Skip to main content
Published Online:


Magnetic resonance imaging (MRI) is highly important for the detection, characterization, and follow-up of focal liver lesions. Several quantitative MRI-based methods have been proposed in addition to qualitative imaging interpretation to improve the diagnostic work-up and prognostics in patients with focal liver lesions. This includes DWI with apparent diffusion coefficient measurements, intravoxel incoherent motion, perfusion imaging, MR elastography, and radiomics. Multiple research studies have reported promising results with quantitative MRI methods in various clinical settings. Nevertheless, applications in everyday clinical practice are limited. This review describes the basic principles of quantitative MRI-based techniques and discusses the main current applications and limitations for the assessment of focal liver lesions.


Focal liver lesions (FLLs) include a wide spectrum of benign and malignant tumors with different pathogeneses, clinical presentations, and outcomes. Magnetic resonance imaging (MRI) is extensively used as a non-invasive tool for the detection, characterization, assessment of treatment response, and follow-up of these lesions. In clinical practice, the assessment of FLLs is mainly based on the morphological and qualitative evaluation of MR images, including unenhanced sequences (T2W, in-phase, and opposed-phase sequences), dynamic imaging after the administration of a gadolinium-based contrast agent, and hepatobiliary phase imaging when a hepatobiliary contrast agent (i.e., gadoxetate disodium or gadobenate dimeglumine) is administered.1

Characterization of FLLs is often challenging in clinical practice due to the overlap in imaging features among different types of tumors, atypical presentations, and uncommon tumor progression over time. Moreover, the qualitative assessment of imaging features is affected by the subjective interpretation of readers, different definitions of features among guidelines, and readers’ experience, which can lead to suboptimal inter-reader agreement. The most challenging tasks for radiologists interpreting liver images include the differential diagnosis of FLLs, prediction of tumor aggressiveness or patient prognosis, and evaluation of treatment response after locoregional and systemic therapies.

In recent years, there has been growing interest in MRI-based quantitative methods to provide a more objective and reproducible assessment of FLLs. Theoretically, quantitative MRI could improve the diagnostic and prognostic accuracy when used in combination with conventional qualitative interpretations.1 However, despite many promising research studies, only a few methods are actually used in routine practice for the assessment of FLLs. Certain techniques are frequently limited to their qualitative approach (e.g., diffusion-weighed imaging with quantification of apparent diffusion coefficient [ADC]), while others are rarely used (e.g., intravoxel incoherent motion, perfusion imaging, MR elastography), or are still limited to the research field (e.g., radiomics). This is confirmed by the absence of quantitative imaging methods in current clinical practice guidelines for the diagnosis and management of FLLs.

The purpose of this review is to describe the basic technical principles of several quantitative MRI-based techniques and to discuss the main current applications and limitations for the assessment of FLLs, placing the gap between research and practice into clinical perspective.

Overview of the main quantitative methods

Diffusion-weighted imaging

DWI provides qualitative and quantitative information based on the random movement of water molecules in human tissues. At present, DWI is integrated into the routine liver imaging protocol using Echo Planar Imaging (EPI) through modification of diffusion gradient strength and magnitude, represented by the b-value and measured in s/mm2.2 High b-values (>500 s/mm2) provide key information on the restricted diffusion of water molecules. In tissues with low cellularity, such as cysts or necrotic areas, water protons have high freedom of movement and show a drop in signal intensity on high b-value DWI compared to low b-value DWI, while solid liver tumors (with high cellular density) restrict the diffusion of water protons and appear hyperintense on high b-value DWI.3,4

The acquisition of two or more b-values allows the point-by-point quantification of tissue diffusivity through the apparent diffusion coefficient (ADC) map, derived using a mono-exponential regression model.5 The ADC map quantifies diffusion in each image voxel, expressed in 10−3 mm2/s5 (Table 1). In clinical practice, the quantitative assessment of diffusion can be easily extrapolated by drawing a region of interest (ROI) on the ADC map. Lesions with increased restricted diffusion will present low values on the corresponding ADC map.3 DWI and the ADC map must be carefully compared in clinical practice. Some FLLs with long T2-relaxation times, such as hepatic hemangiomas, may present with persistent signal intensity on high b-value images, not because of increased restricted diffusion but due to the high signal intensity on T2W sequences that “shines through” to the high b-value DWI, mimicking a malignant lesion.6 Thus, the ADC map makes it possible to differentiate true diffusion restriction from the T2 shine-through effect, since the latter will have high values on the corresponding ADC map.

