Comparison of single breath hyperpolarized ¹²⁹Xe MRI with dynamic ¹⁹F MRI in cystic fibrosis lung disease

Abstract

Purpose

To quantitatively compare dynamic ¹⁹F and single breath hyperpolarized ¹²⁹Xe MRI for the detection of ventilation abnormalities in subjects with mild cystic fibrosis (CF) lung disease.

Methods

Ten participants with stable CF and a baseline FEV1 > 70% completed a single imaging session where dynamic ¹⁹F and single breath ¹²⁹Xe lung ventilation images were acquired on a 3T MRI scanner. Ventilation defect percentages (VDP) values between ¹⁹F early‐breath, ¹⁹F maximum‐ventilation, ¹²⁹Xe low‐resolution, and ¹²⁹Xe high‐resolution images were compared. Dynamic ¹⁹F images were used to determine gas wash‐in/out rates in regions of ventilation congruency and mismatch between ¹²⁹Xe and ¹⁹F.

Results

VDP values from high‐resolution ¹²⁹Xe images were greater than from low‐resolution images (P = .001), although these values were significantly correlated (r = 0.68, P = .03). Early‐breath ¹⁹F VDP and max‐vent ¹⁹F VDP also showed significant correlation (r = 0.75, P = .012), with early‐breath ¹⁹F VDP values being significantly greater (P < .001). No correlation in VDP values were detected between either ¹⁹F method or high‐res ¹²⁹Xe images. In addition, the location and volume of ventilation defects were often different when comparing ¹²⁹Xe and ¹⁹F images from the same subject. Areas of ventilation congruence displayed the expected ventilation kinetics, while areas of ventilation mismatch displayed abnormally slow gas wash‐in and wash‐out.

Conclusion

In CF subjects, ventilation abnormalities are identified by both ¹⁹F and HP ¹²⁹Xe imaging. However, these ventilation abnormalities are not entirely congruent. ¹⁹F and HP ¹²⁹Xe imaging provide complementary information that enable differentiation of normally ventilated, slowly ventilated, and non‐ventilated regions in the lungs.

1. INTRODUCTION

Cystic fibrosis (CF) is the most common fatal monogenic disorder in Caucasians. The pathophysiology of CF includes the production of abnormally viscous mucus that obstructs airways and results in persistent infections and inflammation. ¹ High‐resolution computed tomography (HRCT) is currently the gold standard for assessing structural lung disease in CF; however, the risks associated with repeated radiation exposure limit the use of longitudinal HRCT, particularly in the pediatric population. ² MRI is an emerging modality for analyzing lung structure and function as it is non‐invasive with no ionizing radiation exposure and therefore, well‐suited for longitudinal studies. Historically, structural lung MRI was limited due to short proton‐signal transverse coherence time and low tissue density, but advances in ultrashort echo time (UTE) and zero echo time (ZTE) pulse sequences have allowed for improved structural imaging of lung parenchyma, blood vessels, and mucus plugging. ³ In addition to structural information, MRI has the potential to offer functional information in the form of ventilation imaging. ⁴

Hyperpolarized (HP) gas imaging of the lungs has been performed for the better part of the past two decades with hyperpolarized helium‐3 (³He). More recently, hyperpolarized xenon‐129 (¹²⁹Xe) has emerged as a more cost‐effective and accessible alternative. ⁴ MRI images obtained with hyperpolarized ¹²⁹Xe gas provide regional information about ventilation that standard pulmonary function tests alone cannot provide, ⁵ including a range of functional biomarkers, such as the ventilation defect percentage (VDP), ⁶ apparent diffusion coefficient (ADC), ⁷ and fractional ventilation. ⁸ VDP from ¹²⁹Xe MR imaging correlates with lung clearance index (LCI) in pediatric subjects with CF, ⁶ and can possibly provide more specific and accurate information on CF treatment response in interventional trials. ⁹ In addition, due to the sizable tissue solubility and large chemical shift range (~200 ppm) of ¹²⁹Xe, dissolved‐phase HP ¹²⁹Xe imaging is also feasible, ¹⁰ , ¹¹ , ¹² and could potentially provide valuable insights into the pulmonary gas‐exchange processes in both healthy and CF lungs. Limiting factors for HP ¹²⁹Xe use are the need for technical expertise and special equipment to hyperpolarize the gas, as well as the known anesthetic effect of xenon. ¹³ However, recent studies have clearly shown that xenon inhalation for lung ventilation imaging is safe and well tolerated in adults as well as in children as young as 6 years old. ¹⁴ , ¹⁵

