Lipid Microparticle‐Based Phantoms Modeling Hepatic Steatosis for the Validation of Quantitative Imaging Techniques

Abstract

Metabolic dysfunction‐associated steatotic liver disease (MASLD) typically presents as “macrovesicular steatosis”, where each hepatocyte contains a large fat vacuole (30‐50 µm), indicating a more indolent form. In about 20% of cases, “microvesicular steatosis” occurs, with smaller vacuoles (1‐15 µm) linked to steatohepatitis, cirrhosis progression, and increased risk of liver cancer. Emerging quantitative ultrasound (QUS) liver fat quantification (QUS‐LFQ) tools measure various acoustic properties, but few methods compare techniques and imaging modalities, and the impact of fat vacuole size remains unclear. This study introduces a methodology to create ultrasound (US) phantoms that replicate fat vesicle size in MASLD. While imaging phantoms validate quantitative tools, no model currently links QUS‐LFQ measurements to steatosis severity. Existing homogeneous phantoms assessing properties like attenuation, backscatter, and speed of sound overlook the microstructure of steatosis, despite the known effect of particle size on acoustic interactions. Here, agar‐based phantoms simulate fat accumulation in steatotic hepatocytes using stable peanut oil droplets as analogs for lipid vacuoles. Microscopy and sizing confirm stability at 4 °C, 23 °C, and 50 °C. Both microscopy and US imaging confirm uniform distribution, with QUS‐LFQ measurements reflecting fat content. These phantoms hold promise for validating quantitative imaging methods, particularly for US‐based MASLD screening tools.

This paper presents a novel method for creating ultrasound phantoms that accurately replicate fat vesicle sizes encountered in Metabolic Dysfunction‐Associated Steatotic Liver Disease (MASLD). This approach can improve the understanding of liver fat measurement by creating ultrasound phantoms that link different quantitative imaging techniques to disease states and severity, enhancing diagnosis, treatment, and management of conditions like metabolic syndrome and MASLD.

1. Introduction

MASLD (Metabolic dysfunction‐associated steatotic liver disease), previously known as non‐alcoholic fatty liver disease (NAFLD), is currently the leading cause of new chronic liver disease cases in the United States, surpassing those related to alcohol and hepatitis B and C‐related liver diseases. MASLD affects around 30% of the U.S. population and one billion people globally, mirroring the rise in obesity, in both developed and developing countries.^{[ 1} , ² , ³ , ⁴ , ^{5 ]} Despite its prevalence, MASLD often goes undiagnosed until late stage due to the lack of reliable clinical indicators and effective screening methods.

On grayscale B‐mode ultrasound (US) imaging, steatosis has characteristic findings. Healthy liver parenchyma demonstrates similar echogenicity to kidney. However, in cases of fatty liver, there is increased echogenicity, accentuated attenuation of the US beam, as well as a loss of anatomical differentiation, such as clarity of the diaphragm and vessel walls. The effects of liver fat on sound propagation, although complex, are well described.^{[ 2 ]} These effects alter the acoustic properties of the liver parenchyma, including the speed of sound (SoS), attenuation, and backscatter to varying extents, which in combination lead to degraded image quality, limited depth of liver penetration, and obscured anatomical structures, respectively.

The current gold standard for assessing MASLD and MASH (metabolic associated steatohepatitis) is by histologic analysis of tissue obtained by percutaneous liver biopsy, which evaluates grade of steatosis, ballooning degeneration, lobular inflammation, and fibrosis. However, performing biopsies on the 1.6 billion affected, or even just high‐risk individuals, let alone conducting longitudinal follow‐ups, is impractical. This highlights the urgent need for noninvasive biomarkers to diagnose and grade the severity of steatotic liver disease. Medical imaging already plays an important role in liver fat quantification (LFQ) with multiple US companies in different stages of developing and commercially releasing a variety of quantitative US (QUS) measures.^{[ 2} , ⁶ , ^{7 ]} However, these methods differ across manufacturers, each offering different tools for LFQ, the majority based on measuring SoS, the attenuation coefficient (AC), the backscatter coefficient (BSC), or a combination of these. This variation raises concerns about both intermodality differences (e.g., between US, MRI, and CT) and intramodality differences, particularly between these various QUS‐LFQ techniques. Many potential confounders exist and have not yet been well studied. Even when manufacturers aim to measure the same parameter, inter‐manufacturer variability due to differences in assumptions, measurement techniques, and modeling remain unstudied sources of bias, complicating the clinical reporting of these biomarkers, the establishment of cutoff values, and ultimately the translation of these techniques to widespread use.

Tissue‐mimicking materials, known as “phantoms”, play a crucial role in developing and testing quantitative imaging biomarkers.^{[ 8} , ⁹ , ¹⁰ , ¹¹ , ^{12 ]} Various materials and fabrication techniques have been used to create liver phantoms for MRI, CT, and US^{[ 13} , ¹⁴ , ¹⁵ , ¹⁶ , ^{17 ]} or multimodal imaging.^{[ 18} , ¹⁹ , ^{20 ]} While modifications can be made to adjust a phantom's acoustic properties to simulate conditions of fatty liver, many of these methods alter acoustic properties at the molecular rather than the structural level. Since the acoustic properties are also influenced by the size, number, and distribution of small scatterers (like lipid‐filled hepatocyte vacuoles), it remains essential to consider the microscopic structure of the phantom. Currently, the only known US phantoms that model hepatic steatosis in this way incorporate a mix of milk and cream to vary fat content.^{[ 21 ]} However, milk fat globules (1‐1.5 µm) do not adequately represent the 1–50 µm lipid vacuoles found in MASLD, limiting their ability to accurately mimic in vivo conditions. To overcome this limitation, we developed phantoms containing stable lipid microparticles that can be precisely adjusted to reflect the spectrum of liver disease. These new phantoms will also be compatible with MRI‐based Proton Density Fat Fraction (MRI‐PDFF) techniques, enabling direct comparison with the current noninvasive reference standard for quantifying liver fat.

