Table 5.
Comparison of CRLM segmentation results from our model and the literature.
| Method | Model | Disease diameter and source | No. of sites | Mean DSC | Sensitivity |
|---|---|---|---|---|---|
| Presented method | Hybrid-WNet | <10 mm | 42 | 0.00 | 7% |
| 10–20 mm | 49 | 0.43 | 59% | ||
| 15–20 mm | 18 | 0.68 | 94% | ||
| ≥20 mm | 30 | 0.77 | 98% | ||
| LiTS | 70 patients | 0.810 global | – | ||
| 3D-IRCADb | 15 patients | 0.69 | – | ||
| Vorontsov et al. (8) | FCN | <10 mm | 30 | 0.14 | 10% |
| 10–20 mm | 35 | 0.53 | 71% | ||
| >20 mm | 40 | 0.68 | 85% | ||
| Li et al. (9) | Hybrid Dense UNet | LiTS | 70 patients | 0.824 global | – |
| Seo et al. (10) | mU-Net | 3D-IRCADb | 5 patients | 0.68 | – |
Patients were specified instead of sites for LiTS test submission.