“Effects of Neonatal Lung Abnormalities on Parenchymal R₂* Estimates”

Abstract

Background:

Infants admitted to the neonatal intensive care unit (NICU) often suffer from multifaceted pulmonary morbidities that are not well understood. Ultrashort echo time (UTE) MRI is a promising technique for pulmonary imaging in this population without requiring exposure to ionizing radiation.

Purpose:

To investigate the effect of neonatal pulmonary disease on R₂* and tissue density and to utilize numerical simulations to evaluate the effect of different alveolar structures on predicted R₂*.

Study Type:

Prospective.

Population:

17 neonatal human subjects (5 control, 7 with bronchopulmonary dysplasia (BPD), 5 with congenital diaphragmatic hernia (CDH)). 12 male, 5 female, post-menstrual age (PMA) at MRI 39.7±4.7 weeks.

Field Strength/Sequences:

1.5T/multi-echo 3D UTE MRI.

Assessment:

Pulmonary R₂* and tissue density were compared across disease groups over the whole lung and regionally. A spherical shell alveolar model was used to predict the expected R₂* over a range of tissue densities and tissue susceptibilities.

Statistical Tests:

Tests for significantly different mean R₂* and tissue densities across disease groups were evaluated using analysis of variance (ANOVA), with subsequent pairwise group comparisons performed using t-tests.

Results:

Lung tissue density was lower in the ipsilateral lung in CDH compared to both controls and BPD patients (both P<0.05), while only the contralateral lung in CDH (CDHc) had higher whole lung R₂* than both controls and BPD (both P<0.05). R₂* differences were significant between controls and CDHc within all tissue density ranges (all P<0.05) with the exception of the 80–90% range (P=0.17). Simulations predicted an inverse relationship between alveolar tissue density and R₂* that matches empirical human data. Alveolar wall thickness had no effect on R₂* independent of density (P=1).

Data Conclusion:

The inverse relationship between R₂* and tissue density is influenced by the presence of disease globally and regionally in neonates with BPD and CDH in the NICU.

Introduction

Infants admitted to the neonatal intensive care unit (NICU) often suffer from multifaceted pulmonary morbidities that are not well understood [1,2]; thus, clinical diagnostic pulmonary imaging techniques have the potential to offer significant value. Given that exposure to ionizing radiation associated with computed tomography (CT) carries significant risk to the safety of these patients [3,4], particularly for repeated exposure, researchers have begun developing MRI techniques suitable for imaging of neonatal pulmonary structure [5,6]. Specifically, 3D ultra-short echo time (UTE) MRI [6] allows for self-navigated retrospective respiratory gating [7,8] in order to reduce respiratory- and bulk-motion-related blurring and reconstruction of images at multiple respiratory phases when imaging non-sedated, quiet-breathing infants. This technique has been proven to provide quantitative parenchymal lung density measures that correlate well with CT [9], quantify lung volumes and hyperinflation [10], and predict short-term clinical outcomes in NICU patients with bronchopulmonary dysplasia (BPD), chronic lung disease of prematurity [11]. Recent work has advanced this technique allowing measurement of pulmonary R₂* in NICU patients without lung disease [12], and here we focus on applying this method to NICU patients who have clinical diagnoses of pulmonary disease.

A consistent trend between lung tissue density and R₂* in NICU patients and adults with normal lungs has been previously established [12]. Work in adult humans has demonstrated the dependence of R₂* on inflation state [13,14] and gravitational dependence [15], both of which are associated with bulk tissue density. A relationship between density and R₂* consistent with human experiments has been shown in murine lungs as well [16,17]. These results raise the question as to whether R₂* contains any unique information beyond what can be derived from tissue density alone. Certain pulmonary pathologies are associated with abnormal tissue density, for example emphysema (low density) and fibrosis (high density), and one could expect associated abnormalities in R₂* due to differences in magnetic susceptibility of the lung structures. However, R₂* in the lung is also dependent on water self-diffusion and capillary blood flow [18]. Given that the difference in susceptibility between air and tissue is approximately Δχ = −9.0ppm, while Δχ = −6.0ppm between air and deoxygenized blood [19,20], R₂* could also be influenced by relative blood volume and oxygenation level. This raises the possibility for utilizing R₂* estimates in the lung to detect disease related phenomena that are not necessarily reflected entirely in the density of the tissue.

