Quantitative MRI of Diffuse Liver Diseases: Techniques and Tissue-Mimicking Phantoms

Abstract

Quantitative magnetic resonance imaging (MRI) techniques are emerging as non-invasive alternatives to biopsy for assessment of diffuse liver diseases of iron overload, steatosis and fibrosis. For testing and validating the accuracy of these techniques, phantoms are often used as stand-ins to human tissue to mimic diffuse liver pathologies. However, currently, there is no standardization in the preparation of MRI-based liver phantoms for mimicking iron overload, steatosis, fibrosis or a combination of these pathologies as various sizes and types of materials are used to mimic the same liver disease. Liver phantoms that mimic specific MR features of diffuse liver diseases observed in vivo are important for testing and calibrating new MRI techniques and for evaluating signal models to accurately quantify these features. In this study, we review the liver morphology associated with these diffuse diseases, discuss the quantitative MR techniques for assessing these liver pathologies, and comprehensively examine published liver phantom studies and discuss their benefits and limitations.

INTRODUCTION

Magnetic resonance imaging (MRI) has evolved as a noninvasive alternative to biopsy for the assessment of diffuse liver diseases, primarily hepatic iron overload, steatosis, and fibrosis [1–3]. Hepatic iron overload is a manifestation of hereditary hemochromatosis and transfusional hemosiderosis and is also common in some chronic hepatopathies [4]. Likewise, hepatic steatosis is the abnormal accumulation of fat in the liver and is a hallmark feature of non-alcoholic fatty liver disease (NAFLD), the most common diffuse liver disease that currently affects about 20%-30% of the U.S. population [5]. The co-occurrence of hepatic fat and iron overload, referred to as combined storage disease, is also being increasingly recognized [6–8]. Hepatic fibrosis is another common outcome of various liver pathologies, including hepatitis, NAFLD, and iron overload, and has a high prevalence of up to 25% in the general population [9–11]. The coexistence of fat, iron overload, and/or fibrosis with any pre-existing liver disease can accelerate disease progression. As early stages of these diseases are reversible conditions, techniques that can provide early diagnosis, allow disease monitoring, and guide therapeutic intervention are crucial for better prognosis.

Over the last two decades, several technical developments in quantitative MRI techniques have emerged that enable more precise and frequent monitoring of these diffuse liver pathologies to guide treatment and prevent disease progression and long-term complications. For testing, validating, and standardizing quantitative MRI methods across sites, vendors, and field strengths, phantoms are used as stand-ins to human tissue to mimic various liver pathologies. Quantitative liver phantoms also play an essential role for quality assurance in multi-center clinical trials and drug discovery studies that rely on quantitative imaging biomarkers as primary endpoints [12–14]. Organizations such as Radiological Society of North America (RSNA) Quantitative Imaging Biomarkers Alliance (QIBA) have laid out guidelines in bringing quantitative liver imaging to the clinical sector by collaborating with researchers and healthcare professionals to perform multi-site and multi-vendor validation studies on some promising MR biomarkers using phantoms [15–18]. However, currently, there is no standardization in the preparation of MRI phantoms for emulating diffuse liver diseases, as different studies utilize various types and sizes of iron and fat particles to mimic iron overload and steatosis respectively, different materials for mimicking fibrosis, and different matrices like water, agarose, or gels [19–25]. Further, with MRI, the co-occurrence of liver iron, fat, and fibrosis can confound or interfere with the quantification of each other. To address these challenges, novel MRI acquisition methods and multi-spectral signal models are being developed for accurate and simultaneous assessment of these diffuse liver diseases [26–29]. In addition, recent attempts include creating combination phantoms to mimic coexisting liver pathologies of iron overload, steatosis, and fibrosis observed in vivo for validation and quality assurance of MRI techniques for quantification of diffuse liver diseases [30]. However, simply combining different materials for each pathology to create phantoms mimicking concurrent diffuse liver diseases do not mimic the MRI signal characteristics seen in tissues [30].

The purpose of this study is, therefore, to provide a comprehensive review of the published MRI phantom studies on diffuse liver pathologies of iron overload, steatosis, and fibrosis by first describing the disease morphology and discussing the MRI techniques and biomarkers for each pathology and then examining the phantom materials chosen for mimicking these diffuse liver diseases. In this study, a systematic review is performed using PubMed as the primary database to identify and extract information from publications that constructed MR liver phantoms for mimicking iron overload, steatosis, fibrosis, and the combination of these pathologies. The key words used to search for these publications are phantom, liver, iron overload, steatosis, fibrosis, and MRI.

The following sections describe the desired characteristics and matrix materials of liver phantoms and briefly cover the in vivo morphology and quantitative MRI techniques for each pathology, followed by reviewing the phantom materials used to emulate each disease as well as elucidating their benefits and limitations.

LIVER PHANTOMS: CHARACTERISTICS AND MATRIX MATERIALS

The design requirements of any quantitative tissue-mimicking MRI phantom are typically: (1) characteristic MRI signal behavior and relaxation times that accurately emulate biomarkers of disease while considering field strength dependencies, (2) cover relevant clinical ranges of the disease, (3) multi-site, multi-vendor, and multi-platform reproducibility for standardization, (4) stability over long periods of time, and (5) ease of fabrication and handling [19,31–34]. These tissue-mimicking phantoms can be constructed using a variety of materials to achieve these characteristics. However, phantoms emulating diffuse liver diseases have particular requirements that may limit the materials chosen, and these specifications are described below in the appropriate phantom material sections. Nevertheless, it must be understood that tissue-mimicking phantoms cannot simulate every facet of a tissue, but the phantoms must mimic the MRI signal characteristics observed in vivo of these diffuse liver pathologies.

Matrix Materials

Matrix mediums are a necessary requirement in phantoms for stabilizing the contents used to mimic liver pathologies. The most widely used matrix mediums for making phantoms are water, agar, and agarose. Water is risk-free and easy to handle, but it has some disadvantages as a phantom medium. Water requires a 10-minute settling time to allow suspensions to stabilize and is more susceptible to vibrational effects that may cause motion artifacts [19]. Additionally, water has similar T1 and T2 relaxation times, and therefore, it does not mimic the behavior of human liver tissues in which T1 is much longer than T2 [35–37]. Alternatively, materials such as agarose, carrageenan, polyvinyl alcohol (PVA), polyacrylamide (PAA), and gelatin create a jelly-like medium that will hold the phantom contents in suspension, thereby reducing boundary artifacts and avoiding flow issues [19,38]. Both agarose and agar gel have T2 relaxation times that are similar to human tissue (~30-200 ms) depending on the chosen concentration [19,39]. However, the T1 relaxation times are not comparable to human tissues (~300-1500 ms) but can be adjusted with paramagnetic additives such as CuSO₄, NiCl₂, MnCl₂, or GdCl₃ to reduce T1 [19,39]. An alternative polysaccharide that is utilized in MRI phantoms is carrageenan gel, a more flexible and resistant substance that can be molded into unique shapes, but use of high concentrations can cause the T2 relaxation times to be longer than human tissues [38,40]. Like many polysaccharides, agar and agarose need thermal treatments (80-100°C) during the preparation process, so caution must be taken when using some phantom contents that have restricted temperature ranges [19]. Conversely, PVA-based phantoms benefit from greater longevity and structural stability than most matrix materials, but the need for freezing and thawing cycles to prepare the phantom may pose similar issues [38,40].

While there is no standard guideline to define an acceptable shelf life of a phantom, the phantom stability is greatly affected by the matrix material chosen. Some materials such as agarose and silicone are known to be mechanically stable over long periods of time, assuming no extreme fluctuations in temperature and humidity in storage, while other materials such as PAA are only stable for a few months. [38]. Further, gel phantoms are prone to microbial growth and often use sodium azide or sodium benzoate as preservatives to retard mold formation and increase shelf life [19,22,30,41]. To mimic native tissues, some studies also used homogenized animal liver tissues as a matrix medium instead; however, special handling and storage of these tissue phantoms may be a concern [3,42,43].

