Quantitative magnetic resonance imaging for chronic liver disease

Abstract

Chronic liver disease (CLD) has rapidly increased in prevalence over the past two decades, resulting in significant morbidity and mortality worldwide. Historically, the clinical gold standard for diagnosis, assessment of severity, and longitudinal monitoring of CLD has been liver biopsy with histological analysis, but this approach has limitations that may make it suboptimal for clinical and research settings. Magnetic resonance (MR)-based biomarkers can overcome the limitations by allowing accurate, precise, and quantitative assessment of key components of CLD without the risk of invasive procedures. This review briefly describes the limitations associated with liver biopsy and the need for non-invasive biomarkers. It then discusses the current state-of-the-art for MRI-based biomarkers of liver iron, fat, and fibrosis, and inflammation.

Introduction

Chronic liver disease (CLD) causes significant morbidity and mortality, with downstream societal costs. The prevalence of CLD has rapidly increased over the past two decades worldwide, placing CLD among the main causes of premature mortality.^1,2 The most common etiologies of CLD are viral hepatitis and non-alcoholic fatty liver disease (NAFLD).^2,3

CLD is characterized by distinctive histological abnormalities, which include substance deposition (e.g. fat, iron), inflammation, hepatocellular injury, fibrosis, and ultimately vascular and architectural remodeling.⁴ Currently, the clinical standard for diagnosing the presence and assessing the severity of CLD is liver biopsy with histology analysis. Liver biopsy suffers from diagnostic limitations and is risky, which makes it less than ideal for screening and monitoring. Hence, validation of accurate and precise non-invasive methods to assess CLD is an area of active research. Leading candidates for this purpose include serum or circulating biomarkers, clinical decision rules, and imaging methods. The imaging methods are further defined by modality: ultrasound, CT, and MRI.^5–8

This article focuses on MRI-based biomarkers. It begins with a discussion biopsy and the need for non-invasive assessment of CLD, and it then reviews the current state-of-the-art for MRI-based biomarkers of liver iron, fat, fibrosis, and inflammation.

Current role and limitations of biopsy: argument for non-invasive biomarkers

In the setting of CLD, liver biopsy plays three roles: diagnosis, assessment of disease severity, and longitudinal monitoring. Although considered the reference standard for each of these three roles, biopsy is prone to sampling-associated error and high inter-reader variability.⁴ A study that compared simultaneous intraindividual paired percutaneous liver biopsies for the assessment of NAFLD found that steatosis grades were different in 22% of patients and differed by at least one fibrosis stage in 41% of patients.⁹ Additional analysis showed that if only one of the biopsies was considered for diagnosis, steatohepatitis would have been missed in 24% of patients. Multiple factors contribute to liver biopsy variability, including disease spatial heterogeneity, small and inconsistent tissue sample size, and subjective interpretation with modest intra-reader agreement.^10,11 The semiquantitative nature of histological scoring is an additional limitation. For example, liver steatosis grading is performed by visual estimation of the proportion of hepatocytes containing fat (no steatosis,<5%; Grade 1, 5%–33%; Grade 2, 33–66%; Grade 3 > 66%).¹² These broad brackets of severity complicate statistical analyses in research and limit detectability of small longitudinal changes in liver fat content both clinically and for research. Similarly, semiquantitative scores are used for staging liver fibrosis, which maxes out at cirrhosis (Stage four fibrosis in most scoring systems), effectively grouping the large swath of patients with cirrhosis into a single category. Patients with cirrhosis have a broad biological range of disease, ranging from asymptomatic to severe morbidity and in some cases, deadly complications. To capture this biological range, a quantitative test with a large and continuous dynamic range would be preferable.

