Test-Retest Reliability of Diffusion Tensor Imaging in Huntington's Disease

Abstract

Diffusion tensor imaging (DTI) has shown microstructural abnormalities in patients with Huntington's Disease (HD) and work is underway to characterise how these abnormalities change with disease progression. Using methods that will be applied in longitudinal research, we sought to establish the reliability of DTI in early HD patients and controls. Test-retest reliability, quantified using the intraclass correlation coefficient (ICC), was assessed using region-of-interest (ROI)-based white matter atlas and voxelwise approaches on repeat scan data from 22 participants (10 early HD, 12 controls). T1 data was used to generate further ROIs for analysis in a reduced sample of 18 participants. The results suggest that fractional anisotropy (FA) and other diffusivity metrics are generally highly reliable, with ICCs indicating considerably lower within-subject compared to between-subject variability in both HD patients and controls. Where ICC was low, particularly for the diffusivity measures in the caudate and putamen, this was partly influenced by outliers. The analysis suggests that the specific DTI methods used here are appropriate for cross-sectional research in HD, and give confidence that they can also be applied longitudinally, although this requires further investigation. An important caveat for DTI studies is that test-retest reliability may not be evenly distributed throughout the brain whereby highly anisotropic white matter regions tended to show lower relative within-subject variability than other white or grey matter regions.

Bland-Altman plot for FA values

Bland Altman plots of fractional anisotropy (FA) values to visually assess agreement, systematic bias and proportional bias in scanning technique for T1 ROIs (caudate, putamen, white matter and corpus callosum), for early HD patients (blue triangles) and controls (red circles). FA is a relative value derived from the diffusion tensor, where 0 indicates perfectly isotropic tensor dimensions (i.e. a sphere) and 1 indicates the maximum theoretical level of anisotropy.

Bland-Altman plot for MD values

Bland Altman plots of mean diffusivity (MD) values to visually assess agreement, systematic bias and proportional bias in scanning technique for T1 ROIs (caudate, putamen, white matter and corpus callosum), for early HD patients (blue triangles) and controls (red circles).

Bland-Altman plot for AD values

Bland Altman plots of axial diffusivity (AD) values to visually assess agreement, systematic bias and proportional bias in scanning technique for T1 ROIs (caudate, putamen, white matter and corpus callosum), for early HD patients (blue triangles) and controls (red circles).

Bland-Altman plot for RD values

Bland Altman plots of radial diffusivity (RD) values to visually assess agreement, systematic bias and proportional bias in scanning technique for T1 ROIs (caudate, putamen, white matter and corpus callosum), for early HD patients (blue triangles) and controls (red circles).

Voxelwise ICC maps of FA

Voxelwise distribution of reliability metrics for fractional anisotropy (FA). Panel A) shows sagittal, coronal and axial slices of the mean FA image created during the image processing, included for anatomical reference. B) Equivalent three slices for the intraclass correlation coefficient (ICC) of FA in early HD patients. Higher values reflect higher test-retest reliability. C) ICC of FA in healthy controls.

Voxelwise ICC maps for MD

Voxelwise distribution of reliability metrics for mean diffusivity (MD). Panel A) shows sagittal, coronal and axial slices of the mean FA image created during the image processing, included for anatomical reference. B) Equivalent three slices for the intraclass correlation coefficient (ICC) of MD in early HD patients. Higher values reflect higher test-retest reliability. C) ICC of MD in healthy controls.

Voxelwise ICC maps of AD

Voxelwise distribution of reliability metrics for axial diffusivity (AD). Panel A) shows sagittal, coronal and axial slices of the mean FA image created during the image processing, included for anatomical reference. B) Equivalent three slices for the intraclass correlation coefficient (ICC) of AD in early HD patients. Higher values reflect higher test-retest reliability. C) ICC of AD in healthy controls.

Voxelwise ICC maps of RD

Voxelwise distribution of reliability metrics for radial diffusivity (RD). Panel A) shows sagittal, coronal and axial slices of the mean FA image created during the image processing, included for anatomical reference. B) Equivalent three slices for the intraclass correlation coefficient (ICC) of RD in early HD patients. Higher values reflect higher test-retest reliability. C) ICC of RD in healthy controls.