Table 3 Comparison of the BraTS 2020 segmentation results for the ablation experiment of the 3D channel spatial attention supervision module.
Method | Dice | Sensitivity | ||||||
---|---|---|---|---|---|---|---|---|
ET | WT | TC | Mean | ET | WT | TC | Mean | |
DTASUnet without CSAS | 0.802 | 0.923 | 0.884 | 0.870 | 0.810 | 0.929 | 0.885 | 0.875 |
DTASUnet with 3D CSA | 0.822 | 0.919 | 0.875 | 0.872 | 0.831 | 0.926 | 0.882 | 0.880 |
DTASUnet with supervision loss | 0.812 | 0.924 | 0.893 | 0.876 | 0.819 | 0.931 | 0.893 | 0.881 |
DTASUnet with 3D CSAS | 0.836 | 0.925 | 0.885 | 0.882 | 0.842 | 0.920 | 0.878 | 0.880 |