Introduction

Based on the substantial advances in image quality over the past decade, it is generally accepted that contrast-enhanced dynamic liver magnetic resonance imaging (MRI) outperforms computed tomography in the diagnosis of hepatocellular carcinoma (HCC) [1,2,3]. Initially, the HCC diagnostic criteria was based on extracellular contrast agent (ECA)-enhanced imaging [4]. But then gadoxetic acid (Gd-EOB-DTPA, Primovist [Eovist in the USA], Bayer) opened a new horizon for HCC diagnosis as it offers high sensitivity for HCC detection by providing with hepatobiliary phase (HBP) [5, 6]. However, several limitations in liver MRI remained unsolved: one the most important issue is the long scan time required for imaging that often cannot be achieved within a single breath-hold, now and then causing respiratory motion artifacts. This served as a major reason for poor image quality especially for dynamic contrast-enhanced sequences.

Recently, Akai et al reported that gadoxetic acid increased the respiratory rate rapidly, and the effect on respiration tended to be larger than gadoteridol and gadopentetate dimeglumine [7]. The ghosting artifacts during the arterial phase (AP) of gadoxetic acid-enhanced MRI (EOB-MRI) is reported to range from 4.8 to 19%, leading to suboptimal or nondiagnostic imaging [8,9,10]. Several strategies have been proposed to overcome such artifacts, including shortened scanning time [8, 11], lowered contrast injection rate [12, 13], contrast dilution using saline [14], modified breath-holding or extended breathing method [15, 16], and radial acquisition technique [17]. This is an important issue because the AP is essential for detecting HCC [18]. In addition, since gadoxetic acid is formulated in half the volume (0.1 mL/kg) containing a quarter of the concentration of gadolinium, it is difficult to catch the arterial bolus during the optimum late arterial phase for HCC detection [19].

Multiple APs could increase the probability of capturing the optimal instance of arterial hyperenhancement of HCC [20]. In terms of artifacts, multiple APs might be beneficial for minimizing the ghosting artifact during the AP of EOB-MRI [8]. As a more innovative approach, a recently introduced novel method, known as GRASP (golden-angle radial sparse parallel), for rapid acquisition of data using compressed sensing [21], parallel imaging, and a golden-angle acquisition scheme could potentially overcome the aforementioned limitations in AP imaging [22,23,24,25]. This approach employs a non-Cartesian “stack-of-stars” trajectory to acquire volumetric k-space data. The radial sampling is performed using the golden-angle radial sampling scheme, where the angle of the radial lines is increased continuously by 111.25°, by which a rather uniform coverage of k-space with high temporal incoherence is obtained for any arbitrary number of consecutive lines [26, 27]. This enables dynamic MRI using continuous data acquisition and retrospective reconstruction of an image series with arbitrary temporal resolution by grouping different numbers of consecutive radial lines into temporal frames [28, 29]. The pseudo-golden-angle radial acquisition scheme (PGRA), similar to GRASP, employs a more flexible angle between consecutive lines rather than 111.25°, to provide evenly distributed k-space radial lines for any arbitrary number of projections and in turn reduces streaky artifact and scan time [30].

A few studies have applied GRASP to liver MRI [23,24,25,26, 31]. However, no study has applied a golden-angle acquisition scheme to liver MRI with regard to HCC evaluation. The goal of this prospective study was to determine the feasibility of ECA-enhanced multiphasic liver MRI using PGRA comparing it with EOB-MRI using a conventional Cartesian approach in terms of image quality and the diagnostic performance of HCC.

