The noninvasive diagnosis of hepatocellular carcinoma (HCC) represents a major challenge
for radiologists. In cirrhotic livers, a new focal lesion has a high pre-test probability
of being an HCC.[1]
[2]
[3]
Magnetic resonance imaging (MRI) has a key role in detecting and characterizing HCCs
in cirrhotic livers.[4]
[5] MRI has superior tissue resolution compared with computed tomography (CT). Different
meta-analysis and systemic reviews comparing the diagnostic performance of contrast-enhanced
CT in comparison to contrast-enhanced MRI found that multiphasic MRI is more sensitive
than CT, while both modalities have similar specificity.[6]
[7] MRI contrast agents improve detection by increasing the contrast between the lesion
and the background,[4]
[6] and different types of contrast agents can improve characterization by showing changes
in hepatocyte uptake; biliary excretion; or vascular, extracellular, or intracellular
volumes.[6]
[8] This information can be complemented by information obtained from other sequences
such as restriction on DWI, high signal on T2-weighted images, and presence of intralesional
fat. Nevertheless, despite considerable recent advances in MRI, the final diagnosis
of new focal lesions often relies on histopathology.[4]
Biopsies are required for diagnosis or staging when imaging techniques cannot characterize
new liver lesions in cirrhotic patients. Histopathology is still needed for two main
reasons: first, HCC is heterogeneous on imaging, with different lesions within the
same patient and even the same tumor having different appearances. Second, HCC is
the end-stage of a continuous multistep process,[9] and the MRI findings for pre-malignant lesions and early HCC are not specific and
overlap.[10] Hence, the main question is whether a given nodule is a large regenerative nodule,
a dysplastic nodule, or an HCC. The aim is to avoid overdiagnosis and consequent overtreatment
while avoiding underdiagnosis and consequent undertreatment.
Heterogeneity Using Conventional Imaging Criteria
Hepatocarcinogenesis is a multistep process during which nodules undergo parallel,
simultaneous, and progressive changes, including increased cell density, decreased
cell differentiation, enlargement of nodules, and changes in nodule hemodynamics.[9] The MRI characterization of focal lesions in cirrhotic livers is a complex process
based on analyzing the lesion's signal intensity on different imaging sequences. In
dynamic contrast-enhanced MRI, the signal intensity during the arterial, portal venous,
and delayed venous phases reflects differences in the distribution of contrast material
between the vascular and extravascular spaces in tumors and liver parenchyma; the
noninvasive diagnostic criteria for HCC are based on these characteristics.[4]
[5]
[11]
In the initial phase of hepatocarcinogenesis, normal arterial supply decreases but
portal venous supply is still present.[10]
[12]
[13] In a more advanced stage, intranodular arterial vascularity increases due to the
appearance of unpaired feeding arteries while venous portal blood supply progressively
decreases. These phenomena result in the typical and specific enhancement pattern
of HCC on MRI using nonspecific gadolinium contrast: arterial hypervascularity and
hypointensity or “washout” during the portal venous and/or late dynamic phase. These
features represent the European Association for the Study of the Liver (EASL) and
the American Association for the Study of Liver Diseases (AASLD) noninvasive diagnostic
criteria for HCC larger than 1 cm. Although this pattern is highly specific, it has
limited sensitivity, being found in only ∼60% of HCC smaller than 2 cm,[1]
[2]
[3]
[14]
[15] because early small HCC in which neoangiogenesis is yet to progress is often hypovascular.
