Interleaved, undersampled radial multiple-acquisition SSFP for improved left atrial cine imaging

Abstract

Purpose:

bSSFP left atrial (LA) cine suffers from off-resonance artefacts, particularly in the pulmonary veins (PV). Linear combination or multiple-acquisition SSFP (MA-SSFP) effectively removes banding but greatly increases scan time. We hypothesized that MA-SSFP with interleaved radial undersampling, where each phase-cycling is acquired with an interleaved set of radial projections, would improve image quality of LA cine with a small increase of scan time and streak artefacts.

Methods:

Undersampled radial MA-SSFP with and without interleaving was compared to fully-sampled radial bSSFP via simulations, phantoms, and in vivo imaging. Ten healthy subjects were imaged on a 3T scanner, with bSSFP and MA-SSFP cine of the LA, and B0-mapping. Images were assessed (1=worst, 5=best) by two independent readers, with respect to five qualitative criteria and apparent SNR.

Results:

In healthy subjects, off-resonance differed from the right inferior PVs to the LA cavity by 163Hz±73Hz at 3T. Compared to fully-sampled radial bSSFP, interleaved radial MA-SSFP significantly improved image quality with respect to off-resonance artefacts (3.8±0.6 vs 2.3±1.0, p=0.005), PV conspicuity (2.8±1.0 vs 4.3±0.5, p=0.005), and the number of visualized PVs (1.7±0.4 vs 0.9±0.7, p=0.008), although with greater streak artefacts (3.4±0.4 vs 4.9±0.2, p=0.004) and lower measured aSNR (24±9 vs 69±36, p=0.002). Flow artefacts were similar. Interleaved radial MA-SSFP reduced streaking artefacts and increased aSNR vs. non-interleaved radial.

Conclusions:

Interleaved radial MA-SSFP cine reduces banding artefacts with an acceptable increase of scan time and streak artefacts. The proposed technique improves the LA and PV visualization in bSSFP cine imaging.

Introduction

Conventional LA cine is performed with bSSFP in standard 2-chamber and 4-chamber views. Axial cine might provide volumetric imaging (e.g. using an axial 2D stack or 3D slab), with improved evaluation of LA function, including ejection fraction, volumes, and strain (1). LA function characterization already generates important early markers of diastolic dysfunction and heart disease in many cardiomyopathies (2–7), and improved LA cine would have value. Although it is not well recognized, the LA is one of the most challenging regions for bSSFP, even at 1.5T, and success is limited (8,9). Hu et al. (10) demonstrated that the LA and PVs are locations of very high off-resonance with frequencies ranging from 50Hz to 100Hz at 1.5T. The off-resonance in the LA is likely due to blood inflowing from the highly off-resonant lungs.

bSSFP off-resonance artefacts (11) can be reduced by shortening TR, using lower field strength, optimizing shimming, and using a frequency scout (12), which determines the local optimal centre frequency for bSSFP. However, these modifications are not always practical or consistently successful in a clinical environment. Alternatively GRE might be used, and is recommended when off-resonance artefacts are present (13), but suffers from a poorer SNR and myocardium-blood contrast-to-noise ratio (CNR) due to its inflow-dependent contrast‥ Several methods have been developed to directly reduce the banding artefacts in SSFP (14) (15) (16), although they have not been applied to the LA. Multiple-acquisition bSSFP (MA-SSFP) is one such method, and has been investigated (16–18) regarding SNR efficiency and bSSFP band-removing properties under different numbers of phase-cyclings, and distinct combination methods. The initial MA-SSFP method, (19) was based on linear combination of the complex-valued signal from each phase-cycled image (LC-SSFP), or complex-sum (CS) (16). Many other combination methods (17) (20,21) have also been reported.

MA-SSFP has also been used for a variety of applications and anatomic regions, including radial MA-SSFP (22–25), yet rarely investigated for cardiac cine imaging (26). Prolonged scan time is the major limitation for cardiac applications. Possible acceleration solutions include two-pass MA-SSFP with sum of squares combination (26), and utilization of a random and “disjoint” k-space undersampling for each phase-cycling, combined with compressed sensing (27).

