Table 2.

The average four-fold performance on two public dataset (the performance of our method is described by mean ± std).

Method	DSC (%)	Jaccard (%)	Recall (%)	Precision (%)
NIH dataset
Bottom-up (32)	70.7	57.9	71.6	74.4
Fixed-point (53)	82.4	–	–	–
3D Coarse-to-Fine (54)	84.6	–	–	–
Holistically nested (55)	81.3	68.9	–	–
RSTN (31)	84.5	–	–	–
Recurrent Contextual Learning (39)	83.3	71.8	84.5	82.8
Vnet (56)	80.1	–	–	–
Attention Unet (57)	83.1	–	–	–
DenseASPP (40)	85.4	–	–	–
(46)	84.10	72.86	85.3	83.6
Cascaded FCN (23)	85.9	75.7	85.2	87.6
AX-Unet (Ours)	87.7 ± 3.8	78.2 ± 5.3	90.9 ± 2.2	92.9 ± 6.1
MSD dataset
Unet-64	70.7	–	–	–
Unet-16	67.1	–	–	–
Attention Unet (57)	66.0	–	–	–
MoNet (58)	74.0	68.9	–	–
nn-Unet (27)	80.0	–	–	–
AX-Unet (Ours)	85.9 ± 5.1	77.9 ± 3.4	86.3 ± 5.1	93.1 ± 6.9