IRON OVERLOAD

Pathophysiology & Morphology

Hepatic iron overload is a histological hallmark of primary (hereditary) or secondary hemochromatosis [44,45]. Hereditary hemochromatosis is caused by a defect in the genes that control intestinal iron absorption from food, primarily the HFE gene, and affected patients absorb iron at 5–10 times the normal rate (up to 10mg/day) [4,46]. Secondary hemochromatosis is usually the result of another disease or condition that causes iron overload, e.g., severe chronic hemolysis requiring chronic blood transfusions. Because the body has no physiologic mechanism for clearing excess iron, the extra iron accumulates in organs, primarily the liver, leading to hepatic iron overload. As the liver is a central organ in regulating iron homeostasis, iron is also one of the most common parenchymal deposits in other diffuse liver diseases like NAFLD apart from hemochromatosis [47]. The degree of liver dysfunction is directly dependent on the amount of hepatic iron deposition and progressive iron accumulation eventually leads to hepatomegaly, fibrosis, and finally, cirrhosis and liver failure [48–52].

The liver stores excess iron in hepatocytes and reticuloendothelial Kupffer cells, as ferritin and hemosiderin molecules. Ferritin is the principal iron storage molecule found in cells and is readily available to meet physiologic needs. Ferritin molecules in clusters have a size of 50-70 nm and are not visible individually on Perls iron stain, however, they can be seen as a diffuse blue blush in the cytoplasm of hepatocytes on light microscopy [53–55]. If ferritin levels are persistently elevated, hemosiderin deposits can develop as conglomerates of clumped ferritin particles. Hemosiderin appears as granular, golden brown cytoplasmic deposits on hematoxylin and eosin (H&E) stain and as blue deposits on Perls iron stain as shown in Figure 1. During iron overload, aggregates of hemosiderin molecules are deposited in body tissues and range in size from 0.1 to 2 μm, eventually causing tissue damage and dysregulation of function [56]. Further, with an increase in iron overload severity, iron is deposited as clusters, and the deposition among hepatocytes are not uniform [56].

Figure 1.

Hemosiderin deposits shown in the liver biopsy samples using hematoxylin & eosin (H&E) stain (A) and Perls Prussian blue iron stain (B) at 20x in a severely iron overloaded patient with biopsy hepatic iron content of 16.5 mg Fe/g. Iron deposits are seen as brown granules in the cytoplasm of the Kupffer cells and hepatocytes on the H&E stain and as blue pigment on Perls iron stain.

The amount of iron in the liver, or hepatic iron content (HIC), is a necessary metric for detecting and staging iron overload as HIC significantly correlates with total body iron stores [57]. The current gold standard to quantify HIC directly is by atomic absorption spectrophotometry on liver biopsy specimens, which measures HIC in milligram of iron per gram of dry liver tissue weight (mg Fe/g dry tissue wt.) [58]. Alternatively, iron deposition in biopsy specimens can also be evaluated histologically on a semi-quantitative 5-point grading scale based on Prussian Blue staining of iron granules [4].

Quantitative MR Techniques

R2/R2* Relaxometry

Currently, MRI techniques based on transverse relaxometry (R2/R2*) are considered robust and clinically accepted methods for non-invasively assessing HIC [59]. Iron in the form of ferritin and hemosiderin has paramagnetic properties, causing faster signal decay and increasing the R2 and R2* relaxation rates. R2 can be measured using a series of single spin-echoes, each acquired after a separate excitation [60] or with a train of spin-echoes acquired after a single excitation using a Carr-Purcell-Meiboom-Gill sequence [61–63]. St. Pierre et al. developed a R2-HIC calibration method that is commercially available as FerriScan (Resonance Health, Ltd, Claremont, Australia) for HIC determination [60]. The R2-HIC relationship is however curvilinear which saturates at high HIC, hence making the method less sensitive to severe iron overload.

On the other hand, R2* is measured using a multi-echo gradient-echo (GRE) sequence and fitting a mono-exponential signal model [10,64,65]. The R2*-HIC relationship is linear for the entire pathophysiological range of HIC making R2* measurements desirable for clinical assessments [66,67]. However, for extreme cases of iron overload (HIC > 30 mg Fe/g dry tissue wt. at 1.5T or HIC > 15 mg Fe/g at 3T), the R2* quantification with GRE sequences is limited by the shortest possible echo time (TE) of about 1 ms. To overcome this issue, ultrashort echo-time (UTE) techniques have been developed to improve the dynamic range of R2* relaxometry and have demonstrated high R2* accuracy even for severe iron overload [68–70]. One major limitation with R2* techniques is that the presence of fat (steatosis) introduces oscillations in the signal decay and confounds R2* measurements. To overcome this, multi-spectral signal modeling techniques have been proposed for accounting for both the spectral complexity of fat and R2* decay. These techniques are discussed in more detail in the MR Techniques section under steatosis below.

Quantitative Susceptibility Mapping

Although relaxometry based methods are robust, they are confounded by other pathologies such as fibrosis, thus causing bias in HIC estimations [71,72]. Alternatively, quantitative susceptibility mapping (QSM) techniques can be used to estimate HIC by quantifying magnetic susceptibility, an intrinsic tissue property that is highly sensitive and linearly related to iron content, and is not affected by cellularity changes such as fibrosis [73]. Iron in the form of ferritin and hemosiderin induces a local susceptibility and affects the measured phase on GRE images. The local field map is then extracted from the phase data, and the susceptibility map is generated from the local field map by using a dipole inversion algorithm [74–77]. In contrast to the brain, QSM estimation in the abdomen must account for signals from multiple species (i.e., fat and water) and quantify a relatively wider range of iron contents (5-10 times greater iron deposition in liver iron overload than in brain). Recent studies have incorporated multi-spectral fat-water models to produce a fat-corrected field map estimation and used that in conjunction with QSM processing to generate susceptibility maps and have demonstrated a linear correlation between HIC and susceptibility values [78–82]. However, QSM is still limited to research due to both a lack of standardization of the technique and a not well-established relationship between susceptibility values and HIC [1]. Hence, it is not yet commercially available on MRI scanners for clinical HIC diagnosis [1].

Iron Overload Phantoms

The desired properties of an iron overload phantom are to maintain the characteristic relationship between MRI biomarkers (R2/R2*, susceptibility) and iron concentration, match the clinical ranges of MRI biomarkers in conditions of hepatic iron overload (R2*: ~40–1000 s⁻¹ at 1.5T, susceptibility: ~0.2–6 ppm) [78,80–83] and preserve the size and distribution of iron particles seen in liver tissues throughout the phantom. MRI studies that constructed phantoms to emulate hepatic iron overload using a variety of paramagnetic materials are summarized in Table 1. Of the studies reviewed, the most common materials used for making phantoms mimicking iron overload are iron nanoparticles. Iron oxide nanoparticles exhibit varying levels of paramagnetic and superparamagnetic behavior depending on the oxidation state of iron (ferrous vs. ferric) and particle size, in turn altering the R2* relaxation [84,85]. Ferrous iron, such as that of iron sulfate, can be added to emulate the R2* range [51-434 s⁻¹] of iron content in clinical hepatic iron overload [86]. On the other hand, ferric iron molecules exhibit stronger paramagnetic behavior and have greater susceptibility than ferrous iron [85]. A recent study by Alústiza et al. compared iron oxide nanoparticles with different oxidation states and found that the decrease in MRI signal for ferric iron (FeCl₃) was more linearly correlated with the increase in iron concentration than compared to ferrous iron ((NH₄)₂Fe(SO₄)₂), therefore more closely emulating the linear relationship between R2* and iron content observed in hepatic iron overload [25].

Table 1.

MRI iron phantom studies listing phantom materials used, MR techniques, and R2*/QSM ranges.

