Magnetic resonance imaging of the time course of hyperpolarized ¹²⁹Xe gas exchange in the human lungs and heart

Hyperpolarized xenon-129 (HPX) magnetic resonance imaging (MRI) may be used to evaluate pulmonary ventilation, in patients with respiratory diseases such as chronic obstructive pulmonary disease (COPD) and asthma [1‐8]. Due to the solubility of xenon, HPX MRI may also provide additional functional information about the gas exchange efficiency in the alveolar epithelium, the interstitium, the pulmonary capillary endothelium, and the blood plasma of the lungs [9]. At clinical MR magnetic field strengths (1.5–3 T), the dissolved xenon gas splits into two chemical shifts corresponding to the pulmonary tissue and blood plasma (PTBP) and red blood cell (RBC) compartments of the lung. The vast majority of the previous HP dissolved xenon studies focused on the measurement of diffusion uptake (i.e., gas-transfer dynamics) from xenon in the alveoli into the PTBP and RBC compartments of the entire thoracic cavity using MR spectroscopy techniques [10‐12]. These techniques allow measurement of lung function when used in corroboration with numerical modeling [13‐17]. Recent HPX MRI techniques have enabled the spatial distribution of regional diseases, addressing the challenges arising from low signal-to-noise ratios (SNR) and short transverse signal decay time (T₂^* ≈ 2 ms) of PTBP and RBC peaks [18‐23]. Concurrent imaging of the dissolved phases of xenon and the measurement of gas-transfer dynamics would enable quantitative information about the efficiency or otherwise of gas exchange pulmonary function.

An early dissolved phase imaging technique, the xenon polarization transfer constant (XTC) has been shown to be sensitive to measurement of gas-transfer dynamics combining with the spatial information [24, 25]. In contrast to MR spectroscopy, XTC enabled the indirect measurement of the dissolved phase signal from the estimation of depolarization of the reservoir gas-phase magnetization with respect to various delay times [26, 27]. Unlike other dissolved phase imaging techniques, XTC did not provide imaging of PTBP and RBC compartments of the lungs separately.

MR imaging with iterative decomposition of water and fat with echo asymmetry and least-square estimation (IDEAL) Cartesian k-space sampling has been reported to be an accurate technique for imaging the chemical shift of hydrogen protons in muscle and fat compartments [28]. The IDEAL technique has been used in the metabolic imaging of hyperpolarized (HP) ¹³C using a single-shot spiral k-space sampling, taking advantage of the relatively longer T₂^* (T₂^* ≈ 30 ms) [29]. This IDEAL with spiral k-space sampling technique was subsequently applied to HPX MR for imaging gas, PTBP, and RBC compartments in rat lungs [30]. The short T₂^* of ¹²⁹Xe in the PTBP and RBC compartments was compensated by using an insertable high-performance gradient system (maximum gradient amplitude, 300 mT/m; slew rate, 2500 T/m/s) [31, 32]. While this strategy provided quantitative measurement of the physiological parameters and dynamics of HPX gas diffusion from alveoli air sacs to the heart in a rat model [31, 32], it was difficult to obtain sub-centimeter pixel resolution in humans using this single-shot spiral sampling technique due to the limited gradient performance (maximum gradient amplitude, 50 mT/m; slew rate, 200 T/m/s) of clinical MR scanners.

In this study, HPX MRI single-shot k-space sampling IDEAL technique has now been extended to human lung imaging using multi-interleave spiral k-space sampling in a standard clinical gradient MR scanner. By taking advantage of rapid k-space sampling, the feasibility of capturing the time-series IDEAL gas, PTBP, and RBC compartment images in a single breath-hold across eight different time points has been investigated in six healthy volunteers and four patients with pulmonary emphysema.

Six healthy participants (aged 25–45 years) with no known cardiopulmonary disease, no evidence of emphysema or other pulmonary disease on CT, and with normal lung function were scanned using a 1.5 T MR scanner (Signa HDx Twin Speed, GE Healthcare) and a proton-blocked vest-shaped RF coil (Clinical MR solutions). Enriched xenon gas (87% ¹²⁹Xe) was polarized by a commercially available polarizer (Model 9300, Polarean with 10–15% polarization). Subjects were instructed to inhale 1.0 L of pure xenon gas following a protocol approved by the National Research Ethics Service Committee.

