Echo time-dependent observed T1 and quantitative perfusion in chronic obstructive pulmonary disease using magnetic resonance imaging

Introduction Due to hypoxic vasoconstriction, perfusion is interesting in the lungs. Magnetic Resonance Imaging (MRI) perfusion imaging based on Dynamic Contrast Enhancement (DCE) has been demonstrated in patients with Chronic Obstructive Pulmonary Diseases (COPD) using visual scores, and quantification methods were recently developed further. Inter-patient correlations of echo time-dependent observed T1 [T1(TE)] have been shown with perfusion scores, pulmonary function testing, and quantitative computed tomography. Here, we examined T1(TE) quantification and quantitative perfusion MRI together and investigated both inter-patient and local correlations between T1(TE) and quantitative perfusion. Methods 22 patients (age 68.0 ± 6.2) with COPD were examined using morphological MRI, inversion recovery multi-echo 2D ultra-short TE (UTE) in 1–2 slices for T1(TE) mapping, and 4D Time-resolved angiography With Stochastic Trajectories (TWIST) for DCE. T1(TE) maps were calculated from 2D UTE at five TEs from 70 to 2,300 μs. Pulmonary Blood Flow (PBF) and perfusion defect (QDP) maps were produced from DCE measurements. Lungs were automatically segmented on UTE images and morphological MRI and these segmentations registered to DCE images. DCE images were separately registered to UTE in corresponding slices and divided into corresponding subdivisions. Spearman’s correlation coefficients were calculated for inter-patient correlations using the entire segmented slices and for local correlations separately using registered images and subdivisions for each TE. Median T1(TE) in normal and defect areas according to QDP maps were compared. Results Inter-patient correlations were strongest on average at TE2 = 500 μs, reaching up to |ρ| = 0.64 for T1 with PBF and |ρ| = 0.76 with QDP. Generally, local correlations of T1 with PBF were weaker at TE2 than at TE1 or TE3 and with maximum values of |ρ| = 0.66 (from registration) and |ρ| = 0.69 (from subdivision). In 18 patients, T1 was shorter in defect areas than in normal areas, with the relative difference smallest at TE2. Discussion The inter-patient correlations of T1 with PBF and QDP found show similar strength and TE-dependence as those previously reported for visual perfusion scores and quantitative computed tomography. The local correlations and median T1 suggest that not only base T1 but also the TE-dependence of observed T1 in normal areas is closer to that found previously in healthy volunteers than in defect areas.

Introduction: Due to hypoxic vasoconstriction, perfusion is interesting in the lungs.Magnetic Resonance Imaging (MRI) perfusion imaging based on Dynamic Contrast Enhancement (DCE) has been demonstrated in patients with Chronic Obstructive Pulmonary Diseases (COPD) using visual scores, and quantification methods were recently developed further.Inter-patient correlations of echo time-dependent observed T 1 [T 1 (TE)] have been shown with perfusion scores, pulmonary function testing, and quantitative computed tomography.Here, we examined T 1 (TE) quantification and quantitative perfusion MRI together and investigated both inter-patient and

