Table 4 Classification performance for various ratios between normal and ERM. AUC, accuracy, sensitivity, and specificity were shown for each ratio with and without adding synthesized ERM images.
No. ratio of real dataset (Normal:ERM) | Add synthesized ERM* | AUC | Accuracy | Sensitivity | Specificity |
|---|---|---|---|---|---|
1:1 | No | 0.994 | 0.971 | 0.965 | 0.977 |
1:0.5 | No | 0.988 | 0.963 | 0.955 | 0.972 |
Yes | 0.989 | 0.970 | 0.957 | 0.982 | |
1:0.4 | No | 0.983 | 0.943 | 0.909 | 0.977 |
Yes | 0.994 | 0.970 | 0.942 | 0.997 | |
1:0.3 | No | 0.984 | 0.905 | 0.826 | 0.985 |
Yes | 0.987 | 0.968 | 0.947 | 0.990 | |
1:0.2 | No | 0.943 | 0.874 | 0.843 | 0.904 |
Yes | 0.966 | 0.914 | 0.904 | 0.924 | |
1:0.1 | No | 0.735 | 0.559 | 0.174 | 0.944 |
Yes | 0.909 | 0.739 | 0.508 | 0.970 |