An important, clinically relevant distinction between types of steatosis is the size of lipid particles. Steatotic liver disease is characterized by the number as well as the average size of lipid vacuoles seen histologically within hepatocytes (Table  1 ). The most common form, macrovesicular steatosis, involves a single large vacuole (30‐50 µm) per cell and is generally more indolent. In contrast, microvesicular steatosis, characterized by numerous small (<15 µm) vacuoles throughout the hepatocytes, is less common though more closely linked to inflammation and higher risk of progression to chronic liver disease and cirrhosis.^{[ 22} , ²³ , ^{24 ]} Clinically, many patients exhibit mixed patterns rather than purely one form or another. Importantly, the number (density) of vacuoles as well as their size have significant impacts on the acoustic properties of tissue, particularly backscatter. However, these differences in microparticle (vacuole) size and distribution have not been taken into account in any prior study, nor has their impact on the accuracy of these quantitative measures been accounted for. Interestingly, if these or other novel measures are found to be capable of differentiating microparticle size, then US LFQ could be leveraged to non‐invasively grade and characterize steatotic liver disease, which would be unique relative to methods currently used in MRI and CT.

Table 1.

Microstructural aspects of MASLD as seen histologically.

Macrovesicular steatosis	Microvesicular steatosis
‐Large clear space inside hepatocytes on histology slides ‐Single lipid vacuole (30‐50 µm) per cell ‐More common ‐More commonly quiescent ‐“Simple” or “benign” steatosis	‐Innumerable clear spaces spread throughout the hepatocytes ‐Tiny vacuoles (<15 µm) ‐Less common (20% of cases) ‐More rapid progression of liver disease ‐May indicate additional metabolic risk factors ‐Highly associated with superimposed inflammation

The long‐term purpose of our ongoing work is to develop a multi‐modality tissue‐mimicking imaging phantom that simulates the various degrees and disease states of hepatic steatosis, that is compatible with current and emerging US quantitative techniques as well as established methods in MRI and CT. We hypothesize that by analyzing certain acoustic properties, US could differentiate macrovesicular steatotic pattern, where vacuoles typically measure 30 to 50 µm, from purely microvesicular pattern, with vacuoles measuring 1 to 15 µm. Consequently, our aim is to study how these variations in particle size can be exploited through US imaging to both quantify and characterize the type of steatosis. In this report, we outline the methods used to formulate and characterize oil microparticles, as well as the corresponding phantom models of macro‐ and microvesicular steatosis, that will eventually allow us to test this hypothesis (Figure  1 ). We first show that the materials used to prepare the phantom have tissue‐mimicking acoustic characteristics. We also demonstrate that the oil microparticles have the relevant size and sufficient stability, suitable for the formation of phantoms exhibiting different microstructures that mimic both the macrovesicular and microvesicular steatosis patterns. Finally, microscopy analysis, US imaging, and QUS‐LFQ measurements confirm the homogeneity of the phantoms and accurately reflect their fat content.

Figure 1.

Illustration representing lipid microparticles embedded in US phantoms as surrogate for lipid vacuoles found in MASLD and its potential to impact quantitative US liver fat quantification (adapted with permission).^{[ 2 ]} Inserts highlight the macrovesicular pattern observed on histology slides and the incorporation of similar‐sized microparticles within our phantom. Phantoms designed to simulate either macrovesicular or microvesicular steatosis can be produced.

2. Results and Discussion

2.1. Speed of Sound (SoS), and Attenuation and Backscatter Coefficients (AC and BSC) of Non‐Fatty Liver Mimicking Background Phantom are in the Ranges of those of Normal Human Liver