Introduction
Monitoring disease progression and therapy response in chronic obstructive pulmonary disease (COPD) is commonly based on pulmonary function testing (PFT), which provides global measures for lung function impairment (1).Computed tomography (CT) is used for the regional assessment of lung structure and distinction of COPD phenotypes (2).Magnetic resonance imaging (MRI) of the lungs is particularly challenging due to the low proton density in lung parenchyma, respiratory motion and especially the extremely short T 2 * relaxation time caused by the structure of the lungs (3).MRI was suggested as a radiation-free, non-invasive alternative to PFT and CT and has the additional benefit of techniques for the regional assessment of functional impairment.MRI methods based on image registration and the oscillation of signal intensity allow for the quantification of pulmonary perfusion and ventilation (4,5).However, arguably the most practical and established approach for functional lung MRI in clinical routine is dynamic contrast enhanced (DCE) MRI.In the lungs, local perfusion is regulated by hypoxic pulmonary vasoconstriction which links perfusion directly to ventilation.Accordingly, the detection of perfusion defects can thus be used to identify abnormal ventilation as long as ventilation/perfusion mismatches can be disregarded.Reduction of the pulmonary vascular bed in emphysema also results in perfusion defects.A visual scoring system for DCE-MRI by radiologist readers has been demonstrated to provide a measure of functional lung abnormality that correlates well with other clinical metrics, including lung function tests in COPD and cystic fibrosis (6)(7)(8)(9)(10).
The determination of this visual score requires trained radiologists and is simultaneously limited in spatial resolution to the level of lung lobes.It is also severely limited in numerical resolution since scores cannot practically have fine graduations and it also essentially disregards the difference in volume between lung lobes.Quantification of perfusion from DCE-MRI data has been established in other organs and explored in the lungs as well (11,12).For perfusion quantification, both the calculation of perfusion metrics like pulmonary blood flow (PBF), pulmonary blood volume (PBV) and mean transit time (MTT) as well as the determination of the extent of defects have been shown (13).While these have been examined mainly as global metrics for the entire lung, especially considering the fraction of defects as perfusion defects in percent (QDP), they are also available locally for each image voxel (14).
T 1 quantification in the lungs was originally pursued in the context of oxygen-enhanced imaging (15).In these studies, significantly lower baseline lung T 1 was found in patients with COPD even without oxygen supplementation, and median T 1 was found to correlate with the visual MRI perfusion score (8).In subsequent work, it was further demonstrated that observed lung T 1 [T 1 (TE)] is dependent on the echo time (TE) of the measurement.It was subsequently assumed that T 1 (TE) reflects different compartments of lung tissue protons inside each imaging voxel (protons in blood and extravascular protons), since the inter-patient correlations between T 1 (TE) and the MRI morphology and MRI perfusion score as well as other metrics vary with TE in COPD and cystic fibrosis (16)(17)(18).It was hypothesized that T 1 (TE) has the potential to differentiate tissue abnormalities such as inflammation or emphysema from perfusion abnormalities on the level of image voxels without the need for contrast material.Previous studies have so far examined the relationship between T 1 (TE) and DCE-MRI perfusion using only visual MRI perfusion scores on the level of the entire lungs for median T 1 , and on the level of lobes for visually scored T 1 (8,17,18).The present work was conducted to further investigate the relationship of T 1 (TE) with quantitative perfusion metrics on a local level including voxel-wise correlations.1.Some patients were included in separate previous reports, which did not assess the analyses made in the present study (13,18).
T 1 (TE) mapping was based on an inversion recovery multi-echo 2D ultra-short TE sequence, applied directly before contrast injection (20).Echos at TE 1-5 = 70, 500, 1,200, 1,650, and 2,300 μs were acquired split into two separate measurements with TE 1 , TE 3 , and TE 5 in the first and TE 2 and TE 4 in the second measurement.Imaging parameters were: One or two coronal slices at 15 mm thickness, matrix size 128 × 128, FoV 500 × 500 mm 2 , TR = 5 ms, flip angle 6°, and total acquisition time 4 min (1 slice) or 8 min (2 slices).To cover as much lung parenchyma as possible, the first slice was placed through the descending aorta and the second slice 30 mm ventral.For each measurement, a total of 6,000 radial spokes in a golden angle distribution were acquired, divided into blocks starting with an adiabatic inversion pulse followed by 300 radial spokes and separated by 3 s delays.These were then sorted by inversion time (TI), split using a sliding window 120 spokes wide with 60 spokes step-width and reconstructed using a non-uniform Fourier transform implemented in MATLAB (Mathworks, Natick, United States).To achieve ultra-short echo times, pairs of half sincpulses with opposing slice-selection gradients, 1,250 μs pulse duration and time-bandwidth-product 2 were employed.For each TE, T 1 maps were calculated per voxel from the time-course along TI, based on the UTE images in each slice (21).UTE measurements were acquired during free breathing and gated to expiration using 50% of data, as described in Triphan et al. (20).

