Comparison of quantitative multiple-breath specific ventilation imaging using colocalized 2D oxygen-enhanced MRI and hyperpolarized ³He MRI

Abstract

Two magnetic resonance specific ventilation imaging (SVI) techniques, namely, oxygen-enhanced proton (OE-¹H) and hyperpolarized ³He (HP-³He), were compared in eight healthy supine subjects [age 32 (6) yr]. An in-house radio frequency coil array for ¹H configured with the ³He transmit-receive coil in situ enabled acquisition of SVI data from two nuclei from the same slice without repositioning the subjects. After 3 × 3 voxel downsampling to account for spatial registration errors between the two SV images, the voxel-by-voxel correlation coefficient of two SV maps ranged from 0.11 to 0.63 [0.46 mean (0.17 SD); P < 0.05]. Several indexes were analyzed and compared from the tidal volume-matched SV maps: the mean of SV log-normal distribution (SVmean), the standard deviation of the distribution as a measure of SV heterogeneity (SVwidth), and the gravitational gradient (SVslope). There were no significant differences in SVmean [OE-¹H: 0.28 (0.08) and HP-³He: 0.32 (0.14)], SVwidths [OE-¹H: 0.28 (0.08) and HP-³He: 0.27 (0.10)], and SVslopes [OE-¹H: −0.016 (0.006) cm⁻¹ and HP-³He: −0.013 (0.007) cm⁻¹]. Despite the statistical similarities of the population averages, Bland-Altman analysis demonstrated large individual intertechnique variability. SDs of differences in these indexes were 42% (SVmean), 46% (SVwidths), and 62% (SVslopes) of their corresponding overall mean values. The present study showed that two independent, spatially coregistered, SVI techniques presented a moderate positive voxel-by-voxel correlation. Population averages of SVmean, SVwidth, and SVslope were in close agreement. However, the lack of agreement when the data sets were analyzed individually might indicate some fundamental mechanistic differences between the techniques.

NEW & NOTEWORTHY To the best of our knowledge, this is the first cross-comparison of two different specific ventilation (SV) MRI techniques in the human lung (i.e., oxygen-enhanced proton and hyperpolarized ³He). The present study showed that two types of spatially coregistered SV images presented a modest positive correlation. The two techniques also yielded similar population averages of SV indexes such as log-normal mean, SV heterogeneity, and the gravitational slope, albeit with some intersubject variability.

INTRODUCTION

Specific ventilation (SV) is the ratio of the amount of fresh gas reaching a lung unit to the preinspiratory volume of that unit [essentially the tidal volume (Vt)-to-functional residual capacity (FRC) ratio]. In recent years, two specific ventilation imaging (SVI) techniques based on a multiple-breath magnetic resonance imaging (MRI) approach have been introduced: 1) oxygen-enhanced proton (¹H)-based washin-washout (OE-¹H) (33) and 2) hyperpolarized ³He gas washout (HP-³He) (9, 18). These two imaging techniques enable the quantitative mapping of the spatial distribution of SV in the human lung.

Conventionally, the multibreath washout (MBW) technique has been used to estimate SV and its overall heterogeneity within the lung (23), which is an indicator of gas exchange efficiency as well as small airway obstruction (25, 26). The technique consists of measuring the concentration change of an expired tracer gas [e.g., nitrogen, helium, or sulfur hexafluoride (31)] during breathing following a change in inspired gas concentration (i.e., gas washout or washin due to ventilation). However, this technique lacks the ability to yield spatial information. Similar to the conventional MBW technique, both MRI-based SVI techniques are sensitive to the rate of change in a tracer gas concentration due to washin or/and washout, which subsequently alters the MRI signal measured over the multiple breaths.

In OE-¹H MRI, room air and oxygen [fraction of inspired oxygen ($F{I}_{O_2}$) = 1.00] are presented to a human subject alternately every 20 breaths. Inhaled oxygen shortens the longitudinal relaxation time (T1) of lung tissue and blood, resulting in an increase in the MRI signal when using appropriate MR timing parameters (4, 11). The changes in the MRI signal from water molecules in the lung tissue as well as the blood pool due to the series of O₂ washin and washout cycles are thus assumed to reflect the change in the local alveolar oxygen partial pressure ($P{A}_{O_2}$), since the rate of change of $P{A}_{O_2}$ is directly determined by the local SV. The spatial distribution of SV is subsequently computed from the rate of change in the local MRI signal, which is driven by the oxygen washin and washout, and hence the method provides an indirect method of imaging lung ventilation.

With HP-³He, after inhalation of a single dose of hyperpolarized ³He gas mixture, the local SV is determined from the rate of MRI signal decay resulting from the MBW of the gas, once other effects affecting the overall MRI signal decay have been accounted for. In this technique, the signal from the ³He atoms in the lungs is directly monitored, and as such the method is a direct measure of lung ventilation. The scanning procedure and postprocessing allow us to estimate the regional distribution of hyperpolarized ³He gas within the lung and its breath-by-breath reduction in local concentration (18).

The aim of this work was to compare these two SVI techniques with spatially matched data. The two SVI techniques have independently shown physiologically reasonable and reproducible SV distributions, and OE-¹H has been compared against regionally nonspecific methods in terms of quantifying ventilation heterogeneity (32). However, to the best of our knowledge, the techniques have not yet been compared against one another, mainly because of the complexity of experimental setup of both techniques. We hypothesized that the distribution of SV measured with these two techniques would be similar. Use of an in-house radio frequency (RF) torso ¹H array coil configured with the ³He transmit-receive coil in situ (30) enabled the separate acquisition of ¹H and hyperpolarized ³He MR images during a single scanning session. This allowed a direct comparison between the two SVI techniques obtained from a single two-dimensional (2D) image plane of the same human subject without repositioning to mitigate spatial misregistration.