Fluorine‐19 (¹⁹F) imaging has been proposed as an alternative to HP gas for ventilation‐space imaging that does not require on‐site gas polarization. Feasibility studies have produced promising ventilation images using perfluoropropane (C₃F₈ or PFP) combined with 21% oxygen (O₂). ¹⁶ PFP has a high gyromagnetic ratio (~3.4 times ¹²⁹Xe), high natural abundance (100%), higher spin‐density than ¹²⁹Xe, and very short T₁ (12.4 ms compared with ~20 s for ¹²⁹Xe). ¹⁶ , ¹⁷ The short T₁ allows for the use of very short repetition times (TRs) and signal averaging that increases the ¹⁹F image signal‐to‐noise ratio (SNR) without the need for pre‐polarization. Without the need for a complex hyperpolarization protocol, ¹⁹F gas imaging could potentially provide an alternative means of characterizing regional lung ventilation. ¹⁸ Additionally, the absence of signal decay in the lung allows for multiple breath studies without the need of constant fresh supply of hyperpolarized gas, and the combination of low T1 and high spin density of ¹⁹F enables the quantification of gas wash‐in and wash‐out kinetics, thus, providing the ability to detect and localize ventilation abnormalities. ¹⁹ Ventilation defects have been observed in diseased populations using ¹⁹F imaging ²⁰ and promising comparisons between ¹⁹F and ¹²⁹Xe ventilation images have been recently performed in healthy volunteers. ²¹ As single breath‐hold ¹⁹F imaging suffers from reduced SNR, preventing a true single breath‐hold comparison between the two gases, here we compare dynamic ¹⁹F to single breath HP ¹²⁹Xe MRI for the detection of lung ventilation defects in subjects with mild CF to better understand the similarities and differences between these two imaging approaches.

Due to the difference in resolutions between ¹²⁹Xe and ¹⁹F images, we also compared low and high‐resolution ¹²⁹Xe images to isolate the potential effects of image resolution on the measurement of ventilation defects.

2. METHODS

2.1. Study participants and study design

This study was approved by the research ethics board at UNC‐Chapel Hill (IRB 17‐2569) and all participants provided written informed consent. Ten participants with stable CF lung disease, ≥ 18 y of age, non‐smokers (<10 pack‐year history and no active smoking in the past year), and a baseline FEV1 > 70% predicted were enrolled from May 2018 to March 2019. Subject demographics are reported in Table 1. Subjects ranged in age from 21 to 44 years, and had a mean FEV₁ of 81.7 ± 15.0% predicted. Each subject completed a single imaging session during which both ¹⁹F and ¹²⁹Xe ventilation images were acquired in a random order. Prior to the imaging session, all subjects completed spirometry according to American Thoracic Society standards. ²²

TABLE 1.

Study population demographics

Gender	Age	FEV1 (%)	Genotype
F	20	73	F508del/F508del
F	27	72	W1282X/S341P
F	23	82	F508del/F508del
M	26	86	F508del/G551D
F	24	92	F508del/F508del
M	24	71	F508del/621 + 1G‐‐>T
F	44	64*	R75X/R1066H
M	24	85	F508del/F508del
M	35	75	F508del/F508del
F	30	117	F508del/F508del

^*Visit FEV₁ below baseline due to subject not performing their normal airway clearance procedure on the day of the study visit.

2.2. Imaging session

Imaging was performed on a Siemens PRISMA 3T MR scanner (Siemens AG) with multinuclear capabilities. Subjects were randomized to the order of ¹⁹F and ¹²⁹Xe imaging; after the first scan, subjects exited the scanner and were allowed a 15‐minute break, during which a single spirometric maneuver was completed to ensure no change in lung function as a result of either gas inhalation.

3D ¹H scans were completed before each ¹²⁹Xe and ¹⁹F scan to facilitate co‐registration of ¹²⁹Xe images with ¹⁹F images. ¹H images acquired before ¹²⁹Xe scan were acquired during a breath hold of 1 L of medical air to improve matching of the lung volume at which ¹H and ¹²⁹Xe images were obtained and thus, image registration. All ¹H images were acquired by using a 3D stack of spiral VIBE sequence with 5° excitation flip angle, a TR of 2.42 ms, echo time (TE) of 0.05 ms, 224 by 224 acquisition matrix, and a resolution of 2.1 mm × 2.1 mm × 2.5 mm.

