Three Dimensional Imaging of Ventilation Dynamics in Asthmatics Using Multi-echo Projection Acquisition with Constrained Reconstruction

Abstract

The purpose of this work is to detect dynamic gas trapping in 3-dimensions during forced exhalation at isotropic high spatial resolution and high temporal resolution using hyperpolarized helium-3 (HPHe-3) MRI. Ten subjects underwent HPHe-3 MRI and MDCT. MRI was performed throughout inspiration, breath-hold, and forced expiration. A multi-echo 3D projection acquisition was used to improve data collection efficiency and an iterative constrained reconstruction (I-HYPR) was implemented to improve SNR and increase robustness to motion. Two radiologists evaluated the dynamic MRI and breath-held MDCT data for gas and air trapping respectively. Phantom studies showed the proposed technique significantly improved depiction of moving objects compared to view-sharing methods. Gas trapping was detected using MRI in 5 of the 6 asthmatic subjects who displayed air trapping with MDCT. Locations in disagreement were found to represent small to moderate regions of air trapping. The proposed technique provides whole lung 3D imaging of respiration dynamics at high spatial and temporal resolution and compares well to the current standard, MDCT. While MDCT can provide information about static regional air trapping, it is unable to depict dynamics in a setting more comparable to a spirometry maneuver and explore the longitudinal time evolution of the trapped regions.

Introduction

Obstructive lung disease, including asthma, presents a significant burden on the healthcare system. Further, the etiology of asthma in particular is still an active area of research to develop better therapies and early interventions to reduce the number of individuals that develop this chronic disease. Pulmonary function tests are used extensively in the clinical setting, however these are known to lack regional sensitivity to disease heterogeneity. Measurement of regional lung disease and function has been demonstrated using bronchoscopy [1] and non-invasive imaging techniques [2,3,4]. However, bronchoscopy is inherently limited by the accessible locations and number of samples that may be performed. Imaging techniques have enabled whole lung depiction of regional variations in the lungs. Chest radiography and recently multi-detector CT (MDCT) have been widely used to depict lung structure. In particular, MDCT is widely used as the standard for detection of regional air trapping in the lungs [3]. There is a lack of studies documenting the reproducibility of regional air trapping in asthma. However, asthma is known to be a highly heterogeneous disease [5,6].

Evaluation of air trapping is widely used for longitudinal assessment of broncholitis obliterans syndrome commonly associated with chronic rejection following lung transplant, as well as assessment of therapeutics. Air trapping measured using pulmonary function testing has been shown to be correlated with the severe asthma phenotype [7]. Further, recent work has shown that air trapping depicted using MDCT is a positive predictor for hospitalization due to asthma [8].

Studies of air trapping in asthmatics with MDCT and interventions have shown this technique to provide high sensitivity. Studies looking at air trapping reversibility with methacholine and bronchodilators suggest that regions of trapping were caused by airway hyper-reactivity [9, 10]. However studies looking at treatment with corticosteroids [11] and a leukotriene receptor antagonist [12] have shown reversibility trapped regions suggesting that the trapping may be related to airway inflammation. A method to enable longitudinal evaluation of individual response to therapy may be of great clinical value, particularly in the poorly managed subpopulation of the severe asthma phenotype.

MRI techniques can provide an advantage over CT methods as they do not use ionizing radiation, better enabling longitudinal, interventional, and pediatric exams. Imaging of lung function has presented a challenge for traditional MRI techniques due to the low proton density, high susceptibility, rapid time-scales, limited breath-hold times, and heterogeneity of the lungs. Hyperpolarized gas contrast agents including He-3 and Xe-129 have been used to achieve sufficient signal within the airspaces. Breath-held imaging has primarily been used to achieve sufficient spatial resolution, lung coverage, and SNR. However, this significantly limits the depiction of respiratory dynamics and identification of abnormalities in obstructive lung disease. Cine-type respiratory-gated acquisitions, derived from angiography techniques, have been demonstrated for imaging pre-clinical small animal models, effectively increasing the data acquisition time-window to allow imaging of dynamic processes [13, 14]. However, gas cost and supply limitations currently do not allow for data acquisition over multiple breaths in human subjects.

Non-Cartesian accelerated imaging techniques have been demonstrated for imaging respiratory dynamics in humans during single gas-bolus respiration maneuvers in 2D [15] as well as using interleaved multi-slice techniques [16]. Dynamic imaging of the lungs has been used in the study of obstructive lung diseases using accelerated non-Cartesian 2D acquisitions to detect abnormal physiology including gas uptake and gas trapping [16,17,18]. For the purposes of this work, air trapping will be defined as hypo-lucency on MDCT acquired at functional residual capacity (FRC) that is not visible at total lung capacity (TLC). Gas trapping will be defined as residual signal on dynamic He-3 MRI that is not cleared following a forced exhalation to RV.

