129Xe and Free‐Breathing 1H Ventilation MRI in Patients With Cystic Fibrosis: A Dual‐Center Study

Background Free‐breathing 1H ventilation MRI shows promise but only single‐center validation has yet been performed against methods which directly image lung ventilation in patients with cystic fibrosis (CF). Purpose To investigate the relationship between 129Xe and 1H ventilation images using data acquired at two centers. Study type Sequence comparison. Population Center 1; 24 patients with CF (12 female) aged 9–47 years. Center 2; 7 patients with CF (6 female) aged 13–18 years, and 6 healthy controls (6 female) aged 21–31 years. Data were acquired in different patients at each center. Field Strength/Sequence 1.5 T, 3D steady‐state free precession and 2D spoiled gradient echo. Assessment Subjects were scanned with 129Xe ventilation and 1H free‐breathing MRI and performed pulmonary function tests. Ventilation defect percent (VDP) was calculated using linear binning and images were visually assessed by H.M., L.J.S., and G.J.C. (10, 5, and 8 years' experience). Statistical Tests Correlations and linear regression analyses were performed between 129Xe VDP, 1H VDP, FEV1, and LCI. Bland–Altman analysis of 129Xe VDP and 1H VDP was carried out. Differences in metrics were assessed using one‐way ANOVA or Kruskal–Wallis tests. Results 129Xe VDP and 1H VDP correlated strongly with; each other (r = 0.84), FEV1 z‐score (129Xe VDP r = −0.83, 1H VDP r = −0.80), and LCI (129Xe VDP r = 0.91, 1H VDP r = 0.82). Bland–Altman analysis of 129Xe VDP and 1H VDP from both centers had a bias of 0.07% and limits of agreement of −16.1% and 16.2%. Linear regression relationships of VDP with FEV1 were not significantly different between 129Xe and 1H VDP (P = 0.08), while 129Xe VDP had a stronger relationship with LCI than 1H VDP. Data Conclusion 1H ventilation MRI shows large‐scale agreement with 129Xe ventilation MRI in CF patients with established lung disease but may be less sensitive to subtle ventilation changes in patients with early‐stage lung disease. Evidence Level 2 Technical Efficacy Stage 2

Nevertheless, HP gas MRI is not yet widely available and requires additional equipment that is costly.
Free-breathing 1 H MRI can produce surrogate maps of ventilation without the use of a contrast agent. 8 1 H ventilation metrics correlate strongly with lung clearance index (LCI), show good short-term reproducibility, and can detect ventilation changes following antibiotic treatment in children with cystic fibrosis (CF). [9][10][11] However, only single-center validation has been performed against methods which directly image lung ventilation in patients with CF. [11][12][13] Recent work comparing 129 Xe and free-breathing 1 H ventilation MRI in children with CF showed that 129 Xe and 1 H VDP were both significantly greater in CF patients undergoing pulmonary exacerbation than healthy controls, but no significant difference was found between clinically stable CF patients with normal forced expiratory volume in 1 second (FEV 1 ) and controls. 12 Another study from the same site found that both 129 Xe and 1 H VDP decreased significantly following antibiotic treatment in eight children with CF. 11 In both studies 1 H VDP of the single coronal slice acquired with 1 H free-breathing MRI correlated with 129 Xe VDP, lung clearance index (LCI) and forced expiratory volume in 1 second (FEV 1 ). Eight children with CF were also included in a larger study (including 20 patients with chronic obstructive pulmonary disease (COPD) and 6 healthy volunteers) comparing free-breathing 129 Xe and 1 H ventilation MRI that found a close relationship between the imaging techniques. 13 It is increasingly recognized that multiple-center studies and standardization of imaging techniques and outcome metrics are essential to transition novel imaging techniques such as 129 Xe and 1 H ventilation MRI into tools which can be employed for international clinical research and drug development. 14,15 The aim of this work was to investigate the relationship between 129 Xe and 1 H ventilation images across the whole lungs of patients with a broad spectrum of CF lung disease, using data acquired at two centers.

Methods
Center 1: University of Sheffield, UK The study was approved by the Yorkshire and the Humber -Leeds West research ethics committee (16/YH/03391). All adult patients and parents/guardians of children provided written informed consent.

Center 2: Hannover Medical School, Germany
The study was approved by the institutional review board of Hannover Medical School. All adults and parents/guardians of children provided written informed consent.