Recognizing these limitations, several authors have attempted to automate biopsy analysis using quantitative techniques and deep learning algorithms.^13–15 Further research is needed to determine the clinical advantages of using these methods for biopsy analysis; however, the analysis of a single biopsy, no matter how sophisticated, cannot overcome the limitations of spatial variability and complication risk, major impediments for longitudinal or repeated monitoring. In one large series of 2740 percutaneous liver biopsies, approximately 2.3% resulted in complications, which ranged from mild (e.g. transient pain) to severe (e.g. bleeding, punctured gallbladder, pneumothorax).¹⁶ Importantly, of the total number of planned biopsies for that study, 429 procedures were withheld due to low platelet count as a contraindication to the biopsy procedure. Considering that low platelet count is relatively common in patients with advanced CLD, a significant proportion of patients are not suitable for biopsy evaluation.

MR-based biomarkers have been widely adopted to overcome the limitations associated with liver biopsy.¹⁷ Due to their non-invasive nature, these methods are safer, while assessing the entire liver which obviate sampling problems. Further, MRI-based biomarkers yield quantitative measurements that have been proven to be accurate and precise for the diagnosis, assessment of severity, and longitudinal monitoring of CLD.

Liver iron

Background

Although iron is a vital micronutrient playing a pivotal role in the oxygen transport system of hemoglobin, excessive iron accumulation in the liver leads to oxidative stress, mitochondrial dysfunction and DNA damage, all of which result in hepatocyte injury.¹⁷ Several hepatic and extrahepatic chronic conditions may cause liver iron accumulation – including hereditary hemochromatosis, hematological diseases (e.g. thalassemia, sickle cell) and frequent blood transfusions, and primary liver diseases – through a variety of mechanisms as reviewed elsewhere.^18–21 Assessment of liver iron is useful to inform the need for chelation therapies and to monitor treatment response. To date, liver biopsy remains the reference standard for liver iron quantification, which is most often performed semiquantitatively, using a four-point grading scale.²¹ Quantitative measurements can be performed using biochemical techniques allowing for objective measurements of liver iron concentration (LIC); determined as the amount of iron in μmol or grams per gram of dry weight liver tissue. In normal individuals, LIC ranges from 3.6 to 36 μmol/g (0.2 to 2 mg g⁻¹) of dry weight. Values higher than 36 up to 150 μmol/g (8.3 mg g⁻¹) are considered mild overload, 150 to 300 μmol/g (16.7 mg g⁻¹) moderate overload, and >300 μmol/g severe iron overload.^18,22

MRI-based liver iron quantification-Basic concepts

MRI-based methods for liver iron quantification exploit the effect of iron on MRI signal. Iron, a ferromagnetic substance, alters the local magnetic field, accelerates transverse relaxation, and shortens T2 and T2* relaxation time constants. These continuous variables are measured in units of time (milliseconds) and can be converted into relaxivity rates (R2 = 1/T2, R2*=1/T2*, both reported as 1/s). Conversion from relaxation time constants to relaxivity rates is useful as rates are directly related to LIC over a large biologically relevant range. The relationship between R2 and LIC is curvilinear while the relationship between R2* and LIC is linear over a wide biological range (Figure 1). Although iron also accelerates longitudinal relaxation (i.e. it shortens T1 time constants and increases R1 relaxivity rates), this effect is weaker and not commonly applied for iron quantification.

Figure 1.

Conversion from relaxivity to LIC over a large biological range. The relationship between R2 and LIC is curvilinear while the relationship between R2* and LIC is mostly linear. Higher iron is associated with higher relaxivity rates.

Different approaches to estimate LIC using MRI have been proposed, all leveraging the inverse relationship between iron content and signal intensity on T2 or T2*-weighted images. One of the first successful approaches was pioneered by Gandon et al²³. In this method, signal intensity in liver and in a reference tissue with no iron (paraspinal muscle) are measured on multiple acquisitions with increasing TEs using the body coil. Yet, this approach has limitations including the need for multiple acquisitions, the assumption that the reference tissue has constant signal intensity (which may not be true), and a tendency to overestimate LIC.

Quantitative approaches that do not require a reference tissue for comparison were subsequently developed.^24,25 The so-called R2 relaxometry techniques use spin-echo-based acquisitions with a fixed relaxation time (TR) and increasing TEs to estimate R2 relaxivity values. Pioneered by St. Pierre et al, the leading R2 relaxivity method uses a biexponential model and an external reference phantom to estimate R2 and a non-linear regression algorithm to link R2 with LIC.²⁴ The algorithm was validated using LIC measured in contemporaneously obtained liver tissue samples^24,25 and subsequently commercialized (FerriScan^®) ²⁶ ). Disadvantages of the commercial method are its prolonged acquisition time, requirement for an external phantom, and the additional time and cost associated with offline analysis.

