Dynamic Cerebral Contrast-enhanced Perfusion MRI - Biomedical Signal Analysis: Contemporary Methods and Applications - page 308

The area of the cerebrovascular insult in the right basal ganglia for
subject 1 is clearly represented mainly by cluster #7 and also by cluster
#8, which contains other essential areas. The small CTC amplitude is
evident (i.e., the small cluster-specific rCBV, the rCBF, and the large
MTT). Cluster #3 and #4 contain peripheral and adjacent regions.
Clusters #1, #2, #12, #14, and #16 can be attributed to larger vessels
located in the sulci. Figure 11.2 shows the large amplitudes and apparent
recirculation peaks in the corresponding cluster-specific CTCs .
Further, clusters #2, #12, and #11 represent large, intermediate,
and small parenchymal vessels respectively of the nonaffected left side
showing subsequently increasing rCBV and smaller recirculation peaks.
The clustering technique unveils even subtle differences of contrast
agent first-pass times: small time-to-peak differences of clusters #1,
#2, #12, #14, and #16 enable discrimination between left- and right-
side perfusion. Pixels corresponding to regions supplied by a different
arterial input tend to be collected into separate clusters: For example,
clusters #6 and #11 contain many pixels that can be attributed to the
supply region of the left middle cerebral artery, whereas clusters #3
and #4 include regions supplied by the right middle cerebral artery.
Contralateral clusters #6 and #11 versus #3 and #4 show different
cluster-specific MTTs as evidence for an apparent perfusion deficit at
the expense of the right-hand side.
The diffusion-weighted image in figure 11.3a visualizes the structural
lesion. Figs. 11.3b, c, and d represent the conventional pixel-based MTT,
rCBF, and rCBV maps at the same slice position in the region of the
right basal ganglia. A visual inspection of the clustering results in Figs.
11.1 and 11.2 (clusters #7 and #8) shows a close correspondence with
the findings of these parameter maps. In addition, the unsupervised and
self-organized clustering of pixels with similar signal dynamics allows a
deeper insight in the spatiotemporal perfusion properties .
Figure 11.4 visualizes a method for comparative analysis of clustering
results with regard to side differences of brain perfusion. The best-
matching cluster #7, with the diffusion-weighted image corresponding
to the infarct region in figure 11.1 is shown in figure 11.4a.
To better visualize the perfusion asymmetry between the affected
and the nonaffected sides, a spatially connected region of interest (ROI)
can be obtained from the clustering results by spatial low-pass filter-
ing and thresholding of the given pixel cluster. The resulting ROI is