Investigating biases in the measurement of apparent alveolar septal wall thickness with hyperpolarized 129Xe MRI

Abstract

Purpose:

To investigate biases in the measurement of apparent alveolar septal wall thickness (SWT) with hyperpolarized xenon-129 (HXe) as a function of acquisition parameters.

Methods:

HXe MRI scans with simultaneous gas- (GP) and dissolved-phase (DP) excitation were performed using 1D projection scans in mechanically ventilated rabbits. The DP magnetization was periodically saturated, and the DP xenon uptake dynamics were measured at end inspiration (EI) and end expiration (EE) with temporal resolutions up to 10 ms using a Look-Locker-type acquisition. The apparent alveolar septal wall thickness was extracted by fitting the signal to a theoretical model, and the findings were compared to those from the more commonly employed Chemical Shift Saturation Recovery (CSSR) MRI spectroscopy technique with several different delay time arrangements.

Results:

It was found that repeated application of RF saturation pulses in CSSR acquisitions caused exchange-dependent GP saturation that heavily biased the derived SWT value. When this bias was reduced by our proposed method, the SWT dependence on lung inflation disappeared due to an inherent insensitivity of HXe DP MRI to thin alveolar structures with very short T2*. Further, perfusion-based macroscopic gas transport processes were demonstrated to cause increasing apparent SWTs with TE (2.5 μm/ms at EE) and a lung periphery-to-center SWT gradient.

Conclusion:

The apparent SWT measured with HXe MRI was found to be heavily dependent on the acquisition parameters. A method is proposed that can minimize this measurement bias, add limited spatial resolution, and reduce measurement time to a degree that free-breathing studies are feasible.

Introduction

Measuring hyperpolarized xenon-129 (HXe) transfer from the alveolar gas phase (GP) to the lung tissue, the so-called dissolved phase (DP) compartment, can provide important insights into pulmonary function (1–4). In particular, microscopic changes in the alveolar septal wall thickness (SWT) far below the spatial resolution of existing diagnostic imaging modalities are accessible to this technique in the form of altered cumulative xenon DP signal dynamics (5–8). HXe MRI can therefore offer valuable information about lung pathologies associated with parenchymal changes which complements that provided by existing clinical imaging techniques such as chest x-rays, CT or PET, allowing us to more comprehensively characterize disease progression or treatment efficacy.

The DP imaging techniques currently in use can be divided into two general types: 1) static measurements of the regional xenon distribution within the lung parenchyma as dictated by the structure and physiology of the lung on one hand (9–19), and the MR acquisition parameters, in particular the flip angle and the TR (20–22), on the other hand; 2) dynamic measurements that capture the xenon gas uptake by the lung tissue and transport by the pulmonary circulation as a function of time (3,9,23–30). Both approaches have already been shown to be sensitive to various forms of lung disease (31–42). More recently, efforts have been made to combine these methods by adding 2D or 3D spatial information to dynamic gas uptake measurements at the expense of lengthened scan times (43,44). Nevertheless, although xenon uptake MR techniques for quantifying lung function have been in use for over 20 years, little attention has been paid to the impact of the acquisition parameters on the metrics extracted.

Imaging the xenon DP signal in the lung differs from GP imaging, and even more so from conventional proton MRI, in several important aspects. Because xenon magnetization cannot recover through T1 relaxation once hyperpolarized, its inevitable decay needs to be carefully managed throughout the image acquisition process. Given that the DP signal in the lung is approximately 50 times lower than the GP signal, and given that the DP T2* at field strengths of 1.5 T is only on the order of 1–2 ms, DP imaging would seem futile were it not for the rapid gas exchange processes between DP and GP magnetization pools within the alveoli. As the DP magnetization is depleted by RF pulses, it is continuously replenished from the GP compartment.

While this signal recovery through gas exchange has made pulmonary DP imaging feasible, the problem with most existing DP imaging techniques is their implicit assumption of a steady state condition during which the acquisition takes place. This is generally not the case, however, and two mechanisms have the potential to be particularly troublesome: 1) As the DP magnetization recovers through alveolar gas exchange, the GP magnetization is decreased such that regions with high gas exchange experience a more rapid GP signal depletion during imaging than those with low gas exchange (45,46). 2) In addition to the passive diffusional gas exchange, pulmonary circulation provides active gas transport processes that move xenon magnetization from the lung periphery towards the lung center and into the left ventricle of the heart through a network of large pulmonary veins. If not properly accounted for, both effects can introduce undesirable biases into the measurement.