Materials and methods

Patients

This prospective study was approved by our institutional review board and followed the Declaration of Helsinki and subsequent amendments. Written informed consent was obtained from each patient before enrollment in the study. Between November 2017 and February 2018, we recruited adults (≥ 18 years old) with chronic hepatitis or cirrhosis who had been referred to Samsung Medical Center, Seoul, Korea, an academic referral institution, due to a suspected hepatic tumor on ultrasound. The inclusion criteria for this crossover intra-individual comparison of the two MRIs were (a) patients with a first-time diagnosis of a hepatic nodule on an initial MRI and (b) patients who were scheduled to undergo liver surgery. Since our institutional policy prefers EOB-MRI over ECA-MRI for patients suspected of having HCC, most patients had undergone EOB-MRI as the first MR examination, followed by ECA-MRI. We excluded patients who had refused consent (N = 5), and those who had a longer than 1-month time interval between the two MR examinations (N = 6). We finally enrolled 15 patients (14 men, 1 woman; age range, 41–69 years; mean, 57.9 years). The two MR examinations were performed within a 1-month interval (median, 15 days; range, 10 to 22 days). Reference standards for the diagnosis of HCCs were based on histopathological examination of surgical specimens.

MR examination

MRI was acquired using a 3.0-T MR system (Intera Achieva 3.0-T, Philips Healthcare) equipped with 32-channel phased-array receiver coil. Baseline images included a T1-weighted turbo field-echo in-phase and opposed sequence, with breath-hold multishot T2-weighted imaging and respiratory-triggered heavily T2-weighted imaging. Diffusion weighted imaging was acquired using respiratory-triggered single-shot echo planar imaging with a b value of 0, 100, and 800 s/mm2. For EOB-MRI, unenhanced, AP (25–30 s), portal venous phase (PVP; 60 s), 3-min transitional phase, and 20-min HBP images were obtained using a T1-weighted 3D turbo field-echo sequence (T1 high-resolution isotropic volume examination, THRIVE, Philips Healthcare; TR/TE, 3.1/1.5 ms; flip angle, 10°; matrix size, 256 × 256; bandwidth, 995.7 Hz/pixel). The measured voxel size was 1.5 × 1.5 × 4.0 mm, and the reconstructed voxel size was 1.17 × 1.17 × 2.0 mm. Gadoxetic acid was administered intravenously using a power injector at a rate of 1 mL/s for a dose of 0.025 mmol/kg body weight. The time for AP imaging was determined using the MR fluoroscopic bolus detection technique.

For ECA-MRI, a stack-of-stars three-dimensional (3D) radial GRE sequence with the pseudo-golden-angle ordering scheme (Fig. 1) was used during free-breathing, which was initiated at the time of contrast injection. Gadoterate meglumine (Gd-DOTA, Dotarem®, Guerbet) was administered intravenously using a power injector at a rate of 2 mL/s for a dose of 0.1 mmol/kg body weight, followed by a 20-mL saline flush. No timing run or bolus tracking was performed. A total of 379 radial spokes were continuously acquired in 64 s with the following parameters: slice thickness, 4 mm; flip angle, 10°; field of view, 375 × 375 × 200 mm3; matrix size, 252 × 252 × 50; spatial resolution, 1.5 × 1.5 × 4.0 mm3; repetition time/first echo time/s echo time, 3.9 milliseconds (ms)/1.22 ms/2.1 ms; and bandwidth, 1929 Hz/pixel. Fifty partitions were acquired (interpolated) with 0.75 partial-Fourier performed along the slice-encoding dimension. After completion of PGRA, second PVP and 3 min delayed phase were also obtained using a conventional Cartesian approach.

Fig. 1
figure 1

Schema of the sampling scheme for golden-angle radial acquisitions and pseudo-golden-angle radial acquisitions. In golden-angle radial acquisitions, the angle of the acquired spoke is continuously increased by 111.25°, leading to non-equidistantly spaced radial spokes, whereas pseudo-golden-angle radial acquisitions employ more flexible degrees of angle to generate uniform distribution of an arbitrarily determined number of radial spokes. This enables dynamic imaging using continuous data acquisition and retrospective reconstruction of image series with arbitrary temporal resolution by grouping different numbers of consecutive spokes at the desired temporal positions

