Longitudinal 1H and 129Xe Lung MRI in Patients With Post‐COVID Residual Lung Abnormalities Authors: Laura C. Saunders, Guilhem J. Collier, Laurie J. Smith, Ho‐Fung Chan, Paul J. C. Hughes, Scarlett Strickland, Lotta Gustafsson, Thomas Newman, Megan Plowright, Zoë Gabriel, Lily M. Pearce, James T. Grist, Kher Lik Ng, Arthur Harrison, Charlotte E. Bolton, Jody Bray, Helen Marshall, Graham Norquay, Alberto M. Biancardi, Jimmy E. Ball, Neil J. Stewart, Kevin M. Johnson, Andrew J. Swift, Smitha Rajaram, John Blaikley, Stefan Stanel, Paul J. Collini, Gary H. Mills, Rod Lawson, Jonathan Brooke, Amanda T. Goodwin, Iain D. Stewart, Ling‐Pei Ho, Joseph Jacob, Thomas Meersmann, Galina E. Pavlovskaya, Fergus Gleeson, Ian P. Hall, Gisli Jenkins, Jim M. Wild Published: 2026 ABSTRACT ABSTRACTBackgroundIt is unclear how lung function may recover in patients with residual lung abnormalities (RLAs) following COVID‐19 pneumonia.PurposeTo evaluate lung function trends over time in patients with RLAs following hospitalization due to COVID‐19.Study TypeProspective, multicenter longitudinal cohort study.PopulationTwenty‐four participants hospitalized due to COVID‐19 with RLAs identified on CT ≥ 3 months postdischarge (median [IQR] age 69 (15) years; 3 female) underwent at least one MRI at 6 months (n = 16), 1 year (n = 19), or 2 years (n = 14).Field Strength/Sequence1.5 T. Dynamic contrast enhanced (DCE) 3D spoiled gradient echo, 129Xe steady state free precession (ventilation), 129Xe 3D spoiled gradient echo multiple b‐value (diffusion‐weighted), 129Xe 4‐echo flyback 3D radial (dissolved phase).AssessmentPulmonary blood flow, volume, and mean transit time (MTT) were calculated from DCE MRI. The fraction of 129Xe signal in the red blood cells to membrane (RBC:M) was calculated from the dissolved phase 129Xe acquisition. Ventilation defect percentage (VDP) was calculated from the 129Xe ventilation acquisition. Mean diffusive length scale (LmD) was calculated from the 129Xe diffusion‐weighted acquisition.Statistical TestsChanges in metrics with time and associations between metrics were assessed using mixed‐effect linear regression. Correlations were tested using Spearman's correlation coefficient. Regional differences were assessed using a Friedman's test with a Bonferroni adjustment. p < 0.05 was considered significant.ResultsPulmonary blood flow and MTT improved significantly over time (MTT: 6 months, 15.3 (IQR, 2.0); 1 year, 15.6 (1.4); 2 years, 15.0 (5.3); pulmonary blood flow: 6 months, 75.4 (IQR, 22.0); 1 year, 83.2 (47.4); 2 years, 107.3 (51.1)). RBC:M z‐score was low at all three visits (6 months, −2.85 (0.98); 1 year, −2.44 (1.34); 2 years, −2.60 (1.39)), with no improvement with time (p = 0.993). VDP and LmD did not significantly change with time (VDP: p = 0.100; LmD: p = 0.166).Data ConclusionImprovements in lung perfusion were measured; however, there was no corresponding enhancement in RBC:M.Evidence LevelLevel 2.Technical EfficacyStage 3. Introduction Patients hospitalized due to COVID‐19 experience a persisting high symptom burden postdischarge [1]. At 5 months postdischarge, patients exhibit a greater prevalence of multiorgan abnormalities, including a higher incidence of lung MRI abnormalities compared with control groups with comorbidities [2]. Follow‐up studies in COVID‐19 survivors have reported persistent ventilation, perfusion, and transfer capacity of the lung for carbon monoxide (TLCO) abnormalities up to 1 year after discharge [3, 4, 5]. It is estimated that up to 11% of surviving hospitalized patients during the early COVID‐19 pandemic had residual lung abnormalities (RLAs) which may represent either early interstitial lung disease or residual pneumonitis [6], with CT and TLCO abnormalities persisting at 1‐ and 2‐year follow‐up [7, 8]. Early interstitial lung disease can progress rather than resolve. While patients with progressive lung disease are likely to need more intensive follow‐up care and may benefit from antifibrotic therapies, patients whose RLAs resolve over time may require less intensive monitoring and care. It is, therefore, important to characterize functional lung impairment in patients with RLAs following COVID‐19 and whether such impairments resolve or progress over time. 1H and 129Xe lung MRI are used clinically and in research to evaluate lung function [9]. 1H dynamic contrast enhanced imaging enables measurement of pulmonary blood volume, pulmonary blood flow, and mean transit time, while129Xe lung MRI can visualize and quantify lung ventilation. Diffusion‐weighted imaging of 129Xe enables the measurement of the mean diffusive length scale (LmD), a measure of lung microstructure [10]. 