In this work, we implemented a 1D simultaneous DP-GP imaging method with limited left-right spatial encoding but high temporal resolutions of up to 10 ms. By using a Look-Locker technique (47) similar to what has been proposed by Kern et al. (43) to sample the DP signal recovery after DP saturation with 90° RF pulses, we were able to monitor the dynamics of pulmonary xenon gas exchange and transport in much greater detail than previously reported in the literature. More specifically, we investigated how flip angle, TR and active gas transport impact the measured apparent alveolar septal wall thickness in rabbits, and how these measurements compared to similar acquisitions obtained with the more commonly employed chemical shift saturation recovery (CSSR) MRI spectroscopy technique.

Methods

Animal Studies

Experiments were performed using five New Zealand rabbits (3.5 – 4.5 kg each). The animals were sedated with 25–35 mg/kg ketamine and 5 mg/kg xylazine intramuscularly, and 1–5% isoflurane was administered through a mask to maintain deep anaesthesia throughout animal preparation. A peripheral vein was cannulated to maintain general anaesthesia during imaging. Following anesthetization, a tracheotomy tube was inserted using an aseptic surgical procedure and secured using a silk ligature.

After induction of anesthesia, the rabbits were placed in a xenon RF coil (described below) and attached to a home-built, MRI-compatible mechanical ventilator. Anesthesia was maintained throughout the imaging session by an infusion of Propofol (20–80 mg/kg/h); blood pressure was stabilized by an infusion of saline, and body temperature was supported by a circulating warm water pad. Animals were euthanized at the end of the imaging procedure. All experiments were approved by and performed in accordance with the guidelines established by the University of Pennsylvania Institutional Animal Care and Use Committee (IACUC) and the National Institutes of Health (NIH) guidelines for the care and use of laboratory animals.

Imaging was performed on a 1.5 T commercial whole-body scanner (Magnetom Avanto; Siemens Medical Solutions, Malvern, PA, USA) that had been modified by the addition of a broadband amplifier, permitting operation at the resonant frequency of 17.6 MHz. The RF coil was a custom xenon-129 transmit/receive birdcage design (Stark Contrast, Erlangen, Germany), positioned to cover the whole chest of the animal. Low-resolution proton MR scout images were obtained with the built-in body coil.

Gas Polarization and Administration

Enriched xenon gas (87% xenon-129) was polarized by collisional spin exchange with an optically pumped rubidium vapor using a prototype commercial system (XeBox-E10; Xemed, LLC, Durham, NH) that provided gas polarizations of 40–50%. Immediately before MR data acquisition, HXe gas was dispensed into a Tedlar bag (Jensen Inert Products, Coral Springs, FL) inside a pressurizable cylinder that was subsequently connected to and controlled by the ventilator. Animals were ventilated with room air until the beginning of the imaging study, at which time the gas mix was switched to 20% oxygen and 80% HXe (6 ml/kg tidal volume). For breath hold studies, the animals inhaled the xenon gas mixture for up to 5 breaths, then ventilation was suspended for up to 12 s during data acquisition either at end inspiration (EI) or end expiration (EE) before ventilation with room air was resumed. For free breathing experiments, the ventilation gas was switched to the xenon gas mixture for up to 12 consecutive breaths with data collected throughout.

HXe Data Acquisition

The RF excitation flip angle was calibrated in an initial 5 s breath hold, during which 32 xenon spectra were acquired. The ratio between the nominal and the measured flip angle was used to set the reference voltage on the scanner for all subsequent studies as described in (22).

For the studies in this work, CSSR spectroscopy, 1D projection and 2D projection data sets were acquired. During CSSR spectroscopic measurements, 90° Gaussian RF pulses (3-ms duration) centered at 200 ppm were applied to saturate the DP resonances. Following a variable delay time ranging from 5.9 to 500 ms, a 0.7-ms Gaussian RF excitation pulse also centered at 200 ppm was used to generate a free induction decay. This sequence was repeated with up to 40 different delay times during a single breath hold. The signal was sampled for 30.72 ms, with 1024 sampling points apodized by a squared cosine function and zero-filled to 2048 points. Four different delay time orders (CSSR #1: 9 sequentially increasing delay times; CSSR #2: 40 sequentially increasing delay times with small delay time increments at the beginning and larger increments towards the end of measurement; CSSR #3: 40 sequentially increasing delay times with a fixed increment of 10 ms; CSSR #4: 40 sequentially decreasing delay times with a fixed decrement of 10 ms) were implemented. The dissolved- and gas-phase resonances were integrated numerically. To evaluate the TE dependence of the SWT measurements with CSSR #1, only 924 sampling points were included in the analysis. The effective TE was modified by shifting the selection window within the acquired 1024 sampling points.

