The spatial-temporal dynamics of pulmonary blood flow are altered in pulmonary arterial hypertension

Abstract

Global fluctuation dispersion (FDglobal), a spatial-temporal metric derived from serial images of the pulmonary perfusion obtained with MRI-arterial spin labeling, describes temporal fluctuations in the spatial distribution of perfusion. In healthy subjects, FDglobal is increased by hyperoxia, hypoxia, and inhaled nitric oxide. We evaluated patients with pulmonary arterial hypertension (PAH, 4F, aged 47 ± 15, mean pulmonary artery pressure 48 ± 7 mmHg) and healthy controls (CON, 7F, aged 47 ± 12) to test the hypothesis that FDglobal is increased in PAH. Images were acquired at ∼4–5 s intervals during voluntary respiratory gating, inspected for quality, registered using a deformable registration algorithm, and normalized. Spatial relative dispersion (RD = SD/mean) and the percent of the lung image with no measurable perfusion signal (%NMP) were also assessed. FDglobal was significantly increased in PAH (PAH = 0.40 ± 0.17, CON = 0.17 ± 0.02, P = 0.006, a 135% increase) with no overlap in values between the two groups, consistent with altered vascular regulation. Both spatial RD and %NMP were also markedly greater in PAH vs. CON (PAH RD = 1.46 ± 0.24, CON = 0.90 ± 0.10, P = 0.0004; PAH NMP = 13.4 ± 6.1%; CON = 2.3 ± 1.4%, P = 0.001 respectively) consistent with vascular remodeling resulting in poorly perfused regions of lung and increased spatial heterogeneity. The difference in FDglobal between normal subjects and patients with PAH in this small cohort suggests that spatial-temporal imaging of perfusion may be useful in the evaluation of patients with PAH. Since this MR imaging technique uses no injected contrast agents and has no ionizing radiation it may be suitable for use in diverse patient populations.

NEW & NOTEWORTHY Using proton MRI-arterial spin labeling to obtain serial images of pulmonary perfusion, we show that global fluctuation dispersion (FDglobal), a metric of temporal fluctuations in the spatial distribution of perfusion, was significantly increased in female patients with pulmonary arterial hypertension (PAH) compared with healthy controls. This potentially indicates pulmonary vascular dysregulation. Dynamic measures using proton MRI may provide new tools for evaluating individuals at risk of PAH or for monitoring therapy in patients with PAH.

INTRODUCTION

Pulmonary arterial hypertension (PAH; WSPH Group I Pulmonary Hypertension) arises from progressive pulmonary vascular remodeling that increases pulmonary vascular resistance and thus pulmonary arterial pressure. PAH is defined by a mean pulmonary arterial pressure ≥20 and pulmonary arterial wedge pressure ≤15 mmHg, with pulmonary vascular resistance >2 Wood units (1) measured by right heart catheterization. Right heart failure arises secondarily, ultimately leading to transplant or death (2, 3). Although medical treatment of PAH has advanced, diagnosis of PAH is often delayed, and the overall prognosis remains poor, especially for those with more advanced disease at diagnosis (3–5). Thus, the improved ability to screen at-risk patients and monitor response to therapy is highly desirable (6).

Although the primary defect in PAH resides in the pulmonary circulation, the use of lung imaging to assess PAH is limited and while several technologies are in development (7) imaging has been primarily used to exclude other pathologies rather than to diagnose or monitor PAH (for review, see Ref. 8). Chest radiographs and high-resolution CT scanning are used to rule out parenchymal lung disease (WSPH Group III pulmonary hypertension) and ventilation-perfusion scans to rule out chronic thromboembolic pulmonary hypertension (WSPH Group IV pulmonary hypertension) (8). Functional lung imaging has the potential to provide increased sensitivity to detect early pathology as well as changes in function with treatment. This is because of the ability to detect locally impaired function that may be compensated for by normal lung and thus obscured in nonimaging assessments. Arterial spin labeling (ASL) is a functional magnetic resonance imaging technique that uses magnetic field gradients and radiofrequency pulses to magnetically “tag” and image moving protons in blood, allowing imaging of pulmonary perfusion. One metric derived from pulmonary ASL is the relative dispersion (coefficient of variation, standard deviation of the signal intensity divided by the mean signal intensity). We refer to it here as the spatial relative dispersion to distinguish it from an analogous measure derived from temporal data (9). Spatial relative dispersion is a simple metric of spatial heterogeneity that may be a reflection of, but is not specific for, underlying vascular remodeling. We have previously shown that in patients with PAH, the spatial relative dispersion of perfusion is increased compared with control subjects (10) and the differences were not explained by altered gravitational gradients in perfusion occurring from the increase in pulmonary arterial pressure in PAH. Relative dispersion is also increased by a variety of factors including healthy aging (11), and the consequences of head-down tilt (12) or heavy exercise (13, 14).

The pulmonary circulation is a dynamically regulated system with moment-to-moment changes in vascular smooth muscle activity potentially affecting local perfusion (15). A simple analogy is that the flow through the pulmonary microcirculation in a healthy lung twinkles like stars in the sky, as capillary segments are recruited and de-recruited. Since ASL does not rely on the use of ionizing radiation or injected contrast, it provides the ability to make multiple serial measures of the perfusion distribution at the same phase of the cardiac cycle. In essence, one can create a movie of a portion of the pulmonary circulation where the individual frames of the movie represent single-slice images of the volume of blood delivered to the pulmonary capillaries (i.e., perfusion) in one systolic ejection period. This provides an opportunity to evaluate the dynamics of the pulmonary circulation, that may uniquely reflect underlying pulmonary vascular function.