In this study of eight healthy human subjects, correlation between two spatially matched SV images, SV heterogeneity [an indicator of gas exchange efficiency measured as the width of the SV distribution in logarithmic scale (23, 32)], and the vertical gravitational gradient of SV (3, 21, 28) in the supine lung were compared.

METHODS

Human subjects.

Eight healthy human subjects [2 women and 6 men; age = 32 (6) yr old, height = 178 (10) cm, and weight = 82 (12) kg] were imaged, after National Research Ethics Committee and Medicines and Healthcare Products Regulatory Agency regulatory approval, with written informed consent from all subjects. Table 1 shows subject characteristics (sex, age, height, and weight) and spirometry-based pulmonary function data [forced expiratory volume in 1 s (FEV₁), forced vital capacity (FVC), and FEV₁/FVC]. Spirometry data were obtained on a different day.

Table 1.

Subject characteristics, pulmonary function data, average Vt during each MRI session, scanning order, and total scanning time

Subject	Sex	Age, yr	Height, cm	Weight, kg	FEV1, liters (%predicted)	FVC, liters (%predicted)	FEV1/FVC* (%predicted)	Vt, liters (OE-1H)	Vt, liters (HP-3He)	Scanning Order	Total Scan Time, min
1	M	32	184	75.4	3.65 (76.3)	5.10 (87.0)	0.72 (87.1)	0.83	0.98	3He → 1H	46
2	F	32	162	61.0	3.46 (109.8)	4.54 (120.9)	0.76 (90.3)	0.52	0.55	3He → 1H	44
3	M	31	185	94.4	3.65 (79.9)	5.07 (91.81)	0.72 (86.6)	0.81	0.97	1H → 3He	53
4	M	28	182	83.8	4.87 (101.8)	5.88 (101.8)	0.83 (99.8)	0.57	0.73	3He → 1H	27
5	M	30	184	79.4	4.54 (94.7)	5.45 (93.2)	0.83 (101.0)	0.45	0.46	1H → 3He	32
6	M	43	171	103.0	3.55 (92.9)	4.37 (92.2)	0.81 (100.4)	0.74	0.59	1H → 3He	29
7	F	38	165	79.6	3.70 (102.7)	4.57 (104.1)	0.81 (98.5)	0.57	0.89	3He → 1H	37
8	M	24	189	77.3	5.29 (99.3)	6.56 (102.1)	0.81 (96.3)	1.11	1.06	1H → 3He	39
Avg.		32	178	81.7	4.09 (94.7)	5.19 (99.1)	0.79 (95.0)	0.70	0.78		38
SD		6	10	12.6	0.70 (11.5)	0.75 (10.7)	0.05 (6.1)	0.22	0.23		9

Values are subject characteristics, pulmonary function data, average tidal volume (Vt) during each MRI session, scanning order, and total scanning time for all 8 subjects studied. Scanning order: ³He → ¹H: hyperpolarized ³He (HP-³He) experiment was before oxygen-enhanced proton (OE-¹H) experiment; ¹H → ³He: OE-¹H experiment was before HP-³He experiment. Total scan time, total scanning time for each subject from beginning of 1st specific ventilation imaging (SVI) experiment to end of 2nd SVI experiment. FEV₁, forced expiratory volume in 1 s; FVC, forced vital capacity.

^*See Ref. 29.

MRI data collection.

Both OE-¹H and HP-³He MRI data were acquired on a 1.5-T Signa HDx MRI system (General Electric Medical Systems, Milwaukee, WI).

The subject wore two RF torso coils and a quadrature wrap-around transmit-receive chest RF coil tuned to ³He (Clinical MR Solutions, Milwaukee, WI) that was mounted over an in-house four-loop receiver array for ¹H (30).

The sagittal slice within the right lung presenting the largest anterior-posterior dimension (the approximately midclavicular line) was selected as a common imaging plane. The slice thickness was 1.0 cm, field of view was 38 cm × 38 cm, and an imaging matrix was 64 × 64, resulting in a corresponding voxel size of ~0.6 cm × 0.6 cm × 1.0 cm.

Multibreath specific ventilation imaging techniques.

SV is the ratio of the volume of fresh gas (ΔV) entering a region of lung to the end-expiratory volume (V₀) of that region (i.e., SV = ΔV/V₀) and is thus a dimensionless quantity (33). SVI techniques using OE-¹H and HP-³He performed in this study were described in detail in Refs. 33 and 18, respectively. As mentioned in introduction, both SV imaging techniques utilize the basic principle of MBW, where the local SV is computed from the breath-by-breath changes in MRI signal intensity over subsequent images.

Oxygen-enhanced proton imaging.

The subject lay in the supine posture in the MRI scanner wearing a nose clip and a silicone rubber mouthpiece (Hans Rudolph, silicone mouthpiece 602073) attached to a nonrebreathing valve (Hans Rudolph, T-Shape 1410 with a dead-space volume of 10 ml) to deliver a tracer gas for washin/washout experiments inside the MR scanner bore. The total volume of the inspiratory port is 200 ml, which accounts in part for the volume delay for the washin of 100% inhalation O₂. The inspiratory port of the valve was connected to a manually controlled three-way valve (Hans Rudolph, T-Shape 2100) to allow switching of the inspired gas between room air (used for both washout experiments) and 100% O₂ (used only for oxygen washin during the oxygen-enhanced imaging), which was contained in a 170-liter Mylar bag (Hans Rudolph, Non-Diffusing Gas Collection Bag 112449M). One operator was in the MR scanner room at all times and manually controlled the valve. The resistance of the two inspiratory paths (i.e., to room air and to O₂ bag) was matched by employing two identical tubes with same length and inner diameters to avoid changes in FRC resulting from the change in gas source. A pneumotachometer (Hans Rudolph, RSS 100HR Research with a series 3830/4830 linear screen sensor) was placed between the inspiratory port and switching valve so that the subject’s inspiratory Vt as well as the flow rate were monitored during the imaging session. The pneumotachometer was calibrated for each subject with a 3-liter syringe (Hans Rudolph) over a range encompassing the expected inspiratory flow rates for a supine resting subject. The overall system dead space including the volume of inspiratory port and anatomical dead space was estimated to be 350 ml.

