Aggregation-and-Attention Network for brain tumor segmentation

BMC Medical Imaging

Table 3 The ablation experiments of MCS module’s position

1	2	3	4	Dice↑			Precision↑			Sensitivity↑			Hausdorff↓
1	2	3	4	WT	CT	ET	WT	CT	ET	WT	CT	ET	WT	CT	ET
Baseline				0.806	0.886	0.750	0.814	0.938	0.745	0.954	0.913	0.938	1.485	0.942	1.633
√				0.856	0.890	0.799	0.872	0.941	0.803	0.947	0.913	0.929	1.343	0.940	1.498
√	√			0.846	0.890	0.790	0.858	0.949	0.788	0.951	0.907	0.937	1.371	0.953	1.526
√	√	√		0.844	0.896	0.791	0.855	0.946	0.792	0.951	0.919	0.932	1.373	0.924	1.513
√	√	√	√	0.869	0.896	0.814	0.886	0.952	0.818	0.950	0.912	0.934	1.308	0.934	1.456

The best results are marked with bold
^**1 ~ 4 are the positions of the MSC module. The smaller the feature map processed by the MSC module, the larger the corresponding position number value. For example, 1 represents the position connected with the first EDS module with the largest feature map

ISSN: 1471-2342