Background/Purpose: Takayasu arteritis (TA) is the most common
large vessel vasculitis and may presents in
childhood. Different angiographic classification systems based on adults with
TA have been proposed. However, in clinical practice many children with TA do
not fit into any of these classification categories. We sought to study the
vascular involvement patterns in children with TA using unsupervised
data-driven pattern recognition techniques to better characterize homogeneous
subgroups of children with TA. 

Methods: 29 children with
TA were included from a single-center cohort. Non-negative matrix factorization
(NMF) was conducted on baseline imaging data (MR angiography or conventional
angiogram) to identify unique factors that distinguish between various types of
vascular involvement in children. A feature of NMF is that the factors
naturally give rise to clusters of children. 

Results: NMF identified 3
factors and clusters characterized by the following distinct pattern of vessel
involvement: (1) widespread vessel involvement including the aortic
arch/ascending aorta, coronary artery, neck vessels, upper extremity vessels,
abdominal aorta, renal and iliac arteries (7 patients); (2) the abdominal aorta
its branches (coeliac trunk, superior mesenteric artery, bilateral renal) and
the iliac arteries (14 patients); and (3) the entire aorta and its left-sided
neck vessels (subclavian, carotid and vertebral
artery), and involvement of the superior mesenteric and bilateral pulmonary
arteries. The renal arteries are spared (8 patients). 

Conclusion: Our data-driven,
assumption-free analysis recovered unique patterns of vascular
involvement that are distinct from the current adult-based angiographic
classifications. These patterns may help to define homogeneous subgroups of children
with TA and to guide the challenging management.

Figure 1: Contributions of individual arteries to factors. Shades of gray
indicate the degrees to which individual arteries (y-axis) contribute to
factors (x-axis). Shades nearer black indicate stronger contributions. Traces
(lines represent 0 loadings and dots indicating absolute loadings) provide an
alternative means of visualizing the loadings. Arrows indicate the side of
body.