Intravoxel Incoherent Motion Diffusion-weighted MR Imaging of the Liver: Effect of Triggering Methods on Regional Variability and Measurement Repeatability of Quantitative Parameters

Introduction

Qualitative analysis of diffusion-weighted magnetic resonance (MR) images has been increasingly used for the evaluation of various liver diseases. Among the quantitative parameters derived from diffusion-weighted imaging (DWI), the apparent diffusion coefficient (ADC) is used most commonly and has been validated as a potential imaging biomarker for evaluating diffuse liver disease, characterizing focal hepatic lesions, and monitoring the treatment response of malignant tumors (1–4). However, since ADC represents the magnitude of the mobility of water molecules as a whole in a tissue, it may thus not fully account for the tissue characteristics that can be interrogated with DWI techniques (5,6).

On the other hand, intravoxel incoherent motion (IVIM) imaging, which is a method based on DWI with multiple b values, allows for the separate analysis of two components of random water motion in biological tissue—pure molecular diffusion and microcirculation (or perfusion)—with the parameters of pure molecular diffusion coefficient (D), perfusion fraction (f), and perfusion-related diffusion coefficient (D*) (5,7,8). IVIM imaging has recently been used for liver imaging and has been shown to be useful for the evaluation of liver fibrosis, nonalcoholic fatty liver disease, and focal hepatic lesions (9–12).

However, for IVIM imaging to become a valid clinical tool for the evaluation of liver diseases, measurement of the IVIM parameters should be reliable and unbiased by other factors. However, physiological motion, such as respiration and cardiac pulsation, may influence the IVIM parameters in the liver. Previous studies have shown that both respiratory and cardiac motion result in artificial signal intensity loss of the liver on images obtained with DWI, thus leading to ADC measurement errors (13–15). Several previous investigators have also reported the considerable regional variability of ADC values in the liver, which may have resulted from uneven distribution of cardiac motion artifacts in the liver—that is, more severe artifacts in the area of the liver close to the heart (4,14,16). These physiological motion-induced artifacts may also lead to measurement errors and regional variability of IVIM parameters in the liver, which may offset the diagnostic values of IVIM imaging shown in previous studies (9–12). However, no previous investigators have specifically evaluated this issue.

In our study, we hypothesized that the application of optimal triggering methods that synchronize DWI acquisitions with respiratory or cardiac cycles may be able to reduce the physiological motion–induced measurement errors in ADC and IVIM parameters. To prove our hypothesis, we prospectively performed DWI twice by using free breathing (FB) without triggering, electrocardiography (ECG) triggering (ET), and respiratory triggering (RT) methods in healthy volunteers. With this prospective study, we intended to compare the influence of the triggering methods for DWI on the ADC and IVIM parameters in the liver, as well as regional variability and measurement repeatability.

Materials and Methods

Two employees of Siemens (I.S.K. and B.K.) provided technical support for implementation and optimization of DWI sequences in our study. The authors not associated with Siemens (Y.L., S.S.L., N.K., E.K., Y.J.K., S.C.Y., S.H.P., S.Y.K., M.G.L.) maintained full control of the data at all times.

Subjects

Our institutional ethics committee approved the study protocol, and all participants provided written informed consent. From March 2012 to April 2012, 12 healthy volunteers (mean age, 30 years; age range, 27–40 years), including six women (mean age, 29 years; age range, 27–32 years) and six men (mean age, 31 years; age range, 27–40 years) without any history of disease of the abdominal organs, were prospectively enrolled in this study.

DWI Protocols

DWI was performed with a 1.5-T MR imaging system (Magnetom Avanto; Siemens Medical Solutions, Erlangen, Germany) with dedicated six-channel torso array coils by a senior MR technician (Y.J.K., with 10 years of experience in MR examination) under the supervision of a radiologist (Y.L., with 2 years of clinical experience in the interpretation of liver MR images). The maximum gradient specifications were 45 mT/m for amplitude and 200 mT/m/msec for the slew rate. All volunteers underwent DWI of the liver twice on the same day by using the following three methods: FB without triggering, ET, and RT with navigator echo (prospective acquisition correction, or PACE; Siemens). Between the two DWI sessions, the participants were removed from the MR imager and were then placed inside the magnet again. The time interval between two DWI sessions was approximately 5 minutes. Axial diffusion-weighted images were acquired by using the single-shot echo-planar imaging sequence with the parameters shown in Table 1. Instead of bipolar gradients, monopolar diffusion-weighted gradients that allow for imaging with a shorter echo time were used to achieve a higher signal-to-noise ratio and to decrease the susceptibility artifact. An image–based dynamic distortion correction algorithm was used to correct for eddy current–induced distortions. Diffusion-weighted gradients—nine b values of 0, 30, 60, 100, 150, 200, 400, 600, and 900 sec/mm²—were applied in three orthogonal directions and were subsequently averaged. A k-space–based parallel imaging technique (generalized autocalibrating partially parallel acquisition, or GRAPPA; Siemens Medical Solutions) was used with an acceleration factor of two. Fat suppression was achieved by using a chemical-shift–selective fat-suppression technique.

