Diffusion-weighted imaging of the liver: Current applications

Abstract

Diffusion-weighted imaging (DWI) of the liver can be performed using most commercially available machines and is currently accepted in routine sequence. This sequence has some potential as an imaging biomarker for fibrosis, tumor detection/characterization, and following/predicting therapy. To improve reliability including accuracy and reproducibility, researchers have validated this new technique in terms of image acquisition, data sampling, and analysis. The added value of DWI in contrast-enhanced magnetic resonance imaging was established in the detection of malignant liver lesions. However, some limitations remain in terms of lesion characterization and fibrosis detection. Furthermore, the methodologies of image acquisition and data analysis have been inconsistent. Therefore, researchers should make every effort to not only improve accuracy and reproducibility but also standardize imaging parameters.

Core tip: The current application of diffusion-weighted imaging (DWI) is reviewed. DWI has some potential as an imaging biomarker for fibrosis, tumor detection/characterization, and following/predicting therapy. However, some limitations remain in terms of lesion characterization and fibrosis detection. To improve reliability including accuracy and reproducibility, researchers have validated this new technique in terms of image acquisition, data sampling, and analysis.

INTRODUCTION

Diffusion-weighted imaging (DWI) is an imaging method that allows the mapping of the free diffusion of water molecules which reflects the structural differences in disease by restricting diffusion. DWI can be added to the routine examination easily using recently available machines. This imaging method has a good ability to detect liver lesions, and quantitative evaluation can be achieved without contrast media. Therefore, DWI does not require considerations for patients having contrast media allergy and the risk of nephrogenic systemic fibrosis due to renal dysfunction[1].

FUNDAMENTALS AND TECHNIQUES

Theory

When assuming free water, water molecules spread three-dimensionally with time and temperature dependence by Brownian motion. It is represented by the Einstein-Smoluchowski formula: <r²> = 6Dt, D = μK_BT, where r is the average distance, D is the diffusion coefficient, t is time, μ is mobility, K_B is Boltzmann’s constant, and T is the absolute temperature. This spread follows the Gaussian distribution called free diffusion.

Stejskal and Tanner previously measured the diffusion coefficient along with their theory using a binary magnetic field gradient by the spin-echo method[2]. At present, DWI acquisition is commonly performed with a Spin-Echo echo planar imaging (EPI) sequence. Water molecule movement was impeded by the cell membrane, interstitial space, and macromolecules. The movement did not follow the Gaussian probability distribution. When D (diffusion coefficient) is small or time “t” is short, the measured D is the same as that of free diffusion because water molecules rarely interact with barrier structures. On the other hand, there is a high probability of the movement being affected by a barrier structure when time “t” is greater, which causes the measured D to become smaller than that of free diffusion. This state is referred to as restricted diffusion.

High cellularity, distortion of the extracellular space, and density of the hydrophobic cell membrane within the tissue restrict diffusion. In contrast, an intravoxel microvessel which travels disorderly behaves similarly to a diffusion phenomenon. As mentioned above, DWI enables not only pure diffusion but also microvessel perfusion. Therefore, the diffusion coefficient is designated comprehensively as apparent diffusion coefficient (ADC).

As the b-value increases on DWI, the signal decreases in tissues composed chiefly of large diffusion components such as free water owing to phase dispersion, and thus the contrast to tissues that restrict diffusion becomes more clear. b-value is defined by the following equation[2]: b (s/mm²) = - γ²‧G²‧δ²(Δ - δ/3), where γ is the gyromagnetic ratio, G is the diffusion gradient amplitude, δ is the gradient diffusion length, and Δ is the diffusion time.

ADC is calculated using the following formula: S_b/S₀ = exp (- b‧ADC),where S_b and S₀ are the signal intensity with and without the application of the diffusion gradient, respectively. This formula is a monoexponential model which does not fit with actual measurement. This is the reason why the signal intensity in the voxel is affected by blood microcirculation. Le Bihan et al[3] have proposed the theory of intravoxel incoherent motion (IVIM). They considered blood microcirculation as rapid diffusion, and defined pure molecular diffusion coefficient (D) and pseudodiffusion coefficient (D*). This biexponential model was defined using the following formula when multiple b-values are obtained, from low b-values (< 200 s/mm²) to high b-values (> 200 s/mm²): S_b/S₀ = f × exp-[(D* + D) × b] + (1 - f) × (-D × b), where D is the true diffusion coefficient, D* is the pseudodiffusion coefficient, and f is the perfusion fraction. The IVIM model has been applied to the evaluation of liver fibrosis and tumor characterization[4,5]. However, some controversial issues about IVIM have remained. The poor reproducibility of D* has been reported[6,7]. Selection of a fitting model is also crucial for IVIM parameters, because the choice of the b-value and reproducibility may be closely related to the fitting models[8].