OE-¹H MRI was used to quantify regional SV in a lung slice. Dissolved oxygen shortens the T1 of tissue and blood in the lungs, which results in the increase of the local MR signal intensity in T1-weighted proton MR images (11). The rate of turnover in the local alveolar O₂ concentration following a sudden change in $F{I}_{O_2}$ is determined by the local SV (33): lung regions with a higher SV reach a new equilibrium faster than those with lower SV. Therefore, the regional SV is quantified on a voxel-by-voxel basis by determining the rate of change in the MRI signal intensity on T1-weighted images during the oxygen washin ($F{I}_{O_2}$ = 1.00) and washout ($F{I}_{O_2}$ = 0.21).

Single-shot fast spin echo images [half-Fourier acquisition single-shot turbo spin echo (HASTE), effective echo time (TE) = 23 ms, inter-TE = 4.5 ms, inspiratory time (TI) = 1,100 ms] following a global inversion were acquired every 5 s. The inhaled gas was toggled every 20 images between air and 100% oxygen. The air-oxygen cycle was repeated five times, and 20 additional 100% oxygen breaths were added at the end of the fifth cycle.

Each MR image was acquired during a short (i.e., ~1 s) postexpiratory pause at FRC. Subjects were asked to take a normal tidal breath after each image acquisition and return to FRC at a comfortable flow rate in time for the subsequent image to be acquired. Subjects were trained before entering the scanner to voluntarily pace their breathing with the acquisition sound. In addition, the operator in the MR scanner room also coached the subject on the breathing maneuver during the imaging session. The total scanning time for the OE protocol was ~18 min.

Each series of 220 consecutive OE-¹H images was imported into MATLAB (MathWorks, Natick, MA). One representative FRC lung image was chosen for each subject as the reference proton MR image. Image registration based on projective deformation was performed to match the lung field in the reference image to that in subsequent images (2). When the lung area increased or decreased by >10% compared with the reference image, the image was excluded from the SV calculations. On average, there were 11 rejected images per subject (5% of acquired images; only 2 subjects had >10% of images rejected). The local SV was then estimated based on the MRI signal intensity change over 220 breaths as described in detail in Ref. 32. The estimated volume delay is the overall system dead space (i.e., 350 ml).

Hyperpolarized gas ³He washout imaging.

Multibreath hyperpolarized gas ³He washout imaging was performed with a single dose of hyperpolarized ³He gas (~25% polarization) as described in detail in Ref. 18. The gas was prepared with a spin-exchange optical pumping apparatus (GE Healthcare, Amersham, UK) under a site-specific regulatory license. Since ³He gas has a very low solubility in blood, the local rate of tracer gas turnover due to washout is a direct measurement of local SV.

Before the imaging session, each subject’s expiratory reserve volume [ERV: FRC minus residual volume (RV)] was measured in the supine posture with the pneumotachometer. The ERV equivalent volume of hyperpolarized ³He-N₂ gas mixture (10% hyperpolarized ³He and N₂ balance) was given to each subject. The subject inhaled the gas mixture from RV from a 1-liter Tedlar bag (Fisher Scientific). Thus the subject was at a lung volume close to FRC before the gas washout and image acquisition began.

Image acquisition was repeated every 4 s with a single-slice selective spoiled gradient echo sequence (TE = 1.0 ms, repetition time = 3.4 ms, slice thickness 10 mm, and flip angle = 4.0°). The first two images were acquired during the initial breath hold immediately following the inhalation of the gas mixture. These two images were used to calibrate for contributions to the signal decay that are independent of gas washout (17). It is worth noting that the 1.0-cm 2D slice was chosen for direct comparison with the OE ¹H acquisition. With such a 2D experiment the effect of slice mixing of ³He polarization (13, 38) may contribute to the observed signal decay between breaths, which is mitigated in a 3D or 2D whole lung projection scan (18). These effects were simulated for the long-range diffusion coefficients (D) observed previously in Refs. 13 and 38 for ³He in vivo in healthy lungs (D = 0.033 cm²/s), and we found the slice mixing to be negligible for the healthy lungs imaged here. If, however, we were to use the same 2D slice-selective MBW imaging sequence in a more diffusive regime with ³He in free air (D ~ 0.9 cm²/s), these effects would require consideration.

After the calibration scan, the washout phase began (i.e., the second calibration image corresponds to the start of washout experiment). Similar to the OE-¹H scan, subjects were asked to take a normal tidal breath after each image acquisition synchronous to a noise cue generated by the MRI scanner. All images were acquired during a short breath hold at FRC. The operator in the scanner room coached the subject on the breathing maneuver throughout the scan. The image acquisition was repeated until the MR signal was attenuated to low signal-to-noise ratio (SNR < 15) levels. The flip angle was chosen to ensure that at least four washout images were acquired with the MRI signal intensities above the SNR level threshold.

All image data were registered to the first calibration image. The local SV was calculated from the rate of hyperpolarized ³He gas washout estimated from the MRI signal decay after the T1 decay and RF depolarization corrections. The method is as described in full in Ref. 18.

Image registration of OE-¹H MRI and HP-³He MRI.

The two FRC target volumes in OE-¹H and HP-³He images were coregistered to minimize misalignment. Proton MRI image contrast is such that the air spaces in the lung are dark, whereas air spaces ventilated with ³He show a bright signal; this reciprocal signal renders the use of internal reference marks challenging. In addition, variations in FRC as well as subject movement (i.e., the fact that 2 scans were not exactly simultaneous) may also cause the misalignment. Image registration based on projective deformation was therefore performed to match the two lung outlines (2). A region of interest (ROI) comprising the lung was manually selected so that two SV data were obtained from a common ROI.

