datasets, enabling the differentiation between solid phases and soil pores, which is essential for the quantitative characterization of porous media. Local segmentation methods, in particular, demonstrate superior performance and robustness when addressing local image heterogeneity, making them the foundation of most advanced segmentation techniques [84]. Houston et al. [65] evaluated four commonly employed threshold-based segmentation methods in soil research, including thresholding techniques, adaptive window indicator kriging, fully automatic extension methods, and global segmentation. While the first three methods offer consistent image processing capabilities, they differ in their sensitivity to noise, with the initial two methods often leading to the underrepresentation of smaller pores and the smoothing of larger pore surfaces [85]. Wang et al. [86] highlighted that the soil’s porous framework is rarely captured accurately with a single threshold, underscoring the importance of selecting an appropriate threshold to secure the reliability of the segmentation for each CT photograph. Other segmentation methods in use include region growth [87] (e.g., Higo et al. [88] utilized micro-focus X-ray CT combined with digital image correlation (DIC) to analyze the strain localization behavior of partially saturated sand under triaxial compression. The study unveiled the macroscopic and microscopic progression of shear bands, as well as the alterations in the granular structure. A tri-value thresholding technique based on the region-growing method was proposed, which successfully differentiated between the solid phase, water phase, and gas phase. The calculated local porosity ratio and degree of saturation were found to be in good agreement with the measured values.), watershed algorithms [89], and clustering-based unsupervised and supervised learning approaches [90]. Among these, supervised machine learning methods, such as neural networks, random forests, support vector machines, linear and logistic regression, and classification trees, are widely adopted for their ability to learn from well-labeled data and effectively predict outcomes for new data or images [91].

Houston et al. [92] conducted a review of various software tools and methods for evaluating pore size distribution (PSD) using X-ray CT images. They assessed the performance of seven major image analysis techniques (Avizo, CT Analyser, BoneJ, Quantim4, DTM, DFS-FIJI, and 3DMA) on the CT images of different soil types. The evaluation results showed that the first five techniques were consistent with theoretical values in the PSD analysis of pore volume, while DTM and DFS-FIJI exhibited instability when analyzing complex soil samples. Mukunoki et al. [93] employed mathematical morphology techniques for the segmentation of CT images. Utilizing the granulometry method within mathematical morphology, they assessed the shape and size of pores by varying the size of the spherical elements. Furthermore, they verified the accuracy of area measurements and, through the concept of the representative elementary volume (REV), determined the optimal image size for evaluating the geometric properties of sand. Lucas et al. [94] proposed a novel method that combined Euler number X and Γ-curves to study pore diameter connectivity; it was also suitable for analyzing pore diameters spanning three orders of magnitude. Data regarding pore distribution, fractal dimension, and pore connectivity can be extracted using quantitative image analysis software (e.g., AVIZO, www.vsg3d.com; VGStudio-MAX, www.volumegraphics.com; and Image J, imagej.net, all accessed on 17 July 2024) or semi-automatic image processing plugins [95] (e.g., SoilJ). However, Zhao et al. [96] pointed out that many existing forms of integrated software and self-compiled algorithms face challenges in dealing with irregular pores and unclear boundaries, recommending an adaptive fuzzy C-means clustering algorithm (AFCM) with fuzzy decision-making capabilities. Weller et al. [97] created an open soil structure database aimed at facilitating image analysis and the modeling of soil structure and pore-related data.

Here, this paper also summarizes some CT image reconstruction processes and segmentation methods. Figure 1 illustrates an example workflow for reconstructing three-dimensional pore structures in CT soil images, and Figure 2 briefly describes the 3D microstructure reconstruction process of loess CT images.

3. Bibliometric Analysis

Clarivate Analytics’ Web of Science Core Collection (WoSCC) database and the China National Knowledge Infrastructure (CNKI) database are widely recognized as the most authoritative comprehensive academic research platforms, both internationally and domestically. The WoSCC database covers over 12,000 internationally renowned academic journals, while CNKI encompasses thousands of Chinese journals across various disciplines, including natural sciences, engineering technology, medicine, social sciences, and humanities [100]. To conduct bibliometric analysis, this paper, while taking into account previous research findings, selected the aforementioned two databases to screen and extract published academic articles from both international and domestic scopes, establishing two datasets representing the application of CT technology in loess microstructure-related research (sets-1) and its application in geology (sets-2). The search period for both sets was from January 1, 1999, to April 2024. The search criteria for sets-1 were as follows: TS = (‘loess’ + ‘loess microstructure’) AND TS = (CT + ‘X-ray CT’ + ‘computed tomography’ + ‘micro-CT’ + ‘gamma-ray CT’), yielding a total of 202 documents from the two databases, of which 130 were analyzed after manual screening to remove less relevant studies from the literature. The search criteria for sets-2 were as follows: TS = (CT + ‘X-ray CT’ + ‘computed tomography’ + ‘micro-CT’ + ‘gamma-ray CT’) AND SU = (Geology), resulting in a total of 3581 documents after the screening.