For ¹⁹F acquisitions, a PFP‐filled dual‐cylinder lung phantom was scanned prior to each participant for quality assurance and to establish the ¹⁹F center frequency. Subjects were positioned supine in the scanner with a ¹⁹F‐tuned 8‐channel chest coil (ScanMed LLC, NE) around the chest. Subjects inhaled a pre‐mixed, medical grade gas mixture of 79% PFP:21% oxygen (IND 122,215) using a continuous‐breathing custom gas delivery device. ²⁰ PFP was administered with a full‐face non‐invasive ventilation mask and a non‐rebreathing Douglas Bag system. Subjects inhaled and exhaled one tidal volume breath of the contrast gas followed by a maximal inhalation with a 12 second breath‐hold, during which time images were obtained prior to exhalation. A total of five such imaging cycles during PFP inhalation (wash‐in) were performed, followed by up to eight cycles of room air inhalation (wash‐out) until no visible signal was present (for a total for 10 wash‐in breaths and up to 16 wash‐out breaths). Ventilation was coached with a pneumotachometer as a visual aid for the MRI technician, and safety was monitored via blood oxygenation saturation, exhaled CO₂ concentration, and heart rate. ¹⁹F dynamic images were acquired using a coronal 2D gradient echo (GRE) sequence with a 74° flip angle, an TE of 1.61 ms, a TR of 13 ms, 15 mm slice thickness, and a 64 by 64 acquisition matrix with a 130 Hz/pixel bandwidth and an in‐plane resolution of 6.25 mm × 6.25 mm.

¹²⁹Xe imaging was performed using a flexible ¹²⁹Xe‐tuned quadrature chest coil (Clinical MR Solutions, WI). For each subject, two images were acquired, each during a single 12 second breath‐hold of 750 ml isotopically enriched ¹²⁹Xe, polarized up to ~14% with a Polarean 9800 ¹²⁹Xe Polarizer (Polarean, Inc, Durham, NC), mixed with 250 ml of N₂. During the first inhalation, the ¹²⁹Xe center frequency was determined right before the acquisition of a low‐resolution 2D multislice image data set. A high‐resolution 2D multislice image data set was acquired during the second inhalation, approximately 30 minutes after the first inhalation using a second dose of hyperpolarized gas. Throughout the imaging session the subject's heart rate, blood pressure and oxygen saturation level were monitored every 5 min. After each xenon inhalation, subjects were also queried about symptoms associated with neurologic changes (ie, euphoria, numbness, tingling, etc). ¹²⁹Xe images were acquired using 2D GRE multislice sequences with a 10° excitation flip angle, a TR of 9 ms, an TE of 4 ms, a field of view of 280 × 350 (read × phase), and a matrix size of either 128 × 64 with a slice thickness of 21.0 mm, which resulted in an in‐plane resolution of 2.2 mm × 6.4 mm (low resolution) with a total scan time of 7 s, or a matrix size of 128 × 80 with a slice thickness of 10.5 mm, which resulted in an in‐plane resolution of 2.2 mm × 4.4 mm (high resolution) and a total scan time of 14 s.

2.3. Image analysis

Registration and masking of images were performed using MIM Software (Cleveland, OH). The ¹H lung masks corresponding to the ¹²⁹Xe acquisitions were generated via a semi‐automated segmentation of the ¹H lung‐cavity images, using a simple region‐growing algorithm. The ¹H lung masks corresponding to the ¹⁹F acquisitions were generated through visual inspection using the threshold and manual region of interest (ROI) segmentation tools due to difference in lung inflation between the ¹H and ¹⁹F in some subjects. The ¹H lung cavity images were used as lung masks to eliminate noise regions outside the lung ROI. Two sets of ¹⁹F ventilation type images were created for each subject; an early‐breath image and a maximum ventilation (max‐vent) breath image. For the early‐breath image, the first image to show ¹⁹F signal was used ²¹ ; for the max‐vent image, the last wash‐in image was used. ¹²⁹Xe low‐resolution (low‐res) and high‐resolution (high‐res) were obtained as noted above. ¹²⁹Xe and ¹⁹F images were then registered through a rigid registration transform on their respective ¹H images. The registered images and masks were then exported to MATLAB (version 2017b; MathWorks, Natick, MA) for further processing using in‐house scripts.