Full size image

PGRA reconstruction

Raw data was exported after the PGRA scan was terminated. Image reconstruction was performed off-line using the Berkeley Advanced Reconstruction Toolbox (BART) [32] in MATLAB (MathWorks) on a Windows OS with AMD eight-core CPU at 3.6 GHz and 16 GB RAM. The dual echo stack-of-stars 3D golden-angle radial raw data was first sorted into six-dimensional space, in which the dimensions are indicated by (kr, kθ, kz, coil, phase, echo). The number of radial spokes selected determined the number of phases. The reconstructions with different radial spokes of 40, 55, and 70 were performed, and the reconstruction with 55 radial spokes provided a good compromise between temporal fidelity and image SNR. As such, we chose 55 consecutive radial spokes to reconstruct one phase (i.e., temporal window = 8.8 s). When 379 radial spokes were acquired, ⌊379/55⌋ = 6 phases were produced. First, a 1D inverse Fourier transform was performed along kz to obtain (kr, kθ, z, coil, phase, echo). For a given z and echo, compressed sensing parallel imaging reconstruction was performed, where the number of iterations was 40. The regularization parameter values of 0.01, 0.1, and 1 were tested to compare the image quality with regard to spatial blurriness and the degree of streak artifact, and the value of 0.1 was chosen as it produced the most adequate image quality. The temporal total variation regularization parameter was 0.1. Spatial regularization was not performed. After image reconstruction was performed in all slices and echoes, the first echo 3D image data (ITE1) and the second echo 3D image data (ITE2) were separately obtained. ITE1 and ITE2 were used as the input for a water-fat separation algorithm. An in-house water-fat separation method was implemented based on Eggers et al [33].

Image analysis

Two radiologists (J.H.M. and Y.K.K. with 11 and 18 years of experience interpreting liver MRIs, respectively) independently reviewed MRIs in an anonymized and randomized manner, and a consensus was reached afterwards. To minimize any learning bias, there was a 4-week interval between each blinded interpretations.

Image quality analysis

Reviewers assessed image quality of the AP and PVP with regard to respiratory motion artifact or streak artifact, anatomic sharpness of the upper abdominal organ margin (liver, gallbladder, and pancreas), conspicuity of intrahepatic vessels (hepatic artery, portal vein, and hepatic vein), and conspicuity of the hepatic tumor. For the degree of ghosting artifact, we used the following scale: 5, no artifact; 4, minimal artifact with no effect on diagnostic quality; 3, moderate artifact with some effect but no severe effect on diagnostic quality; 2, severe artifact but image was still interpretable; and 1, extensive artifacts resulting in nondiagnostic images [14]. For others, we also used a 5-point scale system in which 1 = unacceptable, 2 = poor, 3 = fair, 4 = good, and 5 = excellent (the detailed definition of scoring system in Supplementary Table E1). Given the similar image quality of the six successively acquired images using PGRA, the third (AP) and sixth (PVP) images were selected for assessment of image quality.

After reviewing individual images, each reader independently reviewed AP and PVP images side-by-side for each patient for pairwise comparison, in which EOB-MRI was the control image, followed by joint evaluation until a consensus was reached. The readers assessed the image sets by considering aforementioned five subjective categories. A semi-quantitative 5-point scale was used (1 = inferior [impairing diagnosis], 2 = slightly inferior [no influence on diagnosis], 3 = equal, 4 = slightly superior [no influence on diagnosis], and 5 = superior [easing diagnosis]) [34].

Diagnostic performance analysis

Reviewers were also asked to record the signal intensity of the hepatic lesion in the AP image, PVP image (two PVP images in PGRA), and 3 min delayed image and categorize the hepatic tumor based on the Liver Imaging Reporting and Data System (LI-RADS 2017 version) [35].

Statistical analysis

Statistical analysis was executed using R version 3.4.3 (The R Foundation for Statistical Computing). All comparisons for subjective image ratings for each image and pairwise comparisons were performed using the Wilcoxon test. For pairwise comparisons, we tested a null hypothesis that the two MRIs equal. The kappa statistic for two observers was calculated to assess the inter-observer agreement regarding pairwise comparison. The p value < 0.05 was considered significant.