Log-normal nature of SV distribution and tidal volume correction.

The SV distribution histogram in a healthy human is generally a single-mode log-normal distribution, a symmetric Gaussian distribution on logarithmic scale described by two indexes, namely, SVmean and SVwidth corresponding to the arithmetic mean and SD of ln(SV) distribution (37).

$$\text{SVmean}=\mathrm{mean}\left[\ln \left(\text{SV}\right)\right]$$ (1)

$$\text{SVwidth}=\text{std}\left[\ln \left(\text{SV}\right)\right]$$ (2)

When two SV distributions (i.e., SV1 and SV2) were obtained from two different measurements (i.e., time, technique, etc.), the following equations determine the relationships between two SVmeans (i.e., SVmean1 and SVmean2) as well as two SVwidths (i.e., SVwidth1 and SVwidth2):

$$\text{SVmean2}={\text{A}}_{\text{1to2}}{\text{SVmean1}}^{{\text{m}}_{\text{1to2}}}$$ (3)

$$\text{SVwidth2}/\text{SVwidth1}={\text{m}}_{\text{1to2}}$$ (4)

in which A_1to2 and m_1to2 serve to transform a log-normal distribution of SV1 to that of SV2.

Assuming that Vt is distributed into local lung compartments in proportion to their local SVs, the change in Vt between SV1 and SV2 appears as a lateral shift (change in SVmean without changing SVwidth) in the logarithmic scale. This assumes that SVwidth (the metric of ventilation heterogeneity) is time invariant among supine normal subjects during tidal breathing (6–8, 16). In this particular case, A_1to2 corresponds to the ratio between two Vt values, whereas m_1to2 equals 1.

In the present study, SV data of HP-³He were corrected to the measured Vt of OE-¹H. Therefore, two SVmeans were compared at a common Vt. To do this, A_1to2 and m_1to2 were set to the measured ratio of Vt values (i.e., Vt_OE-1H/Vt_HP-3He) and 1, resulting in a lateral shift of the HP-³He distribution in the logarithmic scale as described in the previous paragraph.

Comparison between OE-¹H and HP-³He SVI.

SVmean and SVwidth were determined as an arithmetic mean and SD of ln(SV), respectively. The gravitational dependence of SV (the linear relationship between the vertical height of the lung from the most dependent portion and SV) was computed from each SVI data set by calculating the SV slope by linear regression. To directly compare the two SV maps, a Deming regression was performed (least-squares fitting with perpendicular offset) with SV scaled logarithmically. Since there is no a priori “correct” SV map, perpendicular regression was used so as to not bias the fit in one direction or the other.

The original MRI voxel size is 0.6 cm × 0.6 cm × 1.0 cm. A 3 × 3 voxel downsampling was performed; the rationale behind downsampling was that registration of two types of SV images with very different signal characteristics will produce minor spatial alignment errors, which will be spatially filtered at the 1.8 cm × 1.8 cm × 1.0 cm scale.

All data are presented as means (SD). When SV data for the eight subjects were averaged, the SD corresponds to the intersubject variability. Pearson’s correlation was used to determine the relationship between two SV measurements (statistical significance: P < 0.05). A Bland-Altman comparison was performed to determine the average bias and SD of differences between the two types of data.

RESULTS

Subject characteristics (sex, age, height, weight, and pulmonary function data) as well as average Vt values measured during each MRI session are summarized in Table 1. Vt values during OE-¹H and HP-³He scans were 0.70 (0.22) l/breath and 0.78 (0.23) l/breath, respectively. The difference was not statistically significant (P = 0.19, paired t-test, 2 tailed). There was a positive correlation between the two Vt measurements (R = 0.77, P = 0.03). Table 1 also shows the total scan time for each subject as well as the order of scans. The total scan time was 38 (9) min. Four subjects underwent OE-¹H first, and the remaining four underwent HP-³He first.

Typical SVI data are presented in Fig. 1. Figure 1A shows a right lung anatomical image in the sagittal plane obtained from subject 1 in the supine position. It also serves as the subject’s OE-¹H reference image. The solid white line delineates a common ROI encompassing the lung field. As illustrated, the ROI does not encompass the edge of the lung, to avoid potential errors associated with edge effects such as a partial volume effect. OE-¹H and HP-³He SV maps within the ROI are shown in Fig. 1, B and C, respectively. These two images were taken ~20 min apart; in this subject HP-³He (Fig. 1C) was acquired before OE-¹H (Fig. 1B). Figure 1D shows the vertical profiles of spatial distribution of SV in 1-cm bins, measured from the most dependent portion of the supine lung. The two SV slopes determined from the maps in Fig. 1D were −0.012 cm⁻¹ for OE-¹H and −0.023 cm⁻¹ for HP-³He. It should be noted that neither the Vt correction nor the coarse voxel filter was applied to the data in Fig. 1.

Fig. 1.

Quantitative specific ventilation (SV) map within a sagittal slice of the right lung in subject 1 (Table 1) in the supine position. A: anatomical reference [oxygen-enhanced proton (OE-¹H) reference]. The head is located to the right, the diaphragm to the left, and the subject's back corresponds to the bottom portion of the image. Solid white line outlines the selected region of interest. B: quantitative OE-¹H SV image. C: quantitative hyperpolarized ³He (HP-³He) SV image. Common gray scale for both quantitative maps. D: vertical profiles of average SV within every isogravitational 1-cm bin from the most dependent portion of supine lung. Horizontal error bars represent SD of SV within each bin. Two dashed lines correspond to the average slopes determined by linear regression: OE-¹H: −0.012 cm⁻¹ and HP-³He: −0.023 cm⁻¹.

