Table 3.

Overview of existing methods.

	Pre-processing	No. patients	No. segments	Execution time
Location-based
Takemura et al.	re, be, pt	15	12	n.r.
Dunås et al.	re, be	132	14	13 m (CPU)
Shen et al.	re, be, pt	194	9	15.7 s (GPU)
Graph-based
Bilgel et al.	be, tr	30	15	n.r.
Robben et al.	re, tr	50	9	510 s (CPU)
Bogunovic et al.	re, be	50	9	n.r.
Chen et al.	re, be, nm, tr	729	22	0.1 s (GPU)
Other
Zhang et al.	nm, mm	109	9	n.r.
Proposed
Proposed detailed	–	242	24	10 s (GPU)
Proposed detailed + segment washing	be	242	24	10 s (GPU)
Proposed aggregated	–	242	11	8 s (GPU)
Proposed aggregated + segment washing	be	242	11	8 s (GPU)

Methods are compared in terms of required pre-processing steps from the raw image to produce the proposed anatomical labeling (pre-processing), total number of patients included in the study (number of patients), number of arterial segments considered and labeled by the model (number of segments), and best execution time reported by authors (execution time). Due to being a prerequisite of each method, vessel segmentation and centerline extraction are omitted from the list of required pre-processing steps. re, registration; be, bifurcation extraction; pt, additional reference point tracking (Takemura et al.—tracking of center of CoW, Shen et al.—key point tracking); tr, spatial transformation (Robben et al.—scale-space transformation, Chen et al., Zhu et al., and Bilgel et al.—isotropic reslicing); nm, intensity normalization; mm, mesh modeling; nr, not reported.