One metric derived from this dynamic imaging is global fluctuation dispersion, FDglobal (9). FDglobal (described in methods) isolates heterogeneity due to temporal fluctuations in local perfusion, while controlling for changes in overall flow (i.e., cardiac output). FDglobal is increased by regional shifts in blood flow distribution, and/or temporal instability in flow. In healthy subjects, we have previously shown that FDglobal is affected by factors known to affect pulmonary vascular regulation: it is increased by breathing hypoxic or hyperoxic gas (9), as well as by inhaled nitric oxide (16). In addition, FDglobal is increased even when static spatial measures of heterogeneity such as the spatial relative dispersion are unchanged (9). Thus, dynamic imaging may provide the ability to monitor or detect pulmonary vascular disease in PAH, but its utility is unknown.

As a first step toward evaluating spatial-temporal measures of pulmonary blood flow as a tool to aid in diagnosis and monitoring of patients with pulmonary vascular disease, we evaluated the FDglobal in a small population of female patients with PAH and compared them to a group of healthy nonsmoking female controls. We hypothesized that FDglobal would be increased in patients compared with control subjects, consistent with disrupted pulmonary vascular regulation. We also evaluated the spatial relative dispersion in these subjects as well as the percent of the lung image where there is no measurable perfusion signal. We hypothesized that consistent with our previous work these would both be increased in PAH compared with controls.

METHODS

Ethics Approval

The study was approved by the Institutional Review Board at the University of California, San Diego (UCSD, IRB No. 121529). Patients with PAH were recruited from those patients attending UCSD for care who had given permission to be contacted for participation in research. All patients with PAH were ambulatory and receiving standard-of-care therapy; none were receiving parenteral prostanoid therapy. Control subjects were recruited by advertisement and were nonsmoking, had no previously diagnosed history of cardiovascular or pulmonary disease verified by medical history, and had normal spirometry. All subjects had the risks of the study explained to them by an investigator, provided written informed consent, and were screened using an MRI safety questionnaire before participation in the study.

Protocol Overview

Before entering the scanner, subjects were trained in voluntary respiratory gating for the acquisition of the perfusion time series. This was accomplished by playing a recording of the ASL acquisition, which provided audible cues, and coaching the subject to briefly (∼1 s) suspend respiration at functional residual capacity (FRC) beginning just before and ending just after each image acquisition. The timing of the imaging is such that this is close to a normal breathing pattern. The subjects lay supine in a 1.5 T Signa HDx TwinSpeed MRI system (General Electric Medical Systems, Milwaukee, WI), wearing an eight-channel torso coil and an MRI-compatible pulse oximeter. MRI-compatible ECG electrodes (InVivo ECG Quadtrodes) were placed on the left chest to enable cardiac gating of the ASL images from the R wave of ECG derived QRS complex. Reference silicone phantoms were placed on the right chest under the top elements of the torso coil within the field of view to ensure the phantoms and lung were subject to the same coil loading as the lung. We acquired data from a sagittal slice in the right lung, which was positioned in the mid-clavicular line to capture the maximum anterior-posterior diameter of the lung and to avoid potential motion artifacts from the aorta and heart in the left hemithorax.

Lung Density and Coil Sensitivity Profile

We used proton density (M₀) measurements to correct ASL perfusion images for signal heterogeneity in the images introduced by proximity to the coil elements. A multi-echo fast gradient echo (mGRE) sequence uses multiple single echo acquisitions within a single breath-hold and obtains images alternating between two echo times in a single 9-s breath-hold at FRC: six with an echo time of 1.1 ms and six at an echo time of 1.8 ms. Imaging sequence parameters were TR = 10 ms, flip angle = 10°, slice thickness = 15 mm, field of view = 40 cm, receiver bandwidth = 125 kHz, and a full acquisition matrix of 64 × 64. Proton density was determined by back extrapolating the signal to a TE = 0 by fitting data to a single exponential. A correction factor was empirically determined for the static sequence parameters (TR = 10 ms, T₂ decay effects negligible) permitting the mean phantom signal to be used as a reference for water content. The details of this technique have been published (17), and it showed high reproducibility and validity with excellent agreement with gravimetric measures of lung water (18). Image acquisitions were repeated using the torso coil, which along with the density data from the body coil was used to construct a coil sensitivity profile for each subject as previously described (19) (see Image Processing).

Pulmonary Perfusion

Pulmonary perfusion data were acquired using a two-dimensional (2-D) ASL sequence known as FAIRER (Flow sensitive alternating inversion recovery) (20, 21) with a half-Fourier acquisition single-shot turbo spin-echo (HASTE) imaging scheme. This ASL technique is quantitative (22, 23), reproducible (r = 0.95 between repeated measurements), and validated in a flow phantom (23). Briefly, ASL provides an image of the distribution of perfusion derived from the subtraction of two images in which the signal of blood delivered to the imaged slice is manipulated in different ways, leaving one image with delivered blood generating a strong signal (a selective inversion “blood bright” image) and one in which the signal is largely nulled (a nonselective inversion “blood dark” images). The two images differ only in the way that tagging radio frequency pulses change the signal of blood flowing into the imaged section while keeping the signal from the tissue itself unchanged. The signal in a voxel of the subtracted image is proportional to the amount of blood delivered during the previous cardiac cycle (19, 22).

