Wei Zha^{1}, Kevin M Johnson^{1,2}, Robert V Cadman^{1}, Scott K Nagle^{1,2,3}, and Sean B Fain^{1,2,4}

**Three-dimensional
dynamic imaging using free-breathing oxygen-enhanced (OE) ultrashort echo time
(UTE) MRI can measure change of partial pressure of oxygen (ΔPO _{2}) and wash-in/out time
constants. Nine healthy subjects underwent the dynamic OE MRI protocol at 1.5T.
A subset of 4 subjects underwent repeated exams and 2 of these 4 underwent the
same protocol at 3.0T on the same day. The inter-exam variations at 1.5T
suggest good quantitative measurements of lung function and wash-in/out
dynamics with variations observed in ΔPO_{2max}. The parametric maps imply comparable wash-in/out time
constants and normal ventilation measured at 1.5T vs. 3.0T.**

Nine
healthy subjects were enrolled in a HIPAA-compliant study. Seven
subjects underwent pulmonary function tests (PFTs). A subset of 4 subjects were
each imaged at 2 separate visits ≤5months apart at 1.5T (HDxt, GE
Healthcare). Two of these 4 subjects underwent the same dynamic protocol at
1.5T with an 8-channel cardiac coil and 3.0T (MR750, GE Healthcare) with a
32-channel torso coil on the same day. Subjects were positioned supine and
breathed freely via a non-rebreather facemask with 21% (normoxic) or 100% O_{2}
(hyperoxic) flowing at 15 L/min throughout the scan.

The data and analysis workflow for the dynamic OE UTE MRI^{3} is
illustrated in **Figure 1**. Each UTE was
acquired with 32cm FOV, TE=0.08-0.1ms and prospective gating with a real-time
adaptive 50% acceptance window at end expiration. Regional T1 was measured with
normoxic UTE variable flip angle (VFA)^{4}: 5 FAs = 2°, 4°, 6°, 10°/9°(1.5T/3.0T), and 14°, TR=2.86 ms, ~30,000
projections per FA, and scan time=~14 minutes. The dynamic scan was a
continuous acquisition of 8~18 timeframes with total scan time from 6.8 to 9.3
minutes to cover one cycle of wash-in and wash-out dynamics: subjects breathed
21% O_{2} for the first 2 timeframes, and then the oxygen concentration
alternated between 100% and 21% with 3 (70-second temporal), 5 (34-second) or 8
(24-second) timeframes at each oxygen concentration, FA = 8°/7° (1.5T/3.0T),
TR=1.7ms. All UTE images were reconstructed at 1.25mm isotropic for structure
and 1cm isotropic for dynamic-function quantification.

A deep learning technique was used to automatically segment lungs volumes^{5}. After registration, the baseline (timeframe#1)
T1 was used to estimate $$$T_{1}(t)$$$ at subsequent timeframes to derive the change in
partial pressure of oxygen, $$$\triangle PO_{2}(t)=(1/T_1(t)-T_{1baseline})/r_{o2}$$$,
where $$$r_{O2}$$$=2.49×10-4/s mmHg [6]. The O_{2} wash-in and wash-out
portions of $$$\triangle PO_{2}(t)$$$ curves were fitted using exponential functions^{7}, $$$\triangle PO_{2}(t)=\triangle PO_{2max}\times(1-e^{-t/\tau_{up}})$$$ and $$$\triangle PO_{2}(t)=\triangle PO_{2max}\times e^{-t/\tau_{down}}$$$, to compute the time constants ($$$\tau_{up}$$$
and $$$\tau_{down}$$$) respectively, where $$$\triangle PO_{2max}$$$ is the maximum $$$\triangle PO_{2}$$$ after switching air to 100% O_{2}. The maximum intensity
projection (MIP) of median percent signal enhancement (MPSE) over
hyperoxic-breathing timeframes was used to quantify ventilation defect percent
(VDP) automatically^{8}.
The
repeated measurements of $$$MPSE_{max}$$$ , $$$\triangle PO_{2max}$$$, $$$\tau_{up}$$$ and $$$\tau_{down}$$$ were averaged to
calculate the subject group statistics. Bland-Altman analysis with 95% limits
of agreement (LOA) was used to evaluate inter-exam variations at 1.5T.

