Dissolved-phase hyperpolarized xenon-129 imaging provides regional information on gas exchange between the lung airspaces, parenchymal tissues, and blood stream. Current spectroscopic techniques require that the dissolved-phase 129Xe signal be acquired with no gas-phase excitation. However, the short T2* of the dissolved-phase requires short RF pulses, limiting the spectral selectivity achievable. This, combined with the high spin density of the gas-phase relative to the dissolved-phase, leads to unwanted gas-phase excitation. In this work, we retrospectively remove contaminant gas-phase signal from both simulated and human subject images using a multi-echo acquisition with iterative estimation of the contaminant gas-phase signal.
Spectroscopic imaging of hyperpolarized (HP) xenon-129 dissolved in the lung parenchyma and bloodstream provides regional information on gas exchange [1]. Techniques used to separate HP 129Xe in the barrier tissues and red blood cells (RBC) require signal be acquired without gas-phase excitation [2,3]. However, gas-phase excitation can occur despite careful spectrally selective pulse design due to RF pulse sidelobes that weakly excite the higher density gas signal.
Previously, we presented a technique to retrospectively remove gas-phase signal from dissolved-phase images, estimating the contaminant gas signal using an additional delayed echo acquisition [4]. This method was shown to successfully limit contamination by gas signal to <10% of the dissolved-phase signal, but accuracy was limited by residual dissolved-phase signal in the delayed echo. In the present work, we present an improved multi-echo technique that refines the estimate of gas-phase contaminant signal using a more complete signal model. This technique results in dissolved-phase spectroscopic images with <1% contamination from gaseous HP 129Xe signal, improving performance by an order of magnitude compared to the prior method.
Theory: We simultaneously acquire multi-echo data at both the gas-phase and dissolved-phase resonant frequencies, as described in [2]. The acquired dissolved-phase data, contaminated with gas-phase signal, can be modeled as follows.
$$s_{d,raw}(\vec{k},t)=s_{d,true}(\vec{k},t)+\alpha s_{gas}(\vec{k},t)e^{2\pi i \Delta ft}$$
$$$s_{d,raw}$$$ and $$$s_{gas}$$$ are the acquired dissolved- and gas-phase data, respectively; $$$\alpha$$$ is a correction factor accounting for the change in excitation of the gas between the dissolved- and gas-phase acquisitions; and $$$\Delta f$$$ is the change in the receive frequency between the dissolved- and gas-phase acquisitions. If $$$\alpha$$$ can be properly estimated, the contaminating gas signal can be subtracted from the dissolved-phase data.
The dissolved-phase signal $$$s_{d,true}$$$ can be expanded in terms of the RBC and barrier components.
$$s_{d,raw}(\vec{k},t)=\rho_{rbc}(\vec{k})e^{(2\pi if_{rbc}-r_{2_{rbc}}^*)t}+\rho_{bar}e^{(2\pi if_{bar}-r_{2_{bar}}^*)t}+\alpha s_{gas}(\vec{k},t)e^{2\pi i\Delta ft}$$
$$$\rho_{rbc}$$$ and $$$\rho_{bar}$$$ are the Fourier transforms of the RBC and barrier spin densities, respectively; and $$$f$$$ and $$$r_2^*$$$ are the frequency shifts and transverse relaxation rates of the components, estimated from global spectroscopy performed prior to imaging.
From this model, we iteratively solve for $$$\rho_{rbc}$$$, $$$\rho_{bar}$$$, and $$$\alpha$$$ until $$$\alpha$$$ converges, as described in Figure 1. Contaminant gas-phase signal may then be subtracted from the dissolved-phase signal based on the equations above.
Simulation: To analyze the performance of this technique, a digital phantom was used to generate gas- and dissolved-phase signals based on a 3D-radial acquisition with 4 gradient echoes (TEs=0.9ms, 2.0ms, 3.1ms, 4.2ms). Early results revealed that accuracy was increased by limiting the estimation to the more fully sampled central region of k-space. As such, we tested the performance against the region of k-space included in the estimation.
Human Subject Imaging: Gas- and dissolved-phase MRI was acquired at 1.5T in a single 15s breathhold using 1L of HP 129Xe and a 3D-radial sequence with 4 echoes. A 1.2ms SLR RF pulse with a 5kHz bandwidth was designed to minimize off-resonance transmit power. Transmit and receive frequencies were alternated between the gas and RBC resonances, acquiring the same projection angles in k-space for both frequencies [2].
Simulated data indicate the accuracy of this technique is highly dependent on the region of k-space included in the fit, seen in Figure 2, with a minimum error when including the central 40% of the data. Errors increase above 2% in extreme cases of noise (SNR<5), but remain robust for SNR>5. The performance of this technique is unaffected by changes in the contaminant signal. For dissolved-phase images with SNR>7, simulated results indicate reduction of the contaminant gas-phase signal to <1% of the dissolved-phase signal.
As shown in Figure 3, this technique removes the artifacts caused by the off-resonance gas-phase signal. Qualitatively, the performance appears similar to the previously reported two-echo technique. However, there is a measureable difference between the two techniques in agreement with the difference estimated in the simulated results.