Skip to main content
. 2023 Mar 24;23(7):3420. doi: 10.3390/s23073420

Table 2.

Performance comparison on BraTS 2021 dataset.

Methods FLOPs(G) Params(M) Average WT ET TC
HD95 ↓ DSC ↑ HD95 ↓ DSC ↑ HD95 ↓ DSC ↑ HD95 ↓ DSC ↑
Optimized UNet [34] 1345.57 51.23 10.62 87.84 5.66 92.21 17.18 83.67 9.04 87.65
Extending nnUNet [35] 1657.97 87.41 2.99 92.69 3.62 93.40 2.50 90.79 2.86 93.89
UNETR [8] 175.85 102.45 11.04 86.18 6.18 91.40 18.31 81.80 8.63 85.33
Swin UNETR [10] 1218.79 🟉 240.96 9.84 86.13 6.06 90.42 15.27 83.36 8.19 84.62
VT-UNet-B [9] 124.03 20.76 13.36 87.31 8.96 91.20 18.69 83.85 12.44 86.87
HRSTNet-2 107.47 7.19 8.99 87.22 4.15 92.06 15.63 82.64 7.20 86.95
HRSTNet-3 202.76 50.07 8.06 86.94 4.79 91.81 13.91 82.20 5.50 86.80
HRSTNet-4 318.43 266.33 8.94 87.48 4.09 91.90 15.62 82.92 7.11 87.62

🟉 This value is obtained by setting the input size to 4 × 96 × 96 × 96.