This basic ASL-FAIRER sequence was modified to acquire a series of cardiac gated perfusion images allowing the spatial-temporal behavior of pulmonary perfusion to be assessed (9, 16, 24). The temporal acquisition protocol was similar to previous publications (9), such that each image (selective or nonselective inversion) is acquired at the same point in the cardiac cycle, with a delay time equal to 80% of the R-R interval and 4 R-R intervals in between acquisitions. The sampling rate was doubled from our previously reported methods, when alternating selective/nonselective pairs were collected (9, 16, 24), by collecting 10 nonselective inversion images followed by 300 selective inversion images and finishing with 10 nonselective inversion images, exploiting the fact that differences between the nonselective inversion images under a given condition are within measurement noise (9). This resulted in a total of 320 images acquired breath by breath (4–5 s intervals), with a total acquisition time of 25–30 min. The acquisition interval was adjusted to ensure normal tidal breathing, allowing the subjects to maintain a natural lung volume (FRC) at the end of each breath. Images were collected using a 256 × 128 acquisition matrix reconstructed to 256 × 256, a 40-cm field of view, and 15-mm slice thickness. The intrinsic image resolution was 1.56 mm × 3.12 mm × 15 mm which was then reconstructed into voxels 1.56 mm × 1.56 mm × 15 mm.

Image Processing

All image analysis was performed using MATLAB (MathWorks, Natick, MA). To correct for heterogeneity in perfusion images occurring based on proximity to the coil elements a coil sensitivity profile for each subject was applied as previously described (19). Images were registered based on the lung margins using a deformable image registration algorithm using sub-voxel interpolation, as described in Ref. 25 and available at https://github.com/UCSDPulmonaryImaging/Deforminator. Images that had more than a 10% change in lung image area from FRC (25) were discarded, leaving an average of 305 images (95% of images) per subject. Images were visually inspected for artifacts related to mistiming of the cardiac gating (evidenced as reduced enhancement in large vessels and/or linear streaking artifacts) and affected images (average 4/subject) were also discarded from the time series. The images were masked to leave only the lung field and used in the calculation of the FDglobal, reflective of spatial temporal heterogeneity, and other spatial metrics as described later.

Spatial-Temporal Heterogeneity: Calculation of the FDglobal Metric

An outline of the processing pipeline is shown in Fig. 1. For the calculation of FDglobal, the stack of ASL difference images was created by subtraction of an average nonselective inversion image constructed from all the acceptable (up to 20/subject, mean 17 images) nonselective inversion images from the individual selective inversion images as previously described (16). The result was up to 300 difference ASL images (some nonselective inversion images failed the quality control steps) with each image reflective of the total blood delivered to the imaged slice in one cardiac cycle. Signal from the large conduit vessels, which does not represent perfusion was masked out using a previously established threshold-based approach that utilizes the statistical distribution of image intensities (26). Images were smoothed while maintaining the conduit vessel mask boundaries with a Gaussian kernel (FWHM of ∼11 mm), creating an in-plane spatial resolution of ∼1 cm³ minimizing the impact of any residual registration errors. Each image was normalized by its mean to eliminate the confounding effect of signal variations from changes in cardiac stroke volume. As in previous publications (9, 16, 24), a baseline ASL perfusion image was constructed by averaging the first 20 difference images that fulfilled the quality control criteria into a composite. This baseline perfusion image was subsequently subtracted from the remaining perfusion images in the time series resulting in a series of images characterizing fluctuations in perfusion from the initial composite baseline image. The FDglobal is defined as the spatial standard deviation of the flow fluctuation image (9) and was calculated for each image and the results averaged as in the study by Asadi et al. (9). FDglobal represents temporal variability occurring away from the baseline image due to short-term fluctuations at the image acquisition rate (i.e., at 4–5 s intervals), plus variance due to any redistribution of flow across the lung slice at the time scale of the acquisition (20–30 min).

Figure 1.

Processing pipeline for the calculation of global fluctuation dispersion (FDglobal). 1) During voluntary respiratory gating, 320 cardiac gated images are acquired: 10 nonselective inversion images, followed by 300 selective inversion images, finishing with another 10 nonselective inversion images. 2) Images are inspected for quality, segmented, registered, and corrected for heterogeneity based on proximity to coil elements. An average nonselective inversion image is constructed for each subject from all the acceptable nonselective inversion images. 3) Difference images representing blood delivered in one cardiac cycle (perfusion) are obtained by subtracting the average nonselective image from the individual selective inversion images. Signal from the large conduit vessels is masked and each image is smoothed and normalized. 4) A baseline perfusion image is constructed from the average of the first 20 difference images. 5) The baseline perfusion image is subtracted from the remaining differences images in the time series giving images representing fluctuations in perfusion compared with the baseline image. 6) The FDglobal is calculated as the average of the spatial standard deviation for each flow fluctuation image/map.

