A Short-Breath-Hold Technique for Lung pO₂ Mapping with ³He MRI

Abstract

A pulse-sequence strategy was developed for generating regional maps of alveolar pO₂ in a single six-second breath hold, for use in human subjects with impaired lung function. Like previously described methods, pO₂ values are obtained by measuring the oxygen-induced T₁ relaxation of inhaled hyperpolarized ³He. Unlike other methods, only two ³He images are acquired: one with reverse-centric and the other with centric phase-encoding order. This phase-encoding arrangement minimizes the effects of regional flip-angle variations, so that an accurate map of instantaneous pO₂ can be calculated from two images acquired a few seconds apart. By combining this phase-encoding strategy with variable flip angles, the vast majority of the hyperpolarized magnetization goes directly into the T₁ measurement, minimizing noise in the resulting pO₂ map. The short-breath-hold pulse sequence was tested in phantoms containing known O₂ concentrations. The mean difference between measured and prepared pO₂ values was 1 mm Hg. The method was also tested in four healthy volunteers and three lung-transplant patients. Maps of healthy subjects were largely uniform, whereas focal regions of abnormal pO₂ were observed in diseased subjects. Mean pO₂ values varied with inhaled O₂ concentration. Mean pO₂ was consistent with normal steady-state values in subjects that inhaled ³He diluted only with room air.

INTRODUCTION

Regional measurement of oxygen partial pressure (pO₂) in the lung air spaces has emerged as a promising application of hyperpolarized ³He magnetic resonance imaging (1). Alveolar pO₂ is intimately related to the local ventilation-perfusion ratio (V/Q), an important physiological parameter that reflects the efficiency of gas exchange in the lung (2). Since mismatch of ventilation and perfusion is believed to be a major cause of impaired gas exchange in disease, a noninvasive method of measuring regional pO₂ variations would be a valuable tool for studying the pathophysiology of a variety of lung diseases. Existing nuclear-medicine methods for assessing regional ventilation-perfusion relationships, which are primarily used to detect pulmonary emboli, involve ionizing radiation and offer relatively poor spatial resolution.

Hyperpolarized ³He MRI is a safe, non-invasive procedure for imaging ventilated lung airspaces (3,4). The large nuclear magnetization of hyperpolarized ³He, typically five orders of magnitude larger than the thermal equilibrium value at MRI field strengths, allows the gas to be imaged with high signal-to-noise ratio (SNR) despite its low physical density. Relaxation of the ³He magnetization is accelerated by exposure to molecular oxygen (O₂), which is paramagnetic (5). Deninger et al. demonstrated that by measuring the T₁ of inhaled ³He gas, the oxygen-induced relaxation could be used to generate maps of intrapulmonary pO₂ (6). Since hyperpolarized ³He MRI can easily achieve sub-centimeter pixel resolution in acquisition times of a few seconds, pO₂-mapping techniques based on this principle offer the opportunity for significant improvement in acquisition time and spatial resolution over existing methods of characterizing V/Q (7). Furthermore, by observing T₁ changes over the duration of a breath hold, hyperpolarized ³He MR provides sensitivity to the net rate of O₂ uptake into the bloodstream.

Several variations of the basic pO₂-mapping technique have been described (6,8–10). These implementations rely on reproducible inhalations and/or long breath holds to acquire regional measurements of both instantaneous pO₂ and its rate of change. Although a small number of pO₂ mapping studies have been performed in human subjects (8,11), most of the published studies have been performed in animals (9,10,12–15). One potential obstacle to more extensive human application is that that the required breath-hold procedure may not be practical for subjects with impaired lung function. To observe temporal pO₂ evolution, multiple ³He images must be acquired during a single breath hold. It has been recognized that dividing the initially available magnetization among many image acquisitions also degrades SNR in the underlying ³He images, which can result in both noisy pO₂ maps and considerable uncertainty in the measured O₂ depletion rate (14,15).

The purpose of the present work is to explore a strategy for pO₂ mapping that is complementary to previously described methods. Instead of acquiring temporal information, all the inhaled ³He magnetization is devoted to the measurement of instantaneous pO₂. A unique feature of the method described here is the phase-encoding order of the MR pulse sequence, which is arranged in such a way that accurate pO₂ values can be calculated from a single pair of ³He images without the acquisition of an intermediate flip-angle map. This phase-encoding arrangement is used in combination with variable flip angles to generate pO₂ maps with sub-centimeter resolution in a single short breath hold less than six seconds. The short-breath-hold technique is demonstrated in pO₂ phantoms and in healthy and diseased human subjects.

