Tissue Segmentation on Lower extremity CT scans v1 (test)
93.27Dice SATnnUNet ResEnc-L
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| nnUNet ResEnc-LFamily=LegSegNet, Architecture=Residual-encoder UNet (L), Training=full, Input Resolution=192 x 2562026.05 | 93.27 | 92.76 | 72.42 | 97.05 | 88.87 | 0.29 | 0.418 | 1.138 | 0.331 | 0.544 | |
| nnUNet ResEnc-XLFamily=LegSegNet, Architecture=Residual-encoder UNet (XL), Training=full, Input Resolution=192 x 2562026.05 | 93.23 | 92.79 | 73.09 | 97.1 | 89.05 | 0.288 | 0.406 | 1.133 | 0.503 | 0.583 | |
| nnUNet (default 2D)Family=LegSegNet, Architecture=PlainConvUNet, Training=full, Input Resolution=192 x 2562026.05 | 93.19 | 92.92 | 73.98 | 97.15 | 89.31 | 0.299 | 0.404 | 1.062 | 0.296 | 0.515 | |
| SegResNetFamily=CNN-based, Architecture=Residual encoder-decoder, Training=full, Input Resolution=256 x 2562026.05 | 92.61 | 92.63 | 72.19 | 96.8 | 88.56 | 0.435 | 0.445 | 1.146 | 0.623 | 0.662 | |
| Attention UNetFamily=CNN-based, Architecture=UNet + attention gates, Training=full, Input Resolution=256 x 2562026.05 | 92.46 | 92.1 | 70.08 | 96.09 | 87.68 | 0.474 | 0.517 | 1.295 | 0.93 | 0.804 | |
| SwinUNETRFamily=Transformer-based, Architecture=Swin encoder + CNN decoder, Training=full, Input Resolution=256 x 2562026.05 | 92.16 | 91.43 | 69.08 | 95.18 | 86.96 | 0.363 | 0.559 | 1.263 | 0.944 | 0.782 | |
| UNetFamily=CNN-based, Architecture=Residual UNet, Training=full, Input Resolution=256 x 2562026.05 | 92.11 | 91.51 | 70.92 | 96.04 | 87.65 | 0.356 | 0.593 | 1.191 | 0.808 | 0.737 | |
| EffNet-B0 UNetFamily=CNN-based, Architecture=EfficientNet-B0 encoder + UNet decoder, Training=full, Input Resolution=256 x 2562026.05 | 92.04 | 91.63 | 69.66 | 96.16 | 87.37 | 0.364 | 0.532 | 1.266 | 0.687 | 0.712 | |
| SAM ViT-BFamily=Transformer-based / Foundation model, Architecture=ViT-B encoder + SAM decoder + adapters, Pretraining=SA-1B, Training=adapters only2026.05 | 90.93 | 90.78 | 68.55 | 95.35 | 86.4 | 0.399 | 0.592 | 1.175 | 0.483 | 0.662 | |
| MedSAM ViT-BFamily=Transformer-based / Foundation model, Architecture=ViT-B encoder + SAM decoder + adapters, Pretraining=1.5M medical (incl. CT), Training=adapters only2026.05 | 82.48 | 88.21 | 55.02 | 90.96 | 79.17 | 0.774 | 0.852 | 1.792 | 0.724 | 1.036 |