Spatial Heterogeneity: Calculation of Spatial Relative Dispersion and Regions of No Measurable Perfusion Signal

We have previously shown that spatial heterogeneity of perfusion as measured by the spatial relative dispersion is increased in patients with PAH (10) compared with healthy controls, as was the amount of the lung image with no measurable perfusion signal (27). To obtain data for comparison with those prior studies, which were obtained from subtracted blood bright/blood dark pairs of images obtained during single breath-holds, we used the registered time series images as follows: We constructed two sets of average nonselective inversion images, one from the first nonselective inversion images in the series (i.e., images 1–10) and one from the last nonselective inversion images (images 311–320). The ten adjacent selective inversion images for each nonselective inversion were subtracted, i.e., images 11–20 that passed quality control were subtracted from the first composite nonselective inversion image, and similarly, images 301–310 were subtracted from the last composite nonselective inversion images. The result was up to 20 full-resolution (0.036 cm³) difference images. These were used to calculate spatial relative dispersion (coefficient of variation = SD/mean) on each image which, consistent with previous work (10, 19) included signals from large vessels. The results were averaged. Spatial relative dispersion was expressed as both the absolute value and as a percent of the values predicted based on the subject’s age and height as Spatial Relative Dispersion = 0.007 × age (years) + 0.009 × height (cm) – 0.809, as reported by Levin et al. (11). The amount of the lung image with no measurable perfusion signal was conservatively identified as the percent of voxels within the lung region of interest in which the measured blood flow was zero or negative, i.e., not different from noise, and the results similarly averaged.

Statistical Analysis

All data are expressed as means ± SD. Data were compared using Student’s t test. A two-tailed α threshold of 0.05 was used for all significance testing.

RESULTS

A total of 11 female subjects (4 patients with PAH, 7 control subjects), reflecting the strong female preponderance of PAH, participated in the study. Subject group demographics and spirometry are shown in Table 1, physiological data during imaging in Table 2, and individual patient with PAH data in Table 3. The time since the initial diagnosis of PAH varied from a little over one year to ten years in the patient population. All patients with PAH were receiving therapy consisting of an endothelin receptor antagonist and three of the four were receiving a phosphodiesterase-5 (PDE5) inhibitor. One subject also was receiving inhalation therapy with a prostacyclin analog. Despite therapy, the mean pulmonary arterial pressure as assessed by direct pressure measurements and documented in the clinical record in the PAH group was significantly elevated and more than double the threshold for the diagnosis of pulmonary hypertension at 48 ± 7 mmHg. Pulmonary vascular resistance was also abnormally high at 13.8 ± 2.8 Wood Units.

Table 1.

Subject group demographic and pulmonary function data

	PAH, n = 4	CON, n = 7	P
Age, yr	47	47	0.98
±SD	15	12
Height, cm	157	164	0.17
±SD	8	7
Weight, kg	59	78	0.10
±SD	9	19
PAP, mmHg#	48	*
±SD	7
PVR, Wood Units#	14	*
±SD	3
FVC, L	2.54	3.40	0.03
±SD	0.54	0.56
FVC, % Predicted	78	95	0.10
±SD	16	14
FEV1, L	1.84	2.63	0.02
±SD	0.50	0.44
FEV1, % Predicted	71	91	0.08
±SD	22	14
FEV1/FVC	0.71	0.77	0.10
±SD	0.06	0.06
FEV1/FVC, % Predicted	89	96	0.13
±SD	9	5

Data are means ± standard deviations. FEV₁, forced expiratory volume in 1 s; FVC, forced vital capacity; PAH, pulmonary arterial hypertension; PAP, pulmonary arterial pressure; PVR, pulmonary vascular resistance.

#Data from clinical record; *not measured.

Table 2.

Changes in oxygen saturation, heart rate, and mean signal (reflecting stroke volume) over the course of imaging from the first 20 images used in the calculation of FDglobal (Start) to the last 20 images (End)

	PAH, n = 4		Controls, n = 7
	Start	End	Start	End	P (Group × Time)
$\mathrm{S}{\mathrm{p}}_{{\mathrm{O}}_2}$, %	95.0	95.4	97.9	97.4	0.16
±SD	2.4	3.1	1.8	1.7
Heart rate, beats/min	60	58	70	69	0.33
±SD	3	7	12	12
Mean signal, AU	15	18	22	24	0.85
±SD	5	5	4	4

FDglobal, global fluctuation dispersion; PAH, pulmonary arterial hypertension; $\mathrm{S}{\mathrm{p}}_{{\mathrm{O}}_2}$, oxygen saturation.

Table 3.

Individual PAH patient demographics and clinical data

Patient ID	PAH-A	PAH-B	PAH-C	PAH-D
PAH etiology	Idiopathic	Scleroderma	Toxin	Idiopathic
Time since diagnosis, yr	1.3	5.0	5.1	10.0
PAH medications	1	1,2	1,2	1,2,3
Mean PAP, mmHg	46.6	40.1	50.0	56.2
Mean PVR, Wood Units	12.4	10.9	17.3	14.6
FDglobal	0.36	0.27	0.32	0.66
No measurable perfusion, % of Image	8.0	13.8	10.0	21.8
Spatial relative dispersion	1.17	1.56	1.38	1.74
Spatial relative dispersion, % pred	120	136	170	217

PAH Medications 1: endothelin receptor antagonist, 2: phosphodiesterase-5 (PDE5) inhibitor, 3: inhaled prostacyclin analog. FDglobal, global fluctuation dispersion; PAH, pulmonary arterial hypertension; PAP, pulmonary arterial pressure; PVR, pulmonary vascular resistance.