Citation PMID	Phantom Materials	MR Technique (Field Strength)	R2*(s−1)/QSM (ppm) Range
St. Pierre et al. 2005 15256427 [60]	MnCl2 (0.1-3.2 mM) Matrix: Water	Multi-echo SE	10-280 s−1 (R2)
Szurowska et al. 2010 20105330 [42]	FeCl3 Matrix: Pork liver	GRE SIR (1.5T)	n/a
Chandarana et al. 2012 22997371 [93]	Ferumoxytol (Feraheme®, 1/2000-1/500) Matrix: Deionized Water	Multi-echo GRE (1.5T)	25-150 s−1
Tsai et al. 2014 25120622 [220]	Fe(OH)3 (0.1-25 mg/mL) Matrix: Pork liver	Multi-echo GRE (1.5T)	80-400 s−1
Alústiza et al. 2015 25874207 [25]	FeCl3 (0.05-4 mg Fe/mL) Matrix: Water	GRE SIR (1.5T)	n/a
Yokoo et. al 2015 25996989 [96]	MnCl2 (0-25 mM) Matrix: Water	3D GRE (1.5T)	~1-1250 s−1
Krafft et al. 2016 26308155 [20]	1) BNF (0.4-220 μg/mL), 2) DSPIO (0.2-125 μg/mL) 3) MnCl2 (0-3.2 mM) Matrix: Agarose	Multi-echo GRE, FS GRE (1.5T, 3.0T)	1) 20-2700 s−1 2) 20-600 s−1 3) 3-250 s−1
Ibrahim et al. 2016 26861202 [86]	FeSO4 (0.5-4.5g) Matrix: Agarose	Multi-echo GRE (3.0T)	~51-435 s−1
Guimaraes et al. 2016 27190762 [221]	MION 48 (1:50 - 1:10000 of 1 mg Fe/mL) Matrix: Saline	FS TSE, FS CPMG (1.5T)	n/a
Brown et al. 2017 27435747 [91]	USPIO (Resovist®, 0–246.48 μg/mL) Matrix: CAG gel	Multi-echo GRE (3.0T)	22–2829 s−1
Krafft et al. 2017 28090666 [68]	BNF (0.4-220 μg/mL) Matrix - 2% Agar	Multi-echo UTE (1.5T, 3.0T)	20–2640 s−1
Hong et al. 2017 27495266 [21]	Ferumoxides (Feridex®, 2-45 mM) Matrix: 0.9% Agarose	Multi-echo 2D-UTE, 2D IR-UTE, 3D UTE, 3D IR-UTE (3.0T)	~0.1-12 ms−1
Lu et al. 2018 29314215 [89]	1) Gadopentetate Dimeglumine (Magnevist®,1.5-9 mg/mL) 2) Ferumoxides (Feridex®, 2-22mM) Matrix: 0.9% Agarose	Multi-echo 3D UTE (3.0T)	1) ~0.2-2.8 ppm 2) ~0.4-4.5 ms−1 ~0-52 ppm
Tipirneni-Sajja et al. 2019 30358001 [69]	BNF (0.5-220 μg/mL) Matrix - 2% Agar	Multi-echo UTE (1.5T)	~ 20-2500 s−1
Headley et al. 2020 33207043 [222]	FeCl3 (29.4-96.4 mg) Matrix: Water	Multi-echo GRE (1.5T, 3.0T)	103-420 s−1 (1.5T) 144-616 s−1 (3.0T)

Abbreviations: BNF, Bionized NanoFerrite; CAG, carrageenan-strengthened agarose/gadolinium gel; DSPIO, dextran-coated superparamagnetic iron oxide nanoparticles; CPMG: Carr-Purcell-Meiboom-Gill; FeCl₃, Iron(III) chloride; Fe(OH)₃, Iron (III)-hydroxide; FeSO₄, Iron sulphate; FS, fat suppression; GRE, gradient recalled echo; MnCl₂, Manganese chloride; MION 48, dextran-coated magnetic iron oxide nanoparticles; SE, spin echo; SIR, signal intensity ratio; TSE: turbo spin echo; USPIO, ultra-small particles of iron oxide; UTE, ultrashort echo time; IR-UTE: adiabatic inversion recovery prepared UTE.

Another component that impacts the efficacy of iron oxide nanoparticles is the organic coating over the iron core, impacting aqueous stability, conjugation potential, and reducing agglomeration and allowing uniform dispersion of iron particles in the medium [87,88]. For example, pure magnetite powder without a coat forms heterogenous clumps in phantoms [41]. Dextran is one of the most popular coatings used for iron nanoparticles as it is water-soluble and biodegradable [88]. BNF (Bionized NanoFerrite) particles are one example of dextran-coated iron nanoparticles and have been reported to produce a wide span of R2* values (20-2700 s⁻¹), thus covering the entire clinical range of R2* at 1.5T and 3T during iron overload [20]. Figure 2 shows representative MRI iron phantoms made using BNF particles in 2% agar, their corresponding R2* maps obtained using mono-exponential model and the excellent linear correlation obtained between R2* values and iron concentrations in these phantoms.

Figure 2.

Representative MRI iron phantoms mimicking liver iron overload. (A) MRI magnitude images of iron phantoms consisting of BNF iron nanoparticles of size 80 nm in 2% agar, (B) R2* maps calculated by fitting a mono-exponential model [10], and (C) linear regression plot of R2* (s⁻¹) vs iron concentrations (%) demonstrating an excellent linear correlation with R² = 0.9945.

Alternatively, SPIO (superparamagnetic iron oxide) nanoparticles, which have emerged as specialized MRI contrast agents, can also be used for mimicking iron overload. The dextran-coated SPIO contrast agent, ferumoxides (Feridex^®), is a common choice of iron material used in phantoms, yielding R2* and susceptibility values as high as 100-12000 s⁻¹ and 52 ppm respectively at 3.0T [21,22,89,90]. Some previous studies also used the carboxydextran-coated SPIO contrast agent, ferucarbotran (Resovist^®), for mimicking iron overload with R2* values of 22-2829 s⁻¹ at 3.0T [91,92]. Alternatively, ferumoxytol (Feraheme^®) can also be used for making iron overload mimicking phantoms by diluting with deionized water, with a reported R2* range of 25 – 150 s⁻¹; however, the R2* range can be increased with smaller dilutions [93].

Certain non-iron contrast agents such as extracellular gadolinium chelates (e.g., gadopentetate dimeglumine (Magnevist^®)) and manganese-based contrast agents (e.g., manganese chloride) that are used clinically, also have magnetic properties comparable to those of iron, thereby providing a similar linear relationship between contrast concentration and R2*/susceptibility for mimicking iron overload [89,94–96]. Both gadolinium and manganese are water soluble, allowing them to be suspended in solution [94]. However, compared to manganese-based contrast agents (1-1250 s⁻¹), gadolinium has been reported to provide higher R2* values at 1.5T (up to 3000 s⁻¹) [20,96,97].

One problem with the use of most iron particles is that they are produced on the scale of nanometers. However, in cases of iron overload, accumulated hemosiderin particles are on the scale of micrometers (0.1 – 2 μm [56]), a much larger size than iron particles available in commercial products. To mimic in vivo sizes of iron particles, Mobini et al. used insoluble magnetite powder with a mean diameter of 1.17 μm [0.5 – 5 μm] in their phantom design to emulate pathological conditions of liver iron overload [41]. However, the inhomogeneous suspension of magnetite resulted in microscopic iron clustering that altered relaxation times, hence leading to overestimated R2* at only mild iron contents [41]. In a recent study, Zhao et al. incorporated soluble iron microspheres with a diameter of 2.9 ± 0.14 μm, to mimic in vivo hemosiderin particles, and constructed phantoms that produced target R2* values of 100 – 500 s⁻¹ at 1.5T and 3T, which however did not cover the entire clinical spectrum of R2* ranges [30].