2D axial projection images with simultaneous DP-GP excitation were acquired as described previously (11,20,22). Following 5 breaths of xenon gas mixture an 8-second breath hold was induced at EE. To destroy any DP magnetization taken up prior to the data collection and thereby achieve a steady state condition for the DP signal, the sequence was first preceded by two 2-ms Gaussian RF saturation pulses. Next, a series of 700-μs Gaussian RF excitation pulses was applied for 1.5 s using the same TR as the image acquisition. All preparatory RF pulses were also centered at 200 ppm. The sampling was 65% asymmetric with a bandwidth of 120 Hz, which, at the main field strength of 1.5 T and a gyromagnetic ratio for xenon-129 of 11.78 MHz/T, yielded a 30-pixel separation between the GP and DP images in the readout direction. Other sequence parameters included: matrix size 28×80 (interpolated to 112×320 using bilinear image interpolation for display purposes); TR/TE 20/2.6 ms; FOV 220×220 mm². To investigate the distribution dynamics of the DP magnetization, excitation flip angles of 90°, 53°, 37°, 23° or 16° were used, resulting in acquisitions with a TR_90°,equiv (22) of 20, 50, 100, 250 or 500 ms, respectively. Each image acquisition was repeated 10 times during the same breath hold and all measurements exceeding a signal-to-noise ratio of approximately 2 were averaged. All images were rescaled to compensate for the differences in GP and DP flip angles as described previously (22).

The 1D projection sequence had a similar structure to the 2D version, with frequency encoding in a left-to-right direction but without any phase encoding. Instead, multiple measurements of the same 1D projection line permitted observation of the DP signal dynamics with a temporal resolution of the acquisition TR. To sample the dynamics of the DP magnetization accumulation, multiple DP saturation segments consisting of 1–4 90° Gaussian RF pulses were applied every 300 (multi-breath studies) or 500 ms (breath hold studies).

Data Analysis

All image reconstruction, post-processing and data analysis were performed using customized MatLab (R2018a, Mathworks, Natick, MA, USA) scripts. The asymmetrically sampled k-space data of the 2D projection measurements was filled using a Homodyne algorithm (48) before Fourier transform. The DP saturation recovery studies performed with the 1D projection sequence were analyzed line-by-line, either by averaging lines in the left versus the right lung or by averaging all lines within the lung volume.

For qualitative analysis, EE and EI studies were normalized by dividing the DP-GP ratios at all delay time points by their value at 100 ms. Prior to regional qualitative analysis, all spatial projection lines were normalized by the mean signal in the first 10% of sampling points.

The average SWT thickness for different respiratory stages during multi-breath studies was evaluated by dividing the GP signal into 4 bins covering the range between minimum and maximum GP amplitude. Within each GP signal bin, the associated DP signal collected simultaneously was averaged for each delay time. The averaged DP signal recovery curve for each bin was fitted to an analytical gas-uptake model (6).

Results

Figure 1 shows the GP (Figure 1A) and DP (Figure 1B) signal distributions in 2D axial projections of a rabbit lung for TR_90°,equivs ranging from 20 to 500 ms, achieved by reducing the flip angle at the DP resonance from 90° to 16° while holding the TR constant. For high flip angles (short TR_90°,equivs), the visible GP signal is mostly constrained to the airways due to exchange between the largely depolarized DP magnetization and the alveolar GP magnetization. As the flip angle decreases, this effect is diminished and GP signal becomes apparent throughout the lung parenchyma. Simultaneously, increasing TR_90°,equivs means that more time is available for DP magnetization to accumulate. At the 500 ms time point, DP signal is clearly visible in the left ventricle of the rabbit heart.

Figure 1.