Advance of technology

DWI using parallel imaging allows for a shorter echo time, and it facilitates improvement of the signal-to-noise (SNR) ratio and thus decreasing susceptibility to artifact[9]. Furthermore, distortion, blurring, and off-resonance artifact diminish, and this increases the spatial resolution[10]. ADC measurement using parallel imaging is reliable except for ADC measurement in the left lobe of the liver[11]. The SNR increases at a high field strength system, but there are some concerns about the inferiority of image quality owing to artifact or signal decay by B0/B1 inhomogeneity, T2/T2* shortening, and increasing acoustic gradient noise. However, using parallel imaging offsets these disadvantages[12].

Single shot EPI sequence is sensitized to not only the motion of diffusion but also bulk motion. Therefore, the consideration of respiration and pulsation is important in case of the acquisition of liver images. In image acquisition during breath holding, it is unnecessary to consider respiratory artifact, in contrast to some disadvantages such as low spatial resolution, low SNR, distortion, and ghost artifact. On the other hand, the free breathing (FB) method usually takes a few minutes because of the many acquisition times, and as a result the SNR increases. Moreover, a high spatial resolution can be achieved and thin slices can be obtained[13]. However, the disadvantage of the FB method is that it is less reliable if there is heterogeneity in the lesion owing to the averaging and blurring of the image. The navigator-triggered (NT) acquisition is a method for running the image sequence in accordance with the expiratory phase monitoring the movement of the diaphragm on high-speed imaging systems such as FLASH during FB. The NT technique improves image quality and lesion contrast, and increases SNR. Moreover, it enables accurate ADC measurement[14,15]. Artifact also becomes stronger as b-value increases[15]. In addition, a specific artifact reported as hepatic pseudoanisotropy attributed to performing DWI in the respiratory gating (RT) has been reported[16].

ADC was reported to be affected by SNR, susceptibility artifact, or artifact derived from heart beating or liver motion due to respiration. Although FB tends to scatter signals compared with RT, the ADC does not differ[17]. The SNR on RT is higher than that on BH. The ADC is also slightly higher on RT than on BH[14]. In a comparison between NT and FB, both are reportedly similar in terms of the ADC and IVIM parameters[18].

For ADC reproducibility, RT is superior to BH but inferior to FB in healthy liver parenchyma[19]. Similarly, in a comparison study among multiple breath-hold (MBH), FB, RT, and NT, FB showed the best ADC reproducibility[20]. It should be noted that there were differences in the signal acquisition times among those techniques in these comparison studies[20].

Effects of contrast agent administration

Currently, Gd-EOB-DTPA-enhanced MRI has been widely used for the detection of liver lesions. However, it is necessary to wait for about 20 min for optimal liver parenchymal enhancement[21]. To improve the examination throughput, DWI is undertaken after Gd-EOB-DTPA injection. Gd-EOB-DTPA does not have an effect on ADC[22]. Furthermore, considering the biexponential IVIM model, there were also no effects on D, D*, and PF[23]. Based on these facts, even if DWI is not successful prior to contrast administration, the lesion can be evaluated on the images acquired during the waiting time until the hepatobiliary phase.

Weak feature of DWI

Cardiac motion causes negligible artifact (signal loss) on DWI of the liver. This artifact tends to become emphasized with a higher b-value and is closer to the heart. Thus, the artifact in the left lobe around the lower surface of the heart in particular can make an image particularly obscure[19,24]. The liver-to-background contrast is also changed by the cardiac phase of acquisition; it decreases more at the systolic phase and signal loss is larger in the left lobe[25]. The ADC of the left lobe is higher and its reproducibility is worse compared with the right lobe[26]. Some solutions to reduce the effects of cardiac motion have been proposed. These include the postprocessing method[24] or filtering[27] which corrects the image after signal acquisition or cardiac triggering synchronized with the heart cycle[27,28]. ADC reproducibility was reportedly improved using these methods.