Magnetic resonance imaging analysis
Magnetic resonance images were segmented independently for UTE and DCE measurements: In UTE images, segmentation was performed on images at TE 3 using simple region growing (for details, see the online supplement).Since the DCE images are threedimensional and designed to maximize contrast between tissue with and without contrast agent, these were not segmented directly.Instead, a region growing-based segmentation algorithm was applied to morphological images with a higher resolution and the resulting segmentation masks were then registered to the DCE images, as described previously (13,22).
For perfusion quantification, an arterial input function was detected automatically and then used to calculate a residual function R(t) by applying a voxel-wise deconvolution with the time-dependent signal in the DCE images (23).Pulmonary blood flow (PBF) maps were derived from R(t) by determining the maximum along the timecourse in each individual voxel (24).To calculate defect classification maps and QDP, the following steps were performed: The timepoint of maximum contrast enhancement in the entire lungs (t max ) was determined by taking the mean of R(t) in the entire lung for every timepoint.Otsu's method was then applied to the values of R(t max ) within the lungs to determine two thresholds (25).Defect classification maps were produced by classifying all voxels below the lower threshold as perfusion defects and the remainder as normal (13).The fraction of defect voxels within the segmented lungs on these maps relative to the total number of segmented voxels was defined as QDP.For all of the following steps, the three partitions of the 3D DCE images that correspond to the location of the UTE slice(s) were determined and averaged to produce a thicker projection image.Note that t max and the classification thresholds were calculated based on the entire segmented lungs but QDP to be compared to T 1 (TE) only on these slices corresponding to UTE measurements.
To facilitate local (intra-patient) comparison of T 1 (TE) and quantitative perfusion maps, two approaches were investigated: (1) DCE images were averaged over their entire time-course and registered onto the corresponding UTE images using Advanced Normalization Tools (ANTs) (26) using the same parameters for all patients.The deformation fields produced by this registration were then applied to the PBF and defect classification maps as well as the lung segmentations derived from morphological imaging to allow for voxel-wise comparisons.An example of these deformation fields is given in Figures 1A,B.(2) On each slice, the left and right lungs were individually subdivided into areas with the following approach: Each lung was subdivided into 10 strips of equal volume vertically and then each of these strips was separated into 10 blocks of equal volume horizontally.By applying this subdivision algorithm on the segmentations of the DCE images and the UTE independently, this approach yields subdivided areas for both different resolutions and different respiratory states, which nevertheless correspond to the same physical volumes in the lungs and thus allow for correlations and comparisons without a registration step.An example of this subdivision is shown in Figures 1C,D.

Statistical analyses
Inter-patient correlations were calculated for median T 1 (TE) with median PBF (ρ IPBF ) and QDP (ρ IQDP ) within the segmented slices.For each patient, Spearman's correlation coefficients between T 1 (TE) and PBF were calculated separately for registered voxels (ρ regPBF ) and subdivided areas (ρ subPBF ), respectively.Registered voxels in UTE were classified as normal perfusion or perfusion defect according to the corresponding classification map from DCE, and the median T 1 (TE) in each fraction was compared by Wilcoxon rank-sum tests.For subdivided areas, the fraction of defect voxels (i.e., local QDP) in each subdivided area was correlated with median T 1 (TE) in the corresponding area, analogous to PBF, as the local coefficient ρ subQDP .A value of p <0.05 was considered statistically significant.