Center 1
Inclusion criteria: patients with CF older than 5 years who were clinically stable for 4 weeks prior to scanning.

Center 2
Patients with CF, inclusion criteria: diagnosis of CF and aged 12-60 years. Exclusion criteria: respiratory tract exacerbation within the last month, chronic oxygen therapy, any other severe comorbidities that could limit imaging, MRI contraindication or pregnancy.
Healthy controls, inclusion criteria: aged 18-60 years. Exclusion criteria: lung disease within the last month, known history of chronic lung disease, known history of congenital lung disease, MRI contraindication or pregnancy.

Center 1
A total of 24 patients with CF were scanned using a 1.5 T wholebody MRI system (GE Signa HDx, Milwaukee, WI). Patients underwent spirometry, multiple breath washout and body plethysmography on the same day.
CENTER 2. Subjects were positioned supine in a linearly polarized 129 Xe birdcage transmit coil and 16-channel receive coil (Rapid Biomedical, Rimpar, Germany). A mix (1 L) of hyperpolarized 129 Xe (~20%-30% polarization, 87%-92% 129 Xe, 0.45-0.9 L) and N 2 was inhaled from FRC, with gas  H VDP + LVP (%) 29.9 + 12.5 a 28.9 + 10.0 13.9 + 3.5 a a Significant difference between CF patients at center 1 and controls. b Significant difference between CF patients at center 1 and CF patients at center 2. c Significant difference between CF patients at center 2 and controls. n = number; FEV 1 = forced expiratory volume in 1 second; RV = residual volume; TLC = total lung capacity; LCI = lung clearance index; VDP = ventilation defect percent; LVP = low ventilation percent. Presented as mean AE SD for normally distributed data and median (minimum, maximum) for non-normally distributed data.
volumes determined by patient age and height (details in the OR of the Supplementary material S1). Ventilation images were acquired during breath-hold directly after inhalation of the 129 Xe and N 2 mixture using a 3D coronal SSFP sequence with full lung coverage (voxel size = 4.0 Â 4.0 Â 15 mm, flip angle = 10 , TE/TR = 1.72/3.57 msec, BW = 50 kHz, duration = 5.8 seconds; Table 1). The sequence used a stack of stars trajectory with 90 spokes per k-space partition and symmetric readout. Gradient delay correction was performed assuming an isotropic delay using the method described by Herrmann et al. 17 Relative coil sensitivities were estimated from the central portion of k-space and images reconstructed using the parallel imaging/compressed sensing routine in the Berkeley Advanced Reconstruction Toolbox after Fourier transformation of data along the slice direction. 18 Table 1). 19 Image Analysis Phase-resolved functional lung (PREFUL) analysis was performed on the free-breathing 1 H images from both centers, as detailed by Voskrebenzev et al, and included registration, low-pass filtering and calculation of fractional ventilation. 20 PREFUL analysis of center 1 data was performed at center 1 using code provided by center 2. PREFUL analysis of center 2 data was performed at center 2.
All further image analysis took place at center 1. The 1 H anatomical images of the same imaging volume as the 129 Xe ventilation images were registered to the 129 Xe ventilation images using a supervised approach, that selects preregistered images, within a set computed by an in-house software written in MATLAB (Mathworks, Natick, MA), or manually registered images, using the open-source software itksnap, when the alignment of the automatically registered images is not satisfactory (assessed by A.M.B., 5 years' experience). 21 In order to assist segmentation, for each PREFUL image slice, a 1 H free-breathing image corresponding to inspiration (i.e. with low signal in the lung parenchyma to provide high contrast at the lung boundary) was chosen manually for segmentation by H.M. (10 years' experience). Registration had previously been performed on these 1 H free-breathing images as part of the PREFUL reconstruction and so they were intrinsically registered to the PREFUL ventilation images. 20 VDP (regions with no ventilation) and low ventilation percent (LVP, regions with reduced ventilation) were calculated from the 129 Xe and PREFUL ventilation images in the same manner. The co-registered ventilation and anatomical images were segmented automatically using spatial fuzzy C-means thresholding to produce initial lung cavity masks that were then edited manually to remove the large airways and main vessels, and correct any segmentation errors (by L.J.S., 5 years' experience for the 129 Xe images and H.M., 10 years' experience for the PREFUL images). 22 Large vessels were excluded where they were visible in the anatomical 1 H images. Linear binning of the ventilation images was performed with six bins on N4 bias-field-corrected images scaled by the mean signal inside the lung cavity mask. [23][24][25] The resulting ventilation defect region (first bin, with signal <1/3) was used to calculate VDP and the low ventilation region (second bin, with signal <2/3) was used to calculate LVP. 23 VDP and LVP were added to generate VDP + LVP, a further metric of abnormal ventilation. Ventilation images were assessed qualitatively by H.M., L.J.S., and G.J.C. (10, 5, and 8 years' experience, respectively).