In this work, we propose a novel time-efficient radial MA-SSFP cine method, which uses interleaved undersampled radial projections for each phase-cycling, with the hypothesis that interleaving combined with phase-cycling would not only eliminate the banding artefacts from off-resonance, but also suppress the generation of streak artefacts from undersampling. Simulations, phantom imaging, and in vivo imaging were performed to compare the proposed technique to the non-interleaved undersampled radial MA-SSFP, which does not vary radial projections across phase-cyclings, and to the regular fully-sampled radial bSSFP using a single 180° phase-cycling. This study provides three valuable and novel contributions: it addresses the unmet need of bSSFP cine in the LA, where off-resonance is severe; it presents a feasible MA-SSFP cardiac cine method; and it demonstrates and utilizes the improved radial k-space sampling provided by interleaved radial spokes for each phase-cycling.

Methods

Pulse sequence design

Figure 1A shows the schematic of the proposed 4-pass MA-SSFP cine imaging sequence. The four phase-cyclings are sequentially applied, each started with a dummy-heartbeat acquisition followed by the cine acquisition. The dummy heartbeat is used to drive the magnetization into the steady-state for each phase cycle. The number of radial projections per phase-cycling is N_p, which is sampled through N_p/vps heartbeats, where vps (views-per-segment) is the number of projections sampled per heartbeat. In our implementation (N_p=48, vps=16) 3 heartbeats are required in addition to a dummy heartbeat within each phase cycle. For the entire sequence, four complementary RF phase-increments of 0°, 90°, 180°, and 270° were used, matched with 4 uniformly-spaced radial interleaves to sample a total of 4N_p=192 radial projections. This led to a scan time of 16 heartbeats for 1 breath-hold.

Figure 1.

(A) Schematic of the interleaved radial MA-SSFP sequence, which comprises of four phase-cyclings, each composed of one dummy heartbeat and three segmented acquisitions. (B) Each segment sequentially samples a portion of the radial projections belonging to the interleaf, and the combination of the four phase-cyclings/interleaves comprises the total k-space sampling pattern. The orders of phase-cycling and radial interleaving were rearranged to reduce streak artefacts by matching complementary (because most disjoint) radial interleaves (e.g. Interleaf 1 and 3) to closer phase-cycling (e.g. 0◦ and 90◦ phase-cyclings).

The orders of phase-cycling and radial interleaving were rearranged to reduce streak artefacts by matching the most complementary radial interleaves to closer phase-cyclings. Specifically, this was performed (see Figure 1B) by matching interleaf #1 (initial angle ${\varphi }_{\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}}=0$) and #3 (${\varphi }_{\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}}=2\pi /\left(4N_p \right)$) to phase-cycling 0° and 90°, respectively, and by matching interleaf #2 (${\varphi }_{\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}}=1\pi /\left(4N_p \right)$) and #4 (${\varphi }_{\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}}=3\pi /\left(4N_p \right)$) to phase-cycling 180° and 270°, respectively.

Simulation

All simulations were performed in Matlab (Matlab 2014, Natick MA). The performance of the interleaved radial MA-SSFP at different undersampling rates relative to non-interleaved MA-SSFP and fully-sampled bSSFP was studied. A “left atrial” image was simulated with uniform proton density, and a horizontally linearly varying B0 field, which generated an off-resonance range of −300Hz to 300Hz within the LA (Figure 2a). Four bSSFP images (Figure 2a. lower row) with inter-TR phase increment of 0°, 90°, 180°, and 270° were simulated, based on Bloch Equation modelling (28) of the signal vs. frequency in steady state (after 3000 TRs), with the following parameters: TR=3ms, TE=1.5ms, θ=60ͦ and T1/T2=1600ms/200ms simulating the blood. White Gaussian noise was added to the radial k-space data to simulate the impact of noise on the methods’ performance.