Results
Suitable UTE and DCE measurements were available from 22 patients.In four patients, one of the two measurements comprising the UTE acquisition was mismatched in one slice or insufficient time was available to acquire both slices.Figure 2 shows an example set of base images and parameter maps in one patient.
The coefficients of the inter-patient correlation of T 1 (TE) with median PBF for all corresponding lung slices were ρ IPBF1-5 = 0.56, 0.64, 0.59, 0.60, and 0.29 at TE 1-5 and the respective correlations of T 1 (TE) with QDP were ρ IQDP1-5 = −0.56,−0.76, −0.63, −0.53, and −0.26.Of these, only the correlations at TE 1-4 were significant (at p < 0.02).Table 2 shows the mean local correlation coefficients of T 1 (TE) with PBF based on both methods for associating UTE with DCE images, as well as the correlation of T 1 (TE) with QDP.The values of ρ regPBF and ρ subPBF for individual patients are also shown in Figure 3 to illustrate the inter-patient variation.
At TE 1 = 70 μs, in 18 out of 22 patients, T 1 was shorter in voxels classified as perfusion defect when compared to the rest of the lung voxels classified as normal perfusion in the respective individual.In three patients, no significant difference was found and in one patient, T 1 was longer in the defect area.While, as shown in Figure 4A, T 1 in normal (non-defect) areas and ΔT 1 , the relative difference, varied strongly among patients, median T 1 in defect areas was shorter at all TEs when considered over all patients.Notably, as shown in Figure 4B, mean ΔT 1 was smaller at TE 2 than at TE 1 and TE 3 , with ΔT 1 = 6.2, 4.0, 6.2, 9.1, and 16.6%, similar to the local correlations observed.
As demonstrated in Figure 2, visual inspection of parameter maps suggested that stronger local correlations appeared in patients where areas with clearly recognizable perfusion and longer T 1 remained.To investigate this observation, the inter-patient correlation between median T 1 (TE 1 ) and the local correlation coefficients of T 1 and PBF was calculated.At TE 1 = 70 μs, this was ρ regcorr = 0.68 when using registration and ρ subcorr = 0.32 when using subdivided areas.