Table 1 DWI Parameters

*Repetition times for ETDWI varied, depending on the subjects’ cardiac cycle.

^†Data are mean acquisition times, with ranges in parentheses.

For ETDWI, the operators adjusted the trigger delay (mean ± standard deviation, 391.5 msec ± 50.6; range, 250–400 msec) and the acquisition window (mean, 444 msec ± 92.5; range, 200–500 msec) according to each subject’s cardiac cycle (mean, 888.9 msec ± 116.8; range, 690–1111 msec) to acquire image data during the subjects’ ventricular diastolic phases—that is, between the end of the T wave and the next QRS complex, while observing the subject’s ECG waves displayed on an ECG monitor embedded in the operator console (Fig 1). When the T wave was not clearly depicted on the ECG monitor, a trigger delay of 400 msec was chosen on the basis of the reported mean interval between the QRS complex and the T wave (14,17). For all FB, RT, and ETDWI, the imaging parameters were identical for the first and second DWI sessions.

Figure 1:

Download as PowerPointOpen in Image Viewer

Image Analysis

DWI data were analyzed quantitatively according to both monoexponential and biexponential IVIM models (7). The ADC was calculated with least-squares monoexponential fitting of all b value data on a pixel-by-pixel basis according to the following equation (6):

For IVIM-based analysis, biexponential fitting was performed with a least-squares fitting algorithm, as follows (7):

All fitting algorithms were implemented with software (Matlab; Mathworks, Natick, Mass), which allowed extraction of the parametric maps of ADC, D, f, and D* on a pixel-by-pixel basis, as well as region of interest (ROI) measurements of the parameters.

One radiologist (Y.L., with 2 years of clinical experience in the interpretation of liver MR images) performed the quantitative image analysis in six sessions (in the order of first FBDWI, first ETDWI, first RTDWI, second FBDWI, second ETDWI, and second RTDWI examinations). For each patient, 15 1.5-cm² circular ROIs were positioned in the liver—that is, three ROIs on each of the upper, middle, and lower sections of the right hepatic lobe and two ROIs on each of the upper, middle, and lower sections of the left hepatic lobe in areas devoid of large intrahepatic vessels and prominent artifacts. The mean ADC, D, f, and, D* values were recorded for each ROI.

The ROIs were placed in locations as similar as possible on six series of diffusion-weighted images. To achieve this, the locations of the ROIs positioned during the first review session were recorded by using screen capture. During the other image review sessions, ROIs were placed with reference to the ROI locations in the first review session and the visual anatomic landmarks, including portal vein and hepatic vein branches (Fig 2).

Figure 2:

Download as PowerPointOpen in Image Viewer

Statistical Analysis

The ADC, D, f, and D* values were organized according to the study subject (each volunteer), ROIs (15 ROIs), anatomic locations (right hepatic lobe, left hepatic lobe, and the mean of both hepatic lobes), and DWI techniques (first and second examinations with FBDWI, ETDWI, and RTDWI). The difference in the mean ADC, D, f, and D* values between the right and left hepatic lobes was evaluated by using the paired t test. The mean ADC, D, f, and D* values for the right hepatic lobe, the left hepatic lobe, and the mean of both hepatic lobes and the magnitude of the difference in the ADC, D, f, and D* values between the right and left hepatic lobes were compared among the three DWI techniques by using repeated-measures analysis of variance with the Bonferroni multiple comparisons test as a post hoc. The variability of the ADC, D, f, and D* values across the 15 ROIs in the liver was assessed by using the coefficient of variation (CV) for each patient and each DWI examination. The means of the CV values across ROIs were compared among the three DWI techniques by using repeated-measures analysis of variance. The test-retest repeatability of ADC, D, f, and D* was analyzed by using the within-subject CV (within-subject CV = within-subject standard deviation divided by the mean of two measurements), where within-subject standard deviation is the root-mean-square mean of the standard deviations of repeated measurements (18). The statistics were recently adopted by the Quantitative Imaging Biomarkers Alliance as the preferred method for assessing the test-retest repeatability of quantitative imaging methods (19). The 95% confidence intervals of the within-subject CV values and the differences in the within-subject CV values among the three DWI techniques were assessed by using the bootstrap method (1000 times replicates) (20). Statistical analysis was performed by using commercial software (IBM SPSS statistics version 21 software, IBM, New York, NY; and SAS software, SAS Institute, Cary, NC). All reported P values were two sided, and P < .05 was considered to indicate a significant difference.