Figure 2.

(A) Simulation of LA geometry with uniform proton density and horizontally varying off-resonance from −300Hz to 300Hz, which causes banding artefacts in each phase-cycled bSSFP image. (B) The reconstructed MA-SSFP images with or without interleaving. Yellow arrows point to the increased streak artefacts with non-interleaved radial sampling. (C) The variation of normalized aSNR efficiency, which plots signal-to-noise and artefact divided by the square-root of scan time, normalized by the aSNR efficiency of fully-sampled radial bSSFP. MA-SSFP has intrinsically lower SNR efficiency, and undersampling artefacts at low N_p reduce this further. Interleaved radial MA-SSFP improves the SNR efficiency for low N_p. Normalized signal homogeneity, compared to bSSFP, shows that MA-SSFP is resistant to banding artefacts, at all undersampling levels.

Eleven values of N_p (16, 32, 48, 64, 96, 128, 160, 192, 256, 320, 384) and an image size of 192×192 were explored in the simulation. SNR was 100 (measured on fully-sampled radial bSSFP with 192 projections and 180° phase-cycling). The reconstructed bSSFP images for interleaved/non-interleaved undersampling were combined with complex-sum. The interleaved MA-SSFP was compared to non-interleaved MA-SSFP and the standard fully-sampled radial bSSFP, by measuring apparent SNR (aSNR) efficiency (reflecting noise and artefacts, divided by the square root of total numbers of projections used), and homogeneity against banding artefacts. The aSNR was evaluated by dividing the mean signal in an ROI located in the centre of the LA, which encompasses the peak signal at the centre of bSSFP passband, by the standard deviation in the whole background. The banding artefacts were evaluated by the homogeneity of the blood pool, which was the ratio of the mean signal over the entire LA (including the bands) to the standard deviation of the same area.

Phantom imaging

All imaging studies were carried out on a 3.0T MRI scanner (Siemens Trio, Siemens Healthcare, Erlangen, Germany).

Agar phantoms with T2/T1 values similar to those of blood and myocardium (T2/T1 of 50/1050ms, 70/970ms, 70/1070ms, and 160/1260ms) were imaged with fully-sampled 2D radial cine bSSFP and undersampled 2D radial cine MA-SSFP. Imaging parameters were: FOV 320×320mm², image matrix 192×192, slice thickness 5 mm, bandwidth 1000 Hz/pixel, TR/TE = 3.1/1.5ms, flip angle 60°, and 16 vps were acquired. The shimming was adjusted to generate a strong linear gradient along one direction, to visualize the banding artefacts. The N_p per phase-cycling for MA-SSFP used the same range as for simulations, with and without interleaving. aSNR efficiency was evaluated for each undersampling rate.

In vivo study

This study was approved by our institutions IRB and all volunteers provided written, informed consent. Ten healthy volunteers (age 27±3 years, 6 female) were imaged with the fully-sampled radial bSSFP (N_p=192) and the interleaved undersampled radial 4-pass MA-SSFP, with standard spine and body matrix coils. Shimming was manually optimized by placing a 3D shim volume slightly larger than the heart, and using first and second order shims. A frequency scout was used (12) and the bSSFP centre frequency was modified based on the observed off-resonance in the LA cavity. FOV was 360×360mm², image matrix 160×160, 16 views per segment, resolution 2.2×2.2×8mm³, bandwidth 1008Hz/pixel, TR/TE = 2.8/1.4ms, and flip angle 36°−52°. Forty-eight radial projections were used for each MA-SSFP phase-cycling with total breath-hold time of 16 heartbeats, compared with 13 heartbeats for conventional fully-sampled radial bSSFP (N_p=192). A non-interleaved MA-SSFP radial dataset was acquired as a comparison to the interleaved approach in each subject. A Cartesian cine with similar scan parameters was acquired for visual comparison. B0 mapping was acquired with a dual-echo GRE sequence (TE1=3.6ms, TE2=5.0ms, ΔTE=1.4ms) to measure the off-resonance in the left inferior PV (LIPV), right inferior PV (RIPV) and the LA at 3T. To obtain the B0 map in Hz, the second-echo image was divided by the first-echo image, from which the phase difference was extracted, and divided by the ΔTE.

