Table 3.

Results for vessel segmentation.

Dataset	Method	Prec.	Rec.	Dice
Synthetic	DeepVesselNet-FCN	99.84	99.87	99.86
	DeepVesselNet-VNet	99.54	99.59	99.56
	DeepVesselNet-UNet	99.48	99.42	99.45
	VNet	99.48	99.50	99.49
	UNet	99.57	99.52	99.55
	Schneider et al.	99.47	99.56	99.52
TOF MRA	DeepVesselNet-FCN (finetuned)	86.44	86.93	86.68
	DeepVesselNet-FCN (pretrained)	82.76	80.25	81.48
	DeepVesselNet-VNet (finetuned)	85.00	83.51	84.25
	DeepVesselNet-VNet (pretrained)	83.32	77.12	80.10
	DeepVesselNet-UNet (finetuned)	83.56	85.18	84.36
	DeepVesselNet-UNet (pretrained)	83.48	79.27	81.32
	VNet (finetuned)	84.34	85.62	84.97
	VNet (pretrained)	82.41	75.82	78.98
	UNet (finetuned)	84.02	85.35	84.68
	UNet (pretrained)	83.16	80.23	81.67
	Schneider et al.	84.81	82.15	83.46
	Forkert et al.	84.99	73.00	78.57
μCTA	DeepVesselNet-FCN	96.72	95.82	96.27
	DeepVesselNet-VNet	95.83	96.18	96.01
	DeepVesselNet-UNet	95.85	96.06	95.95
	VNet	95.25	95.84	95.55
	UNet	95.27	95.71	95.49
	Schneider et al.	95.15	91.51	93.30

TOF MRA are evaluated within the brain region only.

Pretrained results refers to the scores we obtained on the test set after pretraining, and finetuned results are scores after finetuning with annotated data available for TOF-MRA.

Best performing method in each metric are show in bold.