Recently, R2* relaxometry has become commonplace in clinical practice with availability of the technology on most MR scanners.²² The concept is similar to R2 relaxometry methods, relying on MRI signal decay at increasing TEs, but using a GRE sequence. GRE sequences are more sensitive to the presence of iron since signal decay is caused by both T2 and T2* effects. R2* methods have practical advantages: they are fast (can be acquired in a single breath-hold), do not require offline analysis, and, if designed properly, allow for the simultaneous quantification of liver fat (as discussed later). Further, by fitting the data at the voxel level using appropriately spaced echo times, it is possible to produce parametric maps (R2* maps) that are easily interpreted (Figure 2). A few different approaches for measuring R2* have been proposed.²² To date, these do not perfectly agree mainly because of differences in acquisition parameters and signal modeling.

Figure 2.

Complex-based R2* maps at 3T. Three different patients with mild, moderate and severe liver iron overload

Clinical and research implementation

The clinical and research implementation of MRI-based liver iron quantification techniques is in evolution, with no single MRI-based method universally accepted. All the above-described proposed methods have advantages and disadvantages. Depending on availability and if applied and interpreted with care, each could be utilized to diagnose patients with iron overload, assess the amount of iron overload, identify patients suitable for clinical trials, inform treatment decisions, and monitor patients longitudinally.

While all three methods might be suitable depending on availability, R2 as measured by Ferriscan^® is generally considered the gold standard, and a calibration equation converting R2 to LIC is available in the literature.²⁵ There is not yet a universally established calibration formula to convert R2* to LIC, with different diagnostic cutoffs published so far. These inconsistencies have challenged the acceptance of R2* methods in clinical care and clinical trials.

Fortunately, recent advances in the field, including the development of more accurate and less-error-prone complex-data based techniques,²⁷ suggest that in the next few years a standardized method for measuring R2* and a universal conversion for R2* to LIC will be established. We anticipate that those developments will usher widespread acceptance of R2* methods for clinical care and research.

Pitfalls and future directions

The high sensitivity to susceptibility effects of GRE sequences is an advantage (e.g. more sensitive to the presence of iron), but also a limitation. It may cause inaccurate results in the presence of other contributors to magnetic field inhomogeneity, such as air or metal, and when liver iron is markedly elevated. Extreme levels of iron may even make R2* estimation impossible due to substantial signal loss earlier than the first signal echo, particularly at 3T.²⁸ Defining the iron levels beyond which R2* estimation becomes unreliable is an area of active investigation. Further, emerging data suggest that the presence of liver fat might confound R2* measurements,²⁹ especially in the relatively low iron and high fat range. Two main mechanisms affect R2* estimations in the presence of liver fat: signal intensity oscillation due to fat-water interference and signal intensity decay due to microscopic field inhomogeneities caused by fat droplets. The former mechanism can be addressed through mathematical modeling,²⁷ while the latter is still to be solved. This correlation between liver fat content and R2* is relevant as patients with NAFLD may be erroneously classified as having excess liver iron.

Liver fat

Background

Liver fat accumulation occurs most commonly due to alterations in fat and insulin metabolism, leading to an abnormal buildup of triglycerides within hepatocytes.³⁰ Accumulation of fat within hepatocytes can also occur secondary to cellular injury. Fatty liver is defined when hepatic steatosis (i.e. accumulation of fat within hepatocytes) affects ≥5% hepatocytes.¹² In this section, we will focus on NAFLD, in which fat accumulates in the absence of significant alcohol ingestion. NAFLD has a variable disease course, from non- or slowly progressive isolated steatosis (non-alcoholic fatty liver, NAFL) to its more rapidly progressive form (non-alcoholic steatohepatitis, NASH).³⁰ While the exact relationship between the amount of liver fat and progressive forms of NAFLD is not established, the quantification of liver fat is recognized as an important clinical biomarker to assess disease status and as a marker of response to antisteatogenic drugs.³¹

