Optimization of steady‐state free precession MRI for lung ventilation imaging with 19F C3F8 at 1.5T and 3T

Purpose: To optimize 19F imaging pulse sequences for perfluoropropane (C3F8) gas human lung ventilation MRI considering intrinsic in vivo relaxation parameters at both 1.5T and 3T. Methods: Optimization of the imaging parameters for both 3D spoiled gradient (SPGR) and steady‐state free precession (SSFP) 19F imaging sequences with inhaled 79% C3F8% and 21% oxygen was performed. Phantom measurements were used to validate simulations of SNR. In vivo parameter mapping and sequence optimization and comparison was performed by imaging the lungs of a healthy adult volunteer. T1 and T2 * mapping was performed in vivo to optimize sequence parameters for in vivo lung MRI. The performance of SSFP and SPGR was then evaluated in vivo at 1.5T and 3T. Results: The in vivo T2 * of C3F8 was shown to be dependent upon lung inflation level (2.04 ms ± 36% for residual volume and 3.14 ms ± 28% for total lung capacity measured at 3T), with lower T2 * observed near the susceptibility interfaces of the diaphragm and around pulmonary blood vessels. Simulation and phantom measurements indicate that a factor of ~2‐3 higher SNR can be achieved with SSFP when compared with optimized SPGR. In vivo lung imaging showed a 1.7 factor of improvement in SNR achieved at 1.5T, while the theoretical improvement at 3T was not attained due to experimental SAR constraints, shorter in vivo T1, and B0 inhomogeneity. Conclusion: SSFP imaging provides increased SNR in lung ventilation imaging of C3F8 demonstrated at 1.5T with optimized SSFP similar to the SNR that can be obtained at 3T with optimized SPGR.


| INTRODUCTION
MRI of lung ventilation with inhaled inert hyperpolarized (HP) gases has a proven sensitivity for the assessment of lung ventilation changes in obstructive airways disease. 1 MRI with fluorinated gases (e.g., SF 6 , C 2 F 6 , and C 3 F 8 ) shows promise as a complementary or alternative method for lung ventilation imaging, but in contrast to HP gas MRI, does not require additional polarization equipment. 2 Additionally, fluorinated gases may be mixed with oxygen (O 2 ) and continuously breathed, possibly allowing simpler investigation of dynamic lung physiology, such as the measurement of fractional ventilation by multi-breath washout [3][4][5] without the complication of gas depolarization observed with HP gas. Efforts to improve the quality of fluorinated gas ventilation MRI has been ongoing. 1,2,6,7 However, obtaining high-resolution ventilation images with fluorinated gases at thermal equilibrium is challenging because of the low spin density, short T 2 * and constrained imaging time. 7 Past strategies of fluorinated gas MRI have focused on the use of short TE spoiled gradient (SPGR) sequences with TR relatively close to T 1 . This is due to two constraints: first, the relatively short T 2 * of fluorinated gases in the lungs and, second, the specific absorption ratio (SAR) considerations at 3T (the most common field strength used for imaging of fluorinated gases to date) . For SPGR imaging with the repetition time TR«T 1 , and where the acquisition time (T aq ) is approximately that of T 2 * , 8,9 the SNR per-unit-time is nearly constant with TR due to the competing factors of averaging, T aq and longitudinal recovery. 9,10 For example, TR values of 20 ms 11,12 for SF 6 with T 1 < 2 ms, 13 and 20 ms 14 or 13 ms 12 for C 3 F 8 with T 1 ∼12.4 ms 14 However, if TR is of the same order as T 1 , the optimization of single-echo SPGR sequences generally requires minimizing TR so that the rate of longitudinal recovery for each TR is maximized. 10 More recently, studies of C 3 F 8 imaging have been performed at 1.5T using a 16-element receive array. 15,16 However, imaging was still performed with T 1 ∼TR = 12 ms, and a T aq of 7.1 ms, which is significantly longer than the T 2 * .
Therefore, future fluorinated gas imaging can clearly benefit from imaging parameter optimization as presented here.
In free-gas phantoms, C 3 F 8 gas has a longer T 2 (~17 ms 17 ) when compared with other fluorinated gases (∼4.2 for SF 6 18 and ∼5.9 for C 2 F 6 2 ), so improved signal to noise may potentially be achieved with the use of SSFP. The optimization of imaging parameters for SPGR 8 and SSFP 19 1 H MRI has been detailed previously. Also, the optimization of SSFP imaging parameters 20 has been investigated for the imaging constraints of HP gas ventilation MRI with both 3 He and 129 Xe. 21 Sequence optimization for perfluorocarbon emulsions has also been performed previously, 22,23 but in this instance the T 2 and T 1 relaxation parameters are significantly longer than for gas phase perfluorocarbons.
In this work, we demonstrate the application of SSFP sequences for 19 F lung ventilation imaging using C 3 F 8 /O 2 gas at 1.5T and 3T. Optimization of SSFP and SPGR imaging parameters was carried out by simulation with the specific relaxation parameters of C 3 F 8 /O 2 gas as found in phantoms. The additional consideration of k-space filtering 24 from T 2 * decay was explored by simulation of the 1D point spread function (PSF). 25 Simulations of the SSFP signal were performed and compared experimentally with those achievable with a SPGR sequence. Constraints posed by SAR for in vivo applications are highlighted and the relaxation parameters T 1 and T 2 * were mapped in vivo to verify the parameters used in simulation. Finally, in vivo lung imaging was performed with both sequences at 1.5T and 3T to test the theoretical/ experimental predictions of SNR improvement. This study is to benchmark optimal imaging parameters.