The goal of this work was to explore the influence of lung disease, specifically BPD and congenital diaphragmatic hernia (CDH), on the pulmonary R₂* of NICU patients. We also aimed to provide numerical simulations of the magnetic field distribution within a simple spherical model of the alveolus to explore the effects of tissue density, alveolar wall thickness and susceptibility on the predicted R₂* in an effort to explore the source of any observed disease related differences.

Methods

Numerical Simulation

A single spherical shell was used to represent an alveolus in the magnetic field distribution simulations. Simulations of neonatal and adult alveoli were performed in Matlab (The MathWorks, Inc., Natick MA). The shell was defined on a 3D sampled grid with matrix 512 × 512 × 512, and an isotropic numerical dimension of each grid sample of 1μm³. The simulations were performed with the magnetic susceptibility within the shell defined as Δχ = −9ppm and Δχ = −7.5ppm to reflect the range from air-to-normally oxygenated blood and air-to-fully deoxygenated blood, respectively. These values assume a hematocrit of 0.5 and that lung tissue mass is dominated by blood [26]. Spin density within the shell is assumed to be uniform. In the model, tissue density was defined as the ratio of the shell volume to the total volume, inclusive of the alveolar airspace plus the shell. The frequency shift, dω, within the shell was computed numerically as defined in [27]. The signal evolution over time, S(t), was produced using

$$S\left(t\right)=\sum _{k=0}^n{e}^{\gamma 2\pi Mi\left(1+d\omega \left(k\right)\right)t}$$ (1)

, where γ=42.58 MHz/T is the proton gyromagnetic ratio, M=1.5 T is the main magnetic field strength, dω(k) is the frequency shift (in ppm) in the k^th voxel, and n is the number of voxels in the shell. The R₂* was estimated by sampling S(t) at the echo times used for imaging and fitting a mono-exponential decay to these data. The echo times used were 0.200ms, 0.95ms, 1.700ms, and 2.45ms for the simulations of neonatal alveoli and 0.09ms, 0.59ms, 1.09ms and 1.59ms for simulations of adult alveoli [12].

To establish physiologically relevant upper and lower bounds on the geometry of the shell, dimensions were estimated using average wall thickness and volume estimates from previous work using chemical shift saturation recovery with hyperpolarized Xenon-129 MRI [28]. Specifically, estimates of surface area to volume ratio in adults at total lung capacity (TLC) and functional residual capacity (FRC) + 1L were applied to constrain the geometry of the model. The specific values reported in that work were an inner radius of 260μm and wall thickness of 7.6μm at TLC, and an inner radius of 150μm and wall thickness of 11μm at FRC + 1L. Using these dimensions, we extrapolated from the (FRC) + 1L lung inflation volume to estimate alveolar dimensions at an FRC of 2L (end-tidal exhalation) using the constant mass assumption. This allowed estimation of lung density and alveolar dimension changes, assuming lung volumes increased by approximately 10% between end-tidal inspiration and expiration [12] and maintaining the constant mass assumption. This alveolar and lung inflation model for spherical dimensions in a typical adult lung was extrapolated to approximate neonatal anatomy at end-expiration, specifically an alveolus was generated with a density of 60% that of an adult [12], here assuming alveolar wall mass is the same as an adult at FRC lung volume.

To extend the simple spherical model, the effect of wall thickness was explored independent of density at the fixed lung inflation corresponding to neonatal end-expiration (FRC). Wall thickness was both increased and decreased by 20%, and then alveolar volume was modified to achieve constant density while conserving the mass of the shell. The dimensions and corresponding densities are provided in Table 2 for all of the alveolar geometries simulated.

Table 2.

Alveolus Simulation Geometries and R₂*

Model	Inner Radius (μm)	Wall Thickness (μm)	Density (tissue/air fraction)	R2*(1/s)
				−7.5ppm	−9ppm
TLC	260.8	7.6	0.083	539.3	696.3
FRC+1L	150.0	11.0	0.191	490.5	648.8
Inspiration	134.8	13.3	0.246	462.5	619.0
Expiration	130.6	14.0	0.264	452.7	608.2
Neonatal Insp.	83.7	27.8	0.577	372.9	478.6
Neonatal Exp.	81.0	28.9	0.600	354.4	459.8
Neonatal Exp. (Thicka)	86.2	30.8	0.600	354.4	459.8
Neonatal Exp. (Thina)	75.2	26.9	0.600	354.4	459.8