Previously we have demonstrated a three-phase single acquisition that performs 2D PR imaging during the dynamic processes of gas inhalation and exhalation, as well as a 3D stack of stars acquisition during a short breath-hold performed between [18]. The technique enabled depiction of variations in respiratory dynamics that could be readily compared to the static ventilation images obtained during the same maneuver. This earlier work was the first publication of the detection of gas trapping using dynamic HPHe-3 MR with confirmation using MDCT. More recently we have demonstrated the use of a 3D multi-echo vastly under-sampled isotropic projection acquisition (ME-VIPR) for accelerating data acquisition in HPHe-3 lung imaging [19]. This ME-VIPR acquisition provided sufficient data collection efficiency to allow imaging of respiratory dynamics and retrospectively accommodate lost patient breath-holds with whole lung coverage using angular under-sampling and view sharing techniques.

Images reconstructed using standard reconstruction methods with under-sampled data result in wrap artifacts for conventional Cartesian acquisitions and streak artifacts for under-sampled projection acquisitions. To address these artifacts, methods have been developed including view-sharing [20], keyhole [21,22], TRICKS [23], Tornado filtering [24], and k-t BLAST [25]. However, these reconstruction techniques result in temporal blurring at some or all spatial frequencies. The HighlY constrained back PRojection (HYPR) reconstruction technique exploits spatial-temporal correlations to improve SNR for accelerated contrast enhanced imaging [26] including dynamic ventilation studies [27]. The HYPR technique was developed for sparse data where the SNR and spatial resolution are determined using a composite image and the temporal data is determined by a set of under-sampled projections. An iterative modification of the HYPR algorithm (I-HYPR) has been demonstrated to accommodate less sparse data sets including perfusion [28], ADC [29] and PO₂ mapping [30]. I-HYPR has been show to be more robust to motion during the composite [31]. The I-HYPR technique exploits the expectation maximization algorithm developed by Shepp and Vardi for emission tomography [32].

The work presented here combines the improved data collection efficiency and under-sampling afforded by ME-VIPR with the I-HYPR reconstruction algorithm to improve SNR for the study of 3D ventilation dynamics and gas trapping in asthma. Phantom studies were performed to quantify the performance of the technique in the presence of motion. A case study, from a subject that was later diagnosed with a pulmonary artery aneurysm, is presented to demonstrate correspondence of dynamic HPHe-3 with MDCT in a disease setting that is spatially reproducible. Finally, air and gas trapping studies were performed in a small sample of asthmatic subjects using the dynamic HPHe-3 and MDCT techniques to demonstrate the technical feasibility of the methods.

Methods

MR imaging studies were performed on a 1.5 T clinical scanner (Signa HDx, GE HealthCare, Waukesha, WI) using an excite/receive coil tuned to 48 MHz (Medical Advances). The pulse sequence was the ME-VIPR acquisition described previously [19]. Specific acquisition parameters included a cubic 42 cm FOV, ± 125 kHz receive band width, ~1° constant RF flip angle, TR of 4.4 ms, first TE of 0.22 ms, 8 half-echoes acquired during each TR, using ramp sampling to increase efficiency, and a maximum radial distance from the center in k-space corresponding to 64 evenly spaced points on a Cartesian grid.

A modification was made to the k-space trajectory described in Holmes et al. [19] as the new technique changes sampling angles at the outer edge of k-space, with no change while passing through the center of k-space. This multi-echo trajectory was used to allow sampling of 5 unique angles within each TR with 3 of these being full projections spanning the full diameter of k-space to better accommodate the geometric assumptions required by the I-HYPR reconstruction. Unique interleaved angle sets were acquired every 0.1 s, each set covering the full sphere of k-space. This angle scheme allows cine-type reconstruction to readily accommodate variable patient breath-hold times.

Dynamic I-HYPR images were reconstructed using a composite image constructed with a radially varying temporal filter (tornado filtering) [8] with data from a 10 s window making up the outer regions of k-space and the center made up of 8 s of data. To compensate for under-sampling at the outer regions of k-space in the composite, the composite images were also reconstructed using a radially varying density compensating filter [33]. I-HYPR time-frame data were composed of 1 s of data (220 TRs or 660 sampled angles) and 6 iterations were performed to achieve the final reconstructed images.

Phantom Studies

Dynamic imaging of a gel phantom was performed to determine the robustness of the technique to motion. The phantom measured 6.5 cm by 9 cm with the short axis oriented along the direction of motion. Motion was controlled using a motion stage with a total range of 24 cm and for speeds from 0 to 3.1 cm/s along the z-axis of the MRI scanner. Following image reconstruction, 1-D profiles were measured across the image, along the direction of motion. The root mean square (RMS) difference between consecutive I-HYPR iterations was calculated by equation 1. The RMS difference was then used as a stopping criteria for the reconstruction algorithm to determine the number of iterations to perform. Six iterations were sufficient to reduce the percent change in the RMS difference between subsequent iterations to below 5%, even for the highest velocity studied (3.1 cm/s).

$$\text{Percent}\text{change}\text{in}\text{RMS}\text{difference}=\frac{\sqrt{{\displaystyle \sum _{r=1}^N}{\left(P_{i+1,r}-P_{i,r}\right)}^2}}{\sqrt{{\displaystyle \sum _{r=1}^N}{P}_{i,r}^2}}\ast 100$$ [1]

where P_i,r is the r^th element of the N total element profile measured, measured across the image reconstructed for i-iterations of the I-HYPR algorithm.