| Simulations of SPGR and SSFP signal for C 3 F 8
The two sequences considered here for 3D lung ventilation imaging with 19 F perfluoropropane were SPGR and SSFP. For the SPGR sequence, transverse magnetization is dephased after each RF pulse by application of spoiling gradients. 26 Conversely, in SSFP the phase of the excitation pulse phase alternates by ± π each TR, resulting in recycling of the transverse magnetization, while gradients refocus spins after acquisition for balanced SSFP (bSSFP). 27 In the simulations presented, transverse magnetization (M xy ) was evaluated at TE, which correlates with the center of k-space and thus determines image signal intensity. Simulations of M xy with SSFP were performed according to Hargreaves et al, 20 with an effective transverse decay rate term of T 2 . 28 The steady state M xy with SPGR, using the Ernst angle for maximum signal, 29 was calculated as 8,30,31 , where is the flip angle (FA). For both SSFP and SPGR, the resulting image SNR is related to the transverse magnetization by 8 , where BW is the bandwidth per-pixel, N avg is the number of averages, ΔV is the voxel size, N p is the number of phase-encode steps, T aq is the readout gradient acquisition time and T s (1) is the total imaging time. The factor T aq ∕TR represents the efficiency of the sequence in terms of maximizing the fraction of the TR devoted to sampling the signal. The expected optimal T aq with SPGR is close to T aq ≈ T * 2 . 9 For the sake of a fair SSFP and SPGR comparison, the spatial resolution, imaging time, and y and z phase encoding steps remained the same. The effects of SSFP signal transient behavior on the final SNR were ignored, which was justified by the relatively short T 1 and T 2 when compared with TR, resulting in a steady state being reached rapidly.
With HP 3 He gas it has been demonstrated that dephasing from the imaging gradients has a significant effect on the effective transverse relaxation rate, 24 while the effect is less significant when imaging with 129 Xe because of the much lower diffusion coefficient. 32 Calculations with the even lower diffusion coefficient of C 3 F 8 , 17 with its relatively low T 2 , indicate that this effect is small when compared with the uncertainty/variability in the T 2 and, therefore, the effect of diffusion dephasing due to the imaging gradients themselves was neglected here.
Imaging timing parameters that impact upon TE and T aq include the following: the RF pulse width (T pw ) and imaging gradient encoding/refocusing delays before (T D1 ) and after (T D2 ) frequency encoding. Therefore, TE = T pw To emulate practical imaging sequence timings, the simulated RF pulse widths were matched to the measurement values, while T D1 and T D2 were selected to be 0.6 ms throughout the comparison to closely match those used in measurement.