2D axial GP (A) and DP (B) projections of the rabbit lung for a constant TR of 20 ms and DP flip angles ranging from 90° (left) to 16° (right), resulting in effective TR_90°,equivs ranging from 20 to 500 ms. The DP flip angle was approximately 40 times higher than the effective GP flip angle. As the flip angle decreases (TR_90°,equiv increases), the GP depolarization due to gas exchange with the DP is reduced, since more downstream xenon magnetization can accumulate before being completely saturated by the imaging RF pulses.

To demonstrate that the GP depolarization apparent in Figure 1A also affects SWT measurements, a modified CSSR #1 data set with sequentially increasing delay times following the DP saturation was acquired (Figure 2A). The 50 ms delay was measured repeatedly during breath hold (green circles). While the DP-GP ratio for the same delay time would ideally be constant, a clear downward drift is visible in the data. The same phenomenon becomes apparent when plotting the DP-GP ratios from multiple CSSR experiments with different delay time orders collected from the same animal (Figure 2B). The dark blue line with circle markers (CSSR #1) reflects a sequential CSSR study with 9 delay times and thus only 9 DP saturation blocks. The associated sampled DP-GP ratio is higher than almost all ratios measured with the other three tested CSSR acquisitions that use 40 DP saturation blocks. However, when focusing only on the sample points that were collected early in the acquisition (short delay times in CSSR #2 and #3, long delay times in CSSR #4), there is excellent agreement with the CSSR #1 measurement. Consequently, when fitting the acquired data points to a theoretical uptake model, the extracted SWTs differs widely, ranging from 10.7 μm for CSSR #1 over 7.7 μm and 8.6 μm for CSSR #2 and #3, respectively, to an astonishing 23.3 μm for CSSR #4.

Figure 2.

DP-GP ratio in a CSSR study with 40 spectra of varying delay times between DP saturation and FID acquisition (A). The 50 ms delay was repeated multiple times (green circles), and the associated DP-GP ratio decreased throughout the breath hold due to increased GP depolarization in lung regions with high gas exchange. The same effect resulted in a large variation in delay time dependence of the DP-GP ratio as a function of the order in which the delay times were sampled (B).

The effective TE of the spectroscopic acquisition can be lengthened by removing sampling points at the beginning of the FID. For a 500 ms delay time at EE in rabbits, the hemoglobin and tissue/plasma resonances appear merged into a single, broad peak at short TEs (Figure 3A). As the TE is increased, the peak amplitudes decrease but the two resonances become discernible. In parallel, using the CSSR #1 acquisition scheme, the derived apparent SWT increases linearly at a rate of 2.5 ± 0.12 μm/ms (R² = 0.96) at EE and of 1.5 ± 0.13 μm/ms (R² = 0.67) at EI (see representative measurements in Figure 3B). However, as reflected in the reduced R² value, the linear correlation is less convincing at EI.

Figure 3.

(A) Spectra of the DP region in a rabbit acquired with a 500 ms delay time after an RF saturation pulse as a function of echo time. At the shortest TEs, the two DP resonances associated with xenon bound to hemoglobin and dissolved in lung tissue and blood plasma heavily overlap but separate into easily distinguishable peaks as the TE is increased. (B) Representative plot of the apparent septal wall thickness at EE and EI obtained from a healthy rabbit by fitting an analytical gas uptake model to CSSR datasets with different spectral echo times. The fitted regression line in this animal has a slope of 2.6 μm/ms at EE and 1.4 μm/ms at EI. It intercepts the y-axis at 12.7 μm (R² = 0.96) at EE and at 13.1 μm (R² = 0.67) at EI.

A simultaneous DP-GP acquisition was implemented to add a modest amount of spatial information to the xenon gas-uptake measurements. Figure 4A shows a 2D axial projection variant of this technique, as previously described in the literature (11,20,22,49). Performing the same acquisition scheme without phase-encoding gradients collapses the 2D maps in Figure 4A along the anterior-posterior direction, yielding a 1D projection with right-to-left spatial encoding and a temporal resolution of the chosen TR (Figure 4B). The dark horizontal lines signify the application of DP saturation blocks every 500 ms, after which the DP magnetization recovery due to exchange with the largely undisturbed alveolar GP magnetization pool is sampled with a series of low flip angle pulses. Using fewer than three saturation pulses per segment resulted in insufficient saturation, with the minimum DP-GP ratio after DP saturation drifting downward for the first few saturations. For three or more saturation pulses, however, the DP-GP ratio remained sufficiently stable. Three 90° RF saturation pulses were applied during each saturation block for all subsequent studies. After the final DP saturation, a pulsatile DP signal appears in the prior signal void between right and left lung (white arrow), corresponding to DP xenon reaching the left ventricle.