^aneonatal expiration wall thickness increased or decreased by 20%, then the volume of the alveolus was adjusted to reach a density of 0.6 while conserving mass of the shell

Human Subject Experiments

MRI was performed on a total of 17 NICU patients with approval of the Institutional Review Board and parental informed consent. Within this cohort, N = 5 had no clinical indication of respiratory disease, referred to as controls (2 male, 3 female; post-menstrual age (PMA) at birth 35.8±4.1 weeks), N = 7 had clinical diagnoses of BPD (5 male, 2 female; PMA at birth 26.2±1.8 weeks), and N = 5 had CDH (5 male, 0 female; PMA at birth 36.8±2.3 weeks). More detailed demographics of the full cohort are provided in Table 1. A 1.5T scanner (gradient amplitude and slew rate equal to 33 mT/m and 120 T/m/s, respectively) designed for neonatal imaging and residing in the NICU [20] was used with a quadrature body coil. Infants were prepared for scanning using hearing protection and the “feed and swaddle” method [21] and were imaged without administration of intravenous contrast agents, sedation or a change in respiratory support. Exam success rate was 100%. Heart rates and SpO₂ levels were monitored throughout the exam. All CDH patients were imaged prior to surgical repair of the hernia.

Table 1.

Neonatal Cohort Demographics

Disease Group	Sex	PMA at Birth (weeks)	PMA at MRI (weeks)	Chronological Age at MRI (weeks)
Control	2 Male, 3	35.8±4.2	38.9±2.2	4.1±2.5
	Female	[29.0–39.0]	[37.3–42.2]	[1.9–8.3]
BPD	5 Male, 2	26.2±1.8	40.3±7.1	14.1±7.1
	Female	[23.0–28.7]	[26.1–48.4]	[0.1–21.4]
CDH	5 Male	36.8±2.3	38.5±2.1	1.7±1.3
		[33.0–39.0]	[35.0–40.6]	[0.4–3.6]

Values shown are mean ± standard deviation, with the range provided in square brackets. (PMA = post-menstrual age)

Image Acquisition

Imaging experiments were performed using a UTE sequence [5] modified to acquire each radial view at multiple echo times. All scans were performed during tidal breathing with self-navigated retrospective respiratory gating [7]. Imaging parameters were: FOV = 180 mm, matrix = 128 × 128 × 128, TR = 7.3 ms, TE_min = 203 μs, bandwidth = 125 kHz, flip angle = 5°, and 50,000 – 70,000 radial projections/TE. Radial views were acquired at 4 TEs (TE = 0.20 ms, 0.95 ms, 1.70 ms, 2.45 ms). Total acquisition time was ~25 minutes.

Image Processing and Analysis

Prior to reconstruction, respiratory motion was corrected by retrospectively gating the image data to end-tidal expiration using a 50% acceptance window based on motion modulations in the magnitude of the central sample of k-space, as previously described [7]. Periods of bulk motion were identified and removed, leading to an average 96% data retention rate (>90% in all cases). Slow motion drift was quantified using the maximum deviation from the minimum (expressed as a percentage of the minimum) of the 10 second moving average of the k-space center after bulk motion removal. Image data were then reconstructed through interpolation onto a Cartesian grid [22] using sampling density compensation weights estimated iteratively [23] to account for the non-uniformly sampled k-space coordinates. Images were then generated (separately for each TE) by inverse Fourier transform of the resulting resampled k-space data.

Following reconstruction, the lungs were semi-automatically segmented from the image with the shortest TE using Fiji, an open-source platform for biological image analysis [24]. In the CDH cohort, the lungs were segmented separately and classified as either ipsilateral (CDHi) or contralateral (CDHc) to the diaphragmatic hernia. R₂* and spin density were then computed voxel-wise within the lung using a mono-exponential decay model as described in [12].

Density Corrected R2*

The relationship between tissue density and R₂* was investigated by classifying voxels within the lungs that fell within equal intervals of proton density normalized to a % of muscle tissue [6,9]. Prior to classification, R2* and proton density images were down-sampled to 1 cm isotropic resolution to increase SNR. Each interval covered a 10% range of normalized proton density, and the entire range (0–100%) was spanned, leading to a total of 10 intervals. The mean R₂* within each interval containing data from all 17 subjects was then compared across cohorts, providing an indication of whether regions of the lung with similar normalized proton densities had similar R₂*, independent of disease.