Bibliometric analysis tools can visually analyze large volumes of literature data through performance analysis and scientific mapping, such as CiteSpace, VOSviewer, Histcite 14.1, Gephi 0.10.1, etc. The first two software options each have their own strengths and features. CiteSpace integrates multiple aspects, such as cluster analysis, keyword co-occurrence, and research frontier analysis, offering a degree of summarization and foresight while also boasting a variety of built-in algorithms, rich map adjustment options, and the reliable verification of results; VOSviewer, a visualization analysis tool, is characterized by its neat and aesthetically pleasing output, which can complement and enhance the analysis performed by CiteSpace [101]. This paper combines the advantages of both bibliometric analysis tools to analyze the screened literature, specifically using version 1.6.20 of VOSviewer software and version 6.3.1.0 of CiteSpace software.

3.1. Trends of Annual Publication

Figure 3 reveals that the annual cumulative publication counts of sets-2 and sets-1 follow a nearly identical trend, yet there is more than an order of magnitude difference of over 100 times between them, indicating that the application of CT technology across various fields is already in a stage of vigorous development.

In sets-1, regarding the application of CT technology to the field of loess microstructure, the literature is divided into two parts: WoSCC and CNKI. There is a relatively low volume of publications in this area of research, with the annual publication count being below three pieces until the year 2010. In 2010, the first literature in this field was published in the WoSCC database, after which the publication counts in both databases began to increase and fluctuate, indicating that the application of CT in the loess microstructure was becoming more widespread.

3.2. Co-Occurrence Analysis of Keyword

This study conducted a keyword co-occurrence analysis on the literature within sets-1, filtering the condition of the keyword frequency to be greater than or equal to two times and selecting 147 and 56 keywords from the WoSCC and CNKI databases, respectively. The most frequently occurring keywords in the WoSCC database were “loess”, “X-ray computed tomography”, “soil pore”, “CT scanning”, “micro-computed tomography”, “microstructure”, etc., while in the CNKI database, the most frequently occurring keywords were “loess”, “CT scanning”, “microstructure”, “structural”, “collapsibility”, etc. Figure 4 and Figure 5 are keyword co-occurrence network diagrams generated by the VOSviewer software for visual analysis. In these diagrams, each hue corresponds to a distinct cluster of themes, with the prominence of the text and the intensity of the background shading reflecting the magnitude of the Total Link Strength (TLS). An increased text size signifies a higher TLS value, while the proximity of the keywords to one another denotes a stronger association between the respective research areas. Figure 4 displays five different colored clusters, and Figure 5 displays seven different colored clusters, each representing a corresponding research topic.

Figure 4 features larger clusters in orange, yellow, green, and blue, each representing different directions of the application for CT in the science of loess microstructure. For instance, the orange cluster primarily focuses on the quantitative analysis of pore changes in undisturbed and remolded soils, including keywords such as “pore distribution”, “pore morphology and size”, and “pore connectivity”; the yellow cluster is dedicated to the study of the stress paths of undisturbed loess under soaked loading conditions. Both the green and blue clusters are concerned with the study of structural strength and damage variables of loess; Figure 5 presents the main research directions similar to those in Figure 4.

3.3. Burst Analysis of Keyword

Keyword burstiness refers to the sudden increase or decrease in the frequency of a keyword within a certain time period, which typically signifies a shift in direction within a research field. A keyword with high burstiness can reflect a new perspective with significant influence within a certain time frame, thereby highlighting a stage of academic advancement. The higher the burstiness value, the greater the rate of change in the frequency of occurrence of the keyword during that time period. In simple terms, it means that scholars in a particular research field may not have paid special attention to a certain keyword, but this keyword is actually particularly important to the field. As research accumulates over time, it will suddenly emerge as a hot keyword in that research field at some point [102]. This paper utilizes CiteSpace for burst detection based on the burst detection algorithm to analyze the keywords of the literature from the two databases. After filtering and analyzing the keywords, the top 25 keywords with high burst frequencies were obtained, along with the corresponding burst strength and the start and end years of burstiness values, as shown in Figure 6 and Figure 7.