Conventional proton MR 3-plane localizer images were acquired using the standard fast gradient echo sequence and double inversion-recovery (DIR) black-blood imaging sequence to acquire a diagnostic display of the lung, cardiac, and mediastinal anatomy as previously explained [6]. Following the conventional proton scan, IDEAL HPX gas, PTBP, and RBC compartment images were obtained using a multi-interleave spiral k-space with a three-point IDEAL approach. Briefly, three-point IDEAL approach acquisition produced three different echo images (TE-I, TE-II, and TE-III) with three echo times (TE₁ = 0 μs, TE₂ = 160 μs, and TE₃ = 320 μs) which were calculated using the number of signal averaging approach [30]. Each of the echo images were acquired using 12 RF pulses and spiral interleaves for a particular echo time as shown in Fig. 1a using a standard clinical MR gradient. Imaging parameters were field of view. 34 cm; number of pixels, 32; bandwidth, 62.5 kHz; and readout time duration per interleaf, 4 ms. The scan time required for a single echo image was 12 × TR_min = 240 ms by setting TR_min to the minimum value of 20 ms as shown in Fig. 1a. The TR_min consisted of the time required for the slice selection, RF pulse (a hard pulse), the echo times, readout time (~ 4 ms), rewinders, and spoiler gradients per interleaves.

To obtain time-series IDEAL images, the imaging pulse sequence in Fig. 1a was repeated with additional delay times as shown in Fig. 1b. This allowed the acquisition of time-series IDEAL images with various gas-transfer time points, ∆. To investigate the optimum selection of ∆, time-series IDEAL images were acquired from a healthy subject including late ∆ points ∆_late: ∆_initial, 20, 25, 50, 100, 500, 1000, and 2000 ms leading to a long scan time of ~ 17 s. ∆_initial was the spoiling acquisition used to destroy initial buildup of dissolved phase signal which accumulated during the time between the inhalation of HP ¹²⁹Xe gas and the scan starting. To investigate the feasibility of obtaining XGTD, the time-series IDEAL images of four healthy subjects were collected with the following gas-transfer time points: ∆_short: ∆_initial, 300, 200, 100, 50, 30, 25, and 20 ms leading to a shorter total scan time of ~ 8 s.

IDEAL images were reconstructed onto a 128 × 128 matrix using a similar algorithm to those described in Wiesinger et al [33] and Schulte et al [29]. Image intensity variations in the PTBP and RBC compartments were analyzed selecting three ROIs, the left and right lungs with a matrix size of 80 × 40 and the heart with a matrix size of 30 × 30 (as shown in Supplementary Fig. 1). ROI-left with matrix size 80 × 40 included signal from the left lung and heart. Similarly, ROI-heart with matrix size 30 × 30 included signal from the left lung and heart. The dissolved phase ¹²⁹Xe heart signal in ROI-left and the dissolved phase ¹²⁹Xe lung signal in ROI-heart were separated by solving two variable equations as explained in appendix.

The normalized dissolved phase signals from ROI-left, ROI-right, and ROI-heart in the time-series PTBP and RBC images were fitted to the simple exponential function as described by Mansson et al [13, 21] as explained in Supplementary Fig. 1. Collectively, the normalized data points and the exponential fit were described as XGTD curves. The gas-transfer time constant, τ₁, which corresponded to the rapid exponential increase, S₁, the linear increase, and S_o, the y- intersect of XGTD curve were measured in all subjects as reported in Table 1. Additionally, the gas compartment time-series images were analyzed by fitting the signal loss equation to the gas signal at right and left ROIs (as explained in Supplementary Fig. 1) as demonstrated in the appendix.

The Welch’s t test in GraphPad Prism (GraphPad Prism version 7.00, GraphPad Software) was used to investigate any potential difference in the fitting parameters (i.e,. S₁, S_o, and τ₁) between ROIs within the subjects with a 95% confidence interval. For proof of concept in clinical practice, fours patients currently involved in another HPX study with CT proven pulmonary emphysema and abnormal lung function were then scanned using the same technique. The CT diagnosis of emphysema was made using CT density maps, with a threshold of – 940 Hounsfield units. After optimization of gas-transfer time points from the images were obtained from a healthy subject as explained previously, XGTD curves were compared between the normal (n = 5) and pulmonary emphysema (n = 4) subjects as shown in Table 1.