Within each range of tissue density, classifications of low, normal and high R₂* were determined relative to the density vs. R₂* dependence for the non-diseased cohort [12]. These classifications are useful for identifying whether a particular value of R₂* is either normal, or abnormally high or low, independent of the associated tissue density. We refer to this as the “density corrected” R₂* hereafter. The low/normal/high R₂* cutoffs were calculated by grouping the R₂* from voxels in the non-diseased cohort falling within a particular range of tissue density together and then using k-means [25] to compute low/normal/high R₂* clusters that were then mapped to lung regions.

Statistical Analysis

Whole lung averages of R₂* and normalized tissue density were compared across disease groups. Mean R₂* within each range of normalized tissue density was also compared across groups. Tests for significantly different means across groups were performed using analysis of variance (ANOVA), with subsequent pairwise group comparisons performed using t-tests. A p value <0.05 reflected statistical significance.

Results

Simulation

An example of a slice through the magnetic field distribution in the alveolar model used in the simulations is shown in Figure 1, along with a histogram of the frequency shifts of voxels within the shell and the time evolution of the signal as computed with Equation 1. The estimated exponential decay and the points used to make this estimate are shown as well. Figure 2 presents a scatterplot of density vs. R₂* at each of the simulated alveolar geometries at both Δχ = −9ppm and Δχ = −7.5ppm. The established empirical inverse relationship between proton density and R₂* is well represented by the model at both the upper and lower susceptibility limits. Notably, the R₂* values generated with Δχ = −7.5ppm (the fully deoxygenated condition) correspond very closely with previous empirical measurements [12], which are plotted as separate data points in Figure 2 for reference. It is also noteworthy that alveolar models with equivalent density but different wall thickness and volume produced identical R₂* values, suggesting that R₂* is insensitive to this geometric difference under the assumptions of our model. The R₂* values predicted by the simulation are provided in Table 2 for the modeled geometric dimensions and alveolar densities

Figure 1.

(a) Magnetic field offset distribution (in ppm) in a slice through the spherical shell alveolar model. (b) Histogram of the magnetic field offset for all voxels within the shell. (c) Magnitude of the simulated temporal signal evolution of voxels in the shell (black) and estimated mono-exponential decay with equivalent R₂* (red). The samples used to fit the exponential decay are shown as red stars.

Figure 2.

Simulated R₂* vs. tissue density for a range of alveolar models. Points are labeled with the corresponding physical model. Simulations with Δχ = −9ppm and Δχ = −7.5ppm are shown in blue and red, respectively. Note that points for neonatal inspiration and expiration with thickened walls overlap exactly with the case of normal wall thickness. Empirical data from the literature in healthy human subjects is shown in black for comparison. Dotted lines indicate linear fits to the data of corresponding color.

Human Subject Study

Slow bulk motion drift during image acquisition was found to be minimal; the 10 second moving average of the center of k-space had a deviation of 3.8% ± 2.7% (mean ± standard deviation) after removal of data corrupted by acute bulk motion.

Boxplots of the measured whole lung normalized lung parenchymal density and R₂* within each disease cohort are shown in Figure 3. An ANOVA test rejected the hypothesis of equivalent mean lung parenchymal densities across the cohorts with P<0.05. T-tests for differences between groups indicated significant differences in normalized lung parenchymal density between the ipsilateral lung in CDH (CDHi) and both controls and BPD (both P<0.05). Similarly, the hypothesis of equivalent mean R₂* across the cohorts was rejected (P<0.05). Significant group differences were found between the contralateral lung in CDH (CDHc), but not ipsilateral lung, and both controls (P<0.05) and BPD (P<0.05). BPD patients showed both slightly elevated lung parenchymal density and R₂* relative to controls although this did not reach significance (density, P=0.504; R₂*, P=0.440).

Figure 3.

Boxplots of mean normalized tissue density (left) and R₂* (right) in the lungs of control, BPD and CDH groups. CDH patients are separated into the ipsilateral and contralateral lung. Significant differences between groups (P < 0.05) are indicated by asterisks.