From this, it can be seen that before 2010, research perspectives were primarily focused on using CT scanning to obtain CT images of loess under conditions such as water immersion and collapse, triaxial shear, and lateral loading, and reflecting the evolutionary rules of loess microstructure and damage variables through CT values. After 2010, a large number of research results emerged that were based on continuous sectioning and micro-CT technology, establishing high-precision 3D microstructure models of loess and quantitatively characterizing parameters, such as particles and pores in 3D; these studies paid special attention to analyzing the connection between the collapsibility of loess and microstructure. These achievements mark that the study of loess microstructure entered a new stage.

4. The Application of CT in Loess Microstructure

As CT technology and equipment continue to evolve and their costs decrease, they have been widely applied in the field of loess microstructure research. Li et al. [19] were the first to propose a method for quantitatively evaluating the structural properties of loess using CT technology. They conducted initial scans on loess compacted to different degrees and analyzed the initial structural changes in compacted loess for both CT values and CT images, paving the way for the application of CT in the field of loess structural research. Shao et al. [103] proposed structural parameters for loess based on triaxial tests, offering a new quantitative approach to the study of the loess structure. Meanwhile, Wang et al. [24] combined SEM images with CT images and conducted a supervised classification for the quantitative analysis of loess microstructure, laying the groundwork for many subsequent scholars to use CT technology to quantitatively analyze the microstructural characteristics of loess (e.g., the size, shape, contact relationships, and arrangement of particles; the morphology, distribution, and connectivity of pores, etc.) and to explore the quantitative relationship between micro-parameters and macroscopic collapse.

4.1. Initial Stage: Exploration and Foundation

The concept of soil microstructure was first introduced by Terzaghi in 1925, and its introduction provided a new perspective and method for addressing the macroscopic physical properties of loess-like soil. Reviewing the development of loess microstructure research, it can be observed that significant advancements at various stages have invariably benefited from progress in observational techniques and experimental equipment, as well as the emergence of new methodologies.

In the early stages, due to the limitations of resolution as well as the constraints of scanning equipment and costs, CT technology was primarily used to explore the structural and damage evolution laws of loess, thereby reflecting its impact on the collapsibility of loess. For instance, in 2000, Pu et al. [22] conducted CT scans and image analysis on loess after loading and water immersion, arguing that the infiltration channels formed by loose capillary pores in the loess were the cause of loess collapse, providing a micro perspective for understanding the mechanism of loess collapse. In 2004, Lei et al. [21] performed CT scans on the original loess and, after water immersion collapse tests, used damage theory to analyze the alterations within the microfabric of the loess during the tests; they then studied the hardening and yield failure process of the original loess during the triaxial shear process through CT scanning and defined various damage variables, offering new insights into the application of damage theory in loess research [104]. In 2006, Fang et al. [105] utilized versatile geotechnical triaxial testing apparatus in conjunction with CT to investigate the intrinsic structural alterations within pristine Q₂ loess during the triaxial shear process, finding that the loess had clear initial damage and anisotropy and that CT values and variances could effectively describe the laws of damage evolution. In 2009, Zhu et al. [106] used CT-triple axial collapse tests to reveal the changes in the pore structure of the original collapsible loess during loading and collapse, determining its structural yield stress, proposing structural parameters based on CT values, and establishing equations for damage evolution. In 2010, Li et al. [23] studied the collapse characteristics of the original Q₃ loess through CT-triple axial tests and quantitatively analyzed the impact of structure on collapse using CT values.

4.2. The 3D Characterization of Loess Microstructure

With further advancements in technology, micro-CT with higher scanning precision began to be applied to the study of the microstructure of loess. Some scholars combined high-precision micrometer-level CT images with three-dimensional image reconstruction software to quantitatively analyze the microstructural characteristics of loess, such as particle size, particle sphericity, pore distribution, pore orientation, and pore connectivity, thereby analyzing the macroscopic disaster mechanisms of loess. For example, Wang et al. [107] used computed tomography and fractal theory to propose a quantitative method for studying the three-dimensional morphology of large pores in loess, finding that the fractal dimension reflects the irregularity of the pore contour and is inversely proportional to moisture content, density, and cohesion, and directly proportional to porosity, providing a new method for the study of soil microstructure. Liu et al. [108] conducted CT scanning on saturated fine-grained soil in Daya Bay, using Avizo 9.0 software for the quantification and characterization of the three-dimensional pore structure and analyzing its consolidation and creep mechanisms from a micro- and nano-scale perspective. Yan et al. [109], based on micro-CT scanning, established three-dimensional images, conducted quantitative analysis of the two-dimensional pore area of Malan loess and qualitatively characterized its three-dimensional structure.

