Background: Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has limitations in the assessment of subtle blood-brain barrier (BBB) permeability changes mainly due to noise levels comparable to true signal values. We evaluate the use of Graph-variate signal analysis (GVSA) in selecting regions of interest across the normal-appearing white matter, hypothesised to be reflective of true signal levels in this tissue, to ameliorate these problems.
Materials and methods: We used data from 40 small vessel disease (SVD) patients who had a recent small subcortical infarct (RSSI), obtained at 1-3 months post-stroke. Images of normal-appearing white matter were sparsified into a coherent spatial graph representation by down-sampling via 3D skeletonisation. We then mapped skeletal voxel clusters and pathways into nodes and connecting edges, respectively. The post-contrast time-series at each node was the average of brain fluid flow over its 2-voxel-radius neighbourhood. We averaged the instantaneous correlations between signals connected through the graph to get one value for each node and time-point. Signal activations comprised the 5% highest values over the time-node array. We used the Kruskal-Wallis test to determine whether the Euclidean distances between the spatio-temporal activations and RSSI discriminated patients with different baseline vascular risk factors or SVD burden, and/or patterns of RSSI evolution.
Results: The median distances between the temporal activations in the graph and the centroid of the RSSI discriminated patients stratified according to 1)brain microbleeds, white matter hyperintensities and perivascular spaces, 2)presence/absence and spatial proximity of these imaging features and old stroke lesions with the RSSI (all p<0.001).
Conclusions: The DCE-MRI signal was clearer in locations spatially related to the RSSI with distinctive patterns related to the SVD markers, as expected. Thus, GVSA is promising for identifying regions of interest to study MRI signal dynamics when changes are subtle and signal levels are comparable to noise.

SPATIO-TEMPORAL TOPOLOGY OF THE POST-CONTRAST SIGNAL ENHANCEMENT IN NORMAL-APPEARING WHITE MATTER 1-3 MONTHS POST-STROKE