Results

Clinicopathologic characteristics

Clinicopathologic characteristics of 15 patients with 18 HCCs are demonstrated in Table 1. Most of patients had liver cirrhosis (N = 10) or chronic hepatitis B (N = 4) associated with hepatitis. HCCs ranged from 11 to 34 mm in diameter (mean 21.3 mm); seven (38.9%) were 10–19 mm, and the remaining 11 (61.1%) were ≥ 20 mm in diameter. On pathology, capsule and septum were present in 16 HCCs (88.9%) and absent in two (11 mm and 12 mm in size). According to Edmondson’s classification, 13 (72.2%) HCCs corresponded to grade 2, three (16.7%) were grade 3, and the remaining two (11.1%) were grade 1. The characteristics of a total of 24 observations (18 HCCs, two hemangiomas, and four arterioportal shunts) are shown in Supplementary Table E2.

Table 1 Clinicopathologic characteristics of 15 patients with 18 HCCs
Full size table

Individual analysis for each MRI

Table 2 and Fig. 2 summarize the results of the consensus review for image quality of the two MRIs. With regard to the severity of ghosting or streak artifacts, a ghosting artifact was scored category 2 or 3 in the AP of the EOB-MRI in three patients (Figs. 2 and 3). In the AP using ECA-MRI, there was no ghosting artifact, but a mild streaky artifact (scored as category 4) was noted in all 15 patients (Figs. 2 and 3). The difference in mean value for artifacts between the two MRIs was not significant for the AP (p = 0.307), but it was significant for the PVP image (p = 0.001). The mean value for anatomic sharpness of the upper abdominal organ was slightly higher in EOB-MRI (mean 4.33 for AP and 4.73 for PVP) than in ECA-MRI (mean 4.13 for AP and 4.20 for PVP), although a statistically significant difference was only achieved for the PVP (p = 0.008). With regard to conspicuity of intrahepatic vessels, no significant difference was found either in the AP or PVP (4.87 in EOB-MRI vs. 4.80 in ECA-MRI; p = 0.500 or 1.000). Meanwhile, there was no a significant difference in the mean value for conspicuity of the HCCs in the AP and PVP (p = 0.188 and p = 0.375).

Table 2 Results of subjective image analysis using a 5-point evaluation scale system for each MR imaging
Full size table
Fig. 2
figure 2

Image quality analysis according to MRI modality

Full size image
Fig. 3
figure 3

Comparison between a gadoxetic acid-enhanced conventional MR imaging using a Cartesian approach and b extracellular contrast-enhanced MR imaging using a pseudo-golden-angle radial acquisition scheme. Ghosting artifacts are observed only in the conventional image (arrows in a)

Full size image

Pairwise comparison between the two MR examinations

In pairwise comparisons of the APs between the two MRIs, there were no significant differences between artifact (p = 0.076), organ sharpness (p = 1.000), conspicuity of intrahepatic vessels (p = 0.750), and conspicuity of the liver mass (p = 0.125). In pairwise comparisons of the PVP between the two MRIs, artifact was of EOB-MRI were significantly lower than that of ECA-MRI (p = 0.0001), but other parameters had no significant difference (p > 0.05; Table 3). The kappa values of the two observers ranged from 0.628 to 0.700 for the four imaging parameters, indicating good inter-observer agreement.

Table 3 The result of pairwise comparison between MR sequences
Full size table

Diagnostic performance analysis

Ten HCCs showed arterial hyperenhancement, followed by washout on the both MRIs, categorized as LR-5 (Table 4). Two HCCs showed no arterial hyperenhancement and were considered to be hypovascular HCCs, categorized as LR-4 with both MRIs. Six HCCs showed different signal intensities on the AP or PVP acquired from the two MRI sets. Five of them were assigned as LR-5 with ECA-MRI and LR-4 with EOB-MRI (Table 4). The remaining one was assigned as LR-4 with ECA-MRI and LR-3 with EOB-MRI. Five HCCs showed different signal intensity in the AP image between the two MRIs (Table 4). Of these, a 1.1-cm-sized HCC showed hypointensity on the AP of EOB-MRI due to early acquisition of the AP, but hyperenhancement on AP of ECA-MRI (Fig. 4). In two patients who each had 1.2-cm- and 1.4-cm-sized HCCs, arterial hyperenhancement of the HCCs was clearly seen on ECA-MRI, but not on EOB-MRI, in which the ghosting artifact was assigned as 3 or 2 (Fig. 5). The remaining two HCCs (2.3 and 2.8 cm in size, respectively) did not show arterial hyperenhancement on EOB-MRI due to too early acquisition of the AP, but showed hyperenhancement on the AP of ECA-MRI using PGRA (Fig. 6).