Li et al. [110,111] conducted CT scans on loess and found that the large structural units of loess exhibited significant concentration in the vertical direction. By analyzing the geometric parameters of the loess pore structure, they considered loess to be a geological material with strong anisotropy. Wei et al. [112] used Avizo to establish a pore grid model of loess and paleosol, discovering that the content of clay particles determined their role in the soil skeleton, thereby affecting the pore structure. Through real-time loading–unloading CT triaxial tests, comparing the pore characteristics of original loess, compacted loess, and remolded loess, Zhang et al. [113] observed that the pristine loess exhibits a higher degree of interconnected porosity compared to its compacted counterpart. Furthermore, they noted that compacted loess and the original loess, despite possessing identical dry densities, exhibit distinct pore configurations. Weina Yuan et al. [25] explored the micron-scale three-dimensional microstructure characteristics of loess in the Jingyang area, Shaanxi Province, China. Using X-ray-computed tomography technology, combined with image segmentation methods, they quantitatively analyzed the size, shape, and arrangement of loess particles and pores. The study found that two-dimensional analysis results could not fully reflect the characteristics of the three-dimensional microstructure, which had a reference value for understanding the microstructure of loess and similar granular materials. Meng et al. [114] used high-precision CT to quantify the structural and pore distribution patterns of remolded loess made by compaction and tamping methods. Ning et al. [115] conducted a quantitative analysis of the microstructure, such as the pores and throats of Yan’an Q₂ loess, based on micro-CT. Most of the above studies are based on discussions of static parameter changes between different samples and cannot account for the impact of inter-sample discrepancies on the experimental outcomes. To address this, Yu et al. [116] developed a micro soil sample hydro-mechanical loading system that could perform CT scans on different conditions of the same soil sample; it is a method that can effectively reduce the impact of sample errors on the experiment.

Additionally, some researchers have integrated CT scanning with continuous sectioning techniques to investigate the microstructure of loess. For example, Deng et al. [117] employed continuous sectioning to study the distribution and connectivity of pore sizes in loess from various regions of the Loess Plateau, categorizing the microstructure into three types. Wei et al. [118] reconstructed the 3D microstructure of loess using an enhanced continuous sectioning method and quantitatively analyzed the 3D microstructural characteristics of particles and pores by examining statistical parameters, such as the equivalent diameter, sphericity, morphology, and orientation angle. In recent years, Yang et al. [119] have utilized different amounts of slag micro-powder to amend loess and industrial CT to perform micro-scale experiments on the amended loess, analyzing aspects such as pore distribution, shape, and size to understand the impact of slag micro-powder on the loess’s pore structure, offering insights into the mechanical property changes in loess. Zhang et al. [120] focused on a typical soil profile from the Loess Plateau, using CT scanning to analyze the 3D structure of the plow layer, plow pan, and heart soil layer, and explored how layered soil pore parameters vary with the soil layer structure and the characteristics of the interface layer’s pores. Feng et al. [121] selected the loess in the southern plateau of Jingyang as the research object, combining CT scanning with mercury intrusion porosimetry to analyze the loess sample’s structure, qualitatively analyze the variation in fine and micro pore structure characteristics with the buried depth of loess from 2D and 3D levels, and quantitatively analyze the pore structure based on statistical analysis parameters.

4.3. Relationship between Microstructure and Macroscopic Mechanical Behavior of Loess

As a unique geological material, the relationship between the microstructure and macroscopic mechanical behavior of loess has always been a hot topic in geotechnical engineering and geology. The microstructure characteristics of loess include the morphology, size, arrangement, and orientation of particles and pores, which have a significant influence on the macroscopic mechanical properties of loess, such as strength, deformation characteristics, permeability, and collapsibility.