There were no significant differences between PAH and control groups for age, height, or weight. Forced vital capacity, forced expiratory volume in 1 s was significantly less in patients with PAH than controls. The changes in oxygen saturation, mean heart rate, and mean image signal intensity during image acquisition were small and did not differ between PAH and controls (Table 2).

There were no significant differences between subject groups for the number of images lost due to deviation in lung volume from FRC (P = 0.07) or to gating artifacts (P = 0.2). The area change required to register the images remaining after quality control was very small, and similar between groups (1.91 ± 2.27% PAH, 0.91 ± 1.05% controls, P = 0.4).

Data for spatial and spatial-temporal metrics are shown in Table 3 (patient’s individual data with PAH) and Table 4 (Group mean data) and individual data are shown in Figs. 2 and 3. In the control subjects, the values for FDglobal (mean 0.17 ± 0.02) were similar to that previously reported for healthy subjects (9, 16, 24). FDglobal was significantly increased in the patients with PAH (mean 0.40 ± 0.17, P = 0.006) compared with controls, with no overlap in values between the two groups. In the control subjects, the amount of the lung image with no measurable perfusion (mean 2.3 ± 1.4%) and spatial relative dispersion (mean 0.90 ± 0.10) were also similar to that previously reported for healthy subjects (11). The percent of the lung image with no measurable perfusion was markedly greater in PAH compared with control subjects (13.4 ± 6.1%, P = 0.001) as was spatial relative dispersion (1.46 ± 0.24, P = 0.0004) and percent predicted spatial relative dispersion (161 ± 43% p = 0.002), consistent with vascular remodeling resulting in poorly perfused regions of lung and increased spatial heterogeneity.

Table 4.

Group mean imaging data

	PAH, n = 4	CON, n = 7	P
FDglobal	0.40	0.17	0.006
±SD	0.17	0.02
No measurable perfusion	13.4	2.3	0.001
(% of image) ±SD	6.1	1.4
Spatial relative dispersion	1.46	0.90	0.0004
±SD	0.24	0.10
Spatial relative dispersion*	161	91	0.002
(% Predicted) ±SD	43	8

CON, healthy controls; FDglobal, global fluctuation dispersion; PAH, pulmonary arterial hypertension.

^*Predicted spatial relative dispersion of perfusion regression equation from Levin et al. (11): Spatial relative dispersion = 0.007 × age (yr) + 0.009 × height (cm) − 0.809.

Figure 2.

Individual subject data for the spatial-temporal metric, global fluctuation dispersion (FDglobal), in patients with pulmonary arterial hypertension (PAH) and control subjects. Gray diamonds are subject means ± SD. FDglobal is the average spatial standard deviation of flow fluctuation maps representing the changes in local perfusion over time. FDglobal is markedly increased in PAH, particularly in one subject who had the longest history (10 yr) of the disease.

Figure 3.

Individual subject data for spatial metrics in patients with pulmonary arterial hypertension (PAH) and control subjects. Gray triangles are subject means ± SD. The percent of the lung region of interest that has no measurable perfusion signal was markedly increased in PAH (A). Spatial relative dispersion (B) the SD/mean of signal intensity was also markedly increased in PAH.

There was a tendency for all metrics to track with the duration of the disease and subject PAH-4 with a ten-year history had the highest values for FDglobal, spatial relative dispersion, and percent of lung image with no measurable perfusion although the small number of subjects precluded statistical analysis. Similarly, these metrics tended to track with pulmonary arterial pressure.

DISCUSSION

The pulmonary circulation is a low resistance high capacitance system, built to withstand the stressors of active pre-20th century life, and is overbuilt for modern sedentary lifestyles. Moderate pulmonary arterial pressures are maintained in healthy people in the face of a variety of stressors, such as hypoxia and exercise (28). However, in PAH, vascular remodeling (4) of the precapillary arteries increases resistance, ultimately leading to increased pulmonary arterial pressures (4, 29). PAH is distinct from pulmonary hypertension arising from left heart failure (WSPH Group 2), hypoxemic lung disease/chronic hypoxic exposure (WSPH Group 3), chronic thromboembolic disease (WSPH Group 4), or other miscellaneous causes (2, 3).

In a small cohort of subjects, we show that spatial-temporal heterogeneity of pulmonary perfusion as measured by FDglobal was increased in female patients with PAH, consistent with pulmonary vascular dysregulation. In addition, consistent with our previous work (10), the spatial heterogeneity of pulmonary blood flow as measured by the spatial relative dispersion was also increased in PAH compared with control subjects, as was the amount of the lung image with no measurable perfusion signal. These data suggest that it may be useful to further evaluate spatial and spatial-temporal dynamic imaging of pulmonary circulation with proton MRI as tools for evaluating individuals at risk of PAH or for monitoring therapy in patients with PAH undergoing treatment.

