Simultaneous MRI of lung structure and perfusion in a single breathhold

To develop and demonstrate a breathheld 3D radial ultrashort echo time (UTE) acquisition to visualize co‐registered lung perfusion and vascular structure.

LUNG DISEASES USUALLY cause both structural and functional changes. For example, pulmonary emboli (PE) restrict flow in the pulmonary arteries, leading to decreased perfusion distally. To identify the source and impact of PE on lung function, imaging of the vascular anatomy and perfusion is required in combination with structural images to rule out alternative diagnosis. Traditionally, computed tomography (CT) has been used to assess the lung vasculature and structure, while scintigraphy has been used to evaluate lung perfusion and ventilation. Over the past decade, MRI has shown growing promise as a crosssectional modality for imaging lung structure (1)(2)(3)(4)(5), ventilation (6), and perfusion (7,8). Early pilot studies in different patient populations have suggested that these approaches may have clinical utility. The combination of these techniques within a single modality holds promise for co-registered and complementary functional and structural images within a single exam and without the use of ionizing radiation. However, current MRI protocols for pulmonary emboli require separate scans optimized for evaluation of vascular structure and pulmonary perfusion. Although effective, separate structure and function imaging protocols complicate workflow and diagnosis. Improved efficiency and diagnostic accuracy is likely to be achieved for an MRI technique capable of joint assessment of vascular structure and perfusion in a manner similar to that used routinely with CT.
Highly accelerated contrast-enhanced MRI has enabled the acquisition of high resolution pulmonary perfusion and angiography (MRA) (9)(10)(11). Unfortunately, the conflicting needs for high spatial resolution for MRA ($1.5 mm isotropic) and high temporal resolution for perfusion ($1 s) pose a challenge for a single imaging sequence. Conventionally, to achieve sufficient temporal resolution for perfusion, relatively low spatial resolution is used, typically 3-4 mm in each dimension (11)(12)(13)(14). While these perfusion sequences are likely sufficient to identify clinically significant focal perfusion defects, they are inadequate to directly visualize the corresponding vascular pathology (e.g., filling defects due to pulmonary emboli). Conversely, pulmonary MRA scans are most often acquired nondynamically, precluding their use for robust evaluation of perfusion defects (15,16). For these reasons, evaluation of both vasculature structure and perfusion typically requires two separate breathheld scans.
The precise correlation of vascular structure and perfusion abnormalities on these separate scans requires retrospective image registration. However, image registration is complicated by elastic lung deformation and differing contrasts in perfusion and angiography scans, making it difficult to integrate into routine clinical workflows. In practice, structurefunction correlation is performed subjectively by a radiologist using visual side-by-side inspection of the images. A single breathheld acquisition producing intrinsically co-registered vasculature structure and perfusion images would more efficiently correlate lung structure with function and would enable integration within a reasonable clinical workflow. Furthermore, this strategy would also reduce the number of breathholds, contrast injections, and overall scan time in a patient population that is liable to be dyspneic and unable to tolerate long exams.
Radial sampling enables higher temporal resolution, spatial resolution and coverage compared with Cartesian sampling due to the repeated sampling of the center of k-space. In particular, three-dimensional (3D) radial sampling has been demonstrated for full chest coverage for both dynamic MRA (17) and ultra short echo time (UTE) applications (18). When combined with center-out sampling (as in UTE) marked improvement in the visualization of structural lung disease has been demonstrated (3,4,18). Although recent small animal studies suggest promise for joint lung structure and perfusion visualization, the combination of dynamic 3D radial imaging with the structural imaging capability of UTE has not yet been investigated for dynamic perfusion and structural imaging in large animals or humans.
The purpose of this work was to develop and demonstrate a method to simultaneously acquire both high-resolution vascular and structural images with intrinsically co-registered perfusion images using a single breathheld 3D radial UTE acquisition.

