DIFFUSION-WEIGHTED IMAGING OF THE LIVER: TECHNIQUES AND APPLICATIONS

SYNOPSIS

Diffusion weighted MRI (DWI) is a technique that assesses the cellularity, tortuosity of the extracellular/extravascular space and cell membrane density based upon differences in water proton mobility in tissues. The strength of the diffusion weighting is reflected by the b-value. DWI using several b-values enables quantification of the apparent diffusion coefficient (ADC). DWI is increasingly employed in liver imaging for multiple reasons: it can add useful qualitative and quantitative information to conventional imaging sequences, it is acquired relatively quickly, it is easily incorporated into existing clinical protocols, and it is a non-contrast technique. DWI is useful for focal liver lesion detection and characterization, for the assessment of post-treatment tumor response and for evaluation of diffuse liver disease. ADC quantification can be used to characterize lesions as cystic/necrotic or solid and for predicting tumor response to therapy. Advanced diffusion methods such as IVIM (intravoxel incoherent motion) may have potential for detection, staging and evaluation of the progression of liver fibrosis and for liver lesion characterization. The lack of standardization of DWI technique including choice of b-values and sequence parameters has somewhat limited its widespread adoption.

INTRODUCTION

Diffusion weighted imaging (DWI) is a magnetic resonance imaging (MRI) technique that reports on the physical process of microscopic thermal motion of water molecules in biologic tissues^1,2. The differences in the mobility of water protons create image contrast, which is influenced by the interaction of water molecules with cellular membranes, macromolecules, degree of cellular density and the size of the extracellular extravascular space³. DWI is increasingly being employed in liver MRI given recent technologic advances and improvements in image quality, including the introduction of echo planar imaging (EPI), parallel imaging, multichannel coils and high amplitude gradients.

Accurate lesion detection and characterization is essential for treatment planning for patients with primary or secondary liver tumors, especially in selecting patients who may undergo liver resection, loco-regional or systemic therapies^4,5. DWI can be employed for focal liver lesion detection and characterization, for the assessment of tumor response, and for evaluation of diffuse liver disease³. This sequence is easily incorporated into routine clinical protocols, especially given that DWI is a non-contrast technique and may be performed either before or after contrast administration. DWI can be acquired rapidly within a breath hold, and it provides both qualitative and quantitative information as an adjunct to conventional sequences. Quantification of apparent diffusion coefficient (ADC) has enabled differentiation of solid cellular lesions from cystic or necrotic lesions⁶. ADC quantification has also demonstrated promising results for predicting tumor response to therapy^7,8. The purpose of this chapter is to review the basic principles of DWI, discuss protocol optimization, summarize the performance of DWI compared to conventional sequences, discuss the role of DWI in evaluation of treatment response and examine the potential role of DWI in the assessment of liver fibrosis.

DWI TECHNIQUE

1. Concepts

1.1. Principles of molecular diffusion

Diffusion describes random molecular motion occurring in tissues as a result of thermally activated translations of atoms and molecules. The resulting motion can be described as a stochastic process, with a Gaussian probability distribution given as:

$$P(r,t)~e^{-\frac{r^2}{4Dt}}$$ Eq. 1

where D is the diffusion coefficient and r is the distance traveled by the diffusing molecule during time t. D depends on the species undergoing diffusion and on the medium in which diffusion occurs. For water self-diffusion at 37° C, D = 3.0 mm²/s ⁹. Larger molecules tend to undergo slower diffusion resulting in lower D, as is also the case of diffusion in fluids of higher viscosity. The Gaussian distribution in Eq. 1 describes free diffusion of molecules in an infinite medium. An important variation of this concept is restricted diffusion where molecular motion is restrained within hard boundaries. Restricted diffusion, which is common in tissues, generally leads to lower D and non-Gaussian distributions.

1.2. DWI physics

Nuclear magnetic resonance offers an novel way to measure diffusion, via the application of magnetic field gradients. If such gradients are deployed in pulses of opposed polarity (using a gradient or spin echo sequence), moving spins undergo dephasing while static spins show a null phase at the echo time. By considering the effect of a stochastic diffusion process on the magnetic signal, one can derive the signal attenuation resulting from the application of pulsed gradients in a spin echo experiment¹⁰:

$$S=S_0{e}^{-bD}$$

with S₀ the signal in the absence of gradient and b a function of the applied gradient:

$$b={(\gamma G\delta )}^2\left(\mathrm{\Delta }-\frac{\delta }{3}\right)$$

with γ the spin gyromagnetic ratio, G and δ the gradient strength and length, and Δ the time separating the gradient pair. The factor b, called b-value, determines the strength of the diffusion weighing. The diffusion gradients can be inserted in an imaging experiment as a preparation module, in order to provide additional contrast to the MRI signal and to estimate molecular diffusion in different organs and tissues.

1.3. Quantification of diffusion properties in tissues

In its simplest form, the diffusion experiment involves acquiring two sets of images, one at low or zero b-value, another at high b-value, to derive a voxel-wise diffusion coefficient. Since multiple tissues and compartments may be present in a single voxel, the derived coefficient is referred to as an ADC that reflects a sum of exponential decays rather than a single, pure diffusion constant. In addition to the ADC, which assumes monoexponential decay of diffusion signal, there are models that account for the more complex properties of tissues:

• Flowing blood contributes to diffusion signal and leads to measureable effects on the diffusion decay. The intravoxel incoherent motion (IVIM) approach¹¹ integrates these effects in a biexponential model where a faster decaying exponential, reflecting perfusion effects at low b-values, can be separated from slower exponential decay reflecting true water diffusion. This approach is especially appropriate in highly perfused organs such as the liver^12–18.

• Because of restricted diffusion in tissues, the Gaussian model of Eq. 1 is not valid anymore, and diffusion weighting has a less trivial form. A successful approach has been diffusion kurtosis imaging (DKI) that evaluates restricted diffusion by analyzing the non-Gaussian diffusion distribution, using an additional constant derived from acquisitions at higher b-value¹⁹. Recently, DKI has been investigated in liver explant studies, where results have been correlated with hepatocellular carcinoma (HCC) tumor cellularity²⁰.

• Biological tissues can be anisotropic, and therefore so can diffusion decay. The ADC measured using different diffusion gradient directions may differ in an anisotropic sample, therefore diffusion tensor imaging²¹ (DTI) has been proposed in order to extract direction-specific information using at least 6 different diffusion gradient directions. In the liver, acquisition protocols typically involve acquiring the diffusion trace, which is an average of all 3 principal diffusion directions. DTI has been investigated in abdominal organs^22–25, but there is limited data on anisotropic properties in the liver ^26–28.

1.4. Quality control

There are sources of variability for diffusion quantification across different platforms or centers. First, hardware variations exist between different platforms, leading to different estimates for ADC. Sasaki et al²⁹ reported up to 9% variability in brain ADC in a multicenter study in 12 healthy subjects. Chenevert et al³⁰ measured the ADC of a temperature-controlled ice-water sample in several platforms and reported 5% variability across 3 vendors at 1.5T and 3T. Another source for variability is the choice of b-values used to derive the ADC. Previous studies in liver DWI have used a b-value range of 0–1000 s/mm². Because of confounding effects of perfusion, computing the ADC using a lower range of b-values (e.g. 0–400) may lead to higher values than using an extended range of b-values (e.g. 0–1000), and to different accuracy of the technique for disease detection³¹. According to Xing et al³², the ADC can be reliably estimated using 2 b-values chosen outside of the perfusion range, with magnitude such that b_high−b_low~1.1/D. In order to measure perfusion effects using the IVIM model, more b-values are needed and the b-values distributions must be appropriately chosen to sample the fast pseudo-diffusion (b<100) and slower true diffusion (100<b<1000) decays^33,34.