The voxel-by-voxel relationship and histogram distributions of SV maps of OE-¹H and Vt-corrected HP-³He (corresponding to Fig. 1, B and C) are shown in Fig. 2, A and B, respectively. In Fig. 2A, SV data were plotted on a log₁₀ scale and binned into the conventional Evans-Wagner 50-compartment model, ranging from SV values of 0.01 to 10 (12, 23). The gray scale corresponds to the data frequency within each bin; the darker color code corresponds to the more frequent data bin. The solid gray line is the regression line, for which the slope is 2.49. The correlation coefficient was 0.63 (P < 0.0001). Figure 2B shows a histogram representation of the data from Fig. 2A.

Fig. 2.

Voxel-by-voxel relationship and histogram distributions of specific ventilations (SVs) corresponding to SV maps of oxygen-enhanced proton (OE-¹H) and hyperpolarized ³He (HP-³He) for the subject depicted in Fig. 1. Tidal volume correction and 3 × 3 downsampling procedure were applied to the original OE-¹H and HP-³He data. A: SV data were plotted on a log₁₀ scale and binned into 50 SV compartments. Dashed line corresponds to the identity line; solid line is the regression line. Gray scale corresponds to the data frequency within each bin; the darker color code corresponds to the more frequent data bin. The correlation coefficient was 0.63 (P < 0.0001). B: histogram distributions of SVs (presented on a log₁₀ scale with x-axis representing 50 compartments ranging from SV = 0.01 to 10). B is the projection of the plot data in A on each axis. OE-¹H: arithmetic mean of ln(SV) distribution (SVmean) = 0.26 and SD of ln(SV) distribution (SVwidth) = 0.21. HP-³He: SVmean = 0.28 and SVwidth = 0.40.

Figure 3 shows the relationship between spatially coregistered OE-¹H and Vt-matched HP-³He SV maps at the downsampled voxel size from all eight subjects. Figures 2A and 3A are the same. Table 2 presents the correlation coefficients and slopes obtained from the two Vt-matched SV maps for each subject. The correlation coefficient was 0.46 (0.17). All eight data sets shown in Fig. 3 were combined and presented in Fig. 4. The combined correlation coefficient and slope were 0.32 and 1.53, respectively. Table 2 also includes the individual SV data for SVmean, SVwidth, and SVslope.

Fig. 3.

Voxel-by-voxel relationship of spatially coregistered oxygen-enhanced proton (OE-¹H) and tidal volume-matched hyperpolarized ³He (HP-³He) specific ventilation (SV) maps at the coarse spatial scale (i.e., 3 × 3 downsampling) from each of 8 subjects (A–H). A is from the same subject as in Fig. 2A. However, the plot data were vertically shifted as a result of tidal volume adjustment. Dashed lines in figures correspond to the identity line, and solid lines correspond to the results of Deming regression. The darker color code corresponds to the more frequent data bin for each subject. Table 2 presents correlation coefficients and slopes obtained from the voxel-by-voxel relationship of 2 tidal-volume-matched SV maps.

Table 2.

SVI results

Subject	Voxelwise Relationship		OE-1H			HP-3He
	R	Orthogonal Reg. Slope	SVmean	SVwidth	Grav. slope, cm−1	SVmean	SVwidth	Grav. slope, cm−1
1	0.63	2.49	0.26	0.21	−0.012	0.28	0.40	−0.023
2	0.56	1.93	0.27	0.25	−0.018	0.17	0.37	−0.012
3	0.63	0.96	0.42	0.20	−0.018	0.61	0.19	−0.023
4	0.49	0.30	0.15	0.31	−0.010	0.28	0.16	−0.007
5	0.48	0.46	0.25	0.40	−0.020	0.28	0.28	−0.006
6	0.26	1.58	0.31	0.30	−0.017	0.16	0.35	−0.006
7	0.11*	0.43	0.35	0.18	−0.006	0.33	0.17	−0.009
8	0.50	0.37	0.25	0.39	−0.024	0.41	0.23	−0.015
Avg (SD)	0.46 (0.17)	1.07 (0.83)	0.28 (0.08)	0.28 (0.08)	−0.016 (0.006)	0.32 (0.14)	0.27 (0.10)	−0.013 (0.007)
Comb	0.32	1.53

Pearson correlation coefficient (R) and orthogonal regression determined from the 2 spatially matched specific ventilation (SV) images for each of the 8 subjects and 3 × 3 downsampling data for 8 subjects corresponding to the respective 8 panels in Fig. 3. Avg (SD), intersubject average and ±SD; Comb, combined data plots in Fig. 4. Individual arithmetic mean of ln(SV) distribution (SVmean), SD of ln(SV) distribution (SVwidth), and gravitational slope of SV data for oxygen-enhanced proton (OE-¹H) and hyperpolarized ³He (HP-³He) are also presented. SVI, specific ventilation imaging.

^*Correlation was not statistically significant.

Fig. 4.

Oxygen-enhanced proton and hyperpolarized ³He specific ventilation (SV) comparison over all subjects. All 8 plots in Fig. 3 were combined to form this figure. The correlation coefficient of combined data was 0.32 (P < 0.01). The slope indicated by the solid line was 1.53 for the voxel-by-voxel relationship.

Figure 5 shows three Bland-Altman plots corresponding to three SV metrics, SVmean (Fig. 5A), SVwidth (Fig. 5B), and gravitational slope of SV (Fig. 5C). There was no statistically significant difference between the SVmeans [OE-¹H: 0.28 (0.08) and HP-³He: 0.32 (0.14)]. Figure 5A shows that the average bias of SVmean was 0.03 and the SD of the difference was 0.12. Figure 5B presents the Bland-Altman comparison of the SV heterogeneity (SVwidth). Overall there was a negligible negative bias of 0.01 (HP-³He being slightly smaller) with a SD of 0.13. For reference, the average values for SVwidth over the eight subjects were 0.28 (0.09) for OE-¹H and 0.27 (0.10) for HP-³He, and these values were not statistically different from each other. The gravitational SV slopes determined from OE-¹H and HP-³He SV images were −0.016 (0.006) cm⁻¹ and −0.013 (0.007) cm⁻¹ (difference not significant), respectively. The Bland-Altman bias and SD (Fig. 5C) were 0.003 cm⁻¹ and 0.009 cm⁻¹.