MATERIALS AND METHODS
Nine canine subjects (eight males: one female; weight 10.7 6 1.2 kg; mean age 13 months) underwent dynamic-contrast enhanced MR imaging using a timeresolved 3D radial UTE pulse sequence. Each animal was scanned twice, separated by 2-4 days, resulting in a total of eighteen UTE perfusion acquisitions. Animals were mechanically ventilated and placed in the scanner in the supine position. Due to the requirements of a separate study being conducted simultaneously on the same dogs, the anesthesia used on the two study visits differed. Each dog received midazolam and fentanyl on one visit, and dexmedetomidine on the other. Eight of the animals were healthy and showed no cardiopulmonary abnormality; however, the one female animal had an incidentally discovered membranous ventricular septal defect causing a small left to right cardiac shunt. Our Institutional Animal Care and Use Committee approved this study.

MR Protocol and Postprocessing
All imaging was done on a commercial 3T scanner (MR750, GE Healthcare, Milwaukee, WI) with gradient peak slew rate of 200 T/m/s and maximum gradient strength of 50 mT/m. Twenty elements of a clinical 32-channel chest phased array coil (Torso Array, Neocoil, Pewaukee, WI) were used, providing for complete lung coverage. Two acquisitions were performed: a 33-s dynamic breathhold acquisition during contrast injection and a 5-min respiratory-gated acquisition immediately following.
Temporally interleaved 3D radial UTE images were acquired starting simultaneously with the injection of 2.3 mL ($0.1 mmol/kg) of gadobenate dimeglumine (Multihance; Bracco Diagnostics, Princeton, NJ) followed by a 17 mL saline flush at 2 mL/s in the cubital vein. Relevant acquisition parameters included: TR/ TE ¼ 3.3/0.08 ms, receiver bandwidth ¼ 6 125 kHz, total number of readout points ¼ 682 (1.4 ms readout time), flip angle ¼ 15 , 0.94 mm isotropic spatial resolution, 8000 total projections, and 1 s time frames acquired over a 33 s breathhold at end expiration. The breathhold was performed by stopping mechanical ventilation to allow for scanning. A slab-selective radiofrequency excitation with limited field of view, variable read-out gradients, and radially oversampled (two times oversampled) projections were used to improve the quality of the lung images. Additional details regarding the design and optimization of the 3D radial UTE pulse sequence have been previously published (18). The 3D radial UTE data were reconstructed using an iterative sensitivity encoding algorithm (iSENSE) (16,19) at full 0.94 mm isotropic resolution on a 256 3 matrix. As the breathheld perfusion acquisition is composed of 33 undersampled time frames, each consisting of a pseudo-random interleaved projection scheme for relatively uniform k-space coverage, a kspace adaptive filter was used as a temporal viewsharing method for reconstruction (20). This filter has a width of 1 s at the center of k-space increases to 7 s with a quadratic weighting function from the center to the edge of k-space. Therefore, the effective frame rate is 1 s with a temporal footprint of 7 s.
For structural analysis, a composite dataset consisting of all 8,000 projections acquired during the dynamic acquisition was reconstructed at full 0.94 mm isotropic resolution. This 8,000 projection breathheld composite image acquired in 33 s was compared with the reconstruction of a respiratory-gated image consisting of 38,000 projections acquired over a 5min acquisition; hereafter referred to as "breathheld" and "respiratory-gated" scans, respectively. The respiratory-gated, 3D radial structural UTE scan was acquired with the same pertinent acquisition parameters during tidal breathing after the dynamic breathheld acquisition. Prospective respiratory gating to end-expiration was used with adaptive feedback from the respiratory bellows signal used to define a 50% acceptance window. Due to the 50% acceptance window, the scan efficiency was 50% as the number of projections remained fixed. Each subject's respiratory pattern determined the total number of projections acquired during each expiratory phase.