THEORY

The empirical dependence of ³He longitudinal relaxation rates on temperature and O₂ concentration was quantified by Saam et al. by measuring the T₁ in sealed glass cells filled with varying mixtures of ³He and O₂ (5). Their result is given in terms of amagat density, which is a convenient unit for constant-density measurements in rigid containers, but a units transformation is required to apply this result to in-vivo measurements, in which the gas density is free to change with temperature and ambient pressure. Performing this conversion using the ideal gas law introduces an additional dependence on absolute temperature T:

$$p{O}_2=\frac{0.75T^{1.42}}{T_1}\equiv \frac{{\xi }_T}{T_1},$$ [1]

where T is in Kelvin and pO₂ is given in millibars (mbar). Thus T₁ is inversely proportional to pO₂, with a coefficient of proportionality ξ_T that varies with temperature. At body temperature, the value of this coefficient is ξ_310K = 2.59 bar·s (1940 mm Hg · s), whereas at room temperature the value is somewhat lower: ξ_297K = 2.43 bar·s (1830 mm Hg · s). Since the dominant mechanism for longitudinal relaxation of ³He in the lung is believed to be interaction with O₂ (6,16), a pO₂ map can be computed directly from a T₁ map of the inhaled gas using Eq.[1].

It is straightforward to generate T₁ weighting from a series of hyperpolarized ³He images. Since the thermal equilibrium polarization of ³He is negligible compared with the hyperpolarized value, the T₁ of hyperpolarized ³He essentially characterizes the rate at which the longitudinal magnetization decays toward zero. Thus the pixel-by-pixel ratio of any two magnetization-density images, each obtained at different times within a breath hold, contains information about the longitudinal relaxation that happened in the interim. The primary complication of this approach is that the process of image acquisition also contributes to the loss of magnetization, as the longitudinal magnetization is consumed by repeated RF excitations. Hence the T₁-weighted magnitude ratio also depends on the flip angles of the excitation RF pulses, which may have significant regional variations due to non-uniform B₁ field of the transmission RF coil, and this dependence must be accounted for to extract an accurate T₁ map.

Pulse Sequence Strategy

Fig.1a depicts the decay of longitudinal magnetization during the acquisition of two low-flip-angle, gradient-echo hyperpolarized ³He images separated by an oxygen-sensitization time Δt. It is assumed throughout the following discussion that any coherent transverse magnetization is completely spoiled before each excitation RF pulse. During the oxygen-sensitization time the magnetization decays solely via T₁ relaxation, whereas during image acquisition (shaded regions) the magnetization decays more quickly due to the additional effect of RF consumption. If the phase-encoding order is the same for both acquisitions, as depicted in Fig.1b, and the same nominal flip angle θ is used for each excitation RF pulse, then the ratio of image magnitudes I₁ and I₂ at each pixel location (x, y) is given by

$$\mid \frac{I_2(x,y)}{I_1(x,y)}\mid =\exp \left(\frac{-\Delta t-N\cdot TR}{T_1(x,y)}\right){\cos }^N\theta (x,y),$$ [2]

Where TR is the repetition time, N is the number of phase-encoding steps, and θ(x, y) is the actual flip angle, which in general is a function of position (17,18).

Figure 1.

(a) Longitudinal magnetization versus time, during the oxygen-weighted acquisition of two low-flip-angle, gradient-echo ³He images. Note that the time axis is not to scale, as the oxygen-sensitization time Δt is usually much longer than the image acquisition time (shaded region). (b) Phase-encoding line versus time, if the images are acquired as 2D Cartesian images with sequential ordering. (c) Phase-encoding line versus time, if the first image is acquired with reverse-centric ordering, and the second is acquired with centric ordering. This is the configuration used in our short-breath-hold pulse sequence.

In addition to the desired T₁ dependence, there is a significant flip-angle dependence on the right-hand side of Eq.[2]. Although cosθ ≈1 for small flip angles, when raised to the N-th power the value can deviate significantly from 1, and for reasonable values of θ, N, and Δt the flip-angle contribution to the magnitude ratio is not negligible compared with the T₁ contribution. Thus not only must the flip-angle dependence be corrected for to obtain an accurate T₁ measurement, but a globally uniform correction is generally not sufficient since any regional variations are amplified by the N in the exponent. At least one additional image, acquired with a different flip angle and/or oxygen-sensitization time, is required to separate the flip-angle from T₁ dependence and calculate accurate pO₂ maps. This strategy is used for previously described pO₂-mapping methods.

Consider instead the situation in which the first ³He image is acquired in reverse-centric phase-encoding order while the second is acquired with centric ordering, as depicted in Fig.1c. In this case the flip-angle contribution to the magnitude ratio is different from Eq.[2], since the intensity of the first image depends primarily on the magnetization present at the end of its acquisition, whereas the intensity of the second image depends primarily on the magnetization present at the beginning of its acquisition. In the approximation that the pixel intensity is determined by the zero-spatial-frequency (central) k-space sample, the flip-angle contribution to the magnitude ratio is reduced to a single excitation RF pulse:

$$\mid \frac{I_2(x,y)}{I_1(x,y)}\mid =\exp \left(\frac{-\Delta t-TR}{T_1(x,y)}\right)\cos \theta (x,y).$$ [3]

This relationship is similar to Eq.[2] but without the N-power amplification of the flip-angle term. Thus values of θ and Δt can easily be chosen such that the signal decay due to T₁ relaxation is much larger than the RF consumption.