The algorithmic VDP segmentations commonly used in ¹²⁹Xe studies, such as k‐means clustering, are SNR dependent, and therefore, it is difficult to use these segmentations to make comparisons across modalities with different SNRs. ²³ In this study, the SNR was calculated as the ratio of the 90th percentile of the lung interior distribution to the noise signal SD (taken from an ROI outside the thorax). As expected, ¹⁹F images had markedly lower SNR than ¹²⁹Xe images, and this difference between SNR values drove the decision to use a threshold‐based VDP analysis for all image types. ¹⁷ , ¹⁸ The intensity threshold was subsequently defined as the 95th percentile of the background noise distribution, ¹⁸ measured from an ROI drawn outside of the thorax. VDP was then defined as the percentage of the total lung volume that failed to exceed this intensity threshold. ¹⁸

Analysis next focused on regions where localization of ventilation defects by ¹²⁹Xe and ¹⁹F yielded disparate results. For this analysis, max‐vent ¹⁹F image was fused with the ¹²⁹Xe image using MIM software using a rigid body translation with no size dilations. The use of a rigid transformation versus a non‐affine transformation was dictated by the absence of a theoretical model of lung inflation that is needed to accurately perform a non‐affine transformation on the images. It is also important to note that, in the case of CF, the elastic properties of the lung tissue may be significantly altered by the presence of chronic inflammation. Such theoretical models are, therefore, expected to be subject specific, thus, introducing new confounding variables in the analysis.

After the rigid transformation, ROIs were manually drawn to identify regions where ventilation defects identified by the two techniques were mismatched. These regions included locations where ¹²⁹Xe was present but ¹⁹F was low/missing (¹²⁹Xe + ¹⁹F‐) and locations where ¹²⁹Xe was low/missing but ¹⁹F was present (¹²⁹Xe‐ ¹⁹F+). A set of ROIs for each subject were also examined in regions where both ¹²⁹Xe and ¹⁹F were present (¹²⁹Xe + ¹⁹F+) for comparison.

Since the ¹⁹F dynamic imaging allows calculation of regional wash‐in and wash‐out kinetics, we compared PFP gas wash‐in and wash‐out rate constants in both matched areas (filled with both ¹⁹F and ¹²⁹Xe) and unmatched areas (¹⁹F only or ¹²⁹Xe only). The mean ¹⁹F signal in each ROI at each image acquisition time point was evaluated to generate a signal versus time plot for each ROI. The average of ROI voxels from a pre‐PFP exposure scan was used to estimate the baseline noise level. The mean ¹⁹F signal values were then curve‐fitted as a non‐linear bi‐exponential function ²⁴ using the non‐linear least squares method. The time constants that describe gas wash‐in (τ1(s)) and wash‐out (τ2(s)) kinetics, and the maximum ¹⁹F signal (peak), were derived from this fit.

2.4. Statistical analysis

Statistical analyses were performed using Matlab and SAS v9.4. Linear regressions and Pearson correlations were performed between VDP values measured by low and high‐resolution ¹²⁹Xe and ¹⁹F imaging. VDP mean differences are reported as [absolute] % difference and displayed in Bland‐Altman plots. The goodness of fit between mean raw ¹⁹F values in ROIs and fitted curves were assessed with R‐square values. Repeated measures analysis of variance (ANOVA) was used to compare SNR, mean τ1 and τ2 values in the ¹²⁹Xe + ¹⁹F+, ¹²⁹Xe + ¹⁹F‐ and ¹²⁹Xe‐ ¹⁹F + ROIs, and the peak ¹⁹F signal in each ROI. Comparisons between all groups were corrected with the Tukey‐Kramer adjustment. Results are reported as mean ± SD and displayed in Table 2. P‐values < 0.05 were considered significant.

TABLE 2.

Results of repeated measures ANOVA on time constants τ1, τ2, and peak signal. Comparison of ¹²⁹Xe + ¹⁹F+ (ROI B) and ¹²⁹Xe‐¹⁹F+ (ROI C) showed significant difference for τ2 (P = .0075)

	ROI type (mean ± SD)			Corrected P values (Tukey Kramer) P‐values
	Xe + F19‐ (ROI A)	Xe + F19+ (ROI B)	Xe‐ F19+ (ROI C)	Comparison A‐B	Comparison B‐C	Comparison A‐C
Tau1 (s)	70.46 ± 38.70	36.83 ± 11.56	85.36 ± 56.83	0.3397	0.0589	0.7601
Tau2 (s)	31.16 ± 33.35	18.13 ± 5.52	34.31 ± 11.63	0.4947	0.0075	0.9294
Peak Signal Intensity	9.87 ± 14.55	53.59 ± 17.15	37.01 ± 8.36	<0.0001	0.0192	0.0002

All comparisons of peak values were statistically significant, indicating that the ¹⁹F signal at the end of the dynamic imaging cycle was able to discriminate between the three different types of ROIs.

