Table 2.

Comparison with state-of-the-art methods using full annotation on the OAI ZIB dataset following the data split in Ambellan et al. (2019). Paired t-test values indicate the significance status of the improved performance of our method vs. the ensemble method (Zheng et al., 2020b).

	Femoral bone		Femoral cartilage		Tibial bone		Tibial cartilage
	DSC (%)	ASSD (mm)	DSC (%)	ASSD (mm)	DSC (%)	ASSD (mm)	DSC (%)	ASSD (mm)
CNN-SSM (Ambellan et al., 2019)a	98.60 ± 0.30	0.170 ± 0.050	89.90 ± 3.60	0.160 ± 0.070	98.50 ± 0.33	0.180 ± 0.060	85.60 ± 4.54	0.230 ± 0.120
Ensemble method (Zheng et al., 2020b)	98.40 ± 0.32	0.197 ± 0.054	88.13 ± 2.57	0.193 ± 0.054	98.53 ± 0.34	0.183 ± 0.067	84.64 ± 4.24	0.215 ± 0.085
TransUNet 3D (Chen et al., 2021)	98.33 ± 0.32	0.212 ± 0.068	88.66 ± 2.70	0.183 ± 0.055	98.53 ± 0.36	0.206 ± 0.205	83.86 ± 4.97	0.235 ± 0.101
Attention UNet 3D (Oktay et al., 2018)	98.41 ± 0.34	0.201 ± 0.068	88.90 ± 2.75	0.178 ± 0.056	98.56 ± 0.36	0.181 ± 0.084	84.99 ± 4.67	0.224 ± 0.096
UNet++ 3D (Zhou et al., 2018)	98.24 ± 0.42	0.266 ± 0.134	88.22 ± 2.77	0.192 ± 0.059	98.31 ± 0.53	0.856 ± 1.251	84.31 ± 5.04	0.242 ± 0.118
Our KCB-Net method	98.79 ± 0.30	0.181 ± 0.054	90.33 ± 2.84	0.152 ± 0.051	98.84 ± 0.34	0.164 ± 0.058	86.10 ± 4.50	0.212 ± 0.090
p-value	≪0.001	≪0.001	≪0.001	≪0.001	≪0.001	≪0.001	≪0.001	≪0.001

^aMarks the row in which the results are from the original paper.