Because the reverse-centric/centric phase-encoding arrangement dramatically reduces the flip-angle dependence of the magnitude ratio, this pulse sequence allows accurate T₁ maps to be obtained directly from the pixel-by-pixel ratio of the resulting images, without performing regional flip-angle corrections. That is, a pO₂ “snapshot” can be calculated from a single pair of ³He magnitude images acquired a few seconds apart during the same short breath hold. Furthermore, pO₂ measurement noise can be minimized because none of the available MR signal is diverted to the acquisition of an intermediate flip-angle map. This strategy is the basis for our short-breath-hold method (19,20).

Variable Flip Angles

If the same flip angle is used for every excitation RF pulse of the short-breath-hold acquisition (Fig.1c), then the transverse magnetization generated at each phase-encoding step declines at the same rate as the longitudinal magnetization (Fig.1a). This effect, in combination with the opposite phase-encoding order of the two image acquisitions, leads to highly asymmetric k-space weightings once the phase-encoding lines have been properly re-ordered, as depicted in Fig.2a. Such asymmetric k-space filters would cause significant differences in blurring between the reconstructed images that could compromise the pixel-by-pixel pO₂ calculation.

Figure 2.

Flip angle sequence and resulting k-space filters, for two hyperpolarized ³He images acquired with the reverse-centric/centric phase encoding order depicted in Fig.1c, using (a) the same flip angle for all excitation RF pulses, and (b) a variable-flip-angle sequence that satisfies Eq.[5]. The k-space filters are much more symmetric for the variable-flip-angle case; the slight residual asymmetry is due to T₁ decay during the acquisition.

This situation can be addressed by increasing the flip angle at each excitation RF pulse, to compensate for the declining longitudinal magnetization (21), as depicted in Fig.2b. For an arbitrary sequence of flip angles θ_n (1 ≤ n ≤ 2N), the transverse magnetization $M_n^T$ immediately following application of the n-th excitation RF pulse is given by

$$M_n^T=M_0^L\exp \left(\frac{-t_n}{T_1}\right)\sin {\theta }_n\prod _{j=1}^{n-1}\cos {\theta }_j,$$ [4]

where $M_0^L$ is the initial hyperpolarized magnetization, t_n is the time at which the n-th excitation RF pulse is applied, and it is assumed that any remaining transverse magnetization coherence is spoiled before the next excitation RF pulse. It is easily verified that if the flip angles θ_n satisfy the recursive relationship

$$\tan {\theta }_n=\sin {\theta }_{n+1},$$ [5]

then the flip-angle product in Eq.[4] has the same value for all n, and therefore any decrease in the transverse magnetization is due solely to T₁ relaxation:

$$M_n^T\propto M_0^L\exp \left(\frac{-t_n}{T_1}\right).$$ [6]

Since the T₁ decay is small during the acquisition of a single image, the resulting k-space filters are much more symmetric than in the constant-flip-angle case, as illustrated in Fig.2.

Another consequence of the variable-flip-angle implementation is that T₁ can be calculated directly from the reconstructed images without explicit flip-angle correction. For the short-breath-hold pulse sequence, the formula is:

$$T_1(x,y)=\frac{\Delta t+TR}{\log \mid I_1(x,y)∕I_2(x,y)\mid },$$ [7]

which follows from Eq.[6], using the same line of reasoning that led to Eq.[3]. It is important to note that the lack of an explicit flip-angle term in Eq.[7] does not mean that the results lack all flip-angle dependence. The flip-angle dependence that appeared explicitly in Eq.[3] has simply been absorbed into the requirement that the sequence of flip angles satisfy Eq.[5]. If the actual flip angles do not satisfy this requirement, for example due to regional variations in RF coil sensitivity, there will be a systematic error in the T₁ calculated from Eq.[7].

Effects of Flip-Angle Errors

To compute the variable-flip-angle sequence in practice, a terminal flip angle θ_2N ≤ 90° is selected for the final excitation RF pulse, and the other flip angles in the sequence are calculated backwards from there using Eq.[5]. This sequence is implemented on the MR scanner by changing the amplitude of each excitation RF pulse in proportion to the desired flip angle. If the relationship between the RF pulse amplitude and resulting flip angle is not properly calibrated, however, then the actual flip angles will differ from the prescribed values. The purpose of this subsection is to quantify the resulting error on our pO₂ measurement.