MRI phantoms mimicking iron overload are also available ready-made to be purchased from commercial vendors. Calimetrix provides twelve R2* phantoms in the range of 15-1000 s⁻¹ (R2* phantom, Model 300). These phantoms are designed to match clinical hemochromatosis and other iron overload conditions and specifically for testing and quality control of MR-based R2* quantification. CaliberMRI is another commercial vendor that provides a T2 phantom (ISMRM/NIST System Phantom, Model 130) utilizing MnCl₂ with T2 ranges of 5-550 ms (~2-200 s⁻¹ R2 measured at 3T).

STEATOSIS

Pathophysiology & Morphology

Hepatic steatosis, or fatty liver disease, is the abnormal accumulation of lipids within the cytoplasm of hepatocytes, predominantly in the form of triglycerides. Approximately, 25% of the world population is affected by steatosis [98] and this number is projected to increase further with the increasing epidemics of obesity, diabetes, and metabolic syndrome [99,100]. The liver plays a critical role in lipid metabolism and any disturbances in these metabolic processes can lead to hepatic steatosis, which is a hallmark of many diffuse liver diseases including viral hepatitis and alcoholic and nonalcoholic fatty liver disease [8]. NAFLD can progress to non-alcoholic steatohepatitis (NASH), which is characterized by inflammation, cellular degeneration, and fibrosis and can ultimately lead to cirrhosis [101]. Cirrhosis associated with NASH is estimated to be the leading etiology of liver transplantation in the coming decades [102].

Steatosis can be characterized into two major types: microvesicular and macrovesicular steatosis, depending on the relative size of lipid vacuoles and the displacement of nucleus seen on histology. Macrovesicular steatosis is the predominant form seen in diffuse liver diseases with fat droplet sizes ranging from >1 μm to the order of the sizes of hepatocytes (20-25 μm) [103,104]. Steatosis is typically graded semi-quantitatively on histology based on the visually estimated percentage of hepatocytes infiltrated with fat deposits – less than 5% (normal), 5% to 33% (mild), 34% to 66% (moderate) and greater than 66% (severe) as shown in Figure 3 [105]. However, the clinical histological grading estimates only the percentage of affected hepatocytes but does not reflect the absolute fat percentage in the liver.

Figure 3.

Histology images at 20x of liver biopsy samples stained with hematoxylin and eosin showing the fat droplet morphology and deposition in patients with macrovesicular steatosis. The steatosis grades of these liver biopsy samples based on the NASH CRN scoring system are (A) grade 0 (normal), (B) grade 1 (mild), (C) grade 2 (moderate) and (D) grade 3 (severe).

Quantitative MR Techniques

Magnetic Resonance Spectroscopy

Magnetic resonance spectroscopy (MRS) has been used as a noninvasive alternative to biopsy and is regarded as the most accurate technique to quantify fat [106,107]. MRS is based on the physical principal that MR-active hydrogen protons in specific molecules resonate at a unique Larmor frequency depending on the molecular structure of compounds. For example, fat and water molecules resonate at slightly different frequencies with a frequency difference of 3.4 ppm (224 Hz at 1.5T and 448 Hz at 3T) between the water (−OH) and main lipid peak (methylene, −CH₂) in liver in vivo [108,109]. Based on the priori known chemical shifts, water and fat peaks can be detected using MRS, and their proton densities (proportional to peak amplitudes) can be quantified to estimate fat fraction (FF) [110–112]. However, due to the small sampling volume (~0.2 - 2 cm³ [113]) and the requirement of offline analysis by an experienced user, MRS has limited use for clinical purposes [107]. Alternatively, over the last two decades, chemical shift-based MRI techniques have been developed that provide 2D cross-sectional images or 3D whole liver coverage.

Chemical Shift Based Fat-Water Imaging

Chemical shift-based fat-water imaging is based on the same physical principle as MRS. Due to the difference in the resonance frequencies, fat protons precess at a slower frequency than water protons and develop a phase lag over time compared to that of water protons. For example, the main lipid peak (methylene, −CH₂) and water (hydroxyl, −OH) protons are in-phase at 4.6 ms and out-of-phase at 2.3 ms at 1.5T. The most common chemical shift-based fat-water separation method is 2-point Dixon imaging, which collects images when fat and water signals are in-phase and out-of-phase, and thereby generates fat-only and water-only images from which FF is computed [114,115]. Although 2-point Dixon is a useful technique to assess steatosis [116–119], this method is affected by magnetic field inhomogeneities and T1 bias [120–122], and most importantly, this technique does not consider R2* decay and the spectral complexity of fat which can corrupt FF estimates [120,123]. T1 bias causes overestimation of FF and can be minimized by reducing the flip angle, correcting using variable flip angles, or in post processing [29,124,125]. However, post processing corrections are generally based upon the assumed T1 values of fat and water in the liver, so in vivo pathologies such as iron overload and fibrosis or phantoms that deviate from these T1 values can produce inaccurate FF results [124,125]. As the signal is collected over multiple echo times during which R2* decay occurs, not accounting for the signal decay can lead to underestimation of FF, especially in tissues with short T2* such as those containing excess iron [114]. Various Dixon techniques have been developed over the years that differ in the number of echoes (or points) that are sampled [126]. While traditional 2-point and 3-point Dixon methods are desirable for a shorter scan time, they typically provide only qualitative information about fat content [126]. On the other hand, using six or more echoes increases the scan time, but they allow for accurate modeling and provide quantitative FF results when fitted with multispectral fat-water R2* models described below. [126–128]. These chemical shift-based MRI techniques are well validated against MRS, showing an excellent agreement between MRS and MRI FF values [129–131]. Further, previous studies demonstrated that clinical MRI FF values can range up to ~50% in patients and showed a good correlation to severity of steatosis graded on histology [132–134].

Multi-spectral Fat-Water Models with R2* Correction

Over the last decade, several multi-spectral signal modeling techniques have been proposed for accounting for the spectral complexity of fat and R2* decay. These techniques use multi-echo GRE acquisitions and perform multi-spectral fitting for simultaneous quantification of R2* and FF and thus, can enable the diagnosis of both hepatic iron overload and steatosis. One popular technique is based on nonlinear least squares (NLSQ) fitting of the fat–water signal model either using magnitude or complex data [28]. These models require priori information about relative frequencies and amplitudes of the multiple lipid peaks. Further, most NLSQ models assume a common R2* rate for fat and water for reducing model complexity [28], known as the single R2* model. However, this assumption will be incorrect if the R2* of fat and water are different, especially in cases of iron overload [135]. A dual R2* model has been developed and was shown to improve the FF quantification accuracy as compared to the single R2* model in phantoms [135]. However, the dual R2* model can be unstable and complex due to the estimation of an additional parameter, and the single R2* model has been described to be more accurate in low SNR conditions for FF estimation [136]. Additionally, recent studies have reported that multipeak fat modeling with complex fitting and single R2* correction model performs better over a range of clinically relevant FF values [136,137].

Another multi-spectral signal modeling technique is based on autoregressive moving averaging (ARMA) that does not use any prior information of fat spectrum and determines FF and separate R2* values for water and lipid species [29,138,139]. A recent study demonstrated that both NLSQ and ARMA models accurately quantified R2* and FF values, but ARMA yielded inaccurate values at both high iron and fat conditions, whereas NLSQ overestimated R2* and FF values at high iron and no fat conditions; both models potentially failing due to rapid T2* decay at high iron overload conditions [29].

Steatosis Phantoms

The design considerations for steatosis phantoms are to emulate the MRI characteristic multi-peak fat spectra seen in adipose tissue, match clinical ranges of hepatic steatosis (~0-50% FF), and maintain the size and distribution of fat deposits seen in liver tissues homogeneously throughout the phantom. Table 2 details the lipid substitutes, FF ranges, and MR techniques used in steatosis phantom studies.

Table 2.