3. RESULTS

3.1. Signal to noise

Signal to noise values across the entire lung volume were determined in the early‐breath ¹⁹F, max‐vent ¹⁹F, low‐resolution ¹²⁹Xe, and high‐resolution ¹²⁹Xe images as shown in Figure 1 (early‐breath ¹⁹F: 18.36 ± 3.16; max‐vent ¹⁹F: 25.02 ± 4.02; low‐res ¹²⁹Xe: 73.59 ± 53.82, high‐res ¹²⁹Xe: 30.17 ± 20.82). Significant SNR differences were present between early‐breath ¹⁹F images vs. low‐res ¹²⁹Xe images (P = .041) and between early‐breath ¹⁹F vs. max‐vent ¹⁹F (P < .001). A nearly significant difference between low‐res ¹²⁹Xe vs. High‐res ¹²⁹Xe was observed (P = .052).

FIGURE 1.

Boxplot showing distribution of ventilation images SNRs across all ten subjects, with the median (red line), 25th and 75th percentiles (blue box edges), and range (whiskers) displayed. Statistically significant differences were seen in early‐breath ¹⁹F vs. low‐res ¹²⁹Xe (P = .041) and early‐breath ¹⁹F vs. max‐vent ¹⁹F (P < .001)

3.2. Comparison of ventilation defect values across methods

Figure 2 provides examples of the variability of gas filling between methods. In Subject 1, the ventilation defect location and volume from the early‐breath ¹⁹F was similar to that of the high‐resolution ¹²⁹Xe. In contrast, the max‐vent ¹⁹F showed eventual filling of a majority of these ventilation defects. In Subject 2, while neither ¹²⁹Xe images nor the max‐vent ¹⁹F images demonstrated significant ventilation defects, the early‐breath ¹⁹F images showed substantial defects that likely reflect inadequate SNR at this early time point. In Subject 3, the early‐breath ¹⁹F image again shows large areas of ventilation defect in the right lung, thought to reflect inadequate gas uptake or signal at this early point of the PFP wash‐in cycle. The high‐resolution ¹²⁹Xe image detected more ventilation defects than the low‐resolution ¹²⁹Xe image, especially in the left upper lobe, for example, where volume averaging may be increasing the local mean value above the threshold. Comparison of the max‐vent ¹⁹F and xenon images shows both matched ¹²⁹Xe‐ ¹⁹F‐ and mismatched ¹²⁹Xe‐ ¹⁹F + ventilation defects, suggesting that some lung regions never filled with PFP, whereas others demonstrated delayed filling.

FIGURE 2.

Representative images showing early‐breath ¹⁹F (A), max‐vent ¹⁹F (B), low‐resolution ¹²⁹Xe (C), and high‐resolution ¹²⁹Xe (D) images in three subjects. A threshold was applied to define regions of ventilation defects (red masks). VDPs were calculated as the percentage of lung with a ventilation defect compared with total lung volume calculated by the anatomic mask. VDPs in early‐breath ¹⁹F images were higher than in max‐vent images, likely due to lack of sufficient signal. High‐resolution ¹²⁹Xe images typically displayed higher VDPs than low‐res images

Figure 3 shows correlations and Bland‐Altman plots of VDPs calculated from each image type, referenced against the standard high‐res ¹²⁹Xe VDP method. No correlation between early‐breath ¹⁹F VDP (r = 0.28, P = .43; Figure 3A) or max‐vent ¹⁹F VDP (r = 0.23, P = .52; Figure 3B) and high‐res ¹²⁹Xe VDP values were observed. The Bland‐Altman plot shows a significant mean difference in VDP between max‐vent ¹⁹F and high‐res ¹²⁹Xe (−10.6%, P = .007; Figure 3E). There was no significant difference in mean VDP values between early‐breath ¹⁹F and high‐res ¹²⁹Xe (−0.5%, P = .87; Figure 3D). In contrast, low‐res vs. high‐res ¹²⁹Xe VDP display a significant correlation (r = 0.68, P = .03; Figure 3C), although a significant mean difference in VDP (−10.5%, P = .001; Figure 3F) indicated consistent underestimation of VDP values by low‐res ¹²⁹Xe images.

FIGURE 3.

Scatter (A, B, C) and Bland‐Altman (D, E, F) plots comparing VDP measurements. Regression line (dark gray) shown on scatter plots. Mean bias ± 95% lines of agreement shown on Bland‐Altman plots. High‐resolution ¹²⁹Xe is compared with: early‐breath ¹⁹F images (r = 0.28, P = .43. Estimated bias = −0.5 ± 19.6%, P = .87) (A,D); (B, E). max‐vent ¹⁹F images (r = 0.23, P = .52. Estimated bias = −10.6 ± 18.9%, P = .007) (B,E); low‐resolution ¹²⁹Xe (r = 0.68, P = .03 (C,F). Estimated bias = −10.5% ± 13.8%, P = .001)

Figure 4 shows the comparison between max‐vent ¹⁹F VDP and early‐breath ¹⁹F VDP. There was significant correlation (r = 0.75, P = .012) and a significant mean difference (10.0%, P = 4.9e‐5) indicating overestimation of VDP by early breath ¹⁹F.