Pulmonary Function Tests
At both centers, spirometry and body plethysmography were performed to international standards using recommended reference equations. 26 Table 2. FEV 1 z-score, 129 Xe VDP, 1 H VDP, 129 Xe VDP + LVP, and 1 H VDP + LVP were significantly different between CF patients scanned at center 1 and controls but not between CF patients scanned at center 2 and controls (P = 0.08, P = 0.37, P = 0.21, P = 0.23, and P = 0.05, respectively). 129 Xe LVP and 1 H LVP were significantly greater for CF patients scanned at center 2 than controls. 129 Xe LVP was significantly lower for patients scanned at center 1 than patients scanned at center 2.
For data acquired at center 1, mean 1 H VDP was lower than mean 129 Xe VDP with a bias of 2.8% and limits of agreement at À13.7% and 19.3% (Fig. 1b). For data acquired at center 2, mean 1 H VDP was higher than mean 129 Xe VDP with a bias of À4.9% and limits of agreement at À14.9% and 5.1% (Fig. 1d). When the data from both centers were pooled the bias was 0.07% and limits of agreement were À16.1% and 16.2% (Fig. 1f). When compared to 129 Xe VDP, 1 H ventilation MRI tended to overestimate VDP for milder disease and underestimate VDP for more severe disease. These trends were not apparent when LVP and VDP were combined (OR Fig. S3). Bland-Altman analysis between 129 Xe VDP + LVP and 1 H VDP + LVP showed similar magnitudes of bias and limits of agreement, with mean 1 H VDP + LVP greater than mean 129 Xe VDP + LVP at center 1, at center 2 and when the data from both centers were combined (OR Fig. S3).
All data acquired at center 2 had 1 H VDP greater than 129 Xe VDP and a trend toward increased bias for larger VDP (Fig. 1c). The bias and limits of agreement were larger for patients with CF (Fig. 2b) than healthy controls (Fig. 2c). All patients scanned at center 2 had 129 Xe VDP < 10%. Patients scanned at center 1 with 129 Xe VDP < 10% also showed a negative bias (Fig. 2d), although the magnitude of the bias was less than at center 2 (Fig. 2e). There were significant correlations between 129 Xe VDP and 1 H VDP for patients with 129 Xe VDP < 10% (center 1: r = 0.73, centers 1 and 2: r = 0.56) ( Table 4).
There were nine patients with absolute difference between 129 Xe VDP and 1 H VDP of more than 5%. These patients had significantly larger 1 H VDP, worse FEV 1 and were shorter than the other patients. 129 Xe VDP (P = 0.13), age (P = 0.08), and RV/TLC (P = 0.24) were not significantly different between the groups (Supplementary OR  Table S4).
The relationships of VDP with FEV 1 and LCI are shown in Fig. 3. The slopes of 129 Xe VDP with FEV 1 (À6.10) and 1 H VDP with FEV 1 (À4.42) were not significantly different (P = 0.08). The slope of 129 Xe VDP with LCI (3.27) was significantly greater than the slope of 1 H VDP with LCI (1.82). The slopes of VDP + LVP with FEV 1 and VDP + LVP with LCI were not significantly different for 129 Xe and 1 H (P = 0.52 and P = 0.14, respectively, Supplementary OR Fig. S4 and OR Table S5). While 1 H LVP had significantly stronger relationships with FEV 1 and LCI than 129 Xe LVP had with FEV 1 and LCI (Supplementary OR Fig. S4 and OR Table S5).