In the following analysis, the sequence of flip angles {θ_n} derived from Eq.[5] will denote the intended flip angles, whereas the actual flip angles experienced by the hyperpolarized magnetization are given by the sequence {α(x,y)θ_n}. The error factor α, which may be attributable to an error in the average transmitter calibration and/or to regional variations about this average due to B₁-field inhomogeneity, is in general a function of position. Within this framework, the transverse magnetization generated by the n-th excitation RF pulse, originally given in Eq.[4], can be rewritten as

$$M_n^T=M_0^L\exp \left(\frac{-t_n}{T_1}\right)f_n\left(\alpha \right),$$ [8]

where the flip-angle dependence has been encapsulated in the function

$$f_n\left(\alpha \right)\equiv \sin \alpha {\theta }_n\prod _{j=1}^{n-1}\cos \alpha {\theta }_j.$$ [9]

Using the same line of reasoning that led to Eq.[3] and Eq.[7], T₁ can be extracted from the magnitude ratio of two hyperpolarized ³He images acquired during the same breath hold:

$$\frac{1}{T_1(x,y)}=\frac{1}{t_{n_2}-t_{n_1}}\log \mid \frac{I_1(x,y)\cdot f_{n_2}\left(\alpha \right)}{I_2(x,y)\cdot f_{n_2}\left(\alpha \right)}\mid ,$$ [10]

where n₁ and n₂ are the indices of the excitation RF pulses corresponding to the central phase-encoding lines of the first and second image, respectively. Note that we have not made any assumptions about the phase-encoding order in the derivation of this expression.

For our short-breath-hold pulse sequence, in which the intended flip angles satisfy Eq.[5], if there is no flip-angle error (i.e. α(x, y) ≡ 1) then f_n1 and f_n2 are equal and Eq.[10] reduces to Eq.[7]. In general, however, the local flip-angle error will not be known, and Eq.[7] will be used to calculate T₁. The resulting systematic error when α ≠ 1 can be quantified by comparing T₁ calculated from Eq.[10] with T₁ calculated from Eq.[7]. The resulting pO₂ error is plotted versus α in Fig.3a, assuming typical values for pO₂ and N, for several different combinations of Δt and θ_2N. For reasonable values of these parameters, the fractional pO₂ error is never more than a few percent, even for flip angle errors as large as 20%.

Figure 3.

Fractional pO₂ error versus flip-angle error factor α, if no correction is made for deviations from the nominal variable-flip-angle sequence given in Eq.[5]. Each curve represents a different combination of the oxygen-sensitization time Δt and the terminal flip angle θ_2N. The curves in (a) were generated using the short-breath-hold phase encoding order depicted in Fig.1c, whereas the curves in (b) were generated using sequential ordering as depicted in Fig.1b. Note that although the character of the two sets of curves is similar, the scaling of the y-axes differs greatly. All curves were generated assuming N = 48 phase encoding lines and pO₂ = 100 mm Hg.

To put such errors in perspective, Fig.3b shows the fractional error that results when the pO₂ is still calculated from two ³He images (acquired using the same variable-flip-angle sequence used to generate Fig.3a) but with sequential phase encoding (as depicted in Fig.1b). The systematic error is zero for α = 1, but the error approaches 100% or more for a flip-angle error of 20%. This graph shows that the N-power amplification discussed earlier is much less forgiving of even small flip-angle errors, and demonstrates the need for explicit correction.

Although it is not possible to measure and correct for regional flip-angle variations when acquiring only two ³He images, it is quite feasible to estimate the average flip-angle error from a small number of global signal acquisitions. Thus by appending a few auxiliary signal acquisitions to the short-breath-hold pulse sequence, we are able to correct for the average value of α, so that the residual pO₂ errors shown in Fig.3a arise only from local B₁ variations. Although Eq.[10] could be used to apply this correction, a more elegant approach is to use α in Eq.[9] to compute the relative flip-angle weighting of each k-space line, and to divide each line of k-space data by the corresponding value of f_n(α) before image reconstruction. This procedure eliminates the need to rely on the assumption that the pixel intensities are determined solely by the central k-space acquisition, and allows accurate T₁ values to be calculated directly from the ³He images using Eq.[7].

METHODS

Our short-breath-hold pO₂ mapping technique was tested in phantoms and human subjects. All MR imaging was performed on a 1.5-T whole-body scanner (Magnetom Sonata; Siemens Medical Solutions, Malvern, PA), using a flexible vest-shaped chest coil (IGC Medical Advances, Milwaukee, WI) for human studies and a smaller homebuilt birdcage coil for phantom studies. The ³He gas was polarized to approximately 35% by collisional spin exchange with an optically pumped rubidium vapor using a commercial system (Model 9600 Helium Polarizer; Magnetic Imaging Technologies Inc., Durham, NC). Just before imaging, the desired volume of hyperpolarized ³He gas was dispensed into a Tedlar plastic bag (Jensen Inert Products, Coral Springs, FL) and carried to the scanner. In some cases N₂ was added to the bag to increase the total gas volume.