MR-based liver fat quantification -Basic concepts

The key concept of MR-based methods is that liver fat can be quantified by decomposing the MR signal into fat and water components. Historically, MR spectroscopy (MRS), a non-anatomical method of measuring MR signal in a prescribed volume was considered the gold standard due to its accuracy in quantifying lipid relative to water in biological tissues. Although MRS- and MRI-based techniques share some physics concepts, adequate MRS data can be technically challenging to acquire, may require an expert spectroscopist for analysis and interpretation, and may also suffer from sampling variability.^32,33 By comparison, MRI-based methods are easier to implement and interpret and allow whole-liver coverage; hence, these methods are more commonly used in clinical practice and research. Signal fat fraction, that is, the MR signal attributed to fat, can be measured using fat-suppressed techniques or chemical shift-based techniques. The former separates the MR signal by subtracting a pair of magnitude images acquired with and without fat saturation. As complete and homogeneous fat signal saturation is virtually impossible with in-vivo imaging, this technique is unreliable and rarely used for fat quantification. The latter, chemical shift-based techniques, is the scope of this review and separate the water and fat signal components by acquiring GRE images at appropriately spaced TEs for this purpose. Measurements can be performed using only magnitude data (magnitude-based approach) or both magnitude and phase information (complex-based approach). Advanced variants of these methods address factors that confound the fat signal, including the spectral complexity of fat, T1 bias, T2* signal decay, noise bias, and phase errors such as those caused by eddy currents.³³ When these confounding factors are addressed, the proton density fat fraction (PDFF) is measured.

PDFF is an inherent tissue property and reflects the proportion of mobile protons attributable to fat over the total proton density in a given tissue. MRI-PDFF is independent of field strength and scanner platform, and accurately correlates with histology determined liver triglyceride concentration.¹⁷ Pixel-based measurements can be made to produce parametric PDFF maps that are easy to interpret while displaying the spatial distribution of liver fat (Figures 3 and 4).

Figure 3.

Magnitude-based PDFF maps. Three different patients with mild, moderate and severe liver steatosis.

Figure 4.

46-year-old female, magnitude-based PDFF maps. Segmental distribution of liver fat is displayed. Segment I has lower fat fraction compared to other liver segments.

Clinical and research implementation

Commercially available chemical-shift-based methods can provide whole liver coverage in a single breath-hold and can generate PDFF maps of the entire liver. These methods can be performed before or after administration of contrast agents, which do not impact the results. Since the methods measure and correct for R2* effects, they also generate R2* maps, permitting simultaneous assessment of liver iron.³³ Images are analyzed by placing regions of interest (ROIs) in representative portions of the liver and recording the mean PDFF and R2* values from the ROIs.

Although PDFF correlates closely to triglycerides concentration, these are not equivalent, as a small proportion of triglycerides in biologic tissues is invisible to MRI.³⁴ Similarly, although histology determined steatosis and PDFF are expressed as percentages, these refer to different measurements (the former reflects the percentage of cells containing intracellular fat droplets, while the latter refers to the fat fraction of a given volume). Hence, the relationships between PDFF and steatosis grades are non-linear. Different PDFF cut-offs have been proposed to classify steatosis grades,^35–37 with high-specificity cutoffs of about ≥5.2%, ≥16.3%, and ≥21.7% corresponding to steatosis grades of ≥1, ≥2, and 3, respectively.³⁸ Studies have also investigated the accuracy and reproducibility of MRI-PDFF in different settings and populations.^37–43 Two metanalysis, showed high accuracy, reproducibility, and repeatability of MRI-PDFF for the assessment of steatosis across different field strengths, vendors, and reconstruction methods.^38,42

