Table 1.

Comparison with state-of-the-art methods using full annotation on the SKI10 dataset following the data split in Ambellan et al. (2019) and evaluation metrics used in Heimann et al. (2010). 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		Overall score
	ASSD (mm)	RSSD (mm)	VOE (%)	VD (%)	ASSD (mm)	RSSD (mm)	VOE (%)	VD (%)
CNN-SSM (Ambellan et al., 2019)a	0.430 ± 0.130	0.740 ± 0.270	20.99 ± 5.08	7.18 ± 10.51	0.350 ± 0.070	0.590 ± 0.190	19.06 ± 5.18	4.29 ± 12.34	74.00 ± 7.70
UNet++ 3D (Zhou et al., 2018)	0.541 ± 0.096	0.694 ± 0.252	20.86 ± 5.01	3.90 ± 11.79	0.521 ± 0.164	0.672 ± 0.448	20.07 ± 5.62	5.35 ± 12.35	72.03 ± 8.29
TransUNet 3D (Chen et al., 2021)	0.538 ± 0.072	0.680 ± 0.196	21.36 ± 5.02	5.42 ± 10.31	0.517 ± 0.152	0.654 ± 0.418	20.01 ± 5.44	6.04 ± 12.68	72.20 ± 8.60
Attention UNet 3D (Oktay et al., 2018)	0.519 ± 0.083	0.690 ± 0.454	18.77 ± 4.74	1.36 ± 9.69	0.519 ± 0.259	0.664 ± 0.648	18.14 ± 4.87	6.19 ± 11.21	74.54 ± 6.50
Ensemble method (Zheng et al., 2020b)	0.689 ± 0.858	0.732 ± 0.871	18.47 ± 4.75	4.71 ± 9.73	0.508 ± 0.200	0.640 ± 0.533	18.19 ± 5.11	3.00 ± 11.15	73.82 ± 9.51
Our KCB-Net method	0.498 ± 0.053	0.579 ± 0.104	18.66 ± 4.54	−1.06 ± 9.20	0.504 ± 0.240	0.516 ± 0.602	17.60 ± 4.65	0.92 ± 10.73	75.94 ± 6.08
p-value	0.016	0.011	0.621	0.033	0.547	≪0.001	0.041	≪0.001	0.001

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