129Xe gas is soluble in the pulmonary membrane/tissue (membrane, M) and blood (RBCs), and the fraction of 129Xe MR signal measured in the airspace (gas), M, and RBCs assesses gas transfer between the three compartments [11]. Together, these methods allow a comprehensive evaluation of pulmonary function. Patients who have had COVID‐19 have been found to have both xenon gas transfer and ventilation abnormalities [12, 13, 14], with more severe abnormalities found in hospitalized patients [15]. Although ventilation defects have been found to improve over time in patients post COVID‐19 [16], xenon gas transfer has not [17]. It is currently unclear how lung function abnormalities may progress in patients with RLAs following COVID‐19. The aim of this work was to evaluate lung function trends over time in patients with RLAs following hospitalization due to COVID‐19, using imaging and PFT markers. Materials and Methods Study Design and Participant Recruitment This study was approved by London‐Hampstead Research Ethics Committee (9/LO/1115), North West—Preston Research Ethics Committee (20/NW/0235) and the Tyne & Wear South Research Ethics Committee (19/NE/0330). All patients provided written informed consent. Patients who were hospitalized due to COVID‐19 and had RLAs on their clinical chest CT or research lung MRI at ≥ 3 months after discharge were prospectively recruited from Sheffield, Manchester, Nottingham, and Oxford NHS hospitals between May 2020 and Nov 2022 as part of the XMAS study [18, p. 59]. Participants underwent MRI at either Sheffield, Nottingham or Oxford hospitals. The following inclusion criteria were applied: A positive SARS‐CoV‐2 result from a nasal/pharyngeal or respiratory sampleHospitalization with a diagnosis of pneumonia (chest x‐ray or CT scan consistent with COVID‐19 infection).Evidence of RLAs on CT imaging at least 3 months after hospital admission, identified by the patient's clinical radiology report generated as part of clinical care, which was confirmed later during CT scoring.Participant chest could fit within the 129Xe chest coil. A positive SARS‐CoV‐2 result from a nasal/pharyngeal or respiratory sample A positive SARS‐CoV‐2 result from a nasal/pharyngeal or respiratory sample Hospitalization with a diagnosis of pneumonia (chest x‐ray or CT scan consistent with COVID‐19 infection). Hospitalization with a diagnosis of pneumonia (chest x‐ray or CT scan consistent with COVID‐19 infection). Evidence of RLAs on CT imaging at least 3 months after hospital admission, identified by the patient's clinical radiology report generated as part of clinical care, which was confirmed later during CT scoring. Evidence of RLAs on CT imaging at least 3 months after hospital admission, identified by the patient's clinical radiology report generated as part of clinical care, which was confirmed later during CT scoring. Participant chest could fit within the 129Xe chest coil. Participant chest could fit within the 129Xe chest coil. Standard MRI exclusion criteria were applied to all subjects. In addition, patients were excluded if they were unable to tolerate a test inhalation of 129Xe gas according to the supervising clinicians' judgment. Participants had both 1H and 129Xe MRI examinations together at 6 months, 1 year, and 2 years after hospitalization where possible. Participants could be recruited at any of the 6‐month, 1‐year, or 2‐year time points. MRI Acquisition and Analysis Full MRI sequence details can be found in Table S1. MRI examinations were undertaken on either a 1.5 T Signa HDx or Optima 450w (GE HealthCare, Milwaukee, WI, USA) scanner. MRI sequences were standardized across scanners. 129Xe doses were hyperpolarized to ~30% [19] and 129Xe imaging was acquired using a flexible quadrature transmit/receive vest coils (Clinical MR Solutions, Brookfield, Wisconsin, USA). For all 129Xe imaging sequences, 129Xe was administered using a bag of up to 1 L, with gas volume titrated according to subject height if subjects were < 160 cm, with inhalation starting from functional residual capacity [20]. The study followed standard MRI safety guidelines and excluded patients who were unable to tolerate a test inhalation of 129Xe gas according to the supervising clinicians' assessments. 129Xe lung ventilation (3D steady‐state free precession) MRI [21], 129Xe diffusion‐weighted MRI (3D spoiled gradient echo multiple b‐value sequence) [10], 129Xe dissolved‐phase xenon MRI (3D spectroscopic imaging of the gas and dissolved phase resonances) were acquired using the same sequence [11] and postprocessing pipeline for all sites. A structural 1H MR image was acquired after inhalation of a bag of air to match the lung inflation state of 129Xe MRI acquisitions. 1H dynamic contrast‐enhanced lung perfusion MRI was acquired at expiration (time‐resolved spoiled gradient echo with view sharing and parallel imaging). A half dose (0.05 mL/kg) of Gadovist (Bayer) was administered at an injection rate of 4 mL/s followed by a 20 mL saline flush at 4 mL/s. Patients were instructed to hold their breath for as long as possible and breathe shallowly thereafter. 3D radial ultra‐short echo time images [22] prospectively gated to expiration were also acquired as part of the imaging protocol to enable MR‐based visualization of lung structural abnormalities. 129Xe lung ventilation or 1H dynamic contrast enhanced lung perfusion imaging was not performed on participants scanned at Nottingham or Oxford. Analysis of the 129Xe ventilation images was undertaken [23] from which ventilation defect percentage (VDP) and low ventilation percentage were calculated. The ventilation heterogeneity index of the segmented lung ventilation images was also calculated [20, 24] as a measure of the heterogeneity of the ventilated airspace. The mean diffusive length scale (LmD, a metric of mean acinar airway dimensions) was calculated for each voxel of the 129Xe diffusion‐weighted images using a stretched exponential model of 129Xe gas diffusion [25] and the standard deviation of the mean diffusive length scale (LmD SD) was also calculated as a measure of acinar airway dimension heterogeneity. Maps of the fraction