Fig. 5.

Bland-Altman plots of 3 specific ventilation (SV) metrics (tidal volume matched and 3 × 3 voxel downsampling): arithmetic mean of ln(SV) distribution (SVmean; A), SD of ln(SV) distribution (SVwidth; B), and gravitational slope of SV (SVslope; C). A: average bias of SVmean was 0.03, and SD of the difference was 0.12. OE-¹H, oxygen-enhanced proton; HP-³He, hyperpolarized ³He. B: average bias of SVwidth was −0.01 (slightly smaller in the case of HP-³He) with SD of 0.13. C: average bias and SD were 0.003 cm⁻¹ and 0.009 cm⁻¹, respectively.

DISCUSSION

The focus of this study was to examine the agreement of various spatial as well as nonspatial SV indexes derived from two different SV MR imaging techniques. The OE-¹H and HP-³He imaging comparison we present in this study is at a spatial scale (in plane) of 1.8 cm × 1.8 cm with a through-plane thickness of 1.0 cm corresponding to a volume of ~3.2 cm³. We corrected the ³He data so that the measured Vt values matched between the two measurements, based on measured Vt across the multiple breaths.

Despite significant intrasubject variation, the two techniques yielded similar population averages of SV indexes such as log-normal mean and width (SVmean and SVwidth) as well as gravitational slope. The voxel-by-voxel comparison of SVI maps showed only a modest degree of correlation, a fact that likely results from intrinsically different sensitivity to high and low values of SV between the two techniques.

SV measurement and SV log-normal mean.

Average SV in healthy supine subjects with a Vt of ~650 ml to 1 liter has been reported with different techniques: MBW in whole lung: 0.30–0.36 (28, 32), HP-³He in 2D multibreath imaging: 0.33 (18) and in the whole lung: 0.33 (15), and OE-¹H in 2D sagittal slice: 0.33–0.36 (32–34). Our results of SV log-normal mean in 2D sagittal slices are in close agreement with all of these previous reports: OE-¹H: 0.28 (0.08) and HP-³He with Vt correction: 0.32 (0.14).

Despite the population mean SV values being very similar, rigor demands we attempt to explain the intrasubject variation between the two methods. It is possible that OE-¹H somewhat underestimated the measured SV compared with previously reported OE-¹H. This would arise from the presence of subvoxel SV heterogeneity, as the previously reported OE-¹H studies employed finer in-plane resolution (0.16 cm × 0.16 cm) than the present study. When subvoxel SV heterogeneity is present—for instance, there are two subvoxel SV compartments in which one has greater SV than the other—the overall signal turnover within the voxel is weighted toward the long tail, slower turnover, predominantly from the low-SV compartment. Thus the OE-¹H technique would be biased toward the low-SV compartment within the voxel when coarse spatial resolution is employed. In the present study, to obtain SV maps with two different techniques, it was necessary to set in-plane resolution to the lowest common resolution, which is the resolution of the 2D HP-³He. Also, if SV drifts occur during the 18 min of OE-¹H acquisition, this could result in the underestimation of SV, as the technique biases toward low SV. In the same way that the OE-¹H method as applied here is likely to underestimate high-SV compartments, the HP-³He technique could overweight the high-SV compartments. SNR tends to be greater in a high-SV compartment, as it receives more tracer gas (more signal) during the initial inspiration than the low-SV compartment does. Thus the initial point of the decay signal as well as the entire course of the decay curve is dominated by the high-SNR compartment. This could explain the trend seen in Fig. 4, where the regression line between the two methods shows a relative overestimation at high SV values and a relative underestimation at low SV values for the HP-³He. In addition, SNR in the later breaths could limit the sensitivity of the HP-³He technique in low-SV regions. Figure 4 indicates that the greater dispersion of data plots away from the trend line was apparent in the low-SV region. This suggests that the disagreements in two techniques were somewhat dominated by the low-SV range.

Comparing OE-¹H and HP-³He measures of heterogeneity and gravitational gradient.

Overall, across the study population, comparison of the two SVI techniques demonstrated close average agreement in terms of SV heterogeneity (taken as the log-normal width of the SV distribution; Fig. 5B) and the vertical slope of SV measured within the common imaging slice (Fig. 5C). However, the interval of agreement between the two techniques, as measured by the SD in the Bland-Altman plots, was 0.13 (Fig. 5B) for the log-normal width and 0.009 cm⁻¹ in gravitational slope (Fig. 5C), corresponding to 46% and 62% of each overall average, respectively. These results suggest that the OE-¹H and HP-³He techniques presented the same levels of heterogeneity on average, albeit with considerable intersubject variability.

Comparing SV heterogeneity across techniques is not trivial, as the measured heterogeneity is greatly affected by the spatial resolution of the image. In normal subjects, Sá et al. compared SVwidth obtained with MBW technique and OE-¹H and showed that the average bias between MBW and OE-¹H in determining SV heterogeneity was ~10% of the overall average, when both distributions were scaled to a common measured Vt (32) as was done here. In the present study the average bias in SV heterogeneity between OE-¹H and HP-³He techniques was 5% of the overall average (Fig. 5B), slightly better than the comparison between OE-¹H and MBW (32). This improvement was expected; MBW measures the SV distribution of the whole lung, while OE-¹H (32) sampled a single 1.5-cm sagittal lung slice, accounting for ~8% of the lung volume at FRC. Here both measurements were sampled from the same 1.0-cm slice location and lung volume. Figure 5B also shows that the SD of the differences between techniques was 0.13 (46% of overall average), which is comparable to the intersubject variability of each technique (OE-¹H: 0.08 and HP-³He: 0.10) (33).