To assess the accuracy of the technique in the presence of motion, the full width at half of the maximum (FWHM) and full width at tenth of the maximum (FWTM) of the 1-D profiles were measured at the center of the range of motion for images at each velocity. To estimate the blurring within a single 1 s time frame with sufficient SNR, motion was simulated by convolving a stationary fully sampled image of the phantom with a rectangle equal to the distance traveled in 1 s. The differences between the FWHM and FWTM of the 1 s time-frame simulation and the proposed I-HYPR reconstruction were measured.

Volunteer Studies

Ten subjects (8 females, 2 males) underwent dynamic He-3 MRI and MDCT for analysis of air and gas trapping. Subjects included 8 asthmatics, 1 pulmonary artery aneurysm, and 1 normal with an overall mean age of 27.7 years (maximum of 46 and minimum of 19). Subjects are part of the ongoing Severe Asthma Research Program (SARP) study [34]. Asthma severity was determined based on the American Thoracic Society Workshop on Refractory Asthma [35].

Lung function measures of spirometry and plethysmography have been described elsewhere [7]. Briefly, subjects withheld bronchodilator medications for an appropriate length of time to avoid interference with the spirometry and lung volumes measurements, unless required to manage asthma symptoms. Lung volumes were measured with the Boyle’s Law method in a constant volume body plethysmograph (Jaeger Masterscreen, VIASYS Healthcare, Yorba Linda, CA), using a pant rate of <1 Hz during the mouthpiece occlusion, which was activated after the subject had attained a stable end-expiratory volume for at least 4 breaths; after the brief occlusion, subjects exhaled slowly and maximally to residual lung volume (RV), and then inhaled maximally to total lung capacity (TLC). RV determined during the slow exhalation maneuver is designated as slowRV for this work [36]. Spirometry (Jaeger, VIASYS Healthcare, Yorba Linda, CA) was conducted according to ATS/ERS guidelines [37], recording the best forced expiratory volume in one second (FEV₁) and the best forced vital capacity (FVC) obtained with 3 acceptable measurements. The FEV₁ was expressed as a percent of the predicted value (FEV₁%pred) [38]. FVC was subtracted from TLC to obtain the dynamic air trapping during a forced expiratory maneuver (fastRV) given by equation [2].

$$\text{fastRV}=\text{TLC}-\text{FVC}$$ [2]

Both the RV obtained during slow expiration (slowRV) and the fastRV were expressed as a fraction of TLC as physiological indicators of air trapping. FEV₁ was measured on the day of the imaging and body plethysmography was performed within 1–2 weeks of the imaging. Plethysmography was unavailable for one subject (I in Table 1) due to inconsistent subject efforts.

Table 1.

Summary of spirometry, plethysmography, MRI and MDCT results.

Subject	Clinical Diagnosis	FEV1 %pred	slow- RV/TLC	fast- RV/TLC	Ventilation Defect Score	MR Ventilation Defect Locations	Trapping indicated by MRI	MR Trapping Locations	Trapping indicated by CT	CT Trapping Locations
A	Normal	86.5	0.28	0.19	11	LUL small lesions	no	none	no	none
						RUL small lesions
						RML small lesions
B	Pulmonary aneurysm	87.6	0.28	0.30	4	RML small lesions	yes	RUL whole lobe	yes	RUL whole lobe
						RLL small lesions				RML to RLL small lesions
										LLL very small lesion
C	Mild- moderate asthma	105.1	0.25	0.27	1	RLL small lesion	no	none	no	none
D	Mild- moderate asthma	74.8	0.30	0.26	20	LUL small lesions	no	none	no	none
						LLL small lesions
						RUL small lesions
						RML small lesion
						RLL small lesions
E	Mild- moderate asthma	92.8	0.26	0.27	0	none	yes	LUL moderate lesion	yes
										RLL small lesion
										LLL small lesion
F	Mild- moderate asthma	84.3	0.36	0.34	9	LUL small lesions	yes	LLL moderate lesion	yes
						RUL small lesions
						RML small lesions				RML moderate lesion
						RLL small lesions		RLL small lesion		RLL very small lesion
G	Mild- moderate asthma	108	0.19	0.19	9	LUL small lesions	no	none	yes	LLL 1 cm lesions
						RUL small lesion
						RML small lesions
						RLL small lesions				RLL 1 cm lesions
H	Severe asthma	115	0.29	0.44	13	LUL small lesions	yes	LLL small lesion	yes	LLL small 1cm lesion
						LLL small lesions		RLL small lesion		RLL moderate lesion ~3.5 cm
						RUL small lesion
						RML small lesions
						RLL small lesions
I	Mild- moderate asthma	104	not available	not available	1	RML small lesion	no	none	no	none
J	Mild- moderate asthma	93.8	0.27	0.30	29	LUL small lesions	yes	LUL small lesion	yes	LUL 2 small lesions
						LLL small lesions		LLL large lesion		LLL very large lesion
						RUL small regions				RUL-superior moderate lesion, less intense
						RML small regions
						RLL small regions		RLL small lesion		RLL small lesion
								RLL-base small lesion		RLL-base moderate lesion

