Cardiac MR spectroscopy can reveal metabolic alterations inside the myocardium as a result of dietary interventions or metabolic conditioning, as well as diurnal changes (1–3). To suppress FID components from imperfect 180° refocusing pulses, conventional PRESS sequences use unipolar spoiler gradients that are sensitive to motion (4). Reliable measurements can therefore only be performed in heart phases where the myocardium is stationary. Variations in heart rate during the examination, especially under pharmacologically induced stress, can however result in acquisitions outside the optimal heart phase. In this work we therefore propose the use of bipolar spoiler gradients in a PRESS sequence (PRESS^mc) to mitigate second order gradient moments, thereby making the acquisition insensitive to velocity and acceleration. The proposed method is compared to a conventional PRESS sequence with reduced spoiler gradients (4) (PRESS^rs) using simulations and in-vivo measurements of the human heart.

The phase of moving spins subject to a gradient field $$$G(t)$$$ with spatial position $$$x(t)$$$ is expanded around $$$t_{0}=0$$$ as:

$$\phi=\gamma\int_{}^{}G(t)\sum_{}^{}\frac{1}{n!}\frac{\partial ^{n}}{\partial t^{n}}x(t)\mid_{t=0}t^{n}dt$$

with $$$\gamma$$$ denoting the gyromagnetic ratio and $$$n$$$ the order of motion. A gradient waveform that is motion compensated up to $$$n^{th}$$$ order has to fulfil

$$m_i = \int_{}^{}G(t) t^{i}dt = 0$$

for $$$1\leq i \leq n$$$, where $$$m_{i}$$$ denotes the $$$i_{th}$$$ gradient moment. Unwanted FID components from refocusing pulse imperfections experience phase dispersion that is given by

$$\phi_{spoil}=\gamma a_{spoil} s_{x}$$

and is proportional to the voxel size $$$s_{x}$$$. The effective spoiling areas for both sequences can be calculated as illustrated in Fig-1.

The proposed sequence was implemented on a Philips Achieva 1.5T system equipped with a 40mT/m, 200T/m/s gradient system and a 5-channel cardiac receive coil. The bipolar gradient pairs were scaled such that the first and second gradient moments are nulled at the time of acquisition (Fig-2). An effective spoiling area of 20mTms/m was used for both sequences, resulting in echo times of 22ms (unipolar) and 28ms (bipolar).

A numerical heart motion model was created from a 3D tagging E dataset acquired in a healthy volunteer and interpolated to a temporal resolution of 0.05ms (Fig-3). Intravoxel dephasing was simulated over the full cardiac cycle considering gradient moments up to 3^rd order (Fig-4a).

Five healthy female volunteers (age 21-32) were enrolled in this study. For both sequences a single PRESS voxel of 8x16x32mm³ was positioned inside the septal wall for 8 different heart phases (250ms–600ms). Iterative shimming was performed during one initial breath hold and was kept constant throughout each session. Eight reference water spectra and 48 water suppressed spectra were acquired for each sequence and heart phase using cardiac triggering and pencil beam respiratory gating. Total scan duration was 6:45min per heart phase and sequence, including preparation phases.

The reconstructed spectra were analysed in jMRUI (5) by time-domain fitting in AMARES (6) assuming Lorentzian line shapes after first-order phase correction. Six resonances were fitted for the water suppressed spectra: triglycerides (TG) at 0.9, 1.3 and 2.1ppm, creatine (Cr) at 3.01ppm, trimethylammonium (TMA) at 3.2ppm and the residual water at 4.7ppm. The triglyceride-to-water ratio (TG/W) was calculated as the sum of the fitted TG resonances at 0.9ppm and 1.3ppm divided by the fitted water signal.

Experimental data confirm the heart phase dependency of SNR for the conventional sequence, while PRESS^mc exhibits only small SNR degradation in diastole (Fig-4a).

Mean SNR values of the water signal increased significantly (p < 0.05) from 702±203 (RS) to 924±183 (MC) with significantly reduced coefficients of variance over eight repeat measurements (Fig-4b).

The variation of the measured TG/W ratio over heart phases is significantly reduced with PRESS^mc during diastole (Fig-5a). Mean coefficient of variation is significantly reduced in across volunteers from 0.37±0.26 (RS) to 0.1±0.02 (MC) (Fig-5b).

Comparison of best and worst spectra acquired during diastole with both sequence shows improved spectral quality with second-order motion compensation (Fig-5d).

The proposed sequence yields higher SNR of water and triglyceride resonances, better reproducibility in repeat measurements and improved repeatability of quantification over the cardiac cycle despite an increased TE, which results in 15% (water) / 7.4% (TG) signal loss due to transversal relaxation.

Experimental data indicates residual signal dephasing during diastole, which is not predicted by simulations. This issue is associated with the limited spatial and temporal resolution of the numerical model. Additionally, rotational components of the contractile motion are only partly compensated to a degree where a linear translational approximation is valid.

Second-order motion compensation in PRESS significantly increases signal-to-noise ratio of water and triglyceride signals of the in-vivo heart while simultaneously reducing variability over repeat measurements and heart phases.