Table 3. Quantitative comparison on the BraTS 2021 dataset.
| Methods | WT | TC | ET | Average | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| DSC↑ | HD95↓ (mm) | DSC↑ | HD95↓ (mm) | DSC↑ | HD95↓ (mm) | DSC↑ | HD95↓ (mm) | ||||
| Attention-UNet (9) | 0.910 | 8.990 | 0.869 | 6.572 | 0.841 | 5.302 | 0.873 | 6.952 | |||
| DAUnet (14) | 0.899 | 6.700 | 0.844 | 6.600 | 0.776 | 14.400 | 0.839 | 9.233 | |||
| SegResNet (15) | 0.917 | 6.113 | 0.896 | 5.790 | 0.861 | 5.423 | 0.891 | 5.781 | |||
| nnUNet (16) | 0.926 | 3.550 | 0.874 | 10.560 | 0.837 | 22.440 | 0.879 | 12.183 | |||
| TransBTS (20) | 0.920 | 4.980 | 0.882 | 4.860 | 0.795 | 16.320 | 0.866 | 8.720 | |||
| TransUNet (21) | 0.919 | 6.162 | 0.877 | 7.340 | 0.818 | 13.090 | 0.872 | 8.860 | |||
| UNETR (22) | 0.890 | 14.423 | 0.847 | 10.221 | 0.825 | 8.785 | 0.854 | 11.142 | |||
| SwinUNETR (23) | 0.926 | 5.831 | 0.885 | 3.770 | 0.858 | 6.016 | 0.890 | 5.206 | |||
| SwinBTS (38) | 0.918 | 3.650 | 0.848 | 14.510 | 0.832 | 16.030 | 0.866 | 11.397 | |||
| CKD-TransBTS (24) | 0.923 | 4.230 | 0.881 | 4.390 | 0.848 | 3.160 | 0.884 | 3.927 | |||
| FCFDiff-Net | 0.926* | 2.156* | 0.903* | 1.834* | 0.869* | 1.583* | 0.899* | 1.858* | |||
Higher DSC scores (↑) indicate better segmentation, while lower HD95 values (↓) indicate better performance. The top result is marked with asterisk (*). BraTS, brain tumor segmentation; DSC, Dice similarity coefficient; ET, enhancing tumor; HD95, Hausdorff distance at the 95th percentile; TC, tumor core; WT, whole tumor.