The averages of all time points over the ∆_late in time-series IDEAL gas compartment images including ROI-Left and ROI-Right are shown in Fig. 2a. The corresponding gas signal from each time point in the time-series IDEAL gas compartment images and the corresponding fit from Eq. (3) for the ROI-left and ROI-right are shown in Fig. 2d. As expected, the S_gas decreases continuously during the acquisition of time-series IDEAL images as a function of T₁ = 35 ± 10 s and number of RF pulses (36 per IDEAL image) with α_Gas = 3.6 ± 0.4°. During the five time points between ∆_initial and ∆ = 0.1 s which correspond to the first 8 s of the entire scan time (17 s), the reservoir gas signal drops monotonically. However, for the later three time points between ∆ = 0.5 s and ∆ = 2 s, corresponding to the scan time interval between 8 s and 17 s, the reservoir gas signal drops more rapidly due to T₁. As the scan time increases, the effect of T₁ becomes more noticeable in the reservoir gas signal as also shown in Supplementary Fig. 2.

The average of PTBP and RBC compartment images of all time points over the ∆_late in time-series IDEAL is shown in Fig. 2b, c. The time-series PTBP and RBC compartment imaging of the lungs are also shown for each time points in Supplementary Fig. 2. A strong signal is visible from ROI-heart (Fig. 2b, c) resulting in a different image texture compared to the gas compartment image (Fig. 2a). The temporal PTBP and RBC signal from each time point over ∆_late from ROI-left, ROI-right, and ROI-heart is shown in Fig. 2e, f, respectively. The S_PTBP(ROI-left) and S_PTBP(ROI-right) in Fig. 2e and S_RBC(ROI-left) and S_RBC(ROI-right) in Fig. 2f follow a different trend than S_gas in Fig. 2d corresponding to the replenishment of dissolved phase signal as a function of ∆. In particular, the S_PTBP(ROI-heart) and S_RBC(ROI-heart) increase significantly at time points of ∆ = 0.5 s:2 s but remain low at the early time point of ∆ = 20 ms:100 ms in Fig. 2e. f. For example, S_PTBP(ROI-heart) = 0.2 × 10⁴ is only 5% of the S_PTBP(ROI-right) = 4.4 × 10⁴ at ∆ = 25 ms in Fig. 2e. However, Fig. 2e demonstrates S_PTBP(ROI-heart) = 1 × 10⁴ is 42% of the S_PTBP(ROI-right) = 2.4 × 10⁴ at ∆ = 2 s showing that the arrival of xenon from the alveoli to the heart is longer than the arrival of xenon to the pulmonary tissue and corresponding pulmonary capillary network. This difference in the arrival time of xenon in ROI-heart and ROI-right Fig. 2e, f also confirms that the separation of heart signal was effective (as explained in Appendix: Separation of Heart Signal). After correction of the heart signal, ROI-left and ROI-right were identical in the healthy subject as expected. The delayed arrival of xenon to the heart (ROI-heart) is also shown in Supplementary Fig. 2 in axial and coronal views confirming the delayed arrival of dissolved phase HPX to the heart for the late gas-transfer time points (> 200 ms).

The time-series IDEAL images from another healthy subject are shown in Fig. 3a in which the effect of T₁ signal loss was minimized by capturing more short gas-transfer time points (∆_short: ∆_initial, 300, 200, 100, 50, 30, 25, and 20 ms) resulting in a shorter scan time of ~ 8 s. The signal is strongest in the center of the PTBP and RBC compartment images at the initial acquisition, ∆_initial, in Fig. 3a as shown with arrows. Three coronal slices of ¹H MRI of the thorax from the same subjects are also shown in Fig. 3b–d, including the 3D render of six coronal slices in Fig. 3e. The left ventricle of the heart and the corresponding arterial vasculature vessels are shown with arrows in Fig. 3b and e, respectively. In comparison to the ¹H MR images, the strong dissolved phase signal in the center (Fig. 3a) at ∆_initial is likely to have originated from the dissolved phase ¹²⁹Xe signal in the left ventricle, where the dissolved ¹²⁹Xe accumulated during the time between inhalation of the ¹²⁹Xe gas and the scan starting.