Figure 4.

The proposed 1D simultaneous DP-GP imaging technique is based on an equivalent 2D projection (A) without phase encoding. Instead, the 1D measurement is repeated continuously in order to capture the temporal dynamics of the GP and DP signals throughout the breath hold and respiratory cycle with a temporal resolution of the selected TR of 50 ms (B). By periodically saturating the DP magnetization (red arrows), every 500 ms in this acquisition, the subsequent DP signal recovery can be assessed. Once the delay time after the final DP saturation approaches the transit time of gas transport from the alveolar airspaces to the left ventricle of the heart, pulsatile DP signal oscillations (white arrow) can be observed in the cardiac cavity.

As first noted by Kern et al. (43) and investigated in more detail by Ruppert et al. (22), the behavior of xenon magnetization accumulation in the DP is determined by the TR_90°,equiv imparted by the repetition rate and flip angle of the RF pulses. Figure 5A illustrates the effect of the TR_90°,equiv of the data sampling on the observed total xenon gas uptake in the rabbit lung averaged over 5 saturation-recovery cycles. Once TR_90°,equiv reaches about 1.3 s, the dependence of the uptake curve on TR_90°,equiv is significantly diminished. Most measurements were conducted at a TR_90°,equiv of 1.34 s (flip angle 7°, TR 10 ms) as a compromise between a high temporal resolution and a sufficient signal-to-noise ratio. However, Figure 5B shows that different flip angle – TR combinations with the same TR_90°,equiv did not yield exactly the same DP-GP ratio curves. Instead, acquisitions with larger flip angles resulted in slightly lower DP-GP ratios than those with smaller flip angles. Another important parameter affecting xenon gas uptake was the lung inflation level at which it was measured. At EI, the lung tissue density is lower than at EE and the DP-GP ratio varied accordingly. Further, as demonstrated by the fitted CSSR uptake curves in Figure 5C, the DP-GP ratio dynamics at EE displayed a notable qualitative difference depending on the selected sample order, while they were heavily overlapping at EI. A similar behavior was observed for the fitted uptake curves of selected 1D acquisitions (Figure 5D), although with better qualitative agreement at EE than for the CSSR studies.

Figure 5.

The measured DP-GP ratio as a function of the acquisition parameters. For all 1D measurements TR_{90°, equiv}, flip angle and TR are listed in the respective figure legends. (A) Spatio-temporally averaged 1D acquisitions at EE with different TR_90°,equiv values. (B) Spatio-temporally averaged 1D acquisitions with approximately equal TR_90°,equiv but different flip angle – TR combinations. (C) Fits of CSSR spectroscopy measurements with two different sampling arrangements at EE and EI. (D) Fits of spatio-temporally averaged 1D acquisitions with 4 different flip angle – TR combinations at EE and EI.

Qualitative differences in the xenon uptake curves at EI and EE become more apparent when normalizing the fitted DP-GP ratio curves to their value at 100 ms. The normalized CSSR measurements (Figures 6A and 6B) exhibit clear qualitative differences in uptake behavior over the entire 500 ms measurement range. Interestingly, the EE and EI curves for the four 1D uptake measurements (Figures 6C–6F) strongly overlapped at short delays but diverged towards longer delay times for flip angles of 16° (Figure 6C) and 11° (Figure 6D). This discrepancy at longer delay times vanished for the 7° acquisition in Figure 6E, however, although all three measurements had a TR_90°,equiv of 1.34 s. When further increasing the TR to 30 ms and the TR_90°,equiv to 4.02 s (Figure 6F), the uptake curves at EI and EE were qualitatively identical, effectively removing any observable SWT difference between EI and EE.

Figure 6.

Qualitative comparison of fitted xenon uptake measurements at EE and EI. The fits were normalized by the DP-GP ratio at 100 ms. For all 1D measurements TR_{90°, equiv}, flip angle and TR are listed in the respective figure legends. (A) CSSR study with 8 saturation segments and sequentially increasing delay times. (B) CSSR study with 40 saturation segments and sequentially increasing delay times. (C-F) Spatio-temporally averaged 1D acquisitions with 4 different flip angle – TR combinations.