Moreover, susceptibility artifact occurs at the boundary surfaces between the lungs and the liver parenchyma because of magnetic field inhomogeneity[29]. The artifact is observed as a signal loss in the diaphragm or liver.

Peristaltic movement can produce ghost artifact or blurring on abdominal MRI in the pancreas and liver near the intestinal tract[30]. Hyoscine butylbromide suppresses contraction of the smooth muscles in the intestines and it can reduce ghost artifact (peristaltic artifact). Moreover, it can similarly improve the image quality[31]. As hyoscine butylbromide administration can increase the heart rate, it has also been pointed out that the image quality of the subcardiac area in the hepatic left lobe is reduced on visual evaluation. However, there is no observed significant change in ADC[32]. Thus, it is necessary to address all of the challenges associated with DWI of the liver to achieve higher levels of quantitative and qualitative outcomes and to obtain precise assessments.

EVALUATION OF LIVER FIBROSIS

Clinical application

Liver fibrosis is the accumulation of scar tissue resulting from hepatocyte response to chronic inflammation caused by the hepatitis B or C virus and alcohol consumption, among many other causes[33]. Chronic inflammation activates the stellate cells and induces fibrosis of the extracellular matrix (ECM). In this process, molecules such as glycogen, proteoglycan, and other macromolecules accumulate in the ECM, restricting ECM diffusion[34,35]. Fibrosis leads to cirrhosis, portal hypertension after many years, and possibly eventual death. Liver biopsy is a widely accepted procedure for diagnosing and grading liver fibrosis. However, this procedure is associated with major complications in 0.3% and with mortality in 0.018% of patients[36]. Furthermore, because of the heterogeneity of liver fibrosis, sampling errors can also arise[37,38]. Therefore, alternative noninvasive diagnostic methods that can precisely evaluate liver fibrosis are desirable. Because of convenience and repeatability, the usefulness of some diffusion-weighted MRI parameters (e.g., ADC) and IVIM parameters has been evaluated in several studies. DWI enables the evaluation of restricted diffusion caused by collagen fibers accumulated in the ECM in cirrhotic liver[39-41]. In relation to this, it is important to distinguish METAVIR fibrosis stage 3 or 4 from stages 0 to 2 because patients in the F0-2 grades can be cured by treating the underlying liver disease[42].

Evaluating liver fibrosis using ADC

Several studies have shown that ADC decreases as the liver fibrosis grade progresses[40,41,43,44]. Specifically, the diagnostic performance of detecting METAVIR fibrosis grade 3 or 4 was variable and the area under the ROC curve (AUC) was 0.54-0.92. Some studies have concluded that MR elastography was more reliable than DWI[44,45]. Do et al[46] proposed normalized ADC to improve the diagnostic accuracy of DWI. They calculated normalized ADC as the ratio of liver ADC to spleen ADC and reported that the AUC increased from 0.689 to 0.805 using their methods (Table 1).

Table 1.

