Quantifying the effects of normal ageing on white matter structure using unsupervised tract shape modelling

Abstract

Quantitative tractography may provide insights into regional heterogeneity of changes in white matter structure in normal ageing. Here we examine how brain atrophy and white matter lesions affect correlations between tract shape, tract integrity and age in a range of frontal and non-frontal tracts in 90 non-demented subjects aged over 65 years using an enhanced version of probabilistic neighbourhood tractography. This novel method for automatic single seed point placement employs unsupervised learning and streamline selection to provide reliable and accurate tract segmentation, whilst also indicating how the shape of an individual tract compares to that of a predefined reference tract. There were significant negative correlations between tract shape similarity to reference tracts derived from a young brain white matter atlas and age in genu and splenium of corpus callosum. Controlling for intracranial and lateral ventricle volume, the latter of which increased significantly with age, attenuated these correlations by 40% and 84%, respectively, indicating that this age-related change in callosal tract topology is significantly mediated by global atrophy and ventricular enlargement. In accordance with the "frontal ageing" hypothesis, there was a significant positive correlation between mean diffusivity (D) and age, and a significant negative correlation between fractional anisotropy (FA) and age in corpus callosum genu; correlations not seen in splenium. Significant positive correlations were also observed between D and age in bilateral cingulum cingulate gyri, uncinate fasciculi and right corticospinal tract. This pattern of correlations was not, however, reproduced when those subjects with significant white matter lesion load were analyzed separately from those without. These data therefore suggest that brain atrophy and white matter lesions play a significant role in driving regional patterns of age-related changes in white matter tract shape and integrity.

Figure 1

Maximum intensity projections of standard space group maps of the segmented fasciculi-of-interest overlaid on an MNI white matter volume without (first column) and with (second column) streamline selection. The tracts displayed are genu (a, b; axial), splenium (c, d; axial), and left cingulum cingulate gyrus (e, f; sagittal), left corticospinal tract (g, h; coronal) and left uncinate fasciculus (i, j; axial). The reference tract is superimposed in green in each case. Note how the significant branches heading into unrelated structures evident in the first column are almost entirely removed using streamline selection in the second column, thereby significantly improving tract segmentation.

Figure 2

Maximum intensity projections of standard space group maps of white matter lesion location overlaid on an MNI whole brain volume for the 32 subjects with significant lesion load. The colour bar indicates the number of subjects with white matter lesions at each standard space location.

Figure 3

Maximum intensity projections of standard space group maps of the intersection between segmented fasciculi-of-interest and white matter lesion location overlaid on an MNI white matter volume for the 32 subjects with significant lesion load. The tracts displayed are genu and splenium (a, b; axial), left and right cingulum cingulate gyri (c, d; sagittal), left and right corticospinal tracts (e, f; coronal), and left and right uncinate fasciculi (g, h; axial). The windowing level is the same across all eight tracts, with ‘hotter’ colours indicating larger values of the product of the tract and white matter lesion group maps.

Figure 4

Scatter plot of tract-averaged fractional anisotropy (FA) against mean diffusivity (〈D〉) within all segmented tracts.

Figure 5

Two-dimensional axial projections of the genu (first row) and splenium (second row) overlaid on maps of FA for the atlas reference tracts and the best match tracts for three representative female subjects without significant white matter lesion load aged 66 (a, d), 78 (b, e) and 84 (c, f) years. Values of R are given for each tract, while LVV as a percentage of ICV is given in parentheses. The colours represent the proportion of probabilistic streamlines generated from the seed point (green star) passing through each voxel, as indicated by the colour bars.

Figure 6

Correlations between tract-averaged mean diffusivity (a, c, e), fractional anisotropy (b, d, f) and age for genu (first row), splenium (second row) and right uncinate fasciculus (third row). Open circles indicate subjects with few or no lesions in deep and periventricular white matter, while closed circles indicate subjects with significant lesion load.