Similarities and Differences Between 129 Xe and 1 H Ventilation Images
Ventilation images of controls were mostly uniform with more ventilation heterogeneity in 1 H ventilation images than in 129 Xe ventilation images (Fig. 4). In most cases, regions of 1 H VDP in controls were adjacent to vessels (Fig. 4c) and the heart, however, some were not (Fig. 4b). In one control with FEV 1 z-score at the low end of normal (À1.58), both 129 Xe and 1 H ventilation images showed abnormalities in the posterior right lung (Fig. 4d). There were five controls where ventilation defects were seen on 1 H ventilation images but not on 129 Xe ventilation images. Figure 5 shows images from CF patients with normal FEV 1 and Fig. 6 shows images from CF patients with abnormal FEV 1 . Regions of reduced 1 H ventilation were often associated with 129 Xe ventilation defects or heterogeneity but did not always capture their full extent or detailed patterns evident on 129 Xe images. In general, 1 H ventilation defects tended to be larger than 129 Xe ventilation defects (Figs. 5b,d  and 6c). There were regional similarities and differences between 129 Xe and 1 H ventilation images, but the distribution of medium-to-large scale ventilation abnormalities throughout the lungs was generally similar. There were five CF patients with normal FEV 1 where small ventilation abnormalities (fully unventilated and partially ventilated) were observed in 129 Xe ventilation images without corresponding regions of low or no ventilation present in 1 H ventilation images (Fig. 5a). There was one patient with abnormal FEV 1 where no ventilation defects were observed on 129 Xe ventilation images and minimal ventilation defects were seen on 1 H images.

Discussion
Analyses of VDP and VDP + LVP show strong relationships between 129 Xe and 1 H ventilation images and with LCI and FEV 1 . Data were acquired at two centers from different cohorts of patients, and image analysis was standardized and performed at a single center. In linear regression analyses, relationships of VDP and VDP + LVP with FEV 1 were not significantly different for 129 Xe and 1 H ventilation imaging, although 129 Xe VDP had a significantly stronger relationship with LCI than 1 H VDP had with LCI. There were small-tomoderate differences in VDP between 129 Xe and 1 H images for most subjects; however, there were some cases with substantially greater differences in VDP. Patients with greater differences between 129 Xe and 1 H VDP had higher 1 H VDP, worse FEV 1 and were shorter than patients with smaller differences in VDP. In patients with milder disease (VDP < 10% and normal FEV 1 ) 1 H ventilation MRI overestimated VDP, while in patients with more severe disease (abnormal FEV 1 ) the relationship was less clear with underestimation of VDP at center 1 and overestimation of VDP at center 2. Qualitatively, although the appearance and exact location of ventilation defects differed between 129 Xe and 1 H ventilation images, larger scale ventilation patterns tended to be similar, that is, the agreement between techniques was greater for larger defects. However, statistical analysis found that 1 H and 129 Xe VDP could not be considered equivalent. The five cases (16% of the CF cohort) where small defects and ventilation heterogeneity present on the 129 Xe ventilation images of CF patients with normal FEV 1 were not detected by 1 H ventilation imaging are suggestive that 129 Xe ventilation imaging is more sensitive to early-stage lung disease than 1 H ventilation imaging. The 1 H ventilation defects present in controls, albeit amounting to 1 H VDPs of less than 2.5% in all controls, which tended to be adjacent to vessels, show that 1 H ventilation imaging may have a higher susceptibility to false positive ventilation defects than 129 Xe ventilation imaging. While medium-sized vessels can cause reduced signal due to partial-volume effects in 129 Xe ventilation images, medium-sized vessels appear to have a greater influence on 1 H ventilation images, which can result in reduced or no signal. Regions of VDP where there was a vessel present on the matching anatomical image were removed for both 129 Xe and 1 H ventilation images, but sometimes ventilation defects adjacent to vessels or in the shape and potential location of vessels remained on 1 H ventilation images contributing to VDP. Other studies have also reported 1 H VDP in healthy volunteers despite removal of the large vessels during image analysis. 13,32 Recently, more extensive vessel segmentation using an automatic deep learning method was found to reduce 1 H VDP and increase its reproducibility. 33 Improved vessel segmentation such as this could be implemented to reduce this source of error for 1 H ventilation in the future.
Differences between the ventilation imaging techniques are likely due in part to the fundamentally different sources of image contrast; inhaled 129 Xe gas density and 1 H signal modulation due to respiratory motion. Other factors include differing voxel size, lung coverage, lung volumes during image acquisition and some error in the matching of image planes for visual comparison. The lung volume during image acquisition was greater for 129 Xe than 1 H ventilation imaging (FRC plus 0.5-1 L for 129 Xe and tidal breathing for 1 H), and this is known to affect the appearance of ventilation defects due to airway closure/opening at lower and higher lung volumes, respectively. 34 Most 1 H ventilation images were acquired with a 5 mm gap between slices to reduce overall acquisition time, in contrast to the full lung coverage of 129 Xe ventilation images, however, recently developed 3D 1 H ventilation imaging techniques could now be applied to provide full lung coverage and increased spatial resolution with 1 H ventilation imaging. 35 Key differences in imaging techniques between the two centers were higher spatial resolution 129 Xe images at center 1, a radial stack of stars k-space trajectory at center 2 compared to a Cartesian k-space trajectory at center 1, and the application of parallel imaging for 1 H ventilation imaging at center 2 while auto-calibrated parallel imaging was not available on the MRI system used at center 1. The scanners at the two centers were also made by different vendors. The thinner slices used for 129 Xe ventilation imaging at center 1 may have contributed to higher mean 129 Xe VDP than mean 1 H VDP due to the reduction in partial volume effects with decreasing slice thickness, while at center 2 where both techniques used the same slice thickness mean 129 Xe VDP was lower than mean 1 H VDP.
The application of linear binning in the image analysis allowed user-independent evaluation of VDP with both techniques treated equally. Partially ventilated defects often present in CF patients with milder lung disease were classified as LVP; however, the relationships of LVP with VDP, LCI and FEV 1 were unclear particularly for 129 Xe LVP. This is likely due to signal loss due to partial volume effects and reduced coil sensitivity also contributing to LVP. Despite this, combining LVP with VDP gave an additional metric of ventilation abnormality, which had strong correlations with LCI and FEV 1 , and showed significant differences between controls and CF patients. 1 H ventilation MRI systematically overestimated VDP + LVP compared to 129 Xe ventilation MRI, but there was no trend toward overestimation of VDP + LVP for milder disease and underestimation for more severe disease as was observed for VDP. VDP + LVP was not as specific as VDP, that is, the values were non-negligible for controls due to the LVP caused by imaging effects rather than disease, but included the partially ventilated regions, which are often seen in patients with mild disease.
This dual-center study using alike, but not identical, imaging techniques and systems at two centers, found comparable relationships between 129 Xe and 1 H ventilation images, showing the potential for multi-site studies which will be necessary for these ventilation imaging techniques to be employed in large-scale clinical research and drug development studies. Standardization of some of the imaging parameters was not possible due to the different imaging platforms at the two sites; however, image analysis was fully standardized in this study which was enabled by center 2 sharing PREFUL analysis code with center 1 and the remaining data analysis being performed by center 1.
The correlations observed between 1 H VDP and 129 Xe VDP, LCI and FEV 1 were stronger than in previously published work and Bland-Altman bias and limits of agreement between 1 H VDP and 129 Xe VDP were smaller. 11,12 This could be due to considering the whole lungs when compared to a single slice, different acquisition parameters, field strengths and/or variations in calculating VDP between the studies. The studies also included different patient groups; clinically stable patients with CF and a broad range of disease severity in the current study (median age = 17.7 years, 55% were children), vs. children with CF undergoing pulmonary exacerbations in the study by Couch