Pulse Sequence

For each hyperpolarized ³He scan, a fast gradient-echo pulse sequence was used to acquire two magnetization-density images separated by an oxygen-sensitization time of 2-3 seconds. Large crusher gradients (32 mT/m, duration 2 ms) were applied following each data readout window to de-phase any remaining coherent transverse magnetization before the next excitation RF pulse. (In addition to the conventional de-phasing action provided by these gradients, they impart substantial diffusion attenuation to the MR signal.) The first image was collected in reverse-centric phase-encoding order and the second was collected in centric order, as depicted in Fig.1c. The flip angle was progressively increased at each excitation RF pulse according to Eq.[5], up to a terminal flip angle that was between 6° and 20°.

The reverse-centric/centric phase-encoding arrangement is susceptible to eddy-current differences at the central k-space acquisitions, since for the first image, central k space is covered at the end of a series of acquisition cycles, whereas for the second image, central k space is covered immediately following a long acquisition delay. To maintain eddy currents in a steady state throughout the acquisition of both images, the readout and crusher gradients were executed for several repetition cycles leading up to the first image acquisition and were repeated continuously throughout the entire duration of the oxygen-sensitization time. Preliminary testing of this pulse sequence in water phantoms revealed noticeable phase differences between the two reconstructed images in the absence of this eddy-current stabilization.

To be able to estimate the average flip-angle error, the central k-space line was measured in two additional diagnostic acquisitions: once immediately before acquisition of the first image and again immediately after acquisition of the second image. Thus the central k-space line was acquired a total of four times for each oxygen-weighted pair of ³He images. The pulse sequence also included the option to repeat the entire image-pair acquisition, so that two independently measured pO₂ maps could be obtained from some of the human subjects by increasing the breath-hold duration.

Data Analysis

The raw k-space data was captured from the scanner and analyzed offline using an ordinary personal computer running MATLAB (The MathWorks, Natick, MA). Prior to image reconstruction, the data from each of the four central k-space line acquisitions was inverse-Fourier transformed, and the integrated magnitude was used to provide a relative measure of the total transverse magnetization generated at each of these four excitations. Global estimates of the average flip-angle error factor α and oxygen partial pressure were then computed by performing a least-squares fit of these four measurements to Eq.[8], using the known sequence of intended flip angles and excitation times. This globally estimated flip-angle error factor was used in Eq.[9] to calculate the average flip-angle weighting f_n(α) of each k-space line, and this weighting was corrected for by dividing each raw data line by the corresponding value of f_n. Finally, the corrected k-space data lines were ordered properly and inverse-Fourier transformed to generate a pair of ³He magnetization-density images.

A separate signal mask was computed for each ³He image, by selecting all pixels whose magnitude was greater than 10% of the maximum pixel magnitude, and a composite mask was constructed from the intersection of the individual masks. At each unmasked pixel location, T₁ was calculated using Eq.[7], and pO₂ was then calculated using Eq.[1] using the appropriate temperature-dependent coefficient. A temperature of 310 K was assumed for humans, and the measured room temperature of 297 K was used for phantoms. Mean pO₂ was computed from the simple pixel average over each map.

Phantom Experiments

To verify that regional pO₂ variations can be accurately measured using our short-breath-hold pulse sequence, three phantom experiments were performed. In each experiment, two square, pillow-shaped Tedlar plastic bags (250-ml nominal capacity) were partially filled with different mixtures of N₂ and hyperpolarized ³He gas and placed side-by-side in the RF coil. The prepared gas mixtures are given in Table 1. Ambient pressure was assumed to be 1 bar. Immediately before imaging, a measured volume of O₂ was injected into one or both of the bags. To ensure adequate mixing with the ³He/N₂ mixture, the O₂ was injected forcefully into the bag from a syringe connected to the input tube. The total volume of gas inside each bag was limited to approximately 200 ml to keep the bag slightly under-inflated and thereby ensure that the pO₂ measurement was made at ambient pressure. All scans were acquired as 2D projections using non-selective RF excitations. Pulse-sequence parameters were: TR/TE, 10/3.1 ms; FOV, 160 mm × 320 mm; matrix, 32 × 64; readout bandwidth, 200 Hz/pixel; oxygen-sensitization time, Δt = 3.0 s.

Table 1.