Abbreviations:

RUL right upper lobe

RML right middle lobe

RLL right lower lobe

LLL left lower lobe

LUL left lower lobe

HPHe-3 gas was prepared to a polarization level of ~28% using a commercial polarizer (GE Healthcare, Waukesha, WI). The HPHe-3 gas was then mixed with nitrogen to produce a dose of ~4.5 mM polarized nuclei for an inhaled volume of 14% of the subject’s TLC. Dynamic HPHe-3 MRI was performed during subject respiration including inspiration (~2–6 s), breath-hold (~15 s), forced expiration to RV, and was followed by tidal breathing for a total time of 40–60 s. Prior to the HPHe-3 acquisition, subjects performed a practice respiration maneuver during which a dynamic proton scan was acquired. The subjects’ maneuvers were verbally coached during the acquisition; however the timing was individually adjusted to accommodate subjects that were unable to complete longer breath-holds.

Six iterations of the I-HYPR algorithm were performed for image reconstruction of the volunteer studies, based on the percent change in the RMS difference between subsequent iterations of the maximum velocity (3.1 cm/s) studied in the phantom experiment described above. Quantitative analysis of the dynamic images was performed using custom software written in MATLAB (Mathworks, Natick, MA) with ROIs manually placed in regions of the lungs and trachea to quantify signal kinetics during the respiration maneuver. Kinetic data measured in the parenchymal gas space were corrected for signal decay due to RF and T₁ measured simultaneous to the dynamic acquisition during the breath-hold phase on a voxel by voxel basis. Specifically, the signal modulation during the breath-hold was assumed to be due to a constant rate of RF and T₁, therefore data during this period were used to fit K in the following decay model given by equation [3].

$$S_n=S_0·K^n$$ [3]

S_n is the signal in the nth image, S₀ is the signal in the initial image, and K represents the decay due to RF and T₁ modeled by equation [4].

$$K=cos(\alpha )·exp(-TR/T_1)$$ [4]

where TR is the TR time-rate, and α is the RF flip angle. Equation 2 was then inverted to remove the modulation from RF and T₁ on the kinetic data to better depict signal changes due to ventilation only. Kinetic analysis was not performed near the diaphragm and large airways due to bulk motion effects.

MDCT imaging was performed using GE LightSpeed 64 or 16 slice scanners (GE Healthcare, Waukesha, WI) with 1.25 mm detector width and 0.58 – 0.78 mm × 0.58 – 0.78 mm in-plane spatial resolution. Imaging parameters included 120 kV and 50 mAs. Images were acquired during separate breath-holds of 4–6 s at TLC and FRC to depict regions of potential air trapping. Breath-holds were performed at FRC as this is conventionally used in the clinical setting.

Two radiologists (CJF and TAB) evaluated the time-resolved MR images for evidence of gas trapping by consensus and blinded to MDCT data. The presence of gas trapping was determined by HPHe-3 signal that did not significantly decrease immediately following the forced exhalation maneuver. The time-scale for signal change during the rapid exhalation maneuver is much faster than the slow evolution of pO2 and RF, effectively separating these processes [18]. The forced exhalation maneuver thereby differentiates these dynamic processes in addition to the physiological advantage of inducing dynamic airway closure in obstructive lung disease as is commonly used in pulmonary function testing. Kinetic analysis of the signal evolution was also used in more complex cases.

Following evaluation of the MR images, the radiologists read the MDCT data by consensus. The size of gas trapping regions were categorized based on the diameter of the region as follows: small ≤ 2 cm, 2 cm < moderate ≤ 4 cm, large > 4 cm. The HPHe-3 data acquired during the breath-hold were evaluated using a weighted whole-lung defect score [39]. Each lobe was evaluated for percent of ventilation defects and assigned a score as follows: 0 = no defects, 1 = >0–25%, 2 = >25–50%, 3 = >50–75%, and 4 = > 75–100%. The whole lung score was then calculated as the sum of the weighted scores for each of the lobes. Spatial overlap between the HPHe-3 and MDCT data was evaluated with the two image volumes displayed side-by-side. Testing for statistical significance between the means of subjects displaying trapping on MRI and those showing no trapping was performed using the 2-sided and 2-sample with unequal variance student’s T-test.