Table 1. Main quantitative MRI methods for the assessment of focal liver lesions.

Methods and ParametersDefinitionBiological correlate
Diffusion-weighted imaging
Apparent diffusion coefficient (mean, min, max)Quantification of tissue diffusivity (10−3 mm2/s)Tissue cellularity &
Apparent diffusion coefficient ratioRatio between lesion and liver parenchyma apparent diffusion coefficientsarchitecture
Intravoxel incoherent motion
D or DslowPerfusion-free molecular diffusion restriction (10−3 mm2/s)Tissue cellularity & architecture
Da or DfastPerfusion-related diffusion (10−3 mm2/s)Microperfusion
Perfusion fraction (f)Fractional volume occupied by flowing spins in the voxel (%)% of perfusion
Dynamic contrast-enhanced MRIa
Semi-quantitative parameters
Area under the curveAmount of enhancement during a specific time interval (dimensionless)
Maximum slopeSlope of enhancement (dimensionless)
Maximum enhancementPeak height (dimensionless)
Time-to-peak enhancementTime between arrival of the tracer and maximum enhancement (s)
Quantitative modelsTissue microcirculation
PerfusionQuantification of tissue perfusion (mL/s/g)
Blood volumeQuantification of volume of blood (mL/g)
Blood flowQuantification of flow of blood (mL/s)
KtransTransfer constant between plasma and the extravascular extracellular space (s−1)
νeExtravascular extracellular volume (%)
vpFractional plasma volume (%)
MR elastography
StiffnessQuantification of the of the resistance offered to deformation (kPa)
ViscosityQuantification of the resistance to deformation at a given rate (Pa.s)
Dynamic modulus GRatio of stress to strain under vibratory conditions (kPa)Tissue shear properties
Storage modulus G’Represents stored energy, related to elasticity (kPa)
Loss modulus G’’Represents dissipated energy, related to viscosity (kPa)
First order features
MeanAverage of the pixels within the ROI
Standard deviationDispersion from the mean
SkewnessAsymmetric of the histogram
KurtosisPeakedness/flatness of the histogram
EntropyImage irregularity or complexity
Second order featuresTissue heterogeneity
Grey level co-occurrence matrixArrangements of pairs of pixels with the same values in specific directions
Grey-level run length matrixConsecutive pixels with the same intensity along specific directions
Gray-level zone length matrixSize of homogeneous zones for each gray-level in multiple dimensions
Neighborhood gray-level different matrix
Third order features
Laplacian of Gaussian spatial band-pass filterHighlights image features of different sizes corresponding to the spatial scale of the filter
Wavelet transformEvaluates spatial location of image features in addition to their frequency characteristics

aThe provided list is not exhaustive.

Intravoxel incoherent motion

The DWI signal actually reflects a combination of water diffusion in tissue and perfusion-related effects due to the microcirculation in the normal capillary network.7 Indeed, the liver is characterized by significant microperfusion with random movements of water molecules in blood capillaries within each image voxel, which constitute intravoxel incoherent motion (IVIM). The relative contribution of microperfusion and true diffusion-to-diffusion signal intensity is related to the b-value, with the weight of microperfusion being more significant at lower b-values.7 IVIM analysis, presented by Le Bihan et al,8 makes it possible to separate these two components by applying a bi-exponential model. IVIM analysis provides quantitative parameters that reflect both the diffusion of water molecules and regional microperfusion.8 These parameters (Figure 1) include the true diffusion coefficient (D or Dslow), related to perfusion-free molecular diffusion restriction in tissue (the slow component of diffusion), pseudo-diffusion or perfusion-related diffusion (D* or Dfast), which is linked to microperfusion and incoherent motion of water molecules (fast component of perfusion), and the perfusion fraction (f), representing the fractional volume occupied by flowing spins in the voxel5 (Table 1).

Figure 1.
Figure 1.

