Table 4 Our domain adaptation strategy improves results in a wide range of external test sets.
From: State-of-the-art retinal vessel segmentation with minimalistic models
Training set | DRIVE | CHASE-DB | HRF | STARE | IOSTAR | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
AUC | DICE | MCC | AUC | DICE | MCC | AUC | DICE | MCC | AUC | DICE | MCC | AUC | DICE | MCC | |
DRIVE | 98.09 | 82.82 | 80.27 | 97.22 | 75.13 | 72.44 | 95.90 | 70.39 | 68.05 | 98.11 | 79.48 | 77.30 | 97.97 | 78.77 | 76.47 |
PSEUDO-L | 98.09 | 82.82 | 80.27 | 97.56 | 76.49 | 74.02 | 96.12 | 71.12 | 68.86 | 98.28 | 79.76 | 77.65 | 98.06 | 78.95 | 76.73 |
Training Set | DRiDB | LES-AV | DR HAGIS | AV-WIDE | UoA-DR | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
AUC | DICE | MCC | AUC | DICE | MCC | AUC | DICE | MCC | AUC | DICE | MCC | AUC | DICE | MCC | |
DRIVE | 96.17 | 68.45 | 66.62 | 95.45 | 76.60 | 74.32 | 97.17 | 67.92 | 66.79 | 86.54 | 61.51 | 59.02 | 82.32 | 38.29 | 35.51 |
PSEUDO-L | 96.52 | 68.25 | 66.59 | 97.34 | 77.93 | 75.92 | 97.34 | 68.67 | 67.49 | 87.64 | 62.46 | 59.97 | 82.71 | 37.68 | 34.97 |
