Table 5.

Comparison with state-of-the-art methods using full annotation on the Iowa dataset. Paired t-test values indicate the significance status of the improved performance of our method vs. the ensemble method (Zheng et al., 2020b). “–” denotes that the corresponding results were not reported in the original paper.

	Femural bone		Femoral cartilage		Tibial bone		Tibial cartilage
	DSC (%)	ASSD (mm)	DSC (%)	ASSD (mm)	DSC (%)	ASSD (mm)	DSC (%)	ASSD (mm)
LOGISMOS-4D (Kashyap et al., 2017)a	–	–	−	0.550 ± 0.110	–	–	–	0.600 ± 0.140
Ensemble method (Zheng et al., 2020b)	94.86 ± 1.02	0.649 ± 0.269	84.38 ± 2.40	0.467 ± 0.170	94.40 ± 1.23	0.676 ± 0.234	81.96 ± 4.59	0.577 ± 0.170
CML (Tan et al., 2019)	94.95 ± 1.23	0.651 ± 0.173	83.63 ± 2.33	0.611 ± 0.152	94.44 ± 1.24	0.612 ± 0.204	81.51 ± 4.91	0.583 ± 0.152
UNet++ 3D (Zhou et al., 2018)	95.68 ± 0.91	0.691 ± 0.182	83.29 ± 2.75	0.487 ± 0.123	94.92 ± 1.71	0.658 ± 0.236	81.30 ± 4.30	0.421 ± 0.116
Attention UNet 3D (Oktay et al., 2018)	95.80 ± 1.14	0.645 ± 0.244	84.42 ± 2.71	0.480 ± 0.124	95.09 ± 1.65	0.635 ± 0.221	82.27 ± 4.31	0.413 ± 0.120
TransUNet 3D (Chen et al., 2021)	95.78 ± 0.79	0.663 ± 0.178	83.78 ± 2.85	0.441 ± 0.147	95.23 ± 1.51	0.705 ± 0.209	80.41 ± 4.51	0.480 ± 0.146
Our KCB-Net method	96.47 ± 0.88	0.542 ± 0.178	86.73 ± 2.76	0.349 ± 0.138	96.49 ± 1.59	0.524 ± 0.214	84.34 ± 4.27	0.416 ± 0.131
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.