Example of intravoxel incoherent motion DWI with 10 b-values (ranging from b 0 to b 800), parametric maps of D, D*, f, and logarithmic plot of biexponential signal decay curve (red curve) and fitting curve (blue curve) in a 65-year-old patient with hepatocellular carcinoma developed on HCV-related cirrhosis.

Perfusion imaging

Perfusion imaging is a quantitative technique that provides information about tissue microcirculation at levels below the spatial resolution of conventional imaging techniques.9 In the liver, the most used approached is dynamic contrast-enhanced (DCE) MRI that requires i.v. gadolinium contrast administration as a tracer, followed by acquisition of signal-time curves that quantify changes in contrast concentrations over time.4 Perfusion imaging acquisitions consist of free-breathing 3D perfusion sequences covering the entire liver with a short acquisition time (1–2 s) repeated for up to 5 min after contrast administration.10 The main challenges of hepatic perfusion are related to the dual vascular supply of the liver (25% of perfusion from the hepatic artery and 75% from the portal vein), fenestrated sinusoids that allow rapid perfusion of the contrast agent and respiratory movements.9

DCE-MRI provides information based on the intralesional temporal distribution of contrast agents in lesions that often present with a heterogeneous vascular network. This is different from IVIM perfusion-related diffusion.11 Lesion perfusion can be quantified by two major mathematical methods: a semi-quantitative approach, based on parameters reflecting the shape of time-intensity curves, and a quantitative approach, based on pharmacokinetic models reflecting changes in contrast agent concentrations. Parameters that are often analyzed by the semi-quantitative approach include the the time-to-peak enhancement (TTP; defined as the time between arrival of the tracer and maximum enhancement), area under the curve (amount of enhancement during a specific time interval), maximum enhancement (peak height), and maximum slope (Table 1). Although these are the most widely adopted semi-quantitative parameters, they are affected by acquisition parameters, injection protocols including contrast volume and injection rate, and physiological conditions such as respiratory motion.1

Quantitative models per se rely on the change in concentrations of the contrast agent using pharmacokinetic modeling techniques. Unlike the hepatic parenchyma, most liver tumors have mainly arterial input and no or low portal venous blood supply, which may be best analyzed using single-input models. Dual-compartment models such as the Kety or Tofts and the extended Kety models are the most frequently used.12 In the Kety model, the transfer constant between plasma and the extravascular extracellular space (Ktrans) and the extravascular extracellular volume (νe, %) are calculated. In the extended Katy model Ktrans, ve, and the fractional plasma volume (vp, %) are assessed (Table 1).


Magnetic resonance elastography (MRE) quantifies the mechanical properties of tissues by measuring the propagation of shear waves induced by an external vibrating compression device.13,14 The external driver generates a continuous acoustic vibration that is transmitted throughout the entire abdomen.14 MRE images are acquired using a phase-contrast pulse sequence with motion-encoding gradients synchronized to the mechanical waves.13,14 This sequence detects the shear waves that are converted in quantitative maps (known as elastograms) of tissue stiffness (measured in kPa by drawing region of interests) and colored elastograms used for qualitative interpretation. Other mechanical properties such as tissue viscosity may be assessed (Table 1).


Radiomics is a new quantitative method that extracts a large number of mathematical features that cannot be assessed visually.15 Radiomic features quantify the distribution of signal intensities within an ROI reflecting lesion heterogeneity.16 Therefore, it is not an MR quantitative methods per se, but rather a quantitative approach that can be applied to MRI. Indeed, radiomics was initially applied to CT imaging, but experimental applications are rapidly expanding to MRI and can be theoretically performed in all MRI sequences.17 The radiomic workflow involves a complex multistep process including imaging acquisition, lesion segmentation, feature extraction, reduction and selection, model building, and finally validation in internal and external cohorts.18 Lesion segmentation (Figure 2) and feature extraction are the most critical steps. Several in-house designed, commercially available, or open-source research softwares have been used for lesion segmentation using manual, semi-automatic, or automatic tools. Although manual segmentation by expert radiologists is still considered to be the gold standard, this is time-consuming and may be affected by intra- and inter-reader variability.19 Segmentation of FLLs is usually obtained by drawing an ROI on the largest lesion cross-section or a 3D ROI on the whole lesion. A large number of quantitative mathematical features are then extracted. First-, second-, and third-order radiomic features are then classified (Table 1).18,19

Figure 2.
Figure 2.