Prepared gas mixtures and measured pO₂ values for all three phantom experiments

Exp #	Bag A					Bag B
	Gas mixture [ml]			pO2 [bar]		Gas mixture [ml]			pO2 [bar]
	3He	N2	O2	Prep.	Meas.	3He	N2	O2	Prep.	Meas,
1	40	160	0	0.00	−0.002	40	120	40	0.20	0.209
2	40	140	20	0.10	0.099	40	160	0	0.00	−0.002
3	40	150	10	0.05	0.046	40	130	30	0.15	0.142

Human Experiments

Lung pO₂ maps were acquired in four healthy volunteers (all female, age range 18–24 years) and three subjects (1 female, age 65; 2 male, ages 66 and 60) that had previously undergone a unilateral lung transplant secondary to a clinical diagnosis of chronic obstructive pulmonary disease (COPD). All studies were performed under a Physician’s Investigational New Drug Application (IND 57866) for imaging with hyperpolarized ³He, using a protocol approved by our institutional review board, and informed written consent was obtained from each subject. The subject’s heart rate and O₂ saturation level were monitored throughout each imaging session, and a physician supervised all procedures. Typical pulse sequence parameters included: TR/TE, 7.5/2.1 ms; readout bandwidth, 390 Hz/pixel; readout FOV, 400 mm; readout pixels, 64; in-plane voxel size, 6.5×6.5 mm. The readout gradient was applied in the head-foot direction, and the number of phase-encoding lines ranged from 40 to 48, depending on lung width.

MR imaging was performed at breath hold following inhalation of hyperpolarized ³He plus N₂, O₂, and/or room air. Experimental parameters for each ³He scan, including the inhaled gas mixture, are given in Table 2. For all scans, the indicated amounts of ³He and medical-grade N₂ were combined in a single Tedlar bag and inhaled through a short plastic tube. In some of the studies (Healthy #2, first two scans of Healthy #4, and COPD #1, 2, 3) a second plastic tube was taped parallel to the tube coming from the bag, and the subject was instructed to inhale through both tubes. This allowed the subject to simultaneously draw in room air while inhaling the anoxic bag mixture. To test the reproducibility of this procedure, two such scans were performed in healthy subject #4. To systematically observe the effects of varying the inhaled O₂ concentration, two additional scans were also performed in this subject. For one of these scans the subject inhaled an anoxic mixture of ³He and N₂; for the other she simultaneously inhaled from two different bags containing ³He and O₂, respectively.

Table 2.

Experimental parameters and quantitative results for all 12 human scans

Subject	Inhaled gas mixture	TH of isolated slice	Δ t	Terminal flip angle	Mean pO2 [mm Hg]a		Flip angle error factor
					Pixel mean	Global estimate
Healthy #1	300ml 3He + 400ml N2 380ml 3He + 320ml N2	— 40 mm	2.6s 2.6s	20° 20°	81 86	82 87	1.01 1.00
Healthy #2	300ml 3He + air	—	2.3 s	20°	102	99	0.98
Healthy #3	350ml 3He + 1L N2 Same	— 35 mm	2.4s 2.4s	6° 6°	86, 77 85, 88	86, 81 85, 89	1.11 1.04
Healthy #4	250ml 3He + air 250ml 3He + air 250ml 3He + 1L O2 250ml 3He + 750ml N2	— — — —	2.6s 2.6s 2.6s 2.6s	10° 10° 10° 10°	101 104 228 58	102 105 229 57	0.98 0.99 0.98 1.00
COPD #1	500ml 3He + 400ml N2	40 mm	2.0 s	12°	76, 72	78, 72	0.92
COPD #2	220ml 3He + air	—	2.7s	10°	101, 87	99, 83	0.98
COPD #3	150ml 3He + 630ml N2 + air	—	3.0 s	10°	98	98	1.15

^aFor scans in which two pO₂ maps were collected, mean values for both are given.

To avoid complications arising from ³He diffusion across slice boundaries (11), all image sets were obtained as coronal projections using a non-selective excitation RF pulse. Thinner sections were isolated in some of the human scans (Healthy #1, 3, and COPD #1), by pre-saturating the magnetization on either side of a 35-40 mm-thick coronal section. For these scans, ten large-flip-angle (~90°) slice-selective RF pulses with a sharp slice profile, each followed by a dephasing gradient, were applied on each side of the desired section before acquisition of the ³He projection images. In several of the subjects (Healthy #3, COPD #1, 2) two consecutive pO₂ maps were acquired during each breath hold, by repeating the entire pulse-sequence acquisition. Both pairs of images were then analyzed independently, to assess the repeatability of the acquisition.

RESULTS

Fig.4a shows a representative ³He image and corresponding pO₂ map from the first phantom experiment. The mean measured pO₂ values for all Tedlar bag phantoms are given in Table 1 alongside the prepared O₂ concentrations, and the data are graphed in Fig.4b. Measured values are given in bars, to facilitate comparison with the prepared O₂ concentrations. The mean difference between measured and prepared pO₂ values was −1.4 ± 5.7 mbar (−1.0 ± 4.3 mm Hg, mean ± standard deviation), with a maximum discrepancy of 9 mbar (7 mm Hg).

Figure 4.

Results of phantom experiments. (a) ³He image and corresponding pO₂ map from the first phantom experiment. The Tedlar bag phantoms are viewed from the side. The bag on the right contains 20% O2, while the bag on the left contains no O2. (b) Plot of pO₂ pixel average versus prepared O₂ concentration for all six Tedlar bag phantoms. The vertical error bars represent the standard deviation of pixel values in each pO₂ map, and the horizontal error bars correspond to a 5% error on the prepared gas volumes. The gray line represents the expected results.

