Imaging biomarkers of NAFLD, NASH, and fibrosis

2.1. Ultrasound-based biomarkers of liver fat

While conventional ultrasonography utilizes semi-quantitative ordinal categories of liver fat as mild, moderate, and severe with poor interobserver agreement [17], newer quantitative ultrasound-based techniques demonstrate superior performance. These methods leverage quantitative data based on the acoustic parameters of liver tissue. The attenuation parameter, backscatter coefficient, and speed/wavelength of ultrasonic wave data are the most well-evaluated quantitative characteristics [18]. Furthermore, combining these parameters may lead to increased diagnostic accuracy [19].

The controlled attenuation parameter (CAP) is being increasingly utilized and implemented as an ultrasound-based tool to assess liver fat and can be obtained simultaneous with a liver stiffness measurement (LSM) by vibration-controlled transient elastography (VCTE) commercially marketed as FibroScan. Measurements are recommended to be obtained on patients in the supine position after at least a 2-h fast. Two probes, M and XL, are available and automatic probe selection software included in the device recommends using the XL probe when the skin to liver capsule distance is > 25 mm. Advantages of CAP include that it is a rapid, point-of-care assessment with good sensitivity and specificity for diagnosing fatty liver [20]. In addition, CAP is obtained simultaneously with VCTE. Despite these strengths, specific limitations of CAP include identifying the optimal cut point, implementing criteria for validity, and the impact of probe selection.

Karlas et al. conducted an individual patient meta-analysis on CAP diagnostic accuracy with liver biopsy as the reference and described 248 dB/m as the optimal cut point for identifying fatty liver [20]. Importantly, the majority of patients in the study had non-NAFLD liver disease, and subsequent studies demonstrated higher optimal cut points in well-characterized NAFLD populations. Specifically, Caussy et al. evaluated the optimal CAP cut point for fatty liver with magnetic resonance imaging proton density fat fraction (MRI-PDFF) as the reference and identified 288 dB/M [21], which was then replicated in a population of HIV-associated NAFLD [22]. Eddowes et al. defined the optimal cut point as 302 dB/M with liver histology as the reference [23]. Importantly, while CAP could distinguish mild from moderate to severe steatosis, it has limited reliability to discriminate between moderate and severe steatosis. Furthermore, CAP measurements with a high interquartile range (IQR) have demonstrated reduced validity in studies with histology [24] or MRI-PDFF [21] as the reference standard. Probe selection also impacts CAP values, and optimal cut points for the diagnosis of fatty liver are lower using the M probe vs the XL probe [25]. Taken together, interpreting CAP measurements should consider the probe choice and validity and are a reasonable point-of-care diagnostic for the presence or absence of fatty liver disease.

2.2. Magnetic resonance-based biomarkers of liver fat

Magnetic resonance spectroscopy (MRS) measures proton signals as a function of their resonance frequency to separate fat and water signal fractions. MRS can detect small amounts of liver fat and is considered the most accurate non-invasive method to quantify liver fat [26,27]. However, MRS has significant limitations that prevent its widespread use including the need for specialized expertise to perform and analyze MRS as well as possible sampling errors with restricted spatial coverage. MRI-based assessment of liver fat also utilizes the difference in resonance frequencies between fat and water. MRI-PDFF corrects for imaging confounders that can affect liver fat measurement and has emerged as an accurate, reproducible biomarker of liver fat [[28], [29], [30], [31], [32], [33], [34], [35]]. Importantly, MRI-PDFF overcomes the limitations of MRS by being widely available in commercial MRI systems with PDFF maps automatically reconstructed. Furthermore, MRI-PDFF maps allow for assessing multiple regions of interest (ROI) and following them longitudinally, which can overcome the heterogeneity of liver fat deposition. MRI-PDFF is reported as a continuous measure from a direct measurement of physical properties of the liver, whereas liver fat on a liver biopsy is a visual estimate reported in broad ordinal categories. In a longitudinal assessment, MRI-PDFF was more accurate at detecting changes in liver fat than liver biopsy [36] and has been validated in multiple studies [37].

Head-to-head comparisons have demonstrated that MRI-PDFF outperforms CAP in diagnostic accuracy with steatosis grade on liver biopsy as the reference [33,38]. Currently, the main utility of CAP is diagnosing fatty liver, whereas MRI-PDFF provides additional accurate quantification of the amount of liver fat that may be useful for evaluating a response to an intervention and is frequently deployed in clinical trials. Further studies on the prognostic significance of high liver fat on MRI-PDFF and its cost-effectiveness will be required prior to broadly implementing its use in clinical practice.