The difficulties in differentiating the different stages of HCC progression by imaging,
particularly in differentiating low-grade dysplastic nodules from regenerative nodules
and high-grade dysplastic nodules from early HCC, probably derive from the fact that
the different classifications simply represent parts of a continuum rather than distinct
categories ([Fig. 1]).[9]
Fig. 1 Phenotypic changes along the continuous process of hepatocarcinogenesis that cause
heterogeneity at imaging are marked by the following phases at the histologic level:
(1) low- and high-grade dysplastic nodules, which on magnetic resonance imaging (MRI)
usually appear iso- or hypovascularized (although a few are hypervascular) compared
with liver parenchyma, which histology shows have individual arteries and portal perfusion;
(2) early hepatocellular carcinoma (HCC) with residual portal perfusion, which on
MRI ranges from iso- or hypo-vascular to hypervascular, which histology shows are
well differentiated, vaguely nodular, and contain individual arteries; and (3) progressed
HCC, with characteristic imaging features of hypervascularity in the arterial phase
and hypovascularity in the venous phase, which histology shows are moderately differentiated,
distinctly nodular, with or without vascular invasion, having individual arteries
but no portal tracts.
Another source of heterogeneity in the appearance of small HCCs is the presence of
fat, also known as fatty metamorphosis. Whereas fat is often present in small, well-differentiated
HCCs, small moderately-to-poorly differentiated HCCs rarely contain fat.[2] Moreover, fatty metamorphosis can also be found in dysplastic nodules and benign
entities. Vascular changes sometimes appear in small areas inside a dysplastic nodule
or an early HCC, conferring the classical appearance of a nodule within a nodule.
As HCCs grow larger, the heterogeneity of their appearance decreases as the typical
vascular pattern develops; this pattern can be identified in more than 70% of HCC
between 2 and 3 cm in diameter.[6]
[7]
[16] At the same time, heterogeneity within lesions increases as different foci of HCC
differentiation coexist, conferring a mosaic pattern resulting from areas with different
degrees of vascularization, fatty metamorphosis, and/or hemorrhage.[17]
In 7 to 13% of cases, HCC presents as the diffuse or infiltrative HCC subtype,[18] which is often accompanied by portal vein thrombosis, resulting in vascular phenomena
and abnormal patterns of enhancement. Infiltrative HCC is easily missed on T1-weighted
images and on arterial phase images, where they usually show no enhancement or slight
heterogeneous enhancement within the tumor and the large heterogeneous tumor with
ill-defined margins blends into the heterogeneous cirrhotic liver.[19]
Another source of heterogeneity by imaging are the primary liver carcinomas with both
hepatocyte and cholangiocyte differentiation, referred to as “combined (or mixed)
hepatocellular-cholangiocarcinoma.” Recently, these tumors have increased their attention
because of greater frequency and clinical recognition.[20] There are few publications that report the radiological appearances of mixed hepatocellular–cholangiocarcinoma.
Most series reported imaging features that commonly overlap with those of intrahepatic
cholangiocarcinoma and less frequently with those of classic HCC displaying predominantly
arterial hyperenhancement and followed by diffuse or patchy washout.[20] Imaging techniques may be useful to guide biopsy on the target HCC-like or cholangiocarcinoma-like
areas.
Heterogeneity on Functional Imaging Techniques
Diffusion-Weighted Imaging
Diffusion-weighted imaging (DWI) is a functional MRI sequence that allows the characterization
of biological tissues based on the diffusion properties of water molecules. This sequence
provides indirect information about cell density and about the integrity of cell membranes
in the area of interest. In clinical practice, the concept of “diffusion restriction”
refers to a focal liver lesion showing signal intensity higher than the surrounding
liver parenchyma on DWI.[21]
Diffusion-weighted imaging now forms part of most routine liver MRI protocols in daily
practice because it has a short acquisition time, does not require intravenous contrast
agents, and is more sensitive for detecting focal liver lesions than T2-weighted sequences.[22] Although HCCs typically appear hyperintense at DWI, this finding is not specific
for HCC; thus, DWI should be integrated in MRI protocols together with dynamic sequences,
rather than used alone ([Fig. 2]).[23]
[24]
[25] DWI's sensitivity for detecting HCC in cirrhotic livers varies widely depending
on the size of the tumor and DWI specifications, ranging from 32 to 70% in HCC smaller
than 2 cm[22]
[23]
[24]
[26]
[27]
[28] to more than 90% in larger lesions.[29]
[30]
Fig. 2 A 1.2 cm hepatocellular carcinoma located in segment VIII seen on T1 pre-contrast
in (A) and opposite phase (B), arterial phase (C), and portal venous phase (D) magnetic resonance imaging. On pre-contrast T1-weighted images, the nodule presented
drop in signal on opposite phase relative to liver parenchyma (arrow in A and B) indicating presence of fat. On arterial-phase images, hyperenhancement is seen within
the nodule (arrow in C). On portal-venous-phase images, the nodule is hypointense (washout) (arrow in D). This vascular pattern is specific for hepatocellular carcinoma.