Table 4 Comparison of MRI modality for major features and LI-RADS categories for HCC diagnosis
Full size table
Fig. 4
figure 4

Edmondson grade II hepatocellular carcinoma in a 62-year-old male. a On the axial arterial phase, b portal venous phase, and c hepatobiliary phase images after administration of gadoxetic acid, an 11-mm-sized liver mass (arrows) is seen as hypointensity. d For the six phases successively acquired during the 64 s following administration of extracellular contrast agent, the tumor showed hyperenhancement from the third to sixth phases (arrows). e On portal venous phase and f delayed phase images, the tumor showed washout with capsule (arrows). Mild streaky artifacts were observed in six phases (d), but did not compromise diagnostic image quality. The mass (arrow) shows high signal intensity g on the T2-weighted image and h on DWI with a b value of 800 s/mm2

Full size image
Fig. 5
figure 5

A 12-mm-sized Edmondson grade II hepatocellular carcinoma in a 65-year-old male. a On the axial arterial phase, b portal venous phase, and c hepatobiliary phase images after administration of gadoxetic acid, arterial hyperenhancement is not clear, and hypointensity on the hepatobiliary phase is suspicious (arrows). A motion artifact (thin arrows) is noticeable in the arterial phase image. For the six phases successively acquired during the 64 s following administration of extracellular contrast agent (d), the tumor showed obvious arterial hyperenhancement (arrows), followed by washout on e portal venous phase (arrow). Mild streaky artifacts were observed in six phases (d)

Full size image
Fig. 6
figure 6

A 28-mm-sized Edmondson grade II hepatocellular carcinoma in a 62-year-old male. a On the axial arterial phase, b portal venous phase, and c hepatobiliary phase images after administration of gadoxetic acid, arterial hyperenhancement and washout are not clear, and hypointensity on the hepatobiliary phase is seen (arrows). d For the six phases successively acquired during the 64 s following administration of extracellular contrast agent, the tumor showed obvious arterial hyperenhancement (arrows), followed by washout on e portal venous phase (arrow). Mild streaky artifacts were observed in six phases (d). The mass (arrow) shows high signal intensity f on the T2-weighted image and g on DWI with a b value of 800 s/mm2

Full size image

ECA-MRI identified the presence of capsule in 15 HCCs as an enhancing rim encircling the tumor on the PVP and/or the delayed phase. Meanwhile, EOB-MRI demonstrated the enhancing capsule in only six HCCs. In case of a 2.2-cm HCC, ECA-MRI demonstrated arterial hyperenhancement, delayed washout, and a capsule, whereas EOB-MRI failed to show washout and capsule on the PVP.

Discussion

In the current study, we conducted an intra-individual comparison in a crossover manner, addressing the following two questions raised during clinical examination. (1) Is PGRA actually beneficial for capturing the arterial hyperenhancement of HCC compared with conventional dynamic images based on Cartesian sampling schemes? (2) Is PGRA useful for minimizing ghosting artifacts that can be observed in the AP of conventional MRI using a Cartesian approach? In addition, we examined the difference of the HCC enhancement pattern between EBO-MRI and ECA-MRI, which might lead to a difference in tumor categorization based on LI-RADS.