of xenon signal in the RBCs to airspace (RBC:gas), membrane to airspace (M:gas), and RBCs to membrane (RBC:M) were calculated from the 3D 129Xe dissolved phase images [11]. Z‐scores were calculated for each metric based on published data from a cohort of 62 healthy participants [26]. The transverse relaxation time (T 2*) of the 129Xe in the membrane and RBC was calculated using a triple Lorentzian fit of the data in the frequency domain. Perfusion time‐course images were coregistered to correct for patient motion using ANTs software [27] and for each voxel, signal was smoothed via Savitzky‐Golay filtering and voxel‐wise concentration maps were calculated using relative signal enhancement. The arterial input function was defined using a manual region of interest in the pulmonary artery. Semiquantitative maps of relative pulmonary blood volume, relative pulmonary blood flow and mean transit time were calculated [28], hereafter referred to as pulmonary blood volume and pulmonary blood flow for brevity. The median and interquartile range (IQR) of each whole‐lung map were calculated, with the IQR of each map hereafter referred to as spatial heterogeneity, that is, pulmonary blood volume spatial heterogeneity. Segmentations excluding main pulmonary vessels were applied to the maps. All images underwent a quality‐control evaluation by a team of experienced 129Xe and lung perfusion MRI physicists and medical image software developers (experience range: 10–15 years), where images with low SNR, image artifacts, or significant respiratory motion were excluded from analysis. CT Acquisition and Analysis CT imaging was acquired as part of patients' routine clinical care at each institution and was retrospectively accessed for scoring. CTs were contrast‐enhanced if clinical need required it (n = 6). All CTs were performed at inspiration breath hold. For CT scans acquired in Sheffield (n = 17), all CTs were volumetric acquisitions performed on either a TOSHIBA Aquilion PRIME (n = 6), Canon medical Systems Aquilion ONE (n = 11), at 1.0 mm slice thickness and a 512 × 512 acquisition matrix with standard reconstruction. KVP was either 100 or 120 kV and collimation was 1.00 mm. Resultant exposure was between 47 and 206 mA. For CT scans acquired in Nottingham (n = 3), scanner manufacturer was unavailable; all scans were acquired with either 1.0 mm (n = 2) or 2.0 mm (n = 1) slice thickness and a 512 × 512 acquisition matrix. For CT scans acquired in Oxford (n = 3), all CT acquisitions were performed at 1 L inspiration breath hold on a GE scanner with slice thickness 0.625 mm. Information on acquisition was unavailable for one CT scan. Research ethics enabled sharing of clinical data but not transfer of clinical CT images to another site for scoring. CTs were therefore scored by a different individual at each site. CT scorers were a chest radiologist (SR, 14 years' experience), chest physician (AG, 11 years' experience; KLN, in‐training) depending on the site. CTs were assessed for the following pulmonary abnormalities: total lesions, ground glass opacities, consolidation, reticulation, and fibrotic‐like changes (traction bronchiectasis, parenchymal bands, and/or honeycombing). A semiquantitative CT score was assigned based on the area involved in each of the five lung lobes: 0 (no involvement), 1 (< 5%), 2 (5%–25%), 3 (26%–49%), 4 (50%–75%), and 5 (> 75%). The total CT severity score was calculated by summing the individual lobar scores (possible range from 0 to 25) [29]. Pulmonary Function Testing (PTF) Whenever possible, participants underwent same‐day PFTs. Forced expiratory volume in 1 s (FEV1), forced vital capacity (FVC), FEV1/FVC, the transfer factor for carbon monoxide (TLCO), and the carbon monoxide transfer coefficient (KCO) were measured and presented as z‐scores [30, 31]. If same‐day PTFs were unavailable, the most recent routine outpatient PFT within 4 months of the MRI was used. For participants with multiple visits, outpatient PFTs were matched to the MRI visit closest to their acquisition. Statistical Analysis The primary outcome of this study was whether longitudinal change could be measured from functional lung MRI metrics, which was assessed using mixed‐effect linear regression using a random intercept, with Maximum Likelihood estimation in IBM SPSS Statistics 27 (SPSS, New York, USA). Regression models were adjusted for age and sex if not already adjusted via z‐scores, and ventilation metrics were also adjusted for smoking status. A secondary outcome of this study was whether associations between metrics could be established, which was assessed using repeated measures mixed‐effect linear regression using a random intercept. Within‐subject tests for differences between upper, middle, and lower regions of RBC:M, M:gas, RBC:gas, and LmD were performed at all three visits (6 months, 1 year, 2 years) using a Friedman's test with a Bonferroni adjustment for multiple testing and denoted p adj. CT scores were tested for correlation with MRI and PFT metrics at 6 months using Spearman's correlation coefficient; MRC breathlessness score was tested for correlation with MRI and PFT metrics at 3 months, as well as CT data at 6 months, using Spearman's correlation coefficient. Acute hospital admission duration was correlated with CT metrics and 6‐month MRI and PFT metrics. Data are presented as median (IQR) unless otherwise specified. For data