Due to its non-invasive nature, accuracy, and high reproducibility, MRI-PDFF has emerged as the preferred method for non-invasive liver fat quantification in clinical and research settings¹⁷ and is now used as an endpoint in antisteatotic phase two clinical trials.^31–43 A ≥ 30% relative reduction in PDFF has been proposed as non-invasive indicator of treatment response in NASH clinical trials.⁴⁴ Recent studies suggest that higher PDFF is a risk factor for fibrosis progression in untreated patients and that PDFF is more sensitive than liver biopsy in detecting changes in liver fat content in clinical trials.^43,44

Pitfalls and future directions

Some limitations of MRI-PDFF merit mention. At the low fat-fraction range, liver PDFF estimation can be inaccurate due to the confounding effect of background signal noise. In patients with severe liver iron overload, PDFF estimation may be impossible because rapid signal loss from dephasing obscures fat-water signal oscillation. The level of R2* beyond which PDFF estimation becomes unreliable is not yet known, however. To date, the commercially available sequences have low spatial resolution and are prone to imaging artifacts, which limit their ability to detect or characterize fat in small liver lesions. Emerging advancements include faster image acquisition, free-breathing protocols to improve patient comfort, and deep-learning (DL)- based automated analysis.⁴⁵

Liver fibrosis

Background

Liver fibrosis is a response to repetitive cellular injury. Hepatocyte damage and associated inflammatory response lead to the activation of stellate cells, proliferation of fibroblasts/myofibroblasts, and excessive extracellular matrix collagen deposition. The diagnosis and assessment of hepatic fibrosis is classically performed on histological analysis. The severity of fibrosis is scored as stages, typically five in most histological scoring systems. The stage of fibrosis is an important prognostic marker. It demonstrates a strong positive correlation to all-cause mortality, liver-related mortality, decompensation of liver disease and liver-disease complications.^46,47

Given the limitations of liver biopsy previously outlined, considerable research has been directed at developing reliable non-invasive biomarkers for assessing liver fibrosis, with MRI-based biomarkers emerging as practical tools for this purpose.

The presence of hepatic fibrosis generally causes little anatomic change in the liver until late in the disease. Therefore, conventional anatomic imaging with MRI or other modalities has low sensitivity for detecting fibrosis and cannot be used to reliably assess severity.⁴⁸ Investigators have focused on exploring multiple MRI-based quantitative and semiquantitative biomarkers. Studies have demonstrated that all of these tissue biomarkers have some relationship with liver fibrosis. Recently, DL- and radiomics-based methods have attempted to provide models to best predict the presence of fibrosis with promising results.^49,50 However, among these technologies, magnetic resonance elastography (MRE) has emerged over the last decade a leading non-invasive and quantitative method for detecting and staging liver fibrosis.⁴⁸

MR elastography–Basic concepts

Tissue stiffness is a qualitative term used to describe the ability of tissues to resist deformation from external or internal forces. The term “stiffness” is often used interchangeably with the more formal terms, shear elasticity and shear modulus.⁵¹ MRE measures stiffness of tissues with the results expressed as the magnitude of the complex shear modulus in units of kilopascals (kPa).

A flat plastic drum-like device is fastened against the right lower anterior chest wall over liver using an elastic strap. This device applies low-amplitude vibrations (typically at 60 Hz), which generate propagating shear waves with amplitudes in the range of tens of microns.⁵² A modified phase-contrast pulse sequence with cyclic motion encoding gradients is used to image the pattern of shear wave propagation in the liver. The MRI scanner automatically processes the acquired wave images to generate “elastograms” (Figure 5), which depict the spatial distribution of stiffness in each slice. The system also calculates a “confidence map”, which excludes the pixels with unreliable stiffness values.

Figure 5.

Contrast-enhanced portal venous phase images and correlating color coded elastogram of five different subjects with METAVIR fibrosis stages F0 to F4 confirmed on histopathological analysis.

