Table 3.
Results achieved by varying input pooling size of the HYPER-RETINO system.
| No. | Architecture | SE% | SP% | ACC% | PR% | MCC% | F1% | AUC | E |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 6-P-ND-M | 91.5 | 89.5 | 90 | 89.3 | 61 | 88 | 0.89 | 0.64 |
| 2 | 6-P-D-A | 89.2 | 88.1 | 88 | 89.2 | 58 | 87 | 0.88 | 0.67 |
| 3 | 7-P-ND-M | 87.3 | 86.5 | 86 | 87.3 | 57 | 86 | 0.87 | 0.69 |
| 4 | 7-P-D-A | 86.4 | 85.2 | 85.4 | 86.4 | 55 | 85 | 0.86 | 0.72 |
| 5 | 8-P-ND-M | 84.1 | 83.5 | 83.1 | 84.1 | 54 | 84 | 0.84 | 0.74 |
| 6 | 8-P-D-A | 80.6 | 79.1 | 79.3 | 80.6 | 53 | 81 | 0.80 | 0.76 |
| 7 | 9-P-ND-M | 78.7 | 76.5 | 77.5 | 78.7 | 51 | 79 | 0.78 | 0.78 |
| 8 | 9-P-D-A | 75.5 | 74.6 | 74.4 | 75.5 | 48 | 75 | 0.75 | 0.80 |
| 9 | 10-P-ND-M | 73.3 | 72.5 | 72.1 | 73.3 | 46 | 72 | 0.73 | 0.83 |
| 10 | 10-P-D-A | 72.2 | 71.1 | 71.6 | 72.2 | 45 | 71 | 0.72 | 0.85 |
-P-ND-M: Pooling-no dropout-maximum layers, -P-D-A: Pooling-dropout-Average layers, MCC: Matthews correlation coefficient, SE: Sensitivity, SP: Specificity, F1: F1 score, ACC: Accuracy, E: Training errors and AUC: Area under the receiver operating curve.