A total of 12 breath-hold scans were performed in the 7 human subjects. Fig.5 shows ³He images and pO₂ maps obtained from the first and third healthy volunteers. The posterior-slice pO₂ maps (bottom row) are noisier than the projection maps (top row) due to lower signal-to-noise ratio in the underlying ³He images, but the mean pixel values are similar. In the healthy subjects, pO₂ maps appeared largely homogeneous except near the diaphragm and heart, where anomalously low or high values were occasionally found.

Figure 5.

³He images and corresponding pO₂ maps from (a) healthy subject #1 and (b) healthy subject #3. The images in the top row are coronal projections through the entire lung, while the images in the bottom row are of thinner posterior sections. These sections were isolated by first destroying the hyperpolarized magnetization on either side of a 40-mm coronal section and then acquiring projection images of the remaining hyperpolarized magnetization. The second pO₂ maps in (b) were obtained in the same breath holds as the first, by repeating the entire short-breath-hold pulse-sequence.

Quantitative results from the human scans are summarized in Table 2. For steady-state tidal breathing of room air (21% O₂), the expected alveolar pO₂ is approximately 102 mm Hg (2). Among the breath-hold scans in which a significant amount of room air was inhaled along with undiluted ³He (Healthy #1, first two scans of Healthy #4, and COPD #2), the mean measured pO₂ ranged from 101 to 104 mm Hg, which is remarkably consistent with the expected steady-state value. All the scans in which the ³He was diluted with N₂ before inhalation (Healthy #1, 3, and COPD #1, 3) yielded lower mean pO₂ values, ranging from 76 to 98 mm Hg.

A more dramatic demonstration of the influence of the inhaled O₂ concentration is seen in Fig.6, which shows three pO₂ maps from healthy subject #4. The map on the left, obtained following inhalation of ³He plus room air, has a mean value of 104 mm Hg. By contrast, the map in the middle, obtained following inhalation of a mixture of ³He and O₂, has a much higher mean pO₂ of 228 mm Hg, whereas the map on the right, obtained following inhalation of an anoxic mixture of ³He and N₂, has a much lower mean value of 58 mm Hg.

Figure 6.

Three pO₂ maps from the same healthy subject, each obtained in a separate breath hold, using widely different inhaled O₂ concentrations. The inhaled gas mixtures were: (a) ³He plus room air, (b) ³He plus O₂, and (c) ³He plus N₂. Mean pO₂ values are 104, 228, and 58 mm Hg, respectively. The origin of the regional pO₂ variations in (c) is unknown.

Imaging results from the lung-transplant COPD patients are shown in Fig.7. In all three patients, the native lung appears hyperexpanded and does not fully ventilate, as can be seen by comparing the ¹H and ³He images. In the first patient, the measured pO₂ distribution is more heterogeneous in the native lung than in the transplanted lung, although focal regions of abnormally high or low pO₂ values are present in both lungs (green arrows). The second pO₂ map from this patient, which was obtained during the same breath hold, exhibits similar focal abnormalities, indicating that these variations are not random fluctuations.

Figure 7.

Proton MRI, ³He image, and pO₂ maps from the COPD lung-transplant subjects. (a) Focal regions of abnormally high or low pO₂ are present in both lungs (arrows). (b) Measured pO₂ in the transplanted lung is fairly uniform except for apparent motion artifacts near the diaphragm (arrows). (c) Focal regions of abnormally high pO₂ are present in the native lung (green arrows), and likely motion artifacts are present in the transplanted lung (black arrow).

The ³He image from the second patient shows that the native lung is very poorly ventilated. Although some of the inhaled ³He reaches the lower part of the right lung, none of the pixels pass the signal threshold for pO₂ calculation. In the third patient, the native COPD lung is partially ventilated, and focal regions of elevated pO₂ are apparent (green arrows). In both of these patients, measured pO₂ values are relatively normal and uniform in the transplanted lung except for elevated values near the diaphragm (black arrows).

DISCUSSION

It is certainly reasonable to expect that the alveolar pO₂ will depend on the inhaled O₂ concentration. The mean measured pO₂ was less than 100 mm Hg in all subjects who inhaled a substantial fraction of anoxic gas (³He diluted with N₂), whereas mean values greater than 100 mm Hg were measured in all subjects who inhaled undiluted ³He plus room air or O₂. Although these results support the validity of the measured values, they also demonstrate that the inhalation procedure must be standardized if pO₂ measurements are to be compared among different scans. To minimally disturb steady-state pO₂ while performing the ³He measurement, it would be desirable for the subject to inhale an evenly mixed combination of ³He, N₂, and 21% O₂. Ideally this mixing would occur just before inhalation, to keep the O₂ from prematurely depolarizing the ³He. Other groups have accomplished this using a respirator system (8). Another possibility is to simultaneously inhale from two bags, where one contains ³He and the other contains N₂ plus enough O₂ to bring the total O₂ concentration to 21%. We obtained reasonable results using a double straw configuration, in which a straw open to the room was taped parallel to a straw leading to a bag of pure ³He, and the subject inhaled through both straws simultaneously. The pO₂ maps obtained using this procedure had mean values ranging from 101 to 104 mm Hg, which are consistent with the normal steady-state value of 102 mm Hg. Although the inhaled mixture would always have less than 21% O₂ using our double-straw technique, for small enough ³He volumes the difference might be negligible. Higher ³He polarization (which would permit the use of smaller ³He doses) would help in this regard.

