Table 2.
Results of ISIC 2019 challenge winners from the legacy leaderboard (rows 1-3) and our proposed models (rows 4–6). Among the 1,572 images, there are 170 images from the MED-NODE dataset, 533 from the seven-point dataset, 120 from the PH2 dataset, and the remained data are from our own collected data.
| Team/authors | Extra images | BMCA (%) | Sensitivity (%) | Specificity (%) | AUC |
|---|---|---|---|---|---|
| Gassert et al. [14] | Unknown | 63.6 | 50.7 | 97.7 | 0.923 |
| Cancerless | Unknown | 63.8 | 53.1 | 97.4 | 0.913 |
| ForCure | Unknown | 64.8 | 53.4 | 97.4 | 0.914 |
| Our single model (FL) | 1,572 | 63.9 | 48.8 | 97.9 | 0.899 |
| Our single model (CE) | 1,572 | 65.0 | 55.3 | 97.3 | 0.906 |
| Our ensemble model | 1,572 | 66.2 | 66.2 | 95.2 | 0.915 |
FL: focal loss; CE: cross-entropy loss.