Discussion
In this work, inter-patient correlations of T 1 (TE) with quantitative perfusion measures showed the strongest correlation coefficient at TE 2 = 500 μs.Local voxel-wise correlations instead showed the  Mean local correlations of T1(TE) with PBF, based on image registration for voxel-wise correlation (ρregPBF) and based on subdivisions (ρsubPBF), as well as with QDP based on subdivisions (ρsubQDP).Values are given as means with standard deviations and the number of patients in whom the correlation was significant in brackets.The visual score for perfusion MRI has been established in COPD and cystic fibrosis, and were shown to correlate with clinical metrics (6, 10) and with T 1 at conventional echo times (8).Subsequently, it was demonstrated that the inter-patient correlation of median T 1 (TE) in the entire lungs with this perfusion score and quantitative CT parameters depends on TE.Specifically, the strongest correlation was found at TE = 500 μs (18).This agrees with inter-patient correlations of T 1 (TE) with quantitative perfusion measures shown in this work which, also gave the strongest correlation at TE 2 .
Perfusion defects in percent was developed to compensate for the vulnerabilities in the quantification of PBF related to non-linearity of Gadolinium-concentration with MRI signal, artifacts, limited temporal resolution, and low SNR in the lung parenchyma (which is particularly low in diseased areas of the lung parenchyma) (14,27,28).This is supported by stronger inter-patient correlations between QDP and the visual perfusion score than were found with visual perfusion scores previously (13).In our present work, the inter-patient correlation between T 1 (TE) and QDP was also stronger at TE 2 and stronger than with PBF.However, while the correlations of T 1 with PBF based on subdivision were slightly stronger than those based on image registration, only very weak correlations of T 1 with local QDP within subdivisions were found.This is most likely to due to many subdivided areas containing only defect or only normal voxels and thus not describing larger defect areas well.QDP derived from DCE using a different algorithm was found to locally match QDP derived from contrast-free PREFUL MRI (28).QDP derived from T 1 (TE) (rather than DCE-MRI) was shown to correlate similarly to T 1 (TE) itself in patients with CF (17), but not in patients in COPD (18).Thus, we did not re-investigate defect classification based on T 1 (TE) in the present work.
Unlike the inter-patient correlations, the local correlations found in the present study were highest at TE 1 and higher at TE 3 than at TE 2 .This can be attributed to lung areas more affected by disease having not only shorter T 1 but also exhibiting a different dependence of T 1 on TE.This is supported by the behavior of T 1 in areas classified as perfusion defect or normal, when averaged over the patient collective: There, the average difference between these areas decreased from TE 1 to TE 2 and then increased again at TE 3 .This behavior becomes clearer when considering the data published previously comparing patients with COPD to healthy volunteers (18).There, the TE-dependence of T 1 was much more gradual and observable at longer TEs in healthy volunteers, while the observed T 1 in patients with COPD did not notably increase further after TE = 500 μs, which also meant that the separation between these groups was smallest at TE = 500 μs.Assuming that the TE-dependence of T 1 in areas classified as normal through perfusion is closer to that in healthy volunteers and in defect areas closer to that in the patient group, this is consistent with the results seen in this work.It should however be noted that in the present collective areas classified as normal are still quite abnormal in T 1 compared to values observed in healthy volunteers.Thus, our results show that both base T 1 and its TE-dependence differ in areas of normal and defective perfusion.This further supports the hypothesis that it may be possible to separate the effects of local tissue changes and perfusion on T 1 and investigate both with a single measurement.However, this is not yet possible with the current approach.
A large variation of local correlations was observed, with the strongest observed local correlations reaching the same magnitude as the inter-patient correlations.The overall visual impression of T 1 and PBF parameter maps was that strong correlations were mainly present in patients with some volume of remaining functional lung tissue.In those cases, a notable spread of perfusion values exists that can be correlated in the first place.As such, we assumed that low local correlations might indicate relatively uniformly severely damaged lungs and thus also short T 1 .In order to confirm this assumption, we calculated the inter-patient correlation of median T 1 (TE 1 ) with local correlation coefficients in each patient.That this correlation was significant supports the assertion that weak local correlations were primarily found in patients with more damaged lungs.However, this correlation was much weaker when based on subdivided areas.Since regions with long T 1 are unlikely to coincide with the borders of subdivided areas, they can be assumed to describe these less well than the direct voxel-wise correlations.Similarly, in cases where a perfusion deficit was present all over the lung with corresponding short T 1 , these borders may matter less.At the same time, subdivided areas were less vulnerable to outlier values produced by quantification at very low SNR due to reduced proton density in severe emphysema.In further research, it may be worthwhile to investigate this in milder or presymptomatic cases of COPD, where a larger spread of values to correlate could be investigated.Additionally, MRI in general struggles in patients with severe emphysema simply due to lack of proton signal, though we might surmise that T 1 mapping at TE 1 = 70 μs should be better able to cope.However, in this work we were limited by the possibilities provided by DCE imaging.Since in COPD patients, the TE-dependence of T 1 is shifted to shorter TEs in comparison to healthy subjects, a tight spacing of TEs was necessary to depict the effects of interest in this work (16,18): The TE-dependence of local and inter-patient correlations discussed here was visible at TE 1 = 70 μs and TE 2 = 500 μs, but it is not possible to measure T 1 at both TEs in one TR due to the time required for the read dephasing gradient in between.Accordingly, the acquisition had to be split into two sets of echoes, which were implemented in separate measurements.This had the additional problem that the dc-signal used for respiratory gating is not necessarily aligned between those two measurements in case the patient changes how deep they breathe between them.Additionally, the two measurements could be misaligned due to operator error, making the entire slice unusable.After this study was concluded, we redesigned the sequence such that different sets of TEs could be acquired in alternating successive TRs which has neither of these disadvantages while still providing the required tight spacing of TEs.