Results

Phantom Studies

When using I-HYPR, the reconstructed phantom position was qualitatively correct for all velocities studied. A single slice of the image volume of the stationary phantom is shown in Figure 1a. An example is shown for the phantom study with a velocity of 2.7 cm/s demonstrating the results from the I-HYPR reconstruction (Fig. 1b, top row), the 1 s time-frame data (Fig. 1b, middle row), and the 10 s tornado filtered reconstruction (Fig. 1b, bottom row) used as the initial composite for the I-HYPR reconstruction. Improvement in the depiction of the true phantom location and size is evident for the I-HYPR images compared to the 10 s tornado filtered reconstruction. The data from the 1 s time-frame shows significant streak artifacts due to angular undersampling, while the noise and artifact are evident in the 10 s view-sharing reconstruction due to inconsistent projection data. Quantitative comparison of the I-HYPR images demonstrated decrease of RMS difference for the profiles across the moving phantom for up to 20 iterations. FWHM measures of the phantom profile using I-HYPR were within +/−1 cm of the expected value (Figure 2a), FWTM were found to show blurring on the order of 2.3 cm for 0.6 cm/s and 4.2 cm for 2.7 cm/s (Figure 2b). For comparison, view sharing methods depict a 2.5 cm and 9.4 cm FWTM blurring, respectively. The increase in FWHM of the view sharing technique at velocities greater than 1.5 cm/s is due to the distance of travel in 5 s being greater than the total width of the phantom (6.5 cm). Note that the level of streak artifact present in the iterative result is reduced compared to both the 1 s time-frame and 10 s tornado filtered images.

Figure 1.

a) Single slice of the image volume depicting the stationary phantom. b) Motion study demonstrating improvement of depiction of motion using the iterative HYPR method at a velocity of 2.7 cm/s. Note significant motion averaging is still evident in the view sharing reconstruction, however by the 6^th iteration the depiction of the object is improved. Dotted bars indicate limits of phantom motion during the 5 s of data depicted.

Figure 2.

a) Phantom results depicting measured difference between FWHM in reconstructed data and simulated time-frame data. Improvement in object depiction using I-HYPR over view sharing is shown with increasing object speed. b) Difference in FWTM with increasing object speed. Improvement with the use of iterations is greatest for higher velocities.

The FWHM was well represented using six iterations of the HYPR algorithm with relatively little change in the width with subsequent iterations. The artifact level represented by the FWTM was found to decrease asymptotically with iteration number. However, even with significant iteration numbers the FWHM of the measured and simulated data did not converge to the same width, i.e. the 1–2 cm difference was maintained. These phantom results demonstrate the applicability of the I-HYPR algorithm to situations involving subject motion.

Volunteer Studies

General imaging results are summarized in Table 1 and will be presented first, followed by a presentation of two case studies with significant air and gas trapping detected on MRI and MDCT. Dynamic studies of human subjects using the ME-VIPR acquisition and I-HYPR reconstruction showed good depiction of inhalation and breath-hold. Coronal MIPs from a typical HPHe-3 data set are shown in Figure 3a depicting filling of the large airways followed by parenchymal enhancement. Individual 3.28 mm slices from the 3D volumes at a 1 s time-resolution taken during expiration demonstrate upward diaphragm motion and compression of the chest wall during gas clearance from the lungs with only moderate spatial blurring during peak motion (Figure 3b arrows). Axial slices of the MRI data are shown during the breath-hold and expiration phase depicting less intense gas trapping (Figure 3c and d, arrows). In this subject, MDCT images at TLC and FRC were not found to show air trapping (Figure 3e and f).

Figure 3.

a) Subset of maximum intensity projection images of time-frames showing inspiration (2.5 s –5.5 s), breath-hold (white boxes, 10 s and 13 s), and expiration (16 s – 20 s) for a subject with mild-moderate asthma (subject E). b) Subset of frames during dynamic expiration with a single 3.3 mm slice of each 3D image volume shown. Upward motion of the diaphragm (arrows, 18.5 s) and compression of the chest wall (arrows, 19.5 s) are visible. Reformatted axial image slices showing the breath-hold and expiration depict a region of slower gas clearance c) and d) (arrows), that is not visualized in the corresponding MDCT images e) and f). Note time-frames have been individually windowed and leveled to better depict regional variations in signal.

A summary of the results of the radiologists’ readings is given in Table 1. Small ventilation defects were observed in all subjects except subject E. The diagnosis of air trapping was established using MDCT in six of the ten subjects. Gas trapping was detected in five of these same six subjects using MRI, with significant trapping found in two subjects (one asthmatic and one pulmonary artery aneurysm). Of the 5 subjects that displayed air and gas trapping on both MRI and MDCT, 47 % of the locations of trapping were detected using both modalities. The regions in disagreement were small to moderate regions of trapping on MDCT or MRI (size of trapped regions < 4 cm). Only one subject was found to exhibit small regions of air trapping on MDCT with no trapping detected using MRI (size of trapped regions ≤ 1 cm). Focal regions of abnormally rapid gas uptake were observed during inhalation in five of the subjects on MRI (Subjects B,D,E, I, and J). In this study, regions of ventilation defects were not found to correspond to regions of air or gas trapping on MDCT or MRI in all cases except subject J. For subject J, the extensive and complex patterns of trapping and ventilation defects were found to lead to small regions of spatial overlap. HPHe-3 ventilation score was also not found to correlate with air or gas trapping.

