| | CV | ICC | Sensitivity | Specificity | Kappa |
|---|
AD3RI
| 28,8% | 0,92 | 0,68 | 0,96 | 0,58 |
THRESH
| 41,1% | 0,67 | 0,74 | 0,94 | 0,49 |
OP1
| 28,3% | 0,86 | 0,58 | 0,98 | 0,55 |
OP2
| 23,6% | 0,79 | 0,66 | 0,98 | 0,61 |
OP3
| 15,0% | 0,92 | 0,69 | 0,97 | 0,64 |
OP4
| 19,6% | 0,92 | 0,61 | 0,96 | 0,53 |
TE1
| 16,3% | 0,92 | 0,66 | 0,98 | 0,64 |
TE2
| 14,2% | 0,89 | 0,65 | 0,98 | 0,62 |
TE3
| 21,9% | 0,90 | 0,76 | 0,96 | 0,64 |
TE4
| 41,2% | 0,82 | 0,77 | 0,93 | 0,57 |
- This table presents the average CV, ICC, sensitivity, specificity and kappa coefficient obtained by AD3RI and by the Thresholding method compared to the manual gradings done by each of the experts (OP - Ophthalmologist; TE - Technician). These results exclude the images considered as outliers. As it can be seen, AD3RI and manual gradings obtained similar values; only for CV the AD3RI results were above the average. This is justified by the level of detail of the AD3RI analysis that is usually superior to the manual analysis. AD3RI obtained better results than the Thresholding method on all the indicators except on sensitivity. However, Thresholding sensitivity is higher due to its over-detection of drusen what consequently decreases its performance on the other indicators.
