notesum.ai

Published at November 26

eess.IV

cs.CV

Released Date: November 26, 2024

Authors: Bastian Wittmann¹, Yannick Wattenberg², Tamaz Amiranashvili³, Suprosanna Shit¹, Bjoern Menze¹

Aff.: ¹University of Zurich; ²ETH Zurich; ³Technical University of Munich

Task	Model	OCTA [54, 17]		BvEM [51]		SMILE-UHURA [9]		MSD8 [2]
Task	Model	$\text{Dice}\uparrow$	$\text{clDice}\uparrow$	$\text{Dice}\uparrow$	$\text{clDice}\uparrow$	$\text{Dice}\uparrow$	$\text{clDice}\uparrow$	$\text{Dice}\uparrow$	$\text{clDice}\uparrow$
zero-shot	tUbeNet [24]	36.01	23.64	10.03	11.17	48.32	36.85	5.13	5.84
	VISTA3D [23]	13.60	3.72	0.94	2.03	5.05	1.62	23.83	20.25
	SAM-Med3D [53]	6.74	6.56	5.98	7.38	2.12	1.66	7.94	7.89
	MedSAM-2 [61]	28.56	15.76	10.92	12.27	3.85	5.46	14.53	14.27
	vesselFM (ours)	46.94	67.07	67.49	62.04	74.66	75.27	29.69	36.14
one-shot	tUbeNet [24]	38.09	59.37	10.75	11.53	57.67	53.25	13.66	15.41
	VISTA3D [23]	51.24	25.69	8.25	15.04	56.53	42.42	31.73	32.94
	SAM-Med3D [53]	38.33	54.90	49.47	52.14	38.57	36.94	29.29	36.78
	MedSAM-2 [61]	56.68	50.95	24.07	24.69	19.78	11.87	30.21	23.89
	vesselFM (from scratch)^∗	65.57	73.79	63.85	39.55	37.99	45.72	27.13	29.48
	vesselFM (ours)	72.10	83.73	78.27	79.91	76.43	78.36	36.88	48.65
few-shot	tUbeNet [24]	41.61	57.98	5.41	10.22	56.31	49.28	17.67	18.97
	VISTA3D [23]	54.25	32.59	24.04	38.10	61.17	51.05	41.90	46.45
	SAM-Med3D [53]	37.85	56.94	57.86	66.04	46.59	44.63	31.30	35.48
	MedSAM-2 [61]	56.96	51.99	18.76	19.66	58.15	42.72	29.24	22.38
	vesselFM (from scratch)^∗	67.37	75.79	63.03	56.69	50.51	58.77	32.64	36.03
	vesselFM (ours)	75.70	84.03	78.11	84.54	78.77	79.37	45.04	57.25
^∗Model not pre-trained on $\mathcal{D}_{\text{real}}$ , $\mathcal{D}_{\text{drand}}$ , and $\mathcal{D}_{\text{flow}}$ .