The whole lung measures of FEV1_%pred, slowRV/TLC, and fastRV/TLC are given in Table 1. The mean values of fastRV/TLC were significantly different for subjects displaying gas trapping on MRI (mean = 0.34) from those showing no trapping (mean = 0.23, p = 0.05), however the means were not significant for FEV1%pred (mean for trapping = 80%, mean for non-trapping = 94%, p = 0.16) or slowRV/TLC (mean for trapping = 0.30, mean for non-trapping = 0.25, p = 0.27). The subject diagnosed with a pulmonary artery aneurysm and severe gas air trapping (Subject B) was not included for this analysis to focus on asthma.

Case: Pulmonary Artery Aneurysm

Subject B was initially enrolled in the study as a normal, however the subject was later diagnosed with a pulmonary artery aneurysm following bronchoscopy and a contrast-enhanced MDCT exam (Figure 4a, arrows). HPHe-3 imaging was performed prior to the bronchoscopy and MDCT assessment. Following the clinical evaluation, the pulmonary artery aneurysm was confirmed to be located in the right upper lobe. Imaging studies for air and gas trapping were carried out prior to the bronchoscopy and the clinical MDCT exam. This pulmonary artery aneurysm case provides an opportunity to evaluate the imaging performance of the MRI technique against MDCT in a case of spatially reproducible airway obstruction between the separate imaging exams. MDCT imaging for evaluation of air trapping depicted the aneurysm impinging on the main bronchi down of the right upper lobe. MDCT images of the breath-holds at TLC and FRC depicted hypo-lucency in the right upper lobe below the bronchial branch point with the posterior-superior upper lobe segment 2 (RB2) at both lung volumes suggesting decreased perfusion to this area (Figure 4f black arrow). The mass effect of the pulmonary artery aneurysm resulted in narrowing of the main airway past this branching point. The location of the aneurysm was found to cause no narrowing of the right upper lobe segment 2 branch (RB2). The right upper lobe segment 2 bronchi (RB2) was found to be accessible and the segment displayed normal opacity suggesting normal perfusion (Figure 4b – d white arrows).

Figure 4.

a) Axial image from a contrast enhanced MDCT exam of the subject with a pulmonary artery aneurysm (subject B). The arterial aneurysm is visible (solid arrow) beside a pulmonary vein (dotted arrow). b) – g) Image slices from the MDCT exam for evaluation of gas trapping depict the mass effect of the aneurysm on the bronchi resulting in narrowing and closure (black arrow) however the bronchus feeding the second segment remains open (white arrows).

On dynamic HPHe-3 MRI, this subject showed abnormal signal dynamics with rapid signal increase during inhalation in segment 2 of the upper right lobe (Figure 5a, arrow at 6 s), followed by significant trapping on MRI corresponding to the upper right lobe visualized in coronal MIPs (Figure 5a, arrow at 24.5 s). A reconstruction using only 1 s of data for each time-frame shows significant streaking due to angular undersampling and low SNR during the later time-frames (Figure 5b). The lobe boundaries can be observed in coronal slices from the 3D image volumes depicting the expiratory phase (Figure 5c arrow). The abnormal findings on MDCT and bronchoscopy were found to be in good spatial agreement with those on dynamic He-3 MRI (Figure 6a–d). Analysis of the He-3 signal kinetics showed a trapezoidal pattern in the left upper lobe due to the inhalation, breath-hold, and forced exhalation (Figure 6e green line). The contra-lateral right upper lobe displayed delayed filling reflected in the delayed time to peak signal (Figure 6e, yellow line). Further, elevated residual signal was evident in the upper right lobe following the forced expiration maneuver. A region of high gas uptake on HPHe-3 imaging was found to correspond to the open second segment of the upper right lobe fed by RB2, that did not appear as hypo-lucent on the MDCT (Figure 6e, magenta line).

Figure 5.

a) Subset of time-frames during dynamic ventilation showing early signal enhancement during inspiration (arrow, 6 s) and significant trapping in the upper right lobe in subject B diagnosed with a pulmonary artery aneurysm (arrow, 24.5 s). b) The same subset of time-frames depicted in a), however only data from the surrounding 1 s has been used for each frame. Note significant streaking due to angular undersampling evident in the early time-frames (4.5 s to 7.5 s) and low SNR throughout the exam. c) Subset of frames during dynamic expiration exhibiting gas trapping in the upper right lobe. Note 3.3 mm slices of the 3D image volumes enable depiction of the lobe boundary during dynamic expiration (arrow 23.5 s). Time-frames have been individually windowed and leveled to better depict regional variations in signal.

