Table 6.
Ground-truth and performance comparison for three-stage glaucoma classification.
| Database and number of images used | Ground truth | RDR-based results (DDLS) | CDR-based results (ICD-9) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| N | Mod | S | N | Mod | S | OCA | N | Mod | S | OCA | |
| Drishti-GS (50) | 13 | 32 | 05 | 19 | 31 | 00 | 42/50 | 07 | 42 | 01 | 36/50 |
| Drions-DB (100) | 78 | 22 | 00 | 77 | 20 | 03 | 82/100 | 73 | 25 | 02 | 78/100 |
| Messidor (100) | 58 | 41 | 01 | 60 | 40 | 00 | 84/100 | 52 | 47 | 01 | 70/100 |
| Forus (126) | 100 | 23 | 03 | 97 | 27 | 02 | 121/126 | 88 | 37 | 01 | 106/126 |
| Bosch (60) | 60 | 00 | 00 | 59 | 01 | 00 | 59/60 | 56 | 04 | 00 | 56/60 |
| Consolidated (436 images) | 309 | 118 | 09 | 312 | 119 | 05 | 388/436 (89%) | 312 | 119 | 05 | 346/436(79.36%) |
(N – Normal; Mod – Moderate; S – Severe; OCA – Overall Classification Accuracy).