Figure 7A depicts the DP signal dynamics across the lung for a TR_90°,equiv of 4.02 s with the cardiac cavity zeroed out; the signal for each line has been normalized by the mean signal in the first 10% of sampling points for ease of qualitative comparison. The relative xenon gas uptake increases from the periphery towards the lung center. For further emphasis, only the most peripheral (thin lines) and most central xenon uptake curves for both lungs were plotted in Figure 7B. While the xenon accumulation in the periphery of the lung leveled off following the initial filling phase, the xenon signal at the center of the lung kept increasing. These discrepancies in gas accumulation dynamics were also reflected in an apparent drop in extracted capillary transit time (Figure 7C) and an apparent increase in alveolar SWT (Figure 7D), respectively, from the periphery towards the center.

Figure 7.

(A) Temporally averaged 1D (right-to-left) xenon uptake by the lung parenchyma normalized to the first 10% of sampling points for qualitative comparison. The signal in the cardiac cavity was zeroed out due to low signal levels following DP saturation. (B) Normalized xenon gas uptake in the most peripheral (thin lines) and most central (thick lines) locations in (A). Regional capillary transit time (C) and apparent SWT obtained by fitting a theoretical gas uptake model to (A).

The spacing of the DP saturations ultimately determines the maximum temporal frequency with which SWT can be assessed. For instance, a DP saturation every 300 ms allows us to perform multiple SWT measurements in each respiratory cycle (Figure 8A, TR_90°,equiv 1.34 s). However, fitting the DP signal to an analytical gas uptake model is complicated by the fact that the signal not only changes due to xenon magnetization entering the DP phase, but also due to variations in lung tissue density during respiration. As a result of these dynamic processes, the derived SWTs appear too thin during inspiration, when gas absorption is compensated for by decreasing tissue density, and too thick during expiration. By dividing the DP signal into 4 bins based on the GP signal level (Figure 8A) and fitting the binned data individually, this respiratory bias was largely alleviated. Confirming the findings presented in Figure 6, the apparent SWT for the left, right and total lung remained almost entirely unchanged for all levels of lung inflation in the respiratory cycle (Figure 8B).

Figure 8.

(A) Total GP and DP signal during multiple consecutive xenon breaths. The DP signal was saturated every 300 ms to permit assessment of the apparent SWT throughout the respiratory cycle. To this end, the DP signal was aggregated in 4 bins (EE, 20–50% lung inflation, 50–75% lung inflation, EI) according to the GP signal level at the time of data sampling. (B) Extracted apparent SWT in the left, right and total lung at four different lung inflation levels following binning of the DP signal based on 4 GP signal quartiles.

Discussion

In this study, we investigated how MRI acquisition parameters and physiological processes can introduce bias into the measurement of apparent alveolar SWT. Global CSSR measurements were compared with those of a Look-Locker type acquisition scheme combined with simultaneous DP-GP sampling. The latter resulted in a technique with modest 1D spatial information but high temporal resolution of up to 10 ms. Instead of sampling pulmonary xenon gas uptake one data point at a time, as in conventional CSSR spectroscopy, the 1D method sampled the entire gas uptake curve after each DP saturation. This allows averaging over multiple saturation-recovery cycles to compensate for signal fluctuations related to the heart beat, for example, or the evaluation of the septal wall thickness at multiple time points throughout the respiratory cycle.

Perhaps the most unexpected factor affecting SWT measurements is the TE (Figure 3). As the TE is lengthened, the two DP resonances become more distinct due to the removal of short-T2* DP signal components. Simultaneously, the apparent SWT at EE thickens at a rate of 2.5 ± 0.12 μm per ms of TE increase in rabbits. These findings suggest that thinner wall components have a gas exchange-related shorter T2*, and that the SWT measurement has a TE-dependent sensitivity limit towards the thinnest SWT that contributes to the mean value calculated for the whole lung. Since the STW distribution is reflective of the underlying anatomical lung structure and xenon tissue solubility, it is possible that the slope of this TE-dependence is species specific and might change in lung disease. This result would also indicate that the apparent SWT value is B0-dependent. Consequently, since spatially encoded acquisitions generally have longer TEs than spectroscopy measurements, we predict that SWT maps will shower a thicker mean apparent SWT than their spectroscopic counterparts.