Figure 6.

Comparison of HPHe-3 MRI with MDCT showing abnormal ventilation during the breath-hold (a, white arrow) and hypo-perfusion on MDCT (c, black arrow) in the subject with a pulmonary artery aneurysm (subject B) shown in Figure 5. During the forced exhalation, gas trapping is evident in MRI (b, white arrow) and air trapping is confirmed in MDCT during breath-hold at FRC (d, black arrow). Locations of ROIs are shown for kinetic analysis (a). Plots of signal time-course for the right upper lobe (yellow) compared with left upper lung (green) in the same pulmonary artery aneurysm case. Hyper-intense signal on HPHe-3 was found to correspond to the 2^nd segment that was not blocked by the aneurysm on MDCT (green). Note delayed filling as evident by the later time-to-peak signal enhancement relative to the expected trapezoidal shaped enhancement curve in the contra-lateral left lung region.

Case: Asthma

The second case was a subject with mild to moderate asthma. An example of significant gas trapping in asthma is demonstrated between the breath-hold to expiration He-3 dynamic images for Subject J (Figure 7a, b). Ventilation defects are also evident during the breath-hold phase (Figure 7a arrows). Further, regions of gas trapping on dynamic He-3 MRI (figure 7b arrows) are correlated with regions of significant air trapping detected on MDCT (figure 7d arrows) performed 2 hours prior to the MRI session. However, several less intense or smaller regions of air trapping were detected on MDCT that were not visible on the dynamic He-3 data. Small regions of spatial overlap were observed in this subject with extensive and complex patterns of trapping and ventilation defects.

Figure 7.

HPHe-3 MRI with MDCT showing ventilation defects during the breath-hold (a, white arrows) in a subject with mild-moderate asthma (subject J). During the forced exhalation, residual signal due to gas trapping is evident in MRI (b, white arrows) and air trapping is confirmed in MDCT during breath-hold at FRC (d, black arrows).

Discussion and Conclusions

Dynamic HPHe-3 MRI results were shown for volunteer studies of gas trapping in asthmatic, a subject with a pulmonary artery aneurysm, and normal subjects and compared to MDCT. Trapping was detected with both MRI and MDCT in 5 of the 10 subjects studied and 4 of the subjects were found to show no gas trapping using both methods. One subject was found to have small regions of gas air trapping on MDCT while no air trapping was observed on MRI. However, there were also moderate to small regions of gas trapping that were not found to over-lap in the 5 subjects that displayed trapping on both MDCT and MRI. Comparison of the imaging results with the spirometry and plethysmography highlight the advantage of imaging methods for improved sensitivity to regional variations in lung function. The conventional plethysmography measure of passive air trapping (slowRV/TLC) was not found to provide a significant separation between subjects displaying trapping on MRI and those with no trapping. However, the dynamic air trapping metric designated by fastRV/TLC (Eq. 2) was significantly different between the two groups. Previous breath-held HPHe-3 ventilation imaging has found correlations between the imaging markers of ventilation defects per slice [40] and defect score [41] with FEV1%pred. The measure of FEV1%pred includes contributions from both airflow limitation and air trapping. Subsequently, FEV1%pred is not necessarily expected to correlate with gas trapping detected using dynamic HPHe-3 MRI due to the increased physiologic noise of the spirometry measure. This is particularly evident in the pulmonary artery aneurysm case where the spirometry FEV1%pred appears relatively normal while the imaging depicts significant abnormalities in the lung function in the entire lobe downstream from the aneurysm. Four of the subjects were also found to display focal regions of early filling during the inspiration period. These results are consistent with previous studies that have demonstrated heterogeneous filling patterns in asthmatics [16].

Imaging results were consistent between MDCT and dynamic HPHe-3 MRI for the pulmonary artery aneurysm case where a non-transient structural abnormality was present. The regions of air and gas trapping detected using both modalities on the remaining subjects may represent areas of structural remodeling of the lungs. While MDCT was used to help validate the detection of gas trapping on MRI, the authors recognize that the imaging sessions were performed at different times, during different exams, using different ventilation maneuvers, and at different lung volumes. In particular, it should be noted that the MRI was performed with exhalations down to RV while the MDCT was performed during breath-holds at FRC. This HPHe-3 imaging technique, like spirometry, relies on reproducible subject efforts. The regions of gas and air trapping that were in disagreement between dynamic HPHe-3 MRI and MDCT were also found to be near the resolution limits of the dynamic HPHe-3 exam. Further, small regions of air trapping have been observed on MDCT in normal subjects [42] and may be of less clinical significance. Given these differences between the exams, it is interesting to note the level of correspondence between the dynamic HPHe-3 MRI and MDCT results. The authors are aware of no reproducibility studies of gas air trapping in asthmatics using MDCT. Studies with bronchodilators [9,10] found no change in CT air trapping suggesting the trapping is not due to airway hyper-responsiveness. However, studies using corticosteroids [11] and a leukotriene receptor antagonist [12] have shown reversibility of air trapping suggesting this is due to airway inflammation. Winkler et al. [43] have suggested that spontaneous airway closure in asthma may lead to temporally varying ventilation patterns. De Lang et al. [44] have shown locations of ventilation defects on He-3 MRI in asthma have a reproducibility of 41% suggesting there may be structural remodeling leading to more repeatable abnormalities in addition to the less repeatable spontaneous airway closure.