Results

ADC and IVIM Parameters according to Locations in the Liver and Imaging Techniques

Table 2 summarizes the ADC and IVIM parameters measured with FBDWI, ETDWI, and RTDWI. For all three DWI techniques, the ADC, D, and f values measured in the left hepatic lobe were significantly higher than those measured in the right hepatic lobe (P ≤ .018). However, ETDWI resulted in smaller differences in these parameters between the right and left hepatic lobes than FBDWI and RTDWI, with a significant difference noted for the ADC and f values in both the first and second DWI examinations (P ≤ .009) and for D in the first DWI examination (P = .005) (Figs 3, 4). Although there was a tendency to have higher D* values in the right hepatic lobe than in the left hepatic lobe, the difference was significant only for the first RTDWI examination (P = .044) and the second ETDWI examination (P = .023).

Table 2 ADC and IVIM Parameters for the Liver

Note.—Data are means ± standard deviations, unless indicated otherwise.

*P values were obtained from the comparison of the parameter values among the three DWI techniques by using repeated-measures analysis of variance.

^†A significant difference between ETDWI and FBDWI was noted by using the post hoc Bonferroni multiple comparisons test.

^‡A significant difference between ETDWI and RTDWI was noted by using the post hoc Bonferroni multiple comparisons test.

^§Values are given for the right hepatic lobe minus the values in the left hepatic lobe.

^||P values were obtained from the comparison of the parameter values between the right and left lobe for each DWI technique.

^#A significant difference between FBDWI and RTDWI was noted by using the post hoc Bonferroni multiple comparisons test.

Figure 3:

Download as PowerPointOpen in Image Viewer

Figure 4:

Download as PowerPointOpen in Image Viewer

There were significant differences in the ADC, D, and f values among the three DWI techniques. ETDWI resulted in lower ADC, D, and f values for the left hepatic lobe and for the mean of both hepatic lobes, compared with FBDWI and RTDWI, and with a significant difference noted for the ADC in both the first and second DWI examinations (P < .001) and for D only in the first DWI examination (P < .001); the difference in f values was significant between ETDWI and RTDWI in both the first and second DWI examinations (P ≤ .003) and between ETDWI and FBDWI only in the second DWI examination (P ≤ .037). In the right hepatic lobe, there was also a tendency to have lower ADC and D values for ETDWI compared with those of FBDWI and RTDWI, although the significant differences were noted only in the first DWI examination and for the comparison of ADC between ETDWI and FBDWI (P = .01) and between ETDWI and RTDWI (P = .019) and for the comparison of D between ETDWI and FBDWI (P = .004). There was no significant difference in ADC, D, and f values between FBDWI and RTDWI, regardless of the measurement locations.

Regional Variability of ADC and IVIM Parameters

The CV values across ROIs for the variability of the ADC and IVIM parameters in various locations in the liver are summarized in Table 3. For ADC, D, and f, ETDWI resulted in significantly lower CV values across ROIs than FBDWI and RTDWI (P ≤ .049), thus suggesting less variability of these parameters across different locations in the liver. Figure 5 demonstrates the ADC and IVIM parameter values measured in various locations in the liver by using three DWI techniques. For FBDWI and RTDWI, the ADC, D, and f values showed a tendency toward a greater variability in the left hepatic lobe than in the right hepatic lobe, and the highest values were measured in the left-lobe upper section, followed by the left-lobe middle section, the left lower section, and the sections in the right hepatic lobe. A similar tendency was also noted for ETDWI, although it was less obvious than at FBDWI and RTDWI. For all three DWI techniques, a large regional variability of the D* values (range of CV values across ROIs, 57.82%–138.6%) was noted.

Table 3 CV Values for the Variability of the Parameters across Various Locations in the Liver

Note.—Data are mean CV ± standard deviation, unless indicated otherwise.

*P values were obtained with repeated-measures analysis of variance for the three DWI techniques.

^†Post hoc Bonferroni multiple comparisons test yielded significantly lower CV values for ETDWI compared with FBDWI and RTDWI.