1.5. Reproducibility and repeatability

Reproducibility and repeatability are important concepts for quantitative diffusion evaluation, particularly for longitudinal and multicentre studies. In the liver, Braithwaite et al³⁵ reported mean coefficient of variation (CV=standard deviation divided by mean) of 14%. In hepatic tumors³⁶, similar results were reported for short term reproducibility (95% limits of agreement: 30%), and ADC reproducibility was found to depend on lesion location (lower reproducibility in the left lobe) and size (lower reproducibility in smaller lesions). The reproducibility of liver IVIM was also reported^13,14,17, showing higher reproducibility for D and ADC (CV <20%), compared to perfusion parameters PF (or f) and D* (or Dt) (CV >20%). IVIM reproducibility was further investigated in lesions in a recent study showing lower reproducibility of IVIM parameters in liver metastases¹³ and HCC³⁷ compared to liver parenchyma. These findings may be due to higher variability of physiological processes in HCC, or to limitations of DWI for measurement of finite size lesions (lesions DWI signal is more sensitive to motion artifacts and to limited resolution).

2. Diffusion acquisition

2.1. Imaging strategy

The most widely used strategy for DWI is echo planar imaging (EPI), which allows acquisition of a full slice in a single shot. A typical protocol involves using fat-saturated single shot diffusion weighted EPI played in an interleaved multislice fashion to allow volume coverage. However, the EPI readout is also subject to ghosting and susceptibility artifacts³⁸. Some alternative techniques to EPI include:

• Segmented EPI³⁹, which is performed by distributing the EPI acquisition over repeated cycles. This way, susceptibility artifacts are reduced by decreasing the EPI readout length. Compared to single shot EPI, segmented EPI is more sensitive to motion occurring between consecutive shots, leading to signal inconsistency and artifacts.

• Turbo spin echo⁴⁰ (TSE) and steady state free precession⁴¹ (SSFP), which may offer reduced B₀ and susceptibility artifacts compared to EPI readouts. However this usually leads to increased echo trains, and to higher power deposition of the radiofrequency pulse train.

• PROPELLER⁴², a modified segmented EPI technique where successive segments are acquired in a radial fashion. Compared to segmented EPI, PROPELLER may be more immune to motion artifacts, with options for robust motion correction⁴³.

2.2. Control of physiologic motion

Physiologic motion is inherent to any liver imaging protocol, with breathing and cardiac motion resulting in subject and acquisition-dependent imaging artifacts. Single shot EPI is robust to motion because its acquisition time is faster than physiological processes. However, in a typical DWI acquisition with multiple b-values and signal averaging, residual fluctuations exist between successively acquired EPI images. Either breath hold (BH), free breathing (FB) or respiratory triggering (RT) technique may be employed. BH results in the shortest image acquisition time, with a limitation in the number of b-values that can be used. FB protocols typically result in blurred images. RT acquisition (using navigator echoes or bellows) may improve DWI data quality^14,36,44,45, at the cost of increased imaging time. Cardiac motion artifacts usually appear in the left liver lobe as signal loss at high b-value, as a result of strong dephasing of coherently moving spins under the influence of diffusion gradients. This can be overcome by performing DWI acquisitions at diastole using an ECG or pulse trigger⁴⁶, but this approach results in significantly increased scan time. Another promising approach to mitigate signal loss due to cardiac motion is to use motion-compensated diffusion gradients^47,48 to cancel the dephasing of coherently moving tissues while maintaining diffusion-weighting.

2.3. Parallel imaging

Using multiple receiver radiofrequency coils, k-space acquisition can be accelerated by skipping lines and reconstructing the image using well established algorithms^49–51. Applied to single shot EPI, parallel imaging allows for shorter EPI train, hence lower susceptibility artifacts and achievable higher resolution⁵². In order to avoid residual artifacts and severe signal loss, acceleration factors are typically limited to 3, depending on the number of coil elements. Another more recent application of parallel imaging is multiband imaging that allows for simultaneous excitation and acquisition of multiple slices^53,54. Multiband imaging leads to increased slice coverage compared to regular DWI acquisition, with minimal signal degradation.

2.4. Effect of magnetic field strength: 1.5 vs. 3T

Because of the strong signal attenuation imposed by diffusion gradients, DWI suffers from low signal levels, as quantified by the signal to noise ratio (SNR). Low SNR, especially at high b-values, can be overcome by signal averaging or by increased static magnetic field. In this regard, a magnetic field strength of 3T might result in up to 2-fold SNR increase compared to 1.5T (4-fold reduction in signal averaging). However, Rozenkrantz et al⁵⁵ reported worse image quality for abdominal DWI at 3T when compared to 1.5T. This study was using a first generation 3T system. Physical limitations at 3T may explain the moderate advantages of 3T DWI. First, higher EPI distortions may arise from stronger B₀ and susceptibility variations. Second, the radiofrequency wavelength at 3T, of the same order as the human body, results in inhomogeneity of the excitation field B₁. This in turn may lead to large signal variations across the liver and ghosting artifacts due to incomplete fat saturation. In addition to image quality degradation, previous reports have shown significant differences in ADC of abdominal organs at 1.5T and 3T⁵⁶. Fig. 1 illustrates DWI artifacts at 3T compared to 1.5T.

Fig. 1.

Example of increased SS EPI diffusion artifacts at 3T (A–B) compared to 1.5T (C–D) in the same patients imaged with both systems (b400 images are shown). Image distortion on coronal SS EPI image due to higher B0 inhomogeneity (A) is observed in a 62 y male patient with HCV and fibrosis. Signal drop due to B1 field inhomogeneity at 3T (arrow on B) is observed in a 24 y female healthy subject, compared to 1.5T (D).

LIVER APPLICATIONS

1. Lesion Detection

Detection of malignant liver lesions is predicated on the belief that tumors have increased cellularity compared to background liver parenchyma and, consequentially, higher signal intensity on DWI. DWI is increasingly employed in oncologic imaging for detection of liver metastases, HCC and cholangiocarcinoma, the application of which will be discussed in detail below. Low b-value diffusion images using SS EPI (i.e. b <100) are most valuable for lesion detection given the suppression of vascular flow, producing “black blood” images, resulting in improved conspicuity for lesions located near vessels^57,58. The suppression of intrahepatic vascular flow also avoids confusing a small lesion with a vessel on the T2WI sequence⁵⁷. Compared to high b-value images (i.e. b1000), low b-value images have relatively higher contrast-to-noise and signal-to-noise ratios^58–61. Low b-value images are also less affected by signal loss due to cardiac motion and geometric distortions induced by eddy currents⁶². Prior work has compared DWI lesion detection rates with conventional sequences, including T2 weighted images and post-contrast T1 weighted imaging (CE T1WI) using extracellular gadolinium based contrast agents (GBCAs), liver specific GBCAs and non-GBCAs liver specific agents, as discussed below. Studies comparing DWI to superparamagnetic iron oxide (SPIO) or mangafodipir trisodium (MnDPDP) will not be discussed in this review as these agents are either no longer available or not widely used.