In a typical SWT measurement, post-saturation DP signal accumulation as a function of time is monitored for a period of several hundred milliseconds to about 1 second. In adult rabbits, a significant amount of xenon is transported from the alveolar volume to the heart during the studied 500-ms interval, as evidenced by a strong DP signal in the left ventricle (Figure 1B). However, as the 2D DP images in Figure 1A indicate, the RF pulses with high DP flip angles typically used for DP saturation in gas-uptake experiments have a large impact on the residual GP signal. This well-known selective GP signal depletion is greatly exaggerated by the steady-state preparation prior to image acquisition, but the underlying mechanism has long been exploited by the Xenon-polarization Transfer Contrast (XTC) technique (45,50–52) for contrast generation. Unfortunately, as shown in Figure 2, the XTC effect has undesirable consequences for measurements that assume a constant DP-GP ratio under identical conditions. These problems become especially apparent in HXe techniques that depend on the application of numerous high flip angle RF pulses, such as conventional CSSR studies, where one or more high flip angle RF pulse is used to saturate the DP magnetization (we applied three in our study) prior to acquiring each sampling point.

If gas exchange in the lung were spatially homogeneous, exchange-induced GP depolarization would simply result in an accelerated drop in signal-to-noise ratio over the course of the measurement. However, pulmonary gas exchange can be highly inhomogeneous both micro- and macroscopically even in healthy subjects, with large differences in surface-to-volume ratios between airways and alveoli, as well as significant gravity-induced differences between dependent and independent lung volumes. In a CSSR spectroscopy measurement under such conditions, the GP magnetization in the dependent lung depletes much faster than in the independent lung, resulting in a spatial selection bias that shifts throughout the breath hold. As the DP-GP ratio in the independent lung is typically much lower than in dependent volumes, late sampling points exhibit a reduced DP-GP ratio, pull down the gas uptake curve, and significantly affect the derived SWT in a sampling-order-dependent manner: 1) acquiring the short delay times early results in a steep increase in the DP-GP ratio that levels off towards longer delay times, meaning the extracted SWT will be thinner; 2) acquiring the short delay times late in the measurement generates a shallower slope and produces a thicker extracted SWT (Figure 2B).

Post-saturation, the DP magnetization signal recovers as the magnetization is replenished through exchange with the alveolar GP. Since our dynamic 1D technique acquires the entire gas uptake curve after each DP saturation, only one uptake measurement is needed in principle. However, to increase the signal-to-noise ratio and average out cardiac oscillations (53–55), we performed up to six saturation-recovery cycles per breath hold study. When measuring the DP signal build up with a series of RF excitation pulses, a fraction of this fresh longitudinal magnetization is consumed by the pulses, pushing the observed DP-GP ratio below its true value. We previously introduced the parameter TR_90°,equiv, which converts the flip angle and TR of an acquisition into a more intuitive equivalent xenon accumulation time (22), as a metric for this effect. As illustrated in Figure 5A, the selection of a sufficiently long TR_90°,equiv is crucial for obtaining an accurate xenon gas uptake curve. Nevertheless, as in the CSSR studies discussed above, the exchange-induced depletion of GP magnetization was also noticeable in the proposed 1D acquisition scheme, albeit to a lesser degree (see Figure 5B). This effect was particularly apparent at EE (Figures 5C and 5D), when the differential in pulmonary gas exchange between dependent and independent lung volumes was largest. Importantly, the same bias could further increase in the case of lung disease, when reduced lung compliance or inflammatory/fibrotic thickening of alveolar membranes increases regional GP depolarization even at EI.

Continuous measurements of apparent alveolar SWT taken every few hundred milliseconds can be used to study non-cooperative or severely ill subjects who cannot reliably hold their breath under free-breathing conditions. We successfully demonstrated this approach in a rabbit ventilated with HXe gas for several seconds without breath hold (Figure 8). After binning the DP data to reduce the impact of the lung tissue density variations throughout the respiratory cycle, this study confirmed the results of the breath hold studies (Figure 6)—namely, that apparent SWT assessed by xenon gas uptake in rabbits seemed insensitive to lung inflation. Therefore, as long as the acquisition time is short compared to the respiratory cycle, even a single inhalation of xenon without a breath hold will provide a reasonably reliable reading of apparent SWT.