Spatial-Temporal Metric: FDglobal

In the present study, FDglobal was significantly greater in those with PAH compared with control subjects of similar age and sex, with values that averaged 135% of the control subjects and reaching almost 300% of the control subjects in the subject with the longest disease history. FDglobal is a measure of how regional pulmonary perfusion varies over time. Images are normalized and the initial baseline 20 image average provides a reference image of the average spatial distribution, which is subtracted from subsequent images removing any baseline spatial variance. If the blood flow distribution was subsequently unchanged from this initial average, subtraction would result in images with essentially zero signal. Thus, any spatial variance that persists after the baseline subtraction arises because of regional variations in flow (either positive or negative) away from this baseline. This may occur because of twinkling-like fluctuations, or because of larger-scale quasistatic flow redistribution (9, 16, 24).

Previous work has shown that FDglobal is increased by several factors that are expected to affect the regulation of the pulmonary circulation. For example, both increased ($\text{F}{\text{I}}_{{\text{O}}_2}$ = 0.30) and decreased ($\text{F}{\text{I}}_{{\text{O}}_2}$ = 0.125) inspired oxygen concentrations increase FDglobal (9). In addition, 40 ppm inhaled nitric oxide increased FDglobal during hypoxia (16); the increases with inhaled nitric oxide in normoxia and hyperoxia were smaller and did not reach statistical significance (16). Why PAH, hypoxia, hyperoxia, and inhaled nitric oxide increase FDglobal is unknown but may be related to a loss of tight control of perfusion: hyperoxia and nitric oxide acting to vasodilate the pulmonary circulation reducing regulation and PAH and hypoxia acting to increase pulmonary arterial pressure, increasing the regulatory forces required to maintain homeostasis. Irrespective of the mechanism, the net result is a slow recovery from any moment-to-moment perturbations.

An analogy to consider is the dynamic equilibrium of a self-driving car on a road. How much the car fluctuates in its trajectory is a function of the adequacy of the autopilot function and the condition of the road. In the healthy lung, ventilation-perfusion matching is largely passive and accomplished by airway/vascular branching structure and the influence of gravity, however, there still is a small amount of active regulation (reviewed in Ref. 30). Any small changes in posture, depth of inspiration, pulmonary arterial pressure may change local ventilation-perfusion ratio and thus alveolar Po₂, which acts to activate or deactivate local control, changing local perfusion (30–32). The healthy lung makes adjustments continuously, but they are small because passive matching predominates. In the car analogy, the road is well paved and straight and the autopilot is in good condition, smoothly adjusting course with only minor fluctuations.

In healthy circulation, regulatory mechanisms are in relatively good condition even when challenged (i.e., by inhaled nitric oxide, hyperoxia, or hypoxia), but fluctuations increase, either because there is less active regulation as with NO or hyperoxia (i.e., the autopilot responsiveness is reduced) or there is more disruption to the system as with hypoxia (the road is bumpier). In PAH, patients contend with both a need for increased regulation from increased pressure as well as limited ability to do, so leading to increased fluctuations. In car terms, the patients with PAH contend with a very bumpy and winding road and as well as a poor autopilot, with the net result that the dynamic equilibrium is disturbed and the path fluctuates greatly. In PAH, if therapy reverses some temporal instabilities, this is expected to reduce FDglobal. Conversely, if a particular therapy is not optimal, FDglobal is expected to increase from previous measurements and FDglobal may therefore be useful to optimize or monitor response to treatment. This remains unexplored. Whether FDglobal is altered in other disease states or is specific to the pulmonary circulation is also presently unknown.

Spatial Metrics

In the present study, we found increased spatial heterogeneity as reflected by an increase in spatial relative dispersion and the percent of the lung image with no measurable perfusion, consistent with our previous work (10). Heterogeneity in lung perfusion arises by a number of physiological and pathophysiological mechanisms. For example, perfusion in the lung is not spatially uniform but varies as a result of vascular branching structure (33), local vascular resistances, and because of the effects of gravity both on the distribution of blood flow and on the distribution of lung tissue (for review, see Ref. 30). In addition, local pathology such as in the case of PAH is expected to disrupt local perfusion increasing heterogeneity.

Lung with no measurable perfusion.

In healthy subjects, typically less than 5% of the lung imaged with ASL contain voxels with zero or negative values which acts as a quality control metric (19). In the control subjects from the present study less than 3% of voxels fell into this category, which is consistent with our previous work, and good data quality. Since negative perfusion is not possible, these represent regions where signal characteristics fall below the noise floor, thus leading to negative numbers when two regions of noise are subtracted (19). In normal subjects these regions arise from several sources: The first of these is because blood vessels do not cross the lobar boundaries, so voxels in the region of the fissures will have no perfusion signal. This is also true of voxels within larger airways present in the image. When subtracted these areas of very low/no signal, not different from noise, may return zero or negative values. Second, the inversion pulse used to create the selective inversion image in the ASL experiment is not a perfect square wave function, it is applied to the imaged slice with a small overlap into the adjacent slices. Although this ensures that all the blood in the selective inversion image plane has been inverted, the magnetization of arterial blood protons in the adjacent “gap” have been only partially inverted. The result is that a small amount of blood delivered to the imaged section is not fully labeled, resulting in an underestimation of true perfusion (19, 26), and leading to regions where signal is not measurable. Finally, small misregistrations between tag and control pairs may create small subtraction errors that also may contribute to these regions.

