### 3346

Particle swarm optimization (PSO) and comparison of a water selective T2 preparation module for simultaneous robust fat suppression and tissue contrast enhancement at 3T
Lionel Arn1, Ruud B van Heeswijk1, Andrew J Coristine1, Matthias Stuber1, and Jessica AM Bastiaansen1

1Radiology, Lausanne University Hospital (CHUV), Lausanne, Switzerland

### Synopsis

Gradient echo based pulse sequences at 3T may lack the required contrast to distinguish blood from muscle. To overcome this, T2 preparation (T2-prep) modules are used in cardiac imaging to distinguish the blood pool from the myocardium. To suppress unwanted fat signals, we exploited the additional degrees of freedom that offer the multiple radiofrequency pulses of an adiabatic T2-prep and we used particle swarm optimization (PSO) to develop a T2-prep with robust fat suppression capabilities that works in the presence of flow. Its robustness against B1 and B0 inhomogeneities were predicted by the Bloch simulations for a range of fatty tissue frequencies, and could be confirmed experimentally both in phantoms and in volunteers.

### Introduction

Gradient echo (GRE) based pulse sequences at 3T may lack the required contrast to distinguish blood from muscle. To overcome this, T2 preparation (T2-prep) modules [1] are used in cardiac imaging to distinguish the blood pool from the myocardium. To suppress unwanted fat signals, additional lengthy and specific absorption rate (SAR)-demanding modules are often necessary, as their brightness may distort image quality, or hinder anatomical visualization due to chemical shift artifacts. However, higher magnetic field strengths may complicate conventional fat saturation techniques, such as CHESS [2], which are sensitive to B1 and B0 field inhomogeneities.

In this study, the aim was to exploit the additional degrees of freedom that the multiple radiofrequency (RF) pulses of an adiabatic T2-prep offer and to use particle swarm optimization (PSO) to develop a T2-prep with robust fat suppression capabilities that works in the presence of flow [3]. The new PSO adiabatic T2-prep (PSOA-T2-prep) was compared with a conventional adiabatic T2-prep (CA-T2-prep) [4] (Fig. 1A) and a water-selective adiabatic T2-prep [5] (WSA-T2-prep) (Fig. 1B) through numerical simulations, phantom experiments, and cardiac experiments in volunteers.

### Methods

Simulations: The duration, frequency and amplitude of the RF excitation and restoration pulses, as well as the initial phase of the RF excitation pulses were optimized using particle swarm optimization (PSO)[6,7]. Bloch equation simulations of the T2-prep were performed in Matlab (The Mathworks) to compute its fitness function, which was defined as follows:

$f = SD(Fz,res - 0.05) + SD(Wz,exc) + SD(Wz,res - 1)$

where SD is the standard deviation, Fz and Wz are the longitudinal magnetizations of fat (resonance frequencies spanning from -600 Hz to -200 Hz) and water (frequency = 0 Hz) , subscript exc and res indicates the magnetization after excitation and restoration respectively. For each simulated tissue frequency, the B1 strength was varied between 80% and 120%.

Acquisitions: The three different T2-preps were implemented in a Cartesian, 2D segmented k-space GRE sequence at 3T system (Prisma, Siemens). The developed PSOA-T2-prep was tested against the CA-T2-prep and the WSA-T2-prep on a 3-compartment phantom that mimics the magnetic properties of blood, fat and myocardium. Sequence parameters: pixel size 1.1x1.1mm2 resolution, 14 k-space lines acquired per T2-prep (duration of 40 ms) spaced by 800 ms, RF excitation angle 15°, TE = 3.71 ms, TR = 8.07 ms, slice thickness = 5mm. In addition, ECG-triggered 2D breath-held acquisitions of the heart and thigh with parameters identical to the phantom scan were acquired on volunteers (N=3). The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were computed in signal compartments containing blood, myocardium, chest fat, and epicardial fat. The specific absorption rate (SAR) was monitored during the scans. Differences were tested via a Student’s t-test for paired data, with a p-value < 0.05 considered as significant.

### Results

Simulations: The excitation and restoration pulses parameters obtained by PSO (Fig. 1C) show a complete T2 preparation of water, while suppressing a range of fat frequencies (See animation uploaded as Fig. 5). Bloch simulations predicted an increased robustness in fat suppression of the PSOA-T2-prep in comparison to the CA-T2-prep and the WSA-T2-prep (Fig. 2).

Acquisitions: The PSOA-T2-prep visually reduced fat signal compared to CA-T2-prep and WSA-T2-prep (Fig. 4). Compared to the CA-T2-prep, the PSOA-T2-prep reduced chest and epicardial fat SNR by 90.0% (p=0.01) and 64.8% (p=0.002) respectively (Fig. 3). Compared to the WSA-T2-prep, the PSOA-T2-prep decreased chest and epicardial fat signal by 59.9% (p=0.044) and by 44.9% non-significantly (p=0.051) respectively. The PSOA-T2-prep reduced blood-myocardium CNR by 7.3% (p=0.036) compared to the CA-T2-prep, but increased it by 26.3% (p=0.001) compared to the WSA-T2-prep.