Reconstructions were performed offline in Matlab (Natick, MA). Gridding was performed in k-space (29) followed by complex-sum combination of the four images to generate MA-SSFP results. Coil combination was performed with square-root of sum of squares.

The images were independently evaluated by two readers (CH, DCP), with more than 3 and 15 years of cardiac MR experience respectively, over five criteria including off-resonance artefacts, streak artefacts, flow artefacts, PV conspicuity, the number of visible PVs, using a 5-point scale. For scoring artefacts (flow, streaks, off-resonance), the scale was 1=severe, 2=moderate (with moderate loss of image quality) 3=mild (easily observable but without loss of image quality), 4=visible (just detectable), 5=not visible. For scoring conspicuity, the scale was 1=non-diagnostic, 2=poor, 3=moderate, 4=good, 5=excellent. Blinding was not possible in comparing bSSFP to MA-SSFP, since the MA-SSFP images were recognizable due to eliminated banding of the chest-wall. The scores were averaged before the final evaluation. Additionally, aSNR was measured by dividing the mean of the blood signal in the LA by the standard deviation of the signal in a large ROI anterior to the chest wall in air space. For interleaved and non-interleaved MA-SSFP matched data-sets, the same criteria were scored in a blinded fashion, and aSNR was compared.

Statistical Analysis

Comparisons of paired qualitative criteria were performed with Wilcoxon signed-rank test. Comparisons of continuous data were performed with a two-sided paired t-test. A p-value ≤0.05 was considered significant. For each metric, a single data point was measured per patient per method.

Results

Simulations

Figure 2A shows simulation results of the interleaved/non-interleaved MA-SSFP for different combination methods, with 48 projections, 60º flip angle, image size 192×192.=. Images were identically window-levelled to reveal the streak artefacts. The non-interleaved acquisition led to more streak artefacts (yellow arrows) than the interleaved acquisition.

Figures 2C shows the changes of aSNR efficiency and banding artefacts, all normalized with those of fully-sampled radial bSSFP, with respect to the number of radial projections for different methods. Interleaved MA-SSFP showed robustness of aSNR efficiency to undersampling. SNR efficiency was reduced with MA-SSFP compared to bSSFP, consistent with previous reports about SNR and SNR efficiency (17,18). Banding artefacts were suppressed independent of the level of undersampling, as expected.

Phantom Studies

Figure 3A compares fully-sampled radial bSSFP with MA-SSFP using 192 and 48 N_p, acquired with both interleaved and non-interleaved radial k-space. Banding artefacts observed in bSSFP image were reduced in all MA-SSFP images, independent of the undersampling. Further, non-interleaved MA-SSFP led to visually stronger streak artefacts than interleaved MA-SSFP. Note that the trajectory-error artefact (well-known when imaging over 180°) visible especially for the topmost bottle is due to acquisition window timing errors related to off-resonant spins (30).

Figure 3.

(A) The phantom reconstruction results of the fully-sampled radial bSSFP, and undersampled MA-SSFP, with or without interleaving (N_p=48). The MA-SSFP used the complex sum of four phase-cyclings (4N_p=192). The banding artefacts in bSSFP were generated by adding a shim along the vertical direction. The banding artefacts were greatly suppressed in the MA-SSFP images. Interleaving mitigated the streak artefacts compared to non-interleaving. (B) The variation of normalized aSNR efficiency with the number of projections per phase-cycling. The trend is similar to that found in simulations, with improved aSNR efficiency (reflecting reduced streak artefacts) using interleaving.