presented as z‐scores, data < −1.64 are considered abnormal. A sample size calculation was not performed because this was an exploratory study. p < 0.05 was considered statistically significant. Results A total of 26 participants were recruited and underwent MRI. Of the 26 participants, 1 was not included in this analysis as they were not hospitalized due to COVID‐19 and 1 participant did not have images of sufficient quality for analysis. Therefore, 24 participants were included in the final analysis. MRI visits were undertaken by 16 participants at 6 months, 19 participants at 1 year and 14 participants at 2 years, see Figure 1. Nine participants attended all three visits, two participants attended 6 month and 1 year visits only, five participants attended 1 year and 2 year visits only, five participants attended the 6 month visit only, and three participants attended visit 1 only. Full details of participants' visits and missing data are given in Table S2. A diagram showing participant recruitment. Participants were recruited at both 6 month and 1 year visits. One patient follow‐up was excluded as it was performed on a 3 T field strength scanner. Of the 24 participants, 21 were male with a median age of 69 (15) years and BMI of 31 (6) kg/m2. Smoking history was documented for 22/24 participants; 10 of those participants did not smoke, and 12 participants either did smoke or had previously smoked. Of the 24 participants, 9 had severe acute disease (World Health Organization score ≥ 6) during their initial hospitalization [32]. Table 1 gives further demographic and clinical information. TABLE 1Participant demographics and clinical information.VariablesMedian (range)Age (years)69 (53–79)Sex, N (%)21 male, 3 femaleBMI (kg/m2)30.4 (20.4–38.9)Length of stay (weeks)4 (0–17)Year of initial hospitalization15202092021ISARIC 4C score a 9 (4–18)MRC breathlessness score (6 weeks) b 3 (2–5)MRC breathlessness score (3 months) c 3 (1–5)MRC breathlessness score (1 year) b 3 (1–5)GroupNumber of patientsPrior respiratory historyNo previous respiratory conditions22Asthma2Comorbidities 4C score (2)01015≥ 29World Health Organization score4 (Hospitalized, oxygen mask or nasal prongs)115 (Hospitalized, noninvasive mechanical ventilation or high‐flow nasal cannula)46 (Hospitalized, intubation and invasive mechanical ventilation)87 (Hospitalized, invasive mechanical ventilation + additional support such as pressors or extracardiac membranous oxygenation)1Tobacco use historyUnknown2Nonsmokers10Ever‐smokers12Treatment during admissionAntibiotics16/24Dexamethasone19/24Prednisolone2/24Remdesivir13/24Tocilizumab3/24Baricitinib1/24Plasma2/24Treatment dose dalteparin12/24Colchicine0/24Aspirin0/24Abbreviations: BMI = body mass index; ISARIC = International Severe Acute Respiratory and Emerging Infection Consortium; MRC = Medical Research Council. a 23 patients. b 12 patients. c 19 patients. Participant demographics and clinical information. Abbreviations: BMI = body mass index; ISARIC = International Severe Acute Respiratory and Emerging Infection Consortium; MRC = Medical Research Council. 23 patients. 12 patients. 19 patients. Patient Symptoms Median MRC breathlessness score at 6 weeks, 3 months, and 1 year postdischarge was 3 (range, 2–5), 3 (1–5), and 3 (1–5). Of the 14 participants with multiple MRC breathlessness scores available, 6 showed an improvement in score and 8 showed no change in score. MRC breathlessness score did not worsen in any patients with available data. MRC breathlessness score at 3 months significantly correlated with 3 month RBC:M z‐score (r = −0.85, n = 13), RBC:gas z‐score (r = −0.84, n = 13), and TLCO z‐score (r = −0.68, n = 10). No other significant correlations with MRI or PFT metrics were found. MRC breathlessness score at 3 months correlated significantly with reticulation score at 1 year (r = 0.51, n = 18). No other significant correlations with CT data were found. CT CT scoring was available for 23/24 participants, median time between CT and hospital discharge of 5 (3) months. One participant had CT imaging available for confirmation of adherence to the inclusion criteria, but the data was not able to be scored. Ground glass opacities were present on CT in 21/23 participants, reticulation was present in 22/23 participants, and fibrosis was present in 10/23 participants. Additional details on CT scores are given in Figure S1. CT scores were correlated with 6‐month MRI data. Ground glass opacity score correlated with mean transit time (r = −0.652) and KCO z‐score (r = 0.650), see Figure S2. Reticulation score correlated positively with LmD standard deviation (SD) (r = 0.611), negatively with M:gas z‐score (r = −0.557), and negatively with FVC z‐score (r = −0.704). Fibrosis score correlated with TLCO z‐score and KCO z‐score (r = −0.695 and r = −0.608, respectively); however, only three participants with TLCO and KCO z‐score had non‐zero fibrosis scores so this should be interpreted with caution. Total CT score correlated negatively with RBC T2* (r = −0.631). No other significant correlations between CT scores and MRI or PFT metrics were found. PFTs PFT data were available for 13/16 participants at 6 months, 18/19 participants at 1 year, and 10/14 participants at 2 years. One participant could not achieve a reliable FVC measurement at their 1 year visit due to cough, and their FVC measurement was excluded. Data at visit are summarized in