FIGURE 4.

Scatter plot (A) and Bland‐Altman plot (B) showing the comparison of VDP measurements from max‐vent ¹⁹F VDP and early‐breath ¹⁹F VDP. The correlation was r = 0.75, P = .01. The Bland‐Altman plot shows estimated bias = 10.05 ± 8.6% with P = 4.9e‐5

3.3. Evaluating mismatched ¹⁹F and ¹²⁹Xe ventilation defects

Comparison of low‐resolution and high‐resolution ¹²⁹Xe images did not reveal any region of non‐congruency that could not be attributed to differences in resolution: high‐resolution ventilation images consistently displayed sharper ventilation defects than the low‐resolution images. In contrast, within the same subject, we observed differences in the location and volume of ventilation defects when comparing ¹²⁹Xe and ¹⁹F images (Figure 5). We sought to utilize the dynamic ventilation data afforded by ¹⁹F imaging to further understand the nature of these mismatched regions. In Figure 6, representative mean ¹⁹F signal vs. time curves in matched and unmatched ROIs are shown. Mean ¹⁹F wash‐in (τ1) and wash‐out (τ2) rate constants derived from these plots, along with the goodness of fit (r‐square), from each ROI type (¹²⁹Xe + ¹⁹F+, ¹²⁹Xe + ¹⁹F‐, and ¹²⁹Xe‐¹⁹F+) are shown with box plots in Figure 7. Wash‐in and wash‐out rate constants (τ1 and τ2, respectively) were higher in both types of mismatched ROIs (¹²⁹Xe‐ ¹⁹F + and ¹²⁹Xe + ¹⁹F‐) when compared with matched ROI (¹²⁹Xe + ¹⁹F+) regions, but only ¹²⁹Xe + ¹⁹F+ vs. ¹²⁹Xe‐¹⁹F + τ2 showed significant difference (P = .008). The mean maximum ¹⁹F signal value derived from the modeled signal versus time plot was found to be statistically different between all three different regions (¹²⁹Xe‐ ¹⁹F+, ¹²⁹Xe + ¹⁹F‐, and ¹²⁹Xe + ¹⁹F+), as shown in Table 2. This demonstrates that lung regions with ¹⁹F, but no ¹²⁹Xe, do indeed fill, but at a slower rate and to a lower peak signal (compared with defect‐free ROIs). More surprising is the paradoxical finding of lung regions with ¹²⁹Xe signal after a single breath, but very low ¹⁹F signal after multiple breaths. Examination of ¹⁹F wash‐in/out time constants and peak signals in these regions suggests that they also have abnormal ventilation kinetics, yet are not detected as abnormal by the ¹²⁹Xe method. Because poorly estimated τ1 and τ2 values could lead to misinterpretation of our data, we calculated the goodness of fit for ¹⁹F signal in each type of ROI to assess the quality of the data. Importantly, the r‐square values between raw data and calculated exponential curves were very high in each type of ROI, including those with low ¹⁹F signal, providing confidence in these observations (R² = 0.89 ± 0.18 for ¹²⁹Xe + ¹⁹F‐, 0.99 ± 0.003 for ¹²⁹Xe + ¹⁹F+, and 0.99 ± 0.01 for ¹²⁹Xe‐ ¹⁹F+).

FIGURE 5.

Examples of mismatched ventilation signal from two subjects. A. max‐vent ¹⁹F and B. high‐resolution ¹²³Xe images. The purple ROI outlines ¹²⁹Xe + ¹⁹F‐ regions; the yellow ROI outlines ¹²⁹Xe‐¹⁹F + regions; the cyan ROI outlines a matched ¹²⁹Xe + ¹⁹F+ region used for comparison. The green ROI outlines a matched ¹²⁹Xe‐¹⁹F‐ for additional comparison

FIGURE 6.

