Imaging-Based Assessment of Steatosis, Inflammation and Fibrosis in NAFLD

Due to its wide availability and low cost, conventional ultrasound is the most common first line tool for the assessment of hepatic steatosis. A fatty liver appears “bright” compared with surrounding structures due to hepatic TAG deposition, resulting in increased acoustic interfaces. A recent meta-analysis including 2815 patients from 34 studies assessed the diagnostic accuracy of ultrasound compared with histology in patients with NAFLD. Overall, ultrasound had reasonable sensitivity and specificity (84.8 and 93.6%, respectively) for detecting > 20–30% steatosis [9]. The sensitivity of ultrasound for detecting milder steatosis (5–10%) on liver biopsy was lower, with values as low as 65% [9]. Ultrasound was also inaccurate in differentiating fatty liver alone, from NASH and liver fibrosis [9]. A clear limitation of ultrasound is its operator dependency, and future studies should include detailed reliability assessments, which are currently lacking. Despite this, ultrasound remains a good first line imaging modality for the assessment of fatty liver. However, as ultrasound is insensitive for mild steatosis, if an individual is strongly suspected of having NAFLD clinically and the ultrasound is reported as normal, then second line imaging modalities should be considered to reduce missed diagnoses.

Controlled attenuation parameter (CAP) is a novel ultrasound technique that measures steatosis simultaneously with liver stiffness during vibration-controlled transient elastography using the FibroScan™ (Echosens, Paris, France) M or XL Probe. It is known that hepatic steatosis alters the propagation of ultrasound waves through the liver, and CAP measures these physical changes, giving an estimate of steatosis. Results of CAP range from 100 to 400 dB/m [10]. The first study describing the efficacy of CAP was published in 2010 and included 115 patients with liver disease from multiple etiologies. In that study, there was a close relationship between CAP and histological assessment of steatosis (r ² = 0.81). Moreover, CAP accurately detected > 10 and > 33% steatosis with AUROCs of 0.91 and 0.95, respectively [10]. A subsequent larger study of 153 patients with chronic liver disease and body mass index ≥ 28 kg/m² found that the optimum cut-off for detecting > 10% steatosis was 283 dB/m giving a sensitivity and specificity of 76 and 79%, respectively [11]. However, CAP had limited ability to differentiate between specific grades of steatosis, particularly between grades 2 and 3 [11].

The largest prospective study assessing CAP in NAFLD included 261 patients and explored the relationship between CAP and steatosis grade (S0 to S3) and also looked at discordance rates and factors associated with discordance [12]. Overall, CAP values for S0, S1, S2 and S3 were 264 ± 45, 298 ± 48, 331 ± 37 and 336 ± 31 dB/m, respectively [12]. Significantly higher CAP values were seen in patients with S2/S3 steatosis compared to S1 (333 versus 285 dB/m; p < 0.001), and the AUROC of CAP to detect ≥ S2 and S3 was 0.8 and 0.66, respectively [12]. The optimal CAP cut-off for ≥ S2 disease was 310 dB/m [12], which provided a correct assessment of S2/S3 in 86% of the cohort [12]. Discordance of at least one grade between CAP and histology was observed in 31% of the patients.

A meta-analysis including nine studies assessing CAP in patients with chronic liver disease of varying etiologies found overall summary sensitivity and specificity values of 78 and 79% for ≥ S1, 85 and 79% for ≥ S2 and 83 and 79% for S3 [13]. A limitation of this study was that CAP cut-off values were different across the studies and were not validated [13]. In order to address the limitations of this study, a meta-analysis of individual patient data was conducted to establish CAP cut-off values for the diagnosis of steatosis and define values for distinguishing between steatosis grades [14•]. Data from 19/21 eligible studies were available, including 2735 patients (> 500 with NAFLD) with chronic liver disease [14•]. In this study, optimal cut-offs were 248 and 268 for > S0 and > S1, respectively [14•]. This study also found that disease etiology may influence the CAP reading, with higher CAP values (by 10 dB/m) for the same grade of steatosis seen in patients with NAFLD compared with individuals with hepatitis B and C [14•].

A major limitation of CAP is its failure rate, which occurred in approximately 20% of the patients with NAFLD [13]. This was higher than that reported for other etiologies of liver disease where the rate was 7.7% [15]. Failure of CAP was associated with older age, BMI, presence of metabolic syndrome and female gender, all of which are frequently present in patients with NAFLD [15]. The high failure rate in patients with NAFLD is not surprising as until recently CAP was only available on the FibroScan™ M probe. The development of CAP on the XL probe may overcome these limitations [16, 17], but further validation is needed.