Table 2. TABLE 2Summary of MRI and PFT parameters at 3 timepoints.MRI or PFT metric6 months1 year2 years N Median (IQR) N Median (IQR) N Median (IQR)Time since discharge (weeks)1623 (7.5)1952 (10)14101 (10)Time between MRI and PFT (days)1316 (31)180 (17)100 (0)FEV1 z‐score13−1.31 (0.80)18−0.84 (1.76)10−0.87 (0.61)FVC z‐score13−1.60 (0.82)17−1.04 (1.57)10−1.18 (0.88)FEV1/FVC z‐score130.52 (0.84)170.30 (1.23)100.56 (1.31)TLCO z‐score13−2.19 (2.24)18−2.49 (1.64)10−2.50 (1.64)KCO z‐score13−1.12 (1.56)18−0.96 (1.62)10−0.73 (1.16)VDP (%)131.4 (1.4)122.0 (2.1)122.2 (3.6)Low ventilation percentage (%)1314.0 (2.4)1214.5 (3.7)1213.9 (4.2)Ventilation heterogeneity index139.3 (2.3)129.6 (2.0)129.2 (1.5)LmD (μm)16294 (19)14295 (23)12295 (26)LmD SD (μm)1668.8 (12.8)1464.5 (11.3)1264.6 (11.2)RBC:M140.19 (0.04)170.20 (0.08)130.22 (0.06)RBC:M z‐score14−2.85 (0.98)17−2.44 (1.34)13−2.60 (1.39)M:gas140.011 (0.022)170.009 (0.005)130.010 (0.003)M:gas z‐score140.86 (1.06)170.12 (2.24)140.52 (1.53)RBC:gas140.0021 (0.0012)170.0016 (0.0007)130.0020 (0.0009)RBC:gas z‐score14−0.96 (2.01)17−1.58 (1.27)13−0.82 (1.00)M T2* (ms)142.61 (0.21)172.57 (0.21)132.53 (0.19)RBC T2* (ms)142.41 (0.21)172.34 (0.27)132.24 (0.37)Pulmonary blood volume (a.u.)1119.6 (7.2)1522.1 (10.7)922.7 (9.5)Pulmonary blood volume spatial heterogeneity (a.u.)1120.2 (8.2)1519.4 (10.7)921.4 (10.8)Pulmonary blood flow (a.u.)1175.4 (22.0)1583.2 (47.4)9107.3 (51.1)Pulmonary blood flow spatial heterogeneity (a.u.)1179.1 (42.5)1576.0 (48.9)996.0 (46.4)Mean transit time (s)1115.3 (2.0)1515.6 (1.4)915.0 (5.3)Mean transit time spatial heterogeneity (s)111.7 (0.9)151.3 (0.6)91.1 (0.4)Abbreviations: FEV1: forced expiry volume in 1 s; FVC: forced vital capacity; IQR: interquartile range; KCO: carbon monoxide transfer coefficient; LMD: mean diffusive length scale; M: membrane; PFT: pulmonary function tests; RBC: red blood cell; SD: standard deviation; TLCO: transfer factor for carbon monoxide; VDP: ventilation defect percentage. Summary of MRI and PFT parameters at 3 timepoints. Abbreviations: FEV1: forced expiry volume in 1 s; FVC: forced vital capacity; IQR: interquartile range; KCO: carbon monoxide transfer coefficient; LMD: mean diffusive length scale; M: membrane; PFT: pulmonary function tests; RBC: red blood cell; SD: standard deviation; TLCO: transfer factor for carbon monoxide; VDP: ventilation defect percentage. Spirometry showed abnormalities consistent with a restrictive pattern, with abnormal FVC z‐score in 5/13 participants at 6 months, 3/18 participants at 1 year, and 2/10 participants at 2 years. One participant had spirometry consistent with an obstructive lung disease pattern (abnormal FEV1/FVC z‐score) at 6 months. Lung volumes significantly increased toward normal values with time (Table 3 and Figure 2), with an increase of FVC z‐score and FEV1 z‐score of 0.009 and 0.009 per week, respectively, which equates to an increase of 0.7 and 0.7, respectively, during the 18 month time frame of the study. TABLE 3Effect of time (measured in weeks) on MRI and PFT metrics tested using a mixed‐effect linear regression analysis with a random intercept.Independent variableEstimated coefficientLower 95% CIUpper 95% CI p FEV1 z‐score0.0090.0040.0130.002FVC z‐score0.0090.0040.0140.001FEV1/FVC z‐score−0.002−0.0090.0050.633TLCO z‐score0.005−0.0010.0110.087KCO z‐score−0.001−0.0050.0030.550VDP (%)0.009−0.0020.0200.100Low ventilation percentage (%)−0.014−0.0290.0040.057VHI−0.004−0.0110.0020.194LmD (μm)−0.093−0.2280.0410.166LmD SD (μm)−0.065−0.105−0.0250.003RBC:M z‐score−0.024 × 10−2 −0.0050.0050.993RBC:M−0.013 × 10−3 −0.0003090.0002820.926RBC:gas z‐score0.016 × 10−1 −0.00690.01020.696RBC:gas0.008 × 10−4 −0.0000060.0000040.735M:gas z‐score0.026 × 10−1 −0.00520.01050.490M:gas0.006 × 10−3 −0.0000110.0000220.490M T2* (ms)−0.008 × 10−1 −0.00170.00010.094RBC T2* (ms)−0.025 × 10−1 −0.0044−0.00060.009Pulmonary blood volume (a.u.)0.026−0.0300.0830.350Pulmonary blood volume spatial heterogeneity (a.u.)0.030−0.0270.0880.285Pulmonary blood flow (a.u.)0.2720.0150.530.039Pulmonary blood flow spatial heterogeneity (a.u.)0.2410.0100.4930.032Mean transit time (s)−0.021−0.038−0.0050.015Mean transit time spatial heterogeneity(s)−0.007−0.010−0.004< 0.001 Note: Regression models are adjusted for age and sex unless already adjusted via a z‐score. Ventilation metrics are also adjusted for smoking status.Abbreviations: FEV1: forced expiry volume in 1 s; FVC: forced vital capacity; KCO carbon monoxide transfer coefficient; LMD: mean diffusive length scale; M: membrane; PFT: pulmonary function tests; RBC: red blood cell; SD: standard deviation; TLCO: transfer factor for carbon monoxide; VDP: ventilation defect percentage. Effect of time (measured in weeks) on MRI and PFT metrics tested using a mixed‐effect linear regression analysis with a random intercept. Note: Regression models are adjusted for age and sex unless already adjusted via a z‐score. Ventilation metrics are also adjusted for smoking status. Abbreviations: FEV1: forced expiry volume in 1 s; FVC: forced vital capacity; KCO carbon monoxide transfer coefficient; LMD: mean diffusive length scale; M: membrane; PFT: pulmonary function tests; RBC: red blood cell; SD: standard deviation; TLCO: transfer factor for carbon monoxide; VDP: ventilation defect percentage. Scatter plots of spirometry lung volume (FEV1 and FVC z‐scores) plotted with time (A, B) and with M:gas z‐score (C), with