The MR imaging methods for capturing respiratory dynamics were further validated in moving phantom studies showing good representation of the object FWHM with significant improvement over view-sharing techniques. The FWTM measures also showed significant improvement over view-sharing however there is still evidence of some low signal ghosting remaining. A more accurate depiction of the moving phantom was provided by suppressing the data inconsistency using the iterative reconstruction, compared to the view-sharing tornado filtered data that was used as the initial composite image. Mistretta et al. showed that the SNR of images reconstructed with HYPR depended on the composite image. In the present work, however, the composite image may contain artifacts due to data inconsistency caused by motion. The iterative algorithm was observed to reduce these artifacts, providing additional benefit. However, further study is needed to extend the formalism of Mistretta et al. to the iterative algorithm in the presence of motion.

The data acquired during inspiration for the subject with the pulmonary artery aneurysm shows abnormalities on both inspiration and expiration. It is evident that the upper right lobe is, in general, filling on inspiration in both HPHe-3 and MDCT. The focal region of increased gas uptake, visible on HPHe-3, is consistent with segment 2 of the lobe that is located upstream and not blocked by the aneurysm. The He-3 signal in the segments downstream from the aneurysm show slow filling that is visible after the breath-hold begins. One possible explanation may be the “Pendelluft” effect, or redistribution of gas within the lungs once the flow of gas has stopped at the end of inhalation [45, 46]. This could occur due to negative pleural pressure being created during the inspiration leading to an increase in the aneurysm size as blood is pulled into the thoracic cavity, reducing flow of He-3 gas beyond the aneurysm. The breath-hold at FRC+14% of TLC will result in positive pleural pressure and cause a decrease in the aneurysm blood volume. Aided by the decrease in the aneurysm volume, He-3 gas can re-distribute to achieve equilibrium within the lungs. However the cause of these abnormalities cannot be verified at this time.

While MDCT and dynamic HPHe-3 MRI can provide static information about regional air trapping comparable to regionally measuring the RV of the lungs, dynamic HPHe-3 is able to depict dynamics in a setting more comparable to a FEV₁ spirometry maneuver to detect the dynamic components of obstruction. Further, the specific dynamic MRI acquisition shown in this work provides flexibility to use varying amounts of data acquired during the subject’s breath-hold to improve image quality. In the pulmonary artery aneurysm case, MDCT was unable to distinguish between hyper-inflation at TLC with air trapping at FRC and reduced perfusion in the images at both TLC and FRC. However, He-3 MRI was able to detect filling of the lobe beyond the obstruction as well as depict the presence of gas trapping upon forced exhalation.

Tzeng et al. demonstrated the use of rapid 2DPR imaging for depiction of the large airways to enable assessment of the airway dimensions [47]. The 3D dynamic imaging method presented here allows improved visualization of the complex structures of the large airways and ventilation during breath-hold and dynamic gas trapping. This can provide a significant advantage in planning guided bronchoscopy for assessment of regional differences.

In the current study we show preliminarily results in a sample of 10 volunteers, however further studies in a larger population will be performed to better evaluate the sensitivity and specificity of the method. In addition, studies in normal subjects need to be performed to establish normal regional gas kinetic behavior during these rapid ventilation maneuvers. Future work will include study of the longitudinal reproducibility of dynamic HPHe-3 MRI for assessment of regions of structural remodeling in mechanistic studies of asthma including interventions with airway bronchodilators. Further, quantification of regional gas uptake [48] may be a metric for obstructive lung disease and 3D depiction is likely to improve sensitivity to regional variations. Future work will consist of including a pulmonary perfusion acquisition with the dynamic HPHe-3 acquisition in a single exam to improve characterization of ventilation/perfusion.

This technique enables the acquisition of a wealth of information on inflow and exhalation kinetics, breath-hold ventilation defects, as well as readily accommodating the individual patient’s breath-hold capabilities all within a single comprehensive maneuver. Whole-lung 3D imaging of respiration dynamics has been demonstrated using a combination of an accelerated data acquisition (ME-VIPR) and a constrained reconstruction (I-HYPR) to improve the depiction of lung function. This technique provides the ability to detect regional differences in gas kinetics in 3D at both high spatial and temporal resolution. The demonstrated reconstruction method allows significant improvements in representing dynamic MRI data in 3 dimensions. This work provides significant advantages in eliminating the need for breath-hold coaching during the exam, and will be part of an on-going study of childhood origins of asthma.