Table 1.
Distribution of outcome classes according to the tilt status in the development and test datasets.
| Development dataset | All (n = 2005, k = 1555) |
Non-tilted disc (n = 1198, k = 992) |
Tilted disc (n = 807, k = 640) |
|
|---|---|---|---|---|
| Class, n (%) | Normal | 1336 (66.6) | 749 (62.5) | 587 (72.7) |
| Glaucoma | 382 (19.1) | 230 (19.2) | 152 (18.8) | |
| Optic disc pallor | 196 (9.8) | 141 (11.8) | 55 (6.8) | |
| Optic disc swelling | 91 (4.5) | 78 (6.5) | 13 (1.6) | |
| Test dataset | All (n = 502, k = 464) |
Non-tilted disc (n = 299, k = 282) |
Tilted disc (n = 203, k = 189) |
|
|---|---|---|---|---|
| Class, n (%) | Normal | 335 (66.7) | 179 (59.9) | 156 (76.8) |
| Glaucoma | 95 (18.9) | 63 (21.1) | 32 (15.8) | |
| Optic disc pallor | 49 (9.8) | 38 (12.7) | 11 (5.4) | |
| Optic disc swelling | 23 (4.6) | 19 (6.4) | 4 (2.0) | |
This table presents the number (percent) of images annotated with each outcome class in each dataset category, illustrating the allocation of data for model training and evaluation. n = numbers of images. k = numbers of patients. Note that, some of the collected images belong to the same individual patients.