Synopsis

Keywords: Oxygenation, Motion Correction

A free-breathing 3D stack-of-spiral data acquisition was developed for motion-free cardiac quantitative susceptibility mapping, to tackle the challenge where even a short breath-hold is difficult for patients to perform. ECG and respiratory bellow signal were recorded for retrospective motion binning. A 5D dataset incorporating additional cardiac and respiratory phase dimensions were reconstructed with joint spatiotemporal regularization to generate a motion-free cardiac QSM. In healthy volunteers, this method was compared with a motion-averaged reconstruction and with a separate breath-hold spiral cardiac QSM using Cartesian navigator QSM as reference. Equivalent right-to-left heart chamber differential blood oxygenation was observed among all method studied.

Introduction

Cardiac Quantitative Susceptibility Mapping (QSM)¹ is an emerging cardiac magnetic resonance imaging (CMR) technique, which has been used to non-invasively measure the differential blood oxygen saturation (ΔSaO₂) between the right and left heart chambers.² A non-gated 3D stack-of-spiral multi-echo gradient echo acquisition can be performed within a single breath-hold scan. However, even short breath-holds can be challenging or impractical for certain groups of patients with cardiopulmonary conditions, degrading the image quality and affecting the quantitative imaging results. Many retrospective motion-compensated or motion-corrected CMR methods have been developed to minimize the motion-induced artifacts, e.g., XD-GRASP based approach³. In this study, we extended the 3D stack-of-spiral breath-hold scan to a free-breathing acquisition with retrospective ECG and respiratory bellow signal guided motion binning strategy and compressed sensing (CS) based reconstruction to generate a motion-free cardiac QSM, taking advantage of spiral sampling efficiency as well as motion robustness. The ΔSaO₂ estimated from both motion-free and motion-averaged free-breathing (FB) cardiac QSM were compared with the breath-holding (BH) cardiac QSM as well as the reference Cartesian navigator (NAV) QSM.

Methods

A 3D spoiled multi-echo stack-of-spiral sequence from the previously proposed breath-holding spiral cardiac QSM was extended to a free-breathing setting. External cardiac and respiratory motion signal was recorded for retrospective motion correction from ECG/PG and bellow, respectively. The free-breathing sequence shared the same variable density spiral trajectory with 2-fold center of k-space oversampling and edge of k-space undersampled by a factor of 0.7, designed for one fully-sampled k-space with 36 leaves⁴, except that FB spiral oversamples the k-space continuously using a golden-angle view order to achieve a nearly evenly sampled k-space after retrospective motion binning. The FB and BH spiral were tested on healthy volunteers on a 3T scanner (GE750), with number of spiral leaves N_BH = 36, N_FB = 432; scan time t_BH = 19sec, t_FB = 3min46sec. Other shared imaging parameters include: readout bandwidth ±125kHz, flip angle FA = 12º, number of echoes N_e = 3, TE₁/TR/ΔTE = 0.4/17.8/5.9msec, reconstructed matrix size 256×256×24, image resolution 1.8×1.8×5mm³. Besides, a Cartesian navigator cardiac QSM² was acquired with readout bandwidth ±83.33kHz, flip angle FA = 15º, number of echoes N_e = 5, TE₁/TR/ΔTE = 1.4/18.7/3.4msec and the same resolution. All subjects provided consent for this IRB approved protocol.

First, off-resonance correction was conducted on a slice-by-slice fashion by demodulating the k-space phase along the readout direction with estimated field from a reference FID readout.⁵ Next, 3D k-space data were sorted into cardiac phases with a 71.2msec temporal resolution by the recorded ECG/PG triggering according to their leaf/slice encoding orderings, followed by the bellow signal sorting into 4 respiratory phases (Figure 1). Coil sensitivity maps were estimated from all the acquired data by ESPIRiT.⁶ A compressed sensing method was adopted for the high-dimensional image reconstruction by imposing a wavelet based spatial regularization and a temporal penalty of total variation along both respiratory and cardiac phases³:
$$x^*=arg\min_x ||AFCx-y||^2_2 + \lambda_1 ||\Psi_r x||_1 + \lambda_2 TV_t(x)$$
where $$$C$$$ is the sensitivity encoding, $$$F$$$ is the Fourier operation, $$$A$$$ is non-Cartesian sampling operation, $$$y$$$ is the motion-sorted k-space, $$$\Psi_r$$$ is wavelet operation, $$$TV_t(·)$$$ is total variation. Then the complex multi-echo data were used for QSM reconstruction by total field inversion methods (TFI+0)^2,7,8. ΔSaO₂ is estimated from mean susceptibility difference between the right/left ventricle blood pools scaled by hematocrit.

For comparison, a motion-averaged QSM was reconstructed by all the acquired data. The motion-free and motion-averaged QSM were compared with both BH spiral and NAV QSM by Deming regression and Bland-Altman plots. Image was reconstructed using BART v0.7.00.⁹ QSM were reconstructed on MATLAB R2020b. Statistical analysis was performed using R 4.2.1.

Results

FB spiral was tested on 6 healthy volunteers (age: 38±19, 50% male), as well as BH spiral and NAV QSM. Among all the subjects, the ΔSaO₂ from motion-free spiral QSM was 15.56±4.86%, compared to 17.33±4.44% for motion-averaged spiral, 15.68±5.32% for BH spiral and 15.42±3.49% for NAV QSM. Figure 2 shows a representative cardiac QSM from FB, BH spiral and NAV QSM. Figure 3 shows that motion-free spiral QSM has equivalent ΔSaO₂ compared to BH spiral as well as NAV QSM (ΔSaO_2,f = 0.91ΔSaO_2,bh + 1.26 (%), Pearson’s r = 0.97; ΔSaO_2,f = 1.25ΔSaO_2,nav – 3.5 (%), Pearson’s r = 0.97).

Discussion

Motion-free QSM generated from the free-breathing 3D stack-of-spiral shows well-aligned ΔSaO₂ estimation with both BH spiral and NAV QSM in this study, demonstrating that this free-breathing scheme can be an alternative to the current BH spiral in case subject has trouble with breath-holds. Moreover, even the motion-averaged QSM shows good correlation with BH and NAV results (Figure 4, ΔSaO_2,a = 0.83ΔSaO_2,bh + 4.28 (%), Pearson’s r = 0.98; ΔSaO_2,a = 1.14ΔSaO_2,nav – 0.08 (%), Pearson’s r = 0.95). A larger sample size with additional right heart catheterization (RHC) data will be conducted to further validate this approach. More sophisticated motion correction method on highly undersampled high dimensional k-space will also help improve the overall image quality.

Conclusion

A free-breathing 3D stack-of-spiral sequence with retrospective motion-binning and CS reconstruction was shown to generate motion-free cardiac QSM with equivalent ΔSaO2 estimation to the breath-hold scan.

Acknowledgements

No acknowledgement found.