Clinical and research implementation

MRE requires approximately one minute of acquisition time (between one and four breath-holds of about 15 s) and can be included in a standard MRI exam of the liver. Typically four slices are obtained through the widest portion of the liver. MRE exams are analyzed by drawing regions of interest (ROIs) over the liver on elastograms and recording mean stiffness values. A standardized approach for manual ROI placement is described in the “consensus profile” for “MR Elastography of the Liver”, published by the RSNA Quantitative Biomarkers Alliance (QIBA) in 2018.⁵³ In the research setting, some investigators have proposed automated methods for analysis.⁵⁴

MRE is now considered the most accurate non-invasive technique for detecting and staging liver fibrosis.⁵⁵ Using histological analysis as gold standard, MRE-stiffness cut-offs of 2.7–2.9 kPa, 3.3–3.5 kPa, and 3.8–3.9 kPa have been proposed for detecting any fibrosis, fibrosis stage ≥2, and fibrosis stage ≥3, respectively.^56–58 The exact cut-offs, however, depend on the underlying etiology of CLD and the prevalence and burden of fibrosis in the study population.⁵¹ A number of systematic reviews have demonstrated the validity of MRE in classifying the stage of fibrosis.^58–61 In a systematic review of 12 studies, the area under the receiver operating characteristic curve (AUROC) of MRE was 0.84 for diagnosing stage ≥1, 0.88 for diagnosing stage ≥2, 0.93 for diagnosing stage ≥3 and 0.92 for diagnosing cirrhosis, with sensitivities of 80–98%, specificities of 90–100% and accuracies of 89–99%, depending on the fibrosis classification of interest.⁵⁹

An important metric for any quantitative imaging technology is test-retest repeatability, which expresses how well the technology can detect a biological change. The QIBA MRE Consensus Profile, based on an update of a meta-analysis of test-retest studies, has stated that a change of at least 19% in liver stiffness, measured with the same equipment, is likely to represent a true biological change.⁵³ MRE also benefits from excellent intra- and inter-observer reliability, with stiffness values reproducible on multiple platforms.⁶²

In addition to staging fibrosis, MRE might be useful for risk stratification. Some studies have shown that baseline hepatic stiffness predicts future risk of cirrhosis development in NAFLD,⁶³ clinical decompensation in cirrhotic patients⁶⁴ or HCC development and recurrence.⁶⁵ The utility and cost-effectiveness of applying baseline MRE for these purposes have not yet been established, however.

Longitudinal monitoring of patients with CLD is often required to identify those at risk of developing complications. In a prospective study on patients with NAFLD, a 15% increase in MRE values was associated with histological fibrosis progression on biopsy.⁶⁶ Interval changes in liver stiffness measured with MRE can also predict risk of hepatic decompensation.⁶⁷ While further validation is needed, these emerging results suggest that MRE might be useful for monitoring patients longitudinally and could potentially be used to assess treatment response in clinical trials.

Finally, MRE shows promise for identifying mimics of hepatic fibrosis such as nodular regenerative hyperplasia and for differentiating cirrhotic and non-cirrhotic causes of portal hypertension⁶⁸ (Figure 6).

Figure 6.

An example of MRE as a problem-solving tool. Contrast-enhanced portal venous phase of two separate 30-year-old male patients with abnormal liver function tests (A + B). Morphological changes suggest advanced fibrosis or cirrhosis with portal hypertension in both patients. c: Color-coded elastogram demonstrates only minimally increased liver stiffness. Biopsy demonstrates diffuse nodular regenerative hyperplasia without significant fibrosis. d: Color-coded elastogram demonstrates markedly increased liver stiffness consistent with stage four fibrosis (cirrhosis), which was confirmed on biopsy.

Pitfalls and future directions

Failure of MRE is rare, quoted as 4.3% in a meta-analysis of 12 studies.⁵⁹ In the past, the most common cause of technical failure in hepatic MRE was the presence of excess iron in liver parenchyma. Early versions of MRE used a GRE sequence, which is sensitive to signal loss from the presence of elevated liver iron. This problem was more apparent in 3.0T imagers than in 1.5T systems. The introduction of spin-echo echo planar (SE EPI) MRE sequences, which are less sensitive to liver iron overload, has ameliorated the problem and reduced the technical failure rate to 2%.⁶⁹ Most remaining failures are due to improper placement of the driver device or failure to ensure coupling with the body. These errors can be recognized by properly trained MR operators and in most cases, the problem can be addressed and the acquisition completed promptly.

