Table 6. Sensitivity boost of the ‘un-curated’ dataset with three labels, for healthy and diseased categories.
Margin Max (0.4) Boosting Healthy | Original | Margin Max (0.4) Boosting Diseased | |
---|---|---|---|
accuracy | 0.8680 | 0.8681 | 0.8444 |
Sensitivity Healthy | 0.9831 | 0.9665 | 0.9154 |
Sensitivity Non-referable DR | 0.5503 | 0.5855 | 0.6203 |
Sensitivity Referable DR | 0.5026 | 0.6093 | 0.7522 |
Specificity Healthy | 0.8840 | 0.9005 | 0.923 |
Specificity Non-referable DR | 0.8004 | 0.777 | 0.6707 |
Specificity Referable DR | 0.7624 | 0.6232 | 0.5052 |
F1 Score Healthy | 0.9309 | 0.9323 | 0.9192 |
F1 Score Non-referable DR | 0.6522 | 0.6678 | 0.6445 |
F1 Score Referable DR | 0.6058 | 0.6162 | 0.6044 |
PPV Healthy | 0.8840 | 0.9005 | 0.923 |
PPV Non-referable DR | 0.8004 | 0.7770 | 0.6707 |
PPV Referable DR | 0.7624 | 0.6232 | 0.5052 |
NPV Healthy | 0.9297 | 0.8801 | 0.7659 |
NPV Non-referable DR | 0.8862 | 0.8934 | 0.8978 |
NPV Referable DR | 0.9767 | 0.9814 | 0.9879 |