1.1. Liver metastases

The accurate diagnosis of liver metastases is critical for treatment planning and for identifying patients suitable for liver resection, locoregional therapy or systemic chemotherapy^63–65. MRI provides valuable and accurate information regarding lesion size, segmental location and distribution of liver metastases^4,66. DWI has outperformed T2WI in several studies for liver lesion detection (including metastases), with DWI detection rates as high as 71–87.7% versus 55.6–70.1% for T2WI^57,67–71. A current challenge remains the ability to detect small lesions, especially metastases <2 cm in size. In a prospective study of 129 lesions including 40 metastases, Coenegrachts et al demonstrated that 100% of metastases < 1 cm (40/40) were detected on low b-value DWI (b=20). Similarly, Parikh et al showed excellent metastatic lesion detection rate of 78.5% (57/72) for DWI compared to 45.8% (33/72) for T2WI for lesions 1–2 cm in size. In a retrospective series of 118 hepatic metastases, Bruegel et al reported DWI detection rate of 88–91% versus 45–62% for T2WI, a result that was also impressive for small lesions <1 cm, with DWI detection rate of 85% versus 26–44% for T2WI sequences⁶⁸. The superior detection rates of optimized RT DWI may be explained by improved image quality compared to T2WI with a reduction in eddy current induced artifacts and image blurring⁶. Furthermore, diffusion images also have improved background fat saturation, suppression of intrahepatic vascular flow and liver to lesion contrast compared to T2WI^58,67.

Routine liver MRI protocols typically include post-contrast T1WI for the detection of liver metastases. Both extracellular and combined extracellular-hepatocyte specific GBCAs such as gadoxetic acid (Primovist/Eovist, Bayer Healthcare) are widely used. DWI is not significantly affected by gadoxetic acid injection, and in order to minimize table time, DWI may be obtained between the dynamic and hepatobiliary (HBP) phases⁷². Results comparing DWI with dynamic CE T1WI and/or HBP for the detection of liver metastases have been mixed with most studies evaluating hypovascular metastases. Overall, DWI performs reasonably well compared to CE-MRI, with equivocal or slightly lower detection rates reported in most series. A recent meta-analysis by Wu et al examined the overall diagnostic value of DWI in detecting liver metastases and to compared its performance with CE-MRI⁶². The authors examined 11 studies (9 retrospective, 2 prospective) with 537 patients using either a variety of GBCAs or liver specific agents for comparison with DWI. These results show no major difference in the performance of DWI compared to either extra-cellular GBCAs or combined GBCAs^{68,70,73–81}. DWI used in combination with CE-MRI or HBP imaging demonstrates higher sensitivity compared to DWI alone, with pooled sensitivities across multiple studies of 97% for the combined set versus 86% for DWI alone⁶². The sensitivity of DWI was statistically equivalent when compared to CE-MRI in several recent studies (66.3–84.4% vs. 76.0–79.7%, respectively)^70,75,79. Subgroup analysis in several of these studies also showed that DWI is highly accurate in detecting small lesions. For example, Lowenthal et al demonstrated 92% sensitivity of DWI for detecting lesions <1 cm⁸⁰. In a recent retrospective study of 144 pathologically proven metastases, the authors noted that assessment of DWI was very important in identifying small lesions, which may be missed or misinterpreted as peripheral vessels or imaging artifacts on HBP⁸². Interestingly, per-lesion analysis in the study by Hardie et al⁷⁵ revealed a trend towards improved DWI detection rate for the less experienced observer for both lesions <1 cm and >1 cm, without reaching statistical significance. This finding was thought to be due to the simplicity of interpretation of DWI (Fig. 2, 3). Recent publications have evaluated DWI for detection of hypervascular liver metastases. DWI outperformed CE-MRI in a retrospective series of 41 patients with 162 hypervascular neuroendocrine liver metastases, with sensitivities of 71.6%–71.0% versus 47.5%–48.1% (P= 0.001)⁶⁹. Further work is necessary to compare of the role of DWI for hypervascular and hypovascular liver metastases.

Summary: DWI is highly sensitive for detection of liver metastases, especially for small lesions < 2 cm. Combined DWI and CE T1WI yields the best performance for lesion detection compared to each sequence alone. In addition, DWI is a reasonable alternative to CE MRI in patients who cannot receive GBCAs.

Fig. 2.

51-year-old male with pancreatic cancer and liver metastases. Axial RT SS EPI DWI at 3T with b values of 50 and 800 s/mm² (A, B) and ADC map (C), FSE T2WI (E) and axial late venous phase GRE T1-WI after injection of extracellular contrast medium (E). Both liver metastases (arrows) and the primary pancreatic tumor with secondary pancreatic duct distention (dashed arrow) are identified on b50 images (A), and remain hyperintense on high b value image (B). ADC value of the largest confluent metastasis was 0.97 × 10⁻³ mm²/s.

Fig. 3.

80 year old male with diffuse metastatic melanoma. Axial T1 in-phase (A) and opposed phase images (B), axial RT SS EPI DWI with b value of 1000 s/mm² (C) and ADC map (D). Innumerable non-fat containing T1 hyperintense lesions are present throughout the liver, the largest confluent lesion measuring 5.2 cm in the caudate lobe (arrow). Lesions demonstrate diffusion restriction with low signal on ADC map (ADC value 0.87 × 10⁻³ mm²/s).

1.2. Hepatocellular carcinoma (HCC)

HCC is the most common primary liver cancer and the third most common cause of cancer related death in the United States⁸³. Early diagnosis is essential given improved prognosis of early stage HCC and advances in surgical and locoregional therapies (LRT)⁸⁴. The diagnosis of HCC is made at CE-CT or CE-MRI for lesions > 1 cm that demonstrate arterial hyperenhancement and portal venous/late venous phase washout⁵. Of note, the sensitivity for detecting small HCC <2 cm is considerably lower using CE-CT or CE-MRI^85,86. Several authors have investigated DWI as a tool to improve noninvasive HCC detection. DWI is moderately sensitive for the detection of HCC with detection rates ranging from 45–91.2%^71,87–89. The wide range of reported sensitivities likely reflects differences in study technique and selection bias, and thus, comparisons of reported results are somewhat challenging. Studies with very high reported sensitivities included either larger HCCs or cases with a high pre-test probability of HCC. Overall, DWI was better for HCC detection using low b value DWI (b0, b50) compared to conventional breath hold T2WI sequences (80.5% vs. 54%, p <0.001) and especially for small HCCs < 2 cm (79% vs. 46%, p<0.001)⁷¹. Excellent detection rate of HCCs < 2 cm of 91.2% was shown in a series of 58 HCCs using high b value DWI compared to post-contrast and T2WI⁸⁸. Overall, DWI has shown mixed results compared to CE-MRI. In the study by Piana et al, there was better sensitivity for HCC detection using DWI compared to CE T1WI in 91 patients with 109 HCCs greater than 1 cm in size (72.5–81.7% vs. 59.6–59.6%)⁸⁵. In contrast, CE-MRI was superior for HCC detection in 52 patients with 72 HCCs in the only published liver explant DWI correlation study (87.9% vs. 75.8%, p=0.02)⁹⁰. In another study, DWI and gadoxetic acid images showed equivalent pooled sensitivity for HCC detection (87.1% vs. 91.1%, p > 0.05) in 130 patients with 179 surgically confirmed HCCs in a separate series⁹¹. Fig. 4 demonstrates the classic imaging characteristics of HCC with DWI and CE-MRI. Finally, published data has shown that the combination of DWI, CE-MRI and/or HBP images yields the best sensitivity for HCC detection compared to either sequence alone^90,91. Sensitivities as high as 97.9% may be achieved using this the combination of sequences⁹². The added benefit of the combination of sequences is that the DWI and CE MRI provide complimentary information: DWI is particularly useful in identifying lesions near vessels and CE MRI is useful for identifying subcapsular lesions or lesions in the left lobe. DWI may also be very helpful in detecting infiltrative HCC, which often does not demonstrate typical arterial hypervascularity and delayed wash-out appearance (Fig. 5)⁹³.