Plots of the distribution of R₂* of each cohort within specific ranges of proton density are provided in Figure 4, and Table 3. This provides insight into the comparison of R₂* standardized to equivalent proton density across the disease groups. The inverse relationship between density and R₂* is maintained within each disease group, but is systematically higher for disease vs. control groups. Specifically, R₂* differences were significant (P<0.05) between controls and CDHc within all tissue density ranges with the exception of the 80–90% range (P=0.17). Additionally, R₂* was significantly different in CDHi relative to controls in the 50–60% and 60–70% (both P<0.05) density ranges. No other differences between groups were statistically significant. However, the qualitative order of R₂* follows a systematic pattern of the control group always having the lowest R₂*, followed by BPD, CDHi and finally CDHc with the highest R₂* for each standardized lung parenchymal density.

Figure 4.

Mean +/− the standard deviation of R₂* within each disease group within specific ranges of tissue density. Individual patient data are shown as filled points. Each tissue density range is 10% wide. Also shown are the linear fit from the simulated model at 7.5ppm and 9ppm. Contralateral CDH lung (CDHc) is significantly higher than controls within all ranges of tissue density besides 80–90%. Note that R₂* in controls < BPD < CDHi < CDHc uniformly.

Table 3.

R₂* Comparisons within Ranges of Normalized Proton Density Across Disease Groups.

Proton Density Range	Mean R2* (s−1)
	Control	BPD	CDHi	CDHc
30–40% a	329 ± 52	400 ± 91 (P=.145)	421 ± 61 (P=.083)	486 ± 98 (P=.006)
40–50%	371 ± 64	440 ± 78 (P=.195)	463 ± 97 (P=.111)	505±108 (P=.025)
50–60% a	360 ± 60	436 ± 78 (P=.120)	469 ± 95 (P=.044)	517 ± 87 (P=.006)
60–70% a	333 ± 66	416 ± 66 (P=.056)	439 ± 63 (P=.027)	504 ± 82 (P=.001)
70–80% a	311 ± 80	373 ± 80 (P=.175)	400 ± 62 (P=.078)	469 ± 76 (P=.004)
80–90%	290 ± 96	342±146 (P=.521)	398±145 (P=.218)	411±137 (P=.170)

Values shown are mean ± standard deviation, and P-values for comparisons between diseased groups and controls are provided in parentheses.

^aANOVA indicates means within the indicated range of proton density are significantly different (P<0.05)

Example regional parametric maps of density corrected R₂* classifications are provided in Figure 5. In the control patient in Figure 5a, density appears relatively uniform, and the R₂* classification map is largely normal with a few small, isolated abnormal regions. Images from two different BPD patients with different patterns of R₂* classifications are shown in Figure 5b and Figure 5c. The first is regionally heterogeneous with a region of low density lung in the right anterior that has a mostly normal R₂*, while the right middle lung has normal, homogenous density but contains a large region of high R₂*. The second BPD patient, in Figure 5c, has heterogeneous tissue density, with nearly the entire lung having abnormally high R₂*.

Figure 5.

Axial slices of tissue density and parametric maps of density normalized R₂* classification in 5 different NICU patients with a variety of disease. White arrows in (b) indicate a region of apparently normal tissue density but abnormally high R₂*. White arrows in (d) show a region of high density lung that has correspondingly low density-corrected R₂*. Note that the left and right lung is ipsilateral to the hernia in (d) and (e), respectively.

Figure 5d and Figure 5e show images from 2 CDH patients. In the first, the left lung is ipsilateral to the herniated diaphragm and shows low proton density. The posterior ipsilateral lung appears to have mostly normal R₂*, while R₂* is high in the anterior. Conversely, the contralateral lung appears to have a large high density region in the posterior with abnormally low R₂* but normal R₂* in the anterior. The second CDH patient, in Figure 5e, has a hernia on the opposite (right) side. The ipsilateral lung is so small in this case that it is almost impossible to see. The contralateral lung is visible and has uniformly high R₂* relative to its normalized proton density.

