Table 7.

Average symmetric surface distance (mm) and standard deviation (in parentheses) of the proposed method and the fully-supervised method (FS-model) trained with ground truth labels.

Test set	Method	Mandible	Parotid-R	Parotid-L	Submand-R	Submand-L
TCIA	DIR-majority-vote	0.43 (±0.15)	2.45 (±0.90)	2.34 (±0.82)	2.85 (±0.89)	2.71 (±1.13)
	FS-model	0.40 (±0.08)	2.56 (±1.33)	2.49 (±1.21)	2.39 (±0.78)	1.86 (±0.61)
	Ours(N10M50)	0.40 (±0.14)	2.11 (±0.77)	2.15 (±0.82)	2.30 (±0.59)	2.20 (±1.04)
UTSW	DIR-majority-vote	0.83 (±0.23)	3.33 (±0.86)	3.48 (±1.35)	7.71 (±8.49)	3.10 (±0.93)
	FS-model	0.77 (±0.32)	2.79 (±0.97)	2.25 (±0.44)	7.85 (±9.52)	2.91 (±1.25)
	Ours(N10M50)	0.74 (±0.26)	2.77 (±0.95)	2.47 (±0.75)	6.76 (±7.57)	2.63 (±1.12)
PDDCA	DIR-majority-vote	0.44 (±0.15)	2.03 (±0.97)	2.05 (±0.60)	2.38 (±1.06)	2.14 (±0.74)
	FS-model	0.45 (±0.10)	1.99 (±1.01)	1.95 (±0.53)	2.01 (±0.58)	1.98 (±0.56)
	Ours(N10M50)	0.44 (±0.13)	1.87 (±0.85)	1.92 (±0.46)	2.14 (±0.74)	2.03 (±0.72)