The heterogeneity in the visibility and signal intensity of HCC at DWI is the product
of complex interactions between various biological factors and MRI characteristics.
As the process of hepatocarcinogenesis and tumor differentiation occur in a continuum,
the heterogeneity of HCCs also results in overlapping diffusivity characteristics
that directly affect DWI's ability to detect these lesions. Nasu et al[30] reported that the DWI signal intensity of HCC generally increased with increasing
histologic grade (i.e., decreasing differentiation), although with large overlaps.
Analyzing a series of 98 lesions in 73 patients, Muhi et al[28] found that more than 90% of moderately-to-poorly differentiated HCCs were hyperintense
on DWI, but only 51% of well-differentiated HCCs and none of the dysplastic nodules
were hyperintense. Moreover, the microarchitecture of the background cirrhotic parenchyma
also influences the visibility of HCCs on DWI, increasing the complexity of interpreting
the findings. Diffusivity is also restricted in cirrhotic parenchyma, probably due
to fibrosis, so the DWI signal ratio between HCC and cirrhotic parenchyma is reduced.[31] This may explain why the DWI detection rate for HCC decreases with increasing severity
of cirrhosis.[30]
Although qualitative visual assessment of DWI is of limited usefulness in characterizing
nodules in cirrhotic liver, quantitative approaches such as apparent diffusion coefficients
(ADC) and the more recent intravoxel incoherent motion (IVIM) can assess nodules independent
of cirrhotic liver background, displaying the findings in parametric maps. IVIM is
more reliable; unlike ADC, which ignores the effect of the perfusion fraction in tissue,
IVIM treats DWI signal attenuation as a mixture of tissue perfusion and tissue diffusivity
effects.[32] In a recent study, Zhu et al[32] found a negative correlation (r = −0.62, p < 0.001) between IVIM quantification and the histological grade of HCCs. In a similar
study, Wei et al[33] found that the correlation was even better when IVIM was measured in the whole HCC
volume (r = −0.90, p < 0.001) rather than in a single section (r = −0.84, p < 0.001).
Heterogeneity on Liver-Specific Contrast Agents
Liver-specific (also called hepatobiliary) contrast agents include gadobenate dimeglumine
(Gd-BOPTA) and gadoxetic acid (Gd-EOB-DTPA); the latter is by far the most commonly
used liver-specific contrast agent in clinical practice and research. Gadoxetic acid's
dual behavior is especially useful in liver imaging. After intravenous injection,
it first distributes into the intravascular and extravascular spaces during the arterial,
portal, and late venous phases; afterward, during the so-called hepatobiliary phase,
it progressively distributes into the hepatocytes and bile ducts.
Gadoxetic acid is eliminated through both renal and hepatic excretion pathways in
a proportion of around 50% each in healthy subjects. The uptake of gadoxetic acid
by hepatocytes mainly occurs via the active transporter organic anion transporting
polypeptides (OATP) 1B1 and OATPB3 located at the sinusoidal pole of the hepatocytes,
and the biliary excretion occurs via multidrug resistance protein (MRP) 2 proteins
located in the canalicular membrane.[34] The expression of OATP transporters diminishes during hepatocarcinogenesis[35]
[36]; consequently, most HCCs appear hypointense during the hepatobiliary phase because
the expression of OATP is decreased or absent ([Fig. 3]). The uptake pattern of gadoxetic acid by HCC in the hepatobiliary phase can be
explained by the expression of OATP and MRP2. When OATP expression is maintained,
the intensity of HCC on hepatobiliary-phase images depends on the expression of MRP2
in the canalicular membrane. In most HCCs, MRP2 is present, and the gadoxetic acid
is eliminated to the bile ducts, making these HCCs hypointense on MRI[36] ([Fig. 4]).