Overall, CAP is a relatively simple and inexpensive method for steatosis assessment that is reasonably accurate for the diagnosis of steatosis. When combined with other clinical assessments, it is likely to help clinicians diagnose or exclude steatosis. It also has the advantage that liver stiffness (discussed in the following texts) is measured simultaneously, providing an assessment of fibrosis at the same time. However, CAP appears to have limited ability to differentiate between grades of steatosis and therefore is unlikely to be accurate enough to monitor changes in steatosis with treatment, although this has not been formally assessed.

There are a number of MR techniques that have been described to measure steatosis, including MR spectroscopy (MRS) and conventional MR methods (Dixon in- and out-of-phase imaging) [18]. Until recently, MRS has been the most widely used modality to measure liver fat in large population-based studies and some therapeutic trials [19‐24]. However, this technique requires specific expertise and as a result has not been widely used in clinical practice.

More recently, a new method of MRI has been developed called proton density fat fraction (PDFF) which improves on Dixon in- and out-of-phase imaging [25]. Similar to other MR methods, this technique measures steatosis by assessing the fraction of MRI-visible protons bound to fat, divided by all protons bound to fat and water in the liver. The advantage of magnetic resonance imaging–proton-derived fat fraction (MRI-PDFF) over previous methods is that it gives a global assessment of liver fat, whereas techniques, such as MR spectroscopy, measure fat from regions of interest within the liver, which can be associated with sampling variability [26]. Recent cross-sectional studies have shown that PDFF correlates well with histological assessment of steatosis (r ² = 0.54, p < 0.001 and r ² = 0.69, p < 0.001, respectively) [25, 27] and can accurately distinguish between the presence or absence of hepatic steatosis (AUROC of 0.989) [25]. MRI-PDFF also had good inter-examination accuracy for whole liver assessment (ICC = 0.999; SD < 0.24%, range < 0.45%) [28]. Furthermore, MRI-PDFF was shown to have better inter- and intra-observer agreement compared with histological steatosis grading (p < 0.001) [29].

MRI-PDFF has shown promise in longitudinal studies assessing changes in steatosis with treatment [30‐32]. In the FLINT study, a randomized controlled trial (RCT) of obeticholic acid vs. placebo for 72 weeks, MRI-PDFF accurately differentiated steatosis S0–1 from S2–3 (AUROC 0.95) and S0–2 from S3 (AUROC 0.96) at baseline [33•]. At 90% specificity, MRI-PDFF cut-off values were 16.3% (83% sensitivity) and 21.7% (84% sensitivity) for distinguishing histological steatosis S0–1 from 2 to 3 and S0–2 from S3, respectively [33•]. At week 72 of treatment, 49% of patients had no change in histological steatosis grade (mean PDFF change + 0.3% ± 6.3%), 9% of patients had increased steatosis grade (mean PDFF change + 7.7% ± 6.0%) and 42% of patients had a reduction in steatosis grade (mean PDFF change − 7.4% ± 8.7%) [33•]. PDFF change had AUROCs of 0.81 for both a reduction and increase in steatosis grade, respectively [33•]. There was also a good relationship between change in PDFF and change in histological steatosis grade (r ² = 0.63, p < 0.001) [33•].

A secondary analysis of the MOZART trial (RCT of ezetimibe versus placebo for 24 weeks in NASH; n = 50) showed a strong correlation between MRI-PDFF and total liver volume (TLV) (r = 0.57, p < 0.0001) and MRI-PDFF and total liver fat index (TLFI) (r = 0.94, p < 0.0001) at week 0 [31]. Patients with grade 3 steatosis on index biopsy had significantly higher PDFF, TLV and TFLI than those with grade 1 steatosis on index biopsy [31]. At week 24, the correlations between the MRI measured indices remained similar, indicating longitudinal reliability [31]. The change in PDFF over the 24-week period also correlated significantly with change in TLV [31]. There was no significant association between the change in TLV and change in histological steatosis at 24 weeks [31]. The authors speculated that within the context of other published studies [32] [19], MRI-PDFF may be more responsive than histological steatosis in detecting incremental hepatic fat change. However, this was a small study, and the authors used grade of steatosis, rather than percentage of steatosis, to assess histology, which may have not been sensitive enough to detect minor histological changes in liver fat [34].

Two recent studies have directly compared MRI-PDFF with CAP for quantifying hepatic steatosis [35••, 36••]. Both studies showed superiority of PDFF over CAP. In the study by Park et al., MRI-PDFF more accurately detected any steatosis than CAP (MRI-PDFF: AUROC 0.99 vs. CAP: AUROC 0.85) and was also more accurate than CAP for identifying specific steatosis grades (p = 0.0091) [36••]. Almost identical findings were seen in the other well-conducted study from Japan [35••].