68-year-old male with HBV cirrhosis. Contrast-enhanced MRI on hepatic arterial (A), portal venous (B), and delayed (C) phases shows a 33-mm hepatocellular carcinoma with arterial phase hyperenhancement (A), washout and capsule (B and C). Tumor segmentation (D) was performed on portal venous phase by manually drawing a region of interest within the lesions margin, using a freely available software (LIFEx, Corresponding histogram (E) shows distribution of signal intensities within the region of interest.

Applications in focal liver lesions: results and performance

Tumor detection and characterization

One should keep in mind that the performance of qualitative and visual analysis of MR imaging for the detection and characterization of focal lesions is very high. Therefore, the added value of quantitative methods is expected to be limited.

Diffusion-weighted imaging

DWI is by far the most widely used technique, routinely included in clinical practice. It has been shown to be more sensitive for the detection of FLLs than other MRI sequences. The main advantage of DWI is that it is rapidly acquired with no need for i.v. contrast administration. Nevertheless, and noticeably, the vast majority of published studies use a strict qualitative visual assessment of ADC in FLLs (i.e., hypersignal intensity on high b-value or not).

Quantitative measurements are less routinely performed and reported in clinical practice than qualitative assessments. Historically, ADC values have been extensively studied for the discrimination between benign and malignant FLLs. Hepatic cysts and hemangiomas have been shown to have significantly higher ADC values than non-cystic malignant FLLs, and malignant tumors to have ADC values lower than that of the liver (Figure 3).20 In our opinion, this benign/malignant opposition is frequently clinically irrelevant, and misleading. It does not reflect the management of patients with liver lesions and does not take into account the clinical context.

Figure 3.
Figure 3.

44-year-old male with chronic hepatitis B and two liver lesions on T2W image (A) and contrast-enhanced MRI (B) consistent with hepatocellular carcinoma (arrows) and hepatic hemangioma (arrowheads). On diffusion-weighted imaging (C), both lesions demonstrate high signal intensities at b 800. On apparent diffusion coefficient (ADC) map (D), mean value of hepatocellular carcinoma was 771 mm2/s in the most restricted tumor area, while no diffusion restriction was observed in hepatic hemangioma (mean ADC value of 1746 mm2/s).

For instance, in patients with fortuitously discovered lesions, the vast majority of lesions are benign but may require dramatically different treatments. In one retrospective study, focal nodular hyperplasia and hepatocellular adenoma, which are two benign hepatocellular tumors with different management strategies, were shown to have lower ADC values than the background liver parenchyma, and ADC was of limited help for the differentiation between the two entities.21

Furthermore, ACD measurements are known to be affected by the type of MR scanner and field strength. The ratio between lesions and background parenchyma ADC values has been proposed in some studies to improve the accuracy of the differential diagnosis, with a slightly higher reported accuracy than standard ADC values (92% vs 89%).22 In patients with cirrhosis, it has been used to differentiate low-grade (LGDN), high-grade dysplastic nodules (HGDN) and hepatocellular carcinoma (HCC). In the study by Inchingolo et al, the sensitivity, specificity, and accuracy for the diagnosis of “HCC +HGDN” were 90.9%, 81.0%, and 83.6%, respectively, when lesion-to-liver ratio was <0.95.23

Mean or minimum ADC values have also been evaluated according to different histopathological features of HCC, with histopathological differentiation of HCC showing inverse correlation with the ADC value (r = −0.51, p = 0.012).24 In line with these results, Li et al. reported a sensitivity of 69.6% and a specificity of 73.4% of mean ADC for the differentiation between poorly and non-poorly differentiated HCC (with a cutoff value of 0.96 × 10−3 mm2/s) and a sensitivity of 78.3% and a specificity of 61.5% of minimum ADC (with a cutoff value of 0.90 × 10−3 mm2/s).25 This was expanded by Ogihara et al. using a multivendor approach.26 Finally, Zhao et al, in a large retrospective study, showed a significantly lower ADC values in HCC with microvascular invasion (MVI) on histopathology.27 The mean ADC provided a sensitivity of 79.2% and a specificity of 50.5% (with a cut-off value of 1.19 × 10 mm/s) for the identification of MVI.