The non‐fatty liver‐mimicking background phantom, represented by a phantom with 0% fat, requires acoustic properties similar to those of normal liver. To achieve this, we used previously cited materials in the appropriate proportions to create a liver‐like phantom matrix that provides the SoS, AC, and BSC of normal liver. We built on the established knowledge that SoS, AC, and BSC can be modulated with propylene glycol/glycerol, graphite, and glass beads, respectively,^{[ 9} , ²⁵ , ^{26 ]} to prepare agar‐based phantoms incorporating these materials to model healthy, fat‐free liver.^{[ 27} , ^{28 ]} Briefly, a 2% w/v molten agar solution containing propylene glycol (8% w/v), glycerol (10% w/v) and graphite (1‐2% v/v) was sonicated at 60 °C to eliminate bubbles and poured into cylindrical sample containers with two parallel acoustic transmission windows, referred as “hockey pucks”, already containing glass beads (0.3‐0.45% w/v) (Figure  2a,b). The corresponding hockey pucks fabricated for each formulation were immersed in a water tank at 23 °C. SoS and AC were measured using a narrowband substitution technique,^{[ 29 ]} and BSC, through a planar reflector‐based broad‐band pulse‐echo technique^{[ 30 ]} (Figure 2c,d). SoS ranged between 1590 m ⁻¹s and 1594 m ⁻¹s between 2.25 MHz and 10 MHz for the three samples (Table S1, Supporting Information). Minor decrease of propylene glycol/glycerol content is expected to lower SoS to the expected range of normal liver (≈1560 ms⁻¹).^{[ 28 ]} As presented in Figure 2c,d and Table S1 (Supporting Information), both AC and BSC values obtained (≈0.2 decibels per centimeter per megahertz (dB/cm/MHz) and 1.0 × 10⁻³, at 3 MHz) are close to those reported in the literature for liver (0.5 dB cm⁻¹ MHz⁻¹ and 5.0 × 10⁻⁴ cm⁻¹ sr⁻¹ respectively, at 3 MHz).^{[ 27 ]} As expected, higher graphite content (Figure 2c, green) resulted in higher AC. It should be noted that graphite primarily contributes to attenuation, though does result in some backscatter. Similar, glass beads contribute primarily to backscatter, however do result in some attenuation. Therefore, both components are necessary within the matrix to achieve the AC and BSC for a “normal” (no fat) situation. For future iterations, minor increase of the amount of graphite and decrease of the amount of glass beads should more closely match the desired acoustic properties of healthy liver.

Figure 2.

a) Schematic representation of background phantom formulation. b) Photograph of the phantom material in the hockey puck housing. c) Attenuation (mean and error bars representing the three standard deviations of five measurements). d) Backscatter coefficients obtained at 2–10 MHz at 23 °C.

2.2. Stable Oil Microparticles have Sizes Matching Hepatocyte Lipid Vacuoles

Peanut oil was used as a surrogate for liver fat, as it is commonly employed to mimic fatty tissues, particularly in MRI phantoms, due to its ability to replicate the MRI characteristics (similar T1 and T2 relaxation times) of human fatty tissues.^{[ 31} , ^{32 ]} Oil microparticles were prepared following a emulsification‐solvent evaporation technique. This method involves dissolving peanut oil in dichloromethane (DCM) and then mixing it with an aqueous polyvinyl alcohol (PVA) solution using homogenization (Figure  3a), a vortex mixer (Figure  4a), or sonication (Figure S7, Supporting Information). In this process, PVA acts as a surfactant, stabilizing the oil‐in‐DCM droplets. Once the droplets are formed, the DCM is evaporated, leaving behind stable oil particles. DCM is one of the most widely used solvents in microparticle and nanoparticle production via the emulsification‐solvent evaporation technique^{[ 33} , ^{34 ]} due to its favorable properties. Its high volatility facilitates rapid solvent removal, while its strong solubility for hydrophobic molecules and low interfacial tension reduce the energy barrier between the oil and aqueous phases, promoting the formation of uniform, stable droplets. Additionally, DCM's compatibility with stabilizers enhances emulsion stability. The density of DCM (1.33 g cm⁻³) also plays a critical role in peanut oil encapsulation. Since peanut oil has a lower density than water (0.91 g cm⁻³ at 25 °C), dissolving it in DCM prevents it from accumulating at the liquid‐air interface. This ensures proper encapsulation within stable droplets. Without DCM, peanut oil would likely remain at the air‐water interface, where interfacial tension may hinder the formation of stable, monodisperse oil particles. The resulting stable oil microparticles were then isolated and characterized by microscopy and a Multisizer (Beckman Coulter MS4). Isolated oil microparticles have diameters corresponding to the ranges found in micro‐ and macro‐steatosis respectively (Figures 3b,c and 4b,c).

Figure 3.

a) Schematic representation of the 5–20 µm oil microparticles formulation and isolation process (PVA: polyvinylalcohol, DCM: dichloromethane). b) Representative micrographs. c) Volume‐weighted size distributions (n = 3). Mean diameters ± standard deviations calculated using volume‐weighted and number‐weighted data are presented in Figure S1 (Supporting Information).

Figure 4.

a) Schematic representation of the 20–60 µm oil microparticles formulation and isolation process (PVA: polyvinylalcohol, DCM: dichloromethane). b) Representative micrographs. c) Volume‐weighted size distributions (n = 3). Mean diameters ± standard deviations calculated using volume‐weighted and number‐weighted data are presented in Figure S1 (Supporting Information).

2.3. Oil Encapsulation is Close to 100% for Both Particle Sizes

Encapsulation efficiency (EE) of peanut oil in microparticles as well as total oil content were quantified using Nuclear Magnetic Resonance (NMR) spectroscopy on lyophilized emulsion aliquots using tetramethylsilane (TMS) as internal standard (Figure S2a,b, Supporting Information). As expected, chemical shifts in the samples were identical as the one obtained from pure peanut oil (Figure S2c,d, Supporting Information). A calibration curve, established with known amounts of pure oil (0 – 100 mg), based on peak areas associated with the α‐methylene groups of triacylglycerols (doublet doublet signals between 4‐4.5 ppm), allowed for the calculation of the oil content available in both formulations. EE (n = 3) was found at 95.5 ± 6.0% (20‐60 µm) and 95.4 ± 4.9% (5‐20 µm). These deviations from 100% can be explained for both size ranges. First, it is important to note that no free oil was visible, indicating that the deviations from 100% were not due to particle instability. We hypothesize that small particles in the 20–60 µm range, with less dichloromethane in their core, were likely present in the less dense supernatant PVA phase rather than the dichloromethane infranatant phase. Alternatively, the small particles in the 5–20 µm range, which contained the least amount of peanut oil in their core, may not have been buoyant enough to be separated by centrifugation. Respective amount of encapsulated oil in suspension were 453 ± 70 mg mL⁻¹ and 543 ± 17 mg mL⁻¹. This method provides a precise measurement of the oil content in the microparticle solution, allowing for accurate production of phantoms with known oil quantities.

