The effect of SNR optimization on cell quantification accuracy for fluorine-19 MRI sequences: bSSFP, FSE, and FLASH
Kai D. Ludwig1, Erin B. Adamson1, Christian M. Capitini2,3, and Sean B. Fain1,4,5

1Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, 2Pediatrics, University of Wisconsin-Madison, Madison, WI, United States, 3Carbone Cancer Center, University of Wisconsin-Madison, Madison, WI, United States, 4Radiology, University of Wisconsin-Madison, Madison, WI, United States, 5Biomedical Engineering, University of Wisconsin-Madison, Madison, WI, United States


Several MRI data acquisition methods have been used for fluorine-19 (19F) MRI cell tracking and optimizing the image SNR helps mitigate low sensitivity. An optimization workflow is presented for three 19F pulse sequences based upon relaxation parameters measured in a 19F reference phantom. Bloch simulations reveal signal differences between the reference phantom and pure 19F cellular label for SNR-optimized bSSFP, FSE, and FLASH. The simulated relative errors in 19F signal suggest SNR optimization can compromise signal quantification and thus in vivo cell quantification but could provide insight for improved methods to balance the degree of spin-density weighting and SNR.


To mitigate low sensitivity, several MRI data acquisition methods including fast low angle shot (FLASH), balance steady-state free precession (bSSFP), and fast spin echo (FSE) pulse sequences have been commonly used in 19F MRI cell tracking and optimized to maximize the image signal-to-noise ratio (SNR). A published cell quantification algorithm relies on the ratio of the detected in vivo 19F signal arising from 19F-labeled cells compared to a 19F reference phantom with a known 19F-spin density to estimate the apparent in vivo cell number.1 However, spin-lattice (T1) and spin-spin (T2) relaxation properties can vary due to temperature, magnetic field strength, and chemical environment.2 Choice in the data acquisition strategy may result in 19F signal contrast that is not dominantly spin-density weighted, thereby biasing any quantification with the relative T1 and T2 of the in vivo and phantom 19F spins. This work investigates MR parameter optimization and the resulting differential image weighting which may compromise accuracy of the estimated cell count for three different 19F pulse sequences.


Relaxation parameters were measured on a 19F perfluoropolyether (PFPE) reference phantom (linearized PFPE suspended in agar; CelSense) and free PFPE cellular label (CS-1000-ATM, linearized PFPE in solution; CelSense). 19F spectra from an inversion recovery experiment (six TIs=0.063-2.0s) were fit to a mono-exponential signal recovery model for T1 estimates. A CPMG pulse sequence measured T2 from 19F spectra (six TEs=13.8-440ms). Global FIDs at ten readout delay times (0.5-120.5ms) were acquired for T2* measurements. Both T2 and T2* data were fit to a mono-exponential decay model. All 19F relaxation measurements used a home-built 19F quadrature volumetric coil, 90° global RF excitation, 10kHz receiver bandwidth (rBW), 5.0s TR, and a hot air blower/temperature probe on a 4.7T Agilent MRI system (Agilent Technologies). Bloch simulations3 estimated transverse magnetization for FLASH, bSSFP, and FSE to optimize sequence parameter for SNR. The optimization utilized the relaxation measurements from the PFPE phantom at room temperature. Bloch simulations and theoretical SNR calculations, performed as previously reported,4 were then simulated for the free PFPE at 37°C to demonstrate variance in signal-weighting. Matrix size, spatial resolution and total acquisition time for all simulations were held constant. All simulations and post-processing were performed in MATLAB (MathWorks).