We found significant increases in the amount of the lung image with no measurable signal in PAH. This is unlikely to be solely because of technical differences in image acquisition between the groups. There were no significant differences (P = 0.36) between groups in the amount of registration required and thus increased misregistration resulting in more subtraction error in PAH is unlikely. However, mean cardiac output is typically reduced in PAH (4) and in keeping with this in the present study mean perfusion was significantly less (P < 0.0003) in PAH compared with controls. Thus, we cannot rule out the possibility that the lower overall perfusion in part caused a greater number of lung regions to fall below the noise floor of our measurements as a partial explanation.

However, the remodeling in PAH is patchy (34), and as pulmonary vessels succumb this potentially affects the spatial distribution of perfusion. Studies using dual-energy CT (35–37) and SPECT (38) to evaluate patients with PAH report patchy defects in contrast distribution and the amount of affected lung correlates with disease severity (38, 39). Data from the present study is consistent with this previous work, with a greater percent of the affected lung with low/no perfusion areas in PAH compared with controls. Also, these regions tended to be greater in the PAH subjects with a longer disease history and higher pulmonary arterial pressures. For example, over 20% of the lung image was affected in the subject diagnosed with PAH over 10 years ago. However, given the small number of subjects care must be taken not to overinterpret these data.

Spatial relative dispersion.

As expected from the study selection criteria, the control subjects in the present study had values for spatial relative dispersion that were within normal limits (11). The results of this study are also consistent our previous work showing that spatial relative dispersion is greater in patients with PAH (10). Spatial relative dispersion is increased with healthy aging (11), at a rate of ∼5%/decade of age above age 20 with a moderate positive correlation between age and spatial relative dispersion of R = 0.48. Data from the control subjects in the present study also show a positive relationship between age and spatial relative dispersion (R = 0.74). We have previously shown that spatial relative dispersion is also positively correlated with height, presumably on the basis of lung size (11). Although the regression equations for spatial relative dispersion are based on a smaller number of subjects than normative values for spirometry, the association (R² = 0.44) is relatively strong. For this reason, we reported our results both as absolute values and as a percentage of predicted, based on our previously published regression equations (11), similar to reporting of values for spirometry (40, 41).

Comparison of Spatial and Spatial-Temporal Metrics

The spatial relative dispersion metric has many appealing features: data are relatively quick to acquire, and the metric is conceptually simple, easy to calculate, and highly reproducible (11). It is a useful metric to quantify pulmonary blood flow heterogeneity that, like FDglobal, is altered by multiple factors affecting pulmonary circulation. Spatial relative dispersion is elevated in patients who have had Fontan procedure (10), possibly also because of remodeling of the pulmonary circulation. It is increased with other interventions such as following heavy exercise (13, 14), with interstitial edema induced by rapid infusion of saline (42) and head-down tilt (12), and by mild hypoxia in subjects who have previously experienced high altitude pulmonary edema (43).

The measurement of FDglobal has a longer acquisition time, is more difficult to conceptualize, and is more complex to calculate. It is important to note that any regional fluctuation in perfusion results in a positive value for FDglobal. Since the standard deviation of the mean normalized perfusion fluctuation distribution is used, any deviation away from the initial baseline increases FDglobal, irrespective of whether there are increases or decreases in local flow. It cannot be negative; larger values denote more variability away from the baseline images. In addition, some of the spatial variance such as occurring from static vascular structure is removed by subtracting the initial baseline average, and any remaining spatial variance results from fluctuations in pulmonary blood flow. However, the spatial relative dispersion contains and may be dominated by static spatial variance, and this may act to obscure changes arising from temporal fluctuations. For these reasons, FDglobal may have increased sensitivity compared with spatial relative dispersion to detect differences because of the way that it is calculated. This is supported by previous work in normal subjects, hypoxia significantly increased FDglobal by almost 100% whereas in those same subjects, the change in spatial relative dispersion was small (<10%) and not significant, consistent with other studies (43, 44) evaluating the effects of hypoxia. In addition, hyperoxia increased FDglobal by ∼50% but spatial relative dispersion remained unchanged (9, 16). Consequently, the measurement of FDglobal may offer some advantages that justify this more complex measure, but this is unexplored.

Limitations

There are several limitations to this study that should be considered. First and foremost, the study was conducted with a small number of subjects and care must be taken not to overinterpret the data. The subject cohort was all-female, and there is a significant biological bias in PAH prevalence among women compared with men. Although we have enough data from prior studies in normal subjects to be reasonably confident that the controls are a representative population with respect to both spatial and spatial-temporal metrics, and the results are highly significant statistically, there remains the possibility that the patients with PAH are not representative, that the results might be specific to women with PAH, and that the findings will not be sustained in a larger study.

For robustness, we increased the number of images acquired during scanning by altering the imaging acquisition from previous work (9, 16, 24): we obtained 10 nonselective inversion images, then 300 selective inversion images, followed by a final 10 nonselective inversion images instead of alternating selective and nonselective inversions. We bracketed our selective inversion acquisitions with nonselective inversion acquisitions and used both sets to construct the average nonselective inversion for subtraction. This provided a control for any potential changes in physiological conditions over the course of the measurement. However, a potential concern may arise that group differences for changes in heart rate and cardiac output during imaging might affect FDglobal. This concern is minimized by the following: first, because the signal in the nonselective inversion arises largely from signal from static tissue with T1 recovery from the inversion pulse, this will be unaffected by changes in heart rate or cardiac output. Second, any blood flow signal in the nonselective image is very low because the timing of the acquisition is such that we image close to null point of blood, so any changes in bulk flow would have minimal effect. Third, the average change in heart rate between the beginning and the end of the acquisition was ∼2 beats/min across all subjects. This change was not statistically significant (P = 0.22) and biologically trivial. In addition, the change in mean signal per image (reflecting stroke volume) also did not significantly change (P = 0.19). It should also be noted that we normalize all the subtracted images to control for any potential changes in bulk flow.