Intravoxel incoherent motion

The performance of IVIM parameters for the characterization of FLLs has been assessed in several studies. The true diffusion coefficient (D) of benign FLLs, such as cysts and hemangiomas, was shown to be significantly higher than in primary and secondary malignant tumors, while results on the performance of perfusion-related diffusion (D*) and the perfusion fraction (f) are discordant.28–32 So far, the added value of IVIM over ADC does not seem to be major. There is even a trend toward a better performance of ADC values in some studies to differentiate benign and malignant FLLs.31–34

IVIM parameters have been also evaluated for the prediction of histopathological grade with studies showing that high-grade HCCs have significantly lower D and ADC values than low-grade tumors.35–38 A recent meta-analysis including 16 studies concluded that both ADC and D are highly accurate for the non-invasive grading of HCC with pooled sensitivity and specificity of the ADC value and D for the discrimination of poorly differentiated HCC of 84 and 80%, and 92 and 77%, respectively.39 D was only better for the prediction of poorly differentiated HCC (AUROC was 0.94 for D vs 0.89 for ADC, p = 0.007).39 Low D values have also been identified in HCC with MVI (AUROC of 0.815 (95% CI, 0.740–0.877), while no differences were observed in IVIM perfusion parameters.40,41

Finally, IVIM parameters have been correlated with histopathology and genetic expression of resected colorectal liver metastases. Granata et al. suggested that the diffusion coefficient (AUROC 0.80) and diffusional kurtosis (a measure of the “tailedness” of the distribution) (AUROC 0.80) could help predict the presence of Kras mutation42 while Chiaradia et al43 reported that both D (r = 0.36; p = 0.035) and ADC values (r = 0.4; p = 0.02) were correlated with the degree of tumor necrosis in metastases, but not with that of viable tumor.

Perfusion imaging

There are very few studies of MRI perfusion parameters for the characterization of liver tumors. Certain studies show that semi-quantitative perfusion parameters from multiphase dynamic contrast-enhanced MR differ between hemangioma and malignant tumors (i.e., HCC, cholangiocarcinoma and metastases).44,45

Studies have shown that both HCC and liiver metastases and HCC can be differentiated using perfusion and permeability parameters extracted from perfusion MR sequences.46,47 Similarly, authors reported on ability of distribution volume and perfusion to differentiate liver metastases from neuroendocrine tumors according to their enhancement pattern (i.e., hypo or hyperenhanced).48

In cirrhotic patients, most studies have focued on the differentiation between benign dysplastic nodules and HCC. Authors have shown that both the arterial fraction and the arterial hepatic blood flow were significantly higher in HCC. The portal venous blood flow and the distribution volume were found significantly lower in HCC compared to the surrounding cirrhotic parenchyma, likely due to significant changes in tumor microvascular architecture and angiogenesis.49

Liver metastases induce changes in local hepatic hemodynamic (e.g., decrease in portal venous perfusion, or increase in arterial perfusion that were historically documented by CT and hepatic scintigraphy. More recently, studies using perfusion MRI were able to replicate these results.50 The clinical implication of such studies remain very limited since secondary tumors were already depicted on morphological cross-sectional imaging modalities.


There are only a few reports on tumor stiffness quantification for the characterization of FLLs. Initial studies reported significantly greater stiffness in malignant FLLs than in benign tumors, with the greatest stiffness in intrahepatic cholangiocarcinoma due to abundant intralesional desmoplastic stroma. A cutoff value of 5 kPa was proposed in a preliminary study by Venkatesh et al. to accurately differentiate benign from malignant FLLs.51 Another study analyzed MRE-based viscoelastic parameters for the characterizations of FLLs showing significantly higher stiffness in HCC than in benign hepatocellular tumors.52 Preliminary data also reported a correlation between HCC tumor stiffness (Figure 4) and the grade of HCC, with a trend toward increased tumor stiffness in well/moderately differentiated HCCs (6.5 ± 1.2 kPa; n = 13) compared to poorly differentiated HCCs (4.9 ± 1.2 kPa; n = 8, p < 0.01).53 Finally, ongoing research focus on the association between viscoelastic tumor properties and intrinsic tumor pressure considered both as a possible indicator of drug resistance, and as a prognostic feature.