| Quantification of T 2 * decay induced k x filtering
Insight into the reduction in image quality due to T * 2 filtering during frequency encoding (k x ) was attained by comparison of the 1D PSF of the different sequences. For SPGR the signal decays exponentially from the center of the RF excitation pulse with a time constant T 2 * . For SSFP the signal is modeled as decaying exponentially with time constant T 2 , as well as decaying symmetrically away from TE with the time constant T 2 * , similar to simulation/measurement performed in reference with a spin-echo sequence, 33 as the transverse magnetism ideally decays similarly in a bSSFP sequence. 28

| Relaxation parameters of C 3 F 8 /O 2
For the phantom simulations presented here the T 1 and T 2 of C 3 F 8 gas mixed with 21% O 2 are assumed to be 17 ms 17 Within the lung the T 1 of fluorinated gases is known to depend more upon regional differences in partial-pressure 34,35 of O 2 . Consequently, the mean in vivo T 1 has been reported as 12.4 ms at 3T. 14 Additionally, the intrinsic T 2 of C 3 F 8 gas within the lungs has not been reported, but is expected to remain comparable to T 1 . 17,34 Additionally, the mean in vivo T 2 * relaxation constant of C 3 F 8 has been reported as ∼2.2 ms at 3T. 14 3 | METHODS

| Simulation of steady-state magnetization with SSFP
The relation between steady-state magnetization, FA, and RF frequency offset from resonance were simulated with MATLAB considering the particular relaxation parameters of C 3 F 8 for 3D imaging with a TR of 3.4 ms. Additionally, to assess whether transient oscillations in the magnetization during initial RF excitations are significant, the transverse magnetization for successive RF excitations was simulated for different values of TR Furthermore, to quantify the expected 1D PSF arising from transverse magnetization decay the PSF was simulated for varying T aq .

| Validation of simulated magnetization with phantom SNR measurements
To compare the simulations of signal for C 3 F 8 for SSFP versus SPGR, phantom experiments were carried out with a 2-L glass cylinder (12 cm diameter, 20 cm length) filled with 79% C 3 F 8% and 21% O 2 at 1.4 bar pressure. Rectangular (24 cm × 16 cm) transceive single loop coils were constructed from 11 mm width copper strip, tuned and matched at the 1.5T (GE Signa HDx) (60 MHz) and 3T (Philips Ingenia) (120 MHz) frequencies and centered with the cylinders during imaging. Before the phantom studies at 1.5T and 3T, FA maps were generated by varying the input power in SPGR imaging with TR 100 ms» T 1 and fitting the received signal according to Equation (1), as in Maunder et al. 36 The prescribed FA recorded in Table 1 for the imaging performed with the glass phantoms was based on the fitted FA at the center of the phantom. Furthermore, to ensure that SNR and relaxation parameters were not inaccurately calculated due to B 1 inhomogeneity, voxelwise parameter mapping was calculated using the voxelwise fitted FA map 37 rather than a prescribed mean value. The assumed T 1 and T 2 relaxation parameters were verified by comparing the variation of image SNR and simulated steady-state transverse magnetization with RF excitation frequency offset. The offset frequency was varied from −1/TR to 1/TR (TR = 4.6 ms) in steps of 30 Hz with two FAs (22.5° and 75°) and the SNR was evaluated within a central voxel of the glass cylinder phantom at 3T. Furthermore, the simulated transverse magnetization of SPGR and SSFP sequences were compared with measured image SNR with varying FA. The image SNR was averaged within a central 1.2 × 1.2 × 3 cm 3 voxel with either 3D SSFP or SPGR imaging at 1.5T.
The restrictions on FA due to regulatory SAR contraints, 38 when applying the same imaging sequence in vivo with a thoracic vest transceiver coil 36,39 were also considered in the SPGR and SSFP SNR versus FA comparison. The vest transceiver coil is similar in geometry to the one used here and should have comparable SAR characteristics. For a 1 kW RMS input power, the maximum local 10 g averaged SAR was simulated within a realistic human body model (SIM4LIFE Zurich Med Tech, Duke model 40 ) as 125 W/kg, with a 11.8 T∕ √ kW transmit efficiency at 60 MHz. The global SAR was calculated conservatively as the input power to the coil being completely deposited into a 70 kg patient. A constant 500 s hard pulse width was assumed, while pulse amplitude was varied to match the FA. All simulated FAs were, therefore, achievable with the 4 kW peak power amplifier used in in vivo imaging at 1.5T. The specific imaging parameters for these and the in vivo imaging experiments detailed later are provided in Table 1.

