Free-Breathing Hepatobiliary Phase Gd-EOB-DTPA-Enhanced MR Imaging with Radial VIBE Sequence: Comparison with Conventional Cartesian VIBE Sequence

Gadolinium-Ethoxybenzyl-Diethylenetriaminepentaacetic Acid (Gd-EOB-DTPA) has the potential for both dynamic imaging and liver-specific static MR imaging of hepatocytes with accurate delineation and characterization of liver tumors [1-3]. Approximately half of the injected dose is taken-up by hepatocytes reaching a plateau after approximately 20 minutes and lasting for approximately 2 hours. The hepatobiliary phase of Gd-EOB-DTPA-enhanced MR imaging can visualize focal hepatic lesions with great contrast and assess liver function [1-5]. Therefore, in recent liver MR imaging, taking high quality hepatobiliary phase images is very important. However, motion artifacts such as respiratory motion, cardiac pulsation, and bowel peristalsis can degrade the image quality of abdominal MR examinations [6-8].


Introduction
Gadolinium-Ethoxybenzyl-Diethylenetriaminepentaacetic Acid (Gd-EOB-DTPA) has the potential for both dynamic imaging and liver-specific static MR imaging of hepatocytes with accurate delineation and characterization of liver tumors [1][2][3]. Approximately half of the injected dose is taken-up by hepatocytes reaching a plateau after approximately 20 minutes and lasting for approximately 2 hours. The hepatobiliary phase of Gd-EOB-DTPA-enhanced MR imaging can visualize focal hepatic lesions with great contrast and assess liver function [1][2][3][4][5]. Therefore, in recent liver MR imaging, taking high quality hepatobiliary phase images is very important. However, motion artifacts such as respiratory motion, cardiac pulsation, and bowel peristalsis can degrade the image quality of abdominal MR examinations [6][7][8].
The breath-hold fat-saturated three-dimensional (3D) T1-weighted gradient echo sequence is usually selected for hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging because of its ability to image thin slices of the whole liver within a single acquisition and to reduce motion artifacts [6,[9][10][11]. However, in patients with a diminished breath-hold capacity, such as elderly, debilitated, or pediatric patients, motion artifacts degrade image quality and diagnosis can become difficult [12,13].
Recently, radial volumetric interpolated breath-hold examination (rVIBE), which is a modified version of the conventional Cartesian VIBE (cVIBE) sequence, has been developed [8,[14][15][16][17][18]. This technique uses the "stack-of-stars" scheme to acquire the k-space data and certain data consistency constraints can be applied to reduce motion artifacts [13]. Several researchers have reported that free-breathing rVIBE was feasible for abdominal MR imaging, particularly imaging patients who have difficulty holding their breath [8,17,18]. However, there is a paucity in the research on the feasibility of the rVIBE sequence for hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging [12,18]. Moreover, the optimal number of radial views for rVIBE has not been elucidated.
Therefore, the purpose of our study was to assess the feasibility of three-dimensional fat-suppressed T1-weighted gradient-echo sequence with rVIBE compared with that of cVIBE sequence for free-breathing hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging and to investigate the optimal number of radial views for rVIBE.
All MR images were obtained using a 1.5-Tesla clinical system with 18 channel body array and 32 channel spine coils. Dynamic images using 3D fat-suppressed T1-weighted gradient-echo VIBE were obtained before and after the injection of the intravenous contrast agent. In all patients, 0.025 mmol/kg body weight of Gd-EOB-DTPA (Primovist, Bayer Schering Pharma, Berlin, and Germany) was intravenously administered at a flow rate of 1 mL/s, followed by a 40 mL saline solution flush. Twenty minutes after the administration of Gd-EOB-DTPA, free-breathing cVIBE and rVIBE examinations with the radial views of 256 (rVIBE 256 ), 512 (rVIBE 512 ), and 1024 (rVIBE 1024 ) were obtained. Scan parameters were the same as those used in the phantom study, except for the matrix size of 208× 256 and generalized auto calibrating partially parallel acquisition with an acceleration factor of two for cVIBE ( Table 2). Parallel acceleration was not used for rVIBE acquisition as it is part of the intrinsic nature of rVIBE technique.

Phantom study
The phantoms SNRs were 20.3 for cVIBE, 31.9 for rVIBE 256 , 46.0 for rVIBE 512 , and 69.7 for rVIBE 1024 ; hence, the values were higher in every rVIBE than in cVIBE. The SNR for rVIBE increased as the number of radial views increased.  (Figure 2). All rVIBE radial views showed a significant higher SNR than cVIBE (all, P<0.001). SNR of the liver was significantly lower for rVIBE 256 than for rVIBE 512 (P=0.004) and rVIBE 1024 (P<0.001). The rVIBE 1024 had a higher SNR than rVIBE 512 , but the difference was not statistically significant (P=0.122).

