Introduction

High-performance low-field systems can improve the clinical feasibility of combined structural and functional MRI of the lungs¹. Oxygen-enhanced MRI (OE-MRI) provides regional information on ventilation and perfusion and is more accessible than hyperpolarized gas MRI². OE-MRI is improved with low-field MRI due to the increased T1 relaxivity of oxygen at lower field strengths³, which increases the percent signal enhancement between normoxia and hyperoxia images. Majority of OE-MRI methods rely on breath-held imaging^4,5, and only a few studies have used free-breathing acquisitions^6,7 with respiratory bellows, which are difficult to use and required additional hardware. Breath-holds are difficult for patients with lung disease, and in addition, inconsistent breath-holding position can be a source of quantification error⁶. Breath-held imaging is also limited in its image resolution and signal-to-noise ratio (SNR) because of the requirement to maintain short breath-hold durations.

In this work, we 1) demonstrate the feasibility of self-gated free-breathing stack-of-spirals with an ultra-short echo time (UTE) acquisition for OE-MRI at 0.55T, and 2) compare the measurement repeatability of free-breathing (FB) and breath-held (BH) OE-MRI methods.

Methods

Data Acquisition: We used a prototype 3D T1-weighted UTE spoiled gradient echo stack-of-spirals sequence for imaging during normoxia and hyperoxia on a 0.55T MRI system (prototype MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany). Hyperoxia was induced by inhalation of 15L/min 100% oxygen through a non-rebreather facemask. The transition between normoxia and hyperoxia states was approximately 3 min. Imaging parameters for both BH and FB scans were TE/TR 0.2/6.83 ms, spiral readout = 5 ms, flip angle = 17.5°, FOV = 450x450x320 mm, matrix size = 128x128x32, slice thickness = 10 mm. All scans used a constant density spiral-out readout with a 23 shot design. BH scans were 16s and acquired 61 over-sampled shots with 5.9° angle-increments between shots. FB scans lasted 2:25 min and acquired 591 shots with 137.5° (golden-angle) angle increments. Three navigator readouts were acquired in the FB scans along the superior-inferior (SI) direction for each slice-encoding loop to estimate the respiratory signal.

Reconstruction: Respiratory waveforms were extracted from the navigator using previously published methods⁸. FB data was binned the most stable respiratory phase, chosen as the window of respiratory signal with the smallest gradient (Figure 1B). The amount of accepted data was empirically chosen to be 40%, such that no respiratory motion related blurring was observed in the reconstructed images. Data were reconstructed using Gadgetron⁹ for both FB and BH acquisitions. Density compensation weights were estimated using an iterative method¹⁰. Binned FB data was reconstructed using CG-SENSE.

Experiment: To assess intra-exam repeatability, 4 OE-MRI measurements were performed during a scan session using both the BH and proposed FB methods. Three healthy volunteers were imaged with local institutional review board approval. In 2 volunteers, the scan began with 100% oxygen (O2) and transitioned to room air (RA). In 1 volunteer, the order of physiological states was reversed to adjust for potential bias.

Analysis: OE-MRI analysis was performed in MATLAB (Mathworks, Natick MA). Normoxia images were registered to hyperoxia images using Advanced Normalization Tools (ANTs)¹¹. The Jacobian determinant of the deformation fields from image registration (det(J_T)) was used to correct for local signal intensity changes due to volumetric changes in the lung⁶. Percent signal enhancement (PSE) maps were generated using equation 1. Sixteen PSE maps were estimated from the set of 4 normoxia and 4 hyperoxia, images. Temporal SNR maps (mean signal across repetitions / standard deviation of signal across repetitions) were also calculated for registered and unregistered images to estimate the temporal stability of the lung signal during normoxia and hyperoxia for both BH and FB images.

$$ PSE = \frac{(Im_{hyperoxia} - Im_{normoxia} * det(J_T))}{Im_{normoxia} * det(J_T)} $$

Results

FB stack-of-spirals UTE generated high quality images and PSE maps (Figure 2). FB stack-of-spirals OE-MRI qualitatively showed good repeatability, as demonstrated in two repetitions in Figure 2B,C. The coefficient of variation for the mean PSE was measured to be 30.8 ± 18.42 % for FB OE-MRI across the 3 subjects. By comparison, BH OE-MRI had poor repeatability (coefficient of variation = 114.3 ± 70.1 %). Improved repeatability of FB OE-MRI is attributed to the consistent respiratory phase, resulting in equivalent lung tissue density for each measurement as shown in Figure 3, and higher SNR. Table 1 compares the mean PSE and tSNR in each subject for free-breathing and breath-held protocols. Across all repetitions from 3 healthy volunteers, we measured mean PSE of 11.2% ± 1.1% for our proposed free-breathing protocol, and 17.7% ± 11.9% for breath-held protocol. FB data sets had higher tSNR (27.1+/-13.6) than breath-held data sets (12.6 ± 2.6) which indicates higher sensitivity, as shown in Figure 4, likely due to higher SNR in the FB data.

Discussion and Conclusion

We have presented a novel self-gated FB approach for OE-MRI with T1w 3D stack-of­-spirals UTE imaging. This approach improves the sensitivity and repeatability of OE-MRI and is easier to use in clinic because it doesn’t require breath holds. This is significant because robust OE-MRI can make routine functional pulmonary imaging clinically viable by capitalizing on the unique properties of high-performance lower field MRI.

Acknowledgements

The authors thank Kendall O’Brien and Christine Mancini for their imaging expertise and Margaret (Peg) Lowery, Jennifer Henry and Amelia Nargozian for assistance with patient recruitment. The authors would like to acknowledge the assistance of Siemens Healthcare in the modification of the MRI system for operation at 0.55T under an existing cooperative research agreement (CRADA) between NHLBI and Siemens Healthcare.