Similarly, the findings are unlikely to be because of differing effects between groups of the breathing pattern required for image acquisition. Each acquisition takes ∼800 ms and is separated by 4–5 s. We coach the subjects to breathe in synchrony with the scanner, choosing an acquisition frequency close to normal breathing. This way, the subject inspires normally after an acquisition and returns to functional residual capacity in time for the next image acquisition, pausing momentarily as they would during normal breathing. Thus, disruption to breathing pattern is minimal.

In the calculation of FDglobal, large conduit vessels are removed applying a threshold and because of the slice thickness of 1.5 cm, this approach may miss some partial volume signal. However, it is important to recognize large vessels are completely full of tagged blood and thus reflect blood volume, and do not change signal in response to changes in flow. For this reason, even though these elements may not be completely removed in the individual images, FDglobal is not sensitive to their presence (9), because they do not change from the initial 20 image baseline. We cannot rule out the possibility of a time-dependent change in local blood volume contributing to some of the FDglobal signal, but this contribution, if any, is expected to be small because these voxels occupy a small portion of the total image. Our current protocols in assessing FDglobal are limited to a single representative slice of the lung and multiple-slice acquisition is not currently possible. Thus, there is a possibility that the finding might be different if whole lung imaging was possible. However, given the diffuse nature of the pathology of PAH and the large difference between PAH and controls in the present study, this is unlikely.

The approach we took to calculating the spatial metrics differed from our previous work in that we took the results from subtracted images that were obtained during respiratory gating and not during a single breath-hold. We did this to streamline the acquisition with a view to translation in the future. However, inaccurate registration between tag and control pairs has the potential to cause subtraction errors which would be expected to increase both spatial relative dispersion and the amount of the image with undetectable perfusion. There are several reasons why this is unlikely. First, we average the result from several images to ensure robust results. Second, in the control subjects we also obtained breath-hold images and the spatial relative dispersion averaged 0.91 ± 0.12 compared with 0.90 ± 0.10 with the registration approach. Third, the values for spatial relative dispersion and the percent of the image with undetectable perfusion, for the control subjects, was low and similar to that previously reported for similar populations (11, 19) and the spatial relative dispersion results in PAH confirm the findings of our previous work (10). Fourth, the amount of registration required was very small with less than a 2% area change in both groups and was not different between groups. Combined, these suggest that the acquisition of tag/control images during respiratory gating with postacquisition registration is an acceptable approach to streamline data acquisition.

Although we have offered some theoretical reasons and data from previous studies to support the idea that FDglobal may be a sensitive measure to evaluate pulmonary circulation, this was not the purpose of the present manuscript and was not evaluated in this manuscript. A larger study is required to confirm these findings and determine relationships with clinical markers and longitudinal follow-up of several patients with PAH is required to evaluate the potential ability of FDglobal to monitor response to therapy.

Conclusions

In a small population of female patients with PAH, we have found evidence of markedly increased spatial-temporal heterogeneity as measured by the FDglobal, a proton MRI metric that has been previously shown to be altered by interventions designed to alter pulmonary vascular regulation. These data suggest that FDglobal may be a useful means of evaluating pulmonary circulation in patients with pulmonary vascular disease. Since this MR imaging technique uses no injected contrast agents and has no ionizing radiation it may be useful in diverse patient populations.

DATA AVAILABILITY

Data will be made available upon reasonable request.

GRANTS

Research reported in this publication was supported by the National Heart, Lung, and Blood Institute of the National Institutes of Health under Award Numbers R01HL129990, R01HL119201, R01HL119263, and 1R56HL159710.

DISCLAIMERS

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

DISCLOSURES

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

G. Kim Prisk is an editor of Journal of Applied Physiology and was not involved and did not have access to information regarding the peer-review process or final disposition of this article. An alternate editor oversaw the peer-review and decision-making process for this article.

AUTHOR CONTRIBUTIONS

A.S.K.P., G.K.P., R.C.S., A.K.A., and S.R.H. conceived and designed research; A.S.K.P., G.K.P., A.R.E., B.P., A.K.A., and S.R.H. performed experiments; A.S.K.P., A.R.E., N.H.K., B.P., R.C.S., and S.R.H. analyzed data; A.S.K.P., G.K.P., A.R.E., N.H.K., B.P., R.C.S., A.K.A., and S.R.H. interpreted results of experiments; S.R.H. prepared figures; A.S.K.P. and S.R.H. drafted manuscript; A.S.K.P., G.K.P., A.R.E., N.H.K., B.P., R.C.S., A.K.A., and S.R.H. edited and revised manuscript; A.S.K.P., G.K.P., A.R.E., N.H.K., B.P., R.C.S., A.K.A., and S.R.H. approved final version of manuscript.