Fig. 4.

66-year-old male with HCC. Axial RT SS EPI DWI at 1.5T with b values of 50 and 1000 s/mm² (A,B) and ADC map (C), contrast-enhanced (using extracellular contrast medium) T1WI on arterial phase (D) and late venous phase (E). There is a left hepatic lobe HCC (arrow) which is hyperintense on b50 (A) and b1000 (B) images with low ADC (1.31 × 10⁻³ mm²/s). The lesion demonstrates typical hypervascular hyperenhancement (D) with delayed phase washout and pseudocapsular enhancement (E).

Fig. 5.

59-year-old female with infiltrative HCC invading the portal vein. Axial free breathing SS EPI DWI with b values of 50 and 1000 s/mm² (A, B) and ADC map (C), fat suppressed FSE T2WI (D) and contrast-enhanced (using extracellular contrast medium) T1WI at portal venous phase (E). Tumor thrombus (arrows) demonstrates T2 hyperintensity and restricted diffusion on both b50 (A) and b1000 (B) images with low corresponding ADC value 1.45 × 10⁻³ mm²/s. The extent of tumor is better delineated on the DWI and T2 images (arrows, A–D) compared to post-contrast image (E).

There are several possible explanations for the lower DWI detection rate for HCC. A recognized limitation of DWI sequences for HCC detection is that not all HCCs are hyperintense. Approximately 8.8–38.9% of HCC are either isointense or hypointense on DWI sequences, and smaller lesions (<2 cm) are frequently inconspicuous (Fig. 6) ^85,94,95. Well-differentiated HCC may be more frequently isointense than HCCs with higher tumor grade, however visual analysis of DW images alone is not reliable for this distinction^94,96. The conspicuity and signal intensity of an HCC in a cirrhotic liver ultimately depend on a complex interaction between biologic and technical factors. HCCs are also more difficult to identify in the background of cirrhosis as the distorted liver parenchyma demonstrates diffusion restriction in relation with background fibrosis^97,98. A retrospective study by Kim et al showed a trend towards decreased sensitivity for HCC detection with increasing severity of liver cirrhosis for two observers (Child-Pugh A 93.9–95.6%, Child-Pugh B 77.1–83.0%, Child-Pugh C 60.0–60.6%)⁹⁹. Based on the data presented and the established benefits and limitations of each individual sequences, the optimal MRI protocol for detecting HCC in the cirrhotic liver should include the combination of DWI and CE/HBP MRI, thus yielding a maximized diagnostic accuracy.

Summary: DWI demonstrates moderate sensitivity for HCC detection. DWI should be used in combination with conventional MRI sequences, including post-contrast sequences, for HCC detection. Currently, DWI in isolation is insufficient for detecting HCC in cirrhosis.

Fig. 6.

63-year-old male with hepatitis C cirrhosis and HCC. Axial arterial phase (A), portal venous phase (B), delayed hepatobiliary phase (C), axial free breathing SS EPI DWI with b values of 50 (D) and 500 s/mm² (E) and ADC map (F) after injection of gadoxetic acid. 2.4 cm left hepatic lobe lesion (arrow) is hypervascular with portal venous phase washout/delayed pseudocapsule enhancement and demonstrates hyperintensity on delayed hepatobiliary phase images. The lesion is isointense on both low and high b-value DWI images and ADC map, with ADC value of 1.49 × 10⁻³ mm²/s. A well-differentiated HCC was found at histopathology.

1.3. Cholangiocarcinoma

Intrahepatic cholangiocarcinoma (ICC) is the second most common primary malignant tumor of the liver and is increasing in incidence¹⁰⁰. Similar to HCC, ICC may arise in the context of hepatitis C and chronic liver disease¹⁰¹. There are three types of ICC classified based on morphology and growth pattern, including mass forming, periductal infiltrating type, and intraductal type¹⁰². The mass forming is the most common type and must be differentiated from HCC given the differences in treatment and prognosis. Recent reports have shown that small ICCs (< 3 cm) may have similar imaging appearance and enhancement pattern as HCC^103,104. There is a paucity of data evaluating DWI imaging features and detection rates for ICC. Very few ICCs were included the previous work comparing DWI to T2 and CE-MRI. A recent study by Park et al describes a “target appearance” of ICC on DWI: peripheral hyperintense rim with a central area of hypointensity on high b-value images¹⁰⁵. The central areas of hypointensity corresponded to fibrosis on histopathology. This finding was noted in 75% of ICCs (24/32) versus 3.1% of HCC (1/32) and may be a reliable imaging feature to distinguish small ICC from HCC. Further work is necessary to evaluate the role of DWI in the detection of ICC given its growing incidence and relatively poor prognosis.

Summary: There is limited data regarding the sensitivity of DWI for ICC detection. DWI may be of value in detecting ICC and shows a characteristic morphologic “target” appearance which should be verified in further studies.

2. Lesion characterization

2.1. Qualitative assessment

The noninvasive characterization of liver lesions without using GBCAs is increasingly attractive in the era of nephrogenic systemic fibrosis (NSF)¹⁰⁶. Visual assessment of liver lesion morphology, signal intensity and changes in signal intensity at increasing b-values on DW images is useful for lesion characterization⁶. The changes in signal intensity observed on DW images over a range of b-values depend on true lesion water diffusivity, vascular microperfusion and T2 relaxation time. It is essential to separate true diffusion restriction from “T2 shine through”. “T2 shine through” is the phenomenon present in lesions with long T2 relaxation times, such as cysts or gallbladder content, causing persistent hyperintensity on high b-value images and corresponding high ADC values; thus, the high signal observed is due to the long T2 of the lesion contents (fluid)³. The observed behavior of a lesion on qualitative images must therefore be interpreted along with the corresponding ADC map, which is the graphical representation of the ratio of the diffusion-weighted signal intensities¹⁰⁷. Mean ADC values (mm²/s) are obtained directly from the ADC map by placing a region of interest (ROI). ADC maps are calculated with a linear regression analysis of the function. The ROI used must be large enough to for sufficient SNR and to avoid the effects of volume averaging.

