Table 2.

Results from pretraining DeepVesselNet-FCN on synthetic data and finetuning with the training set from the clinical MRA vs. training DeepVesselNet-FCN from scratch on clinical MRA.

Method	Precision	Recall	Dice	Iterations
With pretraining	86.44	86.93	86.68	1200
Without pretraining	85.87	86.92	86.39	1800

Iterations refers to training iterations required for the network to converge. Although the result in Dice score are not very different, it is clear that the pre-training on synthetic data leads to an earlier convergence of the network.