Qualitative analysis:
The results of qualitative analyses are shown in Tables 3 and 4. Interobserver agreements were excellent (κ value=0.963 for motion artifact, κ value=0.941 for overall image quality).
The scores of motion artifact were significantly higher for all rVIBE radial views than for cVIBE (all, P<0.001; Figure 3). The rVIBE 256 showed a significant lower score than rVIBE 512 and rVIBE 1024 (both, P<0.001). However, no significant difference was obtained between rVIBE 512 and rVIBE 1024 (P=0.968).
All three rVIBE radial views showed significantly higher scores for overall image quality than cVIBE (all, P<0.001; Figure 4). The mean score of the overall image quality was significantly lower for rVIBE 256 than for rVIBE 512 and rVIBE 1024 (both, P<0.001), and there was no significant difference between rVIBE 512 and rVIBE 1024 (P=0.902).
A representative case that underwent hepatobiliary phase Gd-EOB DTPA-enhanced MR imaging with cVIBE and rVIBE is shown in Figure 5.

Discussion
We sought to determine whether the quality of a free-breathing

Phantom study
Phantom Signal-to-Noise Ratios (SNRs) for cVIBE and all rVIBE radial views were measured using the National Electrical Manufacturers Association method [19] by a radiological technologist (M.S.) who had no knowledge of the sequence parameters. In each sequence, two original phantom images were subtracted, and the subtracted image was obtained. Using the first original image, the signal intensity (SI) was measured by the mean signal intensity in ROI covering approximately 80% of the phantom. The noise was the standard deviation (SD) in the same ROI on the subtracted image. The SNR was calculated using the following equation: SNR=√2×SI/SD sub , where the factor of √2 arises because the SD is derived from the subtraction image and not from the original image.

Clinical study
Quantitative analysis: In all 30 patients, liver signal intensities on hepatobiliary phase Gd-EOB DTPA-enhanced MR images were measured by a radiologist who was blinded to the sequence parameters and clinical information. As shown in Figure 1, ROIs were placed over the lateral, medial, anterosuperior, anterioinferior, posterosuperior, and posteroinferior segments (approximately 100 mm 2 ) of the liver avoiding blood vessels. SNRs were calculated as SI/SD within the ROI and were averaged among six hepatic segments.

Qualitative analysis:
To evaluate the image quality of cVIBE, rVIBE 256 , rVIBE 512 , and rVIBE 1024 , two radiologists (Y.F. and Y.K., with 23 yrs and 14 yrs of experience in abdominal radiology, respectively), blinded to the sequence parameters and clinical information, independently scored the images on a 1-5 scale regarding motion artifact including streak artifact (1: extreme, 2: severe, 3: moderate, 4: mild, 5: none) and overall image quality (1: unacceptable, 2: poor, 3: acceptable, 4: good, 5: excellent), with the higher score representing the more desirable examination. For further analyses, the qualitative scores were averaged between the results of two readers.