2.4. Microparticles are Stable for at Least 1 h at 50 °C, 1 Week at 23 °C, and 1 Month at 4 °C

Oil microparticles thermal stability at 50 °C was confirmed by microscopy and Multisizer after 1 h as particle size (5‐20 µm and 20–60 µm) remained unchanged, and no unencapsulated oil was observed. This process confirmed that droplets maintain their size and integrity when briefly heated during the preparation of all steatosis‐mimicking phantoms, as is necessary for when particles are poured into molten agar at 50 °C to prevent premature agar solidification from rapid cooling. It is important to note that the particles were only exposed to 50 °C for a few minutes during phantom preparation, thus evaluations at higher temperatures or for longer durations were not pursued.

Additionally, particle stability was assessed in the same manner after 1 week of incubation at 23 °C and 1 month incubation at 4 °C to demonstrate the stability of the microparticles during short‐term storage between formulation and phantom manufacturing. Once again, the measurements overlapped (Figure  5 ), indicating that the oil particles maintained their size and concentration, with no evidence of unencapsulated oil. Given the density of peanut oil (d = 0.91 g mL⁻¹ at 25 °C), any free oil resulting from particle destabilization would accumulate at the air‐water interface; however, this was never observed. Representative microscopy images as well as individual Multisizer traces are presented in the SI (Figures S3–S5, Supporting Information).

Figure 5.

Representative volume‐weighted size distributions obtained the day of the formulation and after 1 h incubation at 50 °C (a: 5–20 µm particles, b: 20–60 µm particles), 1 week of storage at 4 °C or 23 °C, or one month of storage at 4 °C (c: 5–20 µm particles, d: 20–60 µm particles).

2.5. Oil Microparticles are Stable and Homogeneously Embedded in Phantoms

2.5.1. Microscopy Validation Using Oil Microparticles Stained with Oil Red O

5–20 µm or 20–60 µm microparticles (10% w/v) and graphite (2% w/v) were mixed into a 2% w/v agar solution containing propylene glycol/glycerol (8/10% w/v) at 50 °C and their uniform distribution in the cooled phantom was assessed using optical microscopy. This assessment was conducted after Oil Red O staining, using bright field and fluorescence microscopy on 60 µm cryosection slides at different depths within the phantom. A characteristic orange‐red tint from the Oil Red O dye in brightfield as well as fluorescence signal (rhodamine filters) was only observed in the core of the particles (Figure  6 ), which is consistent with the presence and integrity of the stained microparticles in the phantom. Glass beads were excluded from the mixture due to their impact on slide quality, often damaging the sections produced during cryosection.

Figure 6.

Representative bright field (top), fluorescence microscopy (Rhodamine channel, middle) and merged (bottom) images showing the stable Oil red O‐stained lipid particles (a: 5–20 µm, b: 20–60 µm) embedded in the graphite‐containing agar matrix. Bright field image was recorded using a black and white camera.

2.6. Conventional US Imaging

Phantom blocks containing either no oil or 20% w/v oil with particles mimicking microvesicular (“micro”) or macrovesicular (“macro”) steatosis were immobilized in a secondary larger cylindrical container filled with agar gel (1% w/v) containing cornstarch (1% w/v) and acoustic absorber pads at the base to improve handling, stability, and reduce artifacts. Conventional ultrasound imaging was performed using a curvilinear 1–5 MHz transducer (5C1, Siemens Healthineers), commonly used for abdominal and liver imaging, to assess the impact of particle size on B‐mode imaging appearance (Figure  7 ). As expected, microvesicular‐sized particles increased attenuation, leading to a decreased signal in the deeper (lower) regions of the phantom. In contrast, macrovesicular‐sized particles increased echogenicity (brightness) and produced a more heterogeneous, granular speckle pattern—effects attributed to enhanced backscatter and a decreased speed of sound, respectively. Imaging these phantoms at a higher frequency using a 4–15 MHz linear transducer (15L4, Siemens Healthineers) confirmed a homogeneous distribution of scatterers (Figure S6, Supporting Information).

Figure 7.

a) Schematic representation of the preparation of MASLD phantoms mimicking microvesicular and macrovesicular steatosis. b) Schematic representation of phantoms immobilized within a larger cylindrical container filled with agar gel (1% w/v) containing cornstarch (1% w/v) implemented with acoustic absorber pads at the base and imaged using a 1–5 MHz transducer. c) Representative grayscale B‐mode images of phantoms without lipid microparticles (0%, left), with 20% w/v of microvesicular peanut microparticles (middle), and with 20% w/v of macrovesicular peanut microparticles (right).