Figure 4.
Figure 4.

75-year-old male with hepatocellular carcinoma undergoing preoperative evaluation. Contrast-enhanced MRI shows a 70 mm lesion with arterial phase hyperenhancement (A, arrow), washout and capsule on portal venous phase (B). Elastography acquired for assessment of liver fibrosis demonstrate a stiffness of 2.2 kPa (C). Lesion stiffness was 2.7 kPa which was consistent with the diagnosis of HCC (D). Resection specimen relieved a moderately differentiated HCC arising in non-cirrhotic liver parenchyma.


Recent studies have explored the potential of MRI-based radiomics for characterization and prognostication in FLLs, with promising results. Initial experiences applied radiomics and texture analysis to T2W MRI for the differentiation of benign and malignant FLLs,54–56 with an overall classification accuracy of 90.1% in the study by Gatos et al,55 and 77% in that of Jansen et al.56 The performance of MRI-based texture features was found to be excellent in benign lesions to differentiate focal nodular hyperplasia from hepatocellular adenoma on gadoxetate-disodium enhanced MRI (AUROC 0.869 [95%CI, 0.777–0.933] for the diagnosis of HCA), with added value compared to hepatobiliary phase hypointensity.57

Several radiomic applications have been explored in patients with HCC. Multiple radiomic models extracted from MRI sequences have been shown to be highly accurate in the prediction of the histopathological features of tumor aggressiveness in HCC, such as tumor grade,58,59 Ki-67 expression (a marker of tumor proliferation),60 and cytokeratin 19 expression (a marker associated with poor HCC prognosis).61

Assessment of tumor response and follow-up

The assessment of tumor response relies on various sets of semi-quantitative criteria that monitor changes of tumor characteristics on imaging over time. Recognized limitations of this approach are the need to wait several weeks (or months) after the initiation of treatments, with subsequent delay in treatment modification in non-responders, and the variety of available criteria, focusing of various tumor features (tumor size, tumor viability, etc.), making it more difficult to compare results. Quantitative imaging aims at addressing these limitations by providing an earlier and more standardized response assessment.

Diffusion-weighted imaging

In a surgical series of small resected HCC (≤5 cm), Lee et al. showed that tumor size (p = 0.002) and ADCmin ≤0.773 ×10−3 mm2/s (p < 0.001) were independent risk factors for early HCC recurrence.62 Pre-clinical and clinical studies have shown that ADC values could indicate the degree of tumor necrosis in HCCs treated with loco-regional therapy as necrotic tissue shows higher ADC values than viable tissue.63–68 The meta-analysis by Liu et al. including 12 studies comprising 624 patients and 712 tumors reported a pooled sensitivity, specificity and AUROC of DWI in diagnosing residual or recurrent HCCs after TACE of 85% (95%CI: 74–92%), 83% (95%CI: 75–88%), and 0.90 (95%CI: 0.87–0.92), respectively.69 Interestingly, these alterations can be observed as early as one week after the treatment, thus helping predict further response.67,68,70 Barat et al. suggested that a low ADC value at 1 month after ablation was associated with an early local recurrence of HCC.71 Several teams have also investigated the role of the pre-treatment ADC value in predicting tumor response. Although these series are preliminary, tumor ADC obtained before transarterial chemoembolization or radioembolization can be used to predict tumor response and patient survival.72–74

In patients with liver metastases, ADC could predict poor response to ablation, radioembolization, or systemic therapies.75–77 Nevertheless, increase in ADC values in responding liver metastases that occur within days after the start of treatment appear of smaller magnitude than the variability of ADC measurement, suggesting that it is not reliable enough to predict final response at such an early time point in individual lesions.

Intravoxel incoherent motion

One study suggested that D and D* obtained after treatment may be correlated to lipiodol uptake, early response, and progression-free survival in patients undergoing transarterial chemoembolization (TACE) for the treatment of HCC.78 In a study by Chiaradia et al., D was shown to have a significant positive correlation with the degree of tumor necrosis, but not with that of viable tumor, in resected colorectal hepatic metastases treated with systemic chemotherapy.43 A prospective study by Kim et al79 observed significant increase in ADC and D values of hepatic metastases after chemotherapy in responder patients, while no changes were noted in non-responders.