Statistical Analysis
SNR of the liver, and the scores of the motion artifact and overall image quality were compared between cVIBE and all rVIBE radial views by using the Steel-Dwass test of post hoc nonparametric multiple comparisons. Statistical analyses were performed using JMP version 9 software (SAS Institute Japan, Tokyo, Japan). Values of P<0.05 were considered to indicate a significant difference.  3D fat-suppressed T1-weighted gradient-echo sequence with rVIBE was sufficient for hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging. Our results revealed that radial k-space sampling in a free-breathing hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging leaded to a reduced motion artifact level, and demonstrated superior image quality compared to Cartesian data acquisition using objective and subjective analyses. Good interobserver agreement for the determination of the best sequence underlines this finding.   In our phantom study, all rVIBE (rVIBE 256 , 31.9; rVIBE 512 , 46.0; and rVIBE 1024 ; 69.7) radial views showed higher SNRs than cVIBE (20.3). There have been no previous reports comparing SNR between cVIBE and rVIBE in a phantom study. Although the reasons for the higher SNR with rVIBE are unclear, the reason can be explained by the repeated acquisition around the center of k-space compared with standard Cartesian k-space readout, and uncorrected higher frequency data in kx-ky because of cylindrical shape of stack-of-stars trajectory [21][22][23].
Our clinical study showed that liver SNRs on hepatobiliary phase Gd-EOB-DTPA-enhanced MR images were significantly higher for Figure 4: Visual score of overall image quality on hepatobiliary phase gadolinium-ethoxybenzyl-diethylenetriaminepentaacetic acid-enhanced MR imaging. All three rVIBE radial views show a significantly higher score for overall image quality than cVIBE (P<0.001). There is a significant difference between rVIBE 256 and rVIBE 512 or rVIBE 1024 (both, P<0.001). Note: cVIBE, conventional Cartesian VIBE; rVIBE 256 , radial VIBE with a radial view number of 256; rVIBE 512 , radial VIBE with a radial view number of 512; rVIBE 1024 , radial VIBE with a radial view number of 1024 all rVIBE radial views than for cVIBE (P<0.001), as observed in our phantom study. Shin et al. [13] reported that rVIBE showed higher liver SNR on gadoteric acid-enhanced MR imaging in pediatric patients. Image quality has been reported to be significantly higher for rVIBE on abdominal MR imaging in pediatric or uncooperative patients, compared with cVIBE [8,17]. In our qualitative analyses, rVIBE reduced motion artifacts and achieved better overall image quality (p<0.001) than cVIBE on hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging. The better overall image quality of rVIBE might be attributable not only to the higher SNR because of a part of the intrinsic nature of the rVIBE technique and no use of parallel-imaging methods, but also to the reduction of widespread motion artifacts that resulted from use of radial k-space sampling. Therefore, our study indicated that rVIBE could be more useful than cVIBE for hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging, particularly in patients with diminished breath-hold capacity.
In our phantom and clinical studies, rVIBE improved SNR when we increased the number of radial views. The rVIBE 256 showed statistically poorer qualitative image quality compared with rVIBE 512 and rVIBE 1024 . With conventional Cartesian k-space sampling, object motion translates into dominant motion artifacts (ghosting) along the phase-encoding direction as well as overall image blurring. Such a vulnerable phase-encoding axis does not exist in the rVIBE radial geometry, and motion artifacts present as streak artifacts with the stack-of-stars scheme where radial sampling is performed in plane [8].
One of the disadvantages of radial k-space sampling is streak artifacts [17]. Kim et al. [24] reported that decreasing the number of radial views leads to an increase in streaking artifacts. Block et al. [25] reported that the rVIBE sequence should be used with radial view numbers of 400-800 when using matrix sizes of 224-384 in free-breathing abdominal MR imaging. In our study using the matrix size of 256 × 256, motion artifacts were remarkable on the image with the radial view number of 256 (rVIBE 256 ), and these artifacts improved when using the radial view numbers of 512 (rVIBE 512 ) or 1024 (rVIBE 1024 ). Therefore, the inferior image quality on rVIBE 256 compared with rVIBE 512 or rVIBE 1024 was considered to be due to the insufficient number of radial views compared with the matrix size, which would cause the streak artifact with radial sampling.
As the number of radial views increased, the data acquisition time became longer, although SNR improved with less motion artifacts. In our study, data acquisition times were 84 s for rVIBE 512 and 167 s for rVIBE 1024 . Moreover, no significant difference between rVIBE 512 and rVIBE 1024 was obtained in quantitative (P=0.122) or qualitative analyses (P=0.968 for motion artifact and 0.902 for overall image quality). A shorter acquisition time would be desirable in a busy clinical setting. Therefore, our study suggested that rVIBE 512 might be more convenient for hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging compared with rVIBE 1024 because of the shorter examination time.
Our study has several limitations. First, our study includes a relatively small number of patients. Nevertheless, our results suggested that rVIBE might have fewer artifacts and higher overall image quality than cVIBE for liver imaging in patients with diminished breath-hold capacity. Second, we did not assess the visualization of liver lesions on hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging with cVIBE or rVIBE, and thus, we cannot definitively state the usefulness of rVIBE for detecting liver tumors. However, we believe that higher SNR and better image quality would yield better visualization of liver tumors.

Conclusion
Our results demonstrated that data acquisition using radial k-space sampling reduced motion artifacts, and thus improved robustness for motion. The radial view number of 512 was considered to be more suitable for hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging when using a matrix size of 256 × 256. The rVIBE acquisitions were longer than cVIBE because parallel-imaging methods, which are widely used with cVIBE, have not been established for rVIBE. However, we believe it is possible to add the rVIBE sequence to the scanning protocol in the setting of non-cooperative patients for hepatobiliary phase Gd-EOB-DTPA-enhanced MR imaging, and the rVIBE sequence could be useful for detection and characterization of liver tumors and evaluation of liver function.