2.7. High Correlation Between Phantom's Lipid Content, MRI, US, and CT Results

Phantom block material containing 5% or 20% w/w lipid from 5–20 µm oil particles was immobilized in a secondary larger cylindrical container as above. Quantitative results from MRI Proton Density Fat Fraction (PDFF) and US‐Derived Fat Fraction (UDFF, Siemens Healthineers), a model‐based estimate of lipid content based on AC and BSC, were compared to known lipid content phantoms. Measure of lipid content by UDFF was in agreement with theoretical content (5% and 18% fat calculated versus 5% and 20% theoretical values determined by ¹H NMR) (Figure  8 ). Note that increasing the fat content to 20% resulted in a corresponding increase in acoustic attenuation, as seen by the darker region in the far field of the phantom. In addition, quantitative measure of AC (dB cm⁻¹ MHz⁻¹), showed that attenuation measurements are achievable with a clinical scanner with LFQ tools (Figure 8‐UGAP). MRI images (3T Ingenia, Philips Healthcare) of the same phantom also demonstrated homogeneous appearance, indicating a uniform distribution of graphite, glass beads, and peanut oil microparticles. However, MR‐PDFF demonstrated comparable fat fraction between theoretical and calculated only for the phantoms bearing 5% fat (Figure 8). We hypothesize that while graphite is a well‐established acoustic attenuation modulator,^{[ 35} , ^{36 ]} its strong diamagnetic properties most likely affect fat quantification by MRI due to its impact on dephasing (T2* effect) which confound the calculation of fat fraction. CT images were obtained using a dual‐energy photon‐counting CT (pcCT) imaging system (NAEOTOM Alpha, Siemens Healthineers). Measurements of linear attenuation coefficient demonstrated the expected linear decrease in X‐ray attenuation as the fat content in the phantoms increased. Interestingly, the measured CT numbers for samples representing normal liver (no fat), and fatty livers with 5%, 15%, and 30% fat were highly correlated with PDFF at 5%, 15%, and 30% (Pearson correlation coefficient r = 0.9908, p (two‐tailed) = 0.0092). This correlation is based on a previously established linear relationship between Hounsfield unit (HU) and PDFF, validated in clinical cases, using the pcCT, the same scanner used for our phantom measurements: PDFF (%) = ‐0.58 x CT (HU) + 43.1″, from virtual monoenergetic images at 70 keV) (Figure  9 ).^{[ 37 ]}

Figure 8.

MR Proton Density Fat Fraction (PDFF), Ultrasound‐derived fat fraction (UDFF™) and attenuation coefficient with Ultrasound‐Guided Attenuation Parameter (UGAP™, GE HealthCare) obtained from phantoms with 5% (top) and 20% oil (bottom) immobilized in a secondary container filled with agar gel (1% w/v) containing cornstarch (1% w/v) implemented with acoustic absorber pads at the base.

Figure 9.

a) Photon‐counting CT of abdominal phantom loaded with vials containing phantoms matrix with different oil contents encapsulated in 5–20 µm microparticles. b) X‐ray Attenuation at 70 keV (HU) as a function of oil content.^{[ 37 ]}

2.8. Clinical Premises and Future Opportunities

Fatty liver disease is now the most common cause of chronic liver disease, leading to serious adverse outcomes including cirrhosis and elevated risk of primary liver cancer, yet current diagnostic methods are inadequate for widespread screening and longitudinal testing. Although US is cost‐effective and safe for detecting steatosis, traditional qualitative assessments are subjective and inaccurate. Newer quantitative tools have only been validated in small studies, with limited knowledge about their precision, reproducibility, and consistency across manufacturers. As liver fat reduction serves as a biomarker for treatment response, US could non‐invasively and cost‐effectively assess the effectiveness of therapies targeting metabolic syndrome^{[ 38} , ³⁹ , ⁴⁰ , ⁴¹ , ^{42 ]} (e.g., semaglutide, Novo Nordisk) or steatosis directly (e.g., resmetirom, Madrigal Pharmaceuticals), thereby reducing overall healthcare costs and environmental impact (compared to using such techniques as CT or MRI). However, further validation of US‐based tools is necessary, as differences between methods and manufacturers hinder clinical adoption. We believe our first‐in‐class phantoms are a key component in the ongoing effort to harmonize and standardize existing US imaging tools. In future studies, we plan to develop phantom sets that simulate the full range of steatosis grades, assess the effects of macrovesicular and microvesicular patterns on QUS‐LFQ measurements, and demonstrate the phantoms' utility in inter‐modality comparisons. Efforts will be made to isolate monodisperse microparticles to better control particle size in the phantom, either through tangential flow filtration or by modifying the emulsification parameters. In this regard, we have already successfully generated 0.2‐2 µm PVA‐stabilized oil microparticles using a probe sonicator, a more energetic emulsification method (Figure S7, Supporting Information). Access to this size range is highly relevant, as many patients display mixed steatosis patterns rather than a purely singular form, including the presence of submicron fat vacuoles. However, a more comprehensive study is needed for further validation. The ability to create phantoms with variable yet precise lipid particle size, concentration, and distribution may uncover imaging biomarkers that differentiate between macrovesicular and microvesicular steatosis, the former being associated with a more benign course and the latter linked to a higher risk of inflammation and fibrosis. If current or new US‐based LFQ methods can accurately assess particle size effects, MASLD disease states may be identified without the need for biopsy.