| Phantom relaxation parameters
The T 2 * of C 3 F 8 within the glass cylinder phantoms is not representative of in vivo values measured in the lung where tissue-airspace field inhomogeneity plays a significant role. Therefore, a spatially varying T 2 * inhomogeneity was introduced by placing a paramagnetic wire in close proximity to the glass canister. At 1.5T, T 2 * maps were calculated by varying the TE in multiple image acquisitions, while fixing the BW, FA, and TR, then fitting according to Equation (1

| Simulated and measured optimization of SPGR and SSFP imaging parameters
To determine the optimal TR for 3D SPGR and SSFP imaging sequences, measurements were performed at 1.5T and 3T with varying TR. The same FA was used for SSFP imaging (approximately 90°), while the input RF power was varied with SPGR imaging to maintain the optimal Ernst FA at the center of the phantom. Three different regions of interest covering a range of T 2 * values were investigated.
The simulated steady-state transverse magnetization was multiplied by the factor √ T aq TR to represent the SNR perunit-time efficiency due to trade-off between acquisition bandwidth and averaging. The simulated magnetization and measured SNR were plotted against TR The previously acquired FA maps were used to verify that within the representative voxels the difference in SNR due to potential mismatch in prescribed FA and optimal FA was less than 5%.

| In vivo relaxation parameter mapping
In vivo lung ventilation imaging was performed in a healthy male adult volunteer (29 years old) following informed consent and adhering to protocols approved by UK National research ethics committee. An eight-element in-house constructed transceive array was used for 1.5T 19 F and 1 H in vivo imaging. 42 An elliptical birdcage coil (Rapid Biomedical, Rimpar, Germany) was used for 19 F and 1 H imaging at 3T. The global FA was measured before imaging by performing whole-lungs spectroscopy with a varying input power and long TR (268 ms at 1.5T and 200 ms at 3T) and then fitting the resulting signal according to Equation (1). Saturation of the lungs with the C 3 F 8 /O 2 mixture was achieved by directing the volunteer to take three inhalations from a Douglas bag then perform a breath-hold.
To compare the global and regional variation of T 1 in vivo at 1.5T and 3T with that obtained in phantoms at 3T, T 1 and FA parameter mapping was performed. In the same breathhold two 3D SPGR imaging sequences were performed with a long TR relative to T 1 (TR = 50 ms at 3T and TR = 35 ms at 1.5T) and prescribed mean FAs of ∼90° FA and ∼30° FA (for 1.5T an additional point of ~60° was included). The resulting pixel-wise FA was calculated based on the signal intensity difference according to Equation (1). 43 In a second breathhold, three 3D SPGR imaging sequences were performed with TR shorter than T 1 (TR = 6.5 ms at 3T and TR = 5.7 ms at 1.5T) and the resulting pixel intensity variation used to fit T 1 . 44 In addition, at 3T, T 2 * mapping was performed to corroborate the presumed values. A multi-echo SPGR acquisition was made with TE in the range of 1-6 ms in 1-ms steps, and the resulting images were fit on a voxel by voxel basis. T 2 * mapping was performed at the two lung volumes of total lung capacity (TLC) and residual volume (RV).