The mean of two XGTD curves from the same healthy subject including the standard deviation are shown in Fig. 4a, b demonstrating the separation of PTBP and RBC compartments. The average standard deviation of the repeated measurements was less than 5.2% suggesting that the repeated measurements are reasonably consistent. There was no statistical difference between the ROI-left and ROI-right (p value > 0.175 and p value > 0.128) of five healthy participants. The τ₁ from ROI-heart (τ₁ = 31-37 ms) demonstrated a statistical difference compared to ROI-left (τ₁ = 9.3–13 ms) and ROI-right (τ₁ = 9.4–13 ms) (p value < 0.001) and approximately three times greater than the lung signal due to the delayed arrival of dissolved phase xenon in the heart. XGTD curves from a patient with pulmonary emphysema are also shown in Fig. 4c, d. The amplitude of XGTD curves from the emphysema subjects was substantially smaller than the healthy subjects. The actual time-series IDEAL images of the healthy and the patient with emphysema are also shown in Supplementary Fig. 3.

The XGTD curve fitting parameters were measured with a very high fitting accuracy (R-squared > 0.98) and summarized in Table 2 for the normal and abnormal lungs. For a statistical comparison, S_o in Table 2 was compared between normal and abnormal lungs as shown in Fig. 5a–b (including the p values). S_o values of the patients with emphysema were smaller and statistically different to the normal lungs. The mean of S_o in the normal lungs was nearly two times greater than the emphysema group in Table 2. Interestingly, the ratio of τ₁ between PTBP and RBC compartments was 0.97 ± 0.001 over the healthy subjects suggesting the gas transfer was repeatable in healthy subjects. However, the pulmonary emphysema patients had a greater standard deviation (1.07 ± 0.07) showing a larger variation due emphysema as shown in Fig. 5c.

Particularly, for the subjects in the Supplementary Fig. 3, the ratio of XGTD curve fitting parameters was compared between four ROIs as shown in Sup Fig. 3 a–b, showing the feasibility of regional analysis. The ventilation and corresponding gas compartment signal were comparable between the healthy and pulmonary diseased subjects from the upper and lower right lungs (ratio of Mi = 0.88 and 1.02 in Table 3). The ratio of the XGTD curve fitting parameters (i.e., S_o and S₁ in Table 3) was a factor of 5.69–2.05 higher in the healthy subject compared to pulmonary emphysema subject.

The multi-interleaved spiral k-space sampling IDEAL technique enabled time-series IDEAL imaging of HP ¹²⁹Xe in the gas, PTBP, and RBC compartments in human lungs using a standard clinical MR scanner. As a proof of concept study, the feasibility of measuring the XGTD curves has been successfully demonstrated in both participant groups with normal and abnormal lungs. The proposed technique in this study showed a statistical difference between the subjects with abnormal and normal lungs. To our knowledge, this is the first in-man study showing the transfer of ¹²⁹Xe gas in the alveoli into PTBP and RBC compartments of the lungs, and into the pulmonary vasculature and the left ventricle of the heart. Additionally, the amount of gas transfer from lung alveolar sacs to the PTBP and RBC compartments and thereby XGTD curves were impaired significantly in emphysema subjects compared to the healthy cohort.

The early XTC imaging technique is conceptually most relevant to the time-series IDEAL imaging in this study. As shown in the time-series IDEAL images the dissolved phase (PTBP and RBC), image texture is different to the gas ventilation distribution of HPX gas. Direct imaging of the dissolved phase compartments allows the capture of the diffusion of dissolved phase ¹²⁹Xe from the lungs to the heart. Therefore, the dissolved phase HPX can be directly and quantitatively measured from the actual PTBP and RBC compartment images by selecting ROIs in the heart and lungs separately. In this respect, the time-series IDEAL imaging technique we have developed may provide insights into the measurement of XGTD similar to those of whole lung MR spectroscopy, with the added advantage of regional measurement of gas-transfer dynamics.