Detection of fibrosis using diffusion-weighted imaging

	Tesla	Respiratory	Staging	ROI setting	b-value	Diagnostic accuracy of fibrosis F3 or grater
						AUC	Sensitivity	Specificity
Cece et al[91]	1.5	BH	MTAVIR	5 ROIs, Both	0, 500, 1000	0.888	92.9	79.4
Taouli et al[92]	1.5	BH	MTAVIR	4 ROIs, Both	0, 50	0.717	40	100
					0, 300	0.716	50	94.7
					0, 500	0.835	70	85
					0, 700	0.901	66.7	100
					0, 1000	0.832	80	90
					0, 50, 300, 500, 700, 1000	0.896	88.9	80
Kocakoc et al[93]	1.5	BH	Ishak	3 ROIs, Both	100, 600, 1000	0.759	56.5	99.3
Wu et al[47]	3	RT	MTAVIR	5 ROIs, Right	0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 200, 300, 400, 500, 1000	0.684
Chung et al[48]	1.5	RT	MTAVIR	6 ROIs, Right	0, 100, 200, 900	0.768
					0, 30, 60, 100, 150, 200, 900	0.764
					0, 30, 60, 100, 150, 200, 400, 600, 900	0.754	65.5	82.1
Ding et al[94]	1.5	FB	New Inuyama	Whole right lobe	0, 500	0.61	30.4	90.6
Feier et al[43]	3	NA	MTAVIR	1 ROI, Right	50, 300, 600	0.77	81.08	72.5
Fujimoto et al[95]	1.5	NA	MTAVIR	4 ROIs, Right	0, 1000 (entropy ADC)	0.926	87	84
Do et al[46]	1.5	BT	Ludwig	4 ROIs, Right	0, 50, 500 (normalized ADC)	0.689	56	71
Bonekamp et al[96]	1.5	BT	MTAVIR	9 ROIs, Both	0, 750	0.8	83.9	68.5
Wang et al[44]	1.5	NA	MTAVIR	3 ROIs, Right	50, 500, 1000	0.84	88	76
Lewin et al[41]	1.5	RT	MTAVIR	3 ROIs, Right	0, 200, 400, 800	0.92	87	87
Sandrosegaran et al[40]	1.5	BH		2 ROIs, Both	50, 400	0.656	51.7	71.4

BH: Breath-hold; RT: Respiratory gating.

Evaluating liver fibrosis using IVIM parameters

The efficacy of diagnosing liver fibrosis has been reported by Luciani et al[5]. They found that perfusion-related diffusion parameters (D*: Fast component of diffusion, f: Fraction of the diffusion linked to microcirculation) were significantly related to restricted diffusion in a cirrhotic liver, whereas diffusion-related parameters (D: Slow component of diffusion) were not significantly related. Several studies followed after this study[5,45,47-53]. Including the study of Luciani et al[5], 3 studies[49,52,53] only compared cirrhotic liver with healthy volunteer liver but did not evaluate the fibrosis grade. D* was found to be significantly lower in the cirrhotic liver in all studies and D showed a significantly lower value in 2 studies[52,53]. In these 2 studies, the authors adopted relatively more of high b-values and less of low b-values. On the other hand, Chung et al[48] calculated IVIM parameters using 3 patterns of b-value selection to diagnose high-grade liver fibrosis: b = 0, 30, 60, 100, 150, 200, 400, 600, 900 s/mm²; b = 0, 30, 60, 100, 150, 200, 900 s/mm²; and b = 0, 100, 200, 900 s/mm². They suggested that the number of lower b-values was not crucial for diagnosing high-grade liver fibrosis. Girometti et al[50] have suggested that higher b-values may not be necessary for diagnosing liver fibrosis. Supporting these hypotheses, Wu et al[47] suggested that favorable results were given by b-values 0, 20, 40, 60, 80, 100, 150, 200, 400, and 800 s/mm².

Effects of steatosis

Steatosis has been reported to have possible effects on ADC. Poyraz et al have suggested that steatosis decreases ADC because the increased fat content of hepatocytes and the extracellular fat accumulation reduce the interstitial space and restrict water diffusion[39,54]. Other studies have evaluated the effects of fat deposition by DWI using other methods. These studies estimated that fat has several components that broaden the spectrum and mimic T2* decay at short TE ranges; however, the accurate mechanism is unknown[55,56]. Another study mentioned that IVIM parameters, such as diffusion coefficient and perfusion fraction, are not affected by the fat fraction and have the possibility of evaluating liver fibrosis regardless of the fat deposition[57].

Effects of iron deposition

The most widely used sequence for DWI is EPI, which allows acquisition of a full slice in a single shot. However, the EPI readout is also subject to ghosting and susceptibility artifacts, and may decrease ADC as a result of the T2* shortening effect[8,58]. Chronic liver disease may often have iron overload. Therefore, if extremely low ADCs are obtained, iron overload should be considered[59-63].

DETECTION AND CHARACTERIZATION OF LIVER TUMORS

Detection of liver tumors

DWI has a better contrast-to-noise ratio and better conspicuity by suppression of background vessels in low b-values[64]. DWI has a higher detection rate of liver tumors than T2WI[64,65], particularly in detecting malignant lesions[66]. However, the ADC of benign solid lesions has been reported to be similar to that of the liver parenchyma[67]. Therefore, benign solid lesions may be difficult to detect on DWI.