Previous work suggested that it is possible to characterize liver lesions as benign or malignant using DWI sequences^{71,106,108–111}. Liver lesions are considered benign when demonstrating hyperintensity on low b-value DW images, attenuation of signal on higher b-value DWI images (b >500), and a corresponding high ADC value. Lesions are considered malignant when remaining hyperintense on higher b-value DW with a low corresponding ADC value^71,108 (Fig. 7). Performing visual assessment of lesion characterization has been shown to be highly accurate^71,108,112. In a study of 185 focal liver lesions including 76 metastases and 11 HCCs, Holzapfel et al. demonstrated accuracy of 93% for characterizing small lesions <1 cm¹¹³. However, the majority of the benign lesions were cysts or hemangiomas. Of note, limitations of these reports include the relatively small sample sizes and limited types of liver lesions evaluated¹⁰⁶. DWI may provide added benefit to conventional sequences in the characterization of indeterminate lesions in the cirrhotic liver. For example, distinguishing well-differentiated HCC from benign cirrhotic nodules remains challenging in many cases. In a recent study with histopathologic correlation, high b-value DWI showed higher accuracy (79%) than did hypointensity on HBP for distinguishing well-differentiated HCCs from benign cirrhotic nodules¹¹⁴. DWI was also useful in distinguishing hypervascular HCC from hypervascular pseudolesions, as no pseudolesions were hyperintense at DWI in a series by Motosugi et al¹¹⁵. In addition, indeterminate hypovascular lesions with diffusion restriction often go on to develop hypervascularity and overt imaging features of HCC; thus DWI may be a strong predictor for progression to hypervascular HCC¹¹⁶.

Summary: Assessment of qualitative DW images allows for characterization of liver lesions as benign or malignant with a high degree of accuracy. DWI may play an added role to conventional sequences in characterizing indeterminate lesions in the cirrhotic liver.

Fig. 7.

58 year old female with metastatic colon cancer and hepatic cyst. Axial respiratory-triggered SS EPI DWI at 1.5T with b values of 0, 50, 500 and 800 s/mm² (A–D) and ADC map (E). Both lesions are easily detected on b50 images (B) with excellent suppression of intrahepatic vascular flow. The liver metastasis (solid arrow) remains hyperintense on high b value with low ADC (1.1 × 10⁻³ mm²/s). The cyst (dashed arrow) shows strong signal drop on intermediate b-value images and is not seen on high b value with high ADC (3.2 × 10⁻³ mm²/s).

2.2. Quantitative assessment

The ADC has been investigated as a tool to characterize liver lesions. As described previously in this text, the ADC is calculated from the diffusion sequence using multiple (at least 2) b-values. ADC quantification requires minimum acceptable SNR at higher b values and a minimum lesion size of 1.5–2 times the in-plane resolution in order to avoid partial volume averaging⁶. ADC images must be interpreted in conjunction with qualitative DW images and other sequences. Extensive prior work has shown that benign lesions generally have a significantly higher ADC value compared to malignant lesions^{71,108,110,111,117–119}. Cysts and hemangiomas have statistically higher mean ADC values compared to all solid liver lesions (benign and malignant)¹⁰⁶. In a large study of 542 lesions in 382 patients, the mean ADC values of specific liver lesions were as follows (measured as ×10⁻³ mm²/s): cysts (3.40 ± 0.48), hemangiomas (2.26 ± 0.70), FNHs (1.79 ± 0.39), hepatocellular adenomas (1.49 ± 0.39), metastases (1.50 ± 0.65), and HCC (1.54 ± 0.44)¹⁰⁶. ADC cutoff values between 1.4–1.6 × 10⁻³ mm²/s have been reported in the literature for diagnosing malignant liver lesions, with reported sensitivities of 57.1–100% and specificities of 77–100% (Fig. 8) ^{71,106,108,110,118}. Of note, these ADC cutoff values vary depending on the patient population studied and DWI acquisition parameters. There is significant overlap in ADC measurement between solid benign and malignant lesions, thus making this distinction with DWI challenging. There is no statistically significant difference in reported ADC values for FNH, hepatocellular adenomas, metastases and HCC^106,120,121. There is also a paucity of literature examining differences in ADC value for HCC, dysplastic (DN) and regenerative nodules (RN). A report by Xu et al did find that the ADC values of HCC were significantly lower than DN (1.28 × 10⁻³ mm²/s vs. 1.53 × 10⁻³ mm²/s, respectively, p=0.0082) in a series of 54 patients with 40 HCCs and 19 DNs, with histopathologic confirmation¹²².

Summary: ADC quantification can be helpful in separating liver cysts and hemangiomas from malignant liver lesions. However, DWI with ADC quantification cannot reliably make the distinction between solid benign and malignant lesions, and between different malignant lesions.

Fig. 8.

79-year-old female with subcapsular ovarian cancer metastasis. Axial respiratory triggered SS EPI DWI at 1.5T with b values of 50 and 1000 s/mm2 (A, B) and ADC map (C), and fat suppressed FSE T2WI (D). There is a metastatic lesion that is T2 hyperintense with diffusion restriction. ADC value was 1.0 × 10⁻³ mm²/s.

2.3. Common pitfalls in using DWI for lesion characterization

There are several important limitations to DWI characterization of focal liver lesions. As stated, DWI and ADC characterization cannot distinguish between benign and malignant solid liver lesions given overlap in signal intensity and ADC values (Fig. 9). Furthermore, malignant lesions may be heterogeneous and may contain cystic, mucinous or necrotic components (Fig. 10). The T2 prolongation of these complex components results in elevated ADC values, allowing for mischaracterization of a malignant lesion as benign (Fig. 11)^3,6. In addition, the ADC of liver abscesses can demonstrate significant overlap with both hemangiomas and solid lesions (Fig. 12)¹⁰⁶. Thus, it is essential to rely on clinical history in addition to the imaging features on conventional sequences in order to correctly diagnose inhomogeneous metastasis and hepatic abscess.

Fig. 9.

29-year-old male with biopsy-proven necrotizing granuloma. Axial RT SS EPI DWI with b values of 50 and 1000 s/mm² (A, B) and ADC map (C), fat suppressed FSE T2WI (D) and gadobutrol enhanced GRE T1WI on arterial and equilibrium phases (E, F). There is a T2-weighted hypointense, enhancing lesion in segment 2 with foci of restricted diffusion and adjacent capsular retraction (arrow). ADC value was 1.19 × 10⁻³ mm²/s.

Fig. 10.

61 year old female with cystic metastasis (arrow). Axial T2 SSFSE (A), breath hold axial SS EPI DWI with b 50, 100, 500 (B, C, D) and ADC map (E). There is a 1.2 cm cystic lesion in segment 6 which is markedly hyperintense on T2 SSFSE. The lesion is hyperintense on low b value image with attenuation of signal on high b value images and high ADC signal (ADC value 3.4 × 10⁻³ mm²/s) (arrow). Imaging characteristics are consistent with a benign cyst, however this lesion represented a cystic neuroendocrine metastasis.

Fig. 11.