MRI fat phantom studies listing the phantom materials used, MR techniques, and fat fraction ranges.

Citation PMID	Phantom Materials	MR Technique (Field Strength)	FF (%) Range
Chang et al. 2006 16794122 [43]	Canola Oil Matrix: Calf liver	2-echo GRE, MRS (1.5T)	0-80%
Bernard et al. 2008 18064714 [143]	Soya oil Matrix: Carageenan	2-echo Dixon, chemical shift imaging, MRS (3.0T)	0-100%
Hamilton et al. 2011 21834002 [112]	Safflower oil (Microlipid®)	MRS (3.0T)	50%
Peng et al. 2011 21737754 [144]	Vegetable oil Matrix: Water	2-echo Dixon, chemical shift imaging, MRS (7.0T)	0-100%
Mashhood et al. 2013 23172799 [149]	Intralipid® fat emulsion Matrix: Water	2-, 4-, 6-echo Dixon (1.5T, 3.0T)	0-30%
Leporq et al. 2013 23588583 [145]	Olive oil Matrix: Agar	MRS, Multi-echo GRE (1.5T, 3T)	0-50%
Deng et al. 2014 24840769 [3]	Pork visceral fat Matrix: liver tissue	Multi-echo GRE (1.5T)	3.5-46.6%
Hernando et al. 2017 27080068 [13]	Peanut oil Matrix: 2% Agar	Multi-echo GRE (1.5T, 3.0T)	0-100%
Fukuzawa et al. 2017 28770394 [146] Hayashi et al. 2018 29425901 [147]	Vegetable oil (soybean & rapeseed) Matrix: 1% Agar	MRS, Multi-echo Dixon (3.0T)	0-50%
Pooler et al. 2018 29665193 [140]	Peanut oil Matrix: 2% Agar	MRS, Multi-echo GRE (1.5T)	0-100%
Pickhardt et al. 2018 30016142 [14]	Peanut oil Matrix: 2% Agar	Multi-echo GRE (1.5T, 3.0T)	0-100%
Kim et al. 2019 30430684 [148]	Lard Matrix: Agar	MRS, Multi-echo GRE (1.5T, 3.0T)	0-60%
Jaubert et al. 2020 31736146 [141]	Peanut oil Matrix: 2% Agar	6-echo Dixon (1.5T)	0-100%
Hu et al. 2021 33464181 [15]	Fat Fraction Phantom, Model 300 (Calimetrix)	Multi-echo GRE (1.5T, 3.0T)	0-100%
Schneider et al. 2021 33768291 [142]	Peanut oil Matrix: 2% Agar	Multi-echo GRE (3.0T)	0-100%
Zhao et al. 2021 34105167 [12]	Peanut oil Matrix: 2% Agar	Multi-echo GRE (1.5T, 3.0T)	0-100%

Abbreviations: GRE, gradient recalled echo; MRS, magnetic resonance spectroscopy, SS-EPI, single-shot echo planar imaging.

Peanut oil is the most common material used in fat phantom studies due to its availability, ease of preparation, and similar NMR spectrum as triglycerides in adipose tissue [12–14,140–142]. Other types of vegetable oils, such as canola or soybean, have also produced similar FF results to peanut oil despite slight differences in saturated and unsaturated fat content [43,112,143–147]. Animal products such as lard and pork fat are also used as lipid substitutes and are mixed with appropriate proportions of liver tissue to create a homogenate or heated until in a liquid form for making steatosis phantoms [3,148]. As seen on Table 2, mayonnaise is another animal product used in steatosis phantoms and has been shown to display similar MRI signal behavior as other lipid substitutes when used in combination phantoms [82]. Hybrid plant and animal products can also be used to create steatosis phantoms through mixing vegetable oils and liver tissues [43]. Further, clinical fat emulsion supplements such as Intralipid^® and Microlipid^™ can also be used for constructing steatosis phantoms [112,149]. Alternatively, commercial steatosis phantoms are available from Calimetrix (Fat Fraction Phantom, Model 300) covering FF ranges from 0-50% and 100% [15].

When emulating steatosis, emulsifying agents are typically needed to disperse oil droplets homogeneously throughout the sample. All emulsifiers have amphiphilic properties that interface between the oil and aqueous phases, but some emulsifiers are more soluble in oil than water and vice versa. In the case of phantoms, the most commonly used emulsifier is sodium dodecyl sulfate (SDS), an anionic surfactant [13,22,41,90]. SDS is an extremely hydrophilic emulsifier with a much higher affinity for aqueous media compared to oil-based media [150]. However, some researchers have opted for using oil-soluble emulsifiers such as Span 80 or lecithin [146,151–153]. This allows a lipophilic emulsifier to be added directly to oil during the phantom preparation to stabilize water-in-oil emulsions rather than dissolving a hydrophilic emulsifying agent in water to stabilize oil-in-water emulsions [152]. However, caution must be taken when using emulsifiers in phantom studies as some emulsifying agents may alter the T1 and T2 relaxation times of the phantom [154].

During the production of steatosis phantoms, it is important to note that the mixing temperature of fat and matrix emulsions and other environmental conditions can affect the quality of the final phantom [151]. Notably, caution must be taken to mix oil and polysaccharide emulsions such as agar/agarose in a limited temperature range of ~40-50°C as mixing them at higher temperatures can cause creaming (i.e., separation of oil and aqueous media) while lower temperatures can solidify polysaccharide matrix solutions, both cases producing inhomogeneous suspensions [151]. When scanning, changes in room temperature will greatly change the resonance frequency of water, but the fat resonance frequency will remain relatively unaffected [108]. In fact, the frequency difference between main fat peak and water at room temperature is 3.5 ppm whereas it is 3.4 ppm in in vivo liver imaging [108,109]. However, changing the temperature of the scanning room to mimic body temperature is not practical, so creation of steatosis phantoms that retain the frequency difference of fat and water as in vivo liver is therefore an unmet need. Nevertheless, maintaining the temperature of the phantom throughout the duration of the scan and adjusting the chemical shift accordingly can produce higher reproducibility and accuracy in FF quantification [108].

Another issue in constructing fat phantoms emulating hepatic steatosis is the discrepancies in sizes between fat globules in in vivo hepatocytes and steatosis phantoms. Hines et al. acquired photomicrographs of a fat-water-SPIO phantom at fat fractions of 10% and 50%, and found that average diameters of the fat droplets were 47 ± 19 μm and 234 ± 66 μm, respectively, which were much larger than the average diameter of the hepatocytes, 20-25 μm [22,155]. This dissimilarity is not only different from histologic conditions, but it may also contribute to faster R2* decay than in vivo samples [22,41]. Some recent studies have incorporated alternative mixing methods such as hand-held or high-pressure homogenization to reduce the sizes of fat droplets [30,153]. However, microscopy validation has not been performed in these studies to ensure that the diameter of the fat droplets match those of in vivo hepatocytes [30,153].

FIBROSIS

Pathophysiology & Morphology

In cases of persistent liver injury such as in alcoholic liver disease, viral hepatitis, iron overload, and steatosis, functional liver tissue is progressively replaced primarily by extracellular matrix (ECM) components in a process called fibrosis [156]. In fibrosis, abnormal amounts of scar tissue replace functional hepatic cells and disrupt internal organ structure that over time limits the blood flow as well as the synthetic capabilities of the liver. Collagen is the most abundant ECM protein, and as fibrosis progresses, the type and composition of collagen changes dramatically with loss and over expression [157,158]. Predominantly, nanofibers of type I collagen with diameters of 300–500 nm replace functional hepatic tissues in fibrosis [159,160]. Initially, fibrosis does not cause noticeable symptoms, but as it worsens, it can lead to portal hypertension and cirrhosis [161]. Cirrhosis is the most advanced stage of liver fibrosis and is evidenced by fibrotic bands, parenchymal nodules, and vascular distortion – all of which derail hepatic function leading to liver failure [162]. The severity of fibrosis is graded semi-quantitatively on histology using the NASH CRN scoring system as follows: 0 (no fibrosis), 1A (mild zone 3 perisinusoidal fibrosis), 1B (moderate zone 3 perisinusoidal fibrosis), 1C (portal/periportal fibrosis only), 2 (perisinusoidal and portal/periportal fibrosis), 3 (bridging fibrosis), and 4 (cirrhosis) [163]. Accordingly, Figure 4 shows representative histology images and fibrosis morphology in patients with different fibrosis grades.