Many studies have reported that DWI has an additional value for detecting liver metastasis in combination with Gd-EOB-DTPA (Table 2); however, this remains controversial in hepatocellular carcinoma (HCC). Some authors have reported no additional value because some well-differentiated HCCs could not be detected on DWI as the major reason[68]. Well-differentiated HCCs include variable pathological characteristics like as early HCCs whose pathology is very similar to the surrounding liver parenchyma, steatosis contained lesion and a hypervascular lesion. Kim et al[69] reported that early HCCs showed hyperintensity on DWI which was strongly associated with their progression to hypervascular HCCs.

Table 2.

Detection of liver tumor in combination with Gd-EOB-DTPA- enhanced magnetic resonance imaging

	Tesla	Respiratory	b-value (× 10-3 s/mm2)	Tumor	Results
Kim et al[97]	3	RT	0, 100, 800	Mets	Combined EOB-MRI and DWI yielded better accuracy and sensitivity
				(Various)
Chung et al[98]	3	FB	50, 400, 800	Mets	Combined EOB-MRI and DWI yielded better accuracy and sensitivity
				(colorectal)
Koh et al[99]	1.5	FB	0, 50, 100, 250, 500, 750	Mets	Combined EOB-MRI and DWI improved detection
				(colorectal)
Löwenthal et al[100]	1.5	BH	0, 500	Mets	DWI can detect small lesions
				(colorectal)
Shimada et al[101]	3	RT	0, 500	Mets	EOB-MRI showed higher accuracy
				(Various)
Donati et al[102]	1.5	BH	0, 150, 500	Mets	No added value of DWI
				(Various)
Kim et al[103]	1.5	RT	0, 50, 600	Mets, HCC	DWI increases sensitivity for detecting Mets
					No added value of DWI for HCC detection

DWI: Diffusion-weighted imaging; BH: Breath-hold; RT: Respiratory gating; FB: Free breathing; HCC: Hepatocellular carcinoma.

Characteristic differentiation of liver tumors (benign vs malignant)

In hypercellular tissue, extracellular water cannot diffuse and this results in a reduction in ADC. A cystic component has few structures to restrict diffusion and this result in a high ADC. Cysts can be distinguished from solid lesions easily. The cut-off ADC was reported to be approximately 2.5 × 10^-3 mm²/s for distinguishing cysts from other solid liver tumors[70]. Hemangioma is also relatively easy to distinguish from malignant lesions. The ADC of hemangioma was reported to be approximately 1.4 × 10^-3 mm²/s. However, some overlaps have been recognized which reduce accuracy in distinguishing metastatic lesions[71]. This is particularly true for mucinous carcinoma from the ovary which mimics colorectal carcinoma (Figure 1). However, tumor characterization was reportedly not dependent on size[72] (Table 3).

Figure 1.

A 65-year-old man with metastatic tumor in the liver from colorectal carcinoma. A: T2-weighted imaging shows an obvious hyperintense lesion on segment VII (arrow); B: DWI (b-value of 800 s/mm²) shows hyperintensity; C: Apparent diffusion coefficient map also shows hyperintensity. This finding mimics that for hemangioma. DWI: Diffusion-weighted imaging.

Table 3.