| In vivo comparison of SPGR and SSFP
image SNR At 1.5T, SNR comparisons were made between a SPGR sequence with an approximately optimal prescribed FA (∼45° with TR = 4 ms), and a 3D SSFP imaging sequence with ∼70° FA, which was limited due to SAR constraints. At 3T, SPGR and SSFP imaging were performed with nearly identical imaging parameters, because SAR constraints restricted the FA to 30° with a TR of 4 ms The direct comparison of SPGR and SSFP sequences at each field strength was carried out within the same breath-hold to avoid inconsistencies in coregistration or possible differences in the PFP: air concentration ratio in the lungs that may arise between breath-holds (20 s at 1.5T and 28 s at 3T plus inter-scan delay of approximately 5 s). Images were obtained with fully optimized sequences using the same resolution at both 1.5T and 3T for final comparison. Finally, to accurately compare the imaging methods k-space was filtered with an identical Hamming filter before FFT reconstruction. 11 As a final comparison between the two field strengths, imaging was performed at 1.5T and 3T with the same resolution (10 × 10 × 10 mm 3 ) TR (4 ms) and 8 averages. To reduce the breath-hold time, four averages were obtained in two separate breath-holds to total lung capacity (20 s at 1.5T and 14 s at 3T which included an elliptical shutter).

Experimental Optimization
Simulation of the C 3 F 8 phantom steady state transverse magnetization with a TR = 3.4 ms is shown in Figure 1A, with varying FA and RF excitation offset frequency. Because T 1 is approximated as T 1 = T 2 , the transverse decay is equal to the longitudinal recovery rate and the optimal FA remains 90° for the central (0 Hz) offset frequency in all cases. 27 The simulated oscillating transverse magnetization during the initial series of excitations is shown in Figure 1B for varying TR The rapid longitudinal recovery of C 3 F 8 means that a steady-state is reached within a short number of RF pulses for the TRs shown, reducing the amount of k y & k z filtering to a negligible level when SSFP imaging with C 3 F 8 . 24 Therefore, the application of 10 stabilization excitations before imaging performed in this study reduced the variation in magnetization with subsequent RF pulse excitations to less than 10%, even for a relatively short TR of 3.2 ms.
The simulation of the 1D PSF during frequency encoding readout is shown in Figure 1C for both SPGR and SSFP. The resulting amplitudes of the PSFs for the different sequences is also shown in Figure 1D, and the FWHM of the PSF in Figure 1E. SPGR is deficient in terms of lower PSF amplitude and increased FWHM when compared with the SSFP PSF as the T aq is increased. However, if T aq is kept short relative to the T 2 * the FWHM remains low and blurring is minimal. For SPGR and SSFP sequences with C 3 F 8 if T aq < 2 T 2 * the FWHM of the PSF remains comparable. Figure 2A shows the measured SNR of the SSFP signal at 3T with varying offset excitation frequency. As expected, the simulated magnetization displays a similar trend versus offset frequency when compared with measurement. Central slices are displayed for the varying offset frequency, demonstrating the introduction of banding artifacts arising from field inhomogeneity as the excitation frequency is offset from the center. In Figure 2B