Figure 4.

Histology images at 20x of liver biopsy samples showing the deposition of extracellular matrix (primarily collagen) with Mason’s trichrome stain in patients with different fibrosis grades. The fibrosis grades of these histology samples based on the NASH CRN scoring system are stage 1, perisinusoidal (A), stage 2, perisinusoidal and periportal (B), stage 3, bridging fibrosis (C), and stage 4, cirrhosis (D).

Quantitative MR Techniques

Elastography

One of the most widely validated MRI techniques used in staging liver fibrosis is magnetic resonance elastography (MRE) [164]. Elastography takes advantage of the relationship between tissue stiffness and the speed of mechanical wave propagation, with waves propagating faster in harder tissues compared to soft tissues, to evaluate collagenous stiffening due to hepatic fibrosis [165]. Specifically, MRE methods use a low frequency vibration source (40-80 Hz) to generate mechanical waves, followed by the utilization of phase contrast techniques with motion encoding gradients to capture microscopic displacements caused by wave propagation [166,167]. The resulting wave information is then processed to produce tissue stiffness maps using an inversion algorithm to recover mechanical parameters from the displacement data. [166,167]. MRE has been shown to have a high sensitivity and specificity for distinguishing among NAFLD, NASH, fibrosis, and cirrhosis [164,168] and can image longitudinal changes in stiffness in patients over time [169]. Previous studies demonstrated good correlation between MRE stiffness values with both ultrasound elastography (Fibroscan) [23,170,171] and biopsy fibrosis grades [168,172,173], which are often used for clinical diagnosis, thus, making MRE a promising non-invasive alternative for assessing hepatic fibrosis.

Although hepatic MRE is accurate, it can fail in some pathologies such as iron overload in which significant signal decay occurs when using gradient echo based MRE sequences [166,174]. The spatial resolution of MRE can be improved by increasing the shear wave frequency; however, high-frequency waves attenuate quickly, so there is a tradeoff between spatial resolution and propagation distance, thus restricting the depth that can be imaged [175]. Due to this, MRE techniques are limited in use in obese subjects and patients with ascites [176].

T1 Imaging

T1 imaging is arising as an alternative in staging fibrosis to overcome the limitations with MRE. In fibrotic conditions, widespread hepatocyte injury occurs, consequently inducing the recruitment of inflammatory cells. The progression of fibrosis thus leads to chronic inflammation and increases extracellular volume throughout the liver [177]. Because T1 relaxation is altered by the molecular environment of water molecules in tissue, the increased extracellular volume in fibrosis and cirrhosis also prolongs liver T1 values [178]. Therefore, T1 imaging could be used as a noninvasive MRI biomarker for evaluating fibrosis. In the literature, T1 mapping has been used in staging myocardial fibrosis [179] as well as in many independent studies for assessing hepatic fibrosis [178,180–183]. While T1 has had high diagnostic accuracy, there are some significant drawbacks to this approach. Notably, T1 values can be confounded by other liver pathologies such as inflammation, steatosis, iron overload and portal hypertension [184–187].

Diffusion

Diffusion-weighted imaging (DWI) is one of the first MR techniques used for assessing liver fibrosis. DWI assesses the degree of Brownian motion of water protons by using motion sensitizing gradients and produces image contrast based on the fact that freely moving protons lose signal faster than restricted molecules [188]. Because ECM remodeling leads to a much higher expression of collagen, the flow of water is restricted in fibrotic tissues thus reducing the apparent diffusion coefficient (ADC) values [189,190]. The normalized ADC values for healthy livers are typically ~1.50-1.85 (x10⁻³ mm²/s) while for fibrotic livers are ~1.30-1.45 (x10⁻³ mm²/s) [189,191–194]. However, these ADC measurements can vary between vendors and scanners as they are sensitive to b-values and image noise, thus the cutoff values for staging liver fibrosis are difficult to establish [195]. More importantly, ADC values are also affected by iron overload and steatosis, hence limiting their usage for assessing co-occurring diffuse liver diseases [93,196].

Fibrosis Phantoms

The design requirements of fibrosis phantoms are to maintain characteristic relationship between MRI biomarkers (stiffness, T1, and ADC) and fibrosis biopsy grades and cover a wide range of clinical values of MRI biomarkers seen in conditions of liver fibrosis, stiffness: ~2.5-8.0 kPa [168,172,173,197], T1: ~500-1000 ms [2,36], ADC: 1.3 – 1.85 (x10⁻³ mm²/s) [189,191–194]. A variety of phantom materials are used to emulate the correlation with the three MRI biomarkers used for the evaluation of liver fibrosis as detailed in Table 3. Most frequently, fibrosis phantoms are created with varying stiffnesses by changing the concentration of the medium. The differing Young’s moduli between each sample allows MRE techniques to quantify the degree of stiffness in the phantom, thus allowing the emulation of different fibrosis grades [23]. In vivo, collagen is the primary component that makes up fibrosis pathologies, so some phantom studies have incorporated differing types and concentrations of collagen into their design [24,82]. Apart from collagen, gelatins, hydrogels, and polymers such as B-gel (bovine gelatin), agar, tofu, grenetine gel, Phytagel, Zerdine, silicone, polyvinyl alcohol, and polyacrylamide can also be used in MRE studies for emulating fibrosis [23,26,198–205]. Elastography phantoms are also available from commercial vendors such as Computerized Imaging Reference Systems, Inc. (Zerdine shear wave liver fibrosis phantom), which are designed specifically for ultrasound elastography but have been used in liver MRE studies [201]. Alternatively, commercial MRE phantoms with stiffness values between 2–4 kPa are also available from Resoundant, an MRE manufacturer, for verification and validation of the MRE system. Figure 5 shows a representative set of MRE phantoms with increasing stiffness values made using different concentrations of Phytagel covering the range of stiffness values similar to those seen in liver fibrosis.

Table 3.

MRI fibrosis phantom studies listing the phantom materials used, MR techniques, and stiffness/T1/ADC ranges.