Fig. 3 Correlation between the continuous process of hepatocarcinogenesis and the level
of heterogeneity at imaging obtained by liver-specific contrast agents: (1) low- and
high-grade dysplastic nodules still have large number of organic anion transporting
polypeptides (OATP) and multidrug resistance protein (MRP) transporters that regulate
the uptake and excretion of gadoxetic acid, so these nodules usually appear isointense
relative to liver parenchyma on hepatobiliary-phase images, although some high-grade
dysplastic nodules may appear hypointense. In early and progressed hepatocellular
carcinomas, OATP expression is often decreased or absent, resulting in decreased uptake,
and MRP expression is often increased, resulting in increased excretion. Thus, most
hepatocellular carcinomas appear hypointense during the hepatobiliary phase. However,
around 10% of HCC have increased expression of OATP and decreased expression of MRP,
or a pseudogland configuration, both of which result in hepatocellular carcinoma appearing
hyperintense on hepatobiliary phase images.
Fig. 4 Gadoxetic-acid liver magnetic resonance imaging (MRI) in a 67-year-old man with hepatitis
C virus-related cirrhosis. Liver MRI in the arterial phase shows a 11 mm hypervascular
nodule in segment II of the liver (arrow in A); there was no washout in the portal venous phase (B), and the lesion was hypointense in the hepatobiliary phase (arrow in C). Diffusion-weighted imaging (arrow in D) shows signal restriction. Histopathology after ultrasound-guided biopsy diagnosed
well-differentiated hepatocellular carcinoma.
OATP expression decreases prior to complete neoangiogenesis and the presence of unpaired
hepatic arteries. Consequently, nodules that are hypointense in the hepatobiliary
phase but do not have the features of HCC in earlier postcontrast phases due to a
lack of unpaired arteries are common and may represent high-grade dysplastic nodules
or early HCCs.[37]
[38]
On the other hand, around 10% of HCCs are hyperintense in the hepatobiliary phase.
This contrast uptake is most commonly associated with well-differentiated HCCs and
with either pseudoglandular architecture (classically known as green HCCs) or low
expression of MPR2 transporters.[39]
[40]
Although it has been postulated that hepatobiliary-phase hypointensity generally increases
with increasing HCC grade,[41] the reality is more complex, because both tumor and liver characteristics can affect
gadoxetic acid uptake and the relative signal intensity of HCC.[42]
Patients with significant liver dysfunction due to the cirrhosis may have decreased
parenchymal uptake of gadoxetic acid, reducing the signal contrast between HCC and
the surrounding liver; this phenomenon is particularly problematic for the characterization
of small tumors. Recently, other biochemical and clinical factors such as prothrombin
activity, total bilirubin, and cholesterol levels have been correlated to the quality
of liver enhancement in the hepatobiliary phase, suggesting that better enhancement
may be obtained in patients with HBV than in those with HCV.[43]
[44]
Heterogeneity on Liver Perfusion
Perfusion imaging of the liver with dynamic contrast-enhanced MRI characterizes and
quantifies functional microenvironmental changes in small vessels that are beyond
the resolution of morphological imaging for direct visualization. Dynamic contrast-enhanced
perfusion MRI is based on temporal changes in tissue enhancement after intravenous
administration of contrast media[45] and reflects pathophysiological changes in the microcirculation. Some groups have
reported that dynamic contrast-enhanced MRI can quantify perfusion in HCC and surrounding
cirrhotic liver, finding increased arterial perfusion and decreased portal venous
flow in HCC compared with cirrhotic liver.[46]
