Table 1.

Training, validation, and testing performance of the three FCNN models in terms of DSC. ${\mathrm{DSC}}_{\mathrm{Tr}}$, ${\mathrm{DSC}}_{\mathrm{v}}$, and ${\mathrm{DSC}}_{\mathrm{Ts}}$ are the means and [95% confidence intervals] of DSC values across the training, validation, and test datasets, respectively.

FCNN model	Training reference (# of samples)	Validation/test reference (# of samples)	${\mathrm{DSC}}_{\mathrm{Tr}}$ (%)	${\mathrm{DSC}}_{\mathrm{V}}$ (%)	${\mathrm{DSC}}_{\mathrm{Ts}}$ (%)
Single-reference training (model I)	A (120)	A (9/23)	92.3 [91.9, 92.7]	82.8 [78.3, 86.3]	82.2 [79.5, 84.1]
Single-reference training (model I)	A (120)	B (9/23)	88.7 [87.6, 89.7]	82.3 [78.3, 85.4]	82.2 [80.2, 83.9]
Single-reference training (model II)	B (120)	B (9/23)	90.7 [90.1, 91.3]	82.3 [78.4, 85.6]	82.4 [80.4, 84.2]
Single-reference training (model II)	B (120)	A (9/23)	89.2 [88.3, 90.0]	80.9 [77.7, 83.5]	82.8 [80.4, 84.5]
Multireference training (model III)	A and B (240)	A (9/23)	92.5 [91.9, 93.1]	82.6 [80.0, 84.9]	83.1 [80.3, 85.0]
Multireference training (model III)	A and B (240)	B (9/23)	90.9 [90.2, 91.5]	83.4 [80.6, 85.6]	82.6 [80.1, 84.6]