The measured τ₁ reported in this study was comparable to those of measured with XTC imaging technique (~ 11 ms) [34‐36] and MR spectroscopy (26 ± 17 ms) [15, 16]. The τ₁ corresponded to ~ 50% of the dissolved phase signal compared to the dissolved phase xenon signal at ∆ = 300 ms in Fig. 4a suggesting that most of the diffusion from alveoli to the pulmonary tissue and capillaries occurs in the order of milliseconds. Additionally, the time-series IDEAL technique in this study allowed measurement of τ₁ from the heart. Since the dissolved phase xenon at ROI-heart followed a longer pathway, the τ₁ has been found to be approximately three times longer than that of xenon arriving to the ROI-left and ROI-right. Furthermore, the ratio of τ₁ between PTBP and RBC compartments showed a lower variance in the heathy group than in patients with pulmonary emphysema suggesting that the emphysema affects not only the gas ventilation in the lung air sacs but also the transfer of xenon between the PTBP and RBC compartments.

The S_o corresponding to the y-intersect of the XGTC curves and dissolved phase signal amplitude was lower in the patients with lung emphysema than in the subjects with normal lungs. This is in agreement with the findings reported by Qing et al in a cohort of COPD patients [22]. Accordingly, the decrease in S_o may be attributed to changes caused by emphysema in the abnormal lung group. Physiologically, the decreased S_o would be attributed to the decreased gas transfer surface area between alveolar airs sacs to pulmonary tissue and capillary network of the lungs as a result of emphysema. The decreased gas transfer surface area is likely to be due to the loss of alveolar walls and the destruction of capillaries which may not significantly affect the gas compartment signal but can result in significant changes in XGTD curves. Therefore, in addition to the gas ventilation, the pattern of change in XGTD curves (i.e., S_o, S₁, and τ₁) would allow for additional functional information to be derived and would be dependent on the type of lung disease present. This was demonstrated in the increase of dissolved phase signal associated with the increased interstitial thickening in patients with idiopathic pulmonary fibrosis [18, 37]. Furthermore, in the radiation-induced lung injury model, the XGTD curves from the PTBP and RBC compartments and apparent diffusion coefficient from HPX gas were shown to change independently due to the radiation pneumonitis and inflammation [11, 31, 38‐41]. In the future, this technique (i.e., measured S_o, S₁, and τ₁) may potentially be used to investigate the severity of pulmonary disease, including the quantitative measurement of lung physiological parameters (i.e., tissue thickness, capillary diffusion length, perfusion, and mean transit time) when incorporated into a suitable gas exchange numerical model.

One major limitation of our study is that the XGTD curves were not fitted to the numerical gas exchange models, although the technique was developed for quantitative measurement of lung physiology by fitting to the numerical models. The existing gas exchange models [13‐15, 17, 42] cannot be directly applied to the time-series IDEAL images with multi-interleave spiral k-space acquisition for calculating the lung gas exchange parameters without a theoretical justification. As the existing models were suitable for fitting the MR spectroscopy of the whole lungs where the spatial variation in the dissolved phase images was not considered, single-shot spiral IDEAL data as the replenishment of dissolved phase magnetization due to the TR_min is negligible. Incorporating the fitting parameters (S_o, S₁, and τ₁) reported in Table 2 in a suitable numerical gas exchange model, the lung gas exchange parameters (i.e., tissue thickness, capillary diffusion length, perfusion, and mean transit time) and variation between the same subject scans and the mean of healthy subject scan can be investigated quantitatively. Another limitation was the relatively low polarization (10–13%) limiting the voxel by voxel analysis of the XGTD particularly for the diseased subjects with short gas-transfer times (∆ = 20–50 ms) when the replenishment of dissolved phase magnetization is low. The signal detected should be significantly improved using higher polarization levels and better spatial and frequency selective RF pulses in the future [43].

In conclusion, the multi-interleave spiral k-space sampling IDEAL approach enabled imaging HPX in gas, PTBP, and RBC compartments in human lungs using a standard clinical strength gradient MR scanner within an easily tolerated breath-hold of 8 s. Furthermore, the feasibility of measuring the gas-transfer curves has been successfully demonstrated in five subjects with normal lungs and four subjects with pulmonary emphysema. The proposed XGTD technique suggests that time-series IDEAL imaging can capture the transfer of ¹²⁹Xe gas in the alveoli into the PTBP and RBC compartments of the lungs and into the pulmonary vasculature and to the left ventricle of the heart. This initial small participant study has also shown that this technique may potentially differentiate between patients with normal and abnormal lungs.