TABLE VI:

Quantitative comparison of different semi-supervised methods for PF lesion segmentation. R_an: Ratio of annotated images in the training set.

R an	Method	Dice (%)	RVE (%)	HD95 (mm)
10%	Baseline	54.73±16.34	45.83±24.99	25.79±18.71
	Fan et al. [13]	60.44±15.86	42.40±21.06	17.44±16.00
	Bai et al. [32]	65.03±18.01	32.98±23.16	16.82±14.82
	Cui et al. [33]	61.86±16.20	41.16±20.78	16.72±16.18
	UA-MT [34]	63.94±15.72	38.47±21.22	15.14±15.34
	I-CRAWL	67.18±14.02	32.12±18.50	11.85±9.82
20%	Baseline	70.36±10.14	27.96±21.72	10.87±9.28
	Fan et al. [13]	71.62±10.32	27.57±19.09	10.54±8.41
	Bai et al. [32]	72.13±8.97	27.38±20.39	8.38±4.75
	Cui et al. [33]	70.39±10.42	26.63±24.85	9.48±8.82
	UA-MT [34]	70.46±9.53	27.14±23.17	8.25±4.93
	I-CRAWL	73.04±10.14	26.12±20.37	7.92±4.87 †
50%	Baseline	72.09±12.54	26.49±23.19	9.38±6.43
	Fan et al. [13]	72.69±11.41	26.46±17.71	9.19±6.45
	Bai et al. [32]	72.92±10.41	25.71±19.97†	7.45±4.98†
	Cui et al. [33]	72.72±12.44	24.57±19.96†	8.11±5.76†
	UA-MT [34]	73.40±11.48	24.67±19.08†	7.86±4.56†
	I-CRAWL	74.02±13.77 †	23.93±16.75 †	7.24±5.03 †
Full Supervision		74.58±10.99	23.07± 15.79	7.03±4.09

^†denotes there is no significant difference from full supervision (p-value > 0.05).