Performance comparison between the algorithm proposed in ‘DL-based segmentation of gonioscopic images’14 and ours
| Previous model14 | New model | ||||
| Mean (%) | SD (%) | Mean (%) | SD (%) | ||
| I | Prec. | 96.2 | 0.4 | 92.8 | 1.3 |
| Sens. | 91.0 | 2.2 | 97.4 | 0.8 | |
| Dice | 93.6 | 1.0 | 94.8 | 0.4 | |
| CBB | Prec. | 67.0 | 5.0 | 84.8 | 1.0 |
| Sens. | 75.2 | 4.2 | 64.0 | 4.3 | |
| Dice | 70.4 | 2.0 | 72.8 | 2.9 | |
| SS | Prec. | 53.2 | 4.2 | 68.4 | 1.7 |
| Sens. | 78.6 | 2.9 | 69.6 | 3.3 | |
| Dice | 63.2 | 2.8 | 68.8 | 1.2 | |
| TM | Prec. | 84.0 | 1.9 | 84.6 | 2.0 |
| Sens. | 85.0 | 3.0 | 91.2 | 1.2 | |
| Dice | 84.4 | 1.9 | 87.6 | 0.5 | |
| C | Prec. | 95.8 | 1.2 | 96.4 | 0.5 |
| Sens. | 89.2 | 3.2 | 92.2 | 1.2 | |
| Dice | 92.0 | 1.5 | 94.2 | 0.4 | |
Bold values indicate the best model in each of the comparisons (each row of the table)
C, cornea; CBB, ciliary body band; I, iris root; SS, scleral spur; TM, trabecular meshwork.