Figure 3B shows the aSNR efficiency measured in an ROI in one of the bottles (red circle in Figure 3A), for different N_p values. As in the simulations, all criteria were normalized by those of the fully-sampled radial bSSFP (N_p=192). The results agree well with simulations.

Healthy subjects

Figure 4 compares LA cine imaging with fully-sampled Cartesian bSSFP, fully-sampled radial bSSFP, and undersampled interleaved radial MA-SSFP, at identical phases in two subjects. Partial to complete obliteration of the right LA and PV is evident on bSSFP (arrows)—also a common finding in clinical assessment of these structures—but not on MA-SSFP. Cartesian and radial bSSFP cine are similar in appearance.

Figure 4.

Comparison of fully-sampled Cartesian bSSFP, fully-sampled radial bSSFP, and underesampled MA-SSFP in two subjects. The improvement in PV and LA conspicuity is remarkable. Yellow arrows point to the banding artefacts in the bSSFP images, where the PV and LA cavity are obliterated due to banding.

Figure 5A compares a fully-sampled radial bSSFP image with non-interleaved and interleaved radial MA-SSFP in two subjects. As observed in Figure 4, partial to complete obliteration of the LA and PVs is present, which is strongly ameliorated using MA-SSFP. Interleaving improves the quality visually, and reduces streaking artefacts, compared with non-interleaved MA-SSFP. Off-resonance maps are shown, with strong off-resonance observed in regions of the pulmonary vein ostia. The movie in Supporting Information Video 1 shows improvement with MA-SSFP throughout the cardiac cycle, although bSSFP flow artefacts remain. The artefacts are strongest in systole, and mostly related to out-of-slice bSSFP signal.

Figure 5.

(A) Comparison of fully-sampled radial bSSFP, and MA-SSFP with and without interleaving, along with corresponding B0 maps in two subjects. Yellow arrows point to banding artefacts in the LA, and red arrows show banding artefacts in the chest wall, which are present in bSSFP, but absent in MA-SSFP images. (B) Qualitative comparison of streak artefacts, flow artefacts, banding artefacts, PV conspicuity, the number of visible PVs between fully-sampled radial bSSFP and the interleaved MA-SSFP method. Flow artefacts remain problematic for both methods. 1=worst and 5=best, see text. (C) Comparison of streak artefacts for interleaved and non-interleaved MA-SSFP in all subjects. (D) Comparison of aSNR (apparent SNR) for fully-sampled radial bSSFP, and for interleaved and non-interleaved MA-SSFP.

Figure 5B compares interleaved MA-SSFP with fully-sampled radial bSSFP over 5 qualitative criteria. While bSSFP had less streaking (4.9±0.2 vs 3.4±0.4, p=0.004), the interleaved MA-SSFP dramatically reduced off-resonance banding artefacts (3.8±0.6 vs 2.3±1.0, p=0.005), improved the PV conspicuity (4.3±0.5 vs 2.8±1.0, p=0.005), and increased the number of visible PVs (1.7±0.4 vs. 0.9±0.7, p=0.008). Flow artefacts remained problematic for both methods. Remarkably, an 89% improvement in number of visible PVs was observed.

The in vivo comparison between interleaved and non-interleaved MA-SSFP cine found that scores for flow artefacts (3.1±1.1 vs. 3.1±0.9), banding artefacts (3.7±0.5 vs. 3.6±0.5), PV conspicuity (3.7±0.5 vs. 3.4±0.5, p=0.08), and the number of visible PVs (2±0 vs. 1.9±0.2) did not significantly differ. Interleaving led to a significant improvement (Figure 5C) in streak artefacts (3.4±0.4 vs 2.2±0.4, P=0.008).

Figure 5D compares the measured aSNR between the three methods. aSNR was much lower with MA-SSFP (24±9 vs 69±36, p=0.002), roughly consistent with the simulation and phantom results. Interleaving improved aSNR by around twofold compared to non-interleaving (24±9 vs 13±3, p=0.002) due to reduction of streak artefacts.