Dynamic Contrast Enhanced Measures
To evaluate regional qualitative pulmonary perfusion, 3D peak lung enhancement maps were calculated. The peak lung enhancement maps were calculated by simply performing a maximum intensity projection along the time dimension on a voxel-by-voxel basis. These peak lung enhancement maps were used to calculate the gravity dependent anterior-posterior (AP) gradient within the lungs, measured by comparing the mean signal in a region of interest (ROI) placed in the anterior right lung with the mean signal in an ROI placed in the posterior right lung, each normalized to the mean signal within ROIs placed in nearby muscle. The anatomic locations of the ROIs were determined by axially scrolling through the volume dataset to locate a slice superior to the heart.
The right ventricle (RV) to aorta transit time was calculated by manually placing circular ROI's centered in the RV and descending aorta to determine their enhancement as a function of time. Transit times were determined by the time interval between peak enhancement in the RV and in the aorta.
Relative lung enhancement was calculated on a voxel-wise basis by subtracting precontrast signal derived from the first time frame, and then dividing by this preconstrast signal. The first time frame is more likely a noise measure of stochastic and image artifact than true precontrast signal; therefore these calculations are a mixed metric of both signal and noise. Again, a circular ROI that avoided large vessels was manually positioned within the right lung to determine average relative lung enhancement for each dog.

Vascular Contrast and Structural Measures
Vessel edge width (to assess vessel sharpness), relative vascular signal (to assess the amount of vascular enhancement), and relative lung signal (to assess how well lung tissue is visualized) were used as structural image quality metrics. The dynamic breathheld acquisition and the respiratory-gated acquisition were compared using these measures.
Vessel edge width was chosen to evaluate potential blurring which can occur both due to motion and due to changing contrast dynamics during the scan (21). Using multiplanar reformations, cross-sectional images of two representative pulmonary arteries just distal to their origins were generated: the left main pulmonary artery and the first medially branching segmental pulmonary artery arising from the caudal lobe of the right lung (Fig. 5). The vessel edge width was measured along 12 evenly spaced radial edge profiles beginning at the center of the specified vessel (demonstrated on Fig. 5). Vessel edge width was defined to be the distance from 20% to 80% of the peak value. Edge profiles that failed to drop below 20% were excluded from the analysis due to insufficient contrast between the measured vessel and adjacent background structures.
Relative vascular and lung signal were calculated through manual placement of circular ROIs. Both measurements were normalized to the standard deviation of airway signal as a "pseudo-noise" metric. Relative vasculature signal (rVS) was calculated using the mean signal in ROIs placed in the vessels used for the vessel edge width measurements and in nearby lung tissue: Similarly, relative lung signal (rLS) was calculated using the mean signal in six circular ROIs placed in anterior, middle, and posterior positions in both lungs and an ROI in nearby airway using the equation: Image measurement and statistical analysis was performed using MATLAB (The Mathworks Inc.; Natick, MA, USA). The quantitative measures were reported as mean 6 standard deviation and were compared using paired Student's t-tests. P-values less than 0.05 were considered statistically significant.
In addition to the parametric measurements previously mentioned, two cardiothoracic radiologists with 6 (S.N.) and 7 (C.F.) years of experience in cardiovascular MRI scored the structural breathheld and respiratory-gated composites by consensus. The structural anatomy of the pulmonary arteries was evaluated on a 4-point scale, indicating the smallest well visualized vessels . Two components of structural quality of the lung were scored on a binary scale: one for differentiating airways from the lung and the second for identifying an AP gradient. Lastly, subjective measurements of the noise in the composites were also scored on a 3-point scale. (1 ¼ "Severe: renders images nondiagnostic for PE or consolidation," 2 ¼ "Moderate: diagnostic but decreased confidence," 3 ¼ "Mild: diagnostic with high confidence," 4 ¼ "No significant noise").