The reported gravitational slopes, −0.016 (0.006) cm⁻¹ [−6.0 (2.5)%/cm relative to SVmean] and −0.013 (0.007) cm⁻¹ [−4.3 (2.3)%/cm] for OE-¹H and HP-³He, respectively, are comparable to, if somewhat lower than, previously reported slopes with OE-¹H, HP-³He, ¹³³Xe, and PET (18, 21, 27, 32). The lower gravitational slopes across both OE-¹H and HP-³He in this study likely reflect the youth [32 (6) yr] and good lung health of the population studied.

The SV gravitational slope (SVslope) provides a comparison between techniques that is less sensitive to subject misregistration errors and translations in the anterior-posterior and superior-inferior directions. If the two techniques were measuring exactly the same process of regional SV using exactly the same tidal breathing maneuvers, then we would expect the SVslopes to be very similar in the same subjects in the same posture. Why the slopes do not agree more closely on an intrasubject level suggests to us that there may be fundamental differences in the two methods of measurement of SV (despite the fact that the systematic bias of the mean differences is low). It is worth noting that substantial ventilation heterogeneity is also apparent in the isogravitational (inferior-superior) direction in the SVI maps as well as the vertical profile of SV (Fig. 1). Thus the fitted slope needs to be viewed within the range of the error bars representing SV heterogeneity within the slice in Fig. 1D. In addition to the potential different subvoxel bias between techniques, the differences in breathing maneuver as well as the depth of tidal breathing between the ³He and ¹H measurements may also have an impact on the SVslope.

Test-retest reliability.

Sá et al. reported test-retest reliability of OE-¹H and MBW: the SD of differences in SV heterogeneity between techniques was ~16% of the overall average for both OE-¹H and MBW (32). Thus the intrinsic variability of both OE-¹H and MBW, including the underlying physiological temporal variability of SV heterogeneity (i.e., potential breath-by-breath variability in SV heterogeneity), is ~16%. The OE-¹H reliability (for a single 15-mm slice) showed negligible mean difference between SV distribution metrics (mean and width). The spatial (1 cm³) pixelwise Pearson correlation coefficient averaged over all of their subjects was 0.72 (range 0.40–0.93) (32). Horn et al. reported similar spatial correlation, 0.85 and 0.74 based on repeated 2D and 3D HP-³He SVIs obtained from a single normal subject in the same session (18). It should be noted that the HP-³He study by Horn et al. was performed at a much coarser spatial resolution than the OE-¹H study by Sá et al. Nonetheless, the techniques demonstrated a similar scan-scan repeatability (spatial correlation coefficient of 0.7–0.8).

Potential sources of observed intrasubject differences between techniques.

Overall, the two techniques showed close agreement for the group averages of the SV metrics reported. However, some individuals showed a higher than expected within-subject variation, with significant intertechnique variabilities in measured SVmean, SVwidth, and SVslope. This is clearly illustrated in Figs. 4 and 5. Below we discuss some of the potential factors that may have contributed to the intrasubject differences between techniques.

First of all, fluctuations in the SV distribution between two SV measurements acquired during the 1 h spent in the supine position could potentially play a role in the observed individual differences. However, the previous test-retest studies by Horn et al. and Sá et al. discussed above do not support this idea (18, 32). Hopkins et al. (16), using argon multiple breath washouts, have also reported stable metrics of SV heterogeneity over a ~60-min period (albeit in the upright seated posture). It is thus unlikely that the SV distribution changed significantly between the two separate measurements made in the present study. In addition, since the two different SVI scans were performed in a randomized order among the subjects, the potential effect of time spent in the supine posture is further minimized.

The source of the within-subject differences is therefore likely rooted in fundamental mechanistic differences between the two MRI measurement techniques.

First, a commonly raised question about OE-¹H is the potential influence of regional changes in local blood flow, driven by inspiring 100% oxygen, and how this affects the measurement of SV. The OE method is indeed an indirect measure of ventilation and assumes that all of the signal change in the tissue and blood in the lungs with oxygen washin can be attributed to ventilation. This assumption is valid only if we assume that the fraction of tissue and blood in a given voxel remains constant in response to the change in $P{A}_{O_2}$ and that the T1 enhancement due to dissolved oxygen in both of these compartments scales linearly with dissolved oxygen concentration. A recent study using three different levels of inspired oxygen concentration showed no effect on the pulmonary perfusion distribution measured with arterial spin labeling (1). However, previous work using dynamic contrast-enhanced perfusion MRI and phase-contrast MRI (24) showed changes in pulmonary perfusion and blood flow with 100% oxygen inhalation on the time course of the OE experiments. As such, further work is justified to further understand this mechanism, which will require a means of separating the perfused and tissue compartments and their oxygen enhancement. In contrast, the ³He method directly measures gas concentration in the ventilated air spaces and has no significant contribution from the tissue or blood compartment because of the low solubility of the gas and the fact that the effect of $P{A}_{O_2}$ on the ³He T1 is corrected by the breath hold calibration process (see below for discussion of potential errors introduced by this procedure).

Second, if we consider the quantitative analysis of the time series of images acquired with either technique, the impact of a single “bad” image (e.g., nonrepeatable FRC) could be more pronounced in the HP-³He technique than in the OE-¹H technique. Additionally, the HP-³He technique requires a correction factor derived from a calibration scan (i.e., T1 relaxation of the polarization of the hyperpolarized gas and depolarization of longitudinal magnetization from the small flip angle RF pulses). The estimated variability in the correction factor was ~10–18% of T1 (17). Both points above have the potential to contribute to intertechnique variability within a same subject without affecting the population average because the effect would converge into the average T1 effect.

In contrast, the presence of anatomical and instrumental dead space affects SV estimation with the OE-¹H technique. The onset of the time shift in concentration of the two inspired gases was delayed by the volume summation of inspiratory port and the anatomical dead space. The estimated anatomical dead space used in the present study was 150 ml (22). If the actual anatomical dead space is >150 ml, the signal turnover at the first breath following the change in inspiratory gas appears to be slower than it should be, resulting in the underestimation of SV. If the actual anatomical dead space is <150 ml, it would result in the overestimation of SV. Hence, it would likely not affect the population average because the mean anatomical dead space volume is ~150 ml.