SNR comparison
Maps of the FA homogeneity that all subsequent phantom parameter mappings are based upon are displayed in Figure  3A. The T 1 map for the phantom at 1.5T and 3T is displayed in Figure 3B, and is in agreement with the range reported in Chang and Conradi 17 at 60 MHz. The T 1 is expected to increase slightly with Larmor frequency (<1 ms larger at 176 MHz versus 60 MHz reported in Chang and Conradi 17 ).
Here, the standard deviation in the measurement was greater than the expected increase from 1.5T to 3T. The T 1 maps do not show any regional variation with proximity to the paramagnetic wire. In Figure 3C the T 2 * maps for a central slice of the phantom with the paramagnetic wire added at both 1.5T and 3T. The T 2 * map measured at 3T when the wire is excluded is also shown. The main comparison of image SNR obtained with SPGR and SSFP sequences is displayed for varying TR in Figure  4A (at 1.5T) and Figure 4B (at 3T). The central ROIs were chosen to demonstrate the SNR variation with T 2 * and are displayed on the T 2 * maps in Figure 3C. As TR is varied the measured SNR remains significantly higher for SSFP when compared with SPGR. SPGR optimization is highly dependent on T 2 * , with maximal SNR occurring when the T aq is slightly greater than T 2 * .
The simulated transverse magnetization (normalized for the time available time for acquisition and averaging) closely matches the measured ROI SNR. However, because the pixel ROIs include a range of T 2 * the SNR behavior with TR does not match exactly. In simulation, the signal was assumed to correspond to the transverse magnetization amplitude at k x = 0 (center of the frequency encoding gradient), but in fact, is also dependent on the PSFs as presented in Figure 1D.

| In vivo parameter mapping
Mapping of FA is displayed in Figure 5A, and the corresponding colocalized T 1 map in Figure 5B. The mean T 1 is lower than that found in the phantom (Figure 3), which is in agreement with previously reported in vivo T 1 from whole lungs (12.4 ms at 3T 14 ). Regional variation is apparent, with the greatest variation observed at the lung-tissue interfaces. T 2 * maps are shown in Figure 6A and Figure 6B  but the regional variation and dependence on inflation level is significant. There does not appear to be a visually observable correlation between regions of varying T 2 * and T 1 , while T 2 * seems to be lowest in regions near the susceptibility interfaces of the pulmonary blood vessels and at the inferior portion of lung where perfusion is highest.

| In vivo SNR performance: SSFP versus SPGR
An average increase in SNR by a factor of 1.7 was found at 1.5T ( Figure 7A compared with Figure 7B). However, there are some bands of high versus low increases in SNR ( Figure  7C) demonstrating the possible impact of field inhomogeneity. At 3T, no overall increase in SNR was observed with SSFP when performed under the SAR conservative settings (FA of 30 • and TR of 4 ms) when compared with SPGR imaging ( Figure 7C compared with Figure 7D). The significant regional variation in the SNR increase throughout the lungs with SSFP versus SPGR is reflective of the local B 0 inhomogeneity.  Figure 8. Through the use of a transceive array and increased SNR with SSFP imaging the mean SNR at 1.5T is higher than that of 3T for the same resolution. The increase in SNR is dominated by the regions of increased coil sensitivity at the anterior and posterior regions of the lung and much of the periphery, so the variation is higher at 1.5T as well. Due to B 1 inhomogeneity in the anterior of the lung at 3T, there is significant signal drop-out.