### Discussion

The robust fat suppression capabilities of the PSOA-T2-prep were predicted by the Bloch simulations for a range of fatty tissue frequencies, and could be confirmed experimentally both in phantoms and in volunteers, despite the challenging B0 and B1 inhomogeneities at 3T. Moreover, the on-resonance magnetization was fully T2-prepared, in contrast to the WSA-T2-prep which overtips the magnetization by 30° during the T2-prep and thus effectively protects the magnetization against T2 decay, decreasing the total T2 contrast, which may be particularly important in applications such as T2-mapping [8]. Conversely, the PSOA-T2-Prep may be more sensitive to B1 inhomogeneities in on-resonance tissues (e.g. water); these can seen in simulations (Fig. 2C) and may also explain the small decrease in CNR observed in vivo, as compared to the CA-T2-prep.

### Conclusion

We have developed a novel adiabatic T2 preparation module and demonstrated efficient T2 preparation while simultaneously providing robust fat suppression compared to the water-selective T2 preparation module.

### Acknowledgements

This work was supported by grants from theSwiss National Science Foundation (PZ00P3_167871) as well as by the Centre d’Imagerie BioMédicale (CIBM) of the UNIL, UNIGE, HUG, CHUV, EPFL, and the Leenaards and Jeantet Foundations.

### References

[1] Brittain JH, Hu BS, Wright GA, Meyer CH, Macovski A, Nishimura DG. Coronary angiography with magnetization-prepared T2 contrast. Magn Reson Med 1995;33:689–696.

[2] Haase A, Frahm J, Hanicke W, Matthaei D. 1H NMR chemical shift selective (CHESS) imaging. Phys Med Biol 1985;30:341–344.

[3] Nezafat R, Ouwerkerk R, Derbyshire AJ, Stuber M, McVeigh ER. Spectrally selective B1-insensitive T2 magnetization preparation sequence. Magn Reson Med 2009;61:1326–1335.

[4] Nezafat R, Stuber M, Ouwerkerk R, Gharib AM, Desai MY, Pettigrew RI. B1-insensitive T2 preparation for improved coronary magnetic resonance angiography at 3 T. Magn Reson Med 2006;55:858–864.

[5] Coristine A, van Heeswijk R, Stuber M, Fat Signal Suppression for Coronary MRA at 3T Using a Water-Selective Adiabatic T2-Preparation Technique. Magn Reson Med 2014; 72:763–769.

[6] Kennedy J, Eberhart R, Particle swarm optimization. Proceedings of the IEEE International Conference on Neural Networks, 1995;1942–1948.

[7] Clerc M, Standard Particle Swarm Optimisation. 15 pages. 2012.

[8] van Heeswijk RB, Feliciano H, Bongard C, et al. Free-breathing 3 T magnetic resonance T2-mapping of the heart. JACC Cardiovasc Imaging 2012;5:1231–1239.

### Figures

Figure 1: Overview of RF pulses used in each T2-prep module. All modules are followed by a transverse spoiling gradient. (A) Conventional adiabatic T2-prep consisting of a 90° RF excitation pulse, two adiabatic inversion (180°) pulses with opposite polarity and a 90° restoration pulse[4]. (B) Water selective adiabatic T2-prep[7], with a water selective excitation such that the lipid magnetization is tipped down by the restoration pulse and subsequently spoiled (Fig. 1B). However, this method requires an increase of flip angle of both excitation and restoration pulses to 120° for effective fat suppression, which in turns makes the water signal less T2-prepared. (C) PSO adiabatic T2-preparation.

Figure 2: Longitudinal magnetization as function of B1 strength and tissue off-resonance frequency after application of three different T2 preparation modules. A B1 value of 100% corresponds to the desired applied RF amplitude, with deviations corresponding to B1 inhomogeneities. The shaded region indicates the frequencies that correspond to fat (-200 Hz to -600 Hz). The black arrows indicate the range of B1 strength where the longitudinal magnetization resonating at -450Hz (fat) is within ±10%.

Figure 3: SNR and CNR of tissue compartments in cardiac experiments using different T2 preparations. Results are reported as mean ± standard deviation across participants.

Figure 4: Phantom, thigh and heart images acquired with different T2-prep modules. Arrows indicate the locations of improved fat suppression using the PSOA-T2-prep compared to the WSA-T2-prep.

Figure 5: Simulation of the magnetization of different isochromats during the PSOA-T2-prep module in the on-resonance rotating frame of reference. The blue vector represents water, the red ones fat and the black one B1. After the T2 preparation, the transverse component of the magnetizations is spoiled and only the longitudinal component remains at the start of the imaging sequence.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)
3346