Table 5.
Performance comparison for models trained and tested with CHASE_DB1.
| Methods | SN | SP | Pr | Acc | AUC | kappa | G | MCC | F1 |
|---|---|---|---|---|---|---|---|---|---|
| Human (2nd Observer) | 0.7425 | 0.9793 | 0.8090 | 0.9560 | - | 0.7529 | 0.8527 | 0.7475 | 0.7686 |
| Unsupervised Methods | |||||||||
| Azzopardi et al. [8] | 0.7585 | 0.9587 | - | 0.9387 | 0.9487 | - | 0.8527 | 0.6802 | - |
| Zhang et al. [44] | 0.7626 | 0.9661 | - | 0.9452 | 0.9606 | - | 0.8583 | - | - |
| Roychowdhury et al. [45] | 0.7615± 0.0516 | 0.9575± 0.003 | - | 0.9467± 0.0076 | 0.9623 | - | 0.8539± 0.0124 | - | - |
| Supervised Methods | |||||||||
| Fraz et al. [7] | 0.7224 | 0.9711 | - | 0.9469 | 0.9712 | - | 0.8376 | - | - |
| Li et al. [51] | 0.7507 | 0.9793 | - | 0.9581 | 0.9716 | - | 0.8574 | - | - |
| Liskowski et al. [52] | 0.7816± 0.0178 | 0.9836± 0.0022 | - | 0.9628± 0.0020 | 0.9823± 0.0016 | 0.7908± 0.0111 | 0.8768± 0.0063 | - | - |
| Mo et al. [55] | 0.7661 ± 0.0533 | 0.9816± 0.0076 | - | 0.9599± 0.0050 | 0.9812± 0.0040 | 0.8672± 0.0201 | 0.7689± 0.0263 | - | - |
| Orlando et al. [38] | 0.7277 | 0.9712 | 0.7438 | - | - | - | 0.8403 | 0.7046 | 0.7332 |
| PixelBNN | 0.8618± 0.0232 | 0.8961± 0.0150 | 0.3951± 0.0603 | 0.8936± 0.0138 | 0.878959± 0.0138 | 0.4889± 0.0609 | 0.8787± 0.0140 | 0.5376± 0.0491 | 0.5391± 0.0587 |