Citation PMID	Phantom Materials	MR Technique (Field Strength)	Stiffness(kPa)/T1(ms)/ ADC(mm2/s) Range
Girometti et al. 2007 17440695 [211]	1) Water 2) Acetone	SE-EPI DWI (1.5T)	1) 2.45 ± 0.05 (x10−3 mm2/s), 2) 4.21 ± 0.05 (x10−3 mm2/s) (ADC)
Yin et al. 2007 17654577 [198]	15% B-gel	GRE MRE (1.5T)	11.2 kPa (G)
Oudry et al. 2009 19856447 [23]	B-gel (5-17.5%) Copolymer-in-oil (3-8.5%)	GRE MRE (1.5T)	~1.0 – 8.0 kPa (G)
Cui et al. 2015 26393236 [212]	Ice water	IVIM DWI (1.5T, 3.0T)	1.13 ± 0.03 (x10−3 mm2/s), 1.21 ± 0.02 (x10−3 mm2/s)
Salameh et al. 2016 26915977 [202]	7% Polyvinyl alcohol gel	Ultra-low field MRE (6.5 mT)	2.3 ± 1.0 kPa (G’)
Morisaka et al. 2016 27662640 [26]	Polyacrylamide gel (6.2-7.7%)	SE-EPI, GRE MRE (3.0T)	2.4 - 4.2 kPa (S)
Kishimoto et al. 2017 28057657 [203]	Acrylamide gel	SE-EPI MRE (3.0T)	2.1 - 25 kPa (G’)
Solamen et al. 2018 29877194 [199]	1) 55-70% A341 Silicone soft gel 2) 70%,100% A341 Silicone soft gel 3) Soft Tofu and 0.75% Agar 4) Soft Tofu and 0.65% Agar	NLI-MRE (3.0T)	1) 2.5 - 4.6 kPa (G’) 2) 4 -15 kPa (G’) 3) ~4 kPa (G’) 4) 2.7 - 4.2 kPa (G’)
Tirkes et al. 2019 31165353 [208]	System Phantom Model 130 (High Precision Devices, Inc.)	VFA, MOLLI, IR-SNAPSHOT, SASHA (1.5T)	20 - 1500 ms (T1)
Cho et al. 2020 33110134 [2]	Gadoterate meglumine (0.1-3 mmol/L) Matrix: Water	SE-IR, VFA, MOLLI (1.5T)	83 - 1310 ms (T1)
Usumura et. al 2021 34019551 [204]	Polyacrylamide gel (12 wt % Acrylamide, 45% Glycerin)	SE-EPI MRE (3.0T)	4.8 - 5.01 kPa (G’)
Andoh et al. 2021 34037285 [201]	Zerdine® solid hydrogel (Computerized Imaging Reference Systems, Inc.)	SE MRE (1.5T, 3.0T)	1.7-14.9 kPa (G)
Meneses et al. 2021 34209547 [207]	Human serum albumin (0-20%)	T1 mapping (3.0T)	1800-4000 ms (T1)
Kishimoto et al. 2022 35061118 [205]	Polyacrylamide gel (13%)	SE-EPI MRE (0.3T)	3.1 kPa (G’)

Abbreviations: B-gel, bovine skin powder; DWI, diffusion-weighted imaging; GRE, gradient recalled echo; IR-SNAPSHOT, inversion recovery snapshot FLASH; IVIM, intravoxel incoherent motion; MRE, magnetic resonance elastography; MOLLI, modified look-locker inversion recovery; NLI, non-linear inversion; SASHA, saturation recovery single-shot acquisition; SE-EPI, spin-echo echo planar imaging; SE-IR, spin echo inversion recovery; VFA, variable flip angle. Note: Elastography results are published using different variables: S, stiffness; G, shear modulus; G’, storage modulus

Figure 5:

P-wave, stiffness maps, and shear moduli (G) obtained from four magnetic resonance elastography (MRE) phantoms made of Phytagel (1.25 – 1.875%) for emulating liver fibrosis [219]. The stiffness values obtained with MRE ranged from 1.83 – 9.84 kPa.

MRI fibrosis phantoms can also be created using T1 modulators to manipulate T1 times and emulate fibrosis conditions at different grades. Gadoterate meglumine (Dotarem^®) is an FDA approved paramagnetic contrast agent capable of shortening T1 relaxation times, hence making it suitable for constructing fibrosis phantoms that are assessed using T1 mapping [2,206]. In contrast, a recent study incorporated human serum albumin as a biomarker for fibrosis as it has similar relaxation times to human tissues and used it to modulate T1 times by varying its concentration for emulating fibrosis [207]. T1 phantoms are also commercially available from CaliberMRI (ISMRM/NIST System Phantom, Model 130) and Calimetrix (T1 Phantom, Model 300) [208,209]. Both phantoms use NiCl₂ as a T1 modulator, but the CaliberMRI System Phantom provides a wider coverage of T1 values of 20-1900 ms, whereas the Calimetrix T1 phantom provides a more limited range of 100-1200 ms; both phantoms’ T1 values reported are at 3.0 T.

DWI phantoms are produced from a variety of materials including aqueous solutions (polyvinylpyrrolidone, sucrose, agarose, etc.), polymers (liquid paraffin and alkanes), and pure water. DWI phantoms are also available from commercial vendors such as CaliberMRI (Diffusion Phantom Model 128) and by organizations such as the National Institute of Standards and Technology (NIST) [17,210]. For DWI phantoms mimicking hepatic fibrosis, aqueous solutions of water, acetone or ice water are the most common materials used [190,211,212]. However, phantoms using water and acetone produced higher ADC values than observed in healthy or fibrotic livers but in contrast, using ice water produced ADC values closer to liver tissue [190,211]. Diffusion coefficients have a dependency on field strength and temperature, hence the phantom temperature should be maintained for consistent results [17]. Consequently, DWI phantoms that encompass clinical ranges of fibrosis are still an unmet need in the literature.

COMBINATION PHANTOMS

Because of the common co-occurrence of iron overload, steatosis, and fibrosis, there have been efforts to combine these pathologies into a standardized set of phantoms, as documented in Table 4. Combination phantoms incorporate the same materials as a single pathology phantom. However, due to the confounding effects between iron, fat, and fibrosis on MRI signals, there have been attempts to find unique biomarkers for each pathology, namely R2* or susceptibility for iron, FF for fat, and stiffness or T1 for fibrosis. Despite this effort, there are still challenges concerning these confounding effects that must be overcome to create an ideal set of combination liver phantoms.

Table 4:

MRI iron-fat and iron-fat-fibrosis phantom studies listing the phantom materials, MR techniques, R2*, fat fraction and T1 ranges.

Condition Mimicked	Citation PMID	Phantom Materials	MR Technique (Field Strength)	R2*(s−1)/QSM (ppm) Range	FF (%) Range	T1 (ms) Range
Iron & Fat	Sharma et al. 2009 19243059 [213]	Iron: Ferumoxides (Feridex®, 0-0.5 mM) Fat: Vegetable oil Matrix: 2% Agar	MRS (1.5T)	11-133 s−1 (R2)	10, 30%	n/a
Iron & Fat	Hines et al. 2009, 19856457 [22]	Iron: Ferumoxides (Feridex®, 0-50 μg/mL) Fat: Peanut oil Matrix: 2% Agar	2-echo Dixon, Multi-echo GRE, MRS (1.5T)	~25-375 s−1	0-100%	n/a
Iron & Fat	Boll et al. 2010 20308498 [223]	Iron: Ferumoxides (Feridex®, 0.002-0.027 mL) Fat: Caprylic/capric acid (3–12 mL) Matrix: Polyacrylamide	2-echo Dixon (3.0T)	n/a	10–40%	n/a
Iron & Fat	Lee et al. 2011 21591008 [90]	1) Iron: Ferumoxides (Feridex®, 0, 11.2 μg/mL) Fat: Canola oil (0-50%) Matrix: 2% Agar 2) Iron: Feriumoxides (Feridex®, 0-14 μg/mL) Fat: Canola oil (10%) Matrix - 2% Agar	Multi-echo GRE, MRS (1.5T)	1) n/a 2) ~45–100 s−1	0–50%	n/a
Iron & Fat	Fukuzawa et al. 2017 28770394 [146]	Iron: SPIO (0-0.4 mM) Fat: Vegetable oil Matrix: 1% Agar	MRS, Multi-echo Dixon (3.0T)	33–217 s−1	0, 20%	n/a
Iron & Fat	Tipirneni-Sajja et al. 2019 30761652 [29]	Iron: BNF (0–60 μg/mL) Fat: Peanut oil Matrix: 2% Agar	Multi-echo GRE (1.5T, 3.0T)	0–800 s−1	0–40%	n/a
Iron & Fat	Wang et al. 2019 30859893 [153]	Iron: Iron dextran (0–30 μg/mL) Fat: Vegetable oil Matrix: Water	2-echo GRE (3.0T)	~0–900 s−1	0–80%	n/a
Iron & Fat	Mobini et al. 2019 31732894 [41]	Iron: Magnetite (0–723.6 μg Fe/mL) Fat: Peanut oil Matrix: 0.9% Agar	Multi-echo GRE (3.0T)	~45–1000 s−1	0–100%	n/a
Iron & Fat, Iron & Fibrosis	Li et al 2018 29566449 [82]	1) Iron: Gadolinium (0–10.0 mM) Fibrosis - Collagen (0–30%) Matrix – Water 2) Iron: Gadolinium (0–10.0 mM) Fat: Mayonnaise Matrix: Water	Multi-echo GRE (1.5T, 3.0T)	1) 0–150 s−1 0–3 ppm 2) 0–350 s−1 0–3 ppm	0–43.7%	n/a
Iron, Fat & Fibrosis	Zhao et al. 2021 32783200 [30]	1) Iron: MnCl2 (0.17– 1.17 mM) & Magnetite (1.03-34.53 μg/mL) Fat - Peanut oil Matrix - 2% Agar 2) Iron: Magnetite (1.03-53.65 μg/mL) Fat - Peanut oil Fibrosis: NiCl2 (0.28-2.10 mM) Matrix: 2% Agar	Multi-echo GRE, MRS (1.5T, 3.0T)	1) 100-500 s−1 2) 50-600 s−1	1) 5-30% 2) 0-40%	1) n/a 2) 500-1500 ms
Iron & Fat	Colgan et al. (2021) 33783066 [214]	Iron: Magnetite Fat: Peanut oil Matrix: 2% Agar	Multi-echo GRE (1.5T, 3.0T)	200-1000 s−1	0-30%	n/a