[1] Kruger SJ, Fain SB, Johnson KM, Cadman R V, Nagle SK. Oxygen-enhanced 3D Radial Ultrashort Echo Time Magnetic Resonance Imaging in the Healthy Human Lung. NMR Biomed 2014;27:1535–41. doi:10.1002/nbm.3158.

[2] Zha W, Nagle SK, Cadman R V, Schiebler ML, Fain SB. 3D Isotropic Functional Imaging in Cystic Fibrosis Using Oxygen-enhanced MRI: Comparison with Hyperpolarized Helium-3 MRI. Radiology 2018. doi: 10.1148/radiol.2018181148

[3] Zha W, Cadman R V., Hahn AD, Johnson KM, Fain SB. Dynamic 3D Isotropic Resolution Imaging of Human Lungs Using Oxygen-enhanced Radial UTE MRI. Abstr #4147, 26th ISMRM Paris, France 2018.

[4] Bell LC, Johnson KM, Fain SB, Kruger SJ, Nagle SK. T1 <apping of the Lungs Using DESPOT1 Approach with 3D Radial UTE Acquisition. Abtract# 4234 20th ISMRM Melbourne, Aust 2012.

[5] Zha W, Fain SB, Schiebler ML, Nagle SK, Liu F. Improved Segmentation of Proton MRI Lung Volume Using a 2.5D Deep Convolutional Neural Network. ISMRM Work Mach Learn Part I, Pacific Grove, CA, 2018.

[6] Zaharchuk G, Busse RF, Rosenthal G, Manley GT, Glenn OA, Dillon WP. Noninvasive oxygen partial pressure measurement of human body fluids in vivo using magnetic resonance imaging. Acad Radiol 2006;13:1016–24. doi:10.1016/j.acra.2006.04.016.

[7] Zhang W-J, Niven RM, Young SS, Liu Y-Z, Parker GJM, Naish JH. Dynamic oxygen-enhanced magnetic resonance imaging of the lung in asthma -- initial experience. Eur J Radiol 2015;84:318–26. doi:10.1016/j.ejrad.2014.10.021. [8] Zha W, Kruger SJ, Johnson KM, Cadman R V, Bell LC, Liu F, et al. Pulmonary Ventilation Imaging in Asthma and Cystic Fibrosis Using Oxygen-Enhanced 3D Radial Ultrashort Echo Time MRI. J Magn Reson Imaging 2018;47:1287–97. doi:10.1002/jmri.25877.

[9] Naish JH, Parker GJM, Beatty PC, Jackson A, Young SS, Waterton JC, et al. Improved Quantitative Dynamic Regional Oxygen-enhanced Pulmonary Imaging Using Image Registration. Magn Reson Med 2005;54:464–9. doi:10.1002/mrm.20570.

[10] Martini K, Gygax CM, Benden C, Morgan AR, Parker GJM, Frauenfelder T. Volumetric Dynamic Oxygen-enhanced MRI (OE-MRI): Comparison with CT Brody Score and Lung Function in Cystic Fibrosis Patients. Eur Radiol 2018:1–11. doi:10.1007/s00330-018-5383-5.

[11] Marshall H, Deppe MH, Parra-Robles J, Hillis S, Billings CG, Rajaram S, et al. Direct visualisation of collateral ventilation in COPD with hyperpolarised gas MRI. Thorax 2012;67:613–7. doi:10.1136/thoraxjnl-2011-200864.

[12] Liu F, Feng L, Kijowski R. MANTIS: Model-Augmented Neural neTwork with Incoherent k-space Sampling for Efficient Estimation of MR Parameters. ISMRM Work OnMachine Learn Part II, Cap Hilton, Washington, DC, USA 2018.

[13] Mardani M, Gong E, Cheng JY, Vasanawala S, Zaharchuk G, Alley M, et al. Deep Generative Adversarial Networks for Compressed Sensing Automates MRI. IEEE Trans Med Imaging 2017;PP:1. doi:10.1109/TMI.2018.2858752.