solid lines indicating the linear trend and confidence intervals predicted by the mixed‐effect linear regression analysis. Carbon monoxide gas transfer (TLCO z‐score) was abnormal in 10/13 participants at 6 months, 13/18 participants at 1 year, and 8/10 participants at 2 years. TLCO and KCO did not significantly change with time indicating persistent gas transfer abnormalities for the majority of subjects. MRI 129Xe lung ventilation imaging showed small peripheral ventilation defects and areas of low or heterogeneous ventilation (Figure 3). Median VDP was 1.4 (1.4)% at 6 months, 2.0 (2.1)% at 1 year, and 2.2 (3.6)% at 2 years. There was no overall change in lung ventilation metrics (VDP p = 0.100, low ventilation percentage p = 0.057, or ventilation heterogeneity index p = 0.194) with time (Table 3). Summary of ventilation data: (A) a jitter plot of VDP at each visit, with participant color coded by their smoking status, (B) single central coronal slices from the lung ventilation images in three participant at 6 months and 2 years with whole‐lung VDP shown under each image and areas of ventilation defect indicated with white arrows, (C) a scatter plot of VDP with time with solid lines indicating the linear trend and confidence intervals. Diffusion weighted 129Xe MRI imaging showed that LmD did not change significantly with time (p = 0.166); however, the heterogeneity of acinar airway dimensions (LmD SD) decreased with time (Table 3, Figure 4). Summary of diffusion data: (A) a scatter plot of mean linear intercept, a measure of acinar airway dimensions (LmD) with time, (B) a single representative coronal lung slice from LmD maps for three participant with diffusion‐weighted MRI acquired at both 6 months and 2 year time points, with mean and standard deviation of LmD from the whole lung shown under each representative slice, (C) A scatter plot of the standard deviation of LmD with time. Solid lines in (A) and (C) indicate the linear trend and confidence intervals predicted by the mixed‐effect linear regression analysis. Dissolved phase 129Xe xenon imaging showed RBC:M was abnormally low (median z‐score < −1.64) at all three visits (Table 2, Figure 5). In Figure 5, Participants 2 and 3 showed persistent RBC:M defects in the upper right lung. There was no significant change in RBC:M z‐score with time (p = 0.993). However, the T2* of the 129Xe in the RBCs (RBC T2*) significantly decreased with time (p = 0.009, Table 3). Summary of xenon gas transfer (RBC:M z‐score) data: (A) A scatter plot of RBC:M z‐score with time. (B) A single coronal slice from the RBC:M maps of three participant at 6 months and 2 years visits, two showing persistent upper lobe RBC:M defects and one showing improvement at 2 years, (C) a scatter plot of RBC:M z‐score and TLCO z‐score. Solid lines in (A) and (C) indicate the linear trend and confidence intervals predicted by the mixed‐effect linear regression analysis. M:gas z‐score was within the healthy range at all three visits; however, three participants had low M:gas (z‐score < −1.64) at one or more visits, and six participants had high M:gas at one or more visits (> 1.64). M:gas z‐score did not change significantly with time (p = 0.490, Table 3). Box plots of MRI metrics acquired in participants hospitalized due to COVID‐19 with RLAs and previously published MRI metrics acquired in participants hospitalized due to COVID‐19 with normal structural lung imaging at 3 months after hospitalization [33] are shown in Figure S3. Mean transit time significantly decreased with time and pulmonary blood flow significantly increased with time (Table 3 and Figure 6), indicating improvements in lung perfusion with time. There were no significant changes in pulmonary blood volume over time. A summary of lung perfusion data: (A, C) scatterplots of pulmonary blood volume and mean transit time with time, with solid lines indicating the linear trend and confidence intervals predicted by the mixed‐effect linear regression analysis. (B) Example coronal slices from the peak perfusion maps (left, maximum signal intensity across all time points) and the maps of mean transit time (right, overlaid over the corresponding perfusion image) in a single patient at 6 months, 1 year, and 2 years. Associations Between MRI, Pulmonary Function, and Acute Disease Severity As anticipated, xenon gas transfer between the membrane and RBCs (RBC:M) was significantly associated with TLCO z‐score (see Figure 5) and KCO z‐score. Reduced FVC z‐score was associated with increased xenon uptake by the alveolar membrane (M:gas z‐score) and reduced TLCO z‐score, suggesting that participants with lung restriction had membrane thickening and reduced gas transfer. However, FVC z‐score was not significantly associated with higher VDP. Full details of significant associations of RBC:M z‐score, FVC z‐score, LmD, and M:gas z‐score are shown in Table 4. TABLE 4Repeated measures mixed‐effect linear regression analysis of RBC:M z‐score, FVC z‐score, mean transit time and M:gas z‐score with the following metrics: RBC:M z‐score, M:gas z‐score, LmD, VDP, pulmonary blood volume, pulmonary blood flow, mean transit time, FEV1 z‐score, FVC z‐score, TLCO z‐score, KCO z‐score.VariableEstimate (lower CI, upper CI) p Associations with RBC:M z‐scoreTLCO z‐score0.52 (0.28, 0.77)< 0.001KCO z‐score0.64 (0.30, 0.98)< 0.001Associations with FVC z‐scoreM:gas z‐score−0.30 (−0.59, −0.02)0.036TLCO z‐score0.52 (0.27,0.78)< 0.001FEV1 z‐score0.89 (0.71,1.07)< 0.001Associations with mean transit timePulmonary blood flow (a.u.)