Whole Lung Morphometry with Hyperpolarised 3He Gas Diffusion MRI - 3D Multiple b-value Acquisition and Compressed Sensing
Ho-Fung Chan1, Neil J. Stewart1, Juan Parra-Robles1, Guilhem J. Collier1, and Jim M. Wild1

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


Compressed sensing (CS) was implemented to reduce scan time and facilitate acquisition of 3D multiple b-value 3He diffusion-weighted (DW) MRI data for whole lung morphometry. A fully-sampled 3D DW-MRI dataset was retrospectively undersampled using CS simulations to determine optimal k-space undersampling patterns. Whole lung morphometry measurements derived from prospective 3-fold undersampled 3D multiple b-value DW-MRI were compared to 3D and 2D fully-sampled equivalents. Good agreement was obtained between lung morphometry measurements indicating 3D multiple b-value 3He DW-MRI with CS can provide reliable measurements of whole lung morphometry within a single breath-hold.


Diffusion-weighted MRI (DW-MRI) with hyperpolarised gases has been shown to be sensitive to changes in lung microstructure [1,2]. Fitting a fractional diffusion stretched exponential model to multiple b-value 3He DW-MRI data can provide a robust quantitative measure of mean alveolar dimension (LmD) [3]. Previous 2D 3He DW-MRI sequences allow acquisition of up to six slices and b-values in a single breath-hold (~15s), but do not provide whole lung coverage required for lung morphometry. 3D diffusion sequences with similar slice thickness and b-values would require long breath-holds to obtain whole lung coverage. In this work, compressed sensing (CS) [4] was implemented to reduce scan time and facilitate 3D multiple b-value 3He DW-MRI of the lung within a single breath-hold.


Fully-sampled 3D DW-MRI images of a healthy subject’s lungs (M, 30y) were acquired on a GE HDx 1.5T MR scanner with 500mL of 3He (~25% polarisation) in a 22s breath-hold. Imaging parameters: 3D SPGR, two b-values=0, 1.6 s/cm2, diffusion time Δ=1.6ms, 96x78x24 matrix, FOV= 40x32.5x28.8 cm3, TE/TR= 4.2/5.7ms, flip angle=2°, bandwidth=±31.25 kHz. Pseudo-random under-sampled k-space patterns were generated for acceleration factors (AF) of 2 to 5. These patterns were used to retrospectively undersample the fully-sampled images, and a non-linear algorithm [4] was used for image reconstruction. Reconstructions were optimised by minimising the mean absolute error (MAE) between reconstructed and fully-sampled apparent diffusion coefficient (ADC) maps (MAEADC). Optimal patterns for each AF were selected and resulting MAE and mean ADC values were compared.

To enable quantitative lung morphometry, a prospective 3D 3He CS DW-MRI dataset was acquired with four b-values (0, 2.4, 4.8, 7.2 s/cm2) and AF=3 in a 15s breath-hold, and compared to equivalent fully-sampled 2D multi-slice DW-MRI (same slice thickness and b-values) and fully-sampled 3D DW-MRI (b=0, 2.4 s/cm2) from the same healthy subject. ADC maps were calculated from the first two b-values, whilst LmD was calculated from all four b-values using the stretched exponential model [3]. For further validation, an additional five healthy volunteers and one COPD patient (FEV1=31.2% predicted) were each imaged with 3D multiple b-value CS DW-MRI, fully-sampled 2D and 3D DW-MRI.

Results and Discussion

The MAE between 3D 3He magnitude (b=0) lung images, reconstructed from optimal k-space patterns, and fully-sampled images was found to increase with AF. However, reconstructed images showed no visual artefacts and good preservation of main details. Some blurring of the edges of the lungs was observed on magnitude images at higher AFs due to the greater undersampling of high frequency components. ADC maps were generated (from b=0 and 1.6 s/cm2 images) for each AF and compared to those of the fully sampled dataset (Table 1). A very small increase in mean ADC value with AF was observed, however the maximum global increase in mean ADC (+1% for AF=3 & 5) remained within the range of healthy lung values (~0.20±0.08 cm2/s) at b=1.6 s/cm2 [1,2]. The small increase in global ADC and similar appearance of sample slice ADC maps and histograms (Figure 1) indicated good preservation of quantitative information.

The mean global ADC and LmD values of the prospective 3D CS dataset were 0.191±0.054 cm2/s and 0.226±0.023 μm respectively (Table 2). The CS-derived values were ~3% larger than their fully-sampled equivalents from 2D and 3D datasets. This mismatch was within the standard deviation of the global values, and well below reported ADC values in emphysema lungs (~0.47±0.18 cm2/s) [1,2]. The LmD value is consistent with reported healthy human lung mean linear intercept values obtained from histology [5].

Good agreement was obtained from a slice-by-slice comparison between ADC and LmD values derived from fully-sampled and CS datasets of the six healthy volunteers and one COPD patient. Again, a small positive bias was observed in 3D CS-derived ADC and LmD values and is likely the result of the CS reconstruction process. These results indicate that 3D multiple b-value DW-MRI could be used clinically to assess pathological changes in the microstructure of the whole lung in a single breath-hold. This potential is demonstrated in Figure 2, which depicts images of a healthy and COPD subject from 3D multiple b-value DW-MRI with CS, and LmD values calculated across the entire lung reflect the expected alveolar size of each subject [5].


We have demonstrated that it is feasible to acquire 3D multiple b-value 3He DW-MRI images for whole lung morphometry in a single breath-hold using compressed sensing. Good agreement was obtained between ADC and LmD values derived from prospective CS and fully-sampled datasets indicating that 3D DW-MRI with CS can provide reliable measurements of whole lung morphometry.


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


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

[2] Salerno, M. et al. 2002. Radiology 222(1):252-260.

[3] Parra-Robles, J. et al. 2014. Proc ISMRM:3529.

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

[5] Woods, J. et al. 2006. Magn Reson Med 56(6):1293-1300.


Table 1: A summary of 3D 3He CS simulations ADC results. Global ADC (cm2/s), sample slice ADC (cm2/s), and MAEADC values are shown for each acceleration factor (AF). AF=1 corresponds to the fully-sampled dataset. MAEADC increases with acceleration factor, however global and slice ADC values are consistent between AFs.

Figure 1: Reconstructed lung slice ADC maps and histograms for 3D 3He CS simulations. (Top row) ADC maps for each acceleration factor. (Bottom row) Corresponding lung slice ADC histograms. The similar ADC maps and histograms indicate the preservation of quantitative information in the CS simulations.

Table 2: Summary of single volunteer validation of ADC and LmD values. Global ADC and LmD values are shown for the three imaging methods. The 3D 3He data was acquired with only two interleaves therefore there is no LmD value calculated.

Figure 2: Four sample slice LmD maps calculated from 3D multiple b-value 3He DW-MRI with CS acquired in a healthy (Top row) and COPD subject (Bottom row).

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