From the technical perspective of the first issue, given the fact that the actual gadolinium concentration of the two contrast agents were different (gadoxetic acid 0.025 mmol/kg vs. ECA 0.1 mmol/kg) [19], it is unlikely that the technical superiority of PGRA over the Cartesian approach accounts for the overall trend that we found. Of note, a previous study described that no significant difference existed in terms of delineation of arterial hyperenhancement of HCC between EOB-MRI and ECA-MRI [36]. Thus, we believe that in addition to the higher concentration of gadolinium of ECA compared with that of gadoxetic acid, the continuous data acquisition scheme in PGRA during the 64 s after contrast administration could have served as a contributing factor. The small dose of gadoxetic acid can cause difficulty in catching the optimal scan time for obtaining late AP after bolus administration of the contrast agent which would be most effective for visualizing HCC. In our study, EOB-MRI did not sufficiently depict the arterial hyperenhancement of five small HCCs, of which ECA-MRI using PGRA clearly depicted. The AP of EOB-MRI in three of the five patients was acquired too early that enhancement in the central portal vein was not observed [37]. In addition, in three patients, a ghosting artifact was observed on the AP of EOB-MRI, which might interfere with the detection of HCC. Meanwhile, no patient showed a ghosting artifact in ECA-MRI using PGRA. Thus, our observation indicates that a continuous data acquisition scheme using PGRA can be applied to optimally acquire the late AP and therefore effectively reveal the arterial hyperenhancement of HCC.

One concern with the radial acquisition scheme is the streak artifacts, as undersampling in the azimuthal direction maintains the overall image structure and results in streaks instead of the backfolding or discrete ghosting artifacts which is typically observed with Cartesian sampling [23]. We observed streak artifacts in all 15 patients enrolled in our study particularly on the dome portion of the liver, although this did not seem to compromise with the diagnostic capacity (Fig. 3d). In the AP, there were no significant differences between the two MRI sets with regard to image quality. Notably, there are other factors contributing to artifact in radial acquisition but not in Cartesian acquisition. Off-resonance can contribute to blurring artifact. Field inhomogeneity is relatively large near the patient arms far from the isocenter. It is known that the use of coil element removal, although not performed in this study, is potentially helpful in artifact reduction in an off-line reconstruction. Gradient timing delay can be another factor contributing to image quality degradation. An off-line correction of the gradient timing delay may be helpful in artifact reduction although the calibration for gradient delay correction can be made during pulse sequence development in the axial scan plane, which was the scan orientation in this study.

Given that gadoxetic acid begins to be taken up by hepatocytes approximately 60 s after contrast injection, the washout appearance on EOB-MRI can be optimally assessed only in the PVP prior to the transitional phase or HBP [38, 39]. However, Okamoto et al showed that certain histologic types of HCC, such as those with better differentiation and trabecular architecture, tend to show continuous enhancement even on the PVP [40]. As expected, in our study, two HCCs were isointense on the PVP of the two MRIs, but showed washout with a capsule only on the delayed phase of ECA-MRI. In addition, ECA-MRI revealed an enhancing capsule in 15 HCCs, but EOB-MRI demonstrated a corresponding structure in only 6 HCCs. Thus, if HCC diagnosis was established based on the imaging criteria of LR-5, the sensitivity of ECA-MRI (N = 15, 83.3%) was higher than that of EOB-MRI (N = 10, 55.6%). The results of our study implies that ECA-MRI with PGRA can be more effective in the HCC diagnosis when applying stringent criteria, at the cost of minimal streaky artifact which did not interfere with the detection of tumor hypervascularity. In particular, we believe that free-breathing PGRA scheme is desirable in patients who cannot adequately hold their breath.

There are several limitations in our study that deserve mention. First, the unbalanced comparison between EOB-MRI and ECA-MRI might have partly acted as a bias in determining the pure technical benefit of the PGRA scheme with regard to delineation of arterial hyperenhancement of HCCs and reduce the ghosting artifact. Thus, our results partly support the technical superiority of PGRA over the conventional approach. Second, a limited number of patients were enrolled in the present study. Third, a selection bias may exist with regard to the initial imaging referrals as we recruited only patients suspected of having HCC who were scheduled to undergo surgical resection.

In conclusion, the free-breathing ECA-enhanced multiphasic liver MRI using a PGRA scheme offers higher detectability of arterial hyperenhancement of HCC than conventional EOB-MRI, while offering acceptable image quality for HCC diagnosis.