Current commercially available MRE technology uses a two-dimensional wave propagation model for acquisition and processing to simplify the implementation and accelerate acquisition. While the validity of liver stiffness measurements obtained with this approach has been established in clinical practice and research, the model is subject to error if a significant component of wave is oblique to the acquired transverse imaging planes. More advanced versions of MRE technology acquire and process three-dimensional wavefield data, promising to provide more accurate measurements of stiffness throughout the liver with increased measurement repeatability and reproducibility.⁷⁰ Newly introduced flexible drivers enhance patient comfort and may improve shear wave illumination of the liver, while introduction of automated methods for MRE analysis can improve workflow and measurement reproducibility.^54,71

Liver inflammation

Background

Hepatic fibrosis and cirrhosis are considered as the end point of liver disease. Inflammation from a range of etiologies, dictate the rate and severity of fibrotic change and so reflect the grade of disease activity.¹² As with other components of CLD, histology is considered the gold standard for assessment. In addition to the previously mentioned limitations of biopsy, a limitation specific to the evaluation of inflammation is that histopathology only assesses the cellular components of inflammation and cannot assess other important components such as edema and hyperemia.

MRI biomarkers for assessing inflammation

DWI can assess the random motion of water molecules (diffusion), which is thought to be altered in inflamed hepatic tissue. Apparent diffusion coefficient (ADC), a marker of diffusion, demonstrated promising results in early studies assessing inflammation from varying etiologies, although subsequent studies reported contradictory findings.^72–76 One problem with using ADC is that it is affected by perfusion in addition to diffusion effects, and the two effects may offset or confound each other. Intravoxel incoherent motion (IVIM) is a more advanced method of analyzing the diffusion signal that can estimate diffusion and perfusion effects separately. In principle, perfusion-related parameters such as the perfusion fraction and pseudodiffusion coefficient might reflect microvascular alterations associated with inflammation. The capability of these parameters to detect inflammation non-invasively is limited, with multiple in-vivo studies showing only minimal to no correlation with inflammation.^73,75,77 A recent study, however, did demonstrate a stronger correlation for the perfusion fraction.⁷⁸

Dynamic contrast-enhanced perfusion imaging is another method for assessing perfusion non-invasively. This method measures the signal in liver, portal vein, and other structures before and at multiple time points after injection of a contrast agent. By applying sophisticated tissue compartment models, the arterial and portal perfusion to the liver can be estimated. The association of contrast-enhanced MRI perfusion parameters to hepatic inflammation has been weak, perhaps due to overlapping effects of inflammation and other physiological and structural changes associated with CLD.⁷⁹ A single study did, however, demonstrate the potential of the arterial fraction parameter in distinguishing mild activity from moderate-to-severe activity and for differentiating no activity from mild activity.⁷⁹

Other investigators have explored the relaxation parameters T1, T1ρ, and T2 for assessing hepatic inflammation, with modest success.^80–82 A proprietary technique involving calculation of a corrected T1 relaxation time has shown promise as a biomarker for liver inflammation and cell injury.⁸³ Further, the development of DL-based models and the advent of radiomics may improve the ability to assess inflammation. In a recent retrospective study, a radiomics model based on a T2W sequence showed encouraging results for detecting hepatic inflammation.⁸⁴

The use of 3D MRE with different inversion models has introduced the potential to isolate an inflammatory process from fibrotic change (Figure 7). Hepatic stiffness measured with MRE derives from a static component (elasticity) and dynamic component (viscosity or loss modulus). The dynamic component is thought to be affected by inflammation. Two 3D MRE-derived parameters, the damping ratio and shear loss modulus, have demonstrated encouraging signs in animal models.⁸⁵ In a cohort of 175 bariatric surgery candidates, 81 with histological NASH, a multivariable model involving the damping ratio, shear stiffness and PDFF provided an AUC of 0.73 for diagnosing NASH.⁸⁶ Changes in these parameters were also found to correlate with a resolution of NASH following bariatric surgery.⁸⁷

Figure 7.