Discussion

In this work we have presented evidence to suggest that neonatal pulmonary disease, specifically BPD and CDH, is associated with changes in pulmonary R₂* independent of otherwise associated tissue density abnormalities. Empirical data in NICU patients with BPD and CDH indicated the R₂* within the lung is influenced by the disease in ways that are not entirely explained by the relationship between tissue density and lung inflation, suggesting underlying disease processes. For example, average R₂* was uniformly elevated in disease relative to control patients, yet density was higher in BPD and lower in CDH. Comparisons of R₂* within regions with similar tissue density also indicated differences between disease groups, again suggesting that disease processes are affecting R₂* measurements in ways that are not reflected in tissue density changes due to lung inflation. We also presented a method for generating regional parametric maps of density corrected R₂*, or maps indicating whether R₂* is abnormal (either high or low) independent of the density. These maps might be useful for visualizing how regional R₂* varies between the different diseases and between different patients within disease groups. Finally, simulations were performed to investigate whether simple variations in alveolar geometry, specifically alveolar septal wall thickness (differences in which might be associated with disease), influence the R₂*. The results of our model and measurements support the inverse relationship between tissue density and R₂* predicted by normal lung inflation, consistent with similar simulated findings from previous work [29] and empirical results both in this work and elsewhere [12].

A fundamentally important and yet unanswered question concerns the underlying disease-related physiology driving the observed changes in R₂*. While we have explored a specific abnormal alveolar geometry (i.e. wall thickness/volume ratio) without apparent promise, exploring more elaborate changes to the structural geometry may be more fruitful. Geometric influences on the R₂* (at least in the simple spherical shell model used here) are only determined by the density of the tissue, and not by the thickness of the walls. In other words, in our simple model, 2 alveoli with different wall thicknesses but bulk identical densities (i.e. appropriately different volumes) will have identical R₂*. Moreover, differences in the magnetic susceptibility of the lung parenchyma could be driving the observed changes. Deoxygenated blood has a much lower Δχ than tissue (−6ppm vs −9ppm), thus variations in blood oxygenation or deoxygenated blood volume could explain the observations. In fact, the difference in R₂* we observe between controls and CDHc is approximately the same as that derived from the simulations at Δχ = −9ppm and Δχ = −7.5ppm although the absolute R₂* measured is not well predicted by our model. The water self-diffusion as well as the capillary blood flow, the latter modelled as pseudo-diffusion [18], also influenced the temporal signal dynamics, and thus the R₂*. This effect is not accounted for in our model, but may have a non-negligible influence on the measured R₂*. Partial voluming of alveolar structures by different fractions of macrovascular blood and edema in disease may also explain the systematic bias in R₂* between the model prediction and empirical measures. For example, capillary blood volume is approximately 20% of the total pulmonary blood volume, indicating a large percentage of the blood signal arises from macro-vessels [30]. Presumably this signal has a lower R₂* than that coming from spins located within the alveolar structures, driving down the total R₂*.

Study Limitations

First, the patient cohort is relatively small, and lung disease in BPD and CDH tends to vary widely from patient to patient, which similarly restricted our ability to draw confident conclusions about the consistency of the findings. The small sample size was in part due to the long scan time required, limiting our ability to perform the scan in a majority of NICU cases; reducing overall scan time and improving SNR is an area of active research. Associated with this limitation is the poorly matched ratio of male/female patients, which could introduce sex related bias. Second, it is important to note that the tissue density we have used throughout this work is a normalization of parenchymal signal to adjacent muscle signal. This normalization relies on the assumption that the T₁ relaxation in lung parenchyma and muscle tissue is similar [6,31]. However, disease related changes in the T₁ within lung parenchyma could result in biases in the normalized tissue density. In other words, it is possible that T₁ changes could be driving the appearance of a different relationship between R₂* and normalized tissue density. Measuring both T₁ and T₂* in future work would resolve this question. Finally, the simulations provided are designed to be straightforward and to test a specific hypothesis regarding differences in tissue wall thickness and magnetic susceptibilities. The assumptions and simplicity of the current model are unable to capture more elaborate structural abnormalities that might be present in disease, and thus we cannot confidently link underlying structural change to the R₂* differences found.

Conclusion

The inverse relationship between R₂* and tissue density is influenced by the presence of disease globally and regionally in neonates with BPD and CDH in the NICU. For a given tissue density, R₂* is higher, relative to controls, in all disease groups, most notably so in the lung contralateral to the hernia in CDH. It is yet unclear what underlying structural or physiological mechanism is driving these differences, although a number of clinically-relevant phenomena are potentially associated. This technique is a potentially useful tool for probing and evaluating disease processes in NICU patients with pulmonary disease and warrants further investigation.