3. Conclusion

In summary, stable oil droplets of various sizes were formulated and embedded in a liver‐mimicking matrix to simulate the patterns of hepatic steatosis observed in MASLD. We verified phantom construction by microscopy and chemical analysis validating that our methods are robust and reproducible. Preliminary cross‐validation with clinical MRI PDFF, QUS‐LFQ, and spectral CT techniques shows promising results. This progress paves the way for further investigations across the full spectrum of hepatic steatosis disease states. This report has the potential to serve as the method followed by academic and industry partners to further develop novel US phantoms for direct mapping between QUS from different manufacturers, MRI, CT, and histopathology to advance US‐based liver fat quantification for clinical use. This advancement would significantly improve screening, diagnosis, patient management, and treatment response, particularly for individuals with metabolic syndrome and MASLD.

4. Experimental Section

Materials

Peanut oil, propylene glycol, and polyvinyl alcohol 13–23 kDa were purchased from Sigma Aldrich. Graphite, glycerol, dichloromethane, Oil Red O, and vortex were purchased from Fisher Scientific. Agarose was purchased from Research Products International. Glass beads with a diameter of 50 µm were purchased from Vaniman Manufacturing Inc. Germall Plus was purchased from Hznxolrc.

Formulation of Oil Microparticles (5‐20 µm) for the Histologic Microvesicular Steatosis Pattern

Oil microparticles, ranging from 5 to 20 µm in size, were produced by emulsifying dichloromethane (10 mL) containing 20 g of peanut oil in an aqueous solution of PVA (200 mL, 13–23 kDa, 2% w/v). Emulsification was performed using a T18 digital Ultra‐Turrax homogenizer equipped with a S18N (19G) dispersion element (IKA) at 10 000 rpm for 4 min. After evaporating the dichloromethane under mild stirring (100 rpm) at ambient pressure for 16 h, the stabilized 5–20 µm oil microparticles were collected through centrifugation at 1000 g for 15 min using a swinging buckets centrifuge (Thermo Scientific Sorvall Legend RT+) and stored at 4 °C (Figure 3a).

Formulation of the Microparticles (20‐60 µm) for the Histologic Macrovesicular Steatosis Pattern

Oil microparticles ranging from 20 to 60 µm were produced using a low‐energy emulsification method with a bench‐top vortex (Fisherbrand Analog Vortex Mixer). The biphasic reaction mixture, consisting of 2 g of peanut oil in 1 mL of dichloromethane and 20 mL of PVA (13‐23 kDa, 2% w/v), was vortexed for 2 minutes (speed 10). After mixing, the particles settled at the bottom of the flask by gravity. They were then separated from the 20 mL supernatant, and the dichloromethane was removed by evaporation under mild stirring (100 rpm) at ambient pressure for 6 h (Figure 4a).

Formulation of the Microparticles (0.2‐2 µm) for the Histologic Macrovesicular Steatosis Pattern

Oil microparticles ranging from 0.2 to 2 µm were produced by sonicating 1 g of peanut oil in 0.5 mL of dichloromethane in coexistence with 10 mL of PVA (13‐23 kDa, 2% w/v) using a probe sonicator (Branson digital sonifier SFX 550 equipped with a 0.3 cm tapered microtip) for 30 s (20% power). After mild stirring (100 rpm, 6 h) under ambient pressure to remove the dichloromethane, the particles were concentrated up to 2.5 mL emulsion by tangential flow filtration using a 750 kDa modified polyethersulfone hollow fiber filtration membrane (13 cm², Spectrum Labs) (Figure S7, Supporting Information).

Particle Sizing and Counting

5–20 µm and 20–60 µm microparticles sizing and counting were performed using a Multisizer 4 Coulter Counter system (Beckman Coulter Inc.) using the 100 µm or 400 µm aperture size, capable of analyzing particles with distribution from 3 to 60 µm or 17 to 240 µm, respectively. 0.2‐2 µm microparticles sizing was performed using Dynamic Light Scattering (DLS) measurements with a Zetasizer ZS nano‐sizing system (Malvern Nano ZS, Malvern Instruments Ltd.).

Stability Study at 50 °C, 23 °C, and 4 °C

1 mL of microparticle suspensions were set aside in Dram glass vials and either placed on a dry heat block set at 50 °C, on the lab bench to achieve 23 °C or in the fridge at 4 °C. Particle stability was assessed using the Multisizer and microscopy as described above after 1 h incubation at 50 °C, to show stability during phantom manufacturing, or 1 month incubation at 23 and 4 °C, to show stability of microparticles during short‐term storage between formulation and phantom manufacturing.

Particle Sizing and Counting by Microscopy Leveraging Custom‐Made ImageJ Script

We developed a script to size circular objects on microscope images acquired on our upright microscope (Zeiss Imager A1 m). The core analysis is based on the ellipse split plugin that approximates ellipses on binarized images (see https://imagej.net/plugins/ellipse‐split for details about the parameters). The images are processed as follows: 1) a pseudo flat‐field is applied to correct for uneven background illumination, 2) object brightness is corrected, 3) binarization is applied, 4) ellipse split plugin is used. Finally, the results are presented in a histogram showing the diameter of measured ellipses as well as an image marking all objects included in the analysis.