Clinical applications and feasibility of the two SVI techniques.

Ventilation heterogeneity is increased in patients with lung disease (10, 23, 35, 36). Our SVI techniques provide quantitative and regionally specific information that is missing in the conventional MBW technique. The quantitative nature of the imaging provides added value to conventional single-breath hyperpolarized gas ventilation imaging as well. Thus the clinical application of SVI methods includes but is not limited to quantifying ventilation heterogeneity and detecting the regional distribution of ventilation deficiency. We anticipate that these SVI techniques will be useful in studying quantitative changes in lung ventilation in obstructive airways disease such as cystic fibrosis and asthma. On the negative side, both the HP gas and OE methods are relatively time consuming compared with single-breath hold imaging. As such, they are more suited to detailed research studies where quantification of ventilation changes is paramount rather than for more routine diagnostic questions where a breath hold ventilation-weighted HP gas MRI scan might be better suited for high-throughput imaging.

From the feasibility point of view, the HP-³He technique enables the capture of the spatial distribution of SV in the lung within 5 breaths in both a 2D plane and a 3D volume. The technique is currently only available at a limited number of MRI facilities, but recent work with ¹²⁹Xe MBW imaging suggests that this alternative and cheaper gas isotope could provide the same information as ³He MBW imaging (19). The absolute quantitative nature of MBW imaging also provides added value to conventional single-breath hyperpolarized gas ventilation imaging.

On the other hand, OE-¹H is somewhat easier to implement, requiring 100% oxygen inhalation as well as breathing equipment consisting mostly of nonrebreathing valves or a flow bypass system (5). However, the long MRI scan duration (~18 min) could make it potentially difficult for some patient populations to comply with the paced Vt breathing maneuver. The pathway to clinical implementation of OE-¹H will require shortening the acquisition time (a trade-off with quantification accuracy is possible) and/or development of a framework for free-breathing image acquisition and processing, diminishing the burden on subjects/patients.

Limitations.

The subjects recruited for this study were young and self-reported as healthy. Two subjects presented a FEV₁/FVC value of 0.72, above, but close to, the lower limit of normality (20). A Bland-Altman analysis excluding these two subjects resulted in a slight increased bias in SVslope; however, the overall conclusions remain the same. Moreover, as the primary purpose of this study is the cross-comparison of two SVI techniques it would have potentially helped if our subject population represented a wider spectrum of ventilation heterogeneity.

The Vt correction applied to the HP-³He SVI was based on the assumption that SV distribution in normal healthy lung is unaffected by changes in Vt (35, 37). However, the Vt difference was quite small [0.70 (0.22) for OE-¹H, 0.78 (0.23) for HP-³He], making this an unlikely source of significant error.

Although ³He MBW imaging can be readily performed in 3D (18), the implementation of OE-¹H performed here was limited to a 2D slice and takes ~18 min, making a 3D comparison of SV distribution between the two methods impractical. For that reason, we performed only a single-slice comparison. The scan time also requires trade-offs with the susceptibility to motion artifact on an obtained SV map. The HP-³He technique generates a SV map from a single-exponential decay curve; therefore, it is crucial to obtain a series of images with minimal misalignment caused by breathing motion. The OE-¹H technique addresses this issue to some degree by averaging over multiple oxygen washin-washout steps. Thus rejecting 5–10% of the images would not affect the resultant SV map. Proper coaching on breathing maneuvers during scanning as well as image registration during postprocessing minimizing the error associated with potential motion artifact were used in this study to obtain the best possible comparison. The two SV principal heterogeneity metrics used for comparison in the present study (SVwidth and SVslope) are insensitive to any slight misalignment between OE-¹H and HP-³He SVI by design.

The measured SV value depends on image resolution, and if subvoxel heterogeneity is present the effects on the two techniques are expected to be different. To clarify the effect, SV measurements at different spatial resolutions would be required. The physiological scale of ventilation is generally considered to be the size of an acinus, which is ~0.2 cm³ (14). The spatial resolution used for our study was ~3 cm³, and thus encompasses multiple acini, which may lead to variability between the techniques from subvoxel heterogeneity.

Conclusions.

Two MRI-based multibreath SVI techniques, namely, oxygen-enhanced ¹H-based oxygen washin-washout and hyperpolarized ³He gas washout, were directly compared with an in-house RF torso ¹H array configured with the ³He coil in situ. The two spatially coregistered SVI techniques presented a modest positive correlation. They also yielded similar population averages of SV indexes such as log-normal mean and width (SVmean and SVwidth) as well as the gravitational slope, albeit with intrasubject and intersubject variability. This is, to our knowledge, the first cross-comparison of functional maps of SV in the human lung derived from two different MRI techniques.

GRANTS

This work was supported by the UK National Institute for Health Research (Grant NIHR-RP-R3-12-027) and the National Heart, Lung, and Blood Institute (Grant R01 HL-119263).

DISCLAIMERS

The views expressed in this publication are those of the authors and not necessarily those of the National Health Service, the National Institute for Health Research, or the Department of Health.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

T.J.A., F.C.H., G.J.C., and J.M.W. conceived and designed research; T.J.A., F.C.H., M.R.R., and G.J.C. performed experiments; T.J.A., F.C.H., and R.C.S. analyzed data; T.J.A., F.C.H., R.C.S., R.J.T., G.K.P., and J.M.W. interpreted results of experiments; T.J.A. prepared figures; T.J.A., F.C.H., R.C.S., and J.M.W. drafted manuscript; T.J.A., F.C.H., R.C.S., R.J.T., G.K.P., and J.M.W. edited and revised manuscript; T.J.A., F.C.H., R.C.S., M.R.R., G.J.C., R.J.T., G.K.P., and J.M.W. approved final version of manuscript.