3D Multi-Parametric Acquisition of 3He Lung Ventilation Images, Lung Diffusion Morphometry and T2* Maps with Compressed Sensing
Ho-Fung Chan1, Neil J. Stewart1, Guilhem J. Collier1, and Jim M. Wild1

1Academic Unit of Radiology, University of Sheffield, Sheffield, United Kingdom


Whole-lung coverage 3He ventilation images, maps of ADC, alveolar dimension (LmD), and T2* were acquired in a single breath-hold using a multiple-interleaved 3D sequence with compressing sensing (CS). A fully-sampled three-interleaved ADC and T2* dataset was acquired for CS simulations, to determine the optimal k-space undersampling patterns. A prospective, 3-fold undersampled 3D five-interleaved dataset was acquired with CS and parametric maps were compared to those calculated from fully-sampled datasets. CS-derived ADC and LmD values showed good agreement with fully-sampled equivalents. CS-derived T2­* values were lower than fully-sampled ones due to the smoothing process of the CS reconstruction.


Hyperpolarised 3He gas MRI has been shown to provide quantitative measures of regional lung ventilation [1] and lung microstructure [2,3]. In particular, the 3He apparent diffusion coefficient (ADC) [2], estimates of alveolar dimension (LmD) [3], and T2* [4] are sensitive to changes in lung microstructure and function. Previously, single breath-hold acquisition of 2D 3He ventilation, ADC, T2*, and B1 maps were demonstrated at 3T using compressed sensing (CS) [5]. However, this 2D multi-slice acquisition did not provide whole-lung coverage or multiple b-value data for LmD calculation. In this work, a 3D multiple-interleaved sequence was implemented with CS [6] to acquire whole-lung coverage co-registered 3He ventilation images, parametric maps of ADC, LmD, and T2* within a single breath-hold.


The 3D multiple-interleaved sequence is summarised in Figure 1. The first and second interleaves were used to compute ADC maps, while the first four interleaves were used to calculate LmD by fitting data with the stretched exponential model [3]. Finally, the first and fifth interleaves were used to compute T2* maps.

In order to simulate the optimal k-space sampling pattern for CS, a fully-sampled three-interleaved dataset (interleaves 1, 2 and 5 in Figure 1) was acquired in a healthy volunteer (M, 25y) on a GE HDx 1.5T MR scanner using 400mL of 3He (~25% polarisation). Imaging parameters: 3D SPGR, b=0, 1.6 s/cm2, 80x66x22 matrix, FOV=40x32.5x24.4 cm3, TE1/TR1=4.1/5.7 ms, TE2/TR2=14.8/16.4 ms, diffusion time Δ=1.6ms, flip angle=1.4°, bandwidth=±31.25 kHz. Optimal k-space undersampling patterns were determined for acceleration factors (AFs) of 2 to 5 by minimising the mean absolute error between undersampled and fully-sampled ventilation images (MAE), ADC maps (ADC MAE), and T2* maps (T2* MAE). By retrospectively undersampling the fully-sampled dataset, images were reconstructed from the optimal undersampling patterns [6], and mean ADC and T2* values were compared between each AF.

The full five-interleaved sequence was implemented with CS in a prospective acquisition on the same healthy volunteer. Three-fold undersampling was introduced to reduce the breath-hold time to 18s. Imaging parameters were as above, except four b-values (0, 1.6, 4.2, 7.2 s/cm2) and flip angle=1.9° were used. Prospective CS-sampled ADC and T2* maps were compared to corresponding fully-sampled maps. LmD maps were compared to those acquired previously using 2D fully-sampled multiple b-value diffusion MRI in the same volunteer.

Results and Discussion

Global mean ADC and T2* values from the fully-sampled three-interleaved dataset were consistent with reported values for healthy lungs at b=1.6 s/cm2 and 1.5T [2,7]. A summary of the simulated changes in MAE, global ADC, and T2* values with AF is shown in Table 1. The MAE value between fully-sampled and reconstructed ventilation images (b=0, interleave 1 in Figure 1) increased for each AF, but images showed good preservation of main details and no additional artefacts at high AFs (Figure 2 top row). Reconstructed ADC maps exhibited a similar trend; the global mean ADC value was approximately constant with increasing ADC MAE and sample slice ADC maps appeared visually similar (Figure 2 middle row). The T2* MAE also increased with AF, and corresponding global T2* values decreased (maps shown in Figure 2 bottom row). This decrease in T2* with increased undersampling was observed previously in 2D CS experiments at 3T [5] and was attributed to the smoothing properties of the CS reconstruction.

The prospective CS five-interleaved dataset allowed whole-lung coverage, 3D co-registered 3He ventilation images, parametric maps of ADC, LmD, and T2* to be acquired in a single breath-hold (Figure 3). Prospective CS-derived global mean ADC (0.175±0.076 cm2/s) and LmD (209.0±30.0 μm) values agreed with corresponding 3D and 2D fully-sampled equivalent values (0.173±0.082 cm2/s and 207.2±24.6 μm, respectively). The small positive bias in CS-derived values is consistent with previous observations in CS-derived microstructure measurements, and is attributable to the CS reconstruction process [8]. The global mean T2* value for the prospective dataset was 16.40±12.86 ms, smaller than the fully-sampled mean value (23.70±18.38 ms). This was consistent with the CS simulations results; the prospective CS T2* value was in agreement with the simulated T2* value for AF=3 (17.08±13.73 ms).


We have demonstrated that it is feasible to acquire whole-lung coverage, co-registered images of lung ventilation, ADC, LmD, and T2* in a single breath-hold with a 3D multiple-interleaved sequence and CS. Good agreement of ADC and LmD values was obtained between prospective CS and fully-sampled datasets in a healthy volunteer. Prospective T2* values were smaller than fully-sampled equivalents, a result of smoothing from the CS reconstruction. Further work will investigate the difference between CS-derived and fully-sampled T2* values and evaluate the sequence in patients and with 129Xe.


This work was funded by the University of Sheffield, National Institute for Health Research, and Medical Research Council.


[1] Horn, F. C., et al. (2014). J Appl Physiol 116(2): 129-139.

[2] Saam, B. T., et al. (2000). Magn Reson Med 44(2): 174-179.

[3] Parra-Robles, J., et al. (2014). Proc. Intl. Soc. Mag. Reson. Med: 3529.

[4] Chen, X. J., et al. (1999). Magn Reson Med 42(4): 729-737.

[5] Ajraoui, S., et al. (2012). NMR Biomed 25(1): 44-51.

[6] Lustig, M., et al. (2007). Magn Reson Med 58(6): 1182-1195.

[7] Deppe, M. H., et al. (2009). J Magn Reson Imaging 30(2): 418-423.

[8] Chan, H-F., et al. (2015). ESMRMB 2015 Congress: 135.


Figure 1: 3D multiple-interleave sequence schematic. TE1 and TE2 are the two echo times used for T2* mapping, with respective repetition times TR1 and TR2.

Table 1: Summary of 3D 3He CS simulations results. For each acceleration factor (AF), MAE, ADC MAE, and T2* MAE along with global mean ADC and T2* values are shown. AF=1 corresponds to the fully-sampled dataset.

Figure 2: Sample lung slice ventilation images, ADC maps and T2* maps for each acceleration factor obtained from CS simulations.

Figure 3: Three representative lung slice ventilation images, ADC, LmD and T2* maps from the prospective CS dataset. Mean slice values are indicated under each image.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)