RESULTS
Both the time-resolved dynamic 3D radial UTE perfusion and the respiratory-gated structural UTE acquisition were successfully performed in all 18 experiments (2 experiments per animal). The first two (2) experiments were performed using different scan parameters than the remainder and were therefore excluded, resulting in a total of sixteen (16) datasets for analysis. Reconstruction of both perfusion and structural images from the same dataset eliminated the need for image registration (Fig. 1). Figure 2 shows coronal MIP images of the first pass contrast dynamics from a typical scan, reconstructed at 1.0 frame/s and 2.3 mm isotropic spatial resolution. Figure 3 illustrates the relative enhancement of the right ventricle, lung, and left ventricle over the breathhold time in the same experiment. Mean relative lung enhancement was a factor of 8.4 6 1.5 (Table  1). Mean RV to aorta transit times was 7.4 6 2.0 s (Table 1) with a typical example shown in Figure 3. The normalized anterior lung signal was 0.80 6 0.13 and the normalized posterior lung signal was 1.76 6 0.30 (Table 1), resulting in a statistically significant AP gradient (P ( 0.05).
An unsuspected membranous ventricular septal defect was discovered in one of the presumed healthy dogs using data acquired for an unrelated study. This resulted in altered temporal dynamics compared with the other animals (Fig. 4). The full-width-at-halfmaximum (FWHM) of the right ventricular enhancement peak was greater than measured in the other dogs, indicating a left-to-right cardiac shunt consistent with the animal's ventricular septal defect. As seen in Figure 4 the measured FWHM was 12.6 s compared with an average of 8.7 6 1.4 s for healthy individuals administered under midazolam and fentanyl anesthesia. Similar results were seen for the exams performed under dexmedetomidine anesthesia. This result illustrates that despite the use of a kspace temporal filter, abnormalities in enhancement curves can be detected. Figure 5 demonstrates typical cross-sectional appearances of the left main pulmonary artery and the right segmental pulmonary artery used for structural assessment. Table 2 summarizes the results of the structural analysis of both composite breathheld and respiratorygated acquisitions. The vessel edge width of both large and small arteries differed significantly (P ¼ 0.003) between the acquisitions. The vessel edge width was greater (ie, vessels were more blurred) on the composite breathheld acquisitions for both the large (left main) and small (segmental) pulmonary arteries (P < 0.05 for both vessels). There was no statistically significant difference in relative vascular signal in either the left main pulmonary artery (P ¼ 0. 16) or in the smaller segmental  Radiologist scoring showed similarly excellent depiction of the pulmonary arteries to the subsegmental level in 15/16 (94%) of the breathheld and 16/16 (100%) of the respiratory-gated acquisitions (P ¼ 0.33). One breathheld case was visualized to the segmental level (score of 3) instead of the subsegmental level. Differentiation between lung tissue and airways was possible for both acquisitions, whereas the AP gradient was clearly seen in the breathheld composites. Breathheld composites were significantly more compromised by noise (median score of 3) than the respiratory-gated images (median score of 4) at P < 0.001.

DISCUSSION
This study demonstrates the feasibility of simultaneously acquiring co-registered pulmonary perfusion and pulmonary artery structure images using a 3D radial UTE sequence. Prior results have shown that lung disease, such as pulmonary embolism, can be identified by both the qualitative changes in pulmonary perfusion and by direct visualization of embolus in the pulmonary artery (7,(22)(23)(24). Using a single 3D dataset acquired during a single breathhold to reconstruct images of both lung structure and function allows direct spatial correlation between the findings on each. This avoids the need for retrospective image registration, shortens total scan time by eliminating the need for a second breathhold, and eliminates a second contrast injection.
Reasonable first-pass contrast dynamics and lung enhancement were seen in all of the animals scanned using this dynamic 3D UTE acquisition. The circulation of the contrast bolus through the right ventricle, pulmonary arteries, lung, and left ventricle could be clearly differentiated despite a temporal footprint of seven seconds that incorporated view sharing in the periphery of k-space. Observed temporal dynamics allowed for quantification of the transit time of contrast between the right ventricle and the descending aorta. Lung enhancement showed the expected gravitational gradient in AP direction, consistent with normal physiology (25,26). The presence of contrast during the breathheld scan most likely created a more defined AP gradient than in the respiratory-gated composites. The observed left-to-right shunt in one animal was later confirmed to be a membranous ventricular septal defect, and this serendipitous result suggests that temporal filtering imposed by view sharing may not compromise quantitative measures.