57 year old male with indeterminate liver lesion. There is a 6.9 cm segment 6 lesion (arrow) which is hyperintense on axial HASTE (A), FB SS EPI DWI b50 (B) and b500 (C) with corresponding high signal on ADC map (D). ADC value measured as 1.6 × 10⁻³ mm²/s. The central hypointense area represents susceptibility artifact from intra-lesional hemorrhage (dashed arrow) with corresponding “T2 black out” on DWI images. Given atypical findings on post-contrast T1WI (not shown), lesion was considered indeterminate and biopsy revealed epitheloid angiosarcoma.

Fig. 12.

67-year-old male status post chemotherapy for diffuse large B-cell lymphoma. Axial respiratory triggered SS EPI DWI with b values of 50 and 800 s/mm² (A, B) and ADC map (C), fat suppressed FSE T2WI (D) and gadobutrol enhanced GRE T1WI on late venous phase (E). A thick-walled T2 hyperintense lesion is present in the right hepatic lobe (arrow) demonstrating peripheral rim enhancement (E) and centrally restricted diffusion on both b 50 (A) and b750 (B). FNA guided biopsy of the lesion revealed pus, compatible with hepatic abscess.

Issues related to image quality may also affect interpretation of DWI and ADC. Patients with chronic liver disease and cirrhosis may have concomitant hepatic iron deposition, which may cause a significant T2 shortening effect in the liver that may result in decreased SNR on DW images, most pronounced on higher b values images, with falsely decreased ADC^123–125. For example, ADC values were noted to be significantly lower in patients with hepatic siderosis in a series of 52 patients with cirrhosis¹²⁴. Finally, interpretation of absolute ADC quantification has yet to be clarified as multiple factors influence ADC value (i.e. instrumental, sequencing, biologic). DW images are often of limited quality with relatively poor SNR, spatial resolution, and EPI related artifacts. DWI is still an imaging technique requiring image optimization to ensure consistent high quality performance¹²⁶.

Summary: Qualitative and quantitative information provided by DWI sequence is best used for characterizing cystic/necrotic lesions and highly cellular solid lesions. One must be aware of potential pitfalls and limitations when interpreting DW images and ADC maps.

3. Tumor treatment response

DWI has been investigated as a tool to evaluate tumor response to therapy and potentially predict which lesions will respond to treatment, by providing information regarding tumor viability, cellularity and vascularity¹²⁷. Changes in DWI signal intensity and ADC following treatment can precede changes in lesion size or enhancement and reflect tumor necrosis¹²⁸. The ability to detect early tumor treatment response or lack of response has significant impact in clinical management and treatment planning. Making this distinction would enable repeat treatment or alternative therapy if necessary. Furthermore, given the significant impact and adverse effects of current treatments, including chemotherapy and LRT, knowledge of how likely a lesion is to respond to therapy will also facilitate clinical decision-making in terms of how aggressively to pursue treatment⁸.

Metastatic lesions to the liver have shown a significant increase in ADC values following systemic chemotherapy that precedes changes in lesion size^129,130. Interestingly, a recent study by Marugami et al showed significant overlap in ADC values between responding and non-responding colorectal metastatic lesions following intra-arterial chemotherapy. Using a threshold change in ADC value of 3.5%, ROC analysis showed higher sensitivity and specificity values for % min ADC (100% and 92.6%, respectively) than for % mean ADC (66.7% and 74.1%, respectively) to identify responding lesions¹³¹. The % min ADC indicated the most diffusion restricting voxel within the ROI. Furthermore, liver metastases with low pre-treatment ADC values have responded better to systemic chemotherapy compared to lesions with high pre-treatment ADC values in several series. High pre-treatment ADC value of liver metastases is therefore a potential predictor of poor response to chemotherapy^8,129,132. Metastatic lesions with higher ADC values may in fact have diminished blood supply and consequentially, more baseline cellular hypoxia and necrosis, which may limit the effectiveness of therapy. Another hypothesis to explain poor treatment response in lesions with high ADC value is the variation in local immune response: local host immune response results in increased tissue cellularity and cellular swelling, resulting in a lower ADC⁸. Fig. 13 demonstrates tumor recurrence in a patient with colon cancer who underwent wedge resection of the liver.

Fig. 13.

51 year old male status post partial right wedge resection for metastatic colon cancer with tumor recurrence. Axial respiratory-triggered SS EPI DWI with b value of 500 s/mm2 (A), axial breath-hold fat suppressed T2WI (B), and axial fat-suppressed GRE T1WI after gadoxetic acid injection at portal venous phase (C) show a hypovascular metastatic lesion at the resection margin (arrow), which is hyperintense on DWI and T2WI.

The development of LRT for HCC, including thermal radiofrequency ablation (RFA), trans-arterial chemoembolization (TACE), trans-arterial radioembolization (TARE), and external beam radiation, has revolutionized HCC treatment in non-operative patients or patients with unresectable disease. Lesions treated with RFA or TACE typically undergo coagulative hemorrhagic necrosis with heterogeneous hyperintensity on unenhanced T1WI, making evaluation of CE-MRI challenging¹³³. Several authors have investigated DWI to evaluate HCC response to TACE and TARE^{7,128,134–138}. LRT causes disruption of cellular membranes, cell death and tumor necrosis with the resultant increase in water diffusivity. An increase in ADC values has been reported following TACE in the early post-treatment period (few days up to 2 weeks) with measurable differences before and after treatment (Fig. 14)^7,128,135. ADC values also increase post-TARE according to several preliminary series, but treatment effect has been noted at 1–3 months post-treatment ^137,139,140. ADC has demonstrated a significant correlation with tumor necrosis, especially complete tumor necrosis assessed with histopathology¹⁴¹.

Fig. 14.

66 year old male with hepatitis C related cirrhosis and HCC. 3.2 cm HCC in segment 8 (arrow). Pre-treatment MRI (top) and post-treatment MRI (bottom row). Respiratory-triggered SS EPI DWI at b50, 1000 (A, B) and ADC (C) and arterial phase CE T1WI (gadobutrol) (D) demonstrate HCC in the right lobe with restricted diffusion (ADC value of 1.0 × 10⁻³ mm²/s). Patient underwent right hepatic lobe Y-90 radio-embolization. Follow-up MRI using free breathing SS EPI DWI at b50, b1000 (E, F) and ADC (G) demonstrate a significant increase in ADC value of 1.7 × 10⁻³ mm²/s. Post-contrast subtracted (H) image demonstrates complete interval necrosis.

Furthermore, studies have demonstrated differences in ADC values between viable and necrotic portions of HCCs^7,141. In theory, the viable portion of the tumor restricts diffusion (hyperintense with low ADC value), whereas the necrotic portions will show relatively un-impeded water diffusion (hyperintense with high ADC value)¹⁴². Fig. 15 demonstrates an example of a patient who underwent TACE and RFA to an index HCC with partial necrosis, in additional to additional tumors. Recent publications have also investigated the role of pre-treatment ADC value in predicting response to TACE, with discordant results. Mannelli et al showed that HCCs with low pre-TACE ADC tended to be less responsive to TACE than those with higher ADC, whereas Yuan et al showed that HCCs with higher pre-TACE ADC were less responsive than those with lower ADC to TACE^138,143. Thus, DWI and ADC quantification may play a very important role in the non-invasive, non-contrast assessment of post-treatment response and guiding clinical treatment planning in a select population of patients with HCC¹⁴⁴. Pre-LRT ADC may be useful in predicting response to therapy; however large prospective studies are needed.