Plots of the raw (symbols) and modeled (lines) ¹⁹F signal time course in matched (¹²⁹Xe + ¹⁹F+; blue) and unmatched (¹²⁹Xe‐¹⁹F + and ¹²⁹Xe + ¹⁹F‐; red and green, respectively) ROIs from a single representative subject. In all subjects with mismatched ROIs, a consistent rank order of maximum gas ¹⁹F signal of ¹²⁹Xe + ¹⁹F+> ¹²⁹Xe‐¹⁹F+> ¹²⁹Xe + ¹⁹F‐ was observed despite accentuated ¹²⁹Xe signal (ie, higher than mean lung signal) in some of the ¹²⁹Xe + ¹⁹F‐ ROIs.

FIGURE 7.

Box plots of wash‐in (τ1) (A) and wash‐out (τ2) (B) time constants from the ¹⁹F bi‐exponential fit. c. R‐square indicates the goodness of the fit between raw data and modeled curves. The horizontal brackets indicate statistical significance with corresponding P‐values

4. DISCUSSION

This study reports a first comparison of regional lung ventilation as assessed by dynamic ¹⁹F and HP ¹²⁹Xe MRI imaging in CF subjects. The sequential performance of ¹⁹F and ¹²⁹Xe MRI scans within the same day in subjects with mild CF lung disease provided a powerful platform to make comparisons between these modalities.

As expected, ¹⁹F images were characterized by a markedly lower SNR, which prevented direct comparison of single breath ¹²⁹Xe ventilation images to single breath ¹⁹F images. Therefore, we compared early‐breath ¹⁹F images (ie, after first appearance of signal during the multiple breath wash‐in procedure) to single breath ¹²⁹Xe images. However, when assessed after multiple PFP breaths, SNR increased as expected, and a portion of ventilation defects disappeared, indicating that many ventilation defects detected by single breath ¹²⁹Xe MRI ultimately do fill during the multiple breath wash‐in cycle. What is unclear from this data, however, is whether the filling was due to direct versus collateral ventilation.

When measuring VDP values, image resolution also played a role. In our study, direct comparison of VDP values from high‐resolution ¹²⁹Xe and low‐resolution ¹²⁹Xe ventilation images allowed for the isolation of the effect of resolution on VDP. As expected, relative to high‐res ¹²⁹Xe scans, low‐res ¹²⁹Xe scans were more prone to partial volume effects and consistently underestimated VDP. This underestimation is larger at higher VDPs, with the trend showing low‐res scans underestimate VDP by about 50%. It is important to note that the size of the VDP underestimation in the low‐resolution ¹²⁹Xe images is a function of the specific signal intensity threshold used. In particular, by using a less conservative signal threshold, some additional ventilation defects could be captured in the low‐resolution ¹²⁹Xe images, resulting into higher VDP values.

Our data also suggest that VDP may be overestimated by single breath approaches (early‐breath ¹⁹F, low‐res ¹²⁹Xe, high‐res ¹²⁹Xe), which consistently measured higher VDPs than those determined via max‐vent ¹⁹F images. Since slow filling regions appear as regions of ventilation defects in single‐breath hold images, they cannot be differentiated from regions of true ventilation defects. To this end, the multiple breath protocol employed for ¹⁹F imaging provides the ability to differentiate between true ventilation defects and slow filling regions, by providing wash‐in and wash‐out gas kinetics. In particular, in areas of congruence (¹²⁹Xe + ¹⁹F+ or ¹²⁹Xe‐¹⁹F‐), ventilation kinetics were as expected, with either rapid or absent ventilation, respectively. In regions that were ¹²⁹Xe‐¹⁹F+, on the contrary, the multiple breath PFP protocol was able to demonstrate that these regions were not truly unventilated, but, rather, slow to fill. Because HP ¹²⁹Xe images are acquired using a single breath technique, these slowly filling regions could not be differentiated from true non‐filling regions. As such, dynamic imaging using ¹⁹F provides additional capability to characterize these abnormal regions.