Perfusion imaging

Interesting applications of quantitative MRI perfusion have been investigated for the assessment of HCC, in particular for the monitoring of treatment response after locoregional and systemic therapies.80 Ippolito et al. used a semi-quantitative perfusion approach and found differences between tumors with complete and incomplete TACE treatment.81 This is in line with results published by Taouli et al.49 Authors were able to observe significantly lower portal venous hepatic blood flow and higher arterial fraction in tumors that were not targeted by chemoembolization, compared to those that were. Unfortunately, these studies did not compare results of perfusion studies with morphological criteria, or used them as reference.

Studies have also evaluated the ability of perfusion MR imaging to document early changes after treatment in order to identify further responders. In the study by Braren et al. using a rodent preclinical model,82 the quantification of the extravascular extracellular volume fraction (ve) as soon as one day after trans-arterial embolization was associated with further tumor necrosis. Going one step further, Michielsen et al83 suggested that pre-TACE perfusion parameters assessed with perfusion MRI could help predict of progression-free survival.

Hsu et al84 showed that the Ktrans of advanced HCC was well correlated with tumor response, progression-free survival, and overall survival in patients treated with systemic therapy, suggesting that a reduction in Ktrans may be related to changes in tumor vascularization after anti angiogenetic therapy. Subsequent studies have supported the value of early perfusion changes in advanced HCC for the prediction of overall survival after systemic treatment.85,86

Similar results have been reported with Ktrans and a variety of perfusion parameters in patients with colorectal metastases treated with chemotherapy in combination with various targeted therapies,. Coenegrachts et al87 showed that the constant rate between extravascular extracellular space and blood plasma at baseline (i.e., kep = Ktrans/ve) was significantly higher in responders than in non-responders. They also showed that responders had a significant decrease in kep after six weeks of treatment. In the study by De Bruyne et al.,88 >40% reduction in Ktrans after treatment was associated with a significantly longer progression-free survival. These may be shown early after treatment introduction, as suggested by Hirashima et al.89 Authors observed variations in both kep and Ktrans within a week after treatment that could help predict response to chemotherapy.


Lower tumor stiffness may be associated with greater intralesional necrosis, as observed in HCC treated with locoregional therapies.90 On the other hand, increased HCC stiffness has been evaluated as a potential predictor of early recurrence after hepatic resection.91,92 In this setting, high pre-treatment tumor stiffness may be an additional noninvasive biomarker for a poor prognosis in patients with HCC.93


Texture-based models have been reported to predict tumor response after TACE94 and early recurrence after curative hepatectomy, alone or in combination with morphological criteria.95–97 There has been some research on MRI-based radiomics in hepatic metastases, but further studies are needed to identify the role of this technique in this context.98

Applications in clinical practice: limitations and perspectives

Diffusion-weighted imaging

Although ADC quantification has been shown to be applicable to several clinical situations, ADC values still vary significantly (and thus, possible cutoff values) with different MRI scanners.99–101 Makyarenko et al. used an ice-water phantom and showed a standard deviation of ADC measured across 35 scanners < 2%.100 The day-to-day repeatability of the measurements was within 4.5%, inter-site reproducibility of ADC was within 3%. Sasaki et al. performed a multivendor, multi-institutional comparison study and reported that, with the exception of one vendor, the intervendor variability at 1.5 T in a clinical setting was as high as 7%. Moreover, there was substantial intra-imager variability, up to 8%, depending on the coil systems in certain imagers.101 The ADC ratio between lesions and the liver parenchyma has been proposed to overcome some of these limitations, but it should be noted that the ADC of the background liver parenchyma is affected by its quality (for instance by the presence of fibrosis in patients with chronic liver disease). Moreover, DWI sequences are very sensitive to artifacts that can negatively influence image quality and ADC quantification.

At present, assessment of the ADC map for the characterization of FLLs is usually qualitative. Quantitative rules of thumbs, for instance liver metastases and cholangiocarcinoma usually displaying an ADC ≤ 1.0×10−3 mm2/s are rare, explaining why quantitative measurements are not routinely used for patient management or clinical decision making.