It is important to resolve the question of whether or not the apparent alveolar SWT varies as a function of lung inflation because, if a significant dependence does exist, a standardized breathing maneuver will be necessary to make inter-subject measurements fully comparable. While Stewart et al. (26) have reported a large increase in SWT in humans from approximately 7.5 μm to 11 μm with decreasing lung inflation, our findings in rabbits suggest that at least some of the observed change could have been caused by exchange-induced measurement biases inherent to current CSSR implementations. Nevertheless, high-resolution anatomical scans using synchrotron X-ray imaging clearly demonstrate oscillatory changes in alveolar size in parallel with respiration (56). A solution to this apparent contradiction might be the deduced reduction of T2* in thinner septal walls discussed above: as the walls get thinner during inspiration, their signal contribution decreases such that the signal-weighted mean SWT remains largely unaffected.

Regardless of lung inflation level, the apparent SWT was always considerably larger than what would have been expected based on the histological value of approximately 5 μm in adult rabbits (57). There are most likely several factors that contribute to this apparent deviation, making it difficult to validate the HXe-derived metrics with existing tools: 1) Extracting the apparent SWT from the measured time constant of xenon gas accumulation in the lung parenchyma requires knowledge of the xenon diffusion constant, which is neither precisely known nor necessarily constant across the entire septal wall. 2) Xenon accumulates everywhere in the lung tissue, not just in the septal wall. 3) Rather than being uniform, the xenon tissue distribution is dictated by its differing solubilities in the various cell compartments. 4) Binding and exchange processes further modulate the effective xenon diffusion constant. 5) The exchange-dependent T2* is shorter in thinner structures than in thicker ones, which increases the signal intensity-weighted mean of all thicknesses.

Extending global CSSR spectroscopy to include spatial information would remove some of the inherent volume averaging and could greatly enhance the technique’s sensitivity to heterogenous lung disease patterns. Yet all existing theoretical xenon gas uptake models assume that the measured DP signal build-up is solely due to xenon magnetization transfer from the alveolar volume to the lung tissue. However, it is not feasible to segment out the pulmonary veins that transport xenon-saturated blood to the left ventricle of the heart. Consequently, the potential impact of xenon transport effects on the assessed DP signal build up rises the closer an image voxel is to the lung center, as shown in Figures 7A and 7B. As no outside-in asymmetry of transit time and SWT seems plausible in a healthy lung, the existence of such an asymmetry in Figures 7C and 7D supports the hypothesis that the unaccounted DP signal boost from circulation-based gas transport processes introduces a significant measurement bias. Extracting consistent SWT maps from regional xenon gas uptake measurements might therefore necessitate a fundamental revision of these models.

Interestingly, XTC-derived SWT quantifications (50,58) should be unaffected by gas transport effects and could thus be used as a reference metric. Further, the magnitude of the transport bias on SWT measurements is likely to be species-dependent. In larger animals and adult human subjects, for example, the time required to accumulate and transport significant quantities of xenon magnetization from the periphery towards the center of the lung is likely too long to be readily apparent on the time scale of one second. In small animals, infants, or possibly even young children, on the other hand, circulatory gas transport could cause large distortions in regional SWT measurements.

All studies were conducted in healthy rabbits, which do not display the same clear spectral separation of the two DP resonances typically observed in rats, pigs or humans. Further, a simultaneous GP-DP acquisition technique was employed for all 1D measurements that, while convenient, places constraints on the maximum practical imaging bandwidth and spatial resolution. While the actual impact of the observed parameter dependencies of the derived physiological metrics on overall measurement sensitivity are unknown, all of the presented qualitative findings are general in nature, are not tied to the specifics of the subject species or acquisition technique, and can be extended to other DP imaging methods. For instance, the Look-Locker method is easily implemented within the framework of global CSSR spectroscopy, and the rapid 1D imaging can be performed with other gradient echo-based MRI techniques using a sufficiently large bandwidth to contain the two DP resonances within a single voxel. Regardless of the implementation details, the proposed methods can greatly increase the speed and accuracy of apparent alveolar SWT measurements.

Conclusions

We proposed a rapid 1D acquisition technique to measure apparent alveolar SWT faster and more reliably than current CSSR technique implementations. This type of measurement could be performed quickly enough that no breath hold would be required. We also found that both the choice of acquisition parameters and circulatory xenon gas transport unaccounted for in current theoretical gas uptake models can introduce large biases in measured SWT.