Limitations
The CF cohorts were not well-matched physiologically between the two sites. Despite no significant differences in age, height, FEV 1 z-score or RV/TLC between the two cohorts, only center 1 enrolled patients older than 18 years who would likely have more progressive, severe disease. The group of patients scanned at center 2 had a higher proportion of females than the group of patients scanned at center 1, and there were no healthy controls scanned at center 1. The small number of healthy controls and patients scanned at center 2 is also a limitation. The small sample sizes were a limitation for the statistical methods used. Differences in spatial resolution, lung coverage and lung volumes during image acquisition between 129 Xe and 1 H MRI, and between centers, are limitations in the context of a truly matched comparison between techniques and sites.

Conclusions
Ventilation defect percentage calculated from 1 H freebreathing MRI showed strong correlations with 129 Xe VDP, LCI and FEV 1 in CF patients and controls scanned at two centers. When the data from both centers were pooled, bias between 1 H VDP and 129 Xe VDP was minimal and limits of agreement were moderately large. On a regional level, both similarities and differences between 129 Xe and 1 H ventilation images were observed across the range of disease severity but whole lung patterns of ventilation abnormality were similar in general. Some small defects and patchy ventilation heterogeneity observed in early-stage CF lung disease on 129 Xe ventilation images were not visualized with 1 H ventilation MRI. Furthermore, imaging acquisition and analysis protocols were standardized between two centers (within MRI system capabilities). In summary, this study supports the potential use of 1 H ventilation MRI in children and adults with CF.