We did not perform a systematic investigation of reproducibility, but the small number of repeated scans yielded encouraging results. The first two scans of healthy subject #4, for which identical inhaled mixtures were used, yielded mean pO₂ values that differed by 3%. In the four scans where two independent pO₂ maps were obtained in the same breath hold, both pO₂ maps appeared very similar, with a mean pO₂ difference of −6 mm Hg. This pO₂ decrease, which occurs over approximately 3 seconds, is similar to reported values of the O₂ depletion rate measured in human lungs using hyperpolarized ³He (6,8,11).

Since computation of the pO₂ map involves comparing individual pixel magnitudes from images acquired several seconds apart, any motion during the interim can lead to erroneous pO₂ values. We believe that the elevated pO₂ values at the base of the lung in some of the pO₂ maps result from relaxation of the diaphragm during the oxygen-sensitization time, which causes a local signal decrease that is incorrectly attributed to high O₂ concentration. For instance, close inspection of both sets of ³He images from COPD subject #2 reveals that the diaphragmatic border is at a slightly different position (about one pixel higher) in the second image of each pair. The precise origin of anomalous pO₂ values near the heart is more difficult to explain, but it seems reasonable to expect that the measurement will be unreliable in any region with bulk motion. It might be possible to reduce this problem though the use of cardiac gating, but implementation would not be straightforward in this pulse sequence since it would require the end of the first acquisition block to be synchronized with the beginning of the second.

Aside from anomalous values in the vicinity of bulk motion, all pO₂ maps obtained from healthy subjects after near-normal inhalations were homogeneous. In particular, we did not observe any obvious pO₂ gradients in the head-foot direction. A gravitationally induced pO₂ gradient is known to be present when a person is upright, but this gradient is expected to largely disappear in the supine position (2), which is consistent with our results. The final scan of healthy subject #4, in which a completely anoxic mixture of ³He and N₂ was inhaled, yielded a very inhomogeneous pO₂ map. Although it is possible that the subject shifted position slightly in the middle of the acquisition, such a shift is not obvious from the ³He images. Thus the origin of these apparent pO₂ variations remains mysterious, but it is worth noting that this scan was performed soon after the previous scan, in which a large amount of pure oxygen was inhaled along with the ³He dose.

It has been pointed out elsewhere (11) that the use of slice-selective RF excitations for image acquisition would yield inaccurate pO₂ maps, since the diffusion of “fresh” magnetization across the slice boundary during the oxygen-sensitization time would artificially increase the intensity of the second image. To avoid these complications, all pO₂ maps in the present study were acquired as coronal projections using non-selective excitations. Our use of pre-saturation RF pulses to isolate a thinner section, by destroying the hyperpolarized magnetization on either side of a coronal slice before image acquisition, appeared to work well. Although diffusion perpendicular to the imaging plane will tend to blur the effective slice profile during image acquisition, the pO₂ measurement should not be compromised since all of the unsaturated magnetization is captured by the non-selective excitation RF pulses. In the two healthy subjects that underwent both slice-isolated and whole-lung scans, the mean pO₂ values differed by 6% and 1%, respectively.

CONCLUSION

We have demonstrated that our short-breath-hold pO₂ mapping technique can accurately measure known regional variations in phantoms, and yields plausible pO₂ maps in human subjects. It remains to be seen whether ³He-based pO₂ mapping techniques will yield clinically useful information, but the prospects appear encouraging. The self-consistency of our in-vivo results, in combination with the regional accuracy of phantom measurements, suggests that this method can generate accurate maps of the actual pO₂ present in the lung at the time of the measurement. However, for future studies it would be highly desirable to standardize the breath-hold procedure and inhaled gas mixture, using 21% oxygen, in order to minimally disturb what is being measured, and to allow the measured pO₂ values to be compared across subjects. In the absence of an efficient 3D acquisition, the pre-saturation technique used in this study appears to be a viable strategy for improving through-plane resolution without compromising the measured pO₂ values. Apparently anomalous pO₂ values were often found in regions susceptible to bulk motion during image acquisition. This may prove to be an unavoidable limitation of the underlying T₁-based measurement, and such effects should be carefully considered when interpreting non-uniform pO₂ maps.