Summary: DWI is a potential useful non-invasive tool in evaluating response to therapy for liver metastases and HCC. Changes in ADC value precede changes in lesion size or enhancement, and correspond with tumor necrosis. Pre-treatment ADC values may also be of value in predicting response to treatment. Therefore, DWI and ADC quantification may provide early information valuable for treatment planning and clinical decision-making. However, large prospective studies are needed.

Fig. 15.

79-year-old female status post TACE/RFA of left hepatic lobe HCC. Axial free breathing SS EPI DWI with b values of 50 and 500 s/mm² (A,B) and ADC map (C), fat suppressed FSE T2WI (D) and gadoxetate enhanced GRE T1WI on arterial phase (E) and late phase subtraction images (F). A left lobe approximately 50% necrotic treated HCC and adjacent arterial enhancing nodule (E) with venous wash-out (F) are present (arrow). The viable nodule shows T2-weighted hyperintensity (D) and restricted diffusion (A, B) with low ADC value of 0.70 × 10⁻³ mm²/s. There are additional HCCs in the left and caudate lobes (dashed arrows).

4. Liver fibrosis and cirrhosis

Liver fibrosis and cirrhosis affect parenchymal structure and function due to excessive extracellular matrix deposition. These microscopic changes may lead to reduced diffusion because of a larger proportion of macromolecules such as collagen (inducing restricted motion) and may also lead to reduced perfusion. Previous studies^{16,17,98,145,146} have reported correlations between liver ADC and fibrosis stage assessed by histopathology. In most studies, a decrease in ADC was observed with increasing fibrosis stage and in cirrhosis. Lewin et al¹⁴⁵ reported AUCs of 0.79 and 0.92, for detection of stages F2–F4 and F3–F4, using ADC cutoff values of 1.24 and 1.21 × 10⁻³ mm²/s, respectively. Taouli et al⁹⁸ observed AUCs of 0.896 and 0.896 for detection of stages F2–F4 and F3–F4, using cutoff values of 1.54 and 1.53 × 10⁻³ mm²/s. Wang et al¹⁴⁶, in a study comparing DWI and MR elastography (MRE) for liver fibrosis detection, reported AUCs of 0.78–0.88 using ADC for detection of any stage of fibrosis, vs. AUC of 0.92–0.99 for MRE. Beyond ADC measurements, the IVIM technique has been tested recently for liver fibrosis and cirrhosis detection^14,16,17 with variable differences observed in fibrotic liver for both diffusion and perfusion metrics (Fig. 16). Fibrosis is, however not the only source of altered diffusion properties in the non-cirrhotic and cirrhotic liver. The influence of inflammatory response on ADC has been studied in lesions^147,148 and in the liver²⁶ showing restricted diffusion possibly because of increased viscosity. Increased fat content may also lead to reduced ADC¹⁴⁹.

Fig. 16.

Parametric IVIM diffusion maps obtained with respiratory-triggered SS EPI sequence and 16 b-values (0–800) in a 26-year-old healthy woman and a 55-year-old woman with HCV and stage F3 fibrosis. A, E, Coronal single-shot echo-planar image DW image (b200). B, F: true diffusion coefficient (D) map. C, G: perfusion fraction (PF) map. D, H, D* (pseudodiffusion coefficient) map. D and PF maps show lower values in patient with HCV than in healthy subject. There is no visible difference in D* maps. Corresponding parameter values were: for D, 0.81 10⁻³ mm²/s and 1.06 10⁻³ mm²/s for the patient with HCV and the healthy volunteer, respectively; for PF, 5.1% and 9.8%, respectively; and for D*, 50.5 × 10⁻⁵ mm²/s and 49.4 × 10⁻⁵ mm²/s, respectively.

From: Dyvorne HA, Galea N, Nevers T, et al. Diffusion-weighted imaging of the liver with multiple b values: effect of diffusion gradient polarity and breathing acquisition on image quality and intravoxel incoherent motion parameters--a pilot study. Radiology 2013;266: 920–929; with permission.

LIMITATIONS

DWI sequence has several important limitations in terms of SS EPI image quality and ADC reproducibility. The SS EPI DWI sequence is limited given its relatively low spatial resolution and reduced SNR⁶. EPI artifacts including ghosting and distortion can result in image degradation, which is worse at higher field strength and in the left hepatic lobe. Despite gains in SNR, 3T SS EPI images may be further impaired by susceptibility artifacts caused by B1 inhomogeneities and dielectric/conductive artifacts^55,58. Technical improvements including the application of strong gradients, multi-channel coils, high magnetic fields and advanced software platforms are being used to optimize DWI^6,126.

Significant variability in ADC values has been reported, likely as a result of a combination of hardware and biologic factors⁶. The CV of ADC values measured in liver parenchyma ranges from 3.0–16.2% in several reports^17,35,55,56. Several authors have investigated ADC reproducibility for malignant liver lesions. The DWI acquisition technique (BH, RT or FB), ADC calculation method, and selection of b-values all potentially influence the ADC measurement and the reproducibility for malignant liver lesions¹⁵⁰. ADCs calculated from breath-hold DWI were more reproducible than from RT DWI in a recent study by Kim et al¹⁵⁰. An additional recent study reports 95% limits of agreement of repeated ADC measurements of malignant liver lesions of 28.7–31.3% of the mean³⁶. Greater reproducibility of ADC measurement for larger lesions and for lesions in the right hepatic lobe was also reported in this study³⁶, as well as in a recent study using IVIM³⁷. Further improvements and modifications to the DWI sequence and ADC calculation are needed to optimize ADC measurement reproducibility and accuracy, thus enabling reliable comparison of research studies and evaluation post-treatment tumor response. Standardization of imaging technique and ADC calculation methods across imaging platforms may serve to address these issues.

Summary: DWI sequences are limited due to relative poor SNR and EPI related artifacts including ghosting and distortion. ADC quantification is limited by poor measurement reproducibility for both liver parenchyma and lesions. Future DWI sequence improvements and modifications are needed to address these limitations.

FUTURE DIRECTIONS

5.1. IVIM in the clinic

Emerging research has investigated the role of intravoxel incoherent motion (IVIM) for the following potential clinical applications:

5.1.1. Non-invasive detection of liver fibrosis

Given the experience with DWI and monoexponential ADC calculation, preliminary work has shown that IVIM can distinguish the individual contributions of cellular restricted diffusion and tissue perfusion, as defined by true diffusion coefficient (D), perfusion related pseudodiffusion (PF), and pseudodiffusion coefficient (D*)^{16,17,35,151–153}. In a study comparing RT and FB IVIM using 9 b-values, there was a significant decrease in liver true diffusion coefficient (D), perfusion fraction (PF), pseudodiffusion coefficient (D*) and ADC in cirrhotic livers compared to healthy livers¹⁷. Similarly, a recent study using tri-exponential IVIM analysis also demonstrated a significant reduction in D*, D, and PF in cirrhosis¹⁵². In a prospective pilot study by Dyvorne et al evaluating the effect of diffusion gradient polarity and breathing acquisition on image quality, PF and D were reduced in patients with HCV related liver fibrosis¹⁵³. Overall, these findings reflect the physiologic and microscopic cellular changes to the liver with fibrosis and cirrhosis: namely, the decrease in hepatic blood flow (reduction in portal flow) and the increased deposition of proteins and macromolecules (collagen). A current challenge in the role of IVIM for liver fibrosis quantification remains the optimization of image quality and standardization of the technique and inter-platform reproducibility, for which further research is necessary⁴⁴.