Off-resonance of the LA at 3T

The average off-resonance measured in 10 subjects, in the LA, LIPV, and RIPV, was −31Hz ±127Hz, 74Hz±121Hz, and 115Hz±146Hz, respectively. The average difference between LA and LIPV, and between LA and RIPV, was 105Hz±44Hz and 163Hz±73Hz, respectively. Figure 5A shows the off-resonance maps in two subjects, disclosing strong off-resonance in the PVs.

Discussion

The main finding of this study is that interleaved undersampled radial cine MA-SSFP dramatically improves LA cine at 3T, with a feasible increase of scan time to 16 heartbeats, because the PVs and atrial cavity are visible. This improvement is due to reduction of banding artefacts. Our study confirmed the extent of off-resonance in this anatomical region, and demonstrated the importance of the MA-SSFP cine method, which is more resistant to off-resonance. Our measurements of off-resonance in the LA at 3T agree with those measured at 1.5T (10), being not quite double.

The PVs and LA are of clinical interest both for anatomical imaging as well as functional assessment. Interleaved radial MA-SSFP of the LA has applications both for non-contrast angiography and volumetric cine function, and would enable 3D approaches, which is an important area of on-going research (9,31,32).

While the imaging time was increased, the proposed MA-SSFP cine technique had lower aSNR compared to bSSFP, because of the inherent SNR inefficiency (17) (18), and additionally due to streak artefacts from undersampled radial sampling (33). However, note that due to the intrinsic high SNR of bSSFP, the in vivo aSNR of MA-SSFP remained adequate (>20) (34), and interleaving minimized streaking. For consistency, we investigated MA-SSFP with 48 N_p per phase-cycling (192 N_p total); use of more projections might improve quality within a still feasible breath-hold.

The simulations, phantoms, and in vivo data showed a strong reduction in undersampling artefact using interleaved MA-SSFP, with complex sum. This is likely due to a more fully-sampled k-space, generated by directly adding complementary k-space interleaves acquired from similar magnetization states. The in vivo study showed a significant reduction of streak artefacts both qualitatively and quantitatively, comparing interleaving to non-interleaving. Although undersampling artefacts were not the major limitation in our study, further improvements to our method using parallel imaging and compressed sensing might also reduce the streak artefacts (27).

While banding artefacts are the greatest limitation for LA bSSFP cine, our study identified that flow artefacts are also a challenge. In particular, through-plane flow passing through bSSFP dark bands causes severe artefacts (35,36). Furthermore, the persistence of out-of-slice signal results in prominent artefacts, especially for LA cine due to the ascending and descending aorta. Partial solutions exist for these artefacts(26), but the limitation remains. Additionally, LC-SSFP depends on a stable breath-hold, for combining data sets. Finally, the radial LC-SSFP scan requires a 16 heart-beat breath-hold which is feasible for most patients, but is still longer than for the conventional cine protocol, which is often accelerated with parallel imaging, and requires a <10 second breath-hold.

The study design also has limitations. Firstly, the study was conducted at 3T, where the off-resonance artefacts are more severe than at 1.5T. However, similar artefacts for bSSFP of the LA have been reported at 1.5T (10) (9). Secondly, the radial LA cine protocol differed from typical cine imaging, in that the spatial resolution was lower than typical (2.2 ×2.2mm² vs. 1.5 ×1.5 to 2.0mm²), the temporal resolution was also lower (44ms vs. 36ms), and the scan time was longer than usual for Cartesian bSSFP cine. Finally, further clinical studies are highly warranted.

In conclusion, undersampled radial interleaved MA-SSFP cine provides, with an acceptable increase in breath-hold time, improved LA and pulmonary vein conspicuity and increased image quality when compared with fully-sampled radial bSSFP.

Supplementary Material

Supp MovieS1

Supporting Information Video S1. Cine movies of a single subject, comparing Cartesian bSSFP, fully-sampled radial bSSFP, and interleaved radial MA-SSFP.