Introduction

Recent CSI techniques have quantified regional spectral parameters of ¹²⁹Xe within red blood cells, parenchyma/blood plasma (membrane,M) and airspaces (gas) of the lungs.¹ However, frequency-domain CSI has limited spectral resolution for 10-15s breath-hold scans, primarily due to the necessarily short acquisition windows restricting the number of sample points, which can result in inaccuracies in the fitted parameters. This method also assumes Lorentzian lineshapes for the resonances, while prior research indicates a substantial contribution of Gaussian broadening to the M linewidth.^2,3 To enhance accuracy and enable regional quantification of Gaussian broadening, here we employ time-domain Voigt fitting of 3D CSI data from the lungs of healthy volunteers.

Methods

Imaging was performed on 3 healthy volunteers (HV) on a 1.5T GE 450w scanner with ¹²⁹Xe polarised to ~30% (POLARIS, Sheffield, UK).⁴ Images were acquired during breath-hold after the inhalation of 1L of HP ¹²⁹Xe from functional residual capacity. Sequence details: 3D CSI with voxel size(2.9cm)³,matrix=14×14×7,FOV=40×40×20cm³,FA=10°/0.1° for dissolved-phase/gas resonances,TR/TE=8ms/0.9ms,1799 excitations.¹ In the spectral dimension, 88 sample points were acquired with BW=20kHz. Images were reconstructed to 64x64x20 while the original spectral resolution was preserved. Following application of a mask, data analysis was performed by fitting the real and imaginary parts of the time-domain signal(Fig.1(a)) to a 3-resonance Voigt function defined by$$Re[V(t))]=\sum_{n=1}^{3}A_{n}\text{exp}(-\pi\Gamma_n^Lt)\text{exp}\left[-\left( \frac{\pi\Gamma_n^Gt}{2\sqrt{\text{ln}2}} \right)^2 \right]\text{cos}\left[ \left( \omega_{0,n}-\omega_t \right)t+\phi_n \right]$$
$$Im[V(t))]=\sum_{n=1}^{3}A_{n}\text{exp}(-\pi\Gamma_n^Lt)\text{exp}\left[-\left( \frac{\pi\Gamma_n^Gt}{2\sqrt{\text{ln}2}} \right)^2 \right]\text{sin}\left[ \left( \omega_{0,n}-\omega_t \right)t+\phi_n \right]$$,

where A is the FID the amplitude, Γ^L and Γ^G are Lorentzian and Gaussian linewidths (set to zero for the RBC resonance), ω₀-ω_t=Δω is frequency difference between the ¹²⁹Xe resonance frequencies and RF transmit/receive frequency and φ is the phase of the n^th (n=Gas,M,RBC) ¹²⁹Xe resonant peak. Voigt T₂ values for the gas and membrane signals were calculated using an empirically-derived expression $$T_2^V=\alpha T_2^L[1-\text{exp}(-T_2^G/\alpha T_2^L)]+\beta T_2^L$$
where T₂^L=1/πΓ^L and T₂^G=2√ln2/πΓ^G are the Lorentzian and Gaussian T ₂s, respectively, and α=0.965 and β=0.00012 are empirical constants. The error in this function was computed by calculating the absolute error between synthetic spectra with known T ₂^V and T₂^V calculated with this function over a range of T₂^G and T ₂^L values parameterised by d=(T₂^G-T₂^L)/(T₂^G+T₂^L), i.e. d=1(Gaussian dominated) for T₂^G>>T₂^L and d=-1(Lorentzian dominated) for T₂^G<<T₂^L(Fig.1(b)).

Results

With the exception of Γ^L and Γ^G, all fitted parameters converged to values within the bounds of the fitting function. When Γ^L and Γ^G are respectively ~zero, it indicated purely Gaussian and Lorentzian broadening(Fig.1,bottom). The function for computing T₂^V from measured T₂^G and T₂^L was shown to have an error of <2% for d values -0.75-1(Fig.1(b)). The ¹²⁹Xe-gas resonance was shown to be Gaussian-dominated on average over the lung volume with median Gaussian fractions of 0.74[0.5,0.9],0.77[0.5,0.93] and 0.79[0.54,0.93] for the 3 HVs. The M resonance was shown to have approximately equivalent contributions from Lorentzian and Gasussian broading, with median Gaussian fractions of 0.48[0.35,0.59],0.46[0.27,0.64] and 0.45[0.24,0.61]. The M Voigt T₂s were similar over all HVs, with median values of 2.6ms[2.4,2.7],2.4ms[2.2,2.6] and 2.4ms[2.2,2.6], whereas there was more variation observed in the gas Voigt T₂s, with values of 21ms[16,29], 18ms[14,26] and 14ms[12,20](Fig.5). While there are clear regional areas of elevated/reduced Gaussian broadening for both gas and M resonances across, there was not a clear consistent pattern in any particular area of the lungs across the HVs(Fig.4).

Discussion

Gaussian broadening was found to contribute ~75-80% and ~45% to the total linewdith for gas and M resonances, respectively, in the lungs of HVs. While similar Gaussian broadening values have been observed from whole-lung spectroscopy for the M resonance,^2,3 this study quantifies Gaussian broadening of the gas resonance for the first time, revealing that, on average, Gaussian broadening dominates over Lorentzian broadening for ¹²⁹Xe gas in the lungs, most likely due to sampling inhomogenous B_­0 over the voxel. Incorporating Gaussian components for both gas and M resonances should improve spectral parameter quantification accuracy, and consequently enhance accuracy of disease metrics derived from these parameters. Although no consistent pattern of Gaussian broadening was observed over the lungs in HVs, reduced Gaussian broadening may be present in areas of alveolar damage in e.g. emphysematous lungs due to increased free diffusion of ¹²⁹Xe gas in these regions. In contrast, elevated Gaussian broadening in the M resonance may be present in fibrotic lungs due to increased B₀ inhomogeneity from regions with elevated tissue heterogenity. As such, fractional Gaussian broadening may provide an additional metric for assessing the severity of lung diseases.

Conclusion

We demonstrate regional quantification of ¹²⁹Xe Gaussian broadening in the lungs. Knowledge of this is expected to increase accuracy of computed spectral parameters for evaluating lung disease metrics and may also serve as a valuable metric in its own right.

Acknowledgements

This work was supported by MRC grant MR/M008894/1, the National Institute for Health and Care Research (NIHR) Sheffield Biomedical Research Centre (NIHR203321) and AMS Springboard award R/162501-1. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care.