Introduction:

The majority of hyperpolarized gas ventilation MR-acquisitions employ multi-slice fast gradient recalled echo (GRE) imaging (1-3). Such sequences are fast, robust, and reconstruct in real-time to provide exquisitely detailed images of regional ventilation. However, more recently a 3D-radial acquisition has emerged to provide an all-inclusive, isotropic image of ¹²⁹Xe not only in the gas-phase but also its uptake in interstitial barrier and red blood cell compartments. Here we investigate whether the ventilation images acquired as part of this gas-exchange scan can provide equally meaningful information as derived from a standard GRE acquisition.

Methods:

11 subjects with chronic obstructive pulmonary disease (COPD) (age: 64±6 years, FEV1:45±19) were enrolled for imaging on a 3-Tesla SIEMENS MAGNETOM Trio scanner (VB19) (4). Subjects underwent dedicated ventilation MRI using a multi-slice GRE sequence with 4x4x(12-18) mm³ resolution accompanied by a matching ¹H thoracic cavity image acquired using a steady-state fast spin-echo (SSFSE) sequence. Subsequently 3D-radial ventilation images were acquired as part of a gas exchange sequence (4), using 1000 randomized radial views and reconstructed to 128³ using a field-of-view of 40³ cm³ (4, 5). This was accompanied by a matching 3D-radial ¹H thoracic cavity scan reconstructed from 4701 views. All ventilation images were quantified by linear-binning (2, 6) to derive the ventilation defect percentage (VDP), as well as low-, medium- and high-ventilation percentage (LVP, MVP, HVP). Because 3D ventilation images generate a non-Gaussian ventilation distribution in healthy volunteers, a Box-Cox transform was applied to derive appropriate thresholds for quantification. The resulting quantitative metrics were compared between the two acquisitions using a Bland-Altman test. The comparison was further extended to evaluate the spatial overlap of the various bins by registering these two acquisitions to the same space. To do so with high registration accuracy, the SSFSE proton images, acquired at a native resolution of 192x192x14 were first interpolated to 128³. Subsequently, the 3D-radial proton images were registered to the interpolated SSFSE images and the resulting transformation matrix was applied to the 3D-radial ventilation images. The registered 3D-radial images were then down-sampled to the 128x128x14 matrix size of the GRE ventilation scans (Figure 1) to compare the spatial overlap of the respective binning maps by their Dice coefficients.

Results:

Figure 2 shows a representative case with good qualitative agreement between the two acquisitions prior to co-registration. Over the entire cohort, the metrics derived from 3D-radial vs. GRE-ventilation agreed well: VDP = 33±15% vs 39±12%, LVP=29±5% vs 24±4%, and HVP=6±2% for both. As shown in Figure 3, Bland-Altman analysis yielded limits of agreement for radial vs GRE that, for VDP, were [-31% to 16%] with a -6% bias, for LVP were [-1% to 14%] with a +6% bias, and for HVP were [-2% to 1%] with a -0.5% bias. The resampling effect of the 3D radial images is illustrated in Figure 4, which shows that all key features are preserved. After the resampling, the quantitative metrics derived from 3D-radial ventilation were largely unchanged, with VDP=34±14%, LVP=28±7%, and HVP=7±2%. A representative case is shown in Figure 5, in which a good spatial overlap was found between the two acquisitions with a dice coefficient of 0.6 for VDP. Overall, the spatial overlap between GRE and 3D-radial was modest for VDP (Dice=0.5±0.2), and relatively poor for LVP (0.3±0.1) and HVP (0.1±0.1).

Discussion:

This study shows that in a cohort of COPD patients, both ventilation sequences yield similar quantitative metrics. However, Bland-Altman analysis shows that limits of agreement for VDP are not yet sufficiently tight. The work also shows that resampled 3D-radial images preserve the quantitative metrics and permit pixel-by-pixel comparison to the GRE acquisition. However, the resulting Dice coefficients show only modest agreement for VDP and relatively poor agreement for LVP and HVP. Although some of these differences may be attributable to the images being acquired during different breath-holds, and with different xenon volumes, they also highlight several areas that must be investigated to reconcile these imaging approaches. First, it must be noted that the healthy reference distribution arising from 3D-radial imaging is inherently non-Gaussian and requires using a Box-Cox transformation to set binning thresholds. However, perhaps the most significant difference between the two is the effect of bias-field correction. Although the algorithm and parameter settings applied were identical, it is likely that the physiologic gradients, which are preserved in the 3D-radial acquisition, affect the algorithm differently than for multi-slice acquisition. It is likely that with further reconciliation of bias-field correction, the 3D-radial scan could well provide the necessary information to obviate the need for a dedicated ventilation scan in a majority of cases.

Acknowledgements

NIH/NHLBI R01 HL105643, NIH/NHLBI R01HL126771, and HHSN268201700001C

References

1. Kirby M, Pike D, Coxson HO, McCormack DG, Parraga G. Hyperpolarized He-3 Ventilation Defects Used to Predict Pulmonary Exacerbations in Mild to Moderate Chronic Obstructive Pulmonary Disease. Radiology. 2014; 273(3):887-96. 2. He M, Driehuys B, Que L, Huang Y-CT. Using Hyperpolarized 129Xe MRI to Quantify the Pulmonary Ventilation Distribution. Acad Radiol. 2016. 3. Zha W, Niles DJ, Kruger SJ, et al. Semiautomated Ventilation Defect Quantification in Exercise-induced Bronchoconstriction Using Hyperpolarized Helium-3 Magnetic Resonance Imaging: A Repeatability Study. Acad Radiol. 2016; 23(9):1104-14. 4. Wang Z, He M, Bier EA, et al. Hyperpolarized 129 Xe gas transfer MRI: the transition from 1.5T to 3T. Magn Reson Med. 2018. 5. Robertson SH, Virgincar RS, He M, Freeman MS, Kaushik SS, Driehuys B. Optimizing 3D noncartesian gridding reconstruction for hyperpolarized Xe-129 MRI: focus on preclinical applications. Concept Magn Reson A. 2015; 44(4):190-202. 6. He M, Kaushik SS, Robertson SH, et al. Extending Semiautomatic Ventilation Defect Analysis for Hyperpolarized Xe-129 Ventilation MRI. Academic Radiology. 2014; 21(12):1530-41.