[47] However, in large HCCs, intratumoral heterogeneity, in particular in histological
grade and microvessel density, often hinders characterization by dynamic contrast-enhanced
MRI and makes it difficult to correlate the quantitative findings with the histological
grade.[48]
Heterogeneity in Radiological Response and Tumor Progression
Antiangiogenic drugs are recommended as the first treatment option in cirrhotic patients
with advanced HCC. The radiological response to systemic treatment of HCC is usually
measured by Response Evaluation Criteria in Solid Tumours (RECIST) 1.1 criteria.[49]
After treatment with antiangiogenic drugs, enhancement decreases either diffusely
or in part of the tumor in some HCCs, but remains stable in others. As initially reported
in the SHARP trial, the reduction in tumor burden measured by conventional RECIST
1.1 does not parallel the benefit of antiangiogenic drugs. Thus, a panel of experts
developed a set of guidelines including alternative criteria that take antiangiogenic
drugs' mechanism of action into consideration, the modified RECIST for HCC (mRECIST),[50] thus potentially capturing the heterogeneity on HCC in the response to antiangiogenic
drugs beyond mere changes in size. In the BRISK trial, a randomized phase III trial
comparing brivanib versus placebo in advanced HCC, Lencioni et al[51] analyzed the objective response by mRECIST as a predictor of overall survival in
a time-dependent covariate analysis and found a strong association between log hazard
ratios for overall survival and log odds ratios for objective response ratio (r = −0.80; 95% confidence interval: −1–0.23, p = 0.091). Note that the 95% confidence interval included 0, indicating that the results
are not robust enough to assume that the objective response according to mRECIST predicts
overall survival in advanced HCC or to consider it a surrogate end-point.
Other authors have compared the rate of objective response using RECIST 1.1 versus
mRECIST. Ronot et al[52] reported that of a total of 64 patients with HCC treated by sorafenib, only 2 (3%)
had an objective response according to RECIST 1.1, whereas 18 (28%) had an objective
response according to mRECIST. In a similar study, Gavanier et al[53] found that only 3% had an objective response to sorafenib according to RECIST 1.1,
whereas 7% had an objective response to sorafenib according to mRECIST. However, neither
study found significant differences in overall survival between patients with an objective
response and those with stable disease using either criteria. Interestingly, in both
studies the rate of disease progression was very similar using either RECIST 1.1 or
mRECIST. The RECIST 1.1 definition of progression takes into account heterogeneity
in different situations. Since the response to antiangiogenic treatment is associated
with overall survival and helps identify patients who would potentially benefit from
second-line therapy, defining the different patterns of response and their implications
is important. The BCLC group defined four patterns of radiological progression: intrahepatic
increase ≥ 20% increase in tumor size from the baseline measurement, extrahepatic
increase ≥ 20% increase in tumor size from the baseline measurement, detection of
new intrahepatic lesions, detection of new extrahepatic lesions, and/or vascular invasion.
Each pattern is associated with different post-progression survival; intrahepatic
growth is the pattern associated with the least aggressive progression, whereas new
extrahepatic lesion and/or vascular invasion is associated with the most aggressive
progression.[54]
[55] This heterogeneity might be explained by molecular factors or gene expression and
it deserves further investigation.
Immune modulators are among the most important classes of new anticancer therapeutics.