Select images of 3D MRE and chemical-shift MRI-derived PDFF on subjects with NAFLD. The subject in the top row has biopsy-proven NASH and stage three fibrosis. The subject in the bottom row has NAFL without NASH and fibrosis. Note the elevated liver stiffness and decreased damping ratio in the liver in the subject with NASH.

Clinical and research implication– Towards the non-invasive diagnosis of NASH

Clinically, the differentiation of hepatic inflammation from normal hepatic parenchyma and fibrotic tissue is particularly relevant in the management of NAFLD. Since the presence of NASH portends high risk for progressing to end-stage liver disease,⁸⁸ the identification of patients with NASH is important to trigger more intense intervention and close follow-up.

Histologically, NASH is diagnosed when hepatic steatosis is observed with hepatocyte ballooning and lobular inflammation with other additional histological findings. The NAFLD activity score (NAS) is used to grade disease activity in NAFLD.¹² This is an aggregate score based on the grade of steatosis (0–3), lobular inflammation (0–3), and hepatocellular ballooning (0–2). Scores of 5–8 are sometimes considered diagnostic of NASH.

Multiple MRI biomarkers are being explored to specifically distinguish NASH from NAFL. The use of IVIM has produced mixed results from two separate studies.^89,90 The Liver Inflammation and Fibrosis (LIF) score derived from multiparametric T1 and T2* mapping demonstrated an AUROC of 0.80 in diagnosing NASH.⁸³ MRE using shear stiffness and loss modulus demonstrated correlation with lobular and portal inflammation and excellent diagnostic accuracy in predicting NAS in animal models.⁹¹ A model containing MRE parameters and PDFF can also predict NASH with good diagnostic performance in candidates awaiting bariatric surgery as well as the resolution of NASH following surgery.^86,87

Cost considerations

The charges for performing and interpreting diagnostic medical procedures vary globally, although there is some consistency in the costs of these tests relative to each other. In the U.S., the “resource-based relative value scale” is a standard method used to guide payments for medical procedures. For diagnostic tests, this method takes into account the costs of purchasing and maintaining equipment, performing the procedure, and interpretating and reporting the results. According to the current schedule of the U.S. Centers for Medicare and Medicaide Services (CMS), the cost of an abdominal MRI exam, which could include PDFF, T2*, and MRE to assess liver fat, iron, and fibrosis, respectively, is $476. An MRI exam consisting only of MRE has a listed charge of $284. For comparison, the CMS reimbursement for Vibration Controlled Transient Elastography (VCTE) is approximately $150 and the charge for liver biopsy and histologic examination is at least $2000. The costs between tests relative to each other in the U.K. are similar. A recent publication listed the unit cost of liver MRI at £101, VCTE at £43, a panel of routine blood tests for CLD at £68, and liver biopsy at £643.⁹²

Further, the cost-effectiveness of non-invasive methods for assessing CLD need to be considered in the context of the disease in question. In chronic hepatitis-C infection, modeling studies have suggested that the most cost-effective approach is to treat all patients with any degree of liver fibrosis⁹³; in this context, the main value of biomarkers may be for assessing treatment response rather than guiding treatment decisions. In NAFLD, by comparison, modeling studies support the use of non-invasive biomarkers as a cost-effective strategy to stratify risk and direct management decisions. To that end, accurate methods such as MRE are particularly attractive as they confidently exclude advanced liver disease and thereby reduce unnecessary referrals and procedures in patients with only mild disease.^94,95

Conclusion

As we move toward a more personalized approach to patient care, objective biomarkers are required to standardize the grading of CLD and offer accurate prognostic information. Given the limitations of biopsy, non-invasive MR-based biomarkers provide exciting opportunities with minimal risk. Examples such as PDFF and shear stiffness from MRE provide us with robust, reproducible measures to assess liver fat and fibrosis respectively, which can rival histological accuracy. Additionally, several MRI-based approaches for iron assessment have been developed, and it is anticipated that a standardized approach will be established in coming years. Finally, despite limitations, emerging techniques for the assessment of inflammation and diagnosis of NASH demonstrate potential for future use.