The observation of lung regions with a ¹²⁹Xe + ¹⁹F‐ pattern was unexpected, as multiple breaths of PFP is not expected to be less likely to fill a partially obstructed region than a single breath of ¹²⁹Xe. The presence of ¹²⁹Xe gas suggests that these areas must be receiving some ventilation, despite the paucity of ¹⁹F signal. The reduced ¹⁹F signal in these regions may be explained by the low SNR and diffusivity of the gas, when compared with ¹²⁹Xe, coupled with an increase in airway resistance in the region, similarly to what has already been observed in ¹²⁹Xe/³He comparison studies in COPD patients. ²⁵ The analysis of τ1 and τ2 rate constants, which we were able to characterize with a high degree of confidence despite the relatively low ¹⁹F signal, confirmed that in these lung regions, not only was peak ¹⁹F signal low, but also the rates of both gas wash‐in and wash‐out were markedly slowed. The presence of ¹²⁹Xe signal (in some cases, even an accentuated ¹²⁹Xe signal) in these regions could be explained by a reduced concentration of molecular oxygen in these poorly ventilated regions. This could theoretically be encountered in areas of gas trapping, where some contrast gas is able to enter the region, perhaps by collateral ventilation, but encounters an environment that is relatively devoid of oxygen. This would thereby reduce the rate of HP ¹²⁹Xe depolarization and, in turn, cause a paradoxically higher ¹²⁹Xe signal. If this is the case, ¹²⁹Xe MRI may at times under‐report true ventilation defects, as the result of the complex relationship between ¹²⁹Xe polarization and the local lung oxygen tension. While further testing is required to fully assess gas trapping effects on ¹²⁹Xe MRI images, our results show that the addition of dynamic ¹⁹F imaging in registration with HP ¹²⁹Xe has provided important additional insights into lung ventilation dynamics.

The peak ¹⁹F value at the end of the inhalation sequences was also able to differentiate these three different filling patterns, suggesting that the delayed phase filling with PFP as a single parameter may be the most informative of the ventilation status (Table 2). Further work in other disease states could better inform the clinical utility of the dynamic ¹⁹F approaches. Although we did not perform repeatability scans as a part of this study, others have shown that dynamic ¹⁹F ventilation imaging, albeit with a different scanning protocol, has good repeatability in normal subjects and subjects with chronic obstructive pulmonary diseases. ²⁶

Several technical factors impacted our data, including challenges related to co‐registration of ¹⁹F, ¹²⁹Xe, and ¹H images. Proton images of the thoracic cavity and gas ventilation images were obtained during separate maximal inhalation breath‐holds, making it likely that the same inspiratory capacity might not be reached consistently. As a result, differences in lung inflation volumes can be expected to cause co‐registration errors, both in ¹⁹F and ¹H images. This problem is avoided with the ¹²⁹Xe‐MRI method, as a fixed inhalation volume can be achieved simply through the use of a fixed volume gas delivery bag. Differences in lung inflation may have led to additional ventilation defects around the lung edges. Thus, the ability to simultaneously acquire anatomical and ventilation images within a single breath‐hold, as done for HP ventilation imaging, would likely benefit the ¹⁹F method. ²⁷ , ²⁸ Recent work to accelerate ¹⁹F ventilation imaging, which could allow for acquisition of ¹⁹F/¹H images in a single breath hold, would accomplish this goal. ²⁹ , ³⁰ A related minor limitation for the study was the need to reposition the patient between the two inhaled gas studies.

Finally, the use of a multi‐channel ¹⁹F lung coil led to B₁ inhomogeneities that were subject and coil position dependent and could not be corrected. B₁ inhomogeneities made ventilation thresholds position dependent, preventing the application of typical linear binning techniques that delineate regions of high, medium and low intensity areas based on universal thresholds. ³¹ Moreover, retrospective bias‐field estimation techniques, often used to reduce subject‐dependent B₁ effects in ¹²⁹Xe studies, ³¹ created spurious gas‐filled regions from noise level intensity in low SNR ¹⁹F images. This is not surprising as these techniques have been developed for high‐SNR images and, in ¹⁹F images, where SNR is on the order of the bias‐field correction, they artificially change image intensities by a factor of 2 or 3. ³¹ , ³²

5. CONCLUSIONS

In CF subjects, ventilation abnormalities are identified by both ¹⁹F and HP ¹²⁹Xe imaging but are not entirely congruent across all areas of ventilation. The use of both modalities in this study allowed us to identify an “imaging phenotype” that resulted from normally ventilated, slowly ventilated, and non‐ventilated regions. Although further work is needed to evaluate these techniques in other patient populations, these data strongly suggest the complimentary nature of ¹⁹F and HP ¹²⁹Xe imaging; however, VDPs obtained from each method should not be considered equivalent. These data also highlight the utility of ventilation kinetic analyses with ¹⁹F MRI and the inherent limitations of relying on a single breath VDP assessment for the characterization of airway function.

McCallister A, Chung SH, Antonacci M, et al. Comparison of single breath hyperpolarized ¹²⁹Xe MRI with dynamic ¹⁹F MRI in cystic fibrosis lung disease. Magn Reson Med. 2020;85:1028–1038. 10.1002/mrm.28457

DATA AVAILABILITY STATEMENT

All data and processing algorithms have been uploaded to the UNC Libraries Carolina Digital Repository and is available online at https://cdr.lib.unc.edu/concern/data_sets/m900p0761?locale=en.