Encapsulation Efficiency (EE) and Determination of Oil Content per mL by 1H NMR

We assessed the encapsulation efficiency (EE) and determined the oil content per mL of the microparticle solution using ¹H NMR. Samples Preparation: 0.1 mL of the oil microparticle solution was weighted and subsequently lyophilized. Lyophilized product (mainly pure oil with traces of PVA) was dissolved in pure deuterated chloroform (CDCl₃) up to 0.6 mL before being introduced in an NMR tube (5 mm OD, Thin Wall Precision, 7″ L, Cole‐Parmer). Control samples were prepared by dissolving 10, 25, 50, or 100 mg of pure peanut oil in pure CDCl₃ (total volume in the tube = 0.6 mL). Internal Standard: Tetramethylsilane (TMS) was solubilized in pure CDCl₃ and introduced in a straight capillary tube (35 mg mL⁻¹, 60 µL). TMS tube was inserted inside both control samples and samples containing the lyophilized microparticles before each NMR measurement, to serve as an external standard for quantification (Figure S2, Supporting Information). ¹ H NMR Analysis and Oil Quantification: Automatic scans were performed using the Varian Unity Inova 400 MHz or Varian VNMRS 600 MHz NMR spectrometers. ¹H NMR spectra were analyzed to determine the relative peak areas for the oil and TMS using the signal associated with the α‐methylene groups of the triacylglycerol (“doublet doublet” between 4 and 4.5 ppm). Peak area corresponding to the oil in the microparticle sample was compared to the peak area of the oil in the control sample to calculate the concentration of oil in the microparticle sample. EE was calculated by comparing the amount of oil measured in the microparticle solution to the total amount of oil initially used for encapsulation.

Preparation and Characterization of Background Matrix that Simulates Normal Non Fatty Liver

“Healthy liver‐like” phantoms (0% fat) were prepared by incorporating propylene glycol (8% w/v)/glycerol (10% w/v) to modulate SoS, graphite (15 µm, 1–2% w/v) to modulate attenuation, and silica beads (50 µm, 0.3‐0.45% w/v) to modulate backscatter, in agar (2% w/v). Briefly, agar (6 g) propylene glycol (24 g, 23 mL), and glycerol (30 g, 23.8 mL) were solubilized in water (253 mL) at 90 °C. Once the solution cooled to 60 °C, graphite (15 µm, 3 g) was introduced and dispersed in the agar solution using a bath sonicator for several seconds. When the solution reached 50 °C, Germall Plus (1.5%w/v) was added. Germall Plus is a broad‐spectrum preservative containing Diazolidinyl Urea and Iodopropynyl Butylcarbamate, which prevent the growth of bacteria, mold, and yeast.^{[ 43} , ^{44 ]} Its inclusion is expected to extend the shelf life of the phantoms. The resulting mixture was then poured into cylindrical sample containers (1.0‐inch‐thick, 3.0‐inch inner diameter, and 3.5‐inch outer diameter) containing beads (Figure 2b). These containers, referred to as “hockey pucks”^{[ 45} , ^{46 ]} had two parallel transmission windows made of 25‐µm‐thick Saran Wrap (Dow Chemical, Midland, Michigan). The mixture was cooled while being continuously rotated until gelation was complete. Then, acoustic characterization was performed at 23 °C in a water tank emersion acoustic measurement system (Figure 2c,d; Table S1, Supporting Information). SoS and AC were measured by through transmission in a water tank using a narrowband substitution technique with matched transmission and receiving transducers.^{[ 29 ]} BSC was measured at a scattering angle of 180° at 23 °C through a planar perfect reflector‐based broad‐band technique^{[ 30 ]} at frequencies ranging from 2–10 MHz. Briefly, for measurement of BSC, a sample is placed at the focus of a single‐element transducer. A short US pulse is emitted and backscattered echoes detected by the same transducer. System‐dependent factors are removed by normalizing to a planar reflector. The process is repeated at several frequencies to cover a bandwidth between 2 and 10 MHz, the typical frequency range for adult abdominal US imaging. Assessment of bias and precision was performed using castor oil, saline, and glass‐bead scatterer reference phantoms.

Preparation and characterization of oil microparticles encapsulating phantoms using microscopy with Oil Red O stained cryosections of the phantoms: To maintain precise concentration of agar (2% w/v), propylene glycol (8% w/v), glycerol (10% w/v) and graphite (2% w/v), volumes occupied by all constituents including the oil microparticles, were subtracted to the target final prepared volume (50 mL). Here, 2.5 g oil (5% w/v) was contained in either 4 mL (5‐20 µm) or 5 mL (20‐60 µm) emulsion and was added at 50 °C in the agar solution containing propylene glycol, glycerol, and graphite, as prepared above. Finally, the resulting mixture was poured into a 50 mL conical tube and stored at 4 °C. Obtained phantoms were removed from the tube and a 1 × 1 × 1 cm cube cut in the center and embedded in OCT and flash frozen in liquid nitrogen. 60 µm slices were obtained using a cryostat (Leica CM1950) at – 20 °C. Obtained sections were stained with freshly prepared Oil Red O solution. Oil Red O was dissolved in isopropanol (0.5% w/v) and filtered through a 0.22 µm filter. A working solution of Oil Red O was prepared by mixing 3 parts of the filtered stock solution with 2 parts distilled water. Subsequent staining of the sections was performed by adding 0.1 mL of the Oil Red O solution onto the cryosections. After incubating for 15 minutes at room temperature in the dark, the sample was washed with 5–10 mL of water to remove excess dye. The sections were then imaged in both bright field and fluorescence modes.