Vessel edges were sharper on the respiratory-gated acquisition than on the breathheld acquisition. We suspect that the primary cause for the increased blurring on the breathheld acquisition is the time-varying contrast bolus imposing a low-pass filter during data acquisition. This can introduce artifacts, analogous to the ringing and widening of vessels edges that can be seen in Cartesian acquisitions (21). However the  1.76 6 0.30 *Relative lung enhancement was visualized and calculated for all twelve subjects during the first-pass of contrast. Temporal behaviors of the first-pass of contrast were quantified from the right ventricle to the aorta. Additionally, a statistical difference is seen between the normalized signal measurements in the anterior and poster regions of the pulmonary perfusion maps. blurring of the vessel edge width during the breathhold was relatively mild because good visualization of the pulmonary arteries to the subsegmental level was appreciated by the radiologist readers. Relative vascular signal trended slightly higher in the respiratory-gated images than in the breathheld images. Intuitively, it would seem that the relative vascular signal should be higher in the breathheld acquisition due to the higher concentration of contrast present in the vessel, but the aggressive undersampling introduces more streak artifacts. Despite this, the relative vascular signal did not differ significantly between the two scans. Finally, relative lung signal measurements trended slightly lower in the breathheld versus respiratory-gated images, likely due again, to the streak artifacts arising from the more aggressive radial under-sampling, causing the standard deviation of the airway signal to be higher in the breathheld acquisition. Indeed, scoring by radiologists demonstrated that the breathheld scans were "noisier" than the respiratory-gated scans; however, the breathheld scans were still considered "reasonably diagnostic". Nevertheless, the relative lung signal did not differ significantly between the scans, suggesting that these under-sampling artifacts are relatively minor.
This study has several limitations. As data used to reconstruct each 3D volume in the perfusion timecourse are under-sampled, an adaptive k-space filter was necessary to mitigate radial streak artifact. Therefore, absolute quantification of the arrival times of the contrast bolus may be compromised. Similarly,  quantitative hemodynamic parameters such as pulmonary blood flow, pulmonary blood volume, and mean transit time derived from this acquisition may be similarly compromised. The application of compressed sensing reconstruction methods to this technique may improve the true temporal resolution (i.e., temporal footprint) of the method, potentially enabling this method to be used for more quantitative pulmonary perfusion measurements. Compressed sensing also could mitigate the subjective noise seen in the breathheld composites. In addition, to ensure linearity between contrast agent concentration and signal in the arterial input function (typically the pulmonary artery), quantification demands a small contrast bolus and optimized pulse sequence parameters (e.g., flip angle) both of which were outside the scope of the present study. For these reasons we chose 3D peak enhancement maps to qualitatively assess pulmonary perfusion because this metric has been used in other studies (27). Finally, this feasibility study was performed only in a small population of presumed healthy animals.
The dynamic contrast enhancement measures demonstrated strong peak enhancement of lung tissue relative to the precontrast image (a factor of eight) and the expected physiologic AP gradient in perfusion was apparent. We, therefore, believe that sufficient signal is obtained on the perfusion reconstruction to support feasibility for detection of clinically significant perfusion defects; however, the performance of this imaging approach in the setting of lung disease remains to be assessed and is indeed the subject of an ongoing study. Furthermore, the 33 s breathhold acquisition currently implemented in our animal study will be difficult to translate to human studies. It was chosen as such to insure sufficient number of projections for the composite structural reconstruction. In translating this method to humans, the scan time may be further shortened by constrained reconstruction methods and pulse sequence optimization.
In conclusion, we have demonstrated the feasibility of using time-resolved 3D radial UTE MRI for simultaneous imaging of lung function (perfusion) and coregistered pulmonary vascular structure during a single breathhold. This approach to structurefunction imaging is achieved in a short exam and is relatively robust to motion due to the use of a 3D radial acquisition. It is well suited for clinical situations in which rapid assessment of pulmonary perfusion and pulmonary artery structure is needed because it requires only a single breathheld with a small dose of contrast.