Table 4.

Results for centerline prediction tasks.

Method	Prec.	Rec.	Dice
DeepVesselNet-FCN	77.63	82.35	79.92
DeepVesselNet-VNet	65.15	68.87	66.96
DeepVesselNet-UNet	71.28	72.95	72.10
VNet	76.41	73.30	74.82
UNet	71.25	73.61	72.41
Schneider et al.	48.07	86.03	61.68

Results suggest that architectures with sub-sampling layers suffer fall in performance due to loss of fine details which is crucial in centerline prediction.

Best performing methods in each category in bold.