4.1. Diagnosis of NASH

Current imaging-based biomarkers have limited diagnostic accuracy for NASH [44]. Two primary domains of imaging biomarkers have been evaluated to address this need, corrected T1 (cT1) MRI and elastography. T1 relaxation time on MRI is a function of extracellular fluid that is associated with inflammation and fibrosis. By correcting for the opposing effects of T2∗ sequences, a cT1 value can be derived, which is associated with histologic inflammation and fibrosis with c-statistic for NASH of 0.80, but limited specificity, 52%, in a small study of 71 patients with suspected NAFLD and liver biopsy [45]. Importantly, cT1 cannot discriminate fibrosis and inflammation, limiting the ability to distinguish NASH from fibrosis, but may be associated with long-term liver-related outcomes [46]. Conversely, elastography measures the elasticity of an object and in liver disease is closely associated with the degree of fibrosis. Evaluation of 2-dimensional (2D) magnetic resonance elastography (MRE) and 3-dimensional (3D) MRE revealed excellent diagnostic accuracy for fibrosis, as detailed to follow, but only modest ability to discriminate NAFL from NASH c-statistics 0.73 and 0.75, respectively [47,48]. Multiparametric MRE is a novel imaging approach that combines assessment at multiple frequencies with 3D MRE to provide a more comprehensive analysis of liver stiffness but also has limited diagnostic accuracy for NASH, c-statistic 0.73 [49]. Similar attempts to diagnose NASH using VCTE at a cutoff of 7 kPa have also yielded modest diagnostic accuracy, c-statistic 0.75 [50].

4.2. Ultrasound-based biomarkers of fibrosis

Ultrasound-based measurements of liver stiffness can be integrated into conventional ultrasound devices as in acoustic resonance-forced impulse imaging (ARFI) and shear wave elastography (SWE) or obtained through a dedicated device, most commonly VCTE, which is commercially available as FibroScan. ARFI and SWE use high-frequency ultrasound impulses to generate sheer waves and require the operator to define a region of interest and obtain a series of liver stiffness measurements. Limited studies of SWE and ARFI in patients with NAFLD have demonstrated very good diagnostic accuracy for advanced fibrosis [51,52]. However, additional data on the optimal cut points and quality criteria are needed.

VCTE provides a point-of-care liver stiffness measurement using a single hand-held probe that measures the velocity of low-amplitude shear waves using ultrasound. As previously detailed regarding CAP, an M probe and XL probe are both available and the XL probe measures wave propagation at greater depths to overcome the impact of larger chest wall depths often from adiposity in NAFLD patients. Implementing the XL probe decreased the failure rate associated with FibroScan in NAFLD patients [53]. VCTE has been evaluated extensively leading to the development of quality criteria to guide its use. Specifically, a minimum of 10 valid measurements of which >60% should be valid and the ratio of the median valid liver stiffness measurement to IQR should be less than or equal to 0.3 [54]. Ongoing limitations of VCTE in NAFLD include (1) unclear optimal cut points, (2) inability to scan or an unreliable scans particularly in the morbidly obese or with an inexperienced operator, (3) limited publications utilizing the XL probe, and (4) limited diagnostic accuracy for earlier stages of fibrosis. Importantly, studies reporting cut-offs with fixed sensitivity and fixed specificity illustrate the wide range of measured liver stiffness values for each histologic fibrosis stage. Specifically, a multicenter report from the NASH Clinical Research Network using the M and XL probes demonstrated a low failure rate and high c-statistic for advanced fibrosis [55]. However, to exclude advanced fibrosis at a fixed sensitivity (0.90), a cut-off of 6.5 kPa is used, and to have a high specificity (0.90), a cut-off of 12.1 kPa is required. The optimal cut point using Youden's index of 8.6 is similar to other reports [56] but yielded a low PPV, 0.59, in the study.

4.3. MRI-based biomarkers of fibrosis

Magnetic resonance elastography (MRE) utilizes a mechanical driver to generate shear waves that can be assessed through a modified pulse sequence. This requires the addition of hardware (driver) and proprietary software. Wave images are converted to elastograms, which are cross-sectional maps of liver stiffness. Then a region of interest free of vessels can be selected and a mean liver stiffness measurement in kPa reported. Importantly, despite using the same units, kPa, results of liver stiffness measurements are not interchangeable between different modalities. MRE has a low failure rate [57] and is less affected by obesity with excellent interobserver agreement [58,59]. However, acute inflammation and iron overload can be associated with overestimations of liver stiffness and failed examinations, respectively.

MRE is most often performed at 60 Hz in 2D with excellent diagnostic accuracy for advanced fibrosis at a cut point of 3.63 kPa (c-statistic > 0.90) [33,38,56,60]. However, 3D-MRE can evaluate a larger volume of the liver and may increase the diagnostic accuracy of MRE for advanced fibrosis [48]. Currently, 3D MRE requires significant expertise and further validation before implementation in clinical practice. Importantly, MRE outperforms VCTE in head-to-head comparisons and has a high diagnostic accuracy for earlier stages of fibrosis, including patients with significant fibrosis who are at risk of disease progression and are potential candidates for treatment [33,56]. However, MRE may not be tolerated in a subset of patients who may be too obese for the MRI scanner, have incompatible metallic implants, or have claustrophobia. MRI is more costly than VCTE; however, a cost-effectiveness analysis for detecting cirrhosis in patients with NAFLD demonstrated that FIB-4 + MRE was more cost-effective than proceeding to biopsy after FIB-4 [61]. Further studies of the cost-effectiveness of diagnostic strategies using MRE to diagnose earlier stages of fibrosis are needed.

5.1. Non-invasive biomarkers to identify those in need of treatment

The combinatorial use of biomarkers can be used in a complementary manner to non-invasively identify patients with NAFLD with significant fibrosis and NASH who would benefit most from treatment. Newsome et al. combined VCTE, CAP, and aspartate aminotransferase (AST) in the FAST score and demonstrated good diagnostic accuracy for patients with significant fibrosis and a histologic NAFLD activity score of ≥4 [62]. The test was validated in multiple international cohorts and used an upper and lower cut point; however, 30–40% of the population falls into an indeterminate gray zone. In addition, combining the blood-based enhanced liver fibrosis (ELF) score ≥ 9.8 and VCTE ≥ 14.0 kPa was demonstrated to have a PPV of 87% for stages 3–4 fibrosis in a secondary analysis of the ATLAS trial.

Detection of ≥ stage 2 fibrosis remains a major unmet need. One of the major limitations in the past has been a high NPV but low PPV to rule in patients who are candidates for treatment in NASH (≥2 stage fibrosis). A prospective study including 460 patients who underwent contemporaneous MRE, FIB-4, and liver biopsy assessment demonstrated that combining MRE ≥3.3 kPa and FIB-4 ≥ 1.6 (MEFIB Index) provided an excellent PPV of 97.1% for significant fibrosis and was validated in an independent and ethnically and geographically diverse cohort from Japan. The simultaneous combination of multiple biomarkers is well suited to enriching populations needing treatment, particularly for clinical trials. However, the sequential use of non-invasive tests will likely prove more feasible for broad implementation and minimize indeterminate results [63]. We propose an algorithm for sequential use of non-invasive tests for patients with suspected NAFLD (Figure 1).

5.2. Longitudinal assessment of liver stiffness

Higher baseline liver stiffness [64] and changes in liver stiffness were recently demonstrated to be associated with liver-related and all-cause mortality in multiple studies. VCTE data on 1,039 NAFLD patients with histologically advanced fibrosis or VCTE >10 kPa found that baseline and changes in liver stiffness values, available in 533 patients, were associated with hepatic decompensation, HCC, overall mortality, and liver-related mortality [65]. Importantly, the risk of decompensation increased in a dose-dependent fashion based on the change in liver stiffness over time, ranging from 3.8% in those with a >20% decrease in liver stiffness to 14.4% in those with a >20% increase in liver stiffness (Table 1). Similarly, Han et al. demonstrated that higher baseline liver stiffness on MRE was associated with hepatic decompensation [66]. Gidener et al. recently demonstrated baseline LSM on MRE also predicts future decompensation and death. They then used longitudinal outcome data on the development of cirrhosis to guide the optimal interval between non-invasive monitoring [67]. Regarding changes in MRE, a secondary analysis of 54 patients in a phase II trial of selonsertib demonstrated that any improvement in MRE was associated with improved fibrosis albeit with a limited diagnostic accuracy [68]. A separate longitudinal study with paired biopsy demonstrated that a 15% increase in MRE was clinically significant [69]. Future studies evaluating the clinical relevance of changes in LSM on changes in histology and liver-related outcomes are needed.