Table 1 DSC, HD(mm) and ASD(mm) of results by different comparative methods. T: the target domain dataset (BTR). S: the source domain dataset (NC). S(it): the source domain dataset with intensity transfer only. S\(\rightarrow\)T(it): synthetic images from the source domain (postoperative transfer and intensity transfer)
From: Hippocampus segmentation after brain tumor resection via postoperative region synthesis
Model | Dataset | DSC | HD(mm) | ASD(mm) |
|---|---|---|---|---|
3D U-Net | T | 0.6971 | 17.20 | 1.82 |
3D U-Net | S+T | 0.6986 | 10.90 | 3.32 |
SegNet | S+T | 0.7199 | 10.66 | 1.39 |
Ours w/o VAE-GAN | S(it)+T | 0.7354 | 5.06 | 0.83 |
Ours (full) | S\(\rightarrow\)T(it)+T | 0.7429 | 9.40 | 1.13 |
Ours (full, train jointly) | S\(\rightarrow\)T(it)+T | 0.7546 | 4.68 | 0.75 |