−0.050 (−0.068,‐0.031)< 0.001Associations with M:gas z‐scoreLmD (μm)−0.021 (−0.040,‐0.002)0.030Mean transit time (s)−0.11 (−0.21,‐0.03)0.01FEV1 z‐score−0.46 (−0.82,‐0.09)0.02FVC z‐score−0.45 (−0.82,‐0.08)0.02 Note: Only significant results are shown.Abbreviations: FEV1: forced expiry volume in 1 s; FVC: forced vital capacity; KCO carbon monoxide transfer coefficient; LMD: mean diffusive length scale; M: membrane; RBC: red blood cell; SD: standard deviation; TLCO: transfer factor for carbon monoxide; VDP: ventilation defect percentage. Repeated measures mixed‐effect linear regression analysis of RBC:M z‐score, FVC z‐score, mean transit time and M:gas z‐score with the following metrics: RBC:M z‐score, M:gas z‐score, LmD, VDP, pulmonary blood volume, pulmonary blood flow, mean transit time, FEV1 z‐score, FVC z‐score, TLCO z‐score, KCO z‐score. Note: Only significant results are shown. Abbreviations: FEV1: forced expiry volume in 1 s; FVC: forced vital capacity; KCO carbon monoxide transfer coefficient; LMD: mean diffusive length scale; M: membrane; RBC: red blood cell; SD: standard deviation; TLCO: transfer factor for carbon monoxide; VDP: ventilation defect percentage. At 6 months, participants with a longer stay in hospital had significantly lower FEV1 (r = −0.556), FVC (r = −0.699), TLCO z‐score (r = −0.646), and LmD SD (r = 0.660) at 6 months. There were no other significant correlations between acute hospitalization admission length and MRI metrics or CT scores at 6 months. Regional Variations in Lung Function MRI Figure S4 shows example images and maps from four participants with different regional patterns of lung function changes despite abnormal gas transfer in all participants and demonstrates the heterogeneity in regional lung function patterns across the participants. Discussion This study conducted a comprehensive, longitudinal assessment of lung function using 129Xe and 1H MRI methods in patients with RLAs following hospitalization due to COVID‐19. The abnormalities identified in pulmonary gas transfer and perfusion did not progress between 6 months and 2 years after hospital discharge, suggesting that these participants are not at risk of progressive interstitial lung disease. While lung perfusion significantly improved in this period, gas transfer did not, and xenon gas transfer was not associated with pulmonary perfusion metrics, which suggests persistent underlying alveolar‐capillary membrane dysfunction. Participants in this study were older and predominantly male, which is consistent with the characteristics of patients most at risk of hospitalization and more severe acute COVID‐19 [1]. The participants in this study had reduced TLCO and RBC:M, slightly increased M:gas, and reduced FEV1 and FVC, a pattern consistent with a cluster of long COVID patients previously identified with increased 129Xe membrane uptake, ground glass opacities on CT, restrictive PFT pattern and low RBC:M [34]. Dissolved phase xenon imaging enabled gas transfer between the airspace, membrane, and pulmonary capillaries (RBC) to be measured separately, providing additional insight into the patterns of gas transfer in patients with RLAs following COVID‐19. While abnormalities in xenon gas transfer to the capillaries (RBC:M) were consistent, 6/24 patients (25%) also had high M:gas during at least one visit, consistent with a pattern of “diffusion block” due to membrane thickening [35]. This is broadly consistent with previous findings of RBC:M in patients with idiopathic pulmonary fibrosis (IPF) [35, 36] who have been reported to demonstrate a variety of different patterns of gas transfer, including patients with membrane thickening alongside impaired RBC transfer, but also patients with low or normal membrane transfer but reduced RBC transfer [35]. RBC:M did not change with time which is consistent with previous longitudinal findings in patients after COVID‐19 in a mixed ambulatory and hospitalized cohort [17], with median RBC:M and RBC:gas lower in this cohort of participants than in patients previously hospitalized due to COVID‐19 without RLAs [33]. Previous work has shown correlation between DLCO and perfusion abnormalities on dual‐energy CT in patients with long COVID 6 months after acute infection [37]. However, this work found mean transit time and pulmonary blood flow improved across the same time period. Improvement of pulmonary blood flow without improvement in gas transfer efficiency may occur if blood flow improves within an area of lung that is either unventilated (resulting in worsened VQ matching) or diffusion limited (due to thickened or impaired alveolar capillary membrane). In this study, both ventilation abnormalities and membrane thickening were identified in subsets of patients, and for those patients this is a plausible explanation for these findings. However, methodological differences may also play a part in explaining this disparity, as perfusion imaging visualizes blood flow across the entire pulmonary vascular bed, whereas xenon only dissolves in RBCs within the pulmonary capillaries. Participants in this study had minor ventilation abnormalities that did not change with time. Previous findings on ventilation abnormalities after COVID‐19 are varied, with some findings showing a higher burden of