| DISCUSSION
The close agreement between the simulated and measured SSFP versus SPGR signals, with both varying FA and offset frequency, indicates that the expected parameters of T 2 , T 1 , and T 2 * within the glass phantom are valid. Additionally, the direct measurement of T 1 and T 2 * matched the expected in-phantom values, with measurably smaller mean values of T 1 and T 2 * measured in vivo. T 2 was indirectly validated by the close agreement between SSFP simulations and measurements because measuring T 2 with established spin echo sequences was constrained by the SAR limitations. For short sequence TR, variations in the simulated T 2 and T 1 for C 3 F 8 have minimal influence on the simulated steady-state magnetization, because they are expected to remain comparable. 17 However, lower T 1 results in a predicted greater steady-state magnetization with SPGR. This manifests as a reduction in the relative improvement of SSFP imaging of PFP in the lungs when compared with in a PFP gas phantom, which was observed at both 1.5T and 3T. It was also demonstrated that the improved SNR achieved using SSFP when compared with SPGR is strongly dependent upon the T 2 * expected in vivo and the k x filtering effect of T 2 * reduces the expected image quality when T 2 * < T acq .
The in vivo T 1 44 and T 2 * mapping results add to the data in the literature for C 3 F 8 in lungs. The T 1 of fluorinated gases has previously been attributed to have a direct correlation with ventilation-perfusion. 34,45 Consequently, the differences in mean values for T 1 measured at 1.5T and 3T ( Figure 5B) may be due to the level of saturation with the C 3 F 8 + O 2 mixture. T 2 * correlates with lung inflation/filling level and may be related to alveolar size, 46 which can change in diseases such as emphysema. Therefore, the parameter mapping techniques followed here may have direct relevance for future study. Figure 2B demonstrates that at 1.5T using the optimal imaging parameters, the conservative SAR limits 38 are exceeded because the optimal SSFP FA is high due to the near equivalence of T 1 and T 2 for the gases used in these experiments. Nevertheless, the sequence when run with a suboptimal FA of 72º still provides significant SNR gains over SPGR. However, at the higher field strength of 3T, SAR constraints are expected to further limit the potential advantage of SSFP for human 19 F ventilation imaging. A prescribed FA during in vivo imaging at 3T of 30° was shown in phantom experiments ( Figure 2A) to result in nearly the same SNR with SPGR and SSFP imaging.
There is a likelihood of some off-resonance banding artifacts occurring in routine imaging, as may be observed near the diaphragm in some of the 1.5T SSFP images in Figure 7B and Figure 8A. Even in the geometrically uniform and relatively small cylindrical glass phantoms, banding can be observed at the susceptibility interfaces and as resonant frequency is offset (Figure 2A). Increased B 0 and RF inhomogeneity, especially with FOVs as large as the human torso (38-52 cm), increases this likelihood at the higher field strength of 3T. Previously, B 0 mapping within the lungs with inhaled 3 He gas at 1.5T and 3T demonstrated F I G U R E 7 SNR maps of PFP in the lungs acquired at 1.5T using either A, SPGR or B, SSFP imaging and C, the relative improvement in SNR with SSFP imaging. Additionally, SNR maps acquired at 3T using D, SPGR or E, SSFP sequences with (F) maps of the relative ratio of SNR of SSFP versus SPGR imaging F I G U R E 8 SNR maps for fully optimized imaging at equal resolution A, with SSFP imaging at 1.5T and an 8-element array or B, at 3T with SPGR and a quadrature birdcage coil are shown for final 1.5T and 3T comparison a variation in Larmor frequency at 3T of >120 Hz across the lungs. 47 Therefore, the B 0 inhomogeneity in the lungs makes the application less robust at 3T. Future investigations to test SSFP versus SPGR imaging at higher field strengths may show the expected improvement if the same imaging methods are reproduced in animal MRI where SAR limits are not exceeded and FA and B 0 inhomogeneity can be reduced.