The estimated 19F T1 increased with temperature, as expected, while T2 and T2* did not vary greatly (Table 1). A large 201-262ms difference in T1 and 205-211ms difference in T2 were observed between the PFPE phantom and free PFPE as shown in Figure 1A-B. The workflow for optimizing the pulse sequences for SNR with Bloch simulations is shown in Figure 2 with the resulting optimized parameters listed in Table 2 for each pulse sequence. The simulated signal exhibited a substantial dependence on TR and ETL for FSE, flip angle for bSSFP and FLASH, and rBW for all sequences. The bSSFP led to the greatest simulated SNR but resulted in the largest percent error. The FSE led to the smallest percent error and showed an advantageous boost in SNR for the free PFPE instead of hindering the potential to detect ‘in vivo’ signal in the free PFPE. Using the optimized parameters resulted in considerable differences in the simulated 19F signal between the PFPE phantom (24°C) and the free PFPE (37°C) for FSE, FLASH, and bSSFP.


The T1 measurements of free PFPE corroborate with trends previously reported1 at higher field strengths. Here, the free PFPE is a surrogate for the relaxation parameters that may arise from 19F labeled cells in vivo, which would reasonably be at body temperature (37°C). Off-resonance and regional variation in flip angle are not addressed in this work but would be expected to affect bSSFP and FLASH sequences more so than FSE. Despite potentially high SAR deposition, the FSE’s long optimum TR would enable multiple slices to be acquired with no additional time penalty. The simulated relative errors in 19F signal presented here suggest SNR optimization can compromise signal quantification and thus cell quantification.


An optimization workflow is presented for three common 19F MRI pulse sequences. The simulated signal differences observed between a 19F reference phantom and pure 19F cellular label provide insight on the effects of pulse sequence SNR optimization for 19F cell tracking on the accuracy of in vivo quantification of cell number. Ongoing work will focus on validating simulations with acquired images and investigating the tradeoffs between TR, SNR, and quantification accuracy. These insights could lead to improved methods to balance the degree of spin-density weighting and SNR.


The authors thank our collaborators and colleagues. We gratefully acknowledge UW School of Medicine and Public Health, UW Carbone Comprehensive Cancer Center, American Cancer Society, Alex’s Lemonade Stand Foundation, St. Baldrick’s Foundation and GE Healthcare. The research was also supported under NIH awards UL1TR000427 and TL1TR000429. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.


1 Srinivas M, Morel PA, Ernst LA, Laidlaw DH, Ahrens ET. Fluorine-19 MRI for visualization and quantification of cell migration in a diabetes model. Magn Reson Med. 2007;58(4):725-34.

2 Kadayakkara DK, Damodaran K, Hitchens TK, Bulte JW, Ahrens ET. (19)F spin-lattice relaxation of perfluoropolyethers: Dependence on temperature and magnetic field strength (7.0-14.1T). J Magn Reson. 2014;242:18-22.

3 Hargreaves B. Bloch Equation Simulation. Accessed November 2015.

4 Mastropietro A, De Bernardi E, Breschi GL, Zucca I, et al. Optimization of rapid acquisition with relaxation enhancement (RARE) pulse sequence parameters for (1)(9)F-MRI studies. J Magn Reson Imaging. 2014;40(1):162-70


Table 1: Summary table of measured relaxation properties (T1, T2, and T2*) of 19F agents at room and body temperature.

Figure 1: Data points and mono-exponential fits for the measured relaxation properties (A) T1, (B) T2, (C) T2* for a 19F reference phantom (PFPE Phantom) and in-solution PFPE cellular label (Free PFPE) at room (24°C) and body (37°C) temperature.

Figure 2: 19F MR parameter optimization workflow for FSE (A,B), bSSFP (C,D), and FLASH (E,F) using Bloch simulations to maximize SNR. Note that TE is a function of modifying the rBW.

Table 2: A summary table of the SNR-optimized parameters for the FSE, FLASH, and bSSFP pulse sequences based on relaxation parameters in PFPE phantom at 24°C.

Table 3: Summary table of the simulated relative SNR differences and percent error using an optimized FSE, FLASH, and bSSFP. The starting magnetizations or spin densities were set equal for the PFPE phantom and free PFPE. In an absolute spin density weighted sequence, the relative SNR difference should be zero.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)