Characteristic differentiation of liver tumors

	Tesla	b-value	ADC (×10-3 mm2)
			Benign			Malignant
			Cyst	Hemangioma	All	HCC	Mets	All
Goshima et al[104]	1.5	0, 100, 200, 400, 800	3.70 ± 0.9	1.23 ± 0.2		1.08 ± 0.3	0.99 ± 0.5
Battal et al[105]	1.5	0, 800			1.94 ± 0.61			0.86 ± 0.13
Gurtosoyianni et al[106]	1.5	0, 50, 500, 1000	2.55	1.9	2.55	1.38	0.99	1.04
Testa et al[71]	1.5	0, 600	2.4				1
Miller et al[73]	1.5	0, 500	3.40 ± 0.48	2.26 ± 0.70	2.50 ± 0.86	1.54 ± 0.44	1.50 ± 0.65	1.52 ± 0.55
Namimoto et al[107]	1.5	30, 1200	3.05	1.95		0.99	1.15	1.04
Kim et al[108]	1.5	3, 57, 192, 408, 517, 705, 846	2.91 ± 1.51	2.04 ± 1.01	2.49 ± 1.39	0.97 ± 0.31	1.06 ± 0.50	1.01 ± 0.38
Taouli et al[67]	1.5	0, 500	3.63 ± 0.56	2.95 ± 0.67		1.33 ± 0.13	0.94 ± 0.60
Cieszanowsk et al[109]	1.5	50, 400, 800	2.45	1,55	1.86	0.94	1.05	1.07
Bruegel et al[72]	1.5	50, 300, 600	3.02 ± 0.31	1.92 ± 0.34		1.05 ± 0.09	1.22 ± 0.31
Kandpal et al[13]	1.5	0, 500	2.90 ± 0.51	2.36 ± 0.48		1.27 ± 0.42	1.13 ± 0.41
Demir et al[110]	1.5	0, 1000	3.05 ± 0.26	2.46 ± 0.21	2.57 ± 0.26	0.90 ± 0.10	0.79 ± 0.11	0.86 ± 0.11
Oner et al[111]	1.5	0, 500	2.34 ± 0.36	1.72 ± 0.30			1.03 ± 0.24
Holzapfel et al[112]	1.5	50, 300, 600	2.61 ± 0.57	1.69 ± 0.34	2.36 ± 0.62	1.12 ± 0.28	1.08 ± 0.32	1.09 ± 0.30

ADC: Apparent diffusion coefficient; HCC: Hepatocellular carcinoma.

DWI is reportedly not helpful in differentiating focal nodular hyperplasia and adenoma from solid malignant lesions. The mean ADCs of these benign solid lesions were reported as 1.40-1.79 × 10^-3 mm²/s[72,73]. Notably, the ADCs of these benign solid lesions and those of malignant lesions such as HCCs and metastatic tumors overlap (Figure 2).

Figure 2.

A 45-year-old man with focal nodular hyperplasia. A: Hepatobiliary phase on Gd-EOB-DTPA-enhanced MRI shows mainly hyperintensity on the outer layer and hypointensity on the inner layer. These enhancement patterns are typical radiologic findings of focal nodular hyperplasia; B: DWI (b-value of 800 s/mm²) shows hyperintensity; C: ADC map shows heterogeneous hyperintensity. The ADC is 1.40 × 10^-3 mm²/s. Gd-EOB-DTPA-enhanced MRI is more useful for obtaining a precise diagnosis than DWI alone. MRI: Magnetic resonance imaging; DWI: Diffusion-weighted imaging; ADC: Apparent diffusion coefficient.

Histological differentiation of HCC

Preoperative prediction of the histological grade of HCC can facilitate the estimation of prognosis and contribute to the choice of therapy. There is also a higher incidence of recurrence in poorly differentiated HCCs than in well-differentiated and moderately differentiated HCCs[74,75].

Histological grade correlates with cellularity and structural atypia which includes trabecular, pseudoglandular, solid, and scirrhus. As HCC progresses to poorly differentiated HCC, there is increased cellular density, nuclear/cytoplasmic ratio and intracellular organelles; thickened cellular plates; and shrinkage of the extracellular and intracellular spaces. This may lead to restricted diffusion in poorly differentiated HCC. However, the results have been inconsistent[75-79] (Table 4). One of the main reasons for this inconsistency is the region of interest (ROI) setting. Previous studies showed no significant differences in the ROI setting for each histological grade on whole lesions[76,77]. On the other hand, in cases of the ROI set at the lowest ADC and the ROI set to avoid a necrotic or cystic area, a lower ADC was obtained in poorly differentiated HCC[75,78,79].

Table 4.