5.1.2. Liver lesion characterization

IVIM is being investigated for liver lesion characterization in hopes of addressing the existing challenges for ADC characterization, including the significant overlap in ADC values for benign and malignant lesions and the variability in ADC measurement across imaging platforms. The ability to discriminate and measure diffusion and perfusion contributions may be of value for highly cellular and hypervascular lesions¹⁵¹. A recent study by Yoon et al retrospectively examined 142 patients with 169 focal liver lesions (120 malignant lesions), and found that the diagnostic performance of true diffusion (D) was better than that of ADC_total in the diagnosis of malignancy (AUC 0.971 using a cut off value 1.278 × 10⁻³ mm²/s)¹⁵⁴. In addition, the authors found that the perfusion fraction (PF) and pseudodiffusion coefficient (D*) were significantly higher in hypervascular lesions compared to hypovascular lesions. Microvascular flow in hypervascular lesions thus contaminated the ADC_total value, and therefore, the removal of perfusion contamination may be useful in revealing the increased cellularity in malignant hypervascular lesions thus allowing improved characterization¹⁵⁴. The exciting new application of IVIM is limited by measurement reproducibility, for which further investigation is warranted^37,155.

5.2. Combination and comparison with PET/MRI

Recent technologic advancements have led to the integration of DWI with FDG-PET, a well-established functional modality for various oncologic and non-oncologic applications, allowing for simultaneous data acquisition. PET-MRI has the potential to combine excellent soft tissue contrast resolution, anatomic resolution and functional information with reduction in overall radiation exposure. DWI and PET are predicated upon entirely different fundamental principles and mechanisms, DWI assessing random motion of water and PET assessing cell glucose metabolism within biologic tissues. Maximum standardized uptake value (SUV_max) is a quantitative measure of cellular glucose metabolism and has been down shown to correlate with histopathologic tumor grade and the presence of necrosis^156,157. These modalities can provide complementary information. DWI may be of value for lesions that are associated with low 18F-FDG accumulation and for the assessment of organs with normal high levels of background 18F-FDG accumulation. PET may be of value for lesions close to the heart or spleen¹⁵⁸.

Both DWI and PET can be acquired using FB technique with the ability to perform whole body imaging, multi-planar and 3-dimensional reformats. DWI and PET images are visually similar in appearance. Diagnostic performance of PET-MRI can be improved with acquisition of anatomic MR imaging sequences¹⁵⁹. PET-MRI T1-weighted gradient-echo images and T2-weighted images have more accurate spatial registration than PET/CT images, although the DWI sequence can result in significant image misregistration from EPI-related geometric distortions, which is a potential weakness of the DWI-PET integration¹⁶⁰.

Studies on correlation between PET and DWI are limited, and results of preliminary work have shown an inverse relationship between ADC and SUV_max^161–163. A recent prospective study was the first to examine correlation between SUV and ADC in neoplastic lesions using a simultaneous hybrid PET/MRI system. In this study, there was a significant negative correlation for between lesion SUV ratio (SUV_max/mean liver SUV) and ADC_min (R=−0.46, p<0.04) for non-osseous soft tissue lesions¹⁶⁴. Well-designed prospective studies are needed to provide more insight and experience into the PET-DWI imaging integration, especially in terms of patient tolerance, safety, diagnostic and prognostic capability and cost effectiveness¹⁵⁸. Currently, there are no published clinical studies investigating the role of DWI-PET in liver malignancy.

5.3. New DWI sequences

The SS EPI sequence is the standard pulse sequence for liver MRI given its rapid acquisition time and relative insensitivity to patient motion¹⁶⁵. Given enhancements in gradient systems and the emergence of parallel imaging, additional pulse sequences have been investigated for use in abdominal imaging, including spin-echo, stimulated-echo, fast-spin-echo, and steady-state free-precession sequences¹⁶⁶. The most important alternative sequences used in the abdomen include the single-shot and PROPELLER fast-spin-echo and steady-state free-precession sequences¹⁶⁶. Modifications of imaging parameters including selection of b-values is necessary when using these pulse sequence in order to obtain sufficient image quality and reliable quantification of diffusion parameters¹⁶⁶.

Diffusion tensor imaging (DTI), which employs additional sensitizing gradients to plot the relative degree of diffusion in multiple directions, has been studied in anatomy where diffusion is preferentially restricted in one direction²¹. DTI has been investigated in the central nervous system for evaluation of non-isotropic tissue such as white matter and spinal cord diseases. Limited data is available for DTI in the liver ^25,52. The use of multiple diffusion gradient directions with DTI results in a stronger applied gradient and shorter TE values⁵². Since more dimensions are plotted in DTI compared to conventional DWI, there is a theoretical advantage of more precise ADC calculation and information regarding tissue anisotropy. Preliminary work has investigated DTI in both healthy and fibrotic livers, with mixed results regarding image quality of DTI compared to conventional DWI²⁶. Taouli et al showed no compromise in hepatic ADC values using DTI⁵². Further work is necessary to determine the role of DTI, if any, in the assessment of diffuse liver disease or focal liver lesion.

Summary: Emerging applications of DWI technology including IVIM, DWI/PET integration and new DWI sequences may further expand the role and ability of DWI to accurately detect and characterize diffuse and focal liver disease.

Conclusion

In this review, we have described our current understanding, applications and potential of DWI for the evaluation of focal and diffuse liver disease. DWI has many potential evolving applications, including focal liver lesion detection and characterization, the assessment of tumor response and the evaluation of diffuse liver disease. In our opinion, DWI should be incorporated into routine liver MRI protocols. DWI is best employed in combination with conventional sequences, especially contrast-enhanced sequences, for detection and characterization of malignant liver lesions. DWI currently has a limited role as an isolated technique for lesion characterization. The reader must be cognizant of the potential pitfalls and limitations of this technique. For patients with impaired renal function or GBCA allergy, DWI is a reasonable non-invasive alternative to CE MRI. Further sequence improvements and widespread standardization of technique, b-values and imaging parameters are necessary for DWI to reach its full potential. Future applications including IVIM and DWI/PET MRI show promise and warrant further investigation.

KEY POINTS.

1. DWI (diffusion-weighted imaging) is based upon differences in mobility of water protons in tissues. Single-shot echo planar (SS EPI) DWI sequences are most commonly used in liver imaging.

2. DWI is generally more sensitive than fast spin echo fat suppressed T2-weighted imaging (T2WI) for liver lesion detection. The combination of DWI and CE-T1 weighted imaging (contrast-enhanced T1-WI) is most sensitive for detection of malignant liver lesions.

3. ADC (apparent diffusion coefficient) quantification can be used to characterize liver lesions as cystic/necrotic or solid. However, ADC alone is insufficient for lesion characterization.

4. ADC has potential value in the evaluation of tumor treatment response, with changes in ADC preceding changes in lesion size.

5. ADC quantification and IVIM (intra voxel incoherent motion) DWI have diagnostic value in the non-invasive detection of liver fibrosis and cirrhosis.