Intravoxel incoherent motion

Current clinical applications of IVIM have been limited by several technical factors and the need for specific acquisition protocols.7 For the moment the added diagnostic value of IVIM analysis compared to the ADC map is still a question because IVIM may be affected by MRI field strength, number or choice of b-values, and post-processing calculation models.102 Particularly, number of b-values widely varies in the published studies, with most research using from four to more than ten b-values. Although a greater number of b-values may provide more accurate information, the acquisition time would significantly increase in clinical practice. These practical limitations as well as a diagnostic performance that is similar to the extensively used ADC map could explain the limited clinical use of this quantitative method in tertiary centers for the assessment of FLLs.

Perfusion imaging

There are several reasons for the limited number of clinical applications of perfusion imaging for FLLs. Differences in imaging systems, technical parameters without standardized acquisition protocols, respiratory motions, and different pharmacokinetic models makes it difficult to compare and validate research studies in FLLs.3 Also, perfusion quantification requires prospective acquisition with contrast administration and complex post-processing methods including additional scanning and interpretation time, limiting its use to specialized or tertiary centers. Thus, for the moment, perfusion MR imaging is only used in research settings in specialized centers for the treatment of FLLs.9 Recent commercially available sequences, such as the golden-angle radial sparse parallel (GRASP) imaging, have been shown to allow for a reliable and robust assessment of hepatic perfusion parameters with quantitative results comparable to perfusion CT.103 These new generation of sequences is promising in making perfusion data more easily available, together with a routine visual interpretation of contrast enhancement.


The routine acquisition of MRE requires specific hardware that is often only available in tertiary centers, and requires additional scanning time. This significantly limits the applicability of MRE for the characterization of FLLs, especially compared to the performance of qualitative assessment by up-to-date contrast-enhanced MRI and other quantitative imaging methods. New transducer-free elastography methods using, for instance, intrinsic cardiac motion appear promising to overcome this limitation.


Although the results of radiomics are promising for the assessment of FLLs, there are still challenges that limit its application in clinical practice. Lack of methodological standardization, time-consuming segmentation, and the type of extracted features with different in-house designed or research-based software have limited the reproducibility of radiomic models.104 Moreover, radiomic features are affected by acquisition parameters, in particular, slice thickness.105 Therefore, methodological standardization and reproducibility of features are needed to provide robust radiomic models that can be used in different settings.

Bridging the gap between research and practice

The current situation of quantitative MRI in FLLs is somewhat paradoxical. We hereby discuss results of numerous studies reporting promising results in a wide variety of clinical settings but very few applications exist in clinical practice. As this review shows, quantitative MRI could theoretically be used in combination with qualitative imaging for the assessment of FLLs and could provide solutions in various clinical scenarios. However, the expected application of these results into routine clinical practice has not occurred.

There are many possible explanations for this. First, quantitative imaging is difficult to perform in the liver because it is a mobile, blood-filled, flexible organ with dual vascular input and fenestrated sinusoids. This requires advanced registration techniques and complex microvascular or tissue models. Second, patients have different morphotypes, capacities to hold their breath, possible iron overload, as well as fasting states. Third, MRI sequences are patented or trademarked making direct comparison difficult. Furthermore, the MRI signal is affected by the magnetic field strength and the overall architecture of machines and antennas. Moreover, most of the published studies are single center and retrospective, without standardized acquisition parameters, post-processing methods or predicted outcome. This results in significant intra- and inter-vendor, software, reader, and patient variability, as discussed above. Observed statistical differences in populations are, therefore, rarely applicable to individuals. There are also still concerns about the need for additional scanning time and the complexity of imaging interpretation. Finally, the good to excellent performance of qualitative imaging with up-to-date MR protocols, and easy access to percutaneous biopsy narrows the number of unmet needs that quantitative imaging can address.

Overcoming these limitations will require a significant collective effort. Nevertheless, the medical imaging community has never been more aware of this and has never had more powerful tools to succeed, as illustrated by initiatives such as the Quantitative Imaging Biomarkers Alliance or the Radiological Society of North America106 or the Biomarker Inventory or the European Society of Radiology.107


Volume 94, Issue 1122June 2021

© 2021 The Authors. Published by the British Institute of Radiology


  • ReceivedFebruary 12,2021
  • RevisedApril 28,2021
  • AcceptedMay 04,2021
  • Published onlineMay 14,2021