Figure 5:

Download as PowerPointOpen in Image Viewer

Test-Retest Measurement Repeatability of ADC and IVIM Parameters

Table 4 summarizes the test-retest measurement repeatability of ADC and IVIM parameters. For all three DWI techniques, the strongest within-subject CV was noted for ADC (2.33%–6.96%), followed by D (3.19%–16.17%) and f (7.60%–16.81%). The measurement repeatability of D* was the worst among the ADC and IVIM parameters, with the within-subject CV values ranging from 57.05% to 156.61%. ETDWI showed a tendency toward better repeatability for the measurement of ADC, D, and f, compared with FBDWI and RTDWI, especially in the left hepatic lobe. The measurement repeatability of D in the left hepatic lobe for ETDWI was significantly better than for RTDWI (P = .023) and showed marginally significant improvement over FBDWI (P = .087). For the measurement of f, ETDWI resulted in significantly better repeatability than FBDWI, regardless of the measurement location (P ≤ .032). Compared with FBDWI, RTDWI led to significantly better measurement repeatability of ADC in the right hepatic lobe (P = .022) and f in the left hepatic lobe (P = .023). For FBDWI and RTDWI, the measurement repeatability of D and D* tended to be poorer in the left hepatic lobe than in the right hepatic lobe, whereas this tendency was less noticeable at ETDWI.

Table 4 Test-Retest Measurement Repeatability of the ADC and IVIM Parameters

Note.—Numbers in parentheses are 95% confidence intervals for the within-subject CVs.

*First, second, and third P values are from the pairwise comparison of the within-subject CV between FBDWI and ETDWI, between FBDWI and RTDWI, and between ETDWI and RTDWI, respectively.

^†P ≤ .05.

Discussion

Our study demonstrates the significant influence of triggering methods for DWI and measurement locations on the ADC and IVIM parameters in the liver. In our study, the ADC, D, and f values measured in the left hepatic lobe were larger than those measured in the right hepatic lobe, and the ADC, D, and f values measured by using ETDWI tended to be lower than those measured with FBDWI and RTDWI, especially in the left hepatic lobe. We believe that these results are mainly attributable to the cardiac motion–induced artifacts, which leads to a false overestimation of the ADC and IVIM parameters, based on the following findings in our study. First, the use of ETDWI to acquire image data during the diastolic phases significantly reduced the magnitude of the difference in these parameter values between the right and left hepatic lobes, more so than FBDWI and RTDWI. Second, the ADC, D, and f values for FBDWI and RTDWI had an obvious tendency to be higher when measured in the areas of the liver that are close to the heart and may thus be more prone to cardiac motion effect—that is, the upper sections of the left hepatic lobe—than the areas of the liver that are further from the heart, such as the right hepatic lobe. However, this regional variability in ADC, D, and f values was less in ETDWI. A previous study also demonstrated the effect of cardiac motion on the liver signal intensity on diffusion-weighted images (14). Kwee et al found that diffusion-weighted images (b = 500 sec/mm²) obtained during the cardiac systolic phase showed lower liver-to-background contrast than those obtained during the cardiac diastolic phase (14). Taken together, these findings indicate that the cardiac motion–induced signal intensity loss seen on diffusion-weighted images results in a false overestimation of the ADC, D, and f values of the liver, especially in the left hepatic lobe.

Our study results indicate that the application of ET to DWI can reduce cardiac motion–induced measurement error in the ADC and IVIM parameters in the liver. In our study, ETDWI resulted in more consistent ADC, D, and f values in the right and left hepatic lobes and a smaller variability in these parameters across different locations in the liver, as well as improvement in the measurement repeatability of D and f, especially in the left hepatic lobe, as compared with FBDWI and RTDWI. Consistent with our findings, a previous study by Mürtz et al also demonstrated the advantage of the pulse-triggering method for reducing motion artifacts on diffusion-weighted images and improving the repeatability of the ADC measurements in abdominal organs (21). On the other hand, the advantage of RT for reliable measurement of ADC and IVIM parameters was not as obvious as ET. Although RTDWI resulted in improved measurement repeatability of ADC in the right hepatic lobe and f in the left hepatic lobe, it did not decrease the regional variability of the ADC and IVIM parameters compared with FBDWI.

On the basis of our results, we suggest that ETDWI is the most preferred technique for DWI of the liver to achieve less biased and more repeatable measurements of ADC and IVIM parameters. If DWI of the liver is performed without ET, the effect of the cardiac motion artifacts on the ADC and IVIM parameters should be taken into consideration to ensure proper measurement and interpretation of these parameters. Specifically, given the lower repeatability of D and D* and the overestimation of the ADC, D, and f values in the left hepatic lobe for FBDWI and RTDWI seen in our study, the right hepatic lobe would be a better location for measurement of ADC and IVIM parameters in liver parenchyma. In addition, when the ADC and IVIM parameters are used for evaluating focal hepatic lesions, it should be noted that the ADC, D, and f values for lesions in the left hepatic lobe may be artificially overestimated and therefore may not be comparable to those for lesions in the right hepatic lobe.