The relationship between MRI-PDFF and histological hepatic steatosis becomes complicated when patients develop advanced liver fibrosis. In the MOZART study, mean PDFF trended downwards between F0–2 and F2–4 [31, 34]. It is known that MRI may underestimate the amount of fat in hepatocytes when compared with histology in those with more advanced fibrosis because the number of hepatocytes per volume of liver is reduced, leading to an apparent reduction in liver fat [37]. Therefore, it is likely that fibrosis stage will need to be considered when measuring steatosis in patients with chronic liver disease [37]. Concomitant assessment of liver fibrosis and steatosis by MRI may overcome this need [34], as discussed later in this review.

Overall, MRI-PDFF offers an accurate non-invasive assessment of steatosis and may have a role in monitoring a patient’s response to treatment. However, MR is expensive, and there is already pressure on MRI scanners for other indications; so, it may not be feasible to use these techniques widely to diagnose and monitor steatosis. However, MRI-PDFF is a useful tool to measure changes in steatosis in clinical trials.

Liver fibrosis can be staged using FibroScan™ (Echosens, Paris, France), a one-dimensional ultrasound transient elastography (TE), which measures the velocity of a low-frequency (50 Hz) elastic shear wave, emitted from the ultrasound probe, propagating through the liver [38]. The elastic modulus underpins the measure, i.e. the stiffer the tissue, the faster the shear wave velocity. The liver stiffness (LSM) is measured through a cylindrical volume 1 cm wide and 4 cm long, 2–7 cm below the skin surface. Results are expressed as kilopascals (kPa) using the median of ten valid measurements with a range of 2.5–75 kPa; normal liver has a value of < 5.5 kPa [38]. It is the most commonly used and most validated imaging method for non-invasive assessment of liver fibrosis currently available.

A recent meta-analysis of the use of TE in patients with NAFLD (9 studies, 1047 patients) suggests that TE has excellent diagnostic accuracy for cirrhosis (92% specificity and 92% sensitivity), good accuracy for F3 (82% specificity and 85% sensitivity), but modest accuracy for F2 (75% specificity, 79% sensitivity) [39]. Despite this, TE can rule out cirrhosis with a high NPV (~ 90%) [40]. One of the challenges of using TE is determining the optimum cut-off for diagnosing a specific stage of fibrosis. Generally, a very low LSM is very good at ruling out significant liver fibrosis and a very high LSM indicates cirrhosis with high probability. However, intermediate LSM readings are less accurate, and the choice of a specific cut-off becomes a real balance between specificity and sensitivity. For example, for a diagnosis of cirrhosis, LSM cut-offs between 10.3–17.5 kPa will yield sensitivity of 78 to 100% and specificity of 82 to 98% [39]. Therefore, clinicians may use different cut-offs for different stages of fibrosis in different environments depending whether they want to exclude or diagnose a fibrosis stage with high probability. In general, a cut-off of < 8 kPa (or 7.2 kPa for the XL probe) reliably excludes advanced fibrosis (F3–4) and a cut-off > 9.6 kPa is suggestive of F3–4.

A recent population-based, cross-sectional study illustrates the utility of TE as a non-invasive fibrosis marker in detecting clinically relevant liver fibrosis in the community. A total of 3041 participants (aged ≥ 45 years) from Holland underwent TE with LSM ≥ 8.0 kPa, defined as the surrogate marker of clinically relevant liver fibrosis [41•]. Of those with reliable LSM, overall 169 (5.6%) demonstrated LSM ≥ 8.0 kPa, and this increased to 8.4% in participants with steatosis on ultrasonography and further increased to 17.2% in those with steatosis and type 2 diabetes [41•]. These findings were consistent with another prospective study from Hong Kong in diabetic patients (n = 1918) which found that 17.7% of patients had an increased LSM (≥ 9.6 kPa by M probe and ≥ 9.6 kPa by XL probe) [42•]. Of those patients that underwent liver biopsy, 50% were subsequently diagnosed with advanced fibrosis [42•]. Both these studies indicated the potential to use TE in the community to identify individuals with undiagnosed significant liver disease.