ventilation abnormalities after COVID‐19 [12, 16, 38] and others reporting a low burden [33, 34, 39]. Reasons for this may include differences in pre‐existing ventilation abnormalities and respiratory comorbidities. FEV1 z‐score and FVC z‐score increased between 6 months and 2 years; however, VDP was not significantly associated with FVC z‐score and did not increase over the same period. Therefore, the improvements seen in FEV1 and FVC z‐scores are not due to an increase in ventilated lung, but instead either represent improvements in lung compliance or improvements in respiratory muscle strength. Participants in this study did not have increased LmD compared to data on age and sex trends in a healthy cohort [40] and had smaller acinar airway dimensions (LmD) than patients with IPF of a similar age to the patients in this study [41]. Patients with inflammatory interstitial lung disease have been shown to have similar RBC:M but decreased LmD when compared to patients with fibrotic interstitial lung disease [42]. In addition, the participants in this study have an overall improvement in FVC z‐score, suggesting resolving rather than irreversible lung damage. Together, this suggests that the participants in this study represent an inflammatory rather than fibrotic profile, which is also consistent with the higher prevalence of ground glass opacities than fibrosis seen on CT. Similarly, improvements in pulmonary perfusion metrics suggest microvascular recovery and align more closely with an inflammatory rather than a fibrotic phenotype. Limitations This study faced challenges in participant recruitment, retention, and data missingness, resulting in a small sample size, which is a key limitation of this work. The primary reason for the low recruitment was the progressive reduction in hospital admission rates for severe COVID‐19 as the pandemic evolved, yielding fewer patients who met the inclusion criteria. We therefore performed a linear mixed model analysis, adjusted for participant age and sex where appropriate. While it was assumed that data missingness was random, it is possible that participants who did not experience improvement over the 18‐month time frame were more motivated to continue with the study, leading to an underestimation of potential improvement. Therefore, the statistical results should be interpreted cautiously as this may weaken the statistical power of the study. Despite these limitations, the findings clearly indicate that participants did not deteriorate over the study period. Further, only limited symptom data were recorded, and therefore a more thorough exploration of the link between lung function and symptom recovery could not be performed. Finally, a further limitation is that CT images were read individually at each site. Conclusion In patients hospitalized due to COVID‐19 with RLAs, improvements with time were seen in lung volume and perfusion; however, xenon gas transfer and TLCO remained abnormally low with no change observed between 6 months and 2 years. This persistent gas transfer impairment, despite pulmonary perfusion recovery, suggests that underlying alveolar‐capillary membrane dysfunction may not fully resolve, indicating potential long‐term consequences for gas exchange efficiency and highlights the need for treatments that target this deficit. Funding This independent research was carried out at the National Institute for Health and Care Research (NIHR) Sheffield Biomedical Research Centre (BRC), NIHR Nottingham Biomedical Research Centre, and NIHR Oxford Biomedical Research Centre. This study was supported by NIHR UCLH Biomedical Research Centre. The UK Interstitial Lung Disease Long‐COVID19 study (UKILD‐Long COVID): understanding the burden of Interstitial Lung Disease in Long COVID (Medical Research Council, MR/W006111/1). Expansion of state‐of‐the‐art MR imaging infrastructure for pulmonary disease stratification: POLARIS (Medical Research Council, MR/M008894/1), INSIGNEO Sheffield, J.J. was funded by the Wellcome Trust. J.J. and by the NIHR UCLH Biomedical Research Centre. IPH is an NIHR Senior Investigator. The views expressed are those of the author(s) and not necessarily those of the MRC, Sheffield, Oxford, Nottingham, or UCLH BRC or the Department of Health and Social Care. Supporting information Table S1: Typical imaging parameters used for MRI acquisition. Table S2: Visualization of patient recruitment and missing data. Figure S1: Histograms of CT scores for (A) fibrotic‐like changes, (B) reticulation, (C) ground glass opacities (GGO), and (D) whole lung CT score, where a maximum score is 25. Figure S2: Scatterplots of CT scores and 6‐month MRI and PFT data. Specifically: a) ground glass opacity CT score and KCO z‐score, b) reticulation CT score and FVC z‐score, c) fibrosis CT score and TLCO z‐score and D) total CT score and RBC T2*. Figure S3: Showing LmD (A), VDP (B), RBC:M z‐score (C), M:gas z‐score (D), and RBC:gas z‐score (E) at 6‐month and 1‐year time points in patients hospitalized due to COVID‐19 with RLAs, alongside previously published data acquired in patients hospitalized due to COVID‐19 with normal lung structural imaging at 3 months after hospitalization (patients without RLAs) (8). Figure S4: Example ultra‐short echo time (UTE) and ventilation images, and LmD, RBC:M, M:gas, and MTT maps for four patients with different patterns of regional alignment between functional metrics.