We found that the measurement of IVIM parameters is overall less repeatable than the measurement of ADC. Of all IVIM parameters, D* showed the worst measurement repeatability, with within-subject CV values ranging from 57.1% to 156.6%. Similar to our results, a previous study also showed the poor measurement repeatability of D* for both liver parenchyma and metastases (22). Despite the potential value of D* for assessing perfusion changes seen in liver cirrhosis and nonalcoholic fatty liver disease, as reported in previous studies (10,23), the high measurement error of D* shown in our study suggests that this parameter may not be a reliable imaging biomarker unless future improvement in the data analysis method or the DWI technique allows a precise estimation of D*. There have been several approaches attempted in previous research studies to overcome or circumvent the problems related to the unreliability of D* measurement. Since the precise estimation of D* requires an unrealistically higher signal-to-noise ratio than is available with most clinical abdominal DWI methods, as shown by Pekar et al (24), some researchers discarded D* and used a predetermined, fixed D* value for the data-fitting process to minimize fitting errors stemming from inaccurate estimation of D* (25,26). On the other hand, some researchers have used Bayesian method–based analysis, which accounts for measurement uncertainty, and they reported improved stability and accuracy in estimating IVIM parameters in relation to those seen by using the least-squares fitting method, which is the most commonly used algorithm for IVIM analysis (27,28). There have also been several segmented IVIM analysis procedures proposed in previously published studies (6,10,12,29). Unlike the fully unconstrained, biexponential fitting procedure used in our study, the segmented IVIM analysis procedures are used to calculate the IVIM parameters in a step-by-step manner to improve the mathematic stability by decreasing the degree of freedom of the fitting models. However, since the theoretical advantages of these approaches have not yet been proven, further studies are warranted to determine the most robust and reliable data analysis algorithm.

Our study has limitations. First, for accurate assessment of measurement repeatability, multiple repeated examinations, rather than two repeated examinations, would have been a more desirable study design. In our study, however, we chose to perform two repeated examinations by using three DWI techniques rather than performing multiple repetitions by using a single technique to compare the measurement repeatability of the IVIM parameters in the three DWI techniques. Second, our study did not include breath-hold DWI or combined RT and ETDWI. Although the inclusion of these two DWI techniques would have allowed complete understanding of the influence of respiratory and cardiac motion on IVIM imaging, these techniques may not be practical methods for IVIM imaging of the liver. Breath-hold DWI has limitations in its signal-to-noise ratio and the number of b values because of the limited breath-hold time and therefore has not been used for IVIM imaging of the liver (10–12,22,25,26,28–30). Despite the potential advantage of the combined RT and ET method for suppressing both respiratory and cardiac motion artifacts, combined RT and ETDWI would require a long imaging time, which may decrease its clinical applicability. Third, as all DWI in our study was performed by using the same MR unit from a single vendor, our results do not account for the potential variability of the ADC and IVIM parameters, which depended on the coil system, image systems, and field strengths, and which could potentially influence the reproducibility of these parameters in actual clinical practice. Finally, although we acquired diffusion-weighted images by using nine b values, the use of more b values, especially low b values (0–50 sec/mm), would have been desirable for more accurate estimation of IVIM parameters (31).

In conclusion, cardiac motion artifacts lead to the overestimation of ADC, D, and f in the liver, especially in the left hepatic lobe. ETDWI is a more effective technique for decreasing the regional variability of ADC and IVIM parameters than FBDWI or RTDWI, and it may improve measurement repeatability by reducing cardiac motion–induced measurement error. The measurement repeatability of IVIM parameters is worse than that of ADC, with that of D* being the worst among the IVIM parameters. Further improvement in the IVIM analysis algorithms and the DWI technique is warranted for the successful clinical application of IVIM imaging of the liver.

Author Contributions

Author contributions: Guarantors of integrity of entire study, S.S.L., S.C.Y.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, Y.L., S.S.L., S.H.P., S.Y.K., M.G.L.; clinical studies, S.S.L., Y.J.K., M.G.L.; experimental studies, Y.L., N.K., E.K., I.S.K., S.Y.K.; statistical analysis, Y.L., S.S.L., S.C.Y.; and manuscript editing, Y.L., S.S.L., B.K., I.S.K., S.H.P.