The 4D MRI used for perfusion quantification was a single measurement, which meant that if the contrast injection was administered too early or late or the patient holding their breath for unsufficient time, it was entirely unusable for quantification.Since the UTE measurement was not part of the primary study protocol, we could only plan two slices and did not have time to retry in case of mistakes.However, these were separately acquired and each split into two measurements, as stated above.In cases where one slices was unusable due to operator error in at least one such sub-measurement, we decided to nevertheless use the remaining slice since the total number of patient measurements was so small.Thus, the UTE images being based on 2D slices was a major limitation of this work: a reliable 3D implementation of the multi-echo inversion recovery UTE would allow for a better quality of registration and give better coverage of the lungs, which would also provide improved comparability with perfusion images.
We investigated the approach to derive local correlations from subdivided areas in addition to image registration to avoid reliance on the registration algorithm, as it is not practically possible to guarantee a successful registration: The quality of image registration limits the comparability and thus voxel-wise correlation of parameter maps.The approach using subdivided areas was designed as an alternative that avoids this problem by increasing the size of spatial units to potentially correct for some imprecision in registration but only with the limitation that it is based on the assumption that both lungs have equal volume.This was implemented in order to be able to use the same number and pattern of subdivisions in each patient.In future work, the accuracy of local correlations could be increased by generating a specific subdivision pattern for each patient, possibly in 3D, which employs different number of subdivisions for each lung.Furthermore, the two segmentations may be subtly different beyond the issue of comparing two-with three-dimensional data.Finally, a relevant limitation for both approaches was that DCE images were acquired in inspiratory breath-holds, while UTE images were acquired during free breathing and reconstructed to expiration retrospectively.
As noted above, the approach discussed here relies entirely on using local perfusion as an indicator of lung function.However, even in early COPD stages, ventilation/perfusion (V/Q) mismatches have been reported, suggesting a blocked hypoxic pulmonary vasoconstriction mechanism to inhibit inflammation (29, 30).Nevertheless, there remains a significant association between ventilation and perfusion in COPD patients and correlation of the extent of V/Q mismatches with disease severity was relatively modest (31).To incorporate the effect of local V/Q mismatches, an additional direct measure of ventilation could be combined with perfusion imaging techniques.While non-contrast enhanced lung perfusion imaging methods have been introduced that also depict ventilation, it should be noted that the 2D UTE sequence used in this work was initially developed for oxygen-enhanced imaging since both T 1 and T 2 * (which was not considered in this work but can be quantified from the same measurement) depend on the presence of molecular oxygen (dissolved in tissue and as gas in the alveoli, respectively) (4,20).An additional acquisition while supplying subjects with oxygen could thus be used to determine local ventilation directly and thus investigate V/Q mismatches.
In conclusion, we have shown inter-patient correlations of T 1 (TE) with quantitative perfusion measures that are comparable to those described previously for T 1 (TE) with semi-quantitative visual perfusion scores from perfusion MRI and quantitative CT parameters.Local (intra-patient) correlations were much weaker on average, but the strongest correlations found reached a similar scale as the interpatient correlation coefficients.Importantly, the TE-dependency of the local correlation coefficients implies that not only base T 1 but also its TE-dependency in normal areas is closer to the behavior previously found in healthy volunteers than in defect areas.That the correlation coefficients themselves correlate with overall average T 1 implies that weak local correlations would be observed in patients with few remaining areas of well-functioning lung tissue and thus low contrasts between local observations.Accordingly, it may be interesting to investigate DCE perfusion and T 1 (TE) in a patient cohort with less severe, earlier disease, both individually and in combination.

FIGURE 1
FIGURE 1 Example horizontal (A) and vertical (B) deformation field as determined by image registration, scaled in number of voxels.Segmentations with subdivided areas for a DCE image (C) and the corresponding 2D UTE image (D).Areas that correspond to each other in the subdivision are drawn in identical colors.

FIGURE 3 Spearman
FIGURE 3Spearman's correlation coefficients for the correlation of T 1 (TE) with PBF, (A) voxel-wise correlation based on image registration and (B) based on subdivisions, respectively.Each patient is shown with the same color in both graphs.

FIGURE 4 (
FIGURE 4 (A) Relative difference of T 1 at TE 1 = 70 μs between voxels classified as perfusion defect and voxels classified as normal.Each point represents one individual.(B) Mean of individual median T 1 over all patients at each TE, according to defect classification.

TABLE 1
Patient characteristics.

TABLE 2
Mean local correlations.