Preparation and Characterization of Microvesicular and Macrovesicular Steatosis Mimicking Phantoms by B‐ Mode US Imaging

The protocol described above for the hockey puck was followed for a target volume of 300 mL using 60 g oil (20% w/v), contained in 69 mL and 134 mL cream for the microvesicular and macrovesicular steatosis mimicking phantoms respectively. The resulting mixture was poured into a plastic mold (55 × 90 × 40 mm, 250 mL internal volume) containing glass beads, (50 µm, 0.75 g, 0.3% w/v) and rotated for a few hours at room temperature. Obtained rectangular cuboids were immobilized in a secondary larger cylindrical container filled with agar gel (1% w/v) containing cornstarch (1% w/v) implemented with acoustic absorber pads at the base to improve handling and stability, and reduce artifacts, for use with clinical US scanners. The phantoms were stored at 4 °C until ultrasound imaging was performed, using B mode with grayscale images obtained with 1–5 MHz curvilinear and 4–15 MHz linear transducers (5C1; 15L4, Acuson Sequoia, Siemens Healthineers).

Preparation and Characterization of Microvesicular Steatosis Mimicking Phantoms Using QUS‐LFQ, PDFF, and CT

The protocol described above was also followed here for a target volume of 300 mL using 15 g oil (5% w/v) and 60 g oil (20% w/v), contained in 27 mL and 105 mL cream respectively. Finally, the resulting mixture was poured into a plastic mold (55 × 90 × 40 mm, 250 mL internal volume) containing glass beads, (50 µm, 0.75 g, 0.3% w/v) and rotated for a few hours at room temperature. Obtained rectangular cuboids were immobilized in a secondary larger cylindrical container filled with agar gel (1% w/v) containing cornstarch (1% w/v) implemented with acoustic absorber pads at the base to improve handling and stability, and reduce artifacts, for use with clinical US scanners. Lipid content was measured by two commercially available US‐based techniques, including a bi‐parametric measure of backscatter, US‐Derived Fat Fraction (UDFF, Sequoia scanner, Siemens Healthineers) and quantitative measure of AC (dB cm⁻¹ MHz⁻¹) by Atten Imaging (EPIQ Elite scanner, Philips Healthcare) and Ultrasound‐Guided Attenuation Parameter (UGAP, Logiq E10, GE HealthCare). Lipid content was also obtained from a clinically‐equivalent whole‐body MRI research system (3T Ingenia, Philips Healthcare) using a standard multichannel Torso XL array coil and integrated posterior coil and commercial Proton Density Fat Fraction (PDFF) pulse sequence (mDixon Quant) with PDFF maps reconstructed.

Four phantoms were prepared in 50 mL conical tubes, containing oil concentrations ranging from 0–30% w/v. The phantoms also contained 2% w/v agar, 10% w/v glycerol, 8% w/v propylene glycol, 1% w/v graphite, and 0.3% w/v 50 µm glass beads. Photon‐counting CT scans were performed using the NAEOTOM Alpha photon counting spectral CT system (Siemens Healthineers) with the vials containing matrix embedded with varying oil content mounted in a standard abdominal torso phantom.

Data Processing and Statistical Analysis

All graphs, standard deviations, and statistical analyses were generated using the built‐in tools in GraphPad Prism 10. Size Distribution: Student's t‐tests were conducted to compare three independent batches (macro and micro) based on their mean size (± standard deviations) measured in triplicates. Experimenters were blinded to all imaging studies. AC and BSC Measurements: Each AC measurement was performed five times, with the sample repositioned in the beam path for each trial. Standard deviations were calculated from these five measurements. In the figure provided, the error bars represent three times the standard deviation. BSCs were measured using a broadband technique. Briefly, a single ultrasound wave pulse provided information across all frequencies within the transducer's bandwidth. Three transducers with resonant frequencies of 2.25, 5, and 10 MHz were used, each yielding BSC values within its respective bandwidth. Each puck was scanned at 64 independent locations per transducer by systematically moving the transducer across the puck's cross‐sectional area. At each location, RF echo signals were collected, and a single estimate of the power spectrum (and the BSC) was derived. Confidence intervals (CI) were determined by analyzing the 64 measurements at each frequency within the bandwidth of each transducer. Clinical Imaging Data: For MRI, at least three circular regions of interest (ROI) were placed in different locations near the center of the phantom material, avoiding obvious artifacts, and the mean value and standard deviation were recorded. For US, 6–10 ROI were placed in the center of the phantom, avoiding obvious artifacts, and the median value was recorded. For CT, a circular ROI is placed on each sample, and the mean CT number and standard deviation are measured across five consecutive slices. Correlation analysis was determined using Pearson correlation coefficient on 2‐tailed test. Repeatability testing was not performed for this manuscript, however will be included in future follow up studies focused on the diagnostic performance of each imaging biomarker.

Conflict of Interest

The authors declare no conflict of interest.

Author Contributions

D.F. and C.d.G.L. conceived and designed the study. CdGL developed the formulations’ methodology. I.R.M., T.Y. and L.R. provided expert guidance and conducted the experiments. C.E., S.A., K.S., A.W., S.G., J.S. and M.Z.S. conducted the experiments. C.d.G.L., D.F., I.R.M., L.R. and T.Y. analyzed and interpreted the data. CdGL, D.F. and I.R.M. wrote the manuscript. All authors reviewed and approved the submitted version of the manuscript.

Supporting information

Supporting Information

Endsley C., Ali S., Salhadar K., et al. “Lipid Microparticle‐Based Phantoms Modeling Hepatic Steatosis for the Validation of Quantitative Imaging Techniques.” Small Methods 9, no. 8 (2025): 9, 2500043. 10.1002/smtd.202500043

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.