Although their role in systemic treatment for HCC is yet to be determined, various
ongoing clinical trials are evaluating their efficacy. The novel mechanism of action
of immune checkpoint inhibitors through immune and T cell activation can result in
unusual patterns of response. Some patients treated with this approach showed morphological
changes in lesions that would meet the criteria for disease progression based on traditional
RECIST (either increase in size or appearance of new lesions), but later showed profound
and durable responses. Dubbed pseudoprogression, the increase in the size of lesions
after treatment with immune modulators is the result of an inflammatory-like reaction
due to an increase in the number of immune cells between tumor cells.[56]
Although pseudoprogression is relatively uncommon, it can be challenging to differentiate
this transient phenomenon from true progression that would potentially require a change
in the therapeutic approach. Thus, it is essential to know how patients are being
treated and the mechanisms of action of these treatments to ensure accurate interpretation
of morphological changes observed at MRI or CT and optimal decisions about further
workup. In patients receiving immune-checkpoint inhibitors, HCC should be reassessed
earlier than originally planned (but only between 4 and 8 weeks after the unconfirmed
progression) to determine whether the findings represent real progression or an early
tumor response.[56]
[57]
Future Perspectives
The noninvasive diagnosis of liver nodules by CT and MRI is limited by the high level
of heterogeneity in small HCCs, by the overlapping findings between HCCs and dysplastic
nodules, and by atypical vascular patterns. Another potential limitation of the noninvasive
diagnosis is the inherent difficulties in detecting small nodules during ultrasound
surveillance caused by cirrhotic liver background. A recent controversial point is
whether the strategy of HCC surveillance should be done by US or by other contrast-enhanced
imaging modalities. Surveillance by CT or MR is generally not recommended because
of the radiation risks, the higher cost, and the increase in false positive signals
due to nonspecific findings.[4]
[5]
In the last years, few studies have postulated to use MRI with liver-specific contrast
agent for screening of HCC in patients with high risk. Authors found that MRI yielded
a detection rate of 84.8%, significantly higher than the 27.3% rate detected by routine
US[58]. The same group published another similar study detecting that screening of HCC
using MRI is also cost-effective.[59] However, there are some aspects of these studies that are at least controversial
and may limit the generalization of reported data including (1) the lower rate of
US detection reported by this group (27.3% compared with the 63% of HCC rate detection
of HCC at early stage reported by meta-analysis[60]); (2) the high rate of HCC detection at screening MRI (without histological confirmation)
suggesting that this strategy may induce HCC over-diagnosis and, in consequence, overtreatment
of non-malignant lesions; and (3) concerns on the concept and model for identification
of high-risk cirrhotic patients, and the potential bias in the selection of eastern
population that can be different than in western countries.
These current limitations of imaging to diagnose HCC limitations have not been overcome
with the introduction of functional imaging modalities such as DWI and hepatobiliary
phase gadoxetic–acid imaging, which are also affected by heterogeneity and are thus
potentially an additional source of diagnostic error.
The classic conceptualization of medical images as pictures intended solely for visual
interpretation is an obstacle that must be overcome to improve diagnosis.
Radiomics, which is an emerging field of oncological research, assume that radiologic
image features could predict the prognosis of oncologic patients, as they are associated
with tumor biological characteristics. In the field of HCC, radiomics had been mainly
explored to link image texture features with clinical outcome, by building models
that positively correlate with aggressive tumor phenotypes (future fast increase in
size and vascular invasion) and with increase in recurrence rate and lower survival.
In that regards, some preliminary studies found that the analysis of texture features
on dynamic phases (either arterial or hepatobiliary phases) may help to detect those
HCC with higher potential aggressive evolution at baseline (i.e., differentiation
low- from high-grade HCC,[61] to predict the risk of progression from hypovascular nodules into HCC, or to predict
risk of recurrence[62]) or to predict survival assessed with texture analysis of pre-treatment CT scans.[63]
The analysis of these data can be used to develop decision-support tools for diagnosis
and to devise predictive or prognostic models for outcomes of interest.[64] Potentially, these models may complement current existing staging systems such as
BCLC and would help to plan a more personalized therapeutic option after the initial
diagnosis of solitary HCC in cirrhotic patients. However, a common limitation of radiomics
studies is the inclusion of radiologic scans acquired in one or few machines. Therefore,
expanding the analysis at large multicentric scales might have affected the radiomics
results. Further studies and validations using different scanners, models, and contrast
agents are required before their application in clinical research or practice.