Abbreviations: BNF, Bionized NanoFerrite; GRE, gradient recalled echo; MnCl₂, manganese chloride; NiCl₂, nickel chloride; SPIO, superparamagnetic iron oxide nanoparticles.

Iron-Fat Phantoms

Most literature on combination phantoms features two coexisting pathologies: iron overload and steatosis. For iron overload, iron nanoparticles are most frequently used to emulate the pathology [29,146,153], but contrast agents such as ferumoxides (Feridex^®) [22,90,213] and magnetite microspheres [41,214] have been reported as well. Likewise, the most popular substitution for steatosis is peanut oil [29,22,30,41,214], but other vegetable oils such as canola and soybean oil [90,146,153,213] have also been used in iron-fat phantoms. Iron overload and steatosis combination phantoms, however, present additional complications compared to single-pathology phantoms. When placed in an aqueous environment, lipids will form large globules with signal-generating hydrophobic tails in the interior of the bilayer [22]. Conversely, some iron particles are only water-soluble and remain in aqueous solution [22]. Because of this compartmentalization, iron particles, especially of nanometer sizes, may preferentially affect the R2* of water molecules but not that of the R2* of large fat droplets, thereby causing inaccurate R2* and FF measurements, especially at high iron concentrations when a single R2* fitting model is used [22,135]. However, some recent studies used iron microparticles to produce single-R2* behavior in the presence of fat and water to emulate the in vivo conditions and yield accurate R2* and FF measurements [30,41]. Figure 6 shows a representative set of MRI iron-fat phantoms and their R2* and FF maps obtained using the NLSQ fat-water model with single R2* correction. As shown, both measured R2* and FF values showed an excellent linear correlation with iron concentrations and true FFs respectively, except for the highest iron concentration.

Figure 6.

Representative MRI iron-fat phantoms mimicking both liver iron overload and steatosis. BNF iron particles (size: 80 nm) and peanut oil are used to emulate iron overload and steatosis, respectively. MRI magnitude images (A), R2* and fat fraction (FF) maps calculated by fitting a non-linear least squares (NLSQ) multi-spectral fat-water model with R2* correction (B, C) [28], and scatter plots (with error bars) of R2* (s⁻¹) vs iron concentrations (D) and measured fat fraction (FF) vs true FF values (E) are shown for these iron-fat phantoms. Both measured R2* and FF values showed an excellent linear correlation with iron concentrations and true FFs respectively, except for the highest iron concentration.

Iron-Fat-Fibrosis Phantoms

To date, very few studies have integrated fibrosis into iron-fat combination phantom designs to emulate all three diffuse liver pathologies. Zhao et al. incorporated NiCl₂ as a T1 modulator for emulating fibrosis and used peanut oil and iron microspheres for mimicking steatosis and iron overload, respectively [30]. NiCl₂ is suitable for combination phantoms as it is less sensitive to causing R2/R2* changes in combination iron phantoms [215]. However, iron shortens T1 times, so iron-fibrosis combination phantoms using iron particles in tandem with NiCl₂ should correct or account for the decrease in T1 values due to iron, which in fact mimics the shortening of liver T1 in the presence of iron overload [30,216–218]. Further, in this study, the R2* and T1 values in phantoms did not show the similar field strength dependency of these parameters between 1.5T and 3T that is observed in liver in vivo [30]. To investigate the effectiveness of QSM compared to R2* relaxometry for quantifying iron in coexisting pathologies, Li et al. made a set of iron-fibrosis phantoms using gadolinium and collagen, and iron-fat phantoms using gadolinium and mayonnaise [82]. Although it is well documented that iron and fat both influence R2* decay, this study reported that collagen used in fibrosis phantoms also has a positive association with R2* decay [82] and thus, can confound R2*-based HIC estimations. In contrast, susceptibility measurements were not confounded by the presence of fat or fibrosis in these combination phantoms, demonstrating that QSM is a better biomarker for iron estimation in coexisting pathologies [82].

Outlook

The morphology of in vivo tissues is often overlooked when creating liver phantoms emulating iron overload, steatosis, and fibrosis pathologies. Commercial iron particles are only available on the nano scale, but micron-sized particles are needed to mimic the sizes of hemosiderin deposits. The sizes of fat droplets are currently unvalidated in phantoms studies and may be much larger than the size of fat globules in hepatocytes. Therefore, homogenization of fat droplets is needed to reduce their size and mimic in vivo hepatocyte fat storage. The inclusion of materials that simulate fibrosis is important due to its frequent incidence in iron overload and/or steatosis pathologies. Currently, most phantom studies only focus on mimicking iron overload, steatosis or both. However, there is little research on combining these pathologies with the presence of fibrosis. Likewise, most fibrosis phantoms concentrate primarily on stiffness without considering the coexisting pathologies such as iron overload or steatosis. Hence, there is a gap in our understanding of imaging the complex interplay between these diffuse liver diseases in a standardized phantom model using MRI. Further studies are needed in combining all three pathologies to produce a more precise set of phantoms that correspond to realistic disease progression in in vivo conditions and to investigate and validate MRI techniques for clinical use. Lastly, this study only focused on MRI phantoms mimicking diffuse liver pathologies of iron overload, steatosis, fibrosis and their combination, however there are other features of diffuse liver disease such as inflammation and ballooning as well as other hepatic pathologies (portal hypertension, focal lesions, etc.) and MR techniques such as dynamic contrast enhanced methods that are beyond the scope of this work.

Conclusion

Iron overload, steatosis, and combined overload are common manifestations of diffuse liver disease that may lead to further injury in the form of fibrosis. MRI has opened new avenues for noninvasive quantification of iron overload, steatosis and fibrosis as an alternative to liver biopsy. To standardize MRI techniques, liver phantoms that mimic these pathologies are developed using matrix materials such as water and agar. To simulate iron overload, iron particles are used as R2* modulators or as sources of magnetic susceptibility. Likewise, fatty oils or animal tissues are used to create steatosis phantoms using emulsifiers to ensure homogeneous dispersion. For emulating fibrosis, phantom materials that change stiffness values based on concentration or T1 modulators are used. However, the current phantom studies suffer from morphological discrepancies in phantom design when compared to in vivo sizes of these particles and also from confounding effects on MRI biomarkers in cases of co-occurring iron, fat, and fibrosis pathologies. Future studies should be aimed at selecting phantom materials that emulate the size and distribution of these diffuse liver pathologies in vivo and also consider the complex microenvironment of these materials and how they impact the MRI signal for creating realistic morphological quantitative liver MRI phantoms.