Although the value of TE is clear, it is now recognized that hepatic steatosis has an impact on the accuracy of LSM. This was demonstrated in a recent study of 253 patients with NAFLD, which showed that patients without significant fibrosis (F0–F1) or severe fibrosis (F0–F2) had a higher rate of false-positive LSM results for F3–4 when severe steatosis (≥ 66%) was present compared to those with lesser degrees of steatosis (F0–F1 23.6 vs. 14.9%, F0–F2 33.3 vs. 13.2%, respectively) [43•]. Building on these findings, another study assessed the effect of CAP values on accuracy of LSM in 324 patients with NAFLD using the M probe [44]. CAP values were classified into tertiles (lower 132–298, middle 299–338, higher 339–400 dB/m). Among patients with F0–F2, mean LSM values increased according to CAP tertiles (6.8 versus 8.6 versus 9.4, p < 0.001). Moreover AUROCs for the diagnosis of F3–F4 with LSM progressively reduced with increasing CAP tertiles (0.915, 0.830 and 0.806), and false-positive rates for F3–F4 increased with increasing CAP (7.2% in lower versus 16.6% in middle versus 18.1% in higher CAP tertiles) [44]. Both studies elegantly define the effect of steatosis on fibrosis prediction, but further studies will need to assess the effect of CAP on LSM values using the XL probe.

A major challenge of using TE in patients with NAFLD is the failure rate (no/insufficient valid measurements) or unreliable results (IQR/LSM ≥ 30% in patients with LSM ≥ 7.1 kPa), which was seen in 14.4 and 8.9% of cases, respectively, in one of the recent well-conducted NAFLD studies [45••]. In the largest series to date (≥ 13,000 examinations in 7261 European patients), LSM failure and unreliable results were independently associated with BMI ≥ 30 kg/m², age, hypertension, diabetes, operator experience and female gender [46].

As mentioned previously, the XL probe may overcome some of the challenges posed by obese patients. This probe emits a lower central ultrasound frequency, generating a deeper region of interest. The failure rate of the XL probe was significantly lower than that of the M probe (2 vs. 10%; p = 0.002), and reliable measurements were obtained in 65% of obese individuals [47]. A study of patients from France and Hong Kong showed it was possible to obtain valid LSM readings in 91.2% participants using either the M or XL probe [48•]. However, discordance of ≥ 2 fibrosis stages between histology and LSM occurred in 11% of patients using the XL probe, which was 4- to 5-fold more frequent among patients with severe obesity (BMI ≥ 40 kg/m² 32 vs. 8%) and liver stiffness > 7.0 kPa (20 vs. 4%) [49]. It should be noted that LSM values generated by the XL probe were lower per stage of fibrosis, and so, distinct cut-off values are required [47, 49]. Finally, LSM may predict long-term prognosis. In a study of 556 patients with NAFLD followed for a median of 6.4 years, individuals with higher LSM had significantly worse overall survival and worse survival free from liver-related and extra-hepatic complications [50•].

Acoustic radiation force impulse imaging (ARFI) is a form of elastography that is performed on commercially available B-mode ultrasound machines [51]. ARFI generates short duration acoustic impulses that excite tissues, and the resulting shear wave velocity through the liver is measured giving a liver stiffness measurement. ARFI has the advantage over TE of giving an assessment of the sonographic appearance of the liver and other abdominal structures, as well as giving an estimate of liver fibrosis. A systematic review of seven studies (723 NAFLD patients) showed that ARFI was reasonably accurate for the detection of significant fibrosis (≥ F2) (AUROC 0.898; summary sensitivity and specificity 80.2 and 85.2%, respectively) [52]. However, optimal cut-off values were not assessed in this paper [52]. Much of the work in assessing ARFI as a measure of liver fibrosis has been conducted in patients with viral hepatitis, rather than NAFLD. Therefore, optimum cut-offs for the diagnosis of varying stages of fibrosis in NAFLD need to be validated. There are also different versions of ARFI (such as Siemens Virtual Touch Quantification™ and Phillips ElastPQ™), which may have different diagnostic accuracies and different diagnostic cut-offs. ElastPQ™ gives significantly lower ARFI readings than the Virtual Touch quantification™ [53]; so, these factors need to be considered when using these techniques.

Another variation of ultrasound elastography is Supersonic Shear Imaging (SSI; Aixplorer, France), which uses an ultrafast ultrasonic scanner [51]. This technique has shown real promise in the assessment of liver fibrosis and performed better than TE in patients with hepatitis C [54].

Only one study compared the diagnostic performance of FibroScan™ M probe, ARFI (Siemens) and SSI in 291 patients with NAFLD using histology as the reference. Overall, SSI performed best for the detection of significant fibrosis, advanced fibrosis and cirrhosis (AUROCs for SSI, FibroScan™ and ARFI were 0.86, 0.82 and 0.77 for diagnoses of ≥ F2; 0.89, 0.86 and 0.84 for ≥ F3; and 0.88, 0.87 and 0.84 for F4; respectively) [45••]. The results are similar to prior individual studies [39, 55]. It appears that none of the three imaging methods could accurately predict < F2 fibrosis, so further optimization of ultrasound techniques is needed [45••]. Failure rates were lower for ARFI (0.7%) than for SSI or FibroScan™ (both p < 0.0001), whereas unreliable results were higher for ARFI (18.2%) than for SSI or FibroScan™ (p = 0.0001 and p = 0.001, respectively) [45••]. Further trials are needed to confirm the utility of SSI in NAFLD.