The expected SNR gains of using SSFP over SPGR imaging at 1.5T are comparable to the improvements seen with SPGR when going from 1.5T to the higher field strength of 3T. Therefore, equivalent quality human ventilation images may be obtained with the lower field strength without the same constraints of SAR. Especially, if at 1.5T a multi-channel receive array is used as in this work and others. 5 The use of a receive array for imaging of the thorax/torso may result in further SNR increases in the range of 50-100%, 48,49 with the majority of the increase obtained at the periphery. Therefore, a mean SNR of 15-20 may have been expected at 1.5T by combining the following factors: a measured SNR increase of 70% by use of SSFP, the approximate linear dependence of SNR with field strength, 50 and the use of a receive array. We note that, at 3T k-space was sampled with the use of an elliptical shutter where the corners of k-space were not sampled (22% undersampling). Hence, despite the same nominal resolution of 1.5T and 3T for images in Figure 8 the SNR was slightly enhanced for the 3T images.
Here, the in vivo imaging at 1.5T was performed with a flexible vest coil, 42 which typically would have a worse transmit homogeneity than rigid volume coils as demonstrated with direct comparisons with 3 He hyperpolarized gas imaging at 1.5T in De Zanche et al 51 (the variation was 7.3% within lungs with an asymmetric birdcage coil). Despite the lower frequency of 1.5T the flexible transceive array showed lower in vivo transmit homogeneity during in vivo imaging, while the transmit homogeneity with the birdcage coil at 3T was also not ideal (∼20% variation). The FA variation should not affect the in vivo T 2 * parameter mapping, or the T 1 mapping because the colocalized FA maps were used in the fitting. The in vivo comparison of SPGR and SSFP imaging is confounded by the coil inhomogeneity (FA variation of ±22.4% in Figure 5A) and natural variation of T 1 (24% in Figure 5B) and T 2 * (28% in Figure   6B) throughout the lungs. These three factors lead to the range of variations in improvement with SSFP versus SPGR shown in Figure 7C and Figure 7F, and in the future may be investigated further.
Comparison of the in vivo ventilation image quality obtained here to previous studies is difficult due to differences in the imaging resolutions used and in the method of reporting and measuring SNR in images. Often, SNR is reported within a ROI with the highest signal. Further complicating the comparison, the longer T aq used in previous studies results in broadened PSF as simulated in Figure 1D,E, which imparts a higher image SNR whilst degrading image quality due to blurring, 25 and may be additionally modified by filtering during postprocessing. 14 Additionally, different studies have used different RF transmit/receive coils that may contribute to more than a factor of 3 in SNR variation.
Nonetheless, in our study the measured SNR of 13.1 ± 5.7 throughout the lungs at 1.5T (8-element transceive array with image resolution of 10 × 10 × 10 mm 3 and T aq = 3.3ms) is equivalent to the SNR of ∼30 reported by Gutberlet et al. 15 (transmit birdcage and 16-element receive array with image resolution of 7.8 × 7.8 × 20 mm 3 and T aq = 7.1ms). At 3T, the SNR achieved in our study of 11.6 ± 3.2 throughout the lungs (elliptical birdcage coil with image resolution of 10 × 10 × 10 mm 3 and T aq = 2.1ms) is also comparable to 32 ± 6 in a chosen central region reported by Couch et al. 14 with a transceive vest coil and image resolution of 7.1 × 7.1 × 22 mm 3 and T aq = 7.1ms with half-fourier echo. Although the in-plane resolution reported here at 3T is lower, visual comparison of the images in Figure 8 with those in Couch et al. 14 show more clearly defined edges and features, similar to those obtained by Halaweish et al. 12 at 3T, which did not report SNR values (with transceiver vest coil and image resolution of 6.25 × 6.25 × 15 mm 3 and T aq = 7.7ms).
The benefits shown here for SSFP of C 3 F 8 are less applicable to the other common fluorinated gases of SF 6 or C 2 F 6 because of their shorter T 1 and T 2 values. Therefore, the use of C 3 F 8 over other fluorinated gases has an increased benefit in terms of SNR achieved with SSFP and longer T 2 * .
Consequently, the use of ultrashort TE (UTE) sequences for SF 6 or C 2 F 6 is logical, 14 while not providing as dramatic an improvement for 19 F lung imaging with C 3 F 8 because T 2 * is greater than gradient encoding and RF pulse times that may be used. Additionally, T 2 * filtering in UTE SPGR imaging with fluorinated gases in 3D radial or 1D Cartesian UTE 11 is another concern somewhat circumvented by the use of C 3 F 8 with short TR SPGR or SSFP.

| CONCLUSION
With optimized SSFP images we have demonstrated improved lung ventilation images with 19 F C 3 F 8 gas at 1.5T. We believe the image quality shown here to be equivalent or superior to images published previously at 1.5T or 3T and this work bodes well for the emergence of 19