Zhou et al. [122] analyzed the macroscopic mechanical properties and the evolution of microstructure in loess, derived a quantitative relationship between CT values and density, discovered that the end restraint effect of the sample was significant under isotropic consolidation conditions, and revealed the variation patterns of CT values under different strain conditions. Shao et al. [123] studied the initial structure and structural changes before and after the shearing of undisturbed and remolded saturated loess through CT experiments, establishing a connection between microstructure and macroscopic mechanical behavior. Yuan et al. [124] used X-ray CT and triaxial shear tests to study the impact of the three-dimensional particle structure of Q₂ loess from Jingyang, Shaanxi, on shear behavior. The results showed that the distribution of particle equivalent diameter and sphericity significantly affected shear strength, and the arrangement of particles had an important impact on the shear failure mode. Wei et al. [99,125] used CT scanning technology to establish three-dimensional reconstruction models of Malan loess in the Yan’an New Area during the collapse process under different states and quantitatively characterized the microstructural parameters, revealing the collapse mechanism of Malan loess in the Yan’an New Area through changes in microstructural parameters before and after collapse. Wang et al. [86] constructed a macropore structure model of remolded loess using X-ray CT technology, analyzed the impact of macropores on permeability, and improved image segmentation accuracy through a new workflow. Zheng et al. [98], based on micro-CT scanning, conducted undrained shear tests on undisturbed loess, undisturbed saturated loess, and remolded loess, respectively, and analyzed the characteristics of changes in pore structure before and after shearing, providing a basis for the study of the reduction in loess shear strength and the collapse mechanism. Fan et al. [58] summarized the current state of research on the micro-mechanism of loess collapse from three aspects: the composition, properties, and cementation mode of loess cementing substances, microstructural characteristics, and inter-particle forces, and preliminarily explored the collapse mechanism of Malan loess in the Yan’an New Area through case analysis. Wei et al. [126] conducted indoor collapse tests, CT scanning, and mercury intrusion tests to analyze and compare the variation patterns of pore distribution during the collapse process of Malan loess from different regions to explore the collapse mechanism of Malan loess. Zhou et al. [127] observed the spatial distribution and morphological evolution of cracks related to the shear failure surface through triaxial shear tests and micro-CT, revealing the development process of local shear failure in Malan loess and providing important insights into the multi-scale simulation of loess mechanical behavior. Li et al. [128] captured the seepage and storage state of water in the pores of loess based on CT technology, revealing the seepage evolution pattern of dynamic preferential flow. The study found that the loess had a preferential flow phenomenon in both saturated and unsaturated states, with uniformity and anisotropy.

Significant advancements have been achieved in the field of loess microstructure research through the application of Computerized Tomography (CT) technology, which greatly facilitates the quantitative analysis of loess particle and pore characteristics. Utilizing high-precision µ-CT technology combined with 3D reconstruction software, researchers can delve into the microstructural features of the loess microstructure, such as particle size, shape, contact relationships, arrangement, pore morphology, distribution, and connectivity. Such attributes substantially dictate the macroscopic mechanical response of loess, including collapsibility, strength, deformation characteristics, and permeability. Additionally, the combination of CT technology with other micro-analysis methods, such as SEM, mercury intrusion porosimetry, and serial sectioning techniques, provides new perspectives for multi-scale studies of loess microstructure. Future research should further optimize CT image processing techniques, establish more precise quantitative models that link the microstructure to macroscopic mechanical behavior, and consider the anisotropy and heterogeneity of loess, as well as the impact of environmental factors on loess microstructure and macroscopic mechanical behavior. Concurrently, conducting long-term monitoring studies on loess microstructure, in conjunction with numerical simulation techniques to predict the long-term mechanical behavior of loess under various conditions, can provide a more solid scientific foundation for geotechnical engineering and geological disaster prevention in loess regions.

5. Conclusions

As a type of soil widely distributed throughout the world, the microstructure characteristics of loess have an important influence on engineering characteristics and disaster mechanisms. Computed tomography (CT) technology, due to its non-destructive, quantitative, and high-resolution advantages, has become an important tool for the study of the microstructure of loess. This paper reviews the current status of the application of CT technology in the study of the microstructure of loess. It begins by introducing the development of CT technology, including the origins of X-ray CT, detection principles, equipment evolution, and image processing and analysis methods. Subsequently, bibliometric analysis methods reveal the annual publication trends of CT technology in the study of loess microstructure and perform co-word analysis and burst detection analysis to demonstrate research hotspots and cutting-edge directions. This paper also focuses on the application of CT technology in the 3D characterization of the loess microstructure and its connection to macroscopic mechanical behavior. Finally, it summarizes the shortcomings of CT technology in the study of loess microstructure and proposes prospects for future research directions.