Introduction

Idiopathic pulmonary fibrosis (IPF) is characterized by pathological fibroblastic activity, predominantly in the basal and peripheral lung tissue, leading to destruction of alveolar tissue. Recently, the global apparent diffusion coefficient (ADC) and stretched exponential model derived mean alveolar dimension (Lm_D)¹ from hyperpolarized ³He diffusion-weighted (DW)-MRI was demonstrated to be elevated in lungs afflicted with IPF, indicative of a loss of acinar integrity related to fibrosis². Furthermore, ³He DW-MRI metrics also correlated with visual scoring of interstitial lung disease (ILD) severity on high resolution CT (HRCT) images. However, to elucidate which features of ILD correspond to the observed elevated DW-MRI metrics, a more regional or voxel-wise comparison is required. Image analysis software tools such as CALIPER, developed by the Biomedical Imaging Resource at Mayo Clinic (Rochester, MN, USA), can automatically characterize and quantify HRCT images for patterns of ILD on a voxel-wise level³. Therefore, the aim of this work was to develop a spatial co-registration framework for voxel-wise comparison of ³He DW-MRI and CALIPER classification maps.

Methods

Fifteen participants (mean 71±5 years) with IPF from Chan et al.² were chosen for this retrospective analysis. Each participant had underwent hyperpolarized ³He MRI at 1.5T (GE HDx), including 3D DW-MRI¹ and 3D ventilation with spatially aligned structural ¹H MRI⁴, and HRCT imaging (Figure 1a). The lung parenchyma in HRCT was characterized with CALIPER software into seven classification groups including: normal, ILD patterns (honeycombing, reticular changes and ground-glass opacities), and lower attenuation areas (mild, moderate and severe)³. Reconstructed ³He DW-MRI was spatially co-registered to HRCT indirectly via spatially aligned ³He ventilation and structural ¹H MRI⁵ using Advanced Normalization Tools image registration software⁶ (Figure 1b). ³He DW-MRI was deformed to the spatial domain of the ventilation images, while HRCT was deformed to the spatial domain of structural ¹H MRI. The resultant spatial resolution of co-registered images was approximately 1.8x1.8x5 mm³ (Figure 1c).

Image registration accuracy was assessed by Dice similarity coefficients between the semi-automated binary lung segmentation masks of warped HRCT and structural ¹H, and warped ³He DW-MRI and ³He ventilation, respectively. The same HRCT-¹H deformation was used to warp CALIPER classification maps, and maps of ADC and Lm_D were subsequently derived from the warped ³He multiple b-value DW-MRI (Figure 1d). Overlapping warped HRCT and ADC or Lm_D voxels were compared across all participants for each CALIPER classification group, and statistical differences between each group were assessed with Kruskal-Wallis and Dunn’s multiple comparison tests.

Results and Discussion

Spatial co-registration of HRCT and ³He DW-MRI was successfully implemented in all fifteen IPF participants. Mean Dice similarity coefficients of 0.95±0.01 and 0.91±0.01 were obtained for the HRCT to structural ¹H and ³He DW-MRI to ³He ventilation co-registrations, respectively. Voxel-wise comparison of CALIPER classifications with Lm_D and ADC values in all participants are summarized in Figure 2. Statistically significant differences between ADC and Lm_D values within the different CALIPER classification groups (P<0.001) were observed. Dunn’s multiple comparison tests found significant differences between all CALIPER classification groups with both ADC and Lm_D (P<0.001), except between reticular and moderate lower attenuation area groups for ADC, and between reticular and ground-glass groups for Lm_D. Voxels classified as honeycombing by CALIPER had the largest ADC and Lm_D values out of all the ILD pattern groups; further supporting the hypothesis that elevated hyperpolarized gas DW-MRI metrics are a result of honeycomb cysts².

Small differences in ADC and Lm_D between normal and reticular or ground-glass classification groups may suggest these interstitial changes have minimal accompanying acinar microstructural changes. However, when the Lm_D values (mean=274±35 µm) within the normal classified CALIPER group are compared to ¹²⁹Xe Lm_D⁷ (mean=234±24 µm) in seven older healthy volunteers (mean 66±3 years), a statistically significant difference was observed (P<0.001) (Figure 3). This perhaps suggests that hyperpolarized gas DW-MRI is sensitive to aspects of acinar microstructural changes in IPF lung disease such as microscopic cysts that are not resolved by HRCT.

To further explore this, IPF Lm_D maps with voxels greater than 272 µm, corresponding to the 95% upper limit of older healthy Lm_D values, were classified as abnormal. Lm_D maps classified for abnormality were compared to CALIPER maps where all abnormal classifications were grouped together. Bland-Altman analysis of the percentage of abnormal classified voxels between Lm_D and CALIPER obtained a mean bias of 5% towards Lm_D across all IPF participants (Figure 4a), suggesting more abnormal microstructural changes are detected by DW-MRI. This bias is demonstrated in a representative IPF participant with 42.3% of the lung classified abnormal with CALIPER, in contrast to the 84.6% abnormal as determined from Lm_D (Figure 4b).

Conclusion

A framework for co-registration of ³He DW-MRI and HRCT images was developed and implemented in fifteen IPF participants. Voxel-wise comparison of derived ADC and Lm_D values with CALIPER classifications of ILD patterns confirmed the largest DW-MRI metrics are in honeycombing voxels. Lm_D in voxels classified as normal by CALIPER were larger than those in age-matched healthy volunteers, suggesting DW-MRI may detect microstructural changes in regions determined as macroscopically normal by HRCT. Future work will involve co-registering ¹²⁹Xe DW-MRI with HRCT to explore classification differences in other ILD subtypes.

Acknowledgements

This work was supported by NIHR grant NIHR-RP-R3-12-027 and MRC grant MR/M008894/1. The views expressed in this work are those of the author(s) and not necessarily those of the NHS, the National Institute for Health Research or the Department of Health.