Histological differentiation of hepatocellular carcinoma using apparent diffusion coefficient

	Tesla	Respiratory	b-value (s/mm2)	Well-diff HCC (×10-3 mm2/s)	Mod diff HCC (×10-3 mm2/s)	Poorly diff HCC (×10-3 mm2/s)	Difference
Saito et al[113]	1.5	RT	100, 800	1.25 ± 0.25	1.12 ± 0.22	1.13 ± 0.23	NS
Nasu et al[114]	1.5	RT	0, 500	1.45 ± 0.35	1.46 ± 0.32	1.36 ± 0.29	NS
Heo et al[77]	1.5	FB	0, 1000	1.2 ± 0.22	1.1 ± 0.10	0.9 ± 0.13	p < w, m
Nakanishi et al[80]	1.5	RT	50, 1000	NA	1.29 ± 0.21	1.07 ± 0.15	p < m
Nishie et al[75]	1.5	RT	0, 500, 1000	1.21 ± 0.11	1.14 ± 0.26	0.76 ± 0.10	p < w, m
Guo et al[79]	3.0	BH	0, 600	1.43 ± 0.09	1.34 ± 0.19	1.16 ± 0.16	p < w, m

RT: Respiratory trigger; BH: Breath holding; FB: Free breathing; ADC: Apparent diffusion coefficient. p < w, m: ADC in poorly differentiated HCC was significantly lower than ADC in well-differentiated HCC and moderately differentiated HCC; p < m: ADC in poorly differentiated HCC was significantly lower than ADC in moderately differentiated HCC; NS: No significant difference in ADC was observed for each histological grade; NA: Not applicable.

The current applications are IVIM and ADC minimum. D shows a better diagnostic performance than ADC in distinguishing high-grade HCC from low-grade HCC[80]. ADC contains combined information on cell density (D) and perfusion (f) (microcirculation). Minimum-spot ADC was reported to be significantly lower in poorly differentiated HCC than in well-differentiated HCC and moderately differentiated HCC[75].

MONITORING OF THERAPY

Transarterial chemoembolization in HCC

Tumor necrosis shows a high intensity on the ADC map, representing free diffusion of water molecules[81]. Therefore, DWI can evaluate the therapeutic outcome of transarterial chemoembolization (TACE). In case of a hypervascular lesion without a definite venous washout, DWI has an advantage compared with dynamic MRI and improves the detection of marginal tumor recurrence[14], although dynamic MRI has a more accurate correlation with histopathological findings in necrosis. TACE-induced perilesional parenchymal changes negatively affect DWI in terms of overall accuracy. On the other hand, Kokabi et al[82] reported that an ADC change 3 h after TACE is an accurate predictor of treatment response and survival.

IVIM and diffusion kurtosis imaging (DKI) are current imaging biomarkers. Specifically, D* predicts lipiodol uptake[83]. DKI is reportedly a more reliable imaging biomarker than ADC[84].

Chemotherapy in liver metastasis

Some studies have reported that ADC could predict the response to chemotherapy in liver metastasis[85,86]. Liang et al reported that pretreatment ADC is significantly lower in responders[87]. In contrast, Koh et al[85] reported that a high pretreatment ADC predicted a poor response. Furthermore, ADC increases 3 or 7 d after chemotherapy in responders. Recently, ADC histogram analysis has shown that the mean, 1^st percentile, 10^th percentile, 50^th percentile, 90^th percentile, and 99^th percentile were significantly lower in the responding group than in the nonresponding group. The reason why ADC in responders is lower is that high cell density tumors are well perfused, resulting in the high delivery and retention of chemotherapeutic drugs.

Sorafenib in HCC

IVIM has been proposed for evaluating the therapeutic outcome of sorafenib[88,89]. D before treatment in responders was found to be higher than D before treatment in nonresponders[88]. This might be due to the tumor histological grade. Sorafenib acts more effectively in low-grade HCCs[90]. D can better distinguish low-grade HCCs from high-grade HCCs[80], and a higher D indicates low-grade HCCs. Lewin et al[89] reported that f increased significantly in responders after 2 wk. This perfusion parameter f increases with normalization of tumor vessels.

CONCLUSION

DWI has potential as an imaging biomarker for fibrosis, tumor detection/characterization, and following/predicting therapy outcome. To improve accuracy and reproducibility, researchers have validated this new technique in terms of image acquisition, data sampling, and analysis. The added value of DWI in contrast-enhanced MRI has been established in the detection of malignant lesions of the liver. However, some limitations remain in terms of lesion characterization and fibrosis detection. Furthermore, the methodologies of image acquisition and data analysis have been inconsistent. Therefore, researchers should make every effort not only to improve accuracy and reproducibility but also to standardize the imaging parameters.