TE was the best performing test (highest AUROC [0.83], sensitivity [88%] and NPV [90%]) for F3–4 when compared with eight blood fibrosis tests (BARD, NAFLD Fibrosis Score, Fibrometer^NAFLD, aspartate aminotransferase to platelet ratio index (APRI), FIB4, FibroTest, Hepascore, FibroMeter^V2G) in a well-conducted study of 452 patients with NAFLD [50•].

Magnetic resonance elastography (MRE) is a novel method of elastographic assessment which uses a modified phase-contrast pulse sequence to visualize the propagation of a shear-wave deep into liver tissue, evaluating a large portion of the liver [26]. A recent single-center prospective US study (117 patients) found that MRE very accurately discriminated between F3–4 and F0–2 (AUROC 0.924, p < 0.0001), but it was less accurate for distinguishing individual fibrosis stages [56••]. A threshold of ≥ 3.63 kPa demonstrated a NPV and PPV of 97 and 68%, respectively, for a diagnosis of F3–4 [56••]. The ability of MRE to discriminate between NASH from non-NASH was modest (AUROC 0.73) [56••]. Recently, a systematic review evaluated nine studies (232 NAFLD patients) and found that the summary AUROCs for ≥ F1, ≥ F2, ≥ F3 and F4 were 0.86, 0.87, 0.90 and 0.91, respectively [57].

Given its accuracy, MRE also may offer a good non-invasive tool to monitor changes in liver fibrosis. In a placebo-controlled trial of sitagliptin in NAFLD, MRE was shown to have robust correlation coefficient between baseline and 24 weeks [58]. Longitudinal studies of contemporaneous MRE and liver biopsy are underway, and their results are eagerly awaited.

Two recent cross-sectional studies have compared the effectiveness of MRE vs. TE for the diagnosis of liver fibrosis in patients with NAFLD [35••, 36••]. The first study, from Japan, assessed 142 patients and demonstrated that MRE was superior to TE for the diagnosis of ≥ F2 (AUROCs 0.91 vs. 0.82, respectively; p = 0.001) and F4 (AUROCs 0.97 vs. 0.92; p = 0.049) [35••]. A similar US-based study in 104 patients showed similar findings in a more obese cohort [36••]. In that study, the M or XL probe for TE was used where indicated, whereas the Japanese study used the M probe only. Overall, MRE had an AUROC of 0.82 for a diagnosis of any fibrosis (F1–4), which was superior to TE (AUROC 0.67). Using a threshold of 2.65 kPa, MRE had a sensitivity of 76.5%, specificity of 79.1%, PPV of 81.3% and NPV of 73.9% for a diagnosis of any fibrosis [36••]. MRE was also more effective than TE in diagnosing NASH (AUROC 0.70 vs. 0.35; p = 0.001) [35••, 36••].

Overall, MRE and TE both perform well for a diagnosis of advanced fibrosis and cirrhosis, with MRE being slightly superior. However, MRE outperforms TE for a diagnosis of earlier stages of fibrosis. MRE does not appear to be affected by obesity and ascites, but inflammation and liver iron overload may cause inaccuracies in MRE readings. The major limitation of MRE, when compared with TE, is its cost, and as a result, it is not likely to be widely used for the staging of NAFLD in the community. However, when compared with liver biopsy, the cost of MRE is much more acceptable, and as a result, specialist centers that have the technology available could choose MRE as a second-line investigation for the staging of NAFLD in those where they are considering a liver biopsy. It is also likely that MRE combined with PDFF will continue to be used in clinical trials to monitor response to treatment.

Another novel MRI technique that has been evaluated is multiparametic MRI (liver MultiScan™, Perspectum Diagnostics, Oxford, England), which assesses hepatic steatosis, fibrosis and iron content. In a study of 79 unselected patients referred for liver biopsy, corrected T1 images (cT1) were significantly different between all fibrosis stages, except between mild and moderate fibrosis [59]. In addition, steatosis measured by rapid spectroscopy correlated strongly with semi-quantitative steatosis scores (r ² = 